OpenCloudOS-Kernel/mm/slab.c

4232 lines
106 KiB
C
Raw Normal View History

License cleanup: add SPDX GPL-2.0 license identifier to files with no license Many source files in the tree are missing licensing information, which makes it harder for compliance tools to determine the correct license. By default all files without license information are under the default license of the kernel, which is GPL version 2. Update the files which contain no license information with the 'GPL-2.0' SPDX license identifier. The SPDX identifier is a legally binding shorthand, which can be used instead of the full boiler plate text. This patch is based on work done by Thomas Gleixner and Kate Stewart and Philippe Ombredanne. How this work was done: Patches were generated and checked against linux-4.14-rc6 for a subset of the use cases: - file had no licensing information it it. - file was a */uapi/* one with no licensing information in it, - file was a */uapi/* one with existing licensing information, Further patches will be generated in subsequent months to fix up cases where non-standard license headers were used, and references to license had to be inferred by heuristics based on keywords. The analysis to determine which SPDX License Identifier to be applied to a file was done in a spreadsheet of side by side results from of the output of two independent scanners (ScanCode & Windriver) producing SPDX tag:value files created by Philippe Ombredanne. Philippe prepared the base worksheet, and did an initial spot review of a few 1000 files. The 4.13 kernel was the starting point of the analysis with 60,537 files assessed. Kate Stewart did a file by file comparison of the scanner results in the spreadsheet to determine which SPDX license identifier(s) to be applied to the file. She confirmed any determination that was not immediately clear with lawyers working with the Linux Foundation. Criteria used to select files for SPDX license identifier tagging was: - Files considered eligible had to be source code files. - Make and config files were included as candidates if they contained >5 lines of source - File already had some variant of a license header in it (even if <5 lines). All documentation files were explicitly excluded. The following heuristics were used to determine which SPDX license identifiers to apply. - when both scanners couldn't find any license traces, file was considered to have no license information in it, and the top level COPYING file license applied. For non */uapi/* files that summary was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 11139 and resulted in the first patch in this series. If that file was a */uapi/* path one, it was "GPL-2.0 WITH Linux-syscall-note" otherwise it was "GPL-2.0". Results of that was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 WITH Linux-syscall-note 930 and resulted in the second patch in this series. - if a file had some form of licensing information in it, and was one of the */uapi/* ones, it was denoted with the Linux-syscall-note if any GPL family license was found in the file or had no licensing in it (per prior point). Results summary: SPDX license identifier # files ---------------------------------------------------|------ GPL-2.0 WITH Linux-syscall-note 270 GPL-2.0+ WITH Linux-syscall-note 169 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) 21 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) 17 LGPL-2.1+ WITH Linux-syscall-note 15 GPL-1.0+ WITH Linux-syscall-note 14 ((GPL-2.0+ WITH Linux-syscall-note) OR BSD-3-Clause) 5 LGPL-2.0+ WITH Linux-syscall-note 4 LGPL-2.1 WITH Linux-syscall-note 3 ((GPL-2.0 WITH Linux-syscall-note) OR MIT) 3 ((GPL-2.0 WITH Linux-syscall-note) AND MIT) 1 and that resulted in the third patch in this series. - when the two scanners agreed on the detected license(s), that became the concluded license(s). - when there was disagreement between the two scanners (one detected a license but the other didn't, or they both detected different licenses) a manual inspection of the file occurred. - In most cases a manual inspection of the information in the file resulted in a clear resolution of the license that should apply (and which scanner probably needed to revisit its heuristics). - When it was not immediately clear, the license identifier was confirmed with lawyers working with the Linux Foundation. - If there was any question as to the appropriate license identifier, the file was flagged for further research and to be revisited later in time. In total, over 70 hours of logged manual review was done on the spreadsheet to determine the SPDX license identifiers to apply to the source files by Kate, Philippe, Thomas and, in some cases, confirmation by lawyers working with the Linux Foundation. Kate also obtained a third independent scan of the 4.13 code base from FOSSology, and compared selected files where the other two scanners disagreed against that SPDX file, to see if there was new insights. The Windriver scanner is based on an older version of FOSSology in part, so they are related. Thomas did random spot checks in about 500 files from the spreadsheets for the uapi headers and agreed with SPDX license identifier in the files he inspected. For the non-uapi files Thomas did random spot checks in about 15000 files. In initial set of patches against 4.14-rc6, 3 files were found to have copy/paste license identifier errors, and have been fixed to reflect the correct identifier. Additionally Philippe spent 10 hours this week doing a detailed manual inspection and review of the 12,461 patched files from the initial patch version early this week with: - a full scancode scan run, collecting the matched texts, detected license ids and scores - reviewing anything where there was a license detected (about 500+ files) to ensure that the applied SPDX license was correct - reviewing anything where there was no detection but the patch license was not GPL-2.0 WITH Linux-syscall-note to ensure that the applied SPDX license was correct This produced a worksheet with 20 files needing minor correction. This worksheet was then exported into 3 different .csv files for the different types of files to be modified. These .csv files were then reviewed by Greg. Thomas wrote a script to parse the csv files and add the proper SPDX tag to the file, in the format that the file expected. This script was further refined by Greg based on the output to detect more types of files automatically and to distinguish between header and source .c files (which need different comment types.) Finally Greg ran the script using the .csv files to generate the patches. Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org> Reviewed-by: Philippe Ombredanne <pombredanne@nexb.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-11-01 22:07:57 +08:00
// SPDX-License-Identifier: GPL-2.0
/*
* linux/mm/slab.c
* Written by Mark Hemment, 1996/97.
* (markhe@nextd.demon.co.uk)
*
* kmem_cache_destroy() + some cleanup - 1999 Andrea Arcangeli
*
* Major cleanup, different bufctl logic, per-cpu arrays
* (c) 2000 Manfred Spraul
*
* Cleanup, make the head arrays unconditional, preparation for NUMA
* (c) 2002 Manfred Spraul
*
* An implementation of the Slab Allocator as described in outline in;
* UNIX Internals: The New Frontiers by Uresh Vahalia
* Pub: Prentice Hall ISBN 0-13-101908-2
* or with a little more detail in;
* The Slab Allocator: An Object-Caching Kernel Memory Allocator
* Jeff Bonwick (Sun Microsystems).
* Presented at: USENIX Summer 1994 Technical Conference
*
* The memory is organized in caches, one cache for each object type.
* (e.g. inode_cache, dentry_cache, buffer_head, vm_area_struct)
* Each cache consists out of many slabs (they are small (usually one
* page long) and always contiguous), and each slab contains multiple
* initialized objects.
*
* This means, that your constructor is used only for newly allocated
* slabs and you must pass objects with the same initializations to
* kmem_cache_free.
*
* Each cache can only support one memory type (GFP_DMA, GFP_HIGHMEM,
* normal). If you need a special memory type, then must create a new
* cache for that memory type.
*
* In order to reduce fragmentation, the slabs are sorted in 3 groups:
* full slabs with 0 free objects
* partial slabs
* empty slabs with no allocated objects
*
* If partial slabs exist, then new allocations come from these slabs,
* otherwise from empty slabs or new slabs are allocated.
*
* kmem_cache_destroy() CAN CRASH if you try to allocate from the cache
* during kmem_cache_destroy(). The caller must prevent concurrent allocs.
*
* Each cache has a short per-cpu head array, most allocs
* and frees go into that array, and if that array overflows, then 1/2
* of the entries in the array are given back into the global cache.
* The head array is strictly LIFO and should improve the cache hit rates.
* On SMP, it additionally reduces the spinlock operations.
*
* The c_cpuarray may not be read with enabled local interrupts -
* it's changed with a smp_call_function().
*
* SMP synchronization:
* constructors and destructors are called without any locking.
* Several members in struct kmem_cache and struct slab never change, they
* are accessed without any locking.
* The per-cpu arrays are never accessed from the wrong cpu, no locking,
* and local interrupts are disabled so slab code is preempt-safe.
* The non-constant members are protected with a per-cache irq spinlock.
*
* Many thanks to Mark Hemment, who wrote another per-cpu slab patch
* in 2000 - many ideas in the current implementation are derived from
* his patch.
*
* Further notes from the original documentation:
*
* 11 April '97. Started multi-threading - markhe
* The global cache-chain is protected by the mutex 'slab_mutex'.
* The sem is only needed when accessing/extending the cache-chain, which
* can never happen inside an interrupt (kmem_cache_create(),
* kmem_cache_shrink() and kmem_cache_reap()).
*
* At present, each engine can be growing a cache. This should be blocked.
*
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
* 15 March 2005. NUMA slab allocator.
* Shai Fultheim <shai@scalex86.org>.
* Shobhit Dayal <shobhit@calsoftinc.com>
* Alok N Kataria <alokk@calsoftinc.com>
* Christoph Lameter <christoph@lameter.com>
*
* Modified the slab allocator to be node aware on NUMA systems.
* Each node has its own list of partial, free and full slabs.
* All object allocations for a node occur from node specific slab lists.
*/
#include <linux/slab.h>
#include <linux/mm.h>
#include <linux/poison.h>
#include <linux/swap.h>
#include <linux/cache.h>
#include <linux/interrupt.h>
#include <linux/init.h>
#include <linux/compiler.h>
[PATCH] cpuset memory spread slab cache implementation Provide the slab cache infrastructure to support cpuset memory spreading. See the previous patches, cpuset_mem_spread, for an explanation of cpuset memory spreading. This patch provides a slab cache SLAB_MEM_SPREAD flag. If set in the kmem_cache_create() call defining a slab cache, then any task marked with the process state flag PF_MEMSPREAD will spread memory page allocations for that cache over all the allowed nodes, instead of preferring the local (faulting) node. On systems not configured with CONFIG_NUMA, this results in no change to the page allocation code path for slab caches. On systems with cpusets configured in the kernel, but the "memory_spread" cpuset option not enabled for the current tasks cpuset, this adds a call to a cpuset routine and failed bit test of the processor state flag PF_SPREAD_SLAB. For tasks so marked, a second inline test is done for the slab cache flag SLAB_MEM_SPREAD, and if that is set and if the allocation is not in_interrupt(), this adds a call to to a cpuset routine that computes which of the tasks mems_allowed nodes should be preferred for this allocation. ==> This patch adds another hook into the performance critical code path to allocating objects from the slab cache, in the ____cache_alloc() chunk, below. The next patch optimizes this hook, reducing the impact of the combined mempolicy plus memory spreading hooks on this critical code path to a single check against the tasks task_struct flags word. This patch provides the generic slab flags and logic needed to apply memory spreading to a particular slab. A subsequent patch will mark a few specific slab caches for this placement policy. Signed-off-by: Paul Jackson <pj@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-24 19:16:07 +08:00
#include <linux/cpuset.h>
#include <linux/proc_fs.h>
#include <linux/seq_file.h>
#include <linux/notifier.h>
#include <linux/kallsyms.h>
#include <linux/cpu.h>
#include <linux/sysctl.h>
#include <linux/module.h>
#include <linux/rcupdate.h>
#include <linux/string.h>
#include <linux/uaccess.h>
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#include <linux/nodemask.h>
#include <linux/kmemleak.h>
[PATCH] NUMA policies in the slab allocator V2 This patch fixes a regression in 2.6.14 against 2.6.13 that causes an imbalance in memory allocation during bootup. The slab allocator in 2.6.13 is not numa aware and simply calls alloc_pages(). This means that memory policies may control the behavior of alloc_pages(). During bootup the memory policy is set to MPOL_INTERLEAVE resulting in the spreading out of allocations during bootup over all available nodes. The slab allocator in 2.6.13 has only a single list of slab pages. As a result the per cpu slab cache and the spinlock controlled page lists may contain slab entries from off node memory. The slab allocator in 2.6.13 makes no effort to discern the locality of an entry on its lists. The NUMA aware slab allocator in 2.6.14 controls locality of the slab pages explicitly by calling alloc_pages_node(). The NUMA slab allocator manages slab entries by having lists of available slab pages for each node. The per cpu slab cache can only contain slab entries associated with the node local to the processor. This guarantees that the default allocation mode of the slab allocator always assigns local memory if available. Setting MPOL_INTERLEAVE as a default policy during bootup has no effect anymore. In 2.6.14 all node unspecific slab allocations are performed on the boot processor. This means that most of key data structures are allocated on one node. Most processors will have to refer to these structures making the boot node a potential bottleneck. This may reduce performance and cause unnecessary memory pressure on the boot node. This patch implements NUMA policies in the slab layer. There is the need of explicit application of NUMA memory policies by the slab allcator itself since the NUMA slab allocator does no longer let the page_allocator control locality. The check for policies is made directly at the beginning of __cache_alloc using current->mempolicy. The memory policy is already frequently checked by the page allocator (alloc_page_vma() and alloc_page_current()). So it is highly likely that the cacheline is present. For MPOL_INTERLEAVE kmalloc() will spread out each request to one node after another so that an equal distribution of allocations can be obtained during bootup. It is not possible to push the policy check to lower layers of the NUMA slab allocator since the per cpu caches are now only containing slab entries from the current node. If the policy says that the local node is not to be preferred or forbidden then there is no point in checking the slab cache or local list of slab pages. The allocation better be directed immediately to the lists containing slab entries for the allowed set of nodes. This way of applying policy also fixes another strange behavior in 2.6.13. alloc_pages() is controlled by the memory allocation policy of the current process. It could therefore be that one process is running with MPOL_INTERLEAVE and would f.e. obtain a new page following that policy since no slab entries are in the lists anymore. A page can typically be used for multiple slab entries but lets say that the current process is only using one. The other entries are then added to the slab lists. These are now non local entries in the slab lists despite of the possible availability of local pages that would provide faster access and increase the performance of the application. Another process without MPOL_INTERLEAVE may now run and expect a local slab entry from kmalloc(). However, there are still these free slab entries from the off node page obtained from the other process via MPOL_INTERLEAVE in the cache. The process will then get an off node slab entry although other slab entries may be available that are local to that process. This means that the policy if one process may contaminate the locality of the slab caches for other processes. This patch in effect insures that a per process policy is followed for the allocation of slab entries and that there cannot be a memory policy influence from one process to another. A process with default policy will always get a local slab entry if one is available. And the process using memory policies will get its memory arranged as requested. Off-node slab allocation will require the use of spinlocks and will make the use of per cpu caches not possible. A process using memory policies to redirect allocations offnode will have to cope with additional lock overhead in addition to the latency added by the need to access a remote slab entry. Changes V1->V2 - Remove #ifdef CONFIG_NUMA by moving forward declaration into prior #ifdef CONFIG_NUMA section. - Give the function determining the node number to use a saner name. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-19 09:42:36 +08:00
#include <linux/mempolicy.h>
#include <linux/mutex.h>
#include <linux/fault-inject.h>
#include <linux/rtmutex.h>
2006-12-13 16:34:27 +08:00
#include <linux/reciprocal_div.h>
infrastructure to debug (dynamic) objects We can see an ever repeating problem pattern with objects of any kind in the kernel: 1) freeing of active objects 2) reinitialization of active objects Both problems can be hard to debug because the crash happens at a point where we have no chance to decode the root cause anymore. One problem spot are kernel timers, where the detection of the problem often happens in interrupt context and usually causes the machine to panic. While working on a timer related bug report I had to hack specialized code into the timer subsystem to get a reasonable hint for the root cause. This debug hack was fine for temporary use, but far from a mergeable solution due to the intrusiveness into the timer code. The code further lacked the ability to detect and report the root cause instantly and keep the system operational. Keeping the system operational is important to get hold of the debug information without special debugging aids like serial consoles and special knowledge of the bug reporter. The problems described above are not restricted to timers, but timers tend to expose it usually in a full system crash. Other objects are less explosive, but the symptoms caused by such mistakes can be even harder to debug. Instead of creating specialized debugging code for the timer subsystem a generic infrastructure is created which allows developers to verify their code and provides an easy to enable debug facility for users in case of trouble. The debugobjects core code keeps track of operations on static and dynamic objects by inserting them into a hashed list and sanity checking them on object operations and provides additional checks whenever kernel memory is freed. The tracked object operations are: - initializing an object - adding an object to a subsystem list - deleting an object from a subsystem list Each operation is sanity checked before the operation is executed and the subsystem specific code can provide a fixup function which allows to prevent the damage of the operation. When the sanity check triggers a warning message and a stack trace is printed. The list of operations can be extended if the need arises. For now it's limited to the requirements of the first user (timers). The core code enqueues the objects into hash buckets. The hash index is generated from the address of the object to simplify the lookup for the check on kfree/vfree. Each bucket has it's own spinlock to avoid contention on a global lock. The debug code can be compiled in without being active. The runtime overhead is minimal and could be optimized by asm alternatives. A kernel command line option enables the debugging code. Thanks to Ingo Molnar for review, suggestions and cleanup patches. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Ingo Molnar <mingo@elte.hu> Cc: Greg KH <greg@kroah.com> Cc: Randy Dunlap <randy.dunlap@oracle.com> Cc: Kay Sievers <kay.sievers@vrfy.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-30 15:55:01 +08:00
#include <linux/debugobjects.h>
#include <linux/memory.h>
#include <linux/prefetch.h>
#include <linux/sched/task_stack.h>
#include <net/sock.h>
#include <asm/cacheflush.h>
#include <asm/tlbflush.h>
#include <asm/page.h>
#include <trace/events/kmem.h>
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
#include "internal.h"
#include "slab.h"
/*
* DEBUG - 1 for kmem_cache_create() to honour; SLAB_RED_ZONE & SLAB_POISON.
* 0 for faster, smaller code (especially in the critical paths).
*
* STATS - 1 to collect stats for /proc/slabinfo.
* 0 for faster, smaller code (especially in the critical paths).
*
* FORCED_DEBUG - 1 enables SLAB_RED_ZONE and SLAB_POISON (if possible)
*/
#ifdef CONFIG_DEBUG_SLAB
#define DEBUG 1
#define STATS 1
#define FORCED_DEBUG 1
#else
#define DEBUG 0
#define STATS 0
#define FORCED_DEBUG 0
#endif
/* Shouldn't this be in a header file somewhere? */
#define BYTES_PER_WORD sizeof(void *)
#define REDZONE_ALIGN max(BYTES_PER_WORD, __alignof__(unsigned long long))
#ifndef ARCH_KMALLOC_FLAGS
#define ARCH_KMALLOC_FLAGS SLAB_HWCACHE_ALIGN
#endif
#define FREELIST_BYTE_INDEX (((PAGE_SIZE >> BITS_PER_BYTE) \
<= SLAB_OBJ_MIN_SIZE) ? 1 : 0)
#if FREELIST_BYTE_INDEX
typedef unsigned char freelist_idx_t;
#else
typedef unsigned short freelist_idx_t;
#endif
#define SLAB_OBJ_MAX_NUM ((1 << sizeof(freelist_idx_t) * BITS_PER_BYTE) - 1)
/*
* struct array_cache
*
* Purpose:
* - LIFO ordering, to hand out cache-warm objects from _alloc
* - reduce the number of linked list operations
* - reduce spinlock operations
*
* The limit is stored in the per-cpu structure to reduce the data cache
* footprint.
*
*/
struct array_cache {
unsigned int avail;
unsigned int limit;
unsigned int batchcount;
unsigned int touched;
void *entry[]; /*
* Must have this definition in here for the proper
* alignment of array_cache. Also simplifies accessing
* the entries.
*/
};
struct alien_cache {
spinlock_t lock;
struct array_cache ac;
};
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
/*
* Need this for bootstrapping a per node allocator.
*/
#define NUM_INIT_LISTS (2 * MAX_NUMNODES)
static struct kmem_cache_node __initdata init_kmem_cache_node[NUM_INIT_LISTS];
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#define CACHE_CACHE 0
#define SIZE_NODE (MAX_NUMNODES)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
static int drain_freelist(struct kmem_cache *cache,
struct kmem_cache_node *n, int tofree);
static void free_block(struct kmem_cache *cachep, void **objpp, int len,
int node, struct list_head *list);
static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list);
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp);
2006-11-22 22:55:48 +08:00
static void cache_reap(struct work_struct *unused);
static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
void **list);
static inline void fixup_slab_list(struct kmem_cache *cachep,
struct kmem_cache_node *n, struct page *page,
void **list);
static int slab_early_init = 1;
#define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node))
static void kmem_cache_node_init(struct kmem_cache_node *parent)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
INIT_LIST_HEAD(&parent->slabs_full);
INIT_LIST_HEAD(&parent->slabs_partial);
INIT_LIST_HEAD(&parent->slabs_free);
parent->total_slabs = 0;
parent->free_slabs = 0;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
parent->shared = NULL;
parent->alien = NULL;
parent->colour_next = 0;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
spin_lock_init(&parent->list_lock);
parent->free_objects = 0;
parent->free_touched = 0;
}
#define MAKE_LIST(cachep, listp, slab, nodeid) \
do { \
INIT_LIST_HEAD(listp); \
list_splice(&get_node(cachep, nodeid)->slab, listp); \
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
} while (0)
#define MAKE_ALL_LISTS(cachep, ptr, nodeid) \
do { \
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
MAKE_LIST((cachep), (&(ptr)->slabs_full), slabs_full, nodeid); \
MAKE_LIST((cachep), (&(ptr)->slabs_partial), slabs_partial, nodeid); \
MAKE_LIST((cachep), (&(ptr)->slabs_free), slabs_free, nodeid); \
} while (0)
#define CFLGS_OBJFREELIST_SLAB ((slab_flags_t __force)0x40000000U)
#define CFLGS_OFF_SLAB ((slab_flags_t __force)0x80000000U)
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
#define OBJFREELIST_SLAB(x) ((x)->flags & CFLGS_OBJFREELIST_SLAB)
#define OFF_SLAB(x) ((x)->flags & CFLGS_OFF_SLAB)
#define BATCHREFILL_LIMIT 16
/*
* Optimization question: fewer reaps means less probability for unnessary
* cpucache drain/refill cycles.
*
2005-11-08 23:44:08 +08:00
* OTOH the cpuarrays can contain lots of objects,
* which could lock up otherwise freeable slabs.
*/
#define REAPTIMEOUT_AC (2*HZ)
#define REAPTIMEOUT_NODE (4*HZ)
#if STATS
#define STATS_INC_ACTIVE(x) ((x)->num_active++)
#define STATS_DEC_ACTIVE(x) ((x)->num_active--)
#define STATS_INC_ALLOCED(x) ((x)->num_allocations++)
#define STATS_INC_GROWN(x) ((x)->grown++)
#define STATS_ADD_REAPED(x,y) ((x)->reaped += (y))
#define STATS_SET_HIGH(x) \
do { \
if ((x)->num_active > (x)->high_mark) \
(x)->high_mark = (x)->num_active; \
} while (0)
#define STATS_INC_ERR(x) ((x)->errors++)
#define STATS_INC_NODEALLOCS(x) ((x)->node_allocs++)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#define STATS_INC_NODEFREES(x) ((x)->node_frees++)
#define STATS_INC_ACOVERFLOW(x) ((x)->node_overflow++)
#define STATS_SET_FREEABLE(x, i) \
do { \
if ((x)->max_freeable < i) \
(x)->max_freeable = i; \
} while (0)
#define STATS_INC_ALLOCHIT(x) atomic_inc(&(x)->allochit)
#define STATS_INC_ALLOCMISS(x) atomic_inc(&(x)->allocmiss)
#define STATS_INC_FREEHIT(x) atomic_inc(&(x)->freehit)
#define STATS_INC_FREEMISS(x) atomic_inc(&(x)->freemiss)
#else
#define STATS_INC_ACTIVE(x) do { } while (0)
#define STATS_DEC_ACTIVE(x) do { } while (0)
#define STATS_INC_ALLOCED(x) do { } while (0)
#define STATS_INC_GROWN(x) do { } while (0)
#define STATS_ADD_REAPED(x,y) do { (void)(y); } while (0)
#define STATS_SET_HIGH(x) do { } while (0)
#define STATS_INC_ERR(x) do { } while (0)
#define STATS_INC_NODEALLOCS(x) do { } while (0)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#define STATS_INC_NODEFREES(x) do { } while (0)
#define STATS_INC_ACOVERFLOW(x) do { } while (0)
#define STATS_SET_FREEABLE(x, i) do { } while (0)
#define STATS_INC_ALLOCHIT(x) do { } while (0)
#define STATS_INC_ALLOCMISS(x) do { } while (0)
#define STATS_INC_FREEHIT(x) do { } while (0)
#define STATS_INC_FREEMISS(x) do { } while (0)
#endif
#if DEBUG
/*
* memory layout of objects:
* 0 : objp
* 0 .. cachep->obj_offset - BYTES_PER_WORD - 1: padding. This ensures that
* the end of an object is aligned with the end of the real
* allocation. Catches writes behind the end of the allocation.
* cachep->obj_offset - BYTES_PER_WORD .. cachep->obj_offset - 1:
* redzone word.
* cachep->obj_offset: The real object.
* cachep->size - 2* BYTES_PER_WORD: redzone word [BYTES_PER_WORD long]
* cachep->size - 1* BYTES_PER_WORD: last caller address
* [BYTES_PER_WORD long]
*/
static int obj_offset(struct kmem_cache *cachep)
{
return cachep->obj_offset;
}
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
static unsigned long long *dbg_redzone1(struct kmem_cache *cachep, void *objp)
{
BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
return (unsigned long long*) (objp + obj_offset(cachep) -
sizeof(unsigned long long));
}
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
static unsigned long long *dbg_redzone2(struct kmem_cache *cachep, void *objp)
{
BUG_ON(!(cachep->flags & SLAB_RED_ZONE));
if (cachep->flags & SLAB_STORE_USER)
return (unsigned long long *)(objp + cachep->size -
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
sizeof(unsigned long long) -
REDZONE_ALIGN);
return (unsigned long long *) (objp + cachep->size -
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
sizeof(unsigned long long));
}
static void **dbg_userword(struct kmem_cache *cachep, void *objp)
{
BUG_ON(!(cachep->flags & SLAB_STORE_USER));
return (void **)(objp + cachep->size - BYTES_PER_WORD);
}
#else
#define obj_offset(x) 0
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
#define dbg_redzone1(cachep, objp) ({BUG(); (unsigned long long *)NULL;})
#define dbg_redzone2(cachep, objp) ({BUG(); (unsigned long long *)NULL;})
#define dbg_userword(cachep, objp) ({BUG(); (void **)NULL;})
#endif
/*
* Do not go above this order unless 0 objects fit into the slab or
* overridden on the command line.
*/
#define SLAB_MAX_ORDER_HI 1
#define SLAB_MAX_ORDER_LO 0
static int slab_max_order = SLAB_MAX_ORDER_LO;
static bool slab_max_order_set __initdata;
static inline void *index_to_obj(struct kmem_cache *cache, struct page *page,
unsigned int idx)
{
return page->s_mem + cache->size * idx;
}
#define BOOT_CPUCACHE_ENTRIES 1
/* internal cache of cache description objs */
static struct kmem_cache kmem_cache_boot = {
.batchcount = 1,
.limit = BOOT_CPUCACHE_ENTRIES,
.shared = 1,
.size = sizeof(struct kmem_cache),
.name = "kmem_cache",
};
static DEFINE_PER_CPU(struct delayed_work, slab_reap_work);
static inline struct array_cache *cpu_cache_get(struct kmem_cache *cachep)
{
return this_cpu_ptr(cachep->cpu_cache);
}
/*
* Calculate the number of objects and left-over bytes for a given buffer size.
*/
static unsigned int cache_estimate(unsigned long gfporder, size_t buffer_size,
slab_flags_t flags, size_t *left_over)
{
unsigned int num;
size_t slab_size = PAGE_SIZE << gfporder;
/*
* The slab management structure can be either off the slab or
* on it. For the latter case, the memory allocated for a
* slab is used for:
*
* - @buffer_size bytes for each object
mm/slab: put the freelist at the end of slab page Currently, the freelist is at the front of slab page. This requires extra space to meet object alignment requirement. If we put the freelist at the end of a slab page, objects could start at page boundary and will be at correct alignment. This is possible because freelist has no alignment constraint itself. This gives us two benefits: It removes extra memory space for the freelist alignment and remove complex calculation at cache initialization step. I can't think notable drawback here. I mentioned that this would reduce extra memory space, but, this benefit is rather theoretical because it can be applied to very few cases. Following is the example cache type that can get benefit from this change. size align num before after 32 8 124 4100 4092 64 8 63 4103 4095 88 8 46 4102 4094 272 8 15 4103 4095 408 8 10 4098 4090 32 16 124 4108 4092 64 16 63 4111 4095 32 32 124 4124 4092 64 32 63 4127 4095 96 32 42 4106 4074 before means whole size for objects and aligned freelist before applying patch and after shows the result of this patch. Since before is more than 4096, number of object should decrease and memory waste happens. Anyway, this patch removes complex calculation so looks beneficial to me. [akpm@linux-foundation.org: fix kerneldoc] Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:30 +08:00
* - One freelist_idx_t for each object
*
* We don't need to consider alignment of freelist because
* freelist will be at the end of slab page. The objects will be
* at the correct alignment.
*
* If the slab management structure is off the slab, then the
* alignment will already be calculated into the size. Because
* the slabs are all pages aligned, the objects will be at the
* correct alignment when allocated.
*/
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (flags & (CFLGS_OBJFREELIST_SLAB | CFLGS_OFF_SLAB)) {
num = slab_size / buffer_size;
mm/slab: put the freelist at the end of slab page Currently, the freelist is at the front of slab page. This requires extra space to meet object alignment requirement. If we put the freelist at the end of a slab page, objects could start at page boundary and will be at correct alignment. This is possible because freelist has no alignment constraint itself. This gives us two benefits: It removes extra memory space for the freelist alignment and remove complex calculation at cache initialization step. I can't think notable drawback here. I mentioned that this would reduce extra memory space, but, this benefit is rather theoretical because it can be applied to very few cases. Following is the example cache type that can get benefit from this change. size align num before after 32 8 124 4100 4092 64 8 63 4103 4095 88 8 46 4102 4094 272 8 15 4103 4095 408 8 10 4098 4090 32 16 124 4108 4092 64 16 63 4111 4095 32 32 124 4124 4092 64 32 63 4127 4095 96 32 42 4106 4074 before means whole size for objects and aligned freelist before applying patch and after shows the result of this patch. Since before is more than 4096, number of object should decrease and memory waste happens. Anyway, this patch removes complex calculation so looks beneficial to me. [akpm@linux-foundation.org: fix kerneldoc] Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:30 +08:00
*left_over = slab_size % buffer_size;
} else {
num = slab_size / (buffer_size + sizeof(freelist_idx_t));
mm/slab: put the freelist at the end of slab page Currently, the freelist is at the front of slab page. This requires extra space to meet object alignment requirement. If we put the freelist at the end of a slab page, objects could start at page boundary and will be at correct alignment. This is possible because freelist has no alignment constraint itself. This gives us two benefits: It removes extra memory space for the freelist alignment and remove complex calculation at cache initialization step. I can't think notable drawback here. I mentioned that this would reduce extra memory space, but, this benefit is rather theoretical because it can be applied to very few cases. Following is the example cache type that can get benefit from this change. size align num before after 32 8 124 4100 4092 64 8 63 4103 4095 88 8 46 4102 4094 272 8 15 4103 4095 408 8 10 4098 4090 32 16 124 4108 4092 64 16 63 4111 4095 32 32 124 4124 4092 64 32 63 4127 4095 96 32 42 4106 4074 before means whole size for objects and aligned freelist before applying patch and after shows the result of this patch. Since before is more than 4096, number of object should decrease and memory waste happens. Anyway, this patch removes complex calculation so looks beneficial to me. [akpm@linux-foundation.org: fix kerneldoc] Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:30 +08:00
*left_over = slab_size %
(buffer_size + sizeof(freelist_idx_t));
}
return num;
}
#if DEBUG
#define slab_error(cachep, msg) __slab_error(__func__, cachep, msg)
static void __slab_error(const char *function, struct kmem_cache *cachep,
char *msg)
{
pr_err("slab error in %s(): cache `%s': %s\n",
function, cachep->name, msg);
dump_stack();
add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
}
#endif
/*
* By default on NUMA we use alien caches to stage the freeing of
* objects allocated from other nodes. This causes massive memory
* inefficiencies when using fake NUMA setup to split memory into a
* large number of small nodes, so it can be disabled on the command
* line
*/
static int use_alien_caches __read_mostly = 1;
static int __init noaliencache_setup(char *s)
{
use_alien_caches = 0;
return 1;
}
__setup("noaliencache", noaliencache_setup);
static int __init slab_max_order_setup(char *str)
{
get_option(&str, &slab_max_order);
slab_max_order = slab_max_order < 0 ? 0 :
min(slab_max_order, MAX_ORDER - 1);
slab_max_order_set = true;
return 1;
}
__setup("slab_max_order=", slab_max_order_setup);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
#ifdef CONFIG_NUMA
/*
* Special reaping functions for NUMA systems called from cache_reap().
* These take care of doing round robin flushing of alien caches (containing
* objects freed on different nodes from which they were allocated) and the
* flushing of remote pcps by calling drain_node_pages.
*/
static DEFINE_PER_CPU(unsigned long, slab_reap_node);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
static void init_reap_node(int cpu)
{
per_cpu(slab_reap_node, cpu) = next_node_in(cpu_to_mem(cpu),
node_online_map);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
}
static void next_reap_node(void)
{
int node = __this_cpu_read(slab_reap_node);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
node = next_node_in(node, node_online_map);
__this_cpu_write(slab_reap_node, node);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
}
#else
#define init_reap_node(cpu) do { } while (0)
#define next_reap_node(void) do { } while (0)
#endif
/*
* Initiate the reap timer running on the target CPU. We run at around 1 to 2Hz
* via the workqueue/eventd.
* Add the CPU number into the expiration time to minimize the possibility of
* the CPUs getting into lockstep and contending for the global cache chain
* lock.
*/
static void start_cpu_timer(int cpu)
{
struct delayed_work *reap_work = &per_cpu(slab_reap_work, cpu);
if (reap_work->work.func == NULL) {
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
init_reap_node(cpu);
INIT_DEFERRABLE_WORK(reap_work, cache_reap);
schedule_delayed_work_on(cpu, reap_work,
__round_jiffies_relative(HZ, cpu));
}
}
static void init_arraycache(struct array_cache *ac, int limit, int batch)
{
if (ac) {
ac->avail = 0;
ac->limit = limit;
ac->batchcount = batch;
ac->touched = 0;
}
}
static struct array_cache *alloc_arraycache(int node, int entries,
int batchcount, gfp_t gfp)
{
size_t memsize = sizeof(void *) * entries + sizeof(struct array_cache);
struct array_cache *ac = NULL;
ac = kmalloc_node(memsize, gfp, node);
mm/slab.c: kmemleak no scan alien caches Kmemleak throws endless warnings during boot due to in __alloc_alien_cache(), alc = kmalloc_node(memsize, gfp, node); init_arraycache(&alc->ac, entries, batch); kmemleak_no_scan(ac); Kmemleak does not track the array cache (alc->ac) but the alien cache (alc) instead, so let it track the latter by lifting kmemleak_no_scan() out of init_arraycache(). There is another place that calls init_arraycache(), but alloc_kmem_cache_cpus() uses the percpu allocation where will never be considered as a leak. kmemleak: Found object by alias at 0xffff8007b9aa7e38 CPU: 190 PID: 1 Comm: swapper/0 Not tainted 5.0.0-rc2+ #2 Call trace: dump_backtrace+0x0/0x168 show_stack+0x24/0x30 dump_stack+0x88/0xb0 lookup_object+0x84/0xac find_and_get_object+0x84/0xe4 kmemleak_no_scan+0x74/0xf4 setup_kmem_cache_node+0x2b4/0x35c __do_tune_cpucache+0x250/0x2d4 do_tune_cpucache+0x4c/0xe4 enable_cpucache+0xc8/0x110 setup_cpu_cache+0x40/0x1b8 __kmem_cache_create+0x240/0x358 create_cache+0xc0/0x198 kmem_cache_create_usercopy+0x158/0x20c kmem_cache_create+0x50/0x64 fsnotify_init+0x58/0x6c do_one_initcall+0x194/0x388 kernel_init_freeable+0x668/0x688 kernel_init+0x18/0x124 ret_from_fork+0x10/0x18 kmemleak: Object 0xffff8007b9aa7e00 (size 256): kmemleak: comm "swapper/0", pid 1, jiffies 4294697137 kmemleak: min_count = 1 kmemleak: count = 0 kmemleak: flags = 0x1 kmemleak: checksum = 0 kmemleak: backtrace: kmemleak_alloc+0x84/0xb8 kmem_cache_alloc_node_trace+0x31c/0x3a0 __kmalloc_node+0x58/0x78 setup_kmem_cache_node+0x26c/0x35c __do_tune_cpucache+0x250/0x2d4 do_tune_cpucache+0x4c/0xe4 enable_cpucache+0xc8/0x110 setup_cpu_cache+0x40/0x1b8 __kmem_cache_create+0x240/0x358 create_cache+0xc0/0x198 kmem_cache_create_usercopy+0x158/0x20c kmem_cache_create+0x50/0x64 fsnotify_init+0x58/0x6c do_one_initcall+0x194/0x388 kernel_init_freeable+0x668/0x688 kernel_init+0x18/0x124 kmemleak: Not scanning unknown object at 0xffff8007b9aa7e38 CPU: 190 PID: 1 Comm: swapper/0 Not tainted 5.0.0-rc2+ #2 Call trace: dump_backtrace+0x0/0x168 show_stack+0x24/0x30 dump_stack+0x88/0xb0 kmemleak_no_scan+0x90/0xf4 setup_kmem_cache_node+0x2b4/0x35c __do_tune_cpucache+0x250/0x2d4 do_tune_cpucache+0x4c/0xe4 enable_cpucache+0xc8/0x110 setup_cpu_cache+0x40/0x1b8 __kmem_cache_create+0x240/0x358 create_cache+0xc0/0x198 kmem_cache_create_usercopy+0x158/0x20c kmem_cache_create+0x50/0x64 fsnotify_init+0x58/0x6c do_one_initcall+0x194/0x388 kernel_init_freeable+0x668/0x688 kernel_init+0x18/0x124 ret_from_fork+0x10/0x18 Link: http://lkml.kernel.org/r/20190129184518.39808-1-cai@lca.pw Fixes: 1fe00d50a9e8 ("slab: factor out initialization of array cache") Signed-off-by: Qian Cai <cai@lca.pw> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-03-06 07:42:03 +08:00
/*
* The array_cache structures contain pointers to free object.
* However, when such objects are allocated or transferred to another
* cache the pointers are not cleared and they could be counted as
* valid references during a kmemleak scan. Therefore, kmemleak must
* not scan such objects.
*/
kmemleak_no_scan(ac);
init_arraycache(ac, entries, batchcount);
return ac;
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
static noinline void cache_free_pfmemalloc(struct kmem_cache *cachep,
struct page *page, void *objp)
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
{
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
struct kmem_cache_node *n;
int page_node;
LIST_HEAD(list);
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
page_node = page_to_nid(page);
n = get_node(cachep, page_node);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
spin_lock(&n->list_lock);
free_block(cachep, &objp, 1, page_node, &list);
spin_unlock(&n->list_lock);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
slabs_destroy(cachep, &list);
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
}
/*
* Transfer objects in one arraycache to another.
* Locking must be handled by the caller.
*
* Return the number of entries transferred.
*/
static int transfer_objects(struct array_cache *to,
struct array_cache *from, unsigned int max)
{
/* Figure out how many entries to transfer */
int nr = min3(from->avail, max, to->limit - to->avail);
if (!nr)
return 0;
memcpy(to->entry + to->avail, from->entry + from->avail -nr,
sizeof(void *) *nr);
from->avail -= nr;
to->avail += nr;
return nr;
}
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
#ifndef CONFIG_NUMA
#define drain_alien_cache(cachep, alien) do { } while (0)
#define reap_alien(cachep, n) do { } while (0)
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
static inline struct alien_cache **alloc_alien_cache(int node,
int limit, gfp_t gfp)
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
{
return NULL;
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
}
static inline void free_alien_cache(struct alien_cache **ac_ptr)
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
{
}
static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
return 0;
}
static inline void *alternate_node_alloc(struct kmem_cache *cachep,
gfp_t flags)
{
return NULL;
}
static inline void *____cache_alloc_node(struct kmem_cache *cachep,
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
gfp_t flags, int nodeid)
{
return NULL;
}
mm: remove GFP_THISNODE NOTE: this is not about __GFP_THISNODE, this is only about GFP_THISNODE. GFP_THISNODE is a secret combination of gfp bits that have different behavior than expected. It is a combination of __GFP_THISNODE, __GFP_NORETRY, and __GFP_NOWARN and is special-cased in the page allocator slowpath to fail without trying reclaim even though it may be used in combination with __GFP_WAIT. An example of the problem this creates: commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify") fixed up many users of GFP_THISNODE that really just wanted __GFP_THISNODE. The problem doesn't end there, however, because even it was a no-op for alloc_misplaced_dst_page(), which also sets __GFP_NORETRY and __GFP_NOWARN, and migrate_misplaced_transhuge_page(), where __GFP_NORETRY and __GFP_NOWAIT is set in GFP_TRANSHUGE. Converting GFP_THISNODE to __GFP_THISNODE is a no-op in these cases since the page allocator special-cases __GFP_THISNODE && __GFP_NORETRY && __GFP_NOWARN. It's time to just remove GFP_THISNODE entirely. We leave __GFP_THISNODE to restrict an allocation to a local node, but remove GFP_THISNODE and its obscurity. Instead, we require that a caller clear __GFP_WAIT if it wants to avoid reclaim. This allows the aforementioned functions to actually reclaim as they should. It also enables any future callers that want to do __GFP_THISNODE but also __GFP_NORETRY && __GFP_NOWARN to reclaim. The rule is simple: if you don't want to reclaim, then don't set __GFP_WAIT. Aside: ovs_flow_stats_update() really wants to avoid reclaim as well, so it is unchanged. Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Christoph Lameter <cl@linux.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Jarno Rajahalme <jrajahalme@nicira.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Greg Thelen <gthelen@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-15 06:46:55 +08:00
static inline gfp_t gfp_exact_node(gfp_t flags)
{
mm: thp: set THP defrag by default to madvise and add a stall-free defrag option THP defrag is enabled by default to direct reclaim/compact but not wake kswapd in the event of a THP allocation failure. The problem is that THP allocation requests potentially enter reclaim/compaction. This potentially incurs a severe stall that is not guaranteed to be offset by reduced TLB misses. While there has been considerable effort to reduce the impact of reclaim/compaction, it is still a high cost and workloads that should fit in memory fail to do so. Specifically, a simple anon/file streaming workload will enter direct reclaim on NUMA at least even though the working set size is 80% of RAM. It's been years and it's time to throw in the towel. First, this patch defines THP defrag as follows; madvise: A failed allocation will direct reclaim/compact if the application requests it never: Neither reclaim/compact nor wake kswapd defer: A failed allocation will wake kswapd/kcompactd always: A failed allocation will direct reclaim/compact (historical behaviour) khugepaged defrag will enter direct/reclaim but not wake kswapd. Next it sets the default defrag option to be "madvise" to only enter direct reclaim/compaction for applications that specifically requested it. Lastly, it removes a check from the page allocator slowpath that is related to __GFP_THISNODE to allow "defer" to work. The callers that really cares are slub/slab and they are updated accordingly. The slab one may be surprising because it also corrects a comment as kswapd was never woken up by that path. This means that a THP fault will no longer stall for most applications by default and the ideal for most users that get THP if they are immediately available. There are still options for users that prefer a stall at startup of a new application by either restoring historical behaviour with "always" or pick a half-way point with "defer" where kswapd does some of the work in the background and wakes kcompactd if necessary. THP defrag for khugepaged remains enabled and will enter direct/reclaim but no wakeup kswapd or kcompactd. After this patch a THP allocation failure will quickly fallback and rely on khugepaged to recover the situation at some time in the future. In some cases, this will reduce THP usage but the benefit of THP is hard to measure and not a universal win where as a stall to reclaim/compaction is definitely measurable and can be painful. The first test for this is using "usemem" to read a large file and write a large anonymous mapping (to avoid the zero page) multiple times. The total size of the mappings is 80% of RAM and the benchmark simply measures how long it takes to complete. It uses multiple threads to see if that is a factor. On UMA, the performance is almost identical so is not reported but on NUMA, we see this usemem 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Amean System-1 102.86 ( 0.00%) 46.81 ( 54.50%) Amean System-4 37.85 ( 0.00%) 34.02 ( 10.12%) Amean System-7 48.12 ( 0.00%) 46.89 ( 2.56%) Amean System-12 51.98 ( 0.00%) 56.96 ( -9.57%) Amean System-21 80.16 ( 0.00%) 79.05 ( 1.39%) Amean System-30 110.71 ( 0.00%) 107.17 ( 3.20%) Amean System-48 127.98 ( 0.00%) 124.83 ( 2.46%) Amean Elapsd-1 185.84 ( 0.00%) 105.51 ( 43.23%) Amean Elapsd-4 26.19 ( 0.00%) 25.58 ( 2.33%) Amean Elapsd-7 21.65 ( 0.00%) 21.62 ( 0.16%) Amean Elapsd-12 18.58 ( 0.00%) 17.94 ( 3.43%) Amean Elapsd-21 17.53 ( 0.00%) 16.60 ( 5.33%) Amean Elapsd-30 17.45 ( 0.00%) 17.13 ( 1.84%) Amean Elapsd-48 15.40 ( 0.00%) 15.27 ( 0.82%) For a single thread, the benchmark completes 43.23% faster with this patch applied with smaller benefits as the thread increases. Similar, notice the large reduction in most cases in system CPU usage. The overall CPU time is 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 User 10357.65 10438.33 System 3988.88 3543.94 Elapsed 2203.01 1634.41 Which is substantial. Now, the reclaim figures 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 128458477 278352931 Major Faults 2174976 225 Swap Ins 16904701 0 Swap Outs 17359627 0 Allocation stalls 43611 0 DMA allocs 0 0 DMA32 allocs 19832646 19448017 Normal allocs 614488453 580941839 Movable allocs 0 0 Direct pages scanned 24163800 0 Kswapd pages scanned 0 0 Kswapd pages reclaimed 0 0 Direct pages reclaimed 20691346 0 Compaction stalls 42263 0 Compaction success 938 0 Compaction failures 41325 0 This patch eliminates almost all swapping and direct reclaim activity. There is still overhead but it's from NUMA balancing which does not identify that it's pointless trying to do anything with this workload. I also tried the thpscale benchmark which forces a corner case where compaction can be used heavily and measures the latency of whether base or huge pages were used thpscale Fault Latencies 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Amean fault-base-1 5288.84 ( 0.00%) 2817.12 ( 46.73%) Amean fault-base-3 6365.53 ( 0.00%) 3499.11 ( 45.03%) Amean fault-base-5 6526.19 ( 0.00%) 4363.06 ( 33.15%) Amean fault-base-7 7142.25 ( 0.00%) 4858.08 ( 31.98%) Amean fault-base-12 13827.64 ( 0.00%) 10292.11 ( 25.57%) Amean fault-base-18 18235.07 ( 0.00%) 13788.84 ( 24.38%) Amean fault-base-24 21597.80 ( 0.00%) 24388.03 (-12.92%) Amean fault-base-30 26754.15 ( 0.00%) 19700.55 ( 26.36%) Amean fault-base-32 26784.94 ( 0.00%) 19513.57 ( 27.15%) Amean fault-huge-1 4223.96 ( 0.00%) 2178.57 ( 48.42%) Amean fault-huge-3 2194.77 ( 0.00%) 2149.74 ( 2.05%) Amean fault-huge-5 2569.60 ( 0.00%) 2346.95 ( 8.66%) Amean fault-huge-7 3612.69 ( 0.00%) 2997.70 ( 17.02%) Amean fault-huge-12 3301.75 ( 0.00%) 6727.02 (-103.74%) Amean fault-huge-18 6696.47 ( 0.00%) 6685.72 ( 0.16%) Amean fault-huge-24 8000.72 ( 0.00%) 9311.43 (-16.38%) Amean fault-huge-30 13305.55 ( 0.00%) 9750.45 ( 26.72%) Amean fault-huge-32 9981.71 ( 0.00%) 10316.06 ( -3.35%) The average time to fault pages is substantially reduced in the majority of caseds but with the obvious caveat that fewer THPs are actually used in this adverse workload 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Percentage huge-1 0.71 ( 0.00%) 14.04 (1865.22%) Percentage huge-3 10.77 ( 0.00%) 33.05 (206.85%) Percentage huge-5 60.39 ( 0.00%) 38.51 (-36.23%) Percentage huge-7 45.97 ( 0.00%) 34.57 (-24.79%) Percentage huge-12 68.12 ( 0.00%) 40.07 (-41.17%) Percentage huge-18 64.93 ( 0.00%) 47.82 (-26.35%) Percentage huge-24 62.69 ( 0.00%) 44.23 (-29.44%) Percentage huge-30 43.49 ( 0.00%) 55.38 ( 27.34%) Percentage huge-32 50.72 ( 0.00%) 51.90 ( 2.35%) 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 37429143 47564000 Major Faults 1916 1558 Swap Ins 1466 1079 Swap Outs 2936863 149626 Allocation stalls 62510 3 DMA allocs 0 0 DMA32 allocs 6566458 6401314 Normal allocs 216361697 216538171 Movable allocs 0 0 Direct pages scanned 25977580 17998 Kswapd pages scanned 0 3638931 Kswapd pages reclaimed 0 207236 Direct pages reclaimed 8833714 88 Compaction stalls 103349 5 Compaction success 270 4 Compaction failures 103079 1 Note again that while this does swap as it's an aggressive workload, the direct relcim activity and allocation stalls is substantially reduced. There is some kswapd activity but ftrace showed that the kswapd activity was due to normal wakeups from 4K pages being allocated. Compaction-related stalls and activity are almost eliminated. I also tried the stutter benchmark. For this, I do not have figures for NUMA but it's something that does impact UMA so I'll report what is available stutter 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Min mmap 7.3571 ( 0.00%) 7.3438 ( 0.18%) 1st-qrtle mmap 7.5278 ( 0.00%) 17.9200 (-138.05%) 2nd-qrtle mmap 7.6818 ( 0.00%) 21.6055 (-181.25%) 3rd-qrtle mmap 11.0889 ( 0.00%) 21.8881 (-97.39%) Max-90% mmap 27.8978 ( 0.00%) 22.1632 ( 20.56%) Max-93% mmap 28.3202 ( 0.00%) 22.3044 ( 21.24%) Max-95% mmap 28.5600 ( 0.00%) 22.4580 ( 21.37%) Max-99% mmap 29.6032 ( 0.00%) 25.5216 ( 13.79%) Max mmap 4109.7289 ( 0.00%) 4813.9832 (-17.14%) Mean mmap 12.4474 ( 0.00%) 19.3027 (-55.07%) This benchmark is trying to fault an anonymous mapping while there is a heavy IO load -- a scenario that desktop users used to complain about frequently. This shows a mix because the ideal case of mapping with THP is not hit as often. However, note that 99% of the mappings complete 13.79% faster. The CPU usage here is particularly interesting 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 User 67.50 0.99 System 1327.88 91.30 Elapsed 2079.00 2128.98 And once again we look at the reclaim figures 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 335241922 1314582827 Major Faults 715 819 Swap Ins 0 0 Swap Outs 0 0 Allocation stalls 532723 0 DMA allocs 0 0 DMA32 allocs 1822364341 1177950222 Normal allocs 1815640808 1517844854 Movable allocs 0 0 Direct pages scanned 21892772 0 Kswapd pages scanned 20015890 41879484 Kswapd pages reclaimed 19961986 41822072 Direct pages reclaimed 21892741 0 Compaction stalls 1065755 0 Compaction success 514 0 Compaction failures 1065241 0 Allocation stalls and all direct reclaim activity is eliminated as well as compaction-related stalls. THP gives impressive gains in some cases but only if they are quickly available. We're not going to reach the point where they are completely free so lets take the costs out of the fast paths finally and defer the cost to kswapd, kcompactd and khugepaged where it belongs. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-18 05:19:23 +08:00
return flags & ~__GFP_NOFAIL;
mm: remove GFP_THISNODE NOTE: this is not about __GFP_THISNODE, this is only about GFP_THISNODE. GFP_THISNODE is a secret combination of gfp bits that have different behavior than expected. It is a combination of __GFP_THISNODE, __GFP_NORETRY, and __GFP_NOWARN and is special-cased in the page allocator slowpath to fail without trying reclaim even though it may be used in combination with __GFP_WAIT. An example of the problem this creates: commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify") fixed up many users of GFP_THISNODE that really just wanted __GFP_THISNODE. The problem doesn't end there, however, because even it was a no-op for alloc_misplaced_dst_page(), which also sets __GFP_NORETRY and __GFP_NOWARN, and migrate_misplaced_transhuge_page(), where __GFP_NORETRY and __GFP_NOWAIT is set in GFP_TRANSHUGE. Converting GFP_THISNODE to __GFP_THISNODE is a no-op in these cases since the page allocator special-cases __GFP_THISNODE && __GFP_NORETRY && __GFP_NOWARN. It's time to just remove GFP_THISNODE entirely. We leave __GFP_THISNODE to restrict an allocation to a local node, but remove GFP_THISNODE and its obscurity. Instead, we require that a caller clear __GFP_WAIT if it wants to avoid reclaim. This allows the aforementioned functions to actually reclaim as they should. It also enables any future callers that want to do __GFP_THISNODE but also __GFP_NORETRY && __GFP_NOWARN to reclaim. The rule is simple: if you don't want to reclaim, then don't set __GFP_WAIT. Aside: ovs_flow_stats_update() really wants to avoid reclaim as well, so it is unchanged. Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Christoph Lameter <cl@linux.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Jarno Rajahalme <jrajahalme@nicira.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Greg Thelen <gthelen@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-15 06:46:55 +08:00
}
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
#else /* CONFIG_NUMA */
static void *____cache_alloc_node(struct kmem_cache *, gfp_t, int);
static void *alternate_node_alloc(struct kmem_cache *, gfp_t);
[PATCH] NUMA policies in the slab allocator V2 This patch fixes a regression in 2.6.14 against 2.6.13 that causes an imbalance in memory allocation during bootup. The slab allocator in 2.6.13 is not numa aware and simply calls alloc_pages(). This means that memory policies may control the behavior of alloc_pages(). During bootup the memory policy is set to MPOL_INTERLEAVE resulting in the spreading out of allocations during bootup over all available nodes. The slab allocator in 2.6.13 has only a single list of slab pages. As a result the per cpu slab cache and the spinlock controlled page lists may contain slab entries from off node memory. The slab allocator in 2.6.13 makes no effort to discern the locality of an entry on its lists. The NUMA aware slab allocator in 2.6.14 controls locality of the slab pages explicitly by calling alloc_pages_node(). The NUMA slab allocator manages slab entries by having lists of available slab pages for each node. The per cpu slab cache can only contain slab entries associated with the node local to the processor. This guarantees that the default allocation mode of the slab allocator always assigns local memory if available. Setting MPOL_INTERLEAVE as a default policy during bootup has no effect anymore. In 2.6.14 all node unspecific slab allocations are performed on the boot processor. This means that most of key data structures are allocated on one node. Most processors will have to refer to these structures making the boot node a potential bottleneck. This may reduce performance and cause unnecessary memory pressure on the boot node. This patch implements NUMA policies in the slab layer. There is the need of explicit application of NUMA memory policies by the slab allcator itself since the NUMA slab allocator does no longer let the page_allocator control locality. The check for policies is made directly at the beginning of __cache_alloc using current->mempolicy. The memory policy is already frequently checked by the page allocator (alloc_page_vma() and alloc_page_current()). So it is highly likely that the cacheline is present. For MPOL_INTERLEAVE kmalloc() will spread out each request to one node after another so that an equal distribution of allocations can be obtained during bootup. It is not possible to push the policy check to lower layers of the NUMA slab allocator since the per cpu caches are now only containing slab entries from the current node. If the policy says that the local node is not to be preferred or forbidden then there is no point in checking the slab cache or local list of slab pages. The allocation better be directed immediately to the lists containing slab entries for the allowed set of nodes. This way of applying policy also fixes another strange behavior in 2.6.13. alloc_pages() is controlled by the memory allocation policy of the current process. It could therefore be that one process is running with MPOL_INTERLEAVE and would f.e. obtain a new page following that policy since no slab entries are in the lists anymore. A page can typically be used for multiple slab entries but lets say that the current process is only using one. The other entries are then added to the slab lists. These are now non local entries in the slab lists despite of the possible availability of local pages that would provide faster access and increase the performance of the application. Another process without MPOL_INTERLEAVE may now run and expect a local slab entry from kmalloc(). However, there are still these free slab entries from the off node page obtained from the other process via MPOL_INTERLEAVE in the cache. The process will then get an off node slab entry although other slab entries may be available that are local to that process. This means that the policy if one process may contaminate the locality of the slab caches for other processes. This patch in effect insures that a per process policy is followed for the allocation of slab entries and that there cannot be a memory policy influence from one process to another. A process with default policy will always get a local slab entry if one is available. And the process using memory policies will get its memory arranged as requested. Off-node slab allocation will require the use of spinlocks and will make the use of per cpu caches not possible. A process using memory policies to redirect allocations offnode will have to cope with additional lock overhead in addition to the latency added by the need to access a remote slab entry. Changes V1->V2 - Remove #ifdef CONFIG_NUMA by moving forward declaration into prior #ifdef CONFIG_NUMA section. - Give the function determining the node number to use a saner name. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-01-19 09:42:36 +08:00
static struct alien_cache *__alloc_alien_cache(int node, int entries,
int batch, gfp_t gfp)
{
size_t memsize = sizeof(void *) * entries + sizeof(struct alien_cache);
struct alien_cache *alc = NULL;
alc = kmalloc_node(memsize, gfp, node);
if (alc) {
mm/slab.c: kmemleak no scan alien caches Kmemleak throws endless warnings during boot due to in __alloc_alien_cache(), alc = kmalloc_node(memsize, gfp, node); init_arraycache(&alc->ac, entries, batch); kmemleak_no_scan(ac); Kmemleak does not track the array cache (alc->ac) but the alien cache (alc) instead, so let it track the latter by lifting kmemleak_no_scan() out of init_arraycache(). There is another place that calls init_arraycache(), but alloc_kmem_cache_cpus() uses the percpu allocation where will never be considered as a leak. kmemleak: Found object by alias at 0xffff8007b9aa7e38 CPU: 190 PID: 1 Comm: swapper/0 Not tainted 5.0.0-rc2+ #2 Call trace: dump_backtrace+0x0/0x168 show_stack+0x24/0x30 dump_stack+0x88/0xb0 lookup_object+0x84/0xac find_and_get_object+0x84/0xe4 kmemleak_no_scan+0x74/0xf4 setup_kmem_cache_node+0x2b4/0x35c __do_tune_cpucache+0x250/0x2d4 do_tune_cpucache+0x4c/0xe4 enable_cpucache+0xc8/0x110 setup_cpu_cache+0x40/0x1b8 __kmem_cache_create+0x240/0x358 create_cache+0xc0/0x198 kmem_cache_create_usercopy+0x158/0x20c kmem_cache_create+0x50/0x64 fsnotify_init+0x58/0x6c do_one_initcall+0x194/0x388 kernel_init_freeable+0x668/0x688 kernel_init+0x18/0x124 ret_from_fork+0x10/0x18 kmemleak: Object 0xffff8007b9aa7e00 (size 256): kmemleak: comm "swapper/0", pid 1, jiffies 4294697137 kmemleak: min_count = 1 kmemleak: count = 0 kmemleak: flags = 0x1 kmemleak: checksum = 0 kmemleak: backtrace: kmemleak_alloc+0x84/0xb8 kmem_cache_alloc_node_trace+0x31c/0x3a0 __kmalloc_node+0x58/0x78 setup_kmem_cache_node+0x26c/0x35c __do_tune_cpucache+0x250/0x2d4 do_tune_cpucache+0x4c/0xe4 enable_cpucache+0xc8/0x110 setup_cpu_cache+0x40/0x1b8 __kmem_cache_create+0x240/0x358 create_cache+0xc0/0x198 kmem_cache_create_usercopy+0x158/0x20c kmem_cache_create+0x50/0x64 fsnotify_init+0x58/0x6c do_one_initcall+0x194/0x388 kernel_init_freeable+0x668/0x688 kernel_init+0x18/0x124 kmemleak: Not scanning unknown object at 0xffff8007b9aa7e38 CPU: 190 PID: 1 Comm: swapper/0 Not tainted 5.0.0-rc2+ #2 Call trace: dump_backtrace+0x0/0x168 show_stack+0x24/0x30 dump_stack+0x88/0xb0 kmemleak_no_scan+0x90/0xf4 setup_kmem_cache_node+0x2b4/0x35c __do_tune_cpucache+0x250/0x2d4 do_tune_cpucache+0x4c/0xe4 enable_cpucache+0xc8/0x110 setup_cpu_cache+0x40/0x1b8 __kmem_cache_create+0x240/0x358 create_cache+0xc0/0x198 kmem_cache_create_usercopy+0x158/0x20c kmem_cache_create+0x50/0x64 fsnotify_init+0x58/0x6c do_one_initcall+0x194/0x388 kernel_init_freeable+0x668/0x688 kernel_init+0x18/0x124 ret_from_fork+0x10/0x18 Link: http://lkml.kernel.org/r/20190129184518.39808-1-cai@lca.pw Fixes: 1fe00d50a9e8 ("slab: factor out initialization of array cache") Signed-off-by: Qian Cai <cai@lca.pw> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-03-06 07:42:03 +08:00
kmemleak_no_scan(alc);
init_arraycache(&alc->ac, entries, batch);
spin_lock_init(&alc->lock);
}
return alc;
}
static struct alien_cache **alloc_alien_cache(int node, int limit, gfp_t gfp)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
struct alien_cache **alc_ptr;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
int i;
if (limit > 1)
limit = 12;
alc_ptr = kcalloc_node(nr_node_ids, sizeof(void *), gfp, node);
if (!alc_ptr)
return NULL;
for_each_node(i) {
if (i == node || !node_online(i))
continue;
alc_ptr[i] = __alloc_alien_cache(node, limit, 0xbaadf00d, gfp);
if (!alc_ptr[i]) {
for (i--; i >= 0; i--)
kfree(alc_ptr[i]);
kfree(alc_ptr);
return NULL;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
}
return alc_ptr;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
static void free_alien_cache(struct alien_cache **alc_ptr)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
int i;
if (!alc_ptr)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
return;
for_each_node(i)
kfree(alc_ptr[i]);
kfree(alc_ptr);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
static void __drain_alien_cache(struct kmem_cache *cachep,
struct array_cache *ac, int node,
struct list_head *list)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
struct kmem_cache_node *n = get_node(cachep, node);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
if (ac->avail) {
spin_lock(&n->list_lock);
/*
* Stuff objects into the remote nodes shared array first.
* That way we could avoid the overhead of putting the objects
* into the free lists and getting them back later.
*/
if (n->shared)
transfer_objects(n->shared, ac, ac->limit);
free_block(cachep, ac->entry, ac->avail, node, list);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
ac->avail = 0;
spin_unlock(&n->list_lock);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
}
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
/*
* Called from cache_reap() to regularly drain alien caches round robin.
*/
static void reap_alien(struct kmem_cache *cachep, struct kmem_cache_node *n)
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
{
int node = __this_cpu_read(slab_reap_node);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
if (n->alien) {
struct alien_cache *alc = n->alien[node];
struct array_cache *ac;
if (alc) {
ac = &alc->ac;
if (ac->avail && spin_trylock_irq(&alc->lock)) {
LIST_HEAD(list);
__drain_alien_cache(cachep, ac, node, &list);
spin_unlock_irq(&alc->lock);
slabs_destroy(cachep, &list);
}
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
}
}
}
static void drain_alien_cache(struct kmem_cache *cachep,
struct alien_cache **alien)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
int i = 0;
struct alien_cache *alc;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
struct array_cache *ac;
unsigned long flags;
for_each_online_node(i) {
alc = alien[i];
if (alc) {
LIST_HEAD(list);
ac = &alc->ac;
spin_lock_irqsave(&alc->lock, flags);
__drain_alien_cache(cachep, ac, i, &list);
spin_unlock_irqrestore(&alc->lock, flags);
slabs_destroy(cachep, &list);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
}
}
static int __cache_free_alien(struct kmem_cache *cachep, void *objp,
int node, int page_node)
{
struct kmem_cache_node *n;
struct alien_cache *alien = NULL;
struct array_cache *ac;
LIST_HEAD(list);
n = get_node(cachep, node);
STATS_INC_NODEFREES(cachep);
if (n->alien && n->alien[page_node]) {
alien = n->alien[page_node];
ac = &alien->ac;
spin_lock(&alien->lock);
if (unlikely(ac->avail == ac->limit)) {
STATS_INC_ACOVERFLOW(cachep);
__drain_alien_cache(cachep, ac, page_node, &list);
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
ac->entry[ac->avail++] = objp;
spin_unlock(&alien->lock);
slabs_destroy(cachep, &list);
} else {
n = get_node(cachep, page_node);
spin_lock(&n->list_lock);
free_block(cachep, &objp, 1, page_node, &list);
spin_unlock(&n->list_lock);
slabs_destroy(cachep, &list);
}
return 1;
}
static inline int cache_free_alien(struct kmem_cache *cachep, void *objp)
{
int page_node = page_to_nid(virt_to_page(objp));
int node = numa_mem_id();
/*
* Make sure we are not freeing a object from another node to the array
* cache on this cpu.
*/
if (likely(node == page_node))
return 0;
return __cache_free_alien(cachep, objp, node, page_node);
}
mm: remove GFP_THISNODE NOTE: this is not about __GFP_THISNODE, this is only about GFP_THISNODE. GFP_THISNODE is a secret combination of gfp bits that have different behavior than expected. It is a combination of __GFP_THISNODE, __GFP_NORETRY, and __GFP_NOWARN and is special-cased in the page allocator slowpath to fail without trying reclaim even though it may be used in combination with __GFP_WAIT. An example of the problem this creates: commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify") fixed up many users of GFP_THISNODE that really just wanted __GFP_THISNODE. The problem doesn't end there, however, because even it was a no-op for alloc_misplaced_dst_page(), which also sets __GFP_NORETRY and __GFP_NOWARN, and migrate_misplaced_transhuge_page(), where __GFP_NORETRY and __GFP_NOWAIT is set in GFP_TRANSHUGE. Converting GFP_THISNODE to __GFP_THISNODE is a no-op in these cases since the page allocator special-cases __GFP_THISNODE && __GFP_NORETRY && __GFP_NOWARN. It's time to just remove GFP_THISNODE entirely. We leave __GFP_THISNODE to restrict an allocation to a local node, but remove GFP_THISNODE and its obscurity. Instead, we require that a caller clear __GFP_WAIT if it wants to avoid reclaim. This allows the aforementioned functions to actually reclaim as they should. It also enables any future callers that want to do __GFP_THISNODE but also __GFP_NORETRY && __GFP_NOWARN to reclaim. The rule is simple: if you don't want to reclaim, then don't set __GFP_WAIT. Aside: ovs_flow_stats_update() really wants to avoid reclaim as well, so it is unchanged. Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Christoph Lameter <cl@linux.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Jarno Rajahalme <jrajahalme@nicira.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Greg Thelen <gthelen@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-15 06:46:55 +08:00
/*
mm: thp: set THP defrag by default to madvise and add a stall-free defrag option THP defrag is enabled by default to direct reclaim/compact but not wake kswapd in the event of a THP allocation failure. The problem is that THP allocation requests potentially enter reclaim/compaction. This potentially incurs a severe stall that is not guaranteed to be offset by reduced TLB misses. While there has been considerable effort to reduce the impact of reclaim/compaction, it is still a high cost and workloads that should fit in memory fail to do so. Specifically, a simple anon/file streaming workload will enter direct reclaim on NUMA at least even though the working set size is 80% of RAM. It's been years and it's time to throw in the towel. First, this patch defines THP defrag as follows; madvise: A failed allocation will direct reclaim/compact if the application requests it never: Neither reclaim/compact nor wake kswapd defer: A failed allocation will wake kswapd/kcompactd always: A failed allocation will direct reclaim/compact (historical behaviour) khugepaged defrag will enter direct/reclaim but not wake kswapd. Next it sets the default defrag option to be "madvise" to only enter direct reclaim/compaction for applications that specifically requested it. Lastly, it removes a check from the page allocator slowpath that is related to __GFP_THISNODE to allow "defer" to work. The callers that really cares are slub/slab and they are updated accordingly. The slab one may be surprising because it also corrects a comment as kswapd was never woken up by that path. This means that a THP fault will no longer stall for most applications by default and the ideal for most users that get THP if they are immediately available. There are still options for users that prefer a stall at startup of a new application by either restoring historical behaviour with "always" or pick a half-way point with "defer" where kswapd does some of the work in the background and wakes kcompactd if necessary. THP defrag for khugepaged remains enabled and will enter direct/reclaim but no wakeup kswapd or kcompactd. After this patch a THP allocation failure will quickly fallback and rely on khugepaged to recover the situation at some time in the future. In some cases, this will reduce THP usage but the benefit of THP is hard to measure and not a universal win where as a stall to reclaim/compaction is definitely measurable and can be painful. The first test for this is using "usemem" to read a large file and write a large anonymous mapping (to avoid the zero page) multiple times. The total size of the mappings is 80% of RAM and the benchmark simply measures how long it takes to complete. It uses multiple threads to see if that is a factor. On UMA, the performance is almost identical so is not reported but on NUMA, we see this usemem 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Amean System-1 102.86 ( 0.00%) 46.81 ( 54.50%) Amean System-4 37.85 ( 0.00%) 34.02 ( 10.12%) Amean System-7 48.12 ( 0.00%) 46.89 ( 2.56%) Amean System-12 51.98 ( 0.00%) 56.96 ( -9.57%) Amean System-21 80.16 ( 0.00%) 79.05 ( 1.39%) Amean System-30 110.71 ( 0.00%) 107.17 ( 3.20%) Amean System-48 127.98 ( 0.00%) 124.83 ( 2.46%) Amean Elapsd-1 185.84 ( 0.00%) 105.51 ( 43.23%) Amean Elapsd-4 26.19 ( 0.00%) 25.58 ( 2.33%) Amean Elapsd-7 21.65 ( 0.00%) 21.62 ( 0.16%) Amean Elapsd-12 18.58 ( 0.00%) 17.94 ( 3.43%) Amean Elapsd-21 17.53 ( 0.00%) 16.60 ( 5.33%) Amean Elapsd-30 17.45 ( 0.00%) 17.13 ( 1.84%) Amean Elapsd-48 15.40 ( 0.00%) 15.27 ( 0.82%) For a single thread, the benchmark completes 43.23% faster with this patch applied with smaller benefits as the thread increases. Similar, notice the large reduction in most cases in system CPU usage. The overall CPU time is 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 User 10357.65 10438.33 System 3988.88 3543.94 Elapsed 2203.01 1634.41 Which is substantial. Now, the reclaim figures 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 128458477 278352931 Major Faults 2174976 225 Swap Ins 16904701 0 Swap Outs 17359627 0 Allocation stalls 43611 0 DMA allocs 0 0 DMA32 allocs 19832646 19448017 Normal allocs 614488453 580941839 Movable allocs 0 0 Direct pages scanned 24163800 0 Kswapd pages scanned 0 0 Kswapd pages reclaimed 0 0 Direct pages reclaimed 20691346 0 Compaction stalls 42263 0 Compaction success 938 0 Compaction failures 41325 0 This patch eliminates almost all swapping and direct reclaim activity. There is still overhead but it's from NUMA balancing which does not identify that it's pointless trying to do anything with this workload. I also tried the thpscale benchmark which forces a corner case where compaction can be used heavily and measures the latency of whether base or huge pages were used thpscale Fault Latencies 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Amean fault-base-1 5288.84 ( 0.00%) 2817.12 ( 46.73%) Amean fault-base-3 6365.53 ( 0.00%) 3499.11 ( 45.03%) Amean fault-base-5 6526.19 ( 0.00%) 4363.06 ( 33.15%) Amean fault-base-7 7142.25 ( 0.00%) 4858.08 ( 31.98%) Amean fault-base-12 13827.64 ( 0.00%) 10292.11 ( 25.57%) Amean fault-base-18 18235.07 ( 0.00%) 13788.84 ( 24.38%) Amean fault-base-24 21597.80 ( 0.00%) 24388.03 (-12.92%) Amean fault-base-30 26754.15 ( 0.00%) 19700.55 ( 26.36%) Amean fault-base-32 26784.94 ( 0.00%) 19513.57 ( 27.15%) Amean fault-huge-1 4223.96 ( 0.00%) 2178.57 ( 48.42%) Amean fault-huge-3 2194.77 ( 0.00%) 2149.74 ( 2.05%) Amean fault-huge-5 2569.60 ( 0.00%) 2346.95 ( 8.66%) Amean fault-huge-7 3612.69 ( 0.00%) 2997.70 ( 17.02%) Amean fault-huge-12 3301.75 ( 0.00%) 6727.02 (-103.74%) Amean fault-huge-18 6696.47 ( 0.00%) 6685.72 ( 0.16%) Amean fault-huge-24 8000.72 ( 0.00%) 9311.43 (-16.38%) Amean fault-huge-30 13305.55 ( 0.00%) 9750.45 ( 26.72%) Amean fault-huge-32 9981.71 ( 0.00%) 10316.06 ( -3.35%) The average time to fault pages is substantially reduced in the majority of caseds but with the obvious caveat that fewer THPs are actually used in this adverse workload 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Percentage huge-1 0.71 ( 0.00%) 14.04 (1865.22%) Percentage huge-3 10.77 ( 0.00%) 33.05 (206.85%) Percentage huge-5 60.39 ( 0.00%) 38.51 (-36.23%) Percentage huge-7 45.97 ( 0.00%) 34.57 (-24.79%) Percentage huge-12 68.12 ( 0.00%) 40.07 (-41.17%) Percentage huge-18 64.93 ( 0.00%) 47.82 (-26.35%) Percentage huge-24 62.69 ( 0.00%) 44.23 (-29.44%) Percentage huge-30 43.49 ( 0.00%) 55.38 ( 27.34%) Percentage huge-32 50.72 ( 0.00%) 51.90 ( 2.35%) 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 37429143 47564000 Major Faults 1916 1558 Swap Ins 1466 1079 Swap Outs 2936863 149626 Allocation stalls 62510 3 DMA allocs 0 0 DMA32 allocs 6566458 6401314 Normal allocs 216361697 216538171 Movable allocs 0 0 Direct pages scanned 25977580 17998 Kswapd pages scanned 0 3638931 Kswapd pages reclaimed 0 207236 Direct pages reclaimed 8833714 88 Compaction stalls 103349 5 Compaction success 270 4 Compaction failures 103079 1 Note again that while this does swap as it's an aggressive workload, the direct relcim activity and allocation stalls is substantially reduced. There is some kswapd activity but ftrace showed that the kswapd activity was due to normal wakeups from 4K pages being allocated. Compaction-related stalls and activity are almost eliminated. I also tried the stutter benchmark. For this, I do not have figures for NUMA but it's something that does impact UMA so I'll report what is available stutter 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Min mmap 7.3571 ( 0.00%) 7.3438 ( 0.18%) 1st-qrtle mmap 7.5278 ( 0.00%) 17.9200 (-138.05%) 2nd-qrtle mmap 7.6818 ( 0.00%) 21.6055 (-181.25%) 3rd-qrtle mmap 11.0889 ( 0.00%) 21.8881 (-97.39%) Max-90% mmap 27.8978 ( 0.00%) 22.1632 ( 20.56%) Max-93% mmap 28.3202 ( 0.00%) 22.3044 ( 21.24%) Max-95% mmap 28.5600 ( 0.00%) 22.4580 ( 21.37%) Max-99% mmap 29.6032 ( 0.00%) 25.5216 ( 13.79%) Max mmap 4109.7289 ( 0.00%) 4813.9832 (-17.14%) Mean mmap 12.4474 ( 0.00%) 19.3027 (-55.07%) This benchmark is trying to fault an anonymous mapping while there is a heavy IO load -- a scenario that desktop users used to complain about frequently. This shows a mix because the ideal case of mapping with THP is not hit as often. However, note that 99% of the mappings complete 13.79% faster. The CPU usage here is particularly interesting 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 User 67.50 0.99 System 1327.88 91.30 Elapsed 2079.00 2128.98 And once again we look at the reclaim figures 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 335241922 1314582827 Major Faults 715 819 Swap Ins 0 0 Swap Outs 0 0 Allocation stalls 532723 0 DMA allocs 0 0 DMA32 allocs 1822364341 1177950222 Normal allocs 1815640808 1517844854 Movable allocs 0 0 Direct pages scanned 21892772 0 Kswapd pages scanned 20015890 41879484 Kswapd pages reclaimed 19961986 41822072 Direct pages reclaimed 21892741 0 Compaction stalls 1065755 0 Compaction success 514 0 Compaction failures 1065241 0 Allocation stalls and all direct reclaim activity is eliminated as well as compaction-related stalls. THP gives impressive gains in some cases but only if they are quickly available. We're not going to reach the point where they are completely free so lets take the costs out of the fast paths finally and defer the cost to kswapd, kcompactd and khugepaged where it belongs. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-18 05:19:23 +08:00
* Construct gfp mask to allocate from a specific node but do not reclaim or
* warn about failures.
mm: remove GFP_THISNODE NOTE: this is not about __GFP_THISNODE, this is only about GFP_THISNODE. GFP_THISNODE is a secret combination of gfp bits that have different behavior than expected. It is a combination of __GFP_THISNODE, __GFP_NORETRY, and __GFP_NOWARN and is special-cased in the page allocator slowpath to fail without trying reclaim even though it may be used in combination with __GFP_WAIT. An example of the problem this creates: commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify") fixed up many users of GFP_THISNODE that really just wanted __GFP_THISNODE. The problem doesn't end there, however, because even it was a no-op for alloc_misplaced_dst_page(), which also sets __GFP_NORETRY and __GFP_NOWARN, and migrate_misplaced_transhuge_page(), where __GFP_NORETRY and __GFP_NOWAIT is set in GFP_TRANSHUGE. Converting GFP_THISNODE to __GFP_THISNODE is a no-op in these cases since the page allocator special-cases __GFP_THISNODE && __GFP_NORETRY && __GFP_NOWARN. It's time to just remove GFP_THISNODE entirely. We leave __GFP_THISNODE to restrict an allocation to a local node, but remove GFP_THISNODE and its obscurity. Instead, we require that a caller clear __GFP_WAIT if it wants to avoid reclaim. This allows the aforementioned functions to actually reclaim as they should. It also enables any future callers that want to do __GFP_THISNODE but also __GFP_NORETRY && __GFP_NOWARN to reclaim. The rule is simple: if you don't want to reclaim, then don't set __GFP_WAIT. Aside: ovs_flow_stats_update() really wants to avoid reclaim as well, so it is unchanged. Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Christoph Lameter <cl@linux.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Jarno Rajahalme <jrajahalme@nicira.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Greg Thelen <gthelen@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-15 06:46:55 +08:00
*/
static inline gfp_t gfp_exact_node(gfp_t flags)
{
mm: thp: set THP defrag by default to madvise and add a stall-free defrag option THP defrag is enabled by default to direct reclaim/compact but not wake kswapd in the event of a THP allocation failure. The problem is that THP allocation requests potentially enter reclaim/compaction. This potentially incurs a severe stall that is not guaranteed to be offset by reduced TLB misses. While there has been considerable effort to reduce the impact of reclaim/compaction, it is still a high cost and workloads that should fit in memory fail to do so. Specifically, a simple anon/file streaming workload will enter direct reclaim on NUMA at least even though the working set size is 80% of RAM. It's been years and it's time to throw in the towel. First, this patch defines THP defrag as follows; madvise: A failed allocation will direct reclaim/compact if the application requests it never: Neither reclaim/compact nor wake kswapd defer: A failed allocation will wake kswapd/kcompactd always: A failed allocation will direct reclaim/compact (historical behaviour) khugepaged defrag will enter direct/reclaim but not wake kswapd. Next it sets the default defrag option to be "madvise" to only enter direct reclaim/compaction for applications that specifically requested it. Lastly, it removes a check from the page allocator slowpath that is related to __GFP_THISNODE to allow "defer" to work. The callers that really cares are slub/slab and they are updated accordingly. The slab one may be surprising because it also corrects a comment as kswapd was never woken up by that path. This means that a THP fault will no longer stall for most applications by default and the ideal for most users that get THP if they are immediately available. There are still options for users that prefer a stall at startup of a new application by either restoring historical behaviour with "always" or pick a half-way point with "defer" where kswapd does some of the work in the background and wakes kcompactd if necessary. THP defrag for khugepaged remains enabled and will enter direct/reclaim but no wakeup kswapd or kcompactd. After this patch a THP allocation failure will quickly fallback and rely on khugepaged to recover the situation at some time in the future. In some cases, this will reduce THP usage but the benefit of THP is hard to measure and not a universal win where as a stall to reclaim/compaction is definitely measurable and can be painful. The first test for this is using "usemem" to read a large file and write a large anonymous mapping (to avoid the zero page) multiple times. The total size of the mappings is 80% of RAM and the benchmark simply measures how long it takes to complete. It uses multiple threads to see if that is a factor. On UMA, the performance is almost identical so is not reported but on NUMA, we see this usemem 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Amean System-1 102.86 ( 0.00%) 46.81 ( 54.50%) Amean System-4 37.85 ( 0.00%) 34.02 ( 10.12%) Amean System-7 48.12 ( 0.00%) 46.89 ( 2.56%) Amean System-12 51.98 ( 0.00%) 56.96 ( -9.57%) Amean System-21 80.16 ( 0.00%) 79.05 ( 1.39%) Amean System-30 110.71 ( 0.00%) 107.17 ( 3.20%) Amean System-48 127.98 ( 0.00%) 124.83 ( 2.46%) Amean Elapsd-1 185.84 ( 0.00%) 105.51 ( 43.23%) Amean Elapsd-4 26.19 ( 0.00%) 25.58 ( 2.33%) Amean Elapsd-7 21.65 ( 0.00%) 21.62 ( 0.16%) Amean Elapsd-12 18.58 ( 0.00%) 17.94 ( 3.43%) Amean Elapsd-21 17.53 ( 0.00%) 16.60 ( 5.33%) Amean Elapsd-30 17.45 ( 0.00%) 17.13 ( 1.84%) Amean Elapsd-48 15.40 ( 0.00%) 15.27 ( 0.82%) For a single thread, the benchmark completes 43.23% faster with this patch applied with smaller benefits as the thread increases. Similar, notice the large reduction in most cases in system CPU usage. The overall CPU time is 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 User 10357.65 10438.33 System 3988.88 3543.94 Elapsed 2203.01 1634.41 Which is substantial. Now, the reclaim figures 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 128458477 278352931 Major Faults 2174976 225 Swap Ins 16904701 0 Swap Outs 17359627 0 Allocation stalls 43611 0 DMA allocs 0 0 DMA32 allocs 19832646 19448017 Normal allocs 614488453 580941839 Movable allocs 0 0 Direct pages scanned 24163800 0 Kswapd pages scanned 0 0 Kswapd pages reclaimed 0 0 Direct pages reclaimed 20691346 0 Compaction stalls 42263 0 Compaction success 938 0 Compaction failures 41325 0 This patch eliminates almost all swapping and direct reclaim activity. There is still overhead but it's from NUMA balancing which does not identify that it's pointless trying to do anything with this workload. I also tried the thpscale benchmark which forces a corner case where compaction can be used heavily and measures the latency of whether base or huge pages were used thpscale Fault Latencies 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Amean fault-base-1 5288.84 ( 0.00%) 2817.12 ( 46.73%) Amean fault-base-3 6365.53 ( 0.00%) 3499.11 ( 45.03%) Amean fault-base-5 6526.19 ( 0.00%) 4363.06 ( 33.15%) Amean fault-base-7 7142.25 ( 0.00%) 4858.08 ( 31.98%) Amean fault-base-12 13827.64 ( 0.00%) 10292.11 ( 25.57%) Amean fault-base-18 18235.07 ( 0.00%) 13788.84 ( 24.38%) Amean fault-base-24 21597.80 ( 0.00%) 24388.03 (-12.92%) Amean fault-base-30 26754.15 ( 0.00%) 19700.55 ( 26.36%) Amean fault-base-32 26784.94 ( 0.00%) 19513.57 ( 27.15%) Amean fault-huge-1 4223.96 ( 0.00%) 2178.57 ( 48.42%) Amean fault-huge-3 2194.77 ( 0.00%) 2149.74 ( 2.05%) Amean fault-huge-5 2569.60 ( 0.00%) 2346.95 ( 8.66%) Amean fault-huge-7 3612.69 ( 0.00%) 2997.70 ( 17.02%) Amean fault-huge-12 3301.75 ( 0.00%) 6727.02 (-103.74%) Amean fault-huge-18 6696.47 ( 0.00%) 6685.72 ( 0.16%) Amean fault-huge-24 8000.72 ( 0.00%) 9311.43 (-16.38%) Amean fault-huge-30 13305.55 ( 0.00%) 9750.45 ( 26.72%) Amean fault-huge-32 9981.71 ( 0.00%) 10316.06 ( -3.35%) The average time to fault pages is substantially reduced in the majority of caseds but with the obvious caveat that fewer THPs are actually used in this adverse workload 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Percentage huge-1 0.71 ( 0.00%) 14.04 (1865.22%) Percentage huge-3 10.77 ( 0.00%) 33.05 (206.85%) Percentage huge-5 60.39 ( 0.00%) 38.51 (-36.23%) Percentage huge-7 45.97 ( 0.00%) 34.57 (-24.79%) Percentage huge-12 68.12 ( 0.00%) 40.07 (-41.17%) Percentage huge-18 64.93 ( 0.00%) 47.82 (-26.35%) Percentage huge-24 62.69 ( 0.00%) 44.23 (-29.44%) Percentage huge-30 43.49 ( 0.00%) 55.38 ( 27.34%) Percentage huge-32 50.72 ( 0.00%) 51.90 ( 2.35%) 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 37429143 47564000 Major Faults 1916 1558 Swap Ins 1466 1079 Swap Outs 2936863 149626 Allocation stalls 62510 3 DMA allocs 0 0 DMA32 allocs 6566458 6401314 Normal allocs 216361697 216538171 Movable allocs 0 0 Direct pages scanned 25977580 17998 Kswapd pages scanned 0 3638931 Kswapd pages reclaimed 0 207236 Direct pages reclaimed 8833714 88 Compaction stalls 103349 5 Compaction success 270 4 Compaction failures 103079 1 Note again that while this does swap as it's an aggressive workload, the direct relcim activity and allocation stalls is substantially reduced. There is some kswapd activity but ftrace showed that the kswapd activity was due to normal wakeups from 4K pages being allocated. Compaction-related stalls and activity are almost eliminated. I also tried the stutter benchmark. For this, I do not have figures for NUMA but it's something that does impact UMA so I'll report what is available stutter 4.4.0 4.4.0 kcompactd-v1r1 nodefrag-v1r3 Min mmap 7.3571 ( 0.00%) 7.3438 ( 0.18%) 1st-qrtle mmap 7.5278 ( 0.00%) 17.9200 (-138.05%) 2nd-qrtle mmap 7.6818 ( 0.00%) 21.6055 (-181.25%) 3rd-qrtle mmap 11.0889 ( 0.00%) 21.8881 (-97.39%) Max-90% mmap 27.8978 ( 0.00%) 22.1632 ( 20.56%) Max-93% mmap 28.3202 ( 0.00%) 22.3044 ( 21.24%) Max-95% mmap 28.5600 ( 0.00%) 22.4580 ( 21.37%) Max-99% mmap 29.6032 ( 0.00%) 25.5216 ( 13.79%) Max mmap 4109.7289 ( 0.00%) 4813.9832 (-17.14%) Mean mmap 12.4474 ( 0.00%) 19.3027 (-55.07%) This benchmark is trying to fault an anonymous mapping while there is a heavy IO load -- a scenario that desktop users used to complain about frequently. This shows a mix because the ideal case of mapping with THP is not hit as often. However, note that 99% of the mappings complete 13.79% faster. The CPU usage here is particularly interesting 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 User 67.50 0.99 System 1327.88 91.30 Elapsed 2079.00 2128.98 And once again we look at the reclaim figures 4.4.0 4.4.0 kcompactd-v1r1nodefrag-v1r3 Minor Faults 335241922 1314582827 Major Faults 715 819 Swap Ins 0 0 Swap Outs 0 0 Allocation stalls 532723 0 DMA allocs 0 0 DMA32 allocs 1822364341 1177950222 Normal allocs 1815640808 1517844854 Movable allocs 0 0 Direct pages scanned 21892772 0 Kswapd pages scanned 20015890 41879484 Kswapd pages reclaimed 19961986 41822072 Direct pages reclaimed 21892741 0 Compaction stalls 1065755 0 Compaction success 514 0 Compaction failures 1065241 0 Allocation stalls and all direct reclaim activity is eliminated as well as compaction-related stalls. THP gives impressive gains in some cases but only if they are quickly available. We're not going to reach the point where they are completely free so lets take the costs out of the fast paths finally and defer the cost to kswapd, kcompactd and khugepaged where it belongs. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Rik van Riel <riel@redhat.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-18 05:19:23 +08:00
return (flags | __GFP_THISNODE | __GFP_NOWARN) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
mm: remove GFP_THISNODE NOTE: this is not about __GFP_THISNODE, this is only about GFP_THISNODE. GFP_THISNODE is a secret combination of gfp bits that have different behavior than expected. It is a combination of __GFP_THISNODE, __GFP_NORETRY, and __GFP_NOWARN and is special-cased in the page allocator slowpath to fail without trying reclaim even though it may be used in combination with __GFP_WAIT. An example of the problem this creates: commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify") fixed up many users of GFP_THISNODE that really just wanted __GFP_THISNODE. The problem doesn't end there, however, because even it was a no-op for alloc_misplaced_dst_page(), which also sets __GFP_NORETRY and __GFP_NOWARN, and migrate_misplaced_transhuge_page(), where __GFP_NORETRY and __GFP_NOWAIT is set in GFP_TRANSHUGE. Converting GFP_THISNODE to __GFP_THISNODE is a no-op in these cases since the page allocator special-cases __GFP_THISNODE && __GFP_NORETRY && __GFP_NOWARN. It's time to just remove GFP_THISNODE entirely. We leave __GFP_THISNODE to restrict an allocation to a local node, but remove GFP_THISNODE and its obscurity. Instead, we require that a caller clear __GFP_WAIT if it wants to avoid reclaim. This allows the aforementioned functions to actually reclaim as they should. It also enables any future callers that want to do __GFP_THISNODE but also __GFP_NORETRY && __GFP_NOWARN to reclaim. The rule is simple: if you don't want to reclaim, then don't set __GFP_WAIT. Aside: ovs_flow_stats_update() really wants to avoid reclaim as well, so it is unchanged. Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Christoph Lameter <cl@linux.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Jarno Rajahalme <jrajahalme@nicira.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Greg Thelen <gthelen@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-15 06:46:55 +08:00
}
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#endif
static int init_cache_node(struct kmem_cache *cachep, int node, gfp_t gfp)
{
struct kmem_cache_node *n;
/*
* Set up the kmem_cache_node for cpu before we can
* begin anything. Make sure some other cpu on this
* node has not already allocated this
*/
n = get_node(cachep, node);
if (n) {
spin_lock_irq(&n->list_lock);
n->free_limit = (1 + nr_cpus_node(node)) * cachep->batchcount +
cachep->num;
spin_unlock_irq(&n->list_lock);
return 0;
}
n = kmalloc_node(sizeof(struct kmem_cache_node), gfp, node);
if (!n)
return -ENOMEM;
kmem_cache_node_init(n);
n->next_reap = jiffies + REAPTIMEOUT_NODE +
((unsigned long)cachep) % REAPTIMEOUT_NODE;
n->free_limit =
(1 + nr_cpus_node(node)) * cachep->batchcount + cachep->num;
/*
* The kmem_cache_nodes don't come and go as CPUs
* come and go. slab_mutex is sufficient
* protection here.
*/
cachep->node[node] = n;
return 0;
}
#if (defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)) || defined(CONFIG_SMP)
/*
* Allocates and initializes node for a node on each slab cache, used for
* either memory or cpu hotplug. If memory is being hot-added, the kmem_cache_node
* will be allocated off-node since memory is not yet online for the new node.
* When hotplugging memory or a cpu, existing node are not replaced if
* already in use.
*
* Must hold slab_mutex.
*/
static int init_cache_node_node(int node)
{
int ret;
struct kmem_cache *cachep;
list_for_each_entry(cachep, &slab_caches, list) {
ret = init_cache_node(cachep, node, GFP_KERNEL);
if (ret)
return ret;
}
return 0;
}
#endif
static int setup_kmem_cache_node(struct kmem_cache *cachep,
int node, gfp_t gfp, bool force_change)
{
int ret = -ENOMEM;
struct kmem_cache_node *n;
struct array_cache *old_shared = NULL;
struct array_cache *new_shared = NULL;
struct alien_cache **new_alien = NULL;
LIST_HEAD(list);
if (use_alien_caches) {
new_alien = alloc_alien_cache(node, cachep->limit, gfp);
if (!new_alien)
goto fail;
}
if (cachep->shared) {
new_shared = alloc_arraycache(node,
cachep->shared * cachep->batchcount, 0xbaadf00d, gfp);
if (!new_shared)
goto fail;
}
ret = init_cache_node(cachep, node, gfp);
if (ret)
goto fail;
n = get_node(cachep, node);
spin_lock_irq(&n->list_lock);
if (n->shared && force_change) {
free_block(cachep, n->shared->entry,
n->shared->avail, node, &list);
n->shared->avail = 0;
}
if (!n->shared || force_change) {
old_shared = n->shared;
n->shared = new_shared;
new_shared = NULL;
}
if (!n->alien) {
n->alien = new_alien;
new_alien = NULL;
}
spin_unlock_irq(&n->list_lock);
slabs_destroy(cachep, &list);
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
/*
* To protect lockless access to n->shared during irq disabled context.
* If n->shared isn't NULL in irq disabled context, accessing to it is
* guaranteed to be valid until irq is re-enabled, because it will be
* freed after synchronize_rcu().
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
*/
if (old_shared && force_change)
synchronize_rcu();
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
fail:
kfree(old_shared);
kfree(new_shared);
free_alien_cache(new_alien);
return ret;
}
#ifdef CONFIG_SMP
static void cpuup_canceled(long cpu)
{
struct kmem_cache *cachep;
struct kmem_cache_node *n = NULL;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
int node = cpu_to_mem(cpu);
const struct cpumask *mask = cpumask_of_node(node);
list_for_each_entry(cachep, &slab_caches, list) {
struct array_cache *nc;
struct array_cache *shared;
struct alien_cache **alien;
LIST_HEAD(list);
n = get_node(cachep, node);
if (!n)
continue;
spin_lock_irq(&n->list_lock);
/* Free limit for this kmem_cache_node */
n->free_limit -= cachep->batchcount;
/* cpu is dead; no one can alloc from it. */
nc = per_cpu_ptr(cachep->cpu_cache, cpu);
free_block(cachep, nc->entry, nc->avail, node, &list);
nc->avail = 0;
if (!cpumask_empty(mask)) {
spin_unlock_irq(&n->list_lock);
goto free_slab;
}
shared = n->shared;
if (shared) {
free_block(cachep, shared->entry,
shared->avail, node, &list);
n->shared = NULL;
}
alien = n->alien;
n->alien = NULL;
spin_unlock_irq(&n->list_lock);
kfree(shared);
if (alien) {
drain_alien_cache(cachep, alien);
free_alien_cache(alien);
}
free_slab:
slabs_destroy(cachep, &list);
}
/*
* In the previous loop, all the objects were freed to
* the respective cache's slabs, now we can go ahead and
* shrink each nodelist to its limit.
*/
list_for_each_entry(cachep, &slab_caches, list) {
n = get_node(cachep, node);
if (!n)
continue;
drain_freelist(cachep, n, INT_MAX);
}
}
static int cpuup_prepare(long cpu)
{
struct kmem_cache *cachep;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
int node = cpu_to_mem(cpu);
int err;
/*
* We need to do this right in the beginning since
* alloc_arraycache's are going to use this list.
* kmalloc_node allows us to add the slab to the right
* kmem_cache_node and not this cpu's kmem_cache_node
*/
err = init_cache_node_node(node);
if (err < 0)
goto bad;
/*
* Now we can go ahead with allocating the shared arrays and
* array caches
*/
list_for_each_entry(cachep, &slab_caches, list) {
err = setup_kmem_cache_node(cachep, node, GFP_KERNEL, false);
if (err)
goto bad;
}
SLAB: Fix lockdep annotations for CPU hotplug As reported by Paul McKenney: I am seeing some lockdep complaints in rcutorture runs that include frequent CPU-hotplug operations. The tests are otherwise successful. My first thought was to send a patch that gave each array_cache structure's ->lock field its own struct lock_class_key, but you already have a init_lock_keys() that seems to be intended to deal with this. ------------------------------------------------------------------------ ============================================= [ INFO: possible recursive locking detected ] 2.6.32-rc4-autokern1 #1 --------------------------------------------- syslogd/2908 is trying to acquire lock: (&nc->lock){..-...}, at: [<c0000000001407f4>] .kmem_cache_free+0x118/0x2d4 but task is already holding lock: (&nc->lock){..-...}, at: [<c0000000001411bc>] .kfree+0x1f0/0x324 other info that might help us debug this: 3 locks held by syslogd/2908: #0: (&u->readlock){+.+.+.}, at: [<c0000000004556f8>] .unix_dgram_recvmsg+0x70/0x338 #1: (&nc->lock){..-...}, at: [<c0000000001411bc>] .kfree+0x1f0/0x324 #2: (&parent->list_lock){-.-...}, at: [<c000000000140f64>] .__drain_alien_cache+0x50/0xb8 stack backtrace: Call Trace: [c0000000e8ccafc0] [c0000000000101e4] .show_stack+0x70/0x184 (unreliable) [c0000000e8ccb070] [c0000000000afebc] .validate_chain+0x6ec/0xf58 [c0000000e8ccb180] [c0000000000b0ff0] .__lock_acquire+0x8c8/0x974 [c0000000e8ccb280] [c0000000000b2290] .lock_acquire+0x140/0x18c [c0000000e8ccb350] [c000000000468df0] ._spin_lock+0x48/0x70 [c0000000e8ccb3e0] [c0000000001407f4] .kmem_cache_free+0x118/0x2d4 [c0000000e8ccb4a0] [c000000000140b90] .free_block+0x130/0x1a8 [c0000000e8ccb540] [c000000000140f94] .__drain_alien_cache+0x80/0xb8 [c0000000e8ccb5e0] [c0000000001411e0] .kfree+0x214/0x324 [c0000000e8ccb6a0] [c0000000003ca860] .skb_release_data+0xe8/0x104 [c0000000e8ccb730] [c0000000003ca2ec] .__kfree_skb+0x20/0xd4 [c0000000e8ccb7b0] [c0000000003cf2c8] .skb_free_datagram+0x1c/0x5c [c0000000e8ccb830] [c00000000045597c] .unix_dgram_recvmsg+0x2f4/0x338 [c0000000e8ccb920] [c0000000003c0f14] .sock_recvmsg+0xf4/0x13c [c0000000e8ccbb30] [c0000000003c28ec] .SyS_recvfrom+0xb4/0x130 [c0000000e8ccbcb0] [c0000000003bfb78] .sys_recv+0x18/0x2c [c0000000e8ccbd20] [c0000000003ed388] .compat_sys_recv+0x14/0x28 [c0000000e8ccbd90] [c0000000003ee1bc] .compat_sys_socketcall+0x178/0x220 [c0000000e8ccbe30] [c0000000000085d4] syscall_exit+0x0/0x40 This patch fixes the issue by setting up lockdep annotations during CPU hotplug. Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Tested-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux-foundation.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-11-24 04:01:15 +08:00
return 0;
bad:
cpuup_canceled(cpu);
return -ENOMEM;
}
int slab_prepare_cpu(unsigned int cpu)
{
int err;
mutex_lock(&slab_mutex);
err = cpuup_prepare(cpu);
mutex_unlock(&slab_mutex);
return err;
}
/*
* This is called for a failed online attempt and for a successful
* offline.
*
* Even if all the cpus of a node are down, we don't free the
* kmem_list3 of any cache. This to avoid a race between cpu_down, and
* a kmalloc allocation from another cpu for memory from the node of
* the cpu going down. The list3 structure is usually allocated from
* kmem_cache_create() and gets destroyed at kmem_cache_destroy().
*/
int slab_dead_cpu(unsigned int cpu)
{
mutex_lock(&slab_mutex);
cpuup_canceled(cpu);
mutex_unlock(&slab_mutex);
return 0;
}
[PATCH] mm: slab: eliminate lock_cpu_hotplug from slab Here's an attempt towards doing away with lock_cpu_hotplug in the slab subsystem. This approach also fixes a bug which shows up when cpus are being offlined/onlined and slab caches are being tuned simultaneously. http://marc.theaimsgroup.com/?l=linux-kernel&m=116098888100481&w=2 The patch has been stress tested overnight on a 2 socket 4 core AMD box with repeated cpu online and offline, while dbench and kernbench process are running, and slab caches being tuned at the same time. There were no lockdep warnings either. (This test on 2,6.18 as 2.6.19-rc crashes at __drain_pages http://marc.theaimsgroup.com/?l=linux-kernel&m=116172164217678&w=2 ) The approach here is to hold cache_chain_mutex from CPU_UP_PREPARE until CPU_ONLINE (similar in approach as worqueue_mutex) . Slab code sensitive to cpu_online_map (kmem_cache_create, kmem_cache_destroy, slabinfo_write, __cache_shrink) is already serialized with cache_chain_mutex. (This patch lengthens cache_chain_mutex hold time at kmem_cache_destroy to cover this). This patch also takes the cache_chain_sem at kmem_cache_shrink to protect sanity of cpu_online_map at __cache_shrink, as viewed by slab. (kmem_cache_shrink->__cache_shrink->drain_cpu_caches). But, really, kmem_cache_shrink is used at just one place in the acpi subsystem! Do we really need to keep kmem_cache_shrink at all? Another note. Looks like a cpu hotplug event can send CPU_UP_CANCELED to a registered subsystem even if the subsystem did not receive CPU_UP_PREPARE. This could be due to a subsystem registered for notification earlier than the current subsystem crapping out with NOTIFY_BAD. Badness can occur with in the CPU_UP_CANCELED code path at slab if this happens (The same would apply for workqueue.c as well). To overcome this, we might have to use either a) a per subsystem flag and avoid handling of CPU_UP_CANCELED, or b) Use a special notifier events like LOCK_ACQUIRE/RELEASE as Gautham was using in his experiments, or c) Do not send CPU_UP_CANCELED to a subsystem which did not receive CPU_UP_PREPARE. I would prefer c). Signed-off-by: Ravikiran Thirumalai <kiran@scalex86.org> Signed-off-by: Shai Fultheim <shai@scalex86.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:32:14 +08:00
#endif
static int slab_online_cpu(unsigned int cpu)
{
start_cpu_timer(cpu);
return 0;
}
static int slab_offline_cpu(unsigned int cpu)
{
/*
* Shutdown cache reaper. Note that the slab_mutex is held so
* that if cache_reap() is invoked it cannot do anything
* expensive but will only modify reap_work and reschedule the
* timer.
*/
cancel_delayed_work_sync(&per_cpu(slab_reap_work, cpu));
/* Now the cache_reaper is guaranteed to be not running. */
per_cpu(slab_reap_work, cpu).work.func = NULL;
return 0;
}
#if defined(CONFIG_NUMA) && defined(CONFIG_MEMORY_HOTPLUG)
/*
* Drains freelist for a node on each slab cache, used for memory hot-remove.
* Returns -EBUSY if all objects cannot be drained so that the node is not
* removed.
*
* Must hold slab_mutex.
*/
static int __meminit drain_cache_node_node(int node)
{
struct kmem_cache *cachep;
int ret = 0;
list_for_each_entry(cachep, &slab_caches, list) {
struct kmem_cache_node *n;
n = get_node(cachep, node);
if (!n)
continue;
drain_freelist(cachep, n, INT_MAX);
if (!list_empty(&n->slabs_full) ||
!list_empty(&n->slabs_partial)) {
ret = -EBUSY;
break;
}
}
return ret;
}
static int __meminit slab_memory_callback(struct notifier_block *self,
unsigned long action, void *arg)
{
struct memory_notify *mnb = arg;
int ret = 0;
int nid;
nid = mnb->status_change_nid;
if (nid < 0)
goto out;
switch (action) {
case MEM_GOING_ONLINE:
mutex_lock(&slab_mutex);
ret = init_cache_node_node(nid);
mutex_unlock(&slab_mutex);
break;
case MEM_GOING_OFFLINE:
mutex_lock(&slab_mutex);
ret = drain_cache_node_node(nid);
mutex_unlock(&slab_mutex);
break;
case MEM_ONLINE:
case MEM_OFFLINE:
case MEM_CANCEL_ONLINE:
case MEM_CANCEL_OFFLINE:
break;
}
out:
return notifier_from_errno(ret);
}
#endif /* CONFIG_NUMA && CONFIG_MEMORY_HOTPLUG */
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
/*
* swap the static kmem_cache_node with kmalloced memory
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
*/
static void __init init_list(struct kmem_cache *cachep, struct kmem_cache_node *list,
int nodeid)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
struct kmem_cache_node *ptr;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
ptr = kmalloc_node(sizeof(struct kmem_cache_node), GFP_NOWAIT, nodeid);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
BUG_ON(!ptr);
memcpy(ptr, list, sizeof(struct kmem_cache_node));
/*
* Do not assume that spinlocks can be initialized via memcpy:
*/
spin_lock_init(&ptr->list_lock);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
MAKE_ALL_LISTS(cachep, ptr, nodeid);
cachep->node[nodeid] = ptr;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
/*
* For setting up all the kmem_cache_node for cache whose buffer_size is same as
* size of kmem_cache_node.
*/
static void __init set_up_node(struct kmem_cache *cachep, int index)
{
int node;
for_each_online_node(node) {
cachep->node[node] = &init_kmem_cache_node[index + node];
cachep->node[node]->next_reap = jiffies +
REAPTIMEOUT_NODE +
((unsigned long)cachep) % REAPTIMEOUT_NODE;
}
}
/*
* Initialisation. Called after the page allocator have been initialised and
* before smp_init().
*/
void __init kmem_cache_init(void)
{
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
int i;
kmem_cache = &kmem_cache_boot;
if (!IS_ENABLED(CONFIG_NUMA) || num_possible_nodes() == 1)
use_alien_caches = 0;
for (i = 0; i < NUM_INIT_LISTS; i++)
kmem_cache_node_init(&init_kmem_cache_node[i]);
/*
* Fragmentation resistance on low memory - only use bigger
* page orders on machines with more than 32MB of memory if
* not overridden on the command line.
*/
if (!slab_max_order_set && totalram_pages() > (32 << 20) >> PAGE_SHIFT)
slab_max_order = SLAB_MAX_ORDER_HI;
/* Bootstrap is tricky, because several objects are allocated
* from caches that do not exist yet:
* 1) initialize the kmem_cache cache: it contains the struct
* kmem_cache structures of all caches, except kmem_cache itself:
* kmem_cache is statically allocated.
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
* Initially an __init data area is used for the head array and the
* kmem_cache_node structures, it's replaced with a kmalloc allocated
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
* array at the end of the bootstrap.
* 2) Create the first kmalloc cache.
* The struct kmem_cache for the new cache is allocated normally.
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
* An __init data area is used for the head array.
* 3) Create the remaining kmalloc caches, with minimally sized
* head arrays.
* 4) Replace the __init data head arrays for kmem_cache and the first
* kmalloc cache with kmalloc allocated arrays.
* 5) Replace the __init data for kmem_cache_node for kmem_cache and
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
* the other cache's with kmalloc allocated memory.
* 6) Resize the head arrays of the kmalloc caches to their final sizes.
*/
/* 1) create the kmem_cache */
/*
* struct kmem_cache size depends on nr_node_ids & nr_cpu_ids
*/
create_boot_cache(kmem_cache, "kmem_cache",
offsetof(struct kmem_cache, node) +
nr_node_ids * sizeof(struct kmem_cache_node *),
usercopy: Prepare for usercopy whitelisting This patch prepares the slab allocator to handle caches having annotations (useroffset and usersize) defining usercopy regions. This patch is modified from Brad Spengler/PaX Team's PAX_USERCOPY whitelisting code in the last public patch of grsecurity/PaX based on my understanding of the code. Changes or omissions from the original code are mine and don't reflect the original grsecurity/PaX code. Currently, hardened usercopy performs dynamic bounds checking on slab cache objects. This is good, but still leaves a lot of kernel memory available to be copied to/from userspace in the face of bugs. To further restrict what memory is available for copying, this creates a way to whitelist specific areas of a given slab cache object for copying to/from userspace, allowing much finer granularity of access control. Slab caches that are never exposed to userspace can declare no whitelist for their objects, thereby keeping them unavailable to userspace via dynamic copy operations. (Note, an implicit form of whitelisting is the use of constant sizes in usercopy operations and get_user()/put_user(); these bypass hardened usercopy checks since these sizes cannot change at runtime.) To support this whitelist annotation, usercopy region offset and size members are added to struct kmem_cache. The slab allocator receives a new function, kmem_cache_create_usercopy(), that creates a new cache with a usercopy region defined, suitable for declaring spans of fields within the objects that get copied to/from userspace. In this patch, the default kmem_cache_create() marks the entire allocation as whitelisted, leaving it semantically unchanged. Once all fine-grained whitelists have been added (in subsequent patches), this will be changed to a usersize of 0, making caches created with kmem_cache_create() not copyable to/from userspace. After the entire usercopy whitelist series is applied, less than 15% of the slab cache memory remains exposed to potential usercopy bugs after a fresh boot: Total Slab Memory: 48074720 Usercopyable Memory: 6367532 13.2% task_struct 0.2% 4480/1630720 RAW 0.3% 300/96000 RAWv6 2.1% 1408/64768 ext4_inode_cache 3.0% 269760/8740224 dentry 11.1% 585984/5273856 mm_struct 29.1% 54912/188448 kmalloc-8 100.0% 24576/24576 kmalloc-16 100.0% 28672/28672 kmalloc-32 100.0% 81920/81920 kmalloc-192 100.0% 96768/96768 kmalloc-128 100.0% 143360/143360 names_cache 100.0% 163840/163840 kmalloc-64 100.0% 167936/167936 kmalloc-256 100.0% 339968/339968 kmalloc-512 100.0% 350720/350720 kmalloc-96 100.0% 455616/455616 kmalloc-8192 100.0% 655360/655360 kmalloc-1024 100.0% 812032/812032 kmalloc-4096 100.0% 819200/819200 kmalloc-2048 100.0% 1310720/1310720 After some kernel build workloads, the percentage (mainly driven by dentry and inode caches expanding) drops under 10%: Total Slab Memory: 95516184 Usercopyable Memory: 8497452 8.8% task_struct 0.2% 4000/1456000 RAW 0.3% 300/96000 RAWv6 2.1% 1408/64768 ext4_inode_cache 3.0% 1217280/39439872 dentry 11.1% 1623200/14608800 mm_struct 29.1% 73216/251264 kmalloc-8 100.0% 24576/24576 kmalloc-16 100.0% 28672/28672 kmalloc-32 100.0% 94208/94208 kmalloc-192 100.0% 96768/96768 kmalloc-128 100.0% 143360/143360 names_cache 100.0% 163840/163840 kmalloc-64 100.0% 245760/245760 kmalloc-256 100.0% 339968/339968 kmalloc-512 100.0% 350720/350720 kmalloc-96 100.0% 563520/563520 kmalloc-8192 100.0% 655360/655360 kmalloc-1024 100.0% 794624/794624 kmalloc-4096 100.0% 819200/819200 kmalloc-2048 100.0% 1257472/1257472 Signed-off-by: David Windsor <dave@nullcore.net> [kees: adjust commit log, split out a few extra kmalloc hunks] [kees: add field names to function declarations] [kees: convert BUGs to WARNs and fail closed] [kees: add attack surface reduction analysis to commit log] Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: linux-mm@kvack.org Cc: linux-xfs@vger.kernel.org Signed-off-by: Kees Cook <keescook@chromium.org> Acked-by: Christoph Lameter <cl@linux.com>
2017-06-11 10:50:28 +08:00
SLAB_HWCACHE_ALIGN, 0, 0);
list_add(&kmem_cache->list, &slab_caches);
mm: memcg/slab: postpone kmem_cache memcg pointer initialization to memcg_link_cache() Patch series "mm: reparent slab memory on cgroup removal", v7. # Why do we need this? We've noticed that the number of dying cgroups is steadily growing on most of our hosts in production. The following investigation revealed an issue in the userspace memory reclaim code [1], accounting of kernel stacks [2], and also the main reason: slab objects. The underlying problem is quite simple: any page charged to a cgroup holds a reference to it, so the cgroup can't be reclaimed unless all charged pages are gone. If a slab object is actively used by other cgroups, it won't be reclaimed, and will prevent the origin cgroup from being reclaimed. Slab objects, and first of all vfs cache, is shared between cgroups, which are using the same underlying fs, and what's even more important, it's shared between multiple generations of the same workload. So if something is running periodically every time in a new cgroup (like how systemd works), we do accumulate multiple dying cgroups. Strictly speaking pagecache isn't different here, but there is a key difference: we disable protection and apply some extra pressure on LRUs of dying cgroups, and these LRUs contain all charged pages. My experiments show that with the disabled kernel memory accounting the number of dying cgroups stabilizes at a relatively small number (~100, depends on memory pressure and cgroup creation rate), and with kernel memory accounting it grows pretty steadily up to several thousands. Memory cgroups are quite complex and big objects (mostly due to percpu stats), so it leads to noticeable memory losses. Memory occupied by dying cgroups is measured in hundreds of megabytes. I've even seen a host with more than 100Gb of memory wasted for dying cgroups. It leads to a degradation of performance with the uptime, and generally limits the usage of cgroups. My previous attempt [3] to fix the problem by applying extra pressure on slab shrinker lists caused a regressions with xfs and ext4, and has been reverted [4]. The following attempts to find the right balance [5, 6] were not successful. So instead of trying to find a maybe non-existing balance, let's do reparent accounted slab caches to the parent cgroup on cgroup removal. # Implementation approach There is however a significant problem with reparenting of slab memory: there is no list of charged pages. Some of them are in shrinker lists, but not all. Introducing of a new list is really not an option. But fortunately there is a way forward: every slab page has a stable pointer to the corresponding kmem_cache. So the idea is to reparent kmem_caches instead of slab pages. It's actually simpler and cheaper, but requires some underlying changes: 1) Make kmem_caches to hold a single reference to the memory cgroup, instead of a separate reference per every slab page. 2) Stop setting page->mem_cgroup pointer for memcg slab pages and use page->kmem_cache->memcg indirection instead. It's used only on slab page release, so performance overhead shouldn't be a big issue. 3) Introduce a refcounter for non-root slab caches. It's required to be able to destroy kmem_caches when they become empty and release the associated memory cgroup. There is a bonus: currently we release all memcg kmem_caches all together with the memory cgroup itself. This patchset allows individual kmem_caches to be released as soon as they become inactive and free. Some additional implementation details are provided in corresponding commit messages. # Results Below is the average number of dying cgroups on two groups of our production hosts. They do run some sort of web frontend workload, the memory pressure is moderate. As we can see, with the kernel memory reparenting the number stabilizes in 60s range; however with the original version it grows almost linearly and doesn't show any signs of plateauing. The difference in slab and percpu usage between patched and unpatched versions also grows linearly. In 7 days it exceeded 200Mb. day 0 1 2 3 4 5 6 7 original 56 362 628 752 1070 1250 1490 1560 patched 23 46 51 55 60 57 67 69 mem diff(Mb) 22 74 123 152 164 182 214 241 # Links [1]: commit 68600f623d69 ("mm: don't miss the last page because of round-off error") [2]: commit 9b6f7e163cd0 ("mm: rework memcg kernel stack accounting") [3]: commit 172b06c32b94 ("mm: slowly shrink slabs with a relatively small number of objects") [4]: commit a9a238e83fbb ("Revert "mm: slowly shrink slabs with a relatively small number of objects") [5]: https://lkml.org/lkml/2019/1/28/1865 [6]: https://marc.info/?l=linux-mm&m=155064763626437&w=2 This patch (of 10): Initialize kmem_cache->memcg_params.memcg pointer in memcg_link_cache() rather than in init_memcg_params(). Once kmem_cache will hold a reference to the memory cgroup, it will simplify the refcounting. For non-root kmem_caches memcg_link_cache() is always called before the kmem_cache becomes visible to a user, so it's safe. Link: http://lkml.kernel.org/r/20190611231813.3148843-2-guro@fb.com Signed-off-by: Roman Gushchin <guro@fb.com> Reviewed-by: Shakeel Butt <shakeelb@google.com> Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Waiman Long <longman@redhat.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Andrei Vagin <avagin@gmail.com> Cc: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:56:02 +08:00
memcg_link_cache(kmem_cache, NULL);
slab_state = PARTIAL;
/*
* Initialize the caches that provide memory for the kmem_cache_node
* structures first. Without this, further allocations will bug.
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
*/
mm, slab: combine kmalloc_caches and kmalloc_dma_caches Patch series "kmalloc-reclaimable caches", v4. As discussed at LSF/MM [1] here's a patchset that introduces kmalloc-reclaimable caches (more details in the second patch) and uses them for dcache external names. That allows us to repurpose the NR_INDIRECTLY_RECLAIMABLE_BYTES counter later in the series. With patch 3/6, dcache external names are allocated from kmalloc-rcl-* caches, eliminating the need for manual accounting. More importantly, it also ensures the reclaimable kmalloc allocations are grouped in pages separate from the regular kmalloc allocations. The need for proper accounting of dcache external names has shown it's easy for misbehaving process to allocate lots of them, causing premature OOMs. Without the added grouping, it's likely that a similar workload can interleave the dcache external names allocations with regular kmalloc allocations (note: I haven't searched myself for an example of such regular kmalloc allocation, but I would be very surprised if there wasn't some). A pathological case would be e.g. one 64byte regular allocations with 63 external dcache names in a page (64x64=4096), which means the page is not freed even after reclaiming after all dcache names, and the process can thus "steal" the whole page with single 64byte allocation. If other kmalloc users similar to dcache external names become identified, they can also benefit from the new functionality simply by adding __GFP_RECLAIMABLE to the kmalloc calls. Side benefits of the patchset (that could be also merged separately) include removed branch for detecting __GFP_DMA kmalloc(), and shortening kmalloc cache names in /proc/slabinfo output. The latter is potentially an ABI break in case there are tools parsing the names and expecting the values to be in bytes. This is how /proc/slabinfo looks like after booting in virtme: ... kmalloc-rcl-4M 0 0 4194304 1 1024 : tunables 1 1 0 : slabdata 0 0 0 ... kmalloc-rcl-96 7 32 128 32 1 : tunables 120 60 8 : slabdata 1 1 0 kmalloc-rcl-64 25 128 64 64 1 : tunables 120 60 8 : slabdata 2 2 0 kmalloc-rcl-32 0 0 32 124 1 : tunables 120 60 8 : slabdata 0 0 0 kmalloc-4M 0 0 4194304 1 1024 : tunables 1 1 0 : slabdata 0 0 0 kmalloc-2M 0 0 2097152 1 512 : tunables 1 1 0 : slabdata 0 0 0 kmalloc-1M 0 0 1048576 1 256 : tunables 1 1 0 : slabdata 0 0 0 ... /proc/vmstat with renamed nr_indirectly_reclaimable_bytes counter: ... nr_slab_reclaimable 2817 nr_slab_unreclaimable 1781 ... nr_kernel_misc_reclaimable 0 ... /proc/meminfo with new KReclaimable counter: ... Shmem: 564 kB KReclaimable: 11260 kB Slab: 18368 kB SReclaimable: 11260 kB SUnreclaim: 7108 kB KernelStack: 1248 kB ... This patch (of 6): The kmalloc caches currently mainain separate (optional) array kmalloc_dma_caches for __GFP_DMA allocations. There are tests for __GFP_DMA in the allocation hotpaths. We can avoid the branches by combining kmalloc_caches and kmalloc_dma_caches into a single two-dimensional array where the outer dimension is cache "type". This will also allow to add kmalloc-reclaimable caches as a third type. Link: http://lkml.kernel.org/r/20180731090649.16028-2-vbabka@suse.cz Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Christoph Lameter <cl@linux.com> Acked-by: Roman Gushchin <guro@fb.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Laura Abbott <labbott@redhat.com> Cc: Sumit Semwal <sumit.semwal@linaro.org> Cc: Vijayanand Jitta <vjitta@codeaurora.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-27 06:05:34 +08:00
kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE] = create_kmalloc_cache(
mm, slab: make kmalloc_info[] contain all types of names Patch series "mm, slab: Make kmalloc_info[] contain all types of names", v6. There are three types of kmalloc, KMALLOC_NORMAL, KMALLOC_RECLAIM and KMALLOC_DMA. The name of KMALLOC_NORMAL is contained in kmalloc_info[].name, but the names of KMALLOC_RECLAIM and KMALLOC_DMA are dynamically generated by kmalloc_cache_name(). Patch1 predefines the names of all types of kmalloc to save the time spent dynamically generating names. These changes make sense, and the time spent by new_kmalloc_cache() has been reduced by approximately 36.3%. Time spent by new_kmalloc_cache() (CPU cycles) 5.3-rc7 66264 5.3-rc7+patch 42188 This patch (of 3): There are three types of kmalloc, KMALLOC_NORMAL, KMALLOC_RECLAIM and KMALLOC_DMA. The name of KMALLOC_NORMAL is contained in kmalloc_info[].name, but the names of KMALLOC_RECLAIM and KMALLOC_DMA are dynamically generated by kmalloc_cache_name(). This patch predefines the names of all types of kmalloc to save the time spent dynamically generating names. Besides, remove the kmalloc_cache_name() that is no longer used. Link: http://lkml.kernel.org/r/1569241648-26908-2-git-send-email-lpf.vector@gmail.com Signed-off-by: Pengfei Li <lpf.vector@gmail.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Roman Gushchin <guro@fb.com> Acked-by: David Rientjes <rientjes@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-12-01 09:49:21 +08:00
kmalloc_info[INDEX_NODE].name[KMALLOC_NORMAL],
kmalloc_info[INDEX_NODE].size,
ARCH_KMALLOC_FLAGS, 0,
kmalloc_info[INDEX_NODE].size);
slab_state = PARTIAL_NODE;
slab: correct size_index table before replacing the bootstrap kmem_cache_node This patch moves the initialization of the size_index table slightly earlier so that the first few kmem_cache_node's can be safely allocated when KMALLOC_MIN_SIZE is large. There are currently two ways to generate indices into kmalloc_caches (via kmalloc_index() and via the size_index table in slab_common.c) and on some arches (possibly only MIPS) they potentially disagree with each other until create_kmalloc_caches() has been called. It seems that the intention is that the size_index table is a fast equivalent to kmalloc_index() and that create_kmalloc_caches() patches the table to return the correct value for the cases where kmalloc_index()'s if-statements apply. The failing sequence was: * kmalloc_caches contains NULL elements * kmem_cache_init initialises the element that 'struct kmem_cache_node' will be allocated to. For 32-bit Mips, this is a 56-byte struct and kmalloc_index returns KMALLOC_SHIFT_LOW (7). * init_list is called which calls kmalloc_node to allocate a 'struct kmem_cache_node'. * kmalloc_slab selects the kmem_caches element using size_index[size_index_elem(size)]. For MIPS, size is 56, and the expression returns 6. * This element of kmalloc_caches is NULL and allocation fails. * If it had not already failed, it would have called create_kmalloc_caches() at this point which would have changed size_index[size_index_elem(size)] to 7. I don't believe the bug to be LLVM specific but GCC doesn't normally encounter the problem. I haven't been able to identify exactly what GCC is doing better (probably inlining) but it seems that GCC is managing to optimize to the point that it eliminates the problematic allocations. This theory is supported by the fact that GCC can be made to fail in the same way by changing inline, __inline, __inline__, and __always_inline in include/linux/compiler-gcc.h such that they don't actually inline things. Signed-off-by: Daniel Sanders <daniel.sanders@imgtec.com> Acked-by: Pekka Enberg <penberg@kernel.org> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-06-25 07:55:57 +08:00
setup_kmalloc_cache_index_table();
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
slab_early_init = 0;
/* 5) Replace the bootstrap kmem_cache_node */
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
int nid;
for_each_online_node(nid) {
init_list(kmem_cache, &init_kmem_cache_node[CACHE_CACHE + nid], nid);
mm, slab: combine kmalloc_caches and kmalloc_dma_caches Patch series "kmalloc-reclaimable caches", v4. As discussed at LSF/MM [1] here's a patchset that introduces kmalloc-reclaimable caches (more details in the second patch) and uses them for dcache external names. That allows us to repurpose the NR_INDIRECTLY_RECLAIMABLE_BYTES counter later in the series. With patch 3/6, dcache external names are allocated from kmalloc-rcl-* caches, eliminating the need for manual accounting. More importantly, it also ensures the reclaimable kmalloc allocations are grouped in pages separate from the regular kmalloc allocations. The need for proper accounting of dcache external names has shown it's easy for misbehaving process to allocate lots of them, causing premature OOMs. Without the added grouping, it's likely that a similar workload can interleave the dcache external names allocations with regular kmalloc allocations (note: I haven't searched myself for an example of such regular kmalloc allocation, but I would be very surprised if there wasn't some). A pathological case would be e.g. one 64byte regular allocations with 63 external dcache names in a page (64x64=4096), which means the page is not freed even after reclaiming after all dcache names, and the process can thus "steal" the whole page with single 64byte allocation. If other kmalloc users similar to dcache external names become identified, they can also benefit from the new functionality simply by adding __GFP_RECLAIMABLE to the kmalloc calls. Side benefits of the patchset (that could be also merged separately) include removed branch for detecting __GFP_DMA kmalloc(), and shortening kmalloc cache names in /proc/slabinfo output. The latter is potentially an ABI break in case there are tools parsing the names and expecting the values to be in bytes. This is how /proc/slabinfo looks like after booting in virtme: ... kmalloc-rcl-4M 0 0 4194304 1 1024 : tunables 1 1 0 : slabdata 0 0 0 ... kmalloc-rcl-96 7 32 128 32 1 : tunables 120 60 8 : slabdata 1 1 0 kmalloc-rcl-64 25 128 64 64 1 : tunables 120 60 8 : slabdata 2 2 0 kmalloc-rcl-32 0 0 32 124 1 : tunables 120 60 8 : slabdata 0 0 0 kmalloc-4M 0 0 4194304 1 1024 : tunables 1 1 0 : slabdata 0 0 0 kmalloc-2M 0 0 2097152 1 512 : tunables 1 1 0 : slabdata 0 0 0 kmalloc-1M 0 0 1048576 1 256 : tunables 1 1 0 : slabdata 0 0 0 ... /proc/vmstat with renamed nr_indirectly_reclaimable_bytes counter: ... nr_slab_reclaimable 2817 nr_slab_unreclaimable 1781 ... nr_kernel_misc_reclaimable 0 ... /proc/meminfo with new KReclaimable counter: ... Shmem: 564 kB KReclaimable: 11260 kB Slab: 18368 kB SReclaimable: 11260 kB SUnreclaim: 7108 kB KernelStack: 1248 kB ... This patch (of 6): The kmalloc caches currently mainain separate (optional) array kmalloc_dma_caches for __GFP_DMA allocations. There are tests for __GFP_DMA in the allocation hotpaths. We can avoid the branches by combining kmalloc_caches and kmalloc_dma_caches into a single two-dimensional array where the outer dimension is cache "type". This will also allow to add kmalloc-reclaimable caches as a third type. Link: http://lkml.kernel.org/r/20180731090649.16028-2-vbabka@suse.cz Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Christoph Lameter <cl@linux.com> Acked-by: Roman Gushchin <guro@fb.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Laura Abbott <labbott@redhat.com> Cc: Sumit Semwal <sumit.semwal@linaro.org> Cc: Vijayanand Jitta <vjitta@codeaurora.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-27 06:05:34 +08:00
init_list(kmalloc_caches[KMALLOC_NORMAL][INDEX_NODE],
&init_kmem_cache_node[SIZE_NODE + nid], nid);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
}
create_kmalloc_caches(ARCH_KMALLOC_FLAGS);
slab: setup cpu caches later on when interrupts are enabled Fixes the following boot-time warning: [ 0.000000] ------------[ cut here ]------------ [ 0.000000] WARNING: at kernel/smp.c:369 smp_call_function_many+0x56/0x1bc() [ 0.000000] Hardware name: [ 0.000000] Modules linked in: [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30 #492 [ 0.000000] Call Trace: [ 0.000000] [<ffffffff8149e021>] ? _spin_unlock+0x4f/0x5c [ 0.000000] [<ffffffff8108f11b>] ? smp_call_function_many+0x56/0x1bc [ 0.000000] [<ffffffff81061764>] warn_slowpath_common+0x7c/0xa9 [ 0.000000] [<ffffffff810617a5>] warn_slowpath_null+0x14/0x16 [ 0.000000] [<ffffffff8108f11b>] smp_call_function_many+0x56/0x1bc [ 0.000000] [<ffffffff810f3e00>] ? do_ccupdate_local+0x0/0x54 [ 0.000000] [<ffffffff810f3e00>] ? do_ccupdate_local+0x0/0x54 [ 0.000000] [<ffffffff8108f2be>] smp_call_function+0x3d/0x68 [ 0.000000] [<ffffffff810f3e00>] ? do_ccupdate_local+0x0/0x54 [ 0.000000] [<ffffffff81066fd8>] on_each_cpu+0x31/0x7c [ 0.000000] [<ffffffff810f64f5>] do_tune_cpucache+0x119/0x454 [ 0.000000] [<ffffffff81087080>] ? lockdep_init_map+0x94/0x10b [ 0.000000] [<ffffffff818133b0>] ? kmem_cache_init+0x421/0x593 [ 0.000000] [<ffffffff810f69cf>] enable_cpucache+0x68/0xad [ 0.000000] [<ffffffff818133c3>] kmem_cache_init+0x434/0x593 [ 0.000000] [<ffffffff8180987c>] ? mem_init+0x156/0x161 [ 0.000000] [<ffffffff817f8aae>] start_kernel+0x1cc/0x3b9 [ 0.000000] [<ffffffff817f829a>] x86_64_start_reservations+0xaa/0xae [ 0.000000] [<ffffffff817f837f>] x86_64_start_kernel+0xe1/0xe8 [ 0.000000] ---[ end trace 4eaa2a86a8e2da22 ]--- Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-12 20:58:59 +08:00
}
void __init kmem_cache_init_late(void)
{
struct kmem_cache *cachep;
/* 6) resize the head arrays to their final sizes */
mutex_lock(&slab_mutex);
list_for_each_entry(cachep, &slab_caches, list)
slab: setup cpu caches later on when interrupts are enabled Fixes the following boot-time warning: [ 0.000000] ------------[ cut here ]------------ [ 0.000000] WARNING: at kernel/smp.c:369 smp_call_function_many+0x56/0x1bc() [ 0.000000] Hardware name: [ 0.000000] Modules linked in: [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30 #492 [ 0.000000] Call Trace: [ 0.000000] [<ffffffff8149e021>] ? _spin_unlock+0x4f/0x5c [ 0.000000] [<ffffffff8108f11b>] ? smp_call_function_many+0x56/0x1bc [ 0.000000] [<ffffffff81061764>] warn_slowpath_common+0x7c/0xa9 [ 0.000000] [<ffffffff810617a5>] warn_slowpath_null+0x14/0x16 [ 0.000000] [<ffffffff8108f11b>] smp_call_function_many+0x56/0x1bc [ 0.000000] [<ffffffff810f3e00>] ? do_ccupdate_local+0x0/0x54 [ 0.000000] [<ffffffff810f3e00>] ? do_ccupdate_local+0x0/0x54 [ 0.000000] [<ffffffff8108f2be>] smp_call_function+0x3d/0x68 [ 0.000000] [<ffffffff810f3e00>] ? do_ccupdate_local+0x0/0x54 [ 0.000000] [<ffffffff81066fd8>] on_each_cpu+0x31/0x7c [ 0.000000] [<ffffffff810f64f5>] do_tune_cpucache+0x119/0x454 [ 0.000000] [<ffffffff81087080>] ? lockdep_init_map+0x94/0x10b [ 0.000000] [<ffffffff818133b0>] ? kmem_cache_init+0x421/0x593 [ 0.000000] [<ffffffff810f69cf>] enable_cpucache+0x68/0xad [ 0.000000] [<ffffffff818133c3>] kmem_cache_init+0x434/0x593 [ 0.000000] [<ffffffff8180987c>] ? mem_init+0x156/0x161 [ 0.000000] [<ffffffff817f8aae>] start_kernel+0x1cc/0x3b9 [ 0.000000] [<ffffffff817f829a>] x86_64_start_reservations+0xaa/0xae [ 0.000000] [<ffffffff817f837f>] x86_64_start_kernel+0xe1/0xe8 [ 0.000000] ---[ end trace 4eaa2a86a8e2da22 ]--- Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-12 20:58:59 +08:00
if (enable_cpucache(cachep, GFP_NOWAIT))
BUG();
mutex_unlock(&slab_mutex);
/* Done! */
slab_state = FULL;
#ifdef CONFIG_NUMA
/*
* Register a memory hotplug callback that initializes and frees
* node.
*/
hotplug_memory_notifier(slab_memory_callback, SLAB_CALLBACK_PRI);
#endif
/*
* The reap timers are started later, with a module init call: That part
* of the kernel is not yet operational.
*/
}
static int __init cpucache_init(void)
{
int ret;
/*
* Register the timers that return unneeded pages to the page allocator
*/
ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "SLAB online",
slab_online_cpu, slab_offline_cpu);
WARN_ON(ret < 0);
return 0;
}
__initcall(cpucache_init);
static noinline void
slab_out_of_memory(struct kmem_cache *cachep, gfp_t gfpflags, int nodeid)
{
#if DEBUG
struct kmem_cache_node *n;
unsigned long flags;
int node;
static DEFINE_RATELIMIT_STATE(slab_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
DEFAULT_RATELIMIT_BURST);
if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slab_oom_rs))
return;
pr_warn("SLAB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
nodeid, gfpflags, &gfpflags);
pr_warn(" cache: %s, object size: %d, order: %d\n",
cachep->name, cachep->size, cachep->gfporder);
for_each_kmem_cache_node(cachep, node, n) {
unsigned long total_slabs, free_slabs, free_objs;
spin_lock_irqsave(&n->list_lock, flags);
total_slabs = n->total_slabs;
free_slabs = n->free_slabs;
free_objs = n->free_objects;
spin_unlock_irqrestore(&n->list_lock, flags);
pr_warn(" node %d: slabs: %ld/%ld, objs: %ld/%ld\n",
node, total_slabs - free_slabs, total_slabs,
(total_slabs * cachep->num) - free_objs,
total_slabs * cachep->num);
}
#endif
}
/*
* Interface to system's page allocator. No need to hold the
* kmem_cache_node ->list_lock.
*
* If we requested dmaable memory, we will get it. Even if we
* did not request dmaable memory, we might get it, but that
* would be relatively rare and ignorable.
*/
static struct page *kmem_getpages(struct kmem_cache *cachep, gfp_t flags,
int nodeid)
{
struct page *page;
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
flags |= cachep->allocflags;
page = __alloc_pages_node(nodeid, flags, cachep->gfporder);
if (!page) {
slab_out_of_memory(cachep, flags, nodeid);
return NULL;
}
if (charge_slab_page(page, flags, cachep->gfporder, cachep)) {
memcg: unify slab and other kmem pages charging We have memcg_kmem_charge and memcg_kmem_uncharge methods for charging and uncharging kmem pages to memcg, but currently they are not used for charging slab pages (i.e. they are only used for charging pages allocated with alloc_kmem_pages). The only reason why the slab subsystem uses special helpers, memcg_charge_slab and memcg_uncharge_slab, is that it needs to charge to the memcg of kmem cache while memcg_charge_kmem charges to the memcg that the current task belongs to. To remove this diversity, this patch adds an extra argument to __memcg_kmem_charge that can be a pointer to a memcg or NULL. If it is not NULL, the function tries to charge to the memcg it points to, otherwise it charge to the current context. Next, it makes the slab subsystem use this function to charge slab pages. Since memcg_charge_kmem and memcg_uncharge_kmem helpers are now used only in __memcg_kmem_charge and __memcg_kmem_uncharge, they are inlined. Since __memcg_kmem_charge stores a pointer to the memcg in the page struct, we don't need memcg_uncharge_slab anymore and can use free_kmem_pages. Besides, one can now detect which memcg a slab page belongs to by reading /proc/kpagecgroup. Note, this patch switches slab to charge-after-alloc design. Since this design is already used for all other memcg charges, it should not make any difference. [hannes@cmpxchg.org: better to have an outer function than a magic parameter for the memcg lookup] Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-06 10:49:01 +08:00
__free_pages(page, cachep->gfporder);
return NULL;
}
__SetPageSlab(page);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
/* Record if ALLOC_NO_WATERMARKS was set when allocating the slab */
if (sk_memalloc_socks() && page_is_pfmemalloc(page))
SetPageSlabPfmemalloc(page);
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
return page;
}
/*
* Interface to system's page release.
*/
static void kmem_freepages(struct kmem_cache *cachep, struct page *page)
{
int order = cachep->gfporder;
BUG_ON(!PageSlab(page));
__ClearPageSlabPfmemalloc(page);
__ClearPageSlab(page);
page_mapcount_reset(page);
page->mapping = NULL;
if (current->reclaim_state)
current->reclaim_state->reclaimed_slab += 1 << order;
uncharge_slab_page(page, order, cachep);
__free_pages(page, order);
}
static void kmem_rcu_free(struct rcu_head *head)
{
struct kmem_cache *cachep;
struct page *page;
page = container_of(head, struct page, rcu_head);
cachep = page->slab_cache;
kmem_freepages(cachep, page);
}
#if DEBUG
static bool is_debug_pagealloc_cache(struct kmem_cache *cachep)
{
mm, debug_pagealloc: don't rely on static keys too early Commit 96a2b03f281d ("mm, debug_pagelloc: use static keys to enable debugging") has introduced a static key to reduce overhead when debug_pagealloc is compiled in but not enabled. It relied on the assumption that jump_label_init() is called before parse_early_param() as in start_kernel(), so when the "debug_pagealloc=on" option is parsed, it is safe to enable the static key. However, it turns out multiple architectures call parse_early_param() earlier from their setup_arch(). x86 also calls jump_label_init() even earlier, so no issue was found while testing the commit, but same is not true for e.g. ppc64 and s390 where the kernel would not boot with debug_pagealloc=on as found by our QA. To fix this without tricky changes to init code of multiple architectures, this patch partially reverts the static key conversion from 96a2b03f281d. Init-time and non-fastpath calls (such as in arch code) of debug_pagealloc_enabled() will again test a simple bool variable. Fastpath mm code is converted to a new debug_pagealloc_enabled_static() variant that relies on the static key, which is enabled in a well-defined point in mm_init() where it's guaranteed that jump_label_init() has been called, regardless of architecture. [sfr@canb.auug.org.au: export _debug_pagealloc_enabled_early] Link: http://lkml.kernel.org/r/20200106164944.063ac07b@canb.auug.org.au Link: http://lkml.kernel.org/r/20191219130612.23171-1-vbabka@suse.cz Fixes: 96a2b03f281d ("mm, debug_pagelloc: use static keys to enable debugging") Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Qian Cai <cai@lca.pw> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-14 08:29:20 +08:00
if (debug_pagealloc_enabled_static() && OFF_SLAB(cachep) &&
(cachep->size % PAGE_SIZE) == 0)
return true;
return false;
}
#ifdef CONFIG_DEBUG_PAGEALLOC
static void slab_kernel_map(struct kmem_cache *cachep, void *objp, int map)
{
if (!is_debug_pagealloc_cache(cachep))
return;
kernel_map_pages(virt_to_page(objp), cachep->size / PAGE_SIZE, map);
}
#else
static inline void slab_kernel_map(struct kmem_cache *cachep, void *objp,
int map) {}
#endif
static void poison_obj(struct kmem_cache *cachep, void *addr, unsigned char val)
{
int size = cachep->object_size;
addr = &((char *)addr)[obj_offset(cachep)];
memset(addr, val, size);
*(unsigned char *)(addr + size - 1) = POISON_END;
}
static void dump_line(char *data, int offset, int limit)
{
int i;
unsigned char error = 0;
int bad_count = 0;
pr_err("%03x: ", offset);
for (i = 0; i < limit; i++) {
if (data[offset + i] != POISON_FREE) {
error = data[offset + i];
bad_count++;
}
}
print_hex_dump(KERN_CONT, "", 0, 16, 1,
&data[offset], limit, 1);
if (bad_count == 1) {
error ^= POISON_FREE;
if (!(error & (error - 1))) {
pr_err("Single bit error detected. Probably bad RAM.\n");
#ifdef CONFIG_X86
pr_err("Run memtest86+ or a similar memory test tool.\n");
#else
pr_err("Run a memory test tool.\n");
#endif
}
}
}
#endif
#if DEBUG
static void print_objinfo(struct kmem_cache *cachep, void *objp, int lines)
{
int i, size;
char *realobj;
if (cachep->flags & SLAB_RED_ZONE) {
pr_err("Redzone: 0x%llx/0x%llx\n",
*dbg_redzone1(cachep, objp),
*dbg_redzone2(cachep, objp));
}
if (cachep->flags & SLAB_STORE_USER)
pr_err("Last user: (%pSR)\n", *dbg_userword(cachep, objp));
realobj = (char *)objp + obj_offset(cachep);
size = cachep->object_size;
for (i = 0; i < size && lines; i += 16, lines--) {
int limit;
limit = 16;
if (i + limit > size)
limit = size - i;
dump_line(realobj, i, limit);
}
}
static void check_poison_obj(struct kmem_cache *cachep, void *objp)
{
char *realobj;
int size, i;
int lines = 0;
if (is_debug_pagealloc_cache(cachep))
return;
realobj = (char *)objp + obj_offset(cachep);
size = cachep->object_size;
for (i = 0; i < size; i++) {
char exp = POISON_FREE;
if (i == size - 1)
exp = POISON_END;
if (realobj[i] != exp) {
int limit;
/* Mismatch ! */
/* Print header */
if (lines == 0) {
pr_err("Slab corruption (%s): %s start=%px, len=%d\n",
print_tainted(), cachep->name,
realobj, size);
print_objinfo(cachep, objp, 0);
}
/* Hexdump the affected line */
i = (i / 16) * 16;
limit = 16;
if (i + limit > size)
limit = size - i;
dump_line(realobj, i, limit);
i += 16;
lines++;
/* Limit to 5 lines */
if (lines > 5)
break;
}
}
if (lines != 0) {
/* Print some data about the neighboring objects, if they
* exist:
*/
struct page *page = virt_to_head_page(objp);
unsigned int objnr;
objnr = obj_to_index(cachep, page, objp);
if (objnr) {
objp = index_to_obj(cachep, page, objnr - 1);
realobj = (char *)objp + obj_offset(cachep);
pr_err("Prev obj: start=%px, len=%d\n", realobj, size);
print_objinfo(cachep, objp, 2);
}
if (objnr + 1 < cachep->num) {
objp = index_to_obj(cachep, page, objnr + 1);
realobj = (char *)objp + obj_offset(cachep);
pr_err("Next obj: start=%px, len=%d\n", realobj, size);
print_objinfo(cachep, objp, 2);
}
}
}
#endif
#if DEBUG
static void slab_destroy_debugcheck(struct kmem_cache *cachep,
struct page *page)
{
int i;
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (OBJFREELIST_SLAB(cachep) && cachep->flags & SLAB_POISON) {
poison_obj(cachep, page->freelist - obj_offset(cachep),
POISON_FREE);
}
for (i = 0; i < cachep->num; i++) {
void *objp = index_to_obj(cachep, page, i);
if (cachep->flags & SLAB_POISON) {
check_poison_obj(cachep, objp);
slab_kernel_map(cachep, objp, 1);
}
if (cachep->flags & SLAB_RED_ZONE) {
if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
slab_error(cachep, "start of a freed object was overwritten");
if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
slab_error(cachep, "end of a freed object was overwritten");
}
}
}
#else
static void slab_destroy_debugcheck(struct kmem_cache *cachep,
struct page *page)
{
}
#endif
/**
* slab_destroy - destroy and release all objects in a slab
* @cachep: cache pointer being destroyed
* @page: page pointer being destroyed
*
* Destroy all the objs in a slab page, and release the mem back to the system.
* Before calling the slab page must have been unlinked from the cache. The
* kmem_cache_node ->list_lock is not held/needed.
*/
static void slab_destroy(struct kmem_cache *cachep, struct page *page)
{
void *freelist;
freelist = page->freelist;
slab_destroy_debugcheck(cachep, page);
if (unlikely(cachep->flags & SLAB_TYPESAFE_BY_RCU))
call_rcu(&page->rcu_head, kmem_rcu_free);
else
kmem_freepages(cachep, page);
/*
* From now on, we don't use freelist
* although actual page can be freed in rcu context
*/
if (OFF_SLAB(cachep))
kmem_cache_free(cachep->freelist_cache, freelist);
}
static void slabs_destroy(struct kmem_cache *cachep, struct list_head *list)
{
struct page *page, *n;
list_for_each_entry_safe(page, n, list, slab_list) {
list_del(&page->slab_list);
slab_destroy(cachep, page);
}
}
/**
* calculate_slab_order - calculate size (page order) of slabs
* @cachep: pointer to the cache that is being created
* @size: size of objects to be created in this cache.
* @flags: slab allocation flags
*
* Also calculates the number of objects per slab.
*
* This could be made much more intelligent. For now, try to avoid using
* high order pages for slabs. When the gfp() functions are more friendly
* towards high-order requests, this should be changed.
*
* Return: number of left-over bytes in a slab
*/
static size_t calculate_slab_order(struct kmem_cache *cachep,
size_t size, slab_flags_t flags)
{
size_t left_over = 0;
int gfporder;
for (gfporder = 0; gfporder <= KMALLOC_MAX_ORDER; gfporder++) {
unsigned int num;
size_t remainder;
num = cache_estimate(gfporder, size, flags, &remainder);
if (!num)
continue;
/* Can't handle number of objects more than SLAB_OBJ_MAX_NUM */
if (num > SLAB_OBJ_MAX_NUM)
break;
[PATCH] slab.c: fix offslab_limit bug mm/slab.c's offlab_limit logic is totally broken. Firstly, "offslab_limit" is a global variable while it should either be calculated in situ or should be passed in as a parameter. Secondly, the more serious problem with it is that the condition for calculating it: if (!(OFF_SLAB(sizes->cs_cachep))) { offslab_limit = sizes->cs_size - sizeof(struct slab); offslab_limit /= sizeof(kmem_bufctl_t); is in total disconnect with the condition that makes use of it: /* More than offslab_limit objects will cause problems */ if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) break; but due to offslab_limit being a global variable this breakage was hidden. Up until lockdep came along and perturbed the slab sizes sufficiently so that the first off-slab cache would still see a (non-calculated) zero value for offslab_limit and would panic with: kmem_cache_create: couldn't create cache size-512. Call Trace: [<ffffffff8020a5b9>] show_trace+0x96/0x1c8 [<ffffffff8020a8f0>] dump_stack+0x13/0x15 [<ffffffff8022994f>] panic+0x39/0x21a [<ffffffff80270814>] kmem_cache_create+0x5a0/0x5d0 [<ffffffff80aced62>] kmem_cache_init+0x193/0x379 [<ffffffff80abf779>] start_kernel+0x17f/0x218 [<ffffffff80abf263>] _sinittext+0x263/0x26a Kernel panic - not syncing: kmem_cache_create(): failed to create slab `size-512' Paolo Ornati's config on x86_64 managed to trigger it. The fix is to move the calculation to the place that makes use of it. This also makes slab.o 54 bytes smaller. Btw., the check itself is quite silly. Its intention is to test whether the number of objects per slab would be higher than the number of slab control pointers possible. In theory it could be triggered: if someone tried to allocate 4-byte objects cache and explicitly requested with CFLGS_OFF_SLAB. So i kept the check. Out of historic interest i checked how old this bug was and it's ancient, 10 years old! It is the oldest hidden and then truly triggering bugs i ever saw being fixed in the kernel! Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-02 21:44:58 +08:00
if (flags & CFLGS_OFF_SLAB) {
struct kmem_cache *freelist_cache;
size_t freelist_size;
freelist_size = num * sizeof(freelist_idx_t);
freelist_cache = kmalloc_slab(freelist_size, 0u);
if (!freelist_cache)
continue;
[PATCH] slab.c: fix offslab_limit bug mm/slab.c's offlab_limit logic is totally broken. Firstly, "offslab_limit" is a global variable while it should either be calculated in situ or should be passed in as a parameter. Secondly, the more serious problem with it is that the condition for calculating it: if (!(OFF_SLAB(sizes->cs_cachep))) { offslab_limit = sizes->cs_size - sizeof(struct slab); offslab_limit /= sizeof(kmem_bufctl_t); is in total disconnect with the condition that makes use of it: /* More than offslab_limit objects will cause problems */ if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) break; but due to offslab_limit being a global variable this breakage was hidden. Up until lockdep came along and perturbed the slab sizes sufficiently so that the first off-slab cache would still see a (non-calculated) zero value for offslab_limit and would panic with: kmem_cache_create: couldn't create cache size-512. Call Trace: [<ffffffff8020a5b9>] show_trace+0x96/0x1c8 [<ffffffff8020a8f0>] dump_stack+0x13/0x15 [<ffffffff8022994f>] panic+0x39/0x21a [<ffffffff80270814>] kmem_cache_create+0x5a0/0x5d0 [<ffffffff80aced62>] kmem_cache_init+0x193/0x379 [<ffffffff80abf779>] start_kernel+0x17f/0x218 [<ffffffff80abf263>] _sinittext+0x263/0x26a Kernel panic - not syncing: kmem_cache_create(): failed to create slab `size-512' Paolo Ornati's config on x86_64 managed to trigger it. The fix is to move the calculation to the place that makes use of it. This also makes slab.o 54 bytes smaller. Btw., the check itself is quite silly. Its intention is to test whether the number of objects per slab would be higher than the number of slab control pointers possible. In theory it could be triggered: if someone tried to allocate 4-byte objects cache and explicitly requested with CFLGS_OFF_SLAB. So i kept the check. Out of historic interest i checked how old this bug was and it's ancient, 10 years old! It is the oldest hidden and then truly triggering bugs i ever saw being fixed in the kernel! Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-02 21:44:58 +08:00
/*
* Needed to avoid possible looping condition
* in cache_grow_begin()
[PATCH] slab.c: fix offslab_limit bug mm/slab.c's offlab_limit logic is totally broken. Firstly, "offslab_limit" is a global variable while it should either be calculated in situ or should be passed in as a parameter. Secondly, the more serious problem with it is that the condition for calculating it: if (!(OFF_SLAB(sizes->cs_cachep))) { offslab_limit = sizes->cs_size - sizeof(struct slab); offslab_limit /= sizeof(kmem_bufctl_t); is in total disconnect with the condition that makes use of it: /* More than offslab_limit objects will cause problems */ if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) break; but due to offslab_limit being a global variable this breakage was hidden. Up until lockdep came along and perturbed the slab sizes sufficiently so that the first off-slab cache would still see a (non-calculated) zero value for offslab_limit and would panic with: kmem_cache_create: couldn't create cache size-512. Call Trace: [<ffffffff8020a5b9>] show_trace+0x96/0x1c8 [<ffffffff8020a8f0>] dump_stack+0x13/0x15 [<ffffffff8022994f>] panic+0x39/0x21a [<ffffffff80270814>] kmem_cache_create+0x5a0/0x5d0 [<ffffffff80aced62>] kmem_cache_init+0x193/0x379 [<ffffffff80abf779>] start_kernel+0x17f/0x218 [<ffffffff80abf263>] _sinittext+0x263/0x26a Kernel panic - not syncing: kmem_cache_create(): failed to create slab `size-512' Paolo Ornati's config on x86_64 managed to trigger it. The fix is to move the calculation to the place that makes use of it. This also makes slab.o 54 bytes smaller. Btw., the check itself is quite silly. Its intention is to test whether the number of objects per slab would be higher than the number of slab control pointers possible. In theory it could be triggered: if someone tried to allocate 4-byte objects cache and explicitly requested with CFLGS_OFF_SLAB. So i kept the check. Out of historic interest i checked how old this bug was and it's ancient, 10 years old! It is the oldest hidden and then truly triggering bugs i ever saw being fixed in the kernel! Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-02 21:44:58 +08:00
*/
if (OFF_SLAB(freelist_cache))
continue;
[PATCH] slab.c: fix offslab_limit bug mm/slab.c's offlab_limit logic is totally broken. Firstly, "offslab_limit" is a global variable while it should either be calculated in situ or should be passed in as a parameter. Secondly, the more serious problem with it is that the condition for calculating it: if (!(OFF_SLAB(sizes->cs_cachep))) { offslab_limit = sizes->cs_size - sizeof(struct slab); offslab_limit /= sizeof(kmem_bufctl_t); is in total disconnect with the condition that makes use of it: /* More than offslab_limit objects will cause problems */ if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) break; but due to offslab_limit being a global variable this breakage was hidden. Up until lockdep came along and perturbed the slab sizes sufficiently so that the first off-slab cache would still see a (non-calculated) zero value for offslab_limit and would panic with: kmem_cache_create: couldn't create cache size-512. Call Trace: [<ffffffff8020a5b9>] show_trace+0x96/0x1c8 [<ffffffff8020a8f0>] dump_stack+0x13/0x15 [<ffffffff8022994f>] panic+0x39/0x21a [<ffffffff80270814>] kmem_cache_create+0x5a0/0x5d0 [<ffffffff80aced62>] kmem_cache_init+0x193/0x379 [<ffffffff80abf779>] start_kernel+0x17f/0x218 [<ffffffff80abf263>] _sinittext+0x263/0x26a Kernel panic - not syncing: kmem_cache_create(): failed to create slab `size-512' Paolo Ornati's config on x86_64 managed to trigger it. The fix is to move the calculation to the place that makes use of it. This also makes slab.o 54 bytes smaller. Btw., the check itself is quite silly. Its intention is to test whether the number of objects per slab would be higher than the number of slab control pointers possible. In theory it could be triggered: if someone tried to allocate 4-byte objects cache and explicitly requested with CFLGS_OFF_SLAB. So i kept the check. Out of historic interest i checked how old this bug was and it's ancient, 10 years old! It is the oldest hidden and then truly triggering bugs i ever saw being fixed in the kernel! Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-02 21:44:58 +08:00
/* check if off slab has enough benefit */
if (freelist_cache->size > cachep->size / 2)
continue;
[PATCH] slab.c: fix offslab_limit bug mm/slab.c's offlab_limit logic is totally broken. Firstly, "offslab_limit" is a global variable while it should either be calculated in situ or should be passed in as a parameter. Secondly, the more serious problem with it is that the condition for calculating it: if (!(OFF_SLAB(sizes->cs_cachep))) { offslab_limit = sizes->cs_size - sizeof(struct slab); offslab_limit /= sizeof(kmem_bufctl_t); is in total disconnect with the condition that makes use of it: /* More than offslab_limit objects will cause problems */ if ((flags & CFLGS_OFF_SLAB) && num > offslab_limit) break; but due to offslab_limit being a global variable this breakage was hidden. Up until lockdep came along and perturbed the slab sizes sufficiently so that the first off-slab cache would still see a (non-calculated) zero value for offslab_limit and would panic with: kmem_cache_create: couldn't create cache size-512. Call Trace: [<ffffffff8020a5b9>] show_trace+0x96/0x1c8 [<ffffffff8020a8f0>] dump_stack+0x13/0x15 [<ffffffff8022994f>] panic+0x39/0x21a [<ffffffff80270814>] kmem_cache_create+0x5a0/0x5d0 [<ffffffff80aced62>] kmem_cache_init+0x193/0x379 [<ffffffff80abf779>] start_kernel+0x17f/0x218 [<ffffffff80abf263>] _sinittext+0x263/0x26a Kernel panic - not syncing: kmem_cache_create(): failed to create slab `size-512' Paolo Ornati's config on x86_64 managed to trigger it. The fix is to move the calculation to the place that makes use of it. This also makes slab.o 54 bytes smaller. Btw., the check itself is quite silly. Its intention is to test whether the number of objects per slab would be higher than the number of slab control pointers possible. In theory it could be triggered: if someone tried to allocate 4-byte objects cache and explicitly requested with CFLGS_OFF_SLAB. So i kept the check. Out of historic interest i checked how old this bug was and it's ancient, 10 years old! It is the oldest hidden and then truly triggering bugs i ever saw being fixed in the kernel! Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-02 21:44:58 +08:00
}
/* Found something acceptable - save it away */
cachep->num = num;
cachep->gfporder = gfporder;
left_over = remainder;
/*
* A VFS-reclaimable slab tends to have most allocations
* as GFP_NOFS and we really don't want to have to be allocating
* higher-order pages when we are unable to shrink dcache.
*/
if (flags & SLAB_RECLAIM_ACCOUNT)
break;
/*
* Large number of objects is good, but very large slabs are
* currently bad for the gfp()s.
*/
if (gfporder >= slab_max_order)
break;
/*
* Acceptable internal fragmentation?
*/
if (left_over * 8 <= (PAGE_SIZE << gfporder))
break;
}
return left_over;
}
static struct array_cache __percpu *alloc_kmem_cache_cpus(
struct kmem_cache *cachep, int entries, int batchcount)
{
int cpu;
size_t size;
struct array_cache __percpu *cpu_cache;
size = sizeof(void *) * entries + sizeof(struct array_cache);
cpu_cache = __alloc_percpu(size, sizeof(void *));
if (!cpu_cache)
return NULL;
for_each_possible_cpu(cpu) {
init_arraycache(per_cpu_ptr(cpu_cache, cpu),
entries, batchcount);
}
return cpu_cache;
}
static int __ref setup_cpu_cache(struct kmem_cache *cachep, gfp_t gfp)
{
if (slab_state >= FULL)
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
return enable_cpucache(cachep, gfp);
cachep->cpu_cache = alloc_kmem_cache_cpus(cachep, 1, 1);
if (!cachep->cpu_cache)
return 1;
if (slab_state == DOWN) {
/* Creation of first cache (kmem_cache). */
set_up_node(kmem_cache, CACHE_CACHE);
} else if (slab_state == PARTIAL) {
/* For kmem_cache_node */
set_up_node(cachep, SIZE_NODE);
} else {
int node;
for_each_online_node(node) {
cachep->node[node] = kmalloc_node(
sizeof(struct kmem_cache_node), gfp, node);
BUG_ON(!cachep->node[node]);
kmem_cache_node_init(cachep->node[node]);
}
}
cachep->node[numa_mem_id()]->next_reap =
jiffies + REAPTIMEOUT_NODE +
((unsigned long)cachep) % REAPTIMEOUT_NODE;
cpu_cache_get(cachep)->avail = 0;
cpu_cache_get(cachep)->limit = BOOT_CPUCACHE_ENTRIES;
cpu_cache_get(cachep)->batchcount = 1;
cpu_cache_get(cachep)->touched = 0;
cachep->batchcount = 1;
cachep->limit = BOOT_CPUCACHE_ENTRIES;
return 0;
}
slab_flags_t kmem_cache_flags(unsigned int object_size,
slab_flags_t flags, const char *name,
void (*ctor)(void *))
{
return flags;
}
struct kmem_cache *
__kmem_cache_alias(const char *name, unsigned int size, unsigned int align,
slab_flags_t flags, void (*ctor)(void *))
{
struct kmem_cache *cachep;
cachep = find_mergeable(size, align, flags, name, ctor);
if (cachep) {
cachep->refcount++;
/*
* Adjust the object sizes so that we clear
* the complete object on kzalloc.
*/
cachep->object_size = max_t(int, cachep->object_size, size);
}
return cachep;
}
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
static bool set_objfreelist_slab_cache(struct kmem_cache *cachep,
size_t size, slab_flags_t flags)
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
{
size_t left;
cachep->num = 0;
mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options Patch series "add init_on_alloc/init_on_free boot options", v10. Provide init_on_alloc and init_on_free boot options. These are aimed at preventing possible information leaks and making the control-flow bugs that depend on uninitialized values more deterministic. Enabling either of the options guarantees that the memory returned by the page allocator and SL[AU]B is initialized with zeroes. SLOB allocator isn't supported at the moment, as its emulation of kmem caches complicates handling of SLAB_TYPESAFE_BY_RCU caches correctly. Enabling init_on_free also guarantees that pages and heap objects are initialized right after they're freed, so it won't be possible to access stale data by using a dangling pointer. As suggested by Michal Hocko, right now we don't let the heap users to disable initialization for certain allocations. There's not enough evidence that doing so can speed up real-life cases, and introducing ways to opt-out may result in things going out of control. This patch (of 2): The new options are needed to prevent possible information leaks and make control-flow bugs that depend on uninitialized values more deterministic. This is expected to be on-by-default on Android and Chrome OS. And it gives the opportunity for anyone else to use it under distros too via the boot args. (The init_on_free feature is regularly requested by folks where memory forensics is included in their threat models.) init_on_alloc=1 makes the kernel initialize newly allocated pages and heap objects with zeroes. Initialization is done at allocation time at the places where checks for __GFP_ZERO are performed. init_on_free=1 makes the kernel initialize freed pages and heap objects with zeroes upon their deletion. This helps to ensure sensitive data doesn't leak via use-after-free accesses. Both init_on_alloc=1 and init_on_free=1 guarantee that the allocator returns zeroed memory. The two exceptions are slab caches with constructors and SLAB_TYPESAFE_BY_RCU flag. Those are never zero-initialized to preserve their semantics. Both init_on_alloc and init_on_free default to zero, but those defaults can be overridden with CONFIG_INIT_ON_ALLOC_DEFAULT_ON and CONFIG_INIT_ON_FREE_DEFAULT_ON. If either SLUB poisoning or page poisoning is enabled, those options take precedence over init_on_alloc and init_on_free: initialization is only applied to unpoisoned allocations. Slowdown for the new features compared to init_on_free=0, init_on_alloc=0: hackbench, init_on_free=1: +7.62% sys time (st.err 0.74%) hackbench, init_on_alloc=1: +7.75% sys time (st.err 2.14%) Linux build with -j12, init_on_free=1: +8.38% wall time (st.err 0.39%) Linux build with -j12, init_on_free=1: +24.42% sys time (st.err 0.52%) Linux build with -j12, init_on_alloc=1: -0.13% wall time (st.err 0.42%) Linux build with -j12, init_on_alloc=1: +0.57% sys time (st.err 0.40%) The slowdown for init_on_free=0, init_on_alloc=0 compared to the baseline is within the standard error. The new features are also going to pave the way for hardware memory tagging (e.g. arm64's MTE), which will require both on_alloc and on_free hooks to set the tags for heap objects. With MTE, tagging will have the same cost as memory initialization. Although init_on_free is rather costly, there are paranoid use-cases where in-memory data lifetime is desired to be minimized. There are various arguments for/against the realism of the associated threat models, but given that we'll need the infrastructure for MTE anyway, and there are people who want wipe-on-free behavior no matter what the performance cost, it seems reasonable to include it in this series. [glider@google.com: v8] Link: http://lkml.kernel.org/r/20190626121943.131390-2-glider@google.com [glider@google.com: v9] Link: http://lkml.kernel.org/r/20190627130316.254309-2-glider@google.com [glider@google.com: v10] Link: http://lkml.kernel.org/r/20190628093131.199499-2-glider@google.com Link: http://lkml.kernel.org/r/20190617151050.92663-2-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Acked-by: Kees Cook <keescook@chromium.org> Acked-by: Michal Hocko <mhocko@suse.cz> [page and dmapool parts Acked-by: James Morris <jamorris@linux.microsoft.com>] Cc: Christoph Lameter <cl@linux.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: "Serge E. Hallyn" <serge@hallyn.com> Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Kostya Serebryany <kcc@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Sandeep Patil <sspatil@android.com> Cc: Laura Abbott <labbott@redhat.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Jann Horn <jannh@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marco Elver <elver@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:59:19 +08:00
/*
* If slab auto-initialization on free is enabled, store the freelist
* off-slab, so that its contents don't end up in one of the allocated
* objects.
*/
if (unlikely(slab_want_init_on_free(cachep)))
return false;
if (cachep->ctor || flags & SLAB_TYPESAFE_BY_RCU)
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
return false;
left = calculate_slab_order(cachep, size,
flags | CFLGS_OBJFREELIST_SLAB);
if (!cachep->num)
return false;
if (cachep->num * sizeof(freelist_idx_t) > cachep->object_size)
return false;
cachep->colour = left / cachep->colour_off;
return true;
}
static bool set_off_slab_cache(struct kmem_cache *cachep,
size_t size, slab_flags_t flags)
{
size_t left;
cachep->num = 0;
/*
* Always use on-slab management when SLAB_NOLEAKTRACE
* to avoid recursive calls into kmemleak.
*/
if (flags & SLAB_NOLEAKTRACE)
return false;
/*
* Size is large, assume best to place the slab management obj
* off-slab (should allow better packing of objs).
*/
left = calculate_slab_order(cachep, size, flags | CFLGS_OFF_SLAB);
if (!cachep->num)
return false;
/*
* If the slab has been placed off-slab, and we have enough space then
* move it on-slab. This is at the expense of any extra colouring.
*/
if (left >= cachep->num * sizeof(freelist_idx_t))
return false;
cachep->colour = left / cachep->colour_off;
return true;
}
static bool set_on_slab_cache(struct kmem_cache *cachep,
size_t size, slab_flags_t flags)
{
size_t left;
cachep->num = 0;
left = calculate_slab_order(cachep, size, flags);
if (!cachep->num)
return false;
cachep->colour = left / cachep->colour_off;
return true;
}
/**
* __kmem_cache_create - Create a cache.
* @cachep: cache management descriptor
* @flags: SLAB flags
*
* Returns a ptr to the cache on success, NULL on failure.
* Cannot be called within a int, but can be interrupted.
* The @ctor is run when new pages are allocated by the cache.
*
* The flags are
*
* %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
* to catch references to uninitialised memory.
*
* %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
* for buffer overruns.
*
* %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
* cacheline. This can be beneficial if you're counting cycles as closely
* as davem.
*
* Return: a pointer to the created cache or %NULL in case of error
*/
int __kmem_cache_create(struct kmem_cache *cachep, slab_flags_t flags)
{
size_t ralign = BYTES_PER_WORD;
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
gfp_t gfp;
int err;
unsigned int size = cachep->size;
#if DEBUG
#if FORCED_DEBUG
/*
* Enable redzoning and last user accounting, except for caches with
* large objects, if the increased size would increase the object size
* above the next power of two: caches with object sizes just above a
* power of two have a significant amount of internal fragmentation.
*/
if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
2 * sizeof(unsigned long long)))
flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
if (!(flags & SLAB_TYPESAFE_BY_RCU))
flags |= SLAB_POISON;
#endif
#endif
/*
* Check that size is in terms of words. This is needed to avoid
* unaligned accesses for some archs when redzoning is used, and makes
* sure any on-slab bufctl's are also correctly aligned.
*/
size = ALIGN(size, BYTES_PER_WORD);
if (flags & SLAB_RED_ZONE) {
ralign = REDZONE_ALIGN;
/* If redzoning, ensure that the second redzone is suitably
* aligned, by adjusting the object size accordingly. */
size = ALIGN(size, REDZONE_ALIGN);
}
/* 3) caller mandated alignment */
if (ralign < cachep->align) {
ralign = cachep->align;
}
Revert "slab: Fix missing DEBUG_SLAB last user" This reverts commit 5c5e3b33b7cb959a401f823707bee006caadd76e. The commit breaks ARM thusly: | Mount-cache hash table entries: 512 | slab error in verify_redzone_free(): cache `idr_layer_cache': memory outside object was overwritten | Backtrace: | [<c0227088>] (dump_backtrace+0x0/0x110) from [<c0431afc>] (dump_stack+0x18/0x1c) | [<c0431ae4>] (dump_stack+0x0/0x1c) from [<c0293304>] (__slab_error+0x28/0x30) | [<c02932dc>] (__slab_error+0x0/0x30) from [<c0293a74>] (cache_free_debugcheck+0x1c0/0x2b8) | [<c02938b4>] (cache_free_debugcheck+0x0/0x2b8) from [<c0293f78>] (kmem_cache_free+0x3c/0xc0) | [<c0293f3c>] (kmem_cache_free+0x0/0xc0) from [<c032b1c8>] (ida_get_new_above+0x19c/0x1c0) | [<c032b02c>] (ida_get_new_above+0x0/0x1c0) from [<c02af7ec>] (alloc_vfsmnt+0x54/0x144) | [<c02af798>] (alloc_vfsmnt+0x0/0x144) from [<c0299830>] (vfs_kern_mount+0x30/0xec) | [<c0299800>] (vfs_kern_mount+0x0/0xec) from [<c0299908>] (kern_mount_data+0x1c/0x20) | [<c02998ec>] (kern_mount_data+0x0/0x20) from [<c02146c4>] (sysfs_init+0x68/0xc8) | [<c021465c>] (sysfs_init+0x0/0xc8) from [<c02137d4>] (mnt_init+0x90/0x1b0) | [<c0213744>] (mnt_init+0x0/0x1b0) from [<c0213388>] (vfs_caches_init+0x100/0x140) | [<c0213288>] (vfs_caches_init+0x0/0x140) from [<c0208c0c>] (start_kernel+0x2e8/0x368) | [<c0208924>] (start_kernel+0x0/0x368) from [<c0208034>] (__enable_mmu+0x0/0x2c) | c0113268: redzone 1:0xd84156c5c032b3ac, redzone 2:0xd84156c5635688c0. | slab error in cache_alloc_debugcheck_after(): cache `idr_layer_cache': double free, or memory outside object was overwritten | ... | c011307c: redzone 1:0x9f91102ffffffff, redzone 2:0x9f911029d74e35b | slab: Internal list corruption detected in cache 'idr_layer_cache'(24), slabp c0113000(16). Hexdump: | | 000: 20 4f 10 c0 20 4f 10 c0 7c 00 00 00 7c 30 11 c0 | 010: 10 00 00 00 10 00 00 00 00 00 c9 17 fe ff ff ff | 020: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff | 030: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff | 040: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff | 050: fe ff ff ff fe ff ff ff fe ff ff ff 11 00 00 00 | 060: 12 00 00 00 13 00 00 00 14 00 00 00 15 00 00 00 | 070: 16 00 00 00 17 00 00 00 c0 88 56 63 | kernel BUG at /home/rmk/git/linux-2.6-rmk/mm/slab.c:2928! Reference: https://lkml.org/lkml/2011/2/7/238 Cc: <stable@kernel.org> # 2.6.35.y and later Reported-and-analyzed-by: Russell King <rmk@arm.linux.org.uk> Signed-off-by: Pekka Enberg <penberg@kernel.org>
2011-02-14 23:46:21 +08:00
/* disable debug if necessary */
if (ralign > __alignof__(unsigned long long))
flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
/*
* 4) Store it.
*/
cachep->align = ralign;
cachep->colour_off = cache_line_size();
/* Offset must be a multiple of the alignment. */
if (cachep->colour_off < cachep->align)
cachep->colour_off = cachep->align;
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
if (slab_is_available())
gfp = GFP_KERNEL;
else
gfp = GFP_NOWAIT;
#if DEBUG
/*
* Both debugging options require word-alignment which is calculated
* into align above.
*/
if (flags & SLAB_RED_ZONE) {
/* add space for red zone words */
Revert "slab: Fix missing DEBUG_SLAB last user" This reverts commit 5c5e3b33b7cb959a401f823707bee006caadd76e. The commit breaks ARM thusly: | Mount-cache hash table entries: 512 | slab error in verify_redzone_free(): cache `idr_layer_cache': memory outside object was overwritten | Backtrace: | [<c0227088>] (dump_backtrace+0x0/0x110) from [<c0431afc>] (dump_stack+0x18/0x1c) | [<c0431ae4>] (dump_stack+0x0/0x1c) from [<c0293304>] (__slab_error+0x28/0x30) | [<c02932dc>] (__slab_error+0x0/0x30) from [<c0293a74>] (cache_free_debugcheck+0x1c0/0x2b8) | [<c02938b4>] (cache_free_debugcheck+0x0/0x2b8) from [<c0293f78>] (kmem_cache_free+0x3c/0xc0) | [<c0293f3c>] (kmem_cache_free+0x0/0xc0) from [<c032b1c8>] (ida_get_new_above+0x19c/0x1c0) | [<c032b02c>] (ida_get_new_above+0x0/0x1c0) from [<c02af7ec>] (alloc_vfsmnt+0x54/0x144) | [<c02af798>] (alloc_vfsmnt+0x0/0x144) from [<c0299830>] (vfs_kern_mount+0x30/0xec) | [<c0299800>] (vfs_kern_mount+0x0/0xec) from [<c0299908>] (kern_mount_data+0x1c/0x20) | [<c02998ec>] (kern_mount_data+0x0/0x20) from [<c02146c4>] (sysfs_init+0x68/0xc8) | [<c021465c>] (sysfs_init+0x0/0xc8) from [<c02137d4>] (mnt_init+0x90/0x1b0) | [<c0213744>] (mnt_init+0x0/0x1b0) from [<c0213388>] (vfs_caches_init+0x100/0x140) | [<c0213288>] (vfs_caches_init+0x0/0x140) from [<c0208c0c>] (start_kernel+0x2e8/0x368) | [<c0208924>] (start_kernel+0x0/0x368) from [<c0208034>] (__enable_mmu+0x0/0x2c) | c0113268: redzone 1:0xd84156c5c032b3ac, redzone 2:0xd84156c5635688c0. | slab error in cache_alloc_debugcheck_after(): cache `idr_layer_cache': double free, or memory outside object was overwritten | ... | c011307c: redzone 1:0x9f91102ffffffff, redzone 2:0x9f911029d74e35b | slab: Internal list corruption detected in cache 'idr_layer_cache'(24), slabp c0113000(16). Hexdump: | | 000: 20 4f 10 c0 20 4f 10 c0 7c 00 00 00 7c 30 11 c0 | 010: 10 00 00 00 10 00 00 00 00 00 c9 17 fe ff ff ff | 020: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff | 030: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff | 040: fe ff ff ff fe ff ff ff fe ff ff ff fe ff ff ff | 050: fe ff ff ff fe ff ff ff fe ff ff ff 11 00 00 00 | 060: 12 00 00 00 13 00 00 00 14 00 00 00 15 00 00 00 | 070: 16 00 00 00 17 00 00 00 c0 88 56 63 | kernel BUG at /home/rmk/git/linux-2.6-rmk/mm/slab.c:2928! Reference: https://lkml.org/lkml/2011/2/7/238 Cc: <stable@kernel.org> # 2.6.35.y and later Reported-and-analyzed-by: Russell King <rmk@arm.linux.org.uk> Signed-off-by: Pekka Enberg <penberg@kernel.org>
2011-02-14 23:46:21 +08:00
cachep->obj_offset += sizeof(unsigned long long);
size += 2 * sizeof(unsigned long long);
}
if (flags & SLAB_STORE_USER) {
/* user store requires one word storage behind the end of
* the real object. But if the second red zone needs to be
* aligned to 64 bits, we must allow that much space.
*/
if (flags & SLAB_RED_ZONE)
size += REDZONE_ALIGN;
else
size += BYTES_PER_WORD;
}
#endif
kasan_cache_create(cachep, &size, &flags);
size = ALIGN(size, cachep->align);
/*
* We should restrict the number of objects in a slab to implement
* byte sized index. Refer comment on SLAB_OBJ_MIN_SIZE definition.
*/
if (FREELIST_BYTE_INDEX && size < SLAB_OBJ_MIN_SIZE)
size = ALIGN(SLAB_OBJ_MIN_SIZE, cachep->align);
#if DEBUG
mm/slab: fix unexpected index mapping result of kmalloc_size(INDEX_NODE+1) Commit description is copied from the original post of this bug: http://comments.gmane.org/gmane.linux.kernel.mm/135349 Kernels after v3.9 use kmalloc_size(INDEX_NODE + 1) to get the next larger cache size than the size index INDEX_NODE mapping. In kernels 3.9 and earlier we used malloc_sizes[INDEX_L3 + 1].cs_size. However, sometimes we can't get the right output we expected via kmalloc_size(INDEX_NODE + 1), causing a BUG(). The mapping table in the latest kernel is like: index = {0, 1, 2 , 3, 4, 5, 6, n} size = {0, 96, 192, 8, 16, 32, 64, 2^n} The mapping table before 3.10 is like this: index = {0 , 1 , 2, 3, 4 , 5 , 6, n} size = {32, 64, 96, 128, 192, 256, 512, 2^(n+3)} The problem on my mips64 machine is as follows: (1) When configured DEBUG_SLAB && DEBUG_PAGEALLOC && DEBUG_LOCK_ALLOC && DEBUG_SPINLOCK, the sizeof(struct kmem_cache_node) will be "150", and the macro INDEX_NODE turns out to be "2": #define INDEX_NODE kmalloc_index(sizeof(struct kmem_cache_node)) (2) Then the result of kmalloc_size(INDEX_NODE + 1) is 8. (3) Then "if(size >= kmalloc_size(INDEX_NODE + 1)" will lead to "size = PAGE_SIZE". (4) Then "if ((size >= (PAGE_SIZE >> 3))" test will be satisfied and "flags |= CFLGS_OFF_SLAB" will be covered. (5) if (flags & CFLGS_OFF_SLAB)" test will be satisfied and will go to "cachep->slabp_cache = kmalloc_slab(slab_size, 0u)", and the result here may be NULL while kernel bootup. (6) Finally,"BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache));" causes the BUG info as the following shows (may be only mips64 has this problem): This patch fixes the problem of kmalloc_size(INDEX_NODE + 1) and removes the BUG by adding 'size >= 256' check to guarantee that all necessary small sized slabs are initialized regardless sequence of slab size in mapping table. Fixes: e33660165c90 ("slab: Use common kmalloc_index/kmalloc_size...") Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Reported-by: Liuhailong <liu.hailong6@zte.com.cn> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-10-02 06:36:54 +08:00
/*
* To activate debug pagealloc, off-slab management is necessary
* requirement. In early phase of initialization, small sized slab
* doesn't get initialized so it would not be possible. So, we need
* to check size >= 256. It guarantees that all necessary small
* sized slab is initialized in current slab initialization sequence.
*/
mm, debug_pagealloc: don't rely on static keys too early Commit 96a2b03f281d ("mm, debug_pagelloc: use static keys to enable debugging") has introduced a static key to reduce overhead when debug_pagealloc is compiled in but not enabled. It relied on the assumption that jump_label_init() is called before parse_early_param() as in start_kernel(), so when the "debug_pagealloc=on" option is parsed, it is safe to enable the static key. However, it turns out multiple architectures call parse_early_param() earlier from their setup_arch(). x86 also calls jump_label_init() even earlier, so no issue was found while testing the commit, but same is not true for e.g. ppc64 and s390 where the kernel would not boot with debug_pagealloc=on as found by our QA. To fix this without tricky changes to init code of multiple architectures, this patch partially reverts the static key conversion from 96a2b03f281d. Init-time and non-fastpath calls (such as in arch code) of debug_pagealloc_enabled() will again test a simple bool variable. Fastpath mm code is converted to a new debug_pagealloc_enabled_static() variant that relies on the static key, which is enabled in a well-defined point in mm_init() where it's guaranteed that jump_label_init() has been called, regardless of architecture. [sfr@canb.auug.org.au: export _debug_pagealloc_enabled_early] Link: http://lkml.kernel.org/r/20200106164944.063ac07b@canb.auug.org.au Link: http://lkml.kernel.org/r/20191219130612.23171-1-vbabka@suse.cz Fixes: 96a2b03f281d ("mm, debug_pagelloc: use static keys to enable debugging") Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Matthew Wilcox <willy@infradead.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Qian Cai <cai@lca.pw> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-01-14 08:29:20 +08:00
if (debug_pagealloc_enabled_static() && (flags & SLAB_POISON) &&
size >= 256 && cachep->object_size > cache_line_size()) {
if (size < PAGE_SIZE || size % PAGE_SIZE == 0) {
size_t tmp_size = ALIGN(size, PAGE_SIZE);
if (set_off_slab_cache(cachep, tmp_size, flags)) {
flags |= CFLGS_OFF_SLAB;
cachep->obj_offset += tmp_size - size;
size = tmp_size;
goto done;
}
}
}
#endif
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (set_objfreelist_slab_cache(cachep, size, flags)) {
flags |= CFLGS_OBJFREELIST_SLAB;
goto done;
}
if (set_off_slab_cache(cachep, size, flags)) {
flags |= CFLGS_OFF_SLAB;
goto done;
}
if (set_on_slab_cache(cachep, size, flags))
goto done;
return -E2BIG;
done:
cachep->freelist_size = cachep->num * sizeof(freelist_idx_t);
cachep->flags = flags;
cachep->allocflags = __GFP_COMP;
if (flags & SLAB_CACHE_DMA)
cachep->allocflags |= GFP_DMA;
mm: add support for kmem caches in DMA32 zone Patch series "iommu/io-pgtable-arm-v7s: Use DMA32 zone for page tables", v6. This is a followup to the discussion in [1], [2]. IOMMUs using ARMv7 short-descriptor format require page tables (level 1 and 2) to be allocated within the first 4GB of RAM, even on 64-bit systems. For L1 tables that are bigger than a page, we can just use __get_free_pages with GFP_DMA32 (on arm64 systems only, arm would still use GFP_DMA). For L2 tables that only take 1KB, it would be a waste to allocate a full page, so we considered 3 approaches: 1. This series, adding support for GFP_DMA32 slab caches. 2. genalloc, which requires pre-allocating the maximum number of L2 page tables (4096, so 4MB of memory). 3. page_frag, which is not very memory-efficient as it is unable to reuse freed fragments until the whole page is freed. [3] This series is the most memory-efficient approach. stable@ note: We confirmed that this is a regression, and IOMMU errors happen on 4.19 and linux-next/master on MT8173 (elm, Acer Chromebook R13). The issue most likely starts from commit ad67f5a6545f ("arm64: replace ZONE_DMA with ZONE_DMA32"), i.e. 4.15, and presumably breaks a number of Mediatek platforms (and maybe others?). [1] https://lists.linuxfoundation.org/pipermail/iommu/2018-November/030876.html [2] https://lists.linuxfoundation.org/pipermail/iommu/2018-December/031696.html [3] https://patchwork.codeaurora.org/patch/671639/ This patch (of 3): IOMMUs using ARMv7 short-descriptor format require page tables to be allocated within the first 4GB of RAM, even on 64-bit systems. On arm64, this is done by passing GFP_DMA32 flag to memory allocation functions. For IOMMU L2 tables that only take 1KB, it would be a waste to allocate a full page using get_free_pages, so we considered 3 approaches: 1. This patch, adding support for GFP_DMA32 slab caches. 2. genalloc, which requires pre-allocating the maximum number of L2 page tables (4096, so 4MB of memory). 3. page_frag, which is not very memory-efficient as it is unable to reuse freed fragments until the whole page is freed. This change makes it possible to create a custom cache in DMA32 zone using kmem_cache_create, then allocate memory using kmem_cache_alloc. We do not create a DMA32 kmalloc cache array, as there are currently no users of kmalloc(..., GFP_DMA32). These calls will continue to trigger a warning, as we keep GFP_DMA32 in GFP_SLAB_BUG_MASK. This implies that calls to kmem_cache_*alloc on a SLAB_CACHE_DMA32 kmem_cache must _not_ use GFP_DMA32 (it is anyway redundant and unnecessary). Link: http://lkml.kernel.org/r/20181210011504.122604-2-drinkcat@chromium.org Signed-off-by: Nicolas Boichat <drinkcat@chromium.org> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Will Deacon <will.deacon@arm.com> Cc: Robin Murphy <robin.murphy@arm.com> Cc: Joerg Roedel <joro@8bytes.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Sasha Levin <Alexander.Levin@microsoft.com> Cc: Huaisheng Ye <yehs1@lenovo.com> Cc: Mike Rapoport <rppt@linux.vnet.ibm.com> Cc: Yong Wu <yong.wu@mediatek.com> Cc: Matthias Brugger <matthias.bgg@gmail.com> Cc: Tomasz Figa <tfiga@google.com> Cc: Yingjoe Chen <yingjoe.chen@mediatek.com> Cc: Christoph Hellwig <hch@infradead.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Hsin-Yi Wang <hsinyi@chromium.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-03-29 11:43:42 +08:00
if (flags & SLAB_CACHE_DMA32)
cachep->allocflags |= GFP_DMA32;
if (flags & SLAB_RECLAIM_ACCOUNT)
cachep->allocflags |= __GFP_RECLAIMABLE;
cachep->size = size;
2006-12-13 16:34:27 +08:00
cachep->reciprocal_buffer_size = reciprocal_value(size);
#if DEBUG
/*
* If we're going to use the generic kernel_map_pages()
* poisoning, then it's going to smash the contents of
* the redzone and userword anyhow, so switch them off.
*/
if (IS_ENABLED(CONFIG_PAGE_POISONING) &&
(cachep->flags & SLAB_POISON) &&
is_debug_pagealloc_cache(cachep))
cachep->flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
#endif
if (OFF_SLAB(cachep)) {
cachep->freelist_cache =
kmalloc_slab(cachep->freelist_size, 0u);
}
err = setup_cpu_cache(cachep, gfp);
if (err) {
mm: slab: free kmem_cache_node after destroy sysfs file When slub_debug alloc_calls_show is enabled we will try to track location and user of slab object on each online node, kmem_cache_node structure and cpu_cache/cpu_slub shouldn't be freed till there is the last reference to sysfs file. This fixes the following panic: BUG: unable to handle kernel NULL pointer dereference at 0000000000000020 IP: list_locations+0x169/0x4e0 PGD 257304067 PUD 438456067 PMD 0 Oops: 0000 [#1] SMP CPU: 3 PID: 973074 Comm: cat ve: 0 Not tainted 3.10.0-229.7.2.ovz.9.30-00007-japdoll-dirty #2 9.30 Hardware name: DEPO Computers To Be Filled By O.E.M./H67DE3, BIOS L1.60c 07/14/2011 task: ffff88042a5dc5b0 ti: ffff88037f8d8000 task.ti: ffff88037f8d8000 RIP: list_locations+0x169/0x4e0 Call Trace: alloc_calls_show+0x1d/0x30 slab_attr_show+0x1b/0x30 sysfs_read_file+0x9a/0x1a0 vfs_read+0x9c/0x170 SyS_read+0x58/0xb0 system_call_fastpath+0x16/0x1b Code: 5e 07 12 00 b9 00 04 00 00 3d 00 04 00 00 0f 4f c1 3d 00 04 00 00 89 45 b0 0f 84 c3 00 00 00 48 63 45 b0 49 8b 9c c4 f8 00 00 00 <48> 8b 43 20 48 85 c0 74 b6 48 89 df e8 46 37 44 00 48 8b 53 10 CR2: 0000000000000020 Separated __kmem_cache_release from __kmem_cache_shutdown which now called on slab_kmem_cache_release (after the last reference to sysfs file object has dropped). Reintroduced locking in free_partial as sysfs file might access cache's partial list after shutdowning - partial revert of the commit 69cb8e6b7c29 ("slub: free slabs without holding locks"). Zap __remove_partial and use remove_partial (w/o underscores) as free_partial now takes list_lock which s partial revert for commit 1e4dd9461fab ("slub: do not assert not having lock in removing freed partial") Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com> Suggested-by: Vladimir Davydov <vdavydov@virtuozzo.com> Acked-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-02-18 05:11:37 +08:00
__kmem_cache_release(cachep);
return err;
}
return 0;
}
#if DEBUG
static void check_irq_off(void)
{
BUG_ON(!irqs_disabled());
}
static void check_irq_on(void)
{
BUG_ON(irqs_disabled());
}
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
static void check_mutex_acquired(void)
{
BUG_ON(!mutex_is_locked(&slab_mutex));
}
static void check_spinlock_acquired(struct kmem_cache *cachep)
{
#ifdef CONFIG_SMP
check_irq_off();
assert_spin_locked(&get_node(cachep, numa_mem_id())->list_lock);
#endif
}
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
static void check_spinlock_acquired_node(struct kmem_cache *cachep, int node)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
#ifdef CONFIG_SMP
check_irq_off();
assert_spin_locked(&get_node(cachep, node)->list_lock);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#endif
}
#else
#define check_irq_off() do { } while(0)
#define check_irq_on() do { } while(0)
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
#define check_mutex_acquired() do { } while(0)
#define check_spinlock_acquired(x) do { } while(0)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#define check_spinlock_acquired_node(x, y) do { } while(0)
#endif
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
static void drain_array_locked(struct kmem_cache *cachep, struct array_cache *ac,
int node, bool free_all, struct list_head *list)
{
int tofree;
if (!ac || !ac->avail)
return;
tofree = free_all ? ac->avail : (ac->limit + 4) / 5;
if (tofree > ac->avail)
tofree = (ac->avail + 1) / 2;
free_block(cachep, ac->entry, tofree, node, list);
ac->avail -= tofree;
memmove(ac->entry, &(ac->entry[tofree]), sizeof(void *) * ac->avail);
}
static void do_drain(void *arg)
{
struct kmem_cache *cachep = arg;
struct array_cache *ac;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
int node = numa_mem_id();
struct kmem_cache_node *n;
LIST_HEAD(list);
check_irq_off();
ac = cpu_cache_get(cachep);
n = get_node(cachep, node);
spin_lock(&n->list_lock);
free_block(cachep, ac->entry, ac->avail, node, &list);
spin_unlock(&n->list_lock);
slabs_destroy(cachep, &list);
ac->avail = 0;
}
static void drain_cpu_caches(struct kmem_cache *cachep)
{
struct kmem_cache_node *n;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
int node;
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
LIST_HEAD(list);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
on_each_cpu(do_drain, cachep, 1);
check_irq_on();
for_each_kmem_cache_node(cachep, node, n)
if (n->alien)
drain_alien_cache(cachep, n->alien);
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
for_each_kmem_cache_node(cachep, node, n) {
spin_lock_irq(&n->list_lock);
drain_array_locked(cachep, n->shared, node, true, &list);
spin_unlock_irq(&n->list_lock);
slabs_destroy(cachep, &list);
}
}
/*
* Remove slabs from the list of free slabs.
* Specify the number of slabs to drain in tofree.
*
* Returns the actual number of slabs released.
*/
static int drain_freelist(struct kmem_cache *cache,
struct kmem_cache_node *n, int tofree)
{
struct list_head *p;
int nr_freed;
struct page *page;
nr_freed = 0;
while (nr_freed < tofree && !list_empty(&n->slabs_free)) {
spin_lock_irq(&n->list_lock);
p = n->slabs_free.prev;
if (p == &n->slabs_free) {
spin_unlock_irq(&n->list_lock);
goto out;
}
page = list_entry(p, struct page, slab_list);
list_del(&page->slab_list);
n->free_slabs--;
n->total_slabs--;
/*
* Safe to drop the lock. The slab is no longer linked
* to the cache.
*/
n->free_objects -= cache->num;
spin_unlock_irq(&n->list_lock);
slab_destroy(cache, page);
nr_freed++;
}
out:
return nr_freed;
}
slab, slub: skip unnecessary kasan_cache_shutdown() The kasan quarantine is designed to delay freeing slab objects to catch use-after-free. The quarantine can be large (several percent of machine memory size). When kmem_caches are deleted related objects are flushed from the quarantine but this requires scanning the entire quarantine which can be very slow. We have seen the kernel busily working on this while holding slab_mutex and badly affecting cache_reaper, slabinfo readers and memcg kmem cache creations. It can easily reproduced by following script: yes . | head -1000000 | xargs stat > /dev/null for i in `seq 1 10`; do seq 500 | (cd /cg/memory && xargs mkdir) seq 500 | xargs -I{} sh -c 'echo $BASHPID > \ /cg/memory/{}/tasks && exec stat .' > /dev/null seq 500 | (cd /cg/memory && xargs rmdir) done The busy stack: kasan_cache_shutdown shutdown_cache memcg_destroy_kmem_caches mem_cgroup_css_free css_free_rwork_fn process_one_work worker_thread kthread ret_from_fork This patch is based on the observation that if the kmem_cache to be destroyed is empty then there should not be any objects of this cache in the quarantine. Without the patch the script got stuck for couple of hours. With the patch the script completed within a second. Link: http://lkml.kernel.org/r/20180327230603.54721-1-shakeelb@google.com Signed-off-by: Shakeel Butt <shakeelb@google.com> Reviewed-by: Andrew Morton <akpm@linux-foundation.org> Acked-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Vladimir Davydov <vdavydov.dev@gmail.com> Cc: Alexander Potapenko <glider@google.com> Cc: Greg Thelen <gthelen@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-06 07:21:57 +08:00
bool __kmem_cache_empty(struct kmem_cache *s)
{
int node;
struct kmem_cache_node *n;
for_each_kmem_cache_node(s, node, n)
if (!list_empty(&n->slabs_full) ||
!list_empty(&n->slabs_partial))
return false;
return true;
}
int __kmem_cache_shrink(struct kmem_cache *cachep)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
int ret = 0;
int node;
struct kmem_cache_node *n;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
drain_cpu_caches(cachep);
check_irq_on();
for_each_kmem_cache_node(cachep, node, n) {
drain_freelist(cachep, n, INT_MAX);
ret += !list_empty(&n->slabs_full) ||
!list_empty(&n->slabs_partial);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
return (ret ? 1 : 0);
}
#ifdef CONFIG_MEMCG
void __kmemcg_cache_deactivate(struct kmem_cache *cachep)
{
__kmem_cache_shrink(cachep);
}
mm: memcg/slab: generalize postponed non-root kmem_cache deactivation Currently SLUB uses a work scheduled after an RCU grace period to deactivate a non-root kmem_cache. This mechanism can be reused for kmem_caches release, but requires generalization for SLAB case. Introduce kmemcg_cache_deactivate() function, which calls allocator-specific __kmem_cache_deactivate() and schedules execution of __kmem_cache_deactivate_after_rcu() with all necessary locks in a worker context after an rcu grace period. Here is the new calling scheme: kmemcg_cache_deactivate() __kmemcg_cache_deactivate() SLAB/SLUB-specific kmemcg_rcufn() rcu kmemcg_workfn() work __kmemcg_cache_deactivate_after_rcu() SLAB/SLUB-specific instead of: __kmemcg_cache_deactivate() SLAB/SLUB-specific slab_deactivate_memcg_cache_rcu_sched() SLUB-only kmemcg_rcufn() rcu kmemcg_workfn() work kmemcg_cache_deact_after_rcu() SLUB-only For consistency, all allocator-specific functions start with "__". Link: http://lkml.kernel.org/r/20190611231813.3148843-4-guro@fb.com Signed-off-by: Roman Gushchin <guro@fb.com> Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com> Reviewed-by: Shakeel Butt <shakeelb@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@suse.com> Cc: Waiman Long <longman@redhat.com> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: Andrei Vagin <avagin@gmail.com> Cc: Qian Cai <cai@lca.pw> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:56:09 +08:00
void __kmemcg_cache_deactivate_after_rcu(struct kmem_cache *s)
{
}
#endif
int __kmem_cache_shutdown(struct kmem_cache *cachep)
mm: slab: free kmem_cache_node after destroy sysfs file When slub_debug alloc_calls_show is enabled we will try to track location and user of slab object on each online node, kmem_cache_node structure and cpu_cache/cpu_slub shouldn't be freed till there is the last reference to sysfs file. This fixes the following panic: BUG: unable to handle kernel NULL pointer dereference at 0000000000000020 IP: list_locations+0x169/0x4e0 PGD 257304067 PUD 438456067 PMD 0 Oops: 0000 [#1] SMP CPU: 3 PID: 973074 Comm: cat ve: 0 Not tainted 3.10.0-229.7.2.ovz.9.30-00007-japdoll-dirty #2 9.30 Hardware name: DEPO Computers To Be Filled By O.E.M./H67DE3, BIOS L1.60c 07/14/2011 task: ffff88042a5dc5b0 ti: ffff88037f8d8000 task.ti: ffff88037f8d8000 RIP: list_locations+0x169/0x4e0 Call Trace: alloc_calls_show+0x1d/0x30 slab_attr_show+0x1b/0x30 sysfs_read_file+0x9a/0x1a0 vfs_read+0x9c/0x170 SyS_read+0x58/0xb0 system_call_fastpath+0x16/0x1b Code: 5e 07 12 00 b9 00 04 00 00 3d 00 04 00 00 0f 4f c1 3d 00 04 00 00 89 45 b0 0f 84 c3 00 00 00 48 63 45 b0 49 8b 9c c4 f8 00 00 00 <48> 8b 43 20 48 85 c0 74 b6 48 89 df e8 46 37 44 00 48 8b 53 10 CR2: 0000000000000020 Separated __kmem_cache_release from __kmem_cache_shutdown which now called on slab_kmem_cache_release (after the last reference to sysfs file object has dropped). Reintroduced locking in free_partial as sysfs file might access cache's partial list after shutdowning - partial revert of the commit 69cb8e6b7c29 ("slub: free slabs without holding locks"). Zap __remove_partial and use remove_partial (w/o underscores) as free_partial now takes list_lock which s partial revert for commit 1e4dd9461fab ("slub: do not assert not having lock in removing freed partial") Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com> Suggested-by: Vladimir Davydov <vdavydov@virtuozzo.com> Acked-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-02-18 05:11:37 +08:00
{
return __kmem_cache_shrink(cachep);
mm: slab: free kmem_cache_node after destroy sysfs file When slub_debug alloc_calls_show is enabled we will try to track location and user of slab object on each online node, kmem_cache_node structure and cpu_cache/cpu_slub shouldn't be freed till there is the last reference to sysfs file. This fixes the following panic: BUG: unable to handle kernel NULL pointer dereference at 0000000000000020 IP: list_locations+0x169/0x4e0 PGD 257304067 PUD 438456067 PMD 0 Oops: 0000 [#1] SMP CPU: 3 PID: 973074 Comm: cat ve: 0 Not tainted 3.10.0-229.7.2.ovz.9.30-00007-japdoll-dirty #2 9.30 Hardware name: DEPO Computers To Be Filled By O.E.M./H67DE3, BIOS L1.60c 07/14/2011 task: ffff88042a5dc5b0 ti: ffff88037f8d8000 task.ti: ffff88037f8d8000 RIP: list_locations+0x169/0x4e0 Call Trace: alloc_calls_show+0x1d/0x30 slab_attr_show+0x1b/0x30 sysfs_read_file+0x9a/0x1a0 vfs_read+0x9c/0x170 SyS_read+0x58/0xb0 system_call_fastpath+0x16/0x1b Code: 5e 07 12 00 b9 00 04 00 00 3d 00 04 00 00 0f 4f c1 3d 00 04 00 00 89 45 b0 0f 84 c3 00 00 00 48 63 45 b0 49 8b 9c c4 f8 00 00 00 <48> 8b 43 20 48 85 c0 74 b6 48 89 df e8 46 37 44 00 48 8b 53 10 CR2: 0000000000000020 Separated __kmem_cache_release from __kmem_cache_shutdown which now called on slab_kmem_cache_release (after the last reference to sysfs file object has dropped). Reintroduced locking in free_partial as sysfs file might access cache's partial list after shutdowning - partial revert of the commit 69cb8e6b7c29 ("slub: free slabs without holding locks"). Zap __remove_partial and use remove_partial (w/o underscores) as free_partial now takes list_lock which s partial revert for commit 1e4dd9461fab ("slub: do not assert not having lock in removing freed partial") Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com> Suggested-by: Vladimir Davydov <vdavydov@virtuozzo.com> Acked-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-02-18 05:11:37 +08:00
}
void __kmem_cache_release(struct kmem_cache *cachep)
{
int i;
struct kmem_cache_node *n;
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
cache_random_seq_destroy(cachep);
free_percpu(cachep->cpu_cache);
/* NUMA: free the node structures */
for_each_kmem_cache_node(cachep, i, n) {
kfree(n->shared);
free_alien_cache(n->alien);
kfree(n);
cachep->node[i] = NULL;
}
}
/*
* Get the memory for a slab management obj.
*
* For a slab cache when the slab descriptor is off-slab, the
* slab descriptor can't come from the same cache which is being created,
* Because if it is the case, that means we defer the creation of
* the kmalloc_{dma,}_cache of size sizeof(slab descriptor) to this point.
* And we eventually call down to __kmem_cache_create(), which
* in turn looks up in the kmalloc_{dma,}_caches for the disired-size one.
* This is a "chicken-and-egg" problem.
*
* So the off-slab slab descriptor shall come from the kmalloc_{dma,}_caches,
* which are all initialized during kmem_cache_init().
*/
static void *alloc_slabmgmt(struct kmem_cache *cachep,
struct page *page, int colour_off,
gfp_t local_flags, int nodeid)
{
void *freelist;
void *addr = page_address(page);
page->s_mem = addr + colour_off;
mm/slab: put the freelist at the end of slab page Currently, the freelist is at the front of slab page. This requires extra space to meet object alignment requirement. If we put the freelist at the end of a slab page, objects could start at page boundary and will be at correct alignment. This is possible because freelist has no alignment constraint itself. This gives us two benefits: It removes extra memory space for the freelist alignment and remove complex calculation at cache initialization step. I can't think notable drawback here. I mentioned that this would reduce extra memory space, but, this benefit is rather theoretical because it can be applied to very few cases. Following is the example cache type that can get benefit from this change. size align num before after 32 8 124 4100 4092 64 8 63 4103 4095 88 8 46 4102 4094 272 8 15 4103 4095 408 8 10 4098 4090 32 16 124 4108 4092 64 16 63 4111 4095 32 32 124 4124 4092 64 32 63 4127 4095 96 32 42 4106 4074 before means whole size for objects and aligned freelist before applying patch and after shows the result of this patch. Since before is more than 4096, number of object should decrease and memory waste happens. Anyway, this patch removes complex calculation so looks beneficial to me. [akpm@linux-foundation.org: fix kerneldoc] Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:30 +08:00
page->active = 0;
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (OBJFREELIST_SLAB(cachep))
freelist = NULL;
else if (OFF_SLAB(cachep)) {
/* Slab management obj is off-slab. */
freelist = kmem_cache_alloc_node(cachep->freelist_cache,
local_flags, nodeid);
if (!freelist)
return NULL;
} else {
mm/slab: put the freelist at the end of slab page Currently, the freelist is at the front of slab page. This requires extra space to meet object alignment requirement. If we put the freelist at the end of a slab page, objects could start at page boundary and will be at correct alignment. This is possible because freelist has no alignment constraint itself. This gives us two benefits: It removes extra memory space for the freelist alignment and remove complex calculation at cache initialization step. I can't think notable drawback here. I mentioned that this would reduce extra memory space, but, this benefit is rather theoretical because it can be applied to very few cases. Following is the example cache type that can get benefit from this change. size align num before after 32 8 124 4100 4092 64 8 63 4103 4095 88 8 46 4102 4094 272 8 15 4103 4095 408 8 10 4098 4090 32 16 124 4108 4092 64 16 63 4111 4095 32 32 124 4124 4092 64 32 63 4127 4095 96 32 42 4106 4074 before means whole size for objects and aligned freelist before applying patch and after shows the result of this patch. Since before is more than 4096, number of object should decrease and memory waste happens. Anyway, this patch removes complex calculation so looks beneficial to me. [akpm@linux-foundation.org: fix kerneldoc] Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:30 +08:00
/* We will use last bytes at the slab for freelist */
freelist = addr + (PAGE_SIZE << cachep->gfporder) -
cachep->freelist_size;
}
mm/slab: put the freelist at the end of slab page Currently, the freelist is at the front of slab page. This requires extra space to meet object alignment requirement. If we put the freelist at the end of a slab page, objects could start at page boundary and will be at correct alignment. This is possible because freelist has no alignment constraint itself. This gives us two benefits: It removes extra memory space for the freelist alignment and remove complex calculation at cache initialization step. I can't think notable drawback here. I mentioned that this would reduce extra memory space, but, this benefit is rather theoretical because it can be applied to very few cases. Following is the example cache type that can get benefit from this change. size align num before after 32 8 124 4100 4092 64 8 63 4103 4095 88 8 46 4102 4094 272 8 15 4103 4095 408 8 10 4098 4090 32 16 124 4108 4092 64 16 63 4111 4095 32 32 124 4124 4092 64 32 63 4127 4095 96 32 42 4106 4074 before means whole size for objects and aligned freelist before applying patch and after shows the result of this patch. Since before is more than 4096, number of object should decrease and memory waste happens. Anyway, this patch removes complex calculation so looks beneficial to me. [akpm@linux-foundation.org: fix kerneldoc] Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:30 +08:00
return freelist;
}
slab: fix the type of the index on freelist index accessor Commit a41adfaa23df ("slab: introduce byte sized index for the freelist of a slab") changes the size of freelist index and also changes prototype of accessor function to freelist index. And there was a mistake. The mistake is that although it changes the size of freelist index correctly, it changes the size of the index of freelist index incorrectly. With patch, freelist index can be 1 byte or 2 bytes, that means that num of object on on a slab can be more than 255. So we need more than 1 byte for the index to find the index of free object on freelist. But, above patch makes this index type 1 byte, so slab which have more than 255 objects cannot work properly and in consequence of it, the system cannot boot. This issue was reported by Steven King on m68knommu which would use 2 bytes freelist index: https://lkml.org/lkml/2014/4/16/433 To fix is easy. To change the type of the index of freelist index on accessor functions is enough to fix this bug. Although 2 bytes is enough, I use 4 bytes since it have no bad effect and make things more easier. This fix was suggested and tested by Steven in his original report. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Reported-and-acked-by: Steven King <sfking@fdwdc.com> Acked-by: Christoph Lameter <cl@linux.com> Tested-by: James Hogan <james.hogan@imgtec.com> Tested-by: David Miller <davem@davemloft.net> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-18 15:24:09 +08:00
static inline freelist_idx_t get_free_obj(struct page *page, unsigned int idx)
{
slab: introduce byte sized index for the freelist of a slab Currently, the freelist of a slab consist of unsigned int sized indexes. Since most of slabs have less number of objects than 256, large sized indexes is needless. For example, consider the minimum kmalloc slab. It's object size is 32 byte and it would consist of one page, so 256 indexes through byte sized index are enough to contain all possible indexes. There can be some slabs whose object size is 8 byte. We cannot handle this case with byte sized index, so we need to restrict minimum object size. Since these slabs are not major, wasted memory from these slabs would be negligible. Some architectures' page size isn't 4096 bytes and rather larger than 4096 bytes (One example is 64KB page size on PPC or IA64) so that byte sized index doesn't fit to them. In this case, we will use two bytes sized index. Below is some number for this patch. * Before * kmalloc-512 525 640 512 8 1 : tunables 54 27 0 : slabdata 80 80 0 kmalloc-256 210 210 256 15 1 : tunables 120 60 0 : slabdata 14 14 0 kmalloc-192 1016 1040 192 20 1 : tunables 120 60 0 : slabdata 52 52 0 kmalloc-96 560 620 128 31 1 : tunables 120 60 0 : slabdata 20 20 0 kmalloc-64 2148 2280 64 60 1 : tunables 120 60 0 : slabdata 38 38 0 kmalloc-128 647 682 128 31 1 : tunables 120 60 0 : slabdata 22 22 0 kmalloc-32 11360 11413 32 113 1 : tunables 120 60 0 : slabdata 101 101 0 kmem_cache 197 200 192 20 1 : tunables 120 60 0 : slabdata 10 10 0 * After * kmalloc-512 521 648 512 8 1 : tunables 54 27 0 : slabdata 81 81 0 kmalloc-256 208 208 256 16 1 : tunables 120 60 0 : slabdata 13 13 0 kmalloc-192 1029 1029 192 21 1 : tunables 120 60 0 : slabdata 49 49 0 kmalloc-96 529 589 128 31 1 : tunables 120 60 0 : slabdata 19 19 0 kmalloc-64 2142 2142 64 63 1 : tunables 120 60 0 : slabdata 34 34 0 kmalloc-128 660 682 128 31 1 : tunables 120 60 0 : slabdata 22 22 0 kmalloc-32 11716 11780 32 124 1 : tunables 120 60 0 : slabdata 95 95 0 kmem_cache 197 210 192 21 1 : tunables 120 60 0 : slabdata 10 10 0 kmem_caches consisting of objects less than or equal to 256 byte have one or more objects than before. In the case of kmalloc-32, we have 11 more objects, so 352 bytes (11 * 32) are saved and this is roughly 9% saving of memory. Of couse, this percentage decreases as the number of objects in a slab decreases. Here are the performance results on my 4 cpus machine. * Before * Performance counter stats for 'perf bench sched messaging -g 50 -l 1000' (10 runs): 229,945,138 cache-misses ( +- 0.23% ) 11.627897174 seconds time elapsed ( +- 0.14% ) * After * Performance counter stats for 'perf bench sched messaging -g 50 -l 1000' (10 runs): 218,640,472 cache-misses ( +- 0.42% ) 11.504999837 seconds time elapsed ( +- 0.21% ) cache-misses are reduced by this patchset, roughly 5%. And elapsed times are improved by 1%. Acked-by: Christoph Lameter <cl@linux.com> Acked-by: David Rientjes <rientjes@google.com> Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Pekka Enberg <penberg@kernel.org>
2013-12-02 16:49:42 +08:00
return ((freelist_idx_t *)page->freelist)[idx];
}
static inline void set_free_obj(struct page *page,
slab: fix the type of the index on freelist index accessor Commit a41adfaa23df ("slab: introduce byte sized index for the freelist of a slab") changes the size of freelist index and also changes prototype of accessor function to freelist index. And there was a mistake. The mistake is that although it changes the size of freelist index correctly, it changes the size of the index of freelist index incorrectly. With patch, freelist index can be 1 byte or 2 bytes, that means that num of object on on a slab can be more than 255. So we need more than 1 byte for the index to find the index of free object on freelist. But, above patch makes this index type 1 byte, so slab which have more than 255 objects cannot work properly and in consequence of it, the system cannot boot. This issue was reported by Steven King on m68knommu which would use 2 bytes freelist index: https://lkml.org/lkml/2014/4/16/433 To fix is easy. To change the type of the index of freelist index on accessor functions is enough to fix this bug. Although 2 bytes is enough, I use 4 bytes since it have no bad effect and make things more easier. This fix was suggested and tested by Steven in his original report. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Reported-and-acked-by: Steven King <sfking@fdwdc.com> Acked-by: Christoph Lameter <cl@linux.com> Tested-by: James Hogan <james.hogan@imgtec.com> Tested-by: David Miller <davem@davemloft.net> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-18 15:24:09 +08:00
unsigned int idx, freelist_idx_t val)
{
slab: introduce byte sized index for the freelist of a slab Currently, the freelist of a slab consist of unsigned int sized indexes. Since most of slabs have less number of objects than 256, large sized indexes is needless. For example, consider the minimum kmalloc slab. It's object size is 32 byte and it would consist of one page, so 256 indexes through byte sized index are enough to contain all possible indexes. There can be some slabs whose object size is 8 byte. We cannot handle this case with byte sized index, so we need to restrict minimum object size. Since these slabs are not major, wasted memory from these slabs would be negligible. Some architectures' page size isn't 4096 bytes and rather larger than 4096 bytes (One example is 64KB page size on PPC or IA64) so that byte sized index doesn't fit to them. In this case, we will use two bytes sized index. Below is some number for this patch. * Before * kmalloc-512 525 640 512 8 1 : tunables 54 27 0 : slabdata 80 80 0 kmalloc-256 210 210 256 15 1 : tunables 120 60 0 : slabdata 14 14 0 kmalloc-192 1016 1040 192 20 1 : tunables 120 60 0 : slabdata 52 52 0 kmalloc-96 560 620 128 31 1 : tunables 120 60 0 : slabdata 20 20 0 kmalloc-64 2148 2280 64 60 1 : tunables 120 60 0 : slabdata 38 38 0 kmalloc-128 647 682 128 31 1 : tunables 120 60 0 : slabdata 22 22 0 kmalloc-32 11360 11413 32 113 1 : tunables 120 60 0 : slabdata 101 101 0 kmem_cache 197 200 192 20 1 : tunables 120 60 0 : slabdata 10 10 0 * After * kmalloc-512 521 648 512 8 1 : tunables 54 27 0 : slabdata 81 81 0 kmalloc-256 208 208 256 16 1 : tunables 120 60 0 : slabdata 13 13 0 kmalloc-192 1029 1029 192 21 1 : tunables 120 60 0 : slabdata 49 49 0 kmalloc-96 529 589 128 31 1 : tunables 120 60 0 : slabdata 19 19 0 kmalloc-64 2142 2142 64 63 1 : tunables 120 60 0 : slabdata 34 34 0 kmalloc-128 660 682 128 31 1 : tunables 120 60 0 : slabdata 22 22 0 kmalloc-32 11716 11780 32 124 1 : tunables 120 60 0 : slabdata 95 95 0 kmem_cache 197 210 192 21 1 : tunables 120 60 0 : slabdata 10 10 0 kmem_caches consisting of objects less than or equal to 256 byte have one or more objects than before. In the case of kmalloc-32, we have 11 more objects, so 352 bytes (11 * 32) are saved and this is roughly 9% saving of memory. Of couse, this percentage decreases as the number of objects in a slab decreases. Here are the performance results on my 4 cpus machine. * Before * Performance counter stats for 'perf bench sched messaging -g 50 -l 1000' (10 runs): 229,945,138 cache-misses ( +- 0.23% ) 11.627897174 seconds time elapsed ( +- 0.14% ) * After * Performance counter stats for 'perf bench sched messaging -g 50 -l 1000' (10 runs): 218,640,472 cache-misses ( +- 0.42% ) 11.504999837 seconds time elapsed ( +- 0.21% ) cache-misses are reduced by this patchset, roughly 5%. And elapsed times are improved by 1%. Acked-by: Christoph Lameter <cl@linux.com> Acked-by: David Rientjes <rientjes@google.com> Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Pekka Enberg <penberg@kernel.org>
2013-12-02 16:49:42 +08:00
((freelist_idx_t *)(page->freelist))[idx] = val;
}
static void cache_init_objs_debug(struct kmem_cache *cachep, struct page *page)
{
#if DEBUG
int i;
for (i = 0; i < cachep->num; i++) {
void *objp = index_to_obj(cachep, page, i);
if (cachep->flags & SLAB_STORE_USER)
*dbg_userword(cachep, objp) = NULL;
if (cachep->flags & SLAB_RED_ZONE) {
*dbg_redzone1(cachep, objp) = RED_INACTIVE;
*dbg_redzone2(cachep, objp) = RED_INACTIVE;
}
/*
* Constructors are not allowed to allocate memory from the same
* cache which they are a constructor for. Otherwise, deadlock.
* They must also be threaded.
*/
if (cachep->ctor && !(cachep->flags & SLAB_POISON)) {
kasan_unpoison_object_data(cachep,
objp + obj_offset(cachep));
cachep->ctor(objp + obj_offset(cachep));
kasan_poison_object_data(
cachep, objp + obj_offset(cachep));
}
if (cachep->flags & SLAB_RED_ZONE) {
if (*dbg_redzone2(cachep, objp) != RED_INACTIVE)
slab_error(cachep, "constructor overwrote the end of an object");
if (*dbg_redzone1(cachep, objp) != RED_INACTIVE)
slab_error(cachep, "constructor overwrote the start of an object");
}
/* need to poison the objs? */
if (cachep->flags & SLAB_POISON) {
poison_obj(cachep, objp, POISON_FREE);
slab_kernel_map(cachep, objp, 0);
}
}
#endif
}
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
#ifdef CONFIG_SLAB_FREELIST_RANDOM
/* Hold information during a freelist initialization */
union freelist_init_state {
struct {
unsigned int pos;
mm: reorganize SLAB freelist randomization The kernel heap allocators are using a sequential freelist making their allocation predictable. This predictability makes kernel heap overflow easier to exploit. An attacker can careful prepare the kernel heap to control the following chunk overflowed. For example these attacks exploit the predictability of the heap: - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU) - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95) ***Problems that needed solving: - Randomize the Freelist (singled linked) used in the SLUB allocator. - Ensure good performance to encourage usage. - Get best entropy in early boot stage. ***Parts: - 01/02 Reorganize the SLAB Freelist randomization to share elements with the SLUB implementation. - 02/02 The SLUB Freelist randomization implementation. Similar approach than the SLAB but tailored to the singled freelist used in SLUB. ***Performance data: slab_test impact is between 3% to 4% on average for 100000 attempts without smp. It is a very focused testing, kernbench show the overall impact on the system is way lower. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles 100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles 100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles 100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles 100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles 100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles 100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles 100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles 100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles 100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 70 cycles 100000 times kmalloc(16)/kfree -> 70 cycles 100000 times kmalloc(32)/kfree -> 70 cycles 100000 times kmalloc(64)/kfree -> 70 cycles 100000 times kmalloc(128)/kfree -> 70 cycles 100000 times kmalloc(256)/kfree -> 69 cycles 100000 times kmalloc(512)/kfree -> 70 cycles 100000 times kmalloc(1024)/kfree -> 73 cycles 100000 times kmalloc(2048)/kfree -> 72 cycles 100000 times kmalloc(4096)/kfree -> 71 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles 100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles 100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles 100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles 100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles 100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles 100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles 100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles 100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles 100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 66 cycles 100000 times kmalloc(16)/kfree -> 66 cycles 100000 times kmalloc(32)/kfree -> 66 cycles 100000 times kmalloc(64)/kfree -> 66 cycles 100000 times kmalloc(128)/kfree -> 65 cycles 100000 times kmalloc(256)/kfree -> 67 cycles 100000 times kmalloc(512)/kfree -> 67 cycles 100000 times kmalloc(1024)/kfree -> 64 cycles 100000 times kmalloc(2048)/kfree -> 67 cycles 100000 times kmalloc(4096)/kfree -> 67 cycles Kernbench, before: Average Optimal load -j 12 Run (std deviation): Elapsed Time 101.873 (1.16069) User Time 1045.22 (1.60447) System Time 88.969 (0.559195) Percent CPU 1112.9 (13.8279) Context Switches 189140 (2282.15) Sleeps 99008.6 (768.091) After: Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.47 (0.562732) User Time 1045.3 (1.34263) System Time 88.311 (0.342554) Percent CPU 1105.8 (6.49444) Context Switches 189081 (2355.78) Sleeps 99231.5 (800.358) This patch (of 2): This commit reorganizes the previous SLAB freelist randomization to prepare for the SLUB implementation. It moves functions that will be shared to slab_common. The entropy functions are changed to align with the SLUB implementation, now using get_random_(int|long) functions. These functions were chosen because they provide a bit more entropy early on boot and better performance when specific arch instructions are not available. [akpm@linux-foundation.org: fix build] Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com Signed-off-by: Thomas Garnier <thgarnie@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:21:56 +08:00
unsigned int *list;
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
unsigned int count;
};
struct rnd_state rnd_state;
};
/*
* Initialize the state based on the randomization methode available.
* return true if the pre-computed list is available, false otherwize.
*/
static bool freelist_state_initialize(union freelist_init_state *state,
struct kmem_cache *cachep,
unsigned int count)
{
bool ret;
unsigned int rand;
/* Use best entropy available to define a random shift */
mm: reorganize SLAB freelist randomization The kernel heap allocators are using a sequential freelist making their allocation predictable. This predictability makes kernel heap overflow easier to exploit. An attacker can careful prepare the kernel heap to control the following chunk overflowed. For example these attacks exploit the predictability of the heap: - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU) - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95) ***Problems that needed solving: - Randomize the Freelist (singled linked) used in the SLUB allocator. - Ensure good performance to encourage usage. - Get best entropy in early boot stage. ***Parts: - 01/02 Reorganize the SLAB Freelist randomization to share elements with the SLUB implementation. - 02/02 The SLUB Freelist randomization implementation. Similar approach than the SLAB but tailored to the singled freelist used in SLUB. ***Performance data: slab_test impact is between 3% to 4% on average for 100000 attempts without smp. It is a very focused testing, kernbench show the overall impact on the system is way lower. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles 100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles 100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles 100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles 100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles 100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles 100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles 100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles 100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles 100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 70 cycles 100000 times kmalloc(16)/kfree -> 70 cycles 100000 times kmalloc(32)/kfree -> 70 cycles 100000 times kmalloc(64)/kfree -> 70 cycles 100000 times kmalloc(128)/kfree -> 70 cycles 100000 times kmalloc(256)/kfree -> 69 cycles 100000 times kmalloc(512)/kfree -> 70 cycles 100000 times kmalloc(1024)/kfree -> 73 cycles 100000 times kmalloc(2048)/kfree -> 72 cycles 100000 times kmalloc(4096)/kfree -> 71 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles 100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles 100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles 100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles 100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles 100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles 100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles 100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles 100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles 100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 66 cycles 100000 times kmalloc(16)/kfree -> 66 cycles 100000 times kmalloc(32)/kfree -> 66 cycles 100000 times kmalloc(64)/kfree -> 66 cycles 100000 times kmalloc(128)/kfree -> 65 cycles 100000 times kmalloc(256)/kfree -> 67 cycles 100000 times kmalloc(512)/kfree -> 67 cycles 100000 times kmalloc(1024)/kfree -> 64 cycles 100000 times kmalloc(2048)/kfree -> 67 cycles 100000 times kmalloc(4096)/kfree -> 67 cycles Kernbench, before: Average Optimal load -j 12 Run (std deviation): Elapsed Time 101.873 (1.16069) User Time 1045.22 (1.60447) System Time 88.969 (0.559195) Percent CPU 1112.9 (13.8279) Context Switches 189140 (2282.15) Sleeps 99008.6 (768.091) After: Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.47 (0.562732) User Time 1045.3 (1.34263) System Time 88.311 (0.342554) Percent CPU 1105.8 (6.49444) Context Switches 189081 (2355.78) Sleeps 99231.5 (800.358) This patch (of 2): This commit reorganizes the previous SLAB freelist randomization to prepare for the SLUB implementation. It moves functions that will be shared to slab_common. The entropy functions are changed to align with the SLUB implementation, now using get_random_(int|long) functions. These functions were chosen because they provide a bit more entropy early on boot and better performance when specific arch instructions are not available. [akpm@linux-foundation.org: fix build] Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com Signed-off-by: Thomas Garnier <thgarnie@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:21:56 +08:00
rand = get_random_int();
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
/* Use a random state if the pre-computed list is not available */
if (!cachep->random_seq) {
prandom_seed_state(&state->rnd_state, rand);
ret = false;
} else {
state->list = cachep->random_seq;
state->count = count;
state->pos = rand % count;
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
ret = true;
}
return ret;
}
/* Get the next entry on the list and randomize it using a random shift */
static freelist_idx_t next_random_slot(union freelist_init_state *state)
{
if (state->pos >= state->count)
state->pos = 0;
return state->list[state->pos++];
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
}
mm: reorganize SLAB freelist randomization The kernel heap allocators are using a sequential freelist making their allocation predictable. This predictability makes kernel heap overflow easier to exploit. An attacker can careful prepare the kernel heap to control the following chunk overflowed. For example these attacks exploit the predictability of the heap: - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU) - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95) ***Problems that needed solving: - Randomize the Freelist (singled linked) used in the SLUB allocator. - Ensure good performance to encourage usage. - Get best entropy in early boot stage. ***Parts: - 01/02 Reorganize the SLAB Freelist randomization to share elements with the SLUB implementation. - 02/02 The SLUB Freelist randomization implementation. Similar approach than the SLAB but tailored to the singled freelist used in SLUB. ***Performance data: slab_test impact is between 3% to 4% on average for 100000 attempts without smp. It is a very focused testing, kernbench show the overall impact on the system is way lower. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles 100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles 100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles 100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles 100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles 100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles 100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles 100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles 100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles 100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 70 cycles 100000 times kmalloc(16)/kfree -> 70 cycles 100000 times kmalloc(32)/kfree -> 70 cycles 100000 times kmalloc(64)/kfree -> 70 cycles 100000 times kmalloc(128)/kfree -> 70 cycles 100000 times kmalloc(256)/kfree -> 69 cycles 100000 times kmalloc(512)/kfree -> 70 cycles 100000 times kmalloc(1024)/kfree -> 73 cycles 100000 times kmalloc(2048)/kfree -> 72 cycles 100000 times kmalloc(4096)/kfree -> 71 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles 100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles 100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles 100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles 100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles 100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles 100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles 100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles 100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles 100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 66 cycles 100000 times kmalloc(16)/kfree -> 66 cycles 100000 times kmalloc(32)/kfree -> 66 cycles 100000 times kmalloc(64)/kfree -> 66 cycles 100000 times kmalloc(128)/kfree -> 65 cycles 100000 times kmalloc(256)/kfree -> 67 cycles 100000 times kmalloc(512)/kfree -> 67 cycles 100000 times kmalloc(1024)/kfree -> 64 cycles 100000 times kmalloc(2048)/kfree -> 67 cycles 100000 times kmalloc(4096)/kfree -> 67 cycles Kernbench, before: Average Optimal load -j 12 Run (std deviation): Elapsed Time 101.873 (1.16069) User Time 1045.22 (1.60447) System Time 88.969 (0.559195) Percent CPU 1112.9 (13.8279) Context Switches 189140 (2282.15) Sleeps 99008.6 (768.091) After: Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.47 (0.562732) User Time 1045.3 (1.34263) System Time 88.311 (0.342554) Percent CPU 1105.8 (6.49444) Context Switches 189081 (2355.78) Sleeps 99231.5 (800.358) This patch (of 2): This commit reorganizes the previous SLAB freelist randomization to prepare for the SLUB implementation. It moves functions that will be shared to slab_common. The entropy functions are changed to align with the SLUB implementation, now using get_random_(int|long) functions. These functions were chosen because they provide a bit more entropy early on boot and better performance when specific arch instructions are not available. [akpm@linux-foundation.org: fix build] Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com Signed-off-by: Thomas Garnier <thgarnie@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:21:56 +08:00
/* Swap two freelist entries */
static void swap_free_obj(struct page *page, unsigned int a, unsigned int b)
{
swap(((freelist_idx_t *)page->freelist)[a],
((freelist_idx_t *)page->freelist)[b]);
}
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
/*
* Shuffle the freelist initialization state based on pre-computed lists.
* return true if the list was successfully shuffled, false otherwise.
*/
static bool shuffle_freelist(struct kmem_cache *cachep, struct page *page)
{
mm: reorganize SLAB freelist randomization The kernel heap allocators are using a sequential freelist making their allocation predictable. This predictability makes kernel heap overflow easier to exploit. An attacker can careful prepare the kernel heap to control the following chunk overflowed. For example these attacks exploit the predictability of the heap: - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU) - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95) ***Problems that needed solving: - Randomize the Freelist (singled linked) used in the SLUB allocator. - Ensure good performance to encourage usage. - Get best entropy in early boot stage. ***Parts: - 01/02 Reorganize the SLAB Freelist randomization to share elements with the SLUB implementation. - 02/02 The SLUB Freelist randomization implementation. Similar approach than the SLAB but tailored to the singled freelist used in SLUB. ***Performance data: slab_test impact is between 3% to 4% on average for 100000 attempts without smp. It is a very focused testing, kernbench show the overall impact on the system is way lower. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles 100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles 100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles 100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles 100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles 100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles 100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles 100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles 100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles 100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 70 cycles 100000 times kmalloc(16)/kfree -> 70 cycles 100000 times kmalloc(32)/kfree -> 70 cycles 100000 times kmalloc(64)/kfree -> 70 cycles 100000 times kmalloc(128)/kfree -> 70 cycles 100000 times kmalloc(256)/kfree -> 69 cycles 100000 times kmalloc(512)/kfree -> 70 cycles 100000 times kmalloc(1024)/kfree -> 73 cycles 100000 times kmalloc(2048)/kfree -> 72 cycles 100000 times kmalloc(4096)/kfree -> 71 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles 100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles 100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles 100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles 100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles 100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles 100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles 100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles 100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles 100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 66 cycles 100000 times kmalloc(16)/kfree -> 66 cycles 100000 times kmalloc(32)/kfree -> 66 cycles 100000 times kmalloc(64)/kfree -> 66 cycles 100000 times kmalloc(128)/kfree -> 65 cycles 100000 times kmalloc(256)/kfree -> 67 cycles 100000 times kmalloc(512)/kfree -> 67 cycles 100000 times kmalloc(1024)/kfree -> 64 cycles 100000 times kmalloc(2048)/kfree -> 67 cycles 100000 times kmalloc(4096)/kfree -> 67 cycles Kernbench, before: Average Optimal load -j 12 Run (std deviation): Elapsed Time 101.873 (1.16069) User Time 1045.22 (1.60447) System Time 88.969 (0.559195) Percent CPU 1112.9 (13.8279) Context Switches 189140 (2282.15) Sleeps 99008.6 (768.091) After: Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.47 (0.562732) User Time 1045.3 (1.34263) System Time 88.311 (0.342554) Percent CPU 1105.8 (6.49444) Context Switches 189081 (2355.78) Sleeps 99231.5 (800.358) This patch (of 2): This commit reorganizes the previous SLAB freelist randomization to prepare for the SLUB implementation. It moves functions that will be shared to slab_common. The entropy functions are changed to align with the SLUB implementation, now using get_random_(int|long) functions. These functions were chosen because they provide a bit more entropy early on boot and better performance when specific arch instructions are not available. [akpm@linux-foundation.org: fix build] Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com Signed-off-by: Thomas Garnier <thgarnie@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:21:56 +08:00
unsigned int objfreelist = 0, i, rand, count = cachep->num;
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
union freelist_init_state state;
bool precomputed;
if (count < 2)
return false;
precomputed = freelist_state_initialize(&state, cachep, count);
/* Take a random entry as the objfreelist */
if (OBJFREELIST_SLAB(cachep)) {
if (!precomputed)
objfreelist = count - 1;
else
objfreelist = next_random_slot(&state);
page->freelist = index_to_obj(cachep, page, objfreelist) +
obj_offset(cachep);
count--;
}
/*
* On early boot, generate the list dynamically.
* Later use a pre-computed list for speed.
*/
if (!precomputed) {
mm: reorganize SLAB freelist randomization The kernel heap allocators are using a sequential freelist making their allocation predictable. This predictability makes kernel heap overflow easier to exploit. An attacker can careful prepare the kernel heap to control the following chunk overflowed. For example these attacks exploit the predictability of the heap: - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU) - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95) ***Problems that needed solving: - Randomize the Freelist (singled linked) used in the SLUB allocator. - Ensure good performance to encourage usage. - Get best entropy in early boot stage. ***Parts: - 01/02 Reorganize the SLAB Freelist randomization to share elements with the SLUB implementation. - 02/02 The SLUB Freelist randomization implementation. Similar approach than the SLAB but tailored to the singled freelist used in SLUB. ***Performance data: slab_test impact is between 3% to 4% on average for 100000 attempts without smp. It is a very focused testing, kernbench show the overall impact on the system is way lower. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles 100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles 100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles 100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles 100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles 100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles 100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles 100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles 100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles 100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 70 cycles 100000 times kmalloc(16)/kfree -> 70 cycles 100000 times kmalloc(32)/kfree -> 70 cycles 100000 times kmalloc(64)/kfree -> 70 cycles 100000 times kmalloc(128)/kfree -> 70 cycles 100000 times kmalloc(256)/kfree -> 69 cycles 100000 times kmalloc(512)/kfree -> 70 cycles 100000 times kmalloc(1024)/kfree -> 73 cycles 100000 times kmalloc(2048)/kfree -> 72 cycles 100000 times kmalloc(4096)/kfree -> 71 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles 100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles 100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles 100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles 100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles 100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles 100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles 100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles 100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles 100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 66 cycles 100000 times kmalloc(16)/kfree -> 66 cycles 100000 times kmalloc(32)/kfree -> 66 cycles 100000 times kmalloc(64)/kfree -> 66 cycles 100000 times kmalloc(128)/kfree -> 65 cycles 100000 times kmalloc(256)/kfree -> 67 cycles 100000 times kmalloc(512)/kfree -> 67 cycles 100000 times kmalloc(1024)/kfree -> 64 cycles 100000 times kmalloc(2048)/kfree -> 67 cycles 100000 times kmalloc(4096)/kfree -> 67 cycles Kernbench, before: Average Optimal load -j 12 Run (std deviation): Elapsed Time 101.873 (1.16069) User Time 1045.22 (1.60447) System Time 88.969 (0.559195) Percent CPU 1112.9 (13.8279) Context Switches 189140 (2282.15) Sleeps 99008.6 (768.091) After: Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.47 (0.562732) User Time 1045.3 (1.34263) System Time 88.311 (0.342554) Percent CPU 1105.8 (6.49444) Context Switches 189081 (2355.78) Sleeps 99231.5 (800.358) This patch (of 2): This commit reorganizes the previous SLAB freelist randomization to prepare for the SLUB implementation. It moves functions that will be shared to slab_common. The entropy functions are changed to align with the SLUB implementation, now using get_random_(int|long) functions. These functions were chosen because they provide a bit more entropy early on boot and better performance when specific arch instructions are not available. [akpm@linux-foundation.org: fix build] Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com Signed-off-by: Thomas Garnier <thgarnie@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:21:56 +08:00
for (i = 0; i < count; i++)
set_free_obj(page, i, i);
/* Fisher-Yates shuffle */
for (i = count - 1; i > 0; i--) {
rand = prandom_u32_state(&state.rnd_state);
rand %= (i + 1);
swap_free_obj(page, i, rand);
}
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
} else {
for (i = 0; i < count; i++)
set_free_obj(page, i, next_random_slot(&state));
}
if (OBJFREELIST_SLAB(cachep))
set_free_obj(page, cachep->num - 1, objfreelist);
return true;
}
#else
static inline bool shuffle_freelist(struct kmem_cache *cachep,
struct page *page)
{
return false;
}
#endif /* CONFIG_SLAB_FREELIST_RANDOM */
static void cache_init_objs(struct kmem_cache *cachep,
struct page *page)
{
int i;
void *objp;
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
bool shuffled;
cache_init_objs_debug(cachep, page);
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
/* Try to randomize the freelist if enabled */
shuffled = shuffle_freelist(cachep, page);
if (!shuffled && OBJFREELIST_SLAB(cachep)) {
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
page->freelist = index_to_obj(cachep, page, cachep->num - 1) +
obj_offset(cachep);
}
for (i = 0; i < cachep->num; i++) {
objp = index_to_obj(cachep, page, i);
objp = kasan_init_slab_obj(cachep, objp);
/* constructor could break poison info */
if (DEBUG == 0 && cachep->ctor) {
kasan_unpoison_object_data(cachep, objp);
cachep->ctor(objp);
kasan_poison_object_data(cachep, objp);
}
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
if (!shuffled)
set_free_obj(page, i, i);
}
}
static void *slab_get_obj(struct kmem_cache *cachep, struct page *page)
{
void *objp;
objp = index_to_obj(cachep, page, get_free_obj(page, page->active));
page->active++;
return objp;
}
static void slab_put_obj(struct kmem_cache *cachep,
struct page *page, void *objp)
{
unsigned int objnr = obj_to_index(cachep, page, objp);
#if DEBUG
unsigned int i;
/* Verify double free bug */
for (i = page->active; i < cachep->num; i++) {
if (get_free_obj(page, i) == objnr) {
pr_err("slab: double free detected in cache '%s', objp %px\n",
cachep->name, objp);
BUG();
}
}
#endif
page->active--;
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (!page->freelist)
page->freelist = objp + obj_offset(cachep);
set_free_obj(page, page->active, objnr);
}
/*
* Map pages beginning at addr to the given cache and slab. This is required
* for the slab allocator to be able to lookup the cache and slab of a
* virtual address for kfree, ksize, and slab debugging.
*/
static void slab_map_pages(struct kmem_cache *cache, struct page *page,
void *freelist)
{
page->slab_cache = cache;
page->freelist = freelist;
}
/*
* Grow (by 1) the number of slabs within a cache. This is called by
* kmem_cache_alloc() when there are no active objs left in a cache.
*/
static struct page *cache_grow_begin(struct kmem_cache *cachep,
gfp_t flags, int nodeid)
{
void *freelist;
size_t offset;
gfp_t local_flags;
int page_node;
struct kmem_cache_node *n;
struct page *page;
/*
* Be lazy and only check for valid flags here, keeping it out of the
* critical path in kmem_cache_alloc().
*/
if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
slab: do not panic on invalid gfp_mask Both SLAB and SLUB BUG() when a caller provides an invalid gfp_mask. This is a rather harsh way to announce a non-critical issue. Allocator is free to ignore invalid flags. Let's simply replace BUG() by dump_stack to tell the offender and fixup the mask to move on with the allocation request. This is an example for kmalloc(GFP_KERNEL|__GFP_HIGHMEM) from a test module: Unexpected gfp: 0x2 (__GFP_HIGHMEM). Fixing up to gfp: 0x24000c0 (GFP_KERNEL). Fix your code! CPU: 0 PID: 2916 Comm: insmod Tainted: G O 4.6.0-slabgfp2-00002-g4cdfc2ef4892-dirty #936 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014 Call Trace: dump_stack+0x67/0x90 cache_alloc_refill+0x201/0x617 kmem_cache_alloc_trace+0xa7/0x24a ? 0xffffffffa0005000 mymodule_init+0x20/0x1000 [test_slab] do_one_initcall+0xe7/0x16c ? rcu_read_lock_sched_held+0x61/0x69 ? kmem_cache_alloc_trace+0x197/0x24a do_init_module+0x5f/0x1d9 load_module+0x1a3d/0x1f21 ? retint_kernel+0x2d/0x2d SyS_init_module+0xe8/0x10e ? SyS_init_module+0xe8/0x10e do_syscall_64+0x68/0x13f entry_SYSCALL64_slow_path+0x25/0x25 Link: http://lkml.kernel.org/r/1465548200-11384-2-git-send-email-mhocko@kernel.org Signed-off-by: Michal Hocko <mhocko@suse.com> Cc: Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:22:05 +08:00
flags &= ~GFP_SLAB_BUG_MASK;
pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
invalid_mask, &invalid_mask, flags, &flags);
dump_stack();
}
WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO));
Categorize GFP flags The function of GFP_LEVEL_MASK seems to be unclear. In order to clear up the mystery we get rid of it and replace GFP_LEVEL_MASK with 3 sets of GFP flags: GFP_RECLAIM_MASK Flags used to control page allocator reclaim behavior. GFP_CONSTRAINT_MASK Flags used to limit where allocations can occur. GFP_SLAB_BUG_MASK Flags that the slab allocator BUG()s on. These replace the uses of GFP_LEVEL mask in the slab allocators and in vmalloc.c. The use of the flags not included in these sets may occur as a result of a slab allocation standing in for a page allocation when constructing scatter gather lists. Extraneous flags are cleared and not passed through to the page allocator. __GFP_MOVABLE/RECLAIMABLE, __GFP_COLD and __GFP_COMP will now be ignored if passed to a slab allocator. Change the allocation of allocator meta data in SLAB and vmalloc to not pass through flags listed in GFP_CONSTRAINT_MASK. SLAB already removes the __GFP_THISNODE flag for such allocations. Generalize that to also cover vmalloc. The use of GFP_CONSTRAINT_MASK also includes __GFP_HARDWALL. The impact of allocator metadata placement on access latency to the cachelines of the object itself is minimal since metadata is only referenced on alloc and free. The attempt is still made to place the meta data optimally but we consistently allow fallback both in SLAB and vmalloc (SLUB does not need to allocate metadata like that). Allocator metadata may serve multiple in kernel users and thus should not be subject to the limitations arising from a single allocation context. [akpm@linux-foundation.org: fix fallback_alloc()] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-10-16 16:25:41 +08:00
local_flags = flags & (GFP_CONSTRAINT_MASK|GFP_RECLAIM_MASK);
check_irq_off();
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd __GFP_WAIT has been used to identify atomic context in callers that hold spinlocks or are in interrupts. They are expected to be high priority and have access one of two watermarks lower than "min" which can be referred to as the "atomic reserve". __GFP_HIGH users get access to the first lower watermark and can be called the "high priority reserve". Over time, callers had a requirement to not block when fallback options were available. Some have abused __GFP_WAIT leading to a situation where an optimisitic allocation with a fallback option can access atomic reserves. This patch uses __GFP_ATOMIC to identify callers that are truely atomic, cannot sleep and have no alternative. High priority users continue to use __GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify callers that want to wake kswapd for background reclaim. __GFP_WAIT is redefined as a caller that is willing to enter direct reclaim and wake kswapd for background reclaim. This patch then converts a number of sites o __GFP_ATOMIC is used by callers that are high priority and have memory pools for those requests. GFP_ATOMIC uses this flag. o Callers that have a limited mempool to guarantee forward progress clear __GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall into this category where kswapd will still be woken but atomic reserves are not used as there is a one-entry mempool to guarantee progress. o Callers that are checking if they are non-blocking should use the helper gfpflags_allow_blocking() where possible. This is because checking for __GFP_WAIT as was done historically now can trigger false positives. Some exceptions like dm-crypt.c exist where the code intent is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to flag manipulations. o Callers that built their own GFP flags instead of starting with GFP_KERNEL and friends now also need to specify __GFP_KSWAPD_RECLAIM. The first key hazard to watch out for is callers that removed __GFP_WAIT and was depending on access to atomic reserves for inconspicuous reasons. In some cases it may be appropriate for them to use __GFP_HIGH. The second key hazard is callers that assembled their own combination of GFP flags instead of starting with something like GFP_KERNEL. They may now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless if it's missed in most cases as other activity will wake kswapd. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Vitaly Wool <vitalywool@gmail.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-07 08:28:21 +08:00
if (gfpflags_allow_blocking(local_flags))
local_irq_enable();
/*
* Get mem for the objs. Attempt to allocate a physical page from
* 'nodeid'.
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
*/
page = kmem_getpages(cachep, local_flags, nodeid);
if (!page)
goto failed;
page_node = page_to_nid(page);
n = get_node(cachep, page_node);
mm/slab: racy access/modify the slab color Slab color isn't needed to be changed strictly. Because locking for changing slab color could cause more lock contention so this patch implements racy access/modify the slab color. This is a preparation step to implement lockless allocation path when there is no free objects in the kmem_cache. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 It shows that contention is reduced for object size >= 1024 and performance increases by roughly 15%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:20 +08:00
/* Get colour for the slab, and cal the next value. */
n->colour_next++;
if (n->colour_next >= cachep->colour)
n->colour_next = 0;
offset = n->colour_next;
if (offset >= cachep->colour)
offset = 0;
offset *= cachep->colour_off;
/*
* Call kasan_poison_slab() before calling alloc_slabmgmt(), so
* page_address() in the latter returns a non-tagged pointer,
* as it should be for slab pages.
*/
kasan_poison_slab(page);
/* Get slab management. */
freelist = alloc_slabmgmt(cachep, page, offset,
local_flags & ~GFP_CONSTRAINT_MASK, page_node);
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (OFF_SLAB(cachep) && !freelist)
goto opps1;
slab_map_pages(cachep, page, freelist);
cache_init_objs(cachep, page);
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd __GFP_WAIT has been used to identify atomic context in callers that hold spinlocks or are in interrupts. They are expected to be high priority and have access one of two watermarks lower than "min" which can be referred to as the "atomic reserve". __GFP_HIGH users get access to the first lower watermark and can be called the "high priority reserve". Over time, callers had a requirement to not block when fallback options were available. Some have abused __GFP_WAIT leading to a situation where an optimisitic allocation with a fallback option can access atomic reserves. This patch uses __GFP_ATOMIC to identify callers that are truely atomic, cannot sleep and have no alternative. High priority users continue to use __GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify callers that want to wake kswapd for background reclaim. __GFP_WAIT is redefined as a caller that is willing to enter direct reclaim and wake kswapd for background reclaim. This patch then converts a number of sites o __GFP_ATOMIC is used by callers that are high priority and have memory pools for those requests. GFP_ATOMIC uses this flag. o Callers that have a limited mempool to guarantee forward progress clear __GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall into this category where kswapd will still be woken but atomic reserves are not used as there is a one-entry mempool to guarantee progress. o Callers that are checking if they are non-blocking should use the helper gfpflags_allow_blocking() where possible. This is because checking for __GFP_WAIT as was done historically now can trigger false positives. Some exceptions like dm-crypt.c exist where the code intent is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to flag manipulations. o Callers that built their own GFP flags instead of starting with GFP_KERNEL and friends now also need to specify __GFP_KSWAPD_RECLAIM. The first key hazard to watch out for is callers that removed __GFP_WAIT and was depending on access to atomic reserves for inconspicuous reasons. In some cases it may be appropriate for them to use __GFP_HIGH. The second key hazard is callers that assembled their own combination of GFP flags instead of starting with something like GFP_KERNEL. They may now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless if it's missed in most cases as other activity will wake kswapd. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Vitaly Wool <vitalywool@gmail.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-07 08:28:21 +08:00
if (gfpflags_allow_blocking(local_flags))
local_irq_disable();
return page;
opps1:
kmem_freepages(cachep, page);
failed:
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd __GFP_WAIT has been used to identify atomic context in callers that hold spinlocks or are in interrupts. They are expected to be high priority and have access one of two watermarks lower than "min" which can be referred to as the "atomic reserve". __GFP_HIGH users get access to the first lower watermark and can be called the "high priority reserve". Over time, callers had a requirement to not block when fallback options were available. Some have abused __GFP_WAIT leading to a situation where an optimisitic allocation with a fallback option can access atomic reserves. This patch uses __GFP_ATOMIC to identify callers that are truely atomic, cannot sleep and have no alternative. High priority users continue to use __GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify callers that want to wake kswapd for background reclaim. __GFP_WAIT is redefined as a caller that is willing to enter direct reclaim and wake kswapd for background reclaim. This patch then converts a number of sites o __GFP_ATOMIC is used by callers that are high priority and have memory pools for those requests. GFP_ATOMIC uses this flag. o Callers that have a limited mempool to guarantee forward progress clear __GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall into this category where kswapd will still be woken but atomic reserves are not used as there is a one-entry mempool to guarantee progress. o Callers that are checking if they are non-blocking should use the helper gfpflags_allow_blocking() where possible. This is because checking for __GFP_WAIT as was done historically now can trigger false positives. Some exceptions like dm-crypt.c exist where the code intent is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to flag manipulations. o Callers that built their own GFP flags instead of starting with GFP_KERNEL and friends now also need to specify __GFP_KSWAPD_RECLAIM. The first key hazard to watch out for is callers that removed __GFP_WAIT and was depending on access to atomic reserves for inconspicuous reasons. In some cases it may be appropriate for them to use __GFP_HIGH. The second key hazard is callers that assembled their own combination of GFP flags instead of starting with something like GFP_KERNEL. They may now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless if it's missed in most cases as other activity will wake kswapd. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Vitaly Wool <vitalywool@gmail.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-07 08:28:21 +08:00
if (gfpflags_allow_blocking(local_flags))
local_irq_disable();
return NULL;
}
static void cache_grow_end(struct kmem_cache *cachep, struct page *page)
{
struct kmem_cache_node *n;
void *list = NULL;
check_irq_off();
if (!page)
return;
INIT_LIST_HEAD(&page->slab_list);
n = get_node(cachep, page_to_nid(page));
spin_lock(&n->list_lock);
n->total_slabs++;
if (!page->active) {
list_add_tail(&page->slab_list, &n->slabs_free);
n->free_slabs++;
} else
fixup_slab_list(cachep, n, page, &list);
mm/slab: improve performance of gathering slabinfo stats On large systems, when some slab caches grow to millions of objects (and many gigabytes), running 'cat /proc/slabinfo' can take up to 1-2 seconds. During this time, interrupts are disabled while walking the slab lists (slabs_full, slabs_partial, and slabs_free) for each node, and this sometimes causes timeouts in other drivers (for instance, Infiniband). This patch optimizes 'cat /proc/slabinfo' by maintaining a counter for total number of allocated slabs per node, per cache. This counter is updated when a slab is created or destroyed. This enables us to skip traversing the slabs_full list while gathering slabinfo statistics, and since slabs_full tends to be the biggest list when the cache is large, it results in a dramatic performance improvement. Getting slabinfo statistics now only requires walking the slabs_free and slabs_partial lists, and those lists are usually much smaller than slabs_full. We tested this after growing the dentry cache to 70GB, and the performance improved from 2s to 5ms. Link: http://lkml.kernel.org/r/1472517876-26814-1-git-send-email-aruna.ramakrishna@oracle.com Signed-off-by: Aruna Ramakrishna <aruna.ramakrishna@oracle.com> Acked-by: David Rientjes <rientjes@google.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-28 08:46:32 +08:00
STATS_INC_GROWN(cachep);
n->free_objects += cachep->num - page->active;
spin_unlock(&n->list_lock);
fixup_objfreelist_debug(cachep, &list);
}
#if DEBUG
/*
* Perform extra freeing checks:
* - detect bad pointers.
* - POISON/RED_ZONE checking
*/
static void kfree_debugcheck(const void *objp)
{
if (!virt_addr_valid(objp)) {
pr_err("kfree_debugcheck: out of range ptr %lxh\n",
(unsigned long)objp);
BUG();
}
}
static inline void verify_redzone_free(struct kmem_cache *cache, void *obj)
{
Increase slab redzone to 64bits There are two problems with the existing redzone implementation. Firstly, it's causing misalignment of structures which contain a 64-bit integer, such as netfilter's 'struct ipt_entry' -- causing netfilter modules to fail to load because of the misalignment. (In particular, the first check in net/ipv4/netfilter/ip_tables.c::check_entry_size_and_hooks()) On ppc32 and sparc32, amongst others, __alignof__(uint64_t) == 8. With slab debugging, we use 32-bit redzones. And allocated slab objects aren't sufficiently aligned to hold a structure containing a uint64_t. By _just_ setting ARCH_KMALLOC_MINALIGN to __alignof__(u64) we'd disable redzone checks on those architectures. By using 64-bit redzones we avoid that loss of debugging, and also fix the other problem while we're at it. When investigating this, I noticed that on 64-bit platforms we're using a 32-bit value of RED_ACTIVE/RED_INACTIVE in the 64-bit memory location set aside for the redzone. Which means that the four bytes immediately before or after the allocated object at 0x00,0x00,0x00,0x00 for LE and BE machines, respectively. Which is probably not the most useful choice of poison value. One way to fix both of those at once is just to switch to 64-bit redzones in all cases. Signed-off-by: David Woodhouse <dwmw2@infradead.org> Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Acked-by: David S. Miller <davem@davemloft.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-08 15:22:59 +08:00
unsigned long long redzone1, redzone2;
redzone1 = *dbg_redzone1(cache, obj);
redzone2 = *dbg_redzone2(cache, obj);
/*
* Redzone is ok.
*/
if (redzone1 == RED_ACTIVE && redzone2 == RED_ACTIVE)
return;
if (redzone1 == RED_INACTIVE && redzone2 == RED_INACTIVE)
slab_error(cache, "double free detected");
else
slab_error(cache, "memory outside object was overwritten");
pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
obj, redzone1, redzone2);
}
static void *cache_free_debugcheck(struct kmem_cache *cachep, void *objp,
unsigned long caller)
{
unsigned int objnr;
struct page *page;
BUG_ON(virt_to_cache(objp) != cachep);
objp -= obj_offset(cachep);
kfree_debugcheck(objp);
page = virt_to_head_page(objp);
if (cachep->flags & SLAB_RED_ZONE) {
verify_redzone_free(cachep, objp);
*dbg_redzone1(cachep, objp) = RED_INACTIVE;
*dbg_redzone2(cachep, objp) = RED_INACTIVE;
}
if (cachep->flags & SLAB_STORE_USER)
*dbg_userword(cachep, objp) = (void *)caller;
objnr = obj_to_index(cachep, page, objp);
BUG_ON(objnr >= cachep->num);
BUG_ON(objp != index_to_obj(cachep, page, objnr));
if (cachep->flags & SLAB_POISON) {
poison_obj(cachep, objp, POISON_FREE);
slab_kernel_map(cachep, objp, 0);
}
return objp;
}
#else
#define kfree_debugcheck(x) do { } while(0)
#define cache_free_debugcheck(x,objp,z) (objp)
#endif
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
static inline void fixup_objfreelist_debug(struct kmem_cache *cachep,
void **list)
{
#if DEBUG
void *next = *list;
void *objp;
while (next) {
objp = next - obj_offset(cachep);
next = *(void **)next;
poison_obj(cachep, objp, POISON_FREE);
}
#endif
}
static inline void fixup_slab_list(struct kmem_cache *cachep,
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
struct kmem_cache_node *n, struct page *page,
void **list)
{
/* move slabp to correct slabp list: */
list_del(&page->slab_list);
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (page->active == cachep->num) {
list_add(&page->slab_list, &n->slabs_full);
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
if (OBJFREELIST_SLAB(cachep)) {
#if DEBUG
/* Poisoning will be done without holding the lock */
if (cachep->flags & SLAB_POISON) {
void **objp = page->freelist;
*objp = *list;
*list = objp;
}
#endif
page->freelist = NULL;
}
} else
list_add(&page->slab_list, &n->slabs_partial);
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
/* Try to find non-pfmemalloc slab if needed */
static noinline struct page *get_valid_first_slab(struct kmem_cache_node *n,
struct page *page, bool pfmemalloc)
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
{
if (!page)
return NULL;
if (pfmemalloc)
return page;
if (!PageSlabPfmemalloc(page))
return page;
/* No need to keep pfmemalloc slab if we have enough free objects */
if (n->free_objects > n->free_limit) {
ClearPageSlabPfmemalloc(page);
return page;
}
/* Move pfmemalloc slab to the end of list to speed up next search */
list_del(&page->slab_list);
if (!page->active) {
list_add_tail(&page->slab_list, &n->slabs_free);
n->free_slabs++;
} else
list_add_tail(&page->slab_list, &n->slabs_partial);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
list_for_each_entry(page, &n->slabs_partial, slab_list) {
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
if (!PageSlabPfmemalloc(page))
return page;
}
n->free_touched = 1;
list_for_each_entry(page, &n->slabs_free, slab_list) {
if (!PageSlabPfmemalloc(page)) {
n->free_slabs--;
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
return page;
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
}
return NULL;
}
static struct page *get_first_slab(struct kmem_cache_node *n, bool pfmemalloc)
{
struct page *page;
assert_spin_locked(&n->list_lock);
page = list_first_entry_or_null(&n->slabs_partial, struct page,
slab_list);
if (!page) {
n->free_touched = 1;
page = list_first_entry_or_null(&n->slabs_free, struct page,
slab_list);
if (page)
n->free_slabs--;
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
if (sk_memalloc_socks())
page = get_valid_first_slab(n, page, pfmemalloc);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
return page;
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
static noinline void *cache_alloc_pfmemalloc(struct kmem_cache *cachep,
struct kmem_cache_node *n, gfp_t flags)
{
struct page *page;
void *obj;
void *list = NULL;
if (!gfp_pfmemalloc_allowed(flags))
return NULL;
spin_lock(&n->list_lock);
page = get_first_slab(n, true);
if (!page) {
spin_unlock(&n->list_lock);
return NULL;
}
obj = slab_get_obj(cachep, page);
n->free_objects--;
fixup_slab_list(cachep, n, page, &list);
spin_unlock(&n->list_lock);
fixup_objfreelist_debug(cachep, &list);
return obj;
}
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
/*
* Slab list should be fixed up by fixup_slab_list() for existing slab
* or cache_grow_end() for new slab
*/
static __always_inline int alloc_block(struct kmem_cache *cachep,
struct array_cache *ac, struct page *page, int batchcount)
{
/*
* There must be at least one object available for
* allocation.
*/
BUG_ON(page->active >= cachep->num);
while (page->active < cachep->num && batchcount--) {
STATS_INC_ALLOCED(cachep);
STATS_INC_ACTIVE(cachep);
STATS_SET_HIGH(cachep);
ac->entry[ac->avail++] = slab_get_obj(cachep, page);
}
return batchcount;
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
static void *cache_alloc_refill(struct kmem_cache *cachep, gfp_t flags)
{
int batchcount;
struct kmem_cache_node *n;
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
struct array_cache *ac, *shared;
int node;
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
void *list = NULL;
struct page *page;
check_irq_off();
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
node = numa_mem_id();
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
ac = cpu_cache_get(cachep);
batchcount = ac->batchcount;
if (!ac->touched && batchcount > BATCHREFILL_LIMIT) {
/*
* If there was little recent activity on this cache, then
* perform only a partial refill. Otherwise we could generate
* refill bouncing.
*/
batchcount = BATCHREFILL_LIMIT;
}
n = get_node(cachep, node);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
BUG_ON(ac->avail > 0 || !n);
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
shared = READ_ONCE(n->shared);
if (!n->free_objects && (!shared || !shared->avail))
goto direct_grow;
spin_lock(&n->list_lock);
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
shared = READ_ONCE(n->shared);
/* See if we can refill from the shared array */
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
if (shared && transfer_objects(ac, shared, batchcount)) {
shared->touched = 1;
goto alloc_done;
}
while (batchcount > 0) {
/* Get slab alloc is to come from. */
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
page = get_first_slab(n, false);
if (!page)
goto must_grow;
check_spinlock_acquired(cachep);
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
batchcount = alloc_block(cachep, ac, page, batchcount);
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
fixup_slab_list(cachep, n, page, &list);
}
must_grow:
n->free_objects -= ac->avail;
alloc_done:
spin_unlock(&n->list_lock);
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
fixup_objfreelist_debug(cachep, &list);
mm/slab: lockless decision to grow cache To check whether free objects exist or not precisely, we need to grab a lock. But, accuracy isn't that important because race window would be even small and if there is too much free object, cache reaper would reap it. So, this patch makes the check for free object exisistence not to hold a lock. This will reduce lock contention in heavily allocation case. Note that until now, n->shared can be freed during the processing by writing slabinfo, but, with some trick in this patch, we can access it freely within interrupt disabled period. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that allocation performance decreases for the object size up to 128 and it may be due to extra checks in cache_alloc_refill(). But, with considering improvement of free performance, net result looks the same. Result for other size class looks very promising, roughly, 50% performance improvement. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:31 +08:00
direct_grow:
if (unlikely(!ac->avail)) {
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
/* Check if we can use obj in pfmemalloc slab */
if (sk_memalloc_socks()) {
void *obj = cache_alloc_pfmemalloc(cachep, n, flags);
if (obj)
return obj;
}
page = cache_grow_begin(cachep, gfp_exact_node(flags), node);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
/*
* cache_grow_begin() can reenable interrupts,
* then ac could change.
*/
ac = cpu_cache_get(cachep);
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
if (!ac->avail && page)
alloc_block(cachep, ac, page, batchcount);
cache_grow_end(cachep, page);
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
if (!ac->avail)
return NULL;
}
ac->touched = 1;
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
return ac->entry[--ac->avail];
}
static inline void cache_alloc_debugcheck_before(struct kmem_cache *cachep,
gfp_t flags)
{
mm, page_alloc: distinguish between being unable to sleep, unwilling to sleep and avoiding waking kswapd __GFP_WAIT has been used to identify atomic context in callers that hold spinlocks or are in interrupts. They are expected to be high priority and have access one of two watermarks lower than "min" which can be referred to as the "atomic reserve". __GFP_HIGH users get access to the first lower watermark and can be called the "high priority reserve". Over time, callers had a requirement to not block when fallback options were available. Some have abused __GFP_WAIT leading to a situation where an optimisitic allocation with a fallback option can access atomic reserves. This patch uses __GFP_ATOMIC to identify callers that are truely atomic, cannot sleep and have no alternative. High priority users continue to use __GFP_HIGH. __GFP_DIRECT_RECLAIM identifies callers that can sleep and are willing to enter direct reclaim. __GFP_KSWAPD_RECLAIM to identify callers that want to wake kswapd for background reclaim. __GFP_WAIT is redefined as a caller that is willing to enter direct reclaim and wake kswapd for background reclaim. This patch then converts a number of sites o __GFP_ATOMIC is used by callers that are high priority and have memory pools for those requests. GFP_ATOMIC uses this flag. o Callers that have a limited mempool to guarantee forward progress clear __GFP_DIRECT_RECLAIM but keep __GFP_KSWAPD_RECLAIM. bio allocations fall into this category where kswapd will still be woken but atomic reserves are not used as there is a one-entry mempool to guarantee progress. o Callers that are checking if they are non-blocking should use the helper gfpflags_allow_blocking() where possible. This is because checking for __GFP_WAIT as was done historically now can trigger false positives. Some exceptions like dm-crypt.c exist where the code intent is clearer if __GFP_DIRECT_RECLAIM is used instead of the helper due to flag manipulations. o Callers that built their own GFP flags instead of starting with GFP_KERNEL and friends now also need to specify __GFP_KSWAPD_RECLAIM. The first key hazard to watch out for is callers that removed __GFP_WAIT and was depending on access to atomic reserves for inconspicuous reasons. In some cases it may be appropriate for them to use __GFP_HIGH. The second key hazard is callers that assembled their own combination of GFP flags instead of starting with something like GFP_KERNEL. They may now wish to specify __GFP_KSWAPD_RECLAIM. It's almost certainly harmless if it's missed in most cases as other activity will wake kswapd. Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Michal Hocko <mhocko@suse.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Vitaly Wool <vitalywool@gmail.com> Cc: Rik van Riel <riel@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-07 08:28:21 +08:00
might_sleep_if(gfpflags_allow_blocking(flags));
}
#if DEBUG
static void *cache_alloc_debugcheck_after(struct kmem_cache *cachep,
gfp_t flags, void *objp, unsigned long caller)
{
WARN_ON_ONCE(cachep->ctor && (flags & __GFP_ZERO));
if (!objp)
return objp;
if (cachep->flags & SLAB_POISON) {
check_poison_obj(cachep, objp);
slab_kernel_map(cachep, objp, 1);
poison_obj(cachep, objp, POISON_INUSE);
}
if (cachep->flags & SLAB_STORE_USER)
*dbg_userword(cachep, objp) = (void *)caller;
if (cachep->flags & SLAB_RED_ZONE) {
if (*dbg_redzone1(cachep, objp) != RED_INACTIVE ||
*dbg_redzone2(cachep, objp) != RED_INACTIVE) {
slab_error(cachep, "double free, or memory outside object was overwritten");
pr_err("%px: redzone 1:0x%llx, redzone 2:0x%llx\n",
objp, *dbg_redzone1(cachep, objp),
*dbg_redzone2(cachep, objp));
}
*dbg_redzone1(cachep, objp) = RED_ACTIVE;
*dbg_redzone2(cachep, objp) = RED_ACTIVE;
}
slab: fix oops when reading /proc/slab_allocators Commit b1cb0982bdd6 ("change the management method of free objects of the slab") introduced a bug on slab leak detector ('/proc/slab_allocators'). This detector works like as following decription. 1. traverse all objects on all the slabs. 2. determine whether it is active or not. 3. if active, print who allocate this object. but that commit changed the way how to manage free objects, so the logic determining whether it is active or not is also changed. In before, we regard object in cpu caches as inactive one, but, with this commit, we mistakenly regard object in cpu caches as active one. This intoduces kernel oops if DEBUG_PAGEALLOC is enabled. If DEBUG_PAGEALLOC is enabled, kernel_map_pages() is used to detect who corrupt free memory in the slab. It unmaps page table mapping if object is free and map it if object is active. When slab leak detector check object in cpu caches, it mistakenly think this object active so try to access object memory to retrieve caller of allocation. At this point, page table mapping to this object doesn't exist, so oops occurs. Following is oops message reported from Dave. It blew up when something tried to read /proc/slab_allocators (Just cat it, and you should see the oops below) Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC Modules linked in: [snip...] CPU: 1 PID: 9386 Comm: trinity-c33 Not tainted 3.14.0-rc5+ #131 task: ffff8801aa46e890 ti: ffff880076924000 task.ti: ffff880076924000 RIP: 0010:[<ffffffffaa1a8f4a>] [<ffffffffaa1a8f4a>] handle_slab+0x8a/0x180 RSP: 0018:ffff880076925de0 EFLAGS: 00010002 RAX: 0000000000001000 RBX: 0000000000000000 RCX: 000000005ce85ce7 RDX: ffffea00079be100 RSI: 0000000000001000 RDI: ffff880107458000 RBP: ffff880076925e18 R08: 0000000000000001 R09: 0000000000000000 R10: 0000000000000000 R11: 000000000000000f R12: ffff8801e6f84000 R13: ffffea00079be100 R14: ffff880107458000 R15: ffff88022bb8d2c0 FS: 00007fb769e45740(0000) GS:ffff88024d040000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: ffff8801e6f84ff8 CR3: 00000000a22db000 CR4: 00000000001407e0 DR0: 0000000002695000 DR1: 0000000002695000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000070602 Call Trace: leaks_show+0xce/0x240 seq_read+0x28e/0x490 proc_reg_read+0x3d/0x80 vfs_read+0x9b/0x160 SyS_read+0x58/0xb0 tracesys+0xd4/0xd9 Code: f5 00 00 00 0f 1f 44 00 00 48 63 c8 44 3b 0c 8a 0f 84 e3 00 00 00 83 c0 01 44 39 c0 72 eb 41 f6 47 1a 01 0f 84 e9 00 00 00 89 f0 <4d> 8b 4c 04 f8 4d 85 c9 0f 84 88 00 00 00 49 8b 7e 08 4d 8d 46 RIP handle_slab+0x8a/0x180 To fix the problem, I introduce an object status buffer on each slab. With this, we can track object status precisely, so slab leak detector would not access active object and no kernel oops would occur. Memory overhead caused by this fix is only imposed to CONFIG_DEBUG_SLAB_LEAK which is mainly used for debugging, so memory overhead isn't big problem. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Reported-by: Dave Jones <davej@redhat.com> Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Reviewed-by: Vladimir Davydov <vdavydov@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-24 04:22:06 +08:00
objp += obj_offset(cachep);
if (cachep->ctor && cachep->flags & SLAB_POISON)
cachep->ctor(objp);
if (ARCH_SLAB_MINALIGN &&
((unsigned long)objp & (ARCH_SLAB_MINALIGN-1))) {
pr_err("0x%px: not aligned to ARCH_SLAB_MINALIGN=%d\n",
objp, (int)ARCH_SLAB_MINALIGN);
}
return objp;
}
#else
#define cache_alloc_debugcheck_after(a,b,objp,d) (objp)
#endif
static inline void *____cache_alloc(struct kmem_cache *cachep, gfp_t flags)
{
void *objp;
struct array_cache *ac;
check_irq_off();
ac = cpu_cache_get(cachep);
if (likely(ac->avail)) {
ac->touched = 1;
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
objp = ac->entry[--ac->avail];
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
STATS_INC_ALLOCHIT(cachep);
goto out;
}
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
STATS_INC_ALLOCMISS(cachep);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
objp = cache_alloc_refill(cachep, flags);
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
/*
* the 'ac' may be updated by cache_alloc_refill(),
* and kmemleak_erase() requires its correct value.
*/
ac = cpu_cache_get(cachep);
out:
/*
* To avoid a false negative, if an object that is in one of the
* per-CPU caches is leaked, we need to make sure kmemleak doesn't
* treat the array pointers as a reference to the object.
*/
if (objp)
kmemleak_erase(&ac->entry[ac->avail]);
return objp;
}
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
#ifdef CONFIG_NUMA
/*
* Try allocating on another node if PFA_SPREAD_SLAB is a mempolicy is set.
*
* If we are in_interrupt, then process context, including cpusets and
* mempolicy, may not apply and should not be used for allocation policy.
*/
static void *alternate_node_alloc(struct kmem_cache *cachep, gfp_t flags)
{
int nid_alloc, nid_here;
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
if (in_interrupt() || (flags & __GFP_THISNODE))
return NULL;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
nid_alloc = nid_here = numa_mem_id();
if (cpuset_do_slab_mem_spread() && (cachep->flags & SLAB_MEM_SPREAD))
cpusets: new round-robin rotor for SLAB allocations We have observed several workloads running on multi-node systems where memory is assigned unevenly across the nodes in the system. There are numerous reasons for this but one is the round-robin rotor in cpuset_mem_spread_node(). For example, a simple test that writes a multi-page file will allocate pages on nodes 0 2 4 6 ... Odd nodes are skipped. (Sometimes it allocates on odd nodes & skips even nodes). An example is shown below. The program "lfile" writes a file consisting of 10 pages. The program then mmaps the file & uses get_mempolicy(..., MPOL_F_NODE) to determine the nodes where the file pages were allocated. The output is shown below: # ./lfile allocated on nodes: 2 4 6 0 1 2 6 0 2 There is a single rotor that is used for allocating both file pages & slab pages. Writing the file allocates both a data page & a slab page (buffer_head). This advances the RR rotor 2 nodes for each page allocated. A quick confirmation seems to confirm this is the cause of the uneven allocation: # echo 0 >/dev/cpuset/memory_spread_slab # ./lfile allocated on nodes: 6 7 8 9 0 1 2 3 4 5 This patch introduces a second rotor that is used for slab allocations. Signed-off-by: Jack Steiner <steiner@sgi.com> Acked-by: Christoph Lameter <cl@linux-foundation.org> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Paul Menage <menage@google.com> Cc: Jack Steiner <steiner@sgi.com> Cc: Robin Holt <holt@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:42:49 +08:00
nid_alloc = cpuset_slab_spread_node();
else if (current->mempolicy)
nid_alloc = mempolicy_slab_node();
if (nid_alloc != nid_here)
return ____cache_alloc_node(cachep, flags, nid_alloc);
return NULL;
}
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
/*
* Fallback function if there was no memory available and no objects on a
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
* certain node and fall back is permitted. First we scan all the
* available node for available objects. If that fails then we
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
* perform an allocation without specifying a node. This allows the page
* allocator to do its reclaim / fallback magic. We then insert the
* slab into the proper nodelist and then allocate from it.
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
*/
static void *fallback_alloc(struct kmem_cache *cache, gfp_t flags)
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
{
struct zonelist *zonelist;
mm: have zonelist contains structs with both a zone pointer and zone_idx Filtering zonelists requires very frequent use of zone_idx(). This is costly as it involves a lookup of another structure and a substraction operation. As the zone_idx is often required, it should be quickly accessible. The node idx could also be stored here if it was found that accessing zone->node is significant which may be the case on workloads where nodemasks are heavily used. This patch introduces a struct zoneref to store a zone pointer and a zone index. The zonelist then consists of an array of these struct zonerefs which are looked up as necessary. Helpers are given for accessing the zone index as well as the node index. [kamezawa.hiroyu@jp.fujitsu.com: Suggested struct zoneref instead of embedding information in pointers] [hugh@veritas.com: mm-have-zonelist: fix memcg ooms] [hugh@veritas.com: just return do_try_to_free_pages] [hugh@veritas.com: do_try_to_free_pages gfp_mask redundant] Signed-off-by: Mel Gorman <mel@csn.ul.ie> Acked-by: Christoph Lameter <clameter@sgi.com> Acked-by: David Rientjes <rientjes@google.com> Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <clameter@sgi.com> Cc: Nick Piggin <nickpiggin@yahoo.com.au> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-28 17:12:17 +08:00
struct zoneref *z;
struct zone *zone;
enum zone_type high_zoneidx = gfp_zone(flags);
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
void *obj = NULL;
struct page *page;
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
int nid;
cpuset: mm: reduce large amounts of memory barrier related damage v3 Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") wins a super prize for the largest number of memory barriers entered into fast paths for one commit. [get|put]_mems_allowed is incredibly heavy with pairs of full memory barriers inserted into a number of hot paths. This was detected while investigating at large page allocator slowdown introduced some time after 2.6.32. The largest portion of this overhead was shown by oprofile to be at an mfence introduced by this commit into the page allocator hot path. For extra style points, the commit introduced the use of yield() in an implementation of what looks like a spinning mutex. This patch replaces the full memory barriers on both read and write sides with a sequence counter with just read barriers on the fast path side. This is much cheaper on some architectures, including x86. The main bulk of the patch is the retry logic if the nodemask changes in a manner that can cause a false failure. While updating the nodemask, a check is made to see if a false failure is a risk. If it is, the sequence number gets bumped and parallel allocators will briefly stall while the nodemask update takes place. In a page fault test microbenchmark, oprofile samples from __alloc_pages_nodemask went from 4.53% of all samples to 1.15%. The actual results were 3.3.0-rc3 3.3.0-rc3 rc3-vanilla nobarrier-v2r1 Clients 1 UserTime 0.07 ( 0.00%) 0.08 (-14.19%) Clients 2 UserTime 0.07 ( 0.00%) 0.07 ( 2.72%) Clients 4 UserTime 0.08 ( 0.00%) 0.07 ( 3.29%) Clients 1 SysTime 0.70 ( 0.00%) 0.65 ( 6.65%) Clients 2 SysTime 0.85 ( 0.00%) 0.82 ( 3.65%) Clients 4 SysTime 1.41 ( 0.00%) 1.41 ( 0.32%) Clients 1 WallTime 0.77 ( 0.00%) 0.74 ( 4.19%) Clients 2 WallTime 0.47 ( 0.00%) 0.45 ( 3.73%) Clients 4 WallTime 0.38 ( 0.00%) 0.37 ( 1.58%) Clients 1 Flt/sec/cpu 497620.28 ( 0.00%) 520294.53 ( 4.56%) Clients 2 Flt/sec/cpu 414639.05 ( 0.00%) 429882.01 ( 3.68%) Clients 4 Flt/sec/cpu 257959.16 ( 0.00%) 258761.48 ( 0.31%) Clients 1 Flt/sec 495161.39 ( 0.00%) 517292.87 ( 4.47%) Clients 2 Flt/sec 820325.95 ( 0.00%) 850289.77 ( 3.65%) Clients 4 Flt/sec 1020068.93 ( 0.00%) 1022674.06 ( 0.26%) MMTests Statistics: duration Sys Time Running Test (seconds) 135.68 132.17 User+Sys Time Running Test (seconds) 164.2 160.13 Total Elapsed Time (seconds) 123.46 120.87 The overall improvement is small but the System CPU time is much improved and roughly in correlation to what oprofile reported (these performance figures are without profiling so skew is expected). The actual number of page faults is noticeably improved. For benchmarks like kernel builds, the overall benefit is marginal but the system CPU time is slightly reduced. To test the actual bug the commit fixed I opened two terminals. The first ran within a cpuset and continually ran a small program that faulted 100M of anonymous data. In a second window, the nodemask of the cpuset was continually randomised in a loop. Without the commit, the program would fail every so often (usually within 10 seconds) and obviously with the commit everything worked fine. With this patch applied, it also worked fine so the fix should be functionally equivalent. Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: Miao Xie <miaox@cn.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-22 07:34:11 +08:00
unsigned int cpuset_mems_cookie;
if (flags & __GFP_THISNODE)
return NULL;
cpuset: mm: reduce large amounts of memory barrier related damage v3 Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") wins a super prize for the largest number of memory barriers entered into fast paths for one commit. [get|put]_mems_allowed is incredibly heavy with pairs of full memory barriers inserted into a number of hot paths. This was detected while investigating at large page allocator slowdown introduced some time after 2.6.32. The largest portion of this overhead was shown by oprofile to be at an mfence introduced by this commit into the page allocator hot path. For extra style points, the commit introduced the use of yield() in an implementation of what looks like a spinning mutex. This patch replaces the full memory barriers on both read and write sides with a sequence counter with just read barriers on the fast path side. This is much cheaper on some architectures, including x86. The main bulk of the patch is the retry logic if the nodemask changes in a manner that can cause a false failure. While updating the nodemask, a check is made to see if a false failure is a risk. If it is, the sequence number gets bumped and parallel allocators will briefly stall while the nodemask update takes place. In a page fault test microbenchmark, oprofile samples from __alloc_pages_nodemask went from 4.53% of all samples to 1.15%. The actual results were 3.3.0-rc3 3.3.0-rc3 rc3-vanilla nobarrier-v2r1 Clients 1 UserTime 0.07 ( 0.00%) 0.08 (-14.19%) Clients 2 UserTime 0.07 ( 0.00%) 0.07 ( 2.72%) Clients 4 UserTime 0.08 ( 0.00%) 0.07 ( 3.29%) Clients 1 SysTime 0.70 ( 0.00%) 0.65 ( 6.65%) Clients 2 SysTime 0.85 ( 0.00%) 0.82 ( 3.65%) Clients 4 SysTime 1.41 ( 0.00%) 1.41 ( 0.32%) Clients 1 WallTime 0.77 ( 0.00%) 0.74 ( 4.19%) Clients 2 WallTime 0.47 ( 0.00%) 0.45 ( 3.73%) Clients 4 WallTime 0.38 ( 0.00%) 0.37 ( 1.58%) Clients 1 Flt/sec/cpu 497620.28 ( 0.00%) 520294.53 ( 4.56%) Clients 2 Flt/sec/cpu 414639.05 ( 0.00%) 429882.01 ( 3.68%) Clients 4 Flt/sec/cpu 257959.16 ( 0.00%) 258761.48 ( 0.31%) Clients 1 Flt/sec 495161.39 ( 0.00%) 517292.87 ( 4.47%) Clients 2 Flt/sec 820325.95 ( 0.00%) 850289.77 ( 3.65%) Clients 4 Flt/sec 1020068.93 ( 0.00%) 1022674.06 ( 0.26%) MMTests Statistics: duration Sys Time Running Test (seconds) 135.68 132.17 User+Sys Time Running Test (seconds) 164.2 160.13 Total Elapsed Time (seconds) 123.46 120.87 The overall improvement is small but the System CPU time is much improved and roughly in correlation to what oprofile reported (these performance figures are without profiling so skew is expected). The actual number of page faults is noticeably improved. For benchmarks like kernel builds, the overall benefit is marginal but the system CPU time is slightly reduced. To test the actual bug the commit fixed I opened two terminals. The first ran within a cpuset and continually ran a small program that faulted 100M of anonymous data. In a second window, the nodemask of the cpuset was continually randomised in a loop. Without the commit, the program would fail every so often (usually within 10 seconds) and obviously with the commit everything worked fine. With this patch applied, it also worked fine so the fix should be functionally equivalent. Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: Miao Xie <miaox@cn.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-22 07:34:11 +08:00
retry_cpuset:
cpuset_mems_cookie = read_mems_allowed_begin();
zonelist = node_zonelist(mempolicy_slab_node(), flags);
cpuset: mm: reduce large amounts of memory barrier related damage v3 Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") wins a super prize for the largest number of memory barriers entered into fast paths for one commit. [get|put]_mems_allowed is incredibly heavy with pairs of full memory barriers inserted into a number of hot paths. This was detected while investigating at large page allocator slowdown introduced some time after 2.6.32. The largest portion of this overhead was shown by oprofile to be at an mfence introduced by this commit into the page allocator hot path. For extra style points, the commit introduced the use of yield() in an implementation of what looks like a spinning mutex. This patch replaces the full memory barriers on both read and write sides with a sequence counter with just read barriers on the fast path side. This is much cheaper on some architectures, including x86. The main bulk of the patch is the retry logic if the nodemask changes in a manner that can cause a false failure. While updating the nodemask, a check is made to see if a false failure is a risk. If it is, the sequence number gets bumped and parallel allocators will briefly stall while the nodemask update takes place. In a page fault test microbenchmark, oprofile samples from __alloc_pages_nodemask went from 4.53% of all samples to 1.15%. The actual results were 3.3.0-rc3 3.3.0-rc3 rc3-vanilla nobarrier-v2r1 Clients 1 UserTime 0.07 ( 0.00%) 0.08 (-14.19%) Clients 2 UserTime 0.07 ( 0.00%) 0.07 ( 2.72%) Clients 4 UserTime 0.08 ( 0.00%) 0.07 ( 3.29%) Clients 1 SysTime 0.70 ( 0.00%) 0.65 ( 6.65%) Clients 2 SysTime 0.85 ( 0.00%) 0.82 ( 3.65%) Clients 4 SysTime 1.41 ( 0.00%) 1.41 ( 0.32%) Clients 1 WallTime 0.77 ( 0.00%) 0.74 ( 4.19%) Clients 2 WallTime 0.47 ( 0.00%) 0.45 ( 3.73%) Clients 4 WallTime 0.38 ( 0.00%) 0.37 ( 1.58%) Clients 1 Flt/sec/cpu 497620.28 ( 0.00%) 520294.53 ( 4.56%) Clients 2 Flt/sec/cpu 414639.05 ( 0.00%) 429882.01 ( 3.68%) Clients 4 Flt/sec/cpu 257959.16 ( 0.00%) 258761.48 ( 0.31%) Clients 1 Flt/sec 495161.39 ( 0.00%) 517292.87 ( 4.47%) Clients 2 Flt/sec 820325.95 ( 0.00%) 850289.77 ( 3.65%) Clients 4 Flt/sec 1020068.93 ( 0.00%) 1022674.06 ( 0.26%) MMTests Statistics: duration Sys Time Running Test (seconds) 135.68 132.17 User+Sys Time Running Test (seconds) 164.2 160.13 Total Elapsed Time (seconds) 123.46 120.87 The overall improvement is small but the System CPU time is much improved and roughly in correlation to what oprofile reported (these performance figures are without profiling so skew is expected). The actual number of page faults is noticeably improved. For benchmarks like kernel builds, the overall benefit is marginal but the system CPU time is slightly reduced. To test the actual bug the commit fixed I opened two terminals. The first ran within a cpuset and continually ran a small program that faulted 100M of anonymous data. In a second window, the nodemask of the cpuset was continually randomised in a loop. Without the commit, the program would fail every so often (usually within 10 seconds) and obviously with the commit everything worked fine. With this patch applied, it also worked fine so the fix should be functionally equivalent. Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: Miao Xie <miaox@cn.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-22 07:34:11 +08:00
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
retry:
/*
* Look through allowed nodes for objects available
* from existing per node queues.
*/
for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
nid = zone_to_nid(zone);
slab: fix cpuset check in fallback_alloc fallback_alloc is called on kmalloc if the preferred node doesn't have free or partial slabs and there's no pages on the node's free list (GFP_THISNODE allocations fail). Before invoking the reclaimer it tries to locate a free or partial slab on other allowed nodes' lists. While iterating over the preferred node's zonelist it skips those zones which hardwall cpuset check returns false for. That means that for a task bound to a specific node using cpusets fallback_alloc will always ignore free slabs on other nodes and go directly to the reclaimer, which, however, may allocate from other nodes if cpuset.mem_hardwall is unset (default). As a result, we may get lists of free slabs grow without bounds on other nodes, which is bad, because inactive slabs are only evicted by cache_reap at a very slow rate and cannot be dropped forcefully. To reproduce the issue, run a process that will walk over a directory tree with lots of files inside a cpuset bound to a node that constantly experiences memory pressure. Look at num_slabs vs active_slabs growth as reported by /proc/slabinfo. To avoid this we should use softwall cpuset check in fallback_alloc. Signed-off-by: Vladimir Davydov <vdavydov@parallels.com> Acked-by: Zefan Li <lizefan@huawei.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-12-13 08:58:25 +08:00
if (cpuset_zone_allowed(zone, flags) &&
get_node(cache, nid) &&
get_node(cache, nid)->free_objects) {
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
obj = ____cache_alloc_node(cache,
mm: remove GFP_THISNODE NOTE: this is not about __GFP_THISNODE, this is only about GFP_THISNODE. GFP_THISNODE is a secret combination of gfp bits that have different behavior than expected. It is a combination of __GFP_THISNODE, __GFP_NORETRY, and __GFP_NOWARN and is special-cased in the page allocator slowpath to fail without trying reclaim even though it may be used in combination with __GFP_WAIT. An example of the problem this creates: commit e97ca8e5b864 ("mm: fix GFP_THISNODE callers and clarify") fixed up many users of GFP_THISNODE that really just wanted __GFP_THISNODE. The problem doesn't end there, however, because even it was a no-op for alloc_misplaced_dst_page(), which also sets __GFP_NORETRY and __GFP_NOWARN, and migrate_misplaced_transhuge_page(), where __GFP_NORETRY and __GFP_NOWAIT is set in GFP_TRANSHUGE. Converting GFP_THISNODE to __GFP_THISNODE is a no-op in these cases since the page allocator special-cases __GFP_THISNODE && __GFP_NORETRY && __GFP_NOWARN. It's time to just remove GFP_THISNODE entirely. We leave __GFP_THISNODE to restrict an allocation to a local node, but remove GFP_THISNODE and its obscurity. Instead, we require that a caller clear __GFP_WAIT if it wants to avoid reclaim. This allows the aforementioned functions to actually reclaim as they should. It also enables any future callers that want to do __GFP_THISNODE but also __GFP_NORETRY && __GFP_NOWARN to reclaim. The rule is simple: if you don't want to reclaim, then don't set __GFP_WAIT. Aside: ovs_flow_stats_update() really wants to avoid reclaim as well, so it is unchanged. Signed-off-by: David Rientjes <rientjes@google.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Christoph Lameter <cl@linux.com> Acked-by: Pekka Enberg <penberg@kernel.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Pravin Shelar <pshelar@nicira.com> Cc: Jarno Rajahalme <jrajahalme@nicira.com> Cc: Li Zefan <lizefan@huawei.com> Cc: Greg Thelen <gthelen@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-04-15 06:46:55 +08:00
gfp_exact_node(flags), nid);
if (obj)
break;
}
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
}
if (!obj) {
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
/*
* This allocation will be performed within the constraints
* of the current cpuset / memory policy requirements.
* We may trigger various forms of reclaim on the allowed
* set and go into memory reserves if necessary.
*/
page = cache_grow_begin(cache, flags, numa_mem_id());
cache_grow_end(cache, page);
if (page) {
nid = page_to_nid(page);
obj = ____cache_alloc_node(cache,
gfp_exact_node(flags), nid);
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
/*
* Another processor may allocate the objects in
* the slab since we are not holding any locks.
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
*/
if (!obj)
goto retry;
[PATCH] slab: better fallback allocation behavior Currently we simply attempt to allocate from all allowed nodes using GFP_THISNODE. However, GFP_THISNODE does not do reclaim (it wont do any at all if the recent GFP_THISNODE patch is accepted). If we truly run out of memory in the whole system then fallback_alloc may return NULL although memory may still be available if we would perform more thorough reclaim. This patch changes fallback_alloc() so that we first only inspect all the per node queues for available slabs. If we find any then we allocate from those. This avoids slab fragmentation by first getting rid of all partial allocated slabs on every node before allocating new memory. If we cannot satisfy the allocation from any per node queue then we extend a slab. We now call into the page allocator without specifying GFP_THISNODE. The page allocator will then implement its own fallback (in the given cpuset context), perform necessary reclaim (again considering not a single node but the whole set of allowed nodes) and then return pages for a new slab. We identify from which node the pages were allocated and then insert the pages into the corresponding per node structure. In order to do so we need to modify cache_grow() to take a parameter that specifies the new slab. kmem_getpages() can no longer set the GFP_THISNODE flag since we need to be able to use kmem_getpage to allocate from an arbitrary node. GFP_THISNODE needs to be specified when calling cache_grow(). One key advantage is that the decision from which node to allocate new memory is removed from slab fallback processing. The patch allows to go back to use of the page allocators fallback/reclaim logic. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 12:33:29 +08:00
}
}
cpuset: mm: reduce large amounts of memory barrier related damage v3 Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") wins a super prize for the largest number of memory barriers entered into fast paths for one commit. [get|put]_mems_allowed is incredibly heavy with pairs of full memory barriers inserted into a number of hot paths. This was detected while investigating at large page allocator slowdown introduced some time after 2.6.32. The largest portion of this overhead was shown by oprofile to be at an mfence introduced by this commit into the page allocator hot path. For extra style points, the commit introduced the use of yield() in an implementation of what looks like a spinning mutex. This patch replaces the full memory barriers on both read and write sides with a sequence counter with just read barriers on the fast path side. This is much cheaper on some architectures, including x86. The main bulk of the patch is the retry logic if the nodemask changes in a manner that can cause a false failure. While updating the nodemask, a check is made to see if a false failure is a risk. If it is, the sequence number gets bumped and parallel allocators will briefly stall while the nodemask update takes place. In a page fault test microbenchmark, oprofile samples from __alloc_pages_nodemask went from 4.53% of all samples to 1.15%. The actual results were 3.3.0-rc3 3.3.0-rc3 rc3-vanilla nobarrier-v2r1 Clients 1 UserTime 0.07 ( 0.00%) 0.08 (-14.19%) Clients 2 UserTime 0.07 ( 0.00%) 0.07 ( 2.72%) Clients 4 UserTime 0.08 ( 0.00%) 0.07 ( 3.29%) Clients 1 SysTime 0.70 ( 0.00%) 0.65 ( 6.65%) Clients 2 SysTime 0.85 ( 0.00%) 0.82 ( 3.65%) Clients 4 SysTime 1.41 ( 0.00%) 1.41 ( 0.32%) Clients 1 WallTime 0.77 ( 0.00%) 0.74 ( 4.19%) Clients 2 WallTime 0.47 ( 0.00%) 0.45 ( 3.73%) Clients 4 WallTime 0.38 ( 0.00%) 0.37 ( 1.58%) Clients 1 Flt/sec/cpu 497620.28 ( 0.00%) 520294.53 ( 4.56%) Clients 2 Flt/sec/cpu 414639.05 ( 0.00%) 429882.01 ( 3.68%) Clients 4 Flt/sec/cpu 257959.16 ( 0.00%) 258761.48 ( 0.31%) Clients 1 Flt/sec 495161.39 ( 0.00%) 517292.87 ( 4.47%) Clients 2 Flt/sec 820325.95 ( 0.00%) 850289.77 ( 3.65%) Clients 4 Flt/sec 1020068.93 ( 0.00%) 1022674.06 ( 0.26%) MMTests Statistics: duration Sys Time Running Test (seconds) 135.68 132.17 User+Sys Time Running Test (seconds) 164.2 160.13 Total Elapsed Time (seconds) 123.46 120.87 The overall improvement is small but the System CPU time is much improved and roughly in correlation to what oprofile reported (these performance figures are without profiling so skew is expected). The actual number of page faults is noticeably improved. For benchmarks like kernel builds, the overall benefit is marginal but the system CPU time is slightly reduced. To test the actual bug the commit fixed I opened two terminals. The first ran within a cpuset and continually ran a small program that faulted 100M of anonymous data. In a second window, the nodemask of the cpuset was continually randomised in a loop. Without the commit, the program would fail every so often (usually within 10 seconds) and obviously with the commit everything worked fine. With this patch applied, it also worked fine so the fix should be functionally equivalent. Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: Miao Xie <miaox@cn.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-22 07:34:11 +08:00
if (unlikely(!obj && read_mems_allowed_retry(cpuset_mems_cookie)))
cpuset: mm: reduce large amounts of memory barrier related damage v3 Commit c0ff7453bb5c ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") wins a super prize for the largest number of memory barriers entered into fast paths for one commit. [get|put]_mems_allowed is incredibly heavy with pairs of full memory barriers inserted into a number of hot paths. This was detected while investigating at large page allocator slowdown introduced some time after 2.6.32. The largest portion of this overhead was shown by oprofile to be at an mfence introduced by this commit into the page allocator hot path. For extra style points, the commit introduced the use of yield() in an implementation of what looks like a spinning mutex. This patch replaces the full memory barriers on both read and write sides with a sequence counter with just read barriers on the fast path side. This is much cheaper on some architectures, including x86. The main bulk of the patch is the retry logic if the nodemask changes in a manner that can cause a false failure. While updating the nodemask, a check is made to see if a false failure is a risk. If it is, the sequence number gets bumped and parallel allocators will briefly stall while the nodemask update takes place. In a page fault test microbenchmark, oprofile samples from __alloc_pages_nodemask went from 4.53% of all samples to 1.15%. The actual results were 3.3.0-rc3 3.3.0-rc3 rc3-vanilla nobarrier-v2r1 Clients 1 UserTime 0.07 ( 0.00%) 0.08 (-14.19%) Clients 2 UserTime 0.07 ( 0.00%) 0.07 ( 2.72%) Clients 4 UserTime 0.08 ( 0.00%) 0.07 ( 3.29%) Clients 1 SysTime 0.70 ( 0.00%) 0.65 ( 6.65%) Clients 2 SysTime 0.85 ( 0.00%) 0.82 ( 3.65%) Clients 4 SysTime 1.41 ( 0.00%) 1.41 ( 0.32%) Clients 1 WallTime 0.77 ( 0.00%) 0.74 ( 4.19%) Clients 2 WallTime 0.47 ( 0.00%) 0.45 ( 3.73%) Clients 4 WallTime 0.38 ( 0.00%) 0.37 ( 1.58%) Clients 1 Flt/sec/cpu 497620.28 ( 0.00%) 520294.53 ( 4.56%) Clients 2 Flt/sec/cpu 414639.05 ( 0.00%) 429882.01 ( 3.68%) Clients 4 Flt/sec/cpu 257959.16 ( 0.00%) 258761.48 ( 0.31%) Clients 1 Flt/sec 495161.39 ( 0.00%) 517292.87 ( 4.47%) Clients 2 Flt/sec 820325.95 ( 0.00%) 850289.77 ( 3.65%) Clients 4 Flt/sec 1020068.93 ( 0.00%) 1022674.06 ( 0.26%) MMTests Statistics: duration Sys Time Running Test (seconds) 135.68 132.17 User+Sys Time Running Test (seconds) 164.2 160.13 Total Elapsed Time (seconds) 123.46 120.87 The overall improvement is small but the System CPU time is much improved and roughly in correlation to what oprofile reported (these performance figures are without profiling so skew is expected). The actual number of page faults is noticeably improved. For benchmarks like kernel builds, the overall benefit is marginal but the system CPU time is slightly reduced. To test the actual bug the commit fixed I opened two terminals. The first ran within a cpuset and continually ran a small program that faulted 100M of anonymous data. In a second window, the nodemask of the cpuset was continually randomised in a loop. Without the commit, the program would fail every so often (usually within 10 seconds) and obviously with the commit everything worked fine. With this patch applied, it also worked fine so the fix should be functionally equivalent. Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: Miao Xie <miaox@cn.fujitsu.com> Cc: David Rientjes <rientjes@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-22 07:34:11 +08:00
goto retry_cpuset;
[PATCH] GFP_THISNODE for the slab allocator This patch insures that the slab node lists in the NUMA case only contain slabs that belong to that specific node. All slab allocations use GFP_THISNODE when calling into the page allocator. If an allocation fails then we fall back in the slab allocator according to the zonelists appropriate for a certain context. This allows a replication of the behavior of alloc_pages and alloc_pages node in the slab layer. Currently allocations requested from the page allocator may be redirected via cpusets to other nodes. This results in remote pages on nodelists and that in turn results in interrupt latency issues during cache draining. Plus the slab is handing out memory as local when it is really remote. Fallback for slab memory allocations will occur within the slab allocator and not in the page allocator. This is necessary in order to be able to use the existing pools of objects on the nodes that we fall back to before adding more pages to a slab. The fallback function insures that the nodes we fall back to obey cpuset restrictions of the current context. We do not allocate objects from outside of the current cpuset context like before. Note that the implementation of locality constraints within the slab allocator requires importing logic from the page allocator. This is a mischmash that is not that great. Other allocators (uncached allocator, vmalloc, huge pages) face similar problems and have similar minimal reimplementations of the basic fallback logic of the page allocator. There is another way of implementing a slab by avoiding per node lists (see modular slab) but this wont work within the existing slab. V1->V2: - Use NUMA_BUILD to avoid #ifdef CONFIG_NUMA - Exploit GFP_THISNODE being 0 in the NON_NUMA case to avoid another #ifdef [akpm@osdl.org: build fix] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-09-27 16:50:08 +08:00
return obj;
}
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
/*
* A interface to enable slab creation on nodeid
*/
static void *____cache_alloc_node(struct kmem_cache *cachep, gfp_t flags,
int nodeid)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
struct page *page;
struct kmem_cache_node *n;
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
void *obj = NULL;
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
void *list = NULL;
slab: fix nodeid bounds check for non-contiguous node IDs The bounds check for nodeid in ____cache_alloc_node gives false positives on machines where the node IDs are not contiguous, leading to a panic at boot time. For example, on a POWER8 machine the node IDs are typically 0, 1, 16 and 17. This means that num_online_nodes() returns 4, so when ____cache_alloc_node is called with nodeid = 16 the VM_BUG_ON triggers, like this: kernel BUG at /home/paulus/kernel/kvm/mm/slab.c:3079! Call Trace: .____cache_alloc_node+0x5c/0x270 (unreliable) .kmem_cache_alloc_node_trace+0xdc/0x360 .init_list+0x3c/0x128 .kmem_cache_init+0x1dc/0x258 .start_kernel+0x2a0/0x568 start_here_common+0x20/0xa8 To fix this, we instead compare the nodeid with MAX_NUMNODES, and additionally make sure it isn't negative (since nodeid is an int). The check is there mainly to protect the array dereference in the get_node() call in the next line, and the array being dereferenced is of size MAX_NUMNODES. If the nodeid is in range but invalid (for example if the node is off-line), the BUG_ON in the next line will catch that. Fixes: 14e50c6a9bc2 ("mm: slab: Verify the nodeid passed to ____cache_alloc_node") Signed-off-by: Paul Mackerras <paulus@samba.org> Reviewed-by: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> Reviewed-by: Pekka Enberg <penberg@kernel.org> Acked-by: David Rientjes <rientjes@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-12-03 07:59:48 +08:00
VM_BUG_ON(nodeid < 0 || nodeid >= MAX_NUMNODES);
n = get_node(cachep, nodeid);
BUG_ON(!n);
check_irq_off();
spin_lock(&n->list_lock);
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
page = get_first_slab(n, false);
if (!page)
goto must_grow;
check_spinlock_acquired_node(cachep, nodeid);
STATS_INC_NODEALLOCS(cachep);
STATS_INC_ACTIVE(cachep);
STATS_SET_HIGH(cachep);
BUG_ON(page->active == cachep->num);
obj = slab_get_obj(cachep, page);
n->free_objects--;
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
fixup_slab_list(cachep, n, page, &list);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
spin_unlock(&n->list_lock);
mm/slab: introduce new slab management type, OBJFREELIST_SLAB SLAB needs an array to manage freed objects in a slab. It is only used if some objects are freed so we can use free object itself as this array. This requires additional branch in somewhat critical lock path to check if it is first freed object or not but that's all we need. Benefits is that we can save extra memory usage and reduce some computational overhead by allocating a management array when new slab is created. Code change is rather complex than what we can expect from the idea, in order to handle debugging feature efficiently. If you want to see core idea only, please remove '#if DEBUG' block in the patch. Although this idea can apply to all caches whose size is larger than management array size, it isn't applied to caches which have a constructor. If such cache's object is used for management array, constructor should be called for it before that object is returned to user. I guess that overhead overwhelm benefit in that case so this idea doesn't applied to them at least now. For summary, from now on, slab management type is determined by following logic. 1) if management array size is smaller than object size and no ctor, it becomes OBJFREELIST_SLAB. 2) if management array size is smaller than leftover, it becomes NORMAL_SLAB which uses leftover as a array. 3) if OFF_SLAB help to save memory than way 4), it becomes OFF_SLAB. It allocate a management array from the other cache so memory waste happens. 4) others become NORMAL_SLAB. It uses dedicated internal memory in a slab as a management array so it causes memory waste. In my system, without enabling CONFIG_DEBUG_SLAB, Almost caches become OBJFREELIST_SLAB and NORMAL_SLAB (using leftover) which doesn't waste memory. Following is the result of number of caches with specific slab management type. TOTAL = OBJFREELIST + NORMAL(leftover) + NORMAL + OFF /Before/ 126 = 0 + 60 + 25 + 41 /After/ 126 = 97 + 12 + 15 + 2 Result shows that number of caches that doesn't waste memory increase from 60 to 109. I did some benchmarking and it looks that benefit are more than loss. Kmalloc: Repeatedly allocate then free test /Before/ [ 0.286809] 1. Kmalloc: Repeatedly allocate then free test [ 1.143674] 100000 times kmalloc(32) -> 116 cycles kfree -> 78 cycles [ 1.441726] 100000 times kmalloc(64) -> 121 cycles kfree -> 80 cycles [ 1.815734] 100000 times kmalloc(128) -> 168 cycles kfree -> 85 cycles [ 2.380709] 100000 times kmalloc(256) -> 287 cycles kfree -> 95 cycles [ 3.101153] 100000 times kmalloc(512) -> 370 cycles kfree -> 117 cycles [ 3.942432] 100000 times kmalloc(1024) -> 413 cycles kfree -> 156 cycles [ 5.227396] 100000 times kmalloc(2048) -> 622 cycles kfree -> 248 cycles [ 7.519793] 100000 times kmalloc(4096) -> 1102 cycles kfree -> 452 cycles /After/ [ 1.205313] 100000 times kmalloc(32) -> 117 cycles kfree -> 78 cycles [ 1.510526] 100000 times kmalloc(64) -> 124 cycles kfree -> 81 cycles [ 1.827382] 100000 times kmalloc(128) -> 130 cycles kfree -> 84 cycles [ 2.226073] 100000 times kmalloc(256) -> 177 cycles kfree -> 92 cycles [ 2.814747] 100000 times kmalloc(512) -> 286 cycles kfree -> 112 cycles [ 3.532952] 100000 times kmalloc(1024) -> 344 cycles kfree -> 141 cycles [ 4.608777] 100000 times kmalloc(2048) -> 519 cycles kfree -> 210 cycles [ 6.350105] 100000 times kmalloc(4096) -> 789 cycles kfree -> 391 cycles In fact, I tested another idea implementing OBJFREELIST_SLAB with extendable linked array through another freed object. It can remove memory waste completely but it causes more computational overhead in critical lock path and it seems that overhead outweigh benefit. So, this patch doesn't include it. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:50 +08:00
fixup_objfreelist_debug(cachep, &list);
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
return obj;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
must_grow:
spin_unlock(&n->list_lock);
page = cache_grow_begin(cachep, gfp_exact_node(flags), nodeid);
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
if (page) {
/* This slab isn't counted yet so don't update free_objects */
obj = slab_get_obj(cachep, page);
}
cache_grow_end(cachep, page);
mm/slab: refill cpu cache through a new slab without holding a node lock Until now, cache growing makes a free slab on node's slab list and then we can allocate free objects from it. This necessarily requires to hold a node lock which is very contended. If we refill cpu cache before attaching it to node's slab list, we can avoid holding a node lock as much as possible because this newly allocated slab is only visible to the current task. This will reduce lock contention. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=355/750 Kmalloc N*alloc N*free(64): Average=452/812 Kmalloc N*alloc N*free(128): Average=559/1070 Kmalloc N*alloc N*free(256): Average=1176/980 Kmalloc N*alloc N*free(512): Average=1939/1189 Kmalloc N*alloc N*free(1024): Average=3521/1278 Kmalloc N*alloc N*free(2048): Average=7152/1838 Kmalloc N*alloc N*free(4096): Average=13438/2013 * After Kmalloc N*alloc N*free(32): Average=248/966 Kmalloc N*alloc N*free(64): Average=261/949 Kmalloc N*alloc N*free(128): Average=314/1016 Kmalloc N*alloc N*free(256): Average=741/1061 Kmalloc N*alloc N*free(512): Average=1246/1152 Kmalloc N*alloc N*free(1024): Average=2437/1259 Kmalloc N*alloc N*free(2048): Average=4980/1800 Kmalloc N*alloc N*free(4096): Average=9000/2078 It shows that contention is reduced for all the object sizes and performance increases by 30 ~ 40%. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:29 +08:00
return obj ? obj : fallback_alloc(cachep, flags);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
static __always_inline void *
slab_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid,
unsigned long caller)
{
unsigned long save_flags;
void *ptr;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
int slab_node = numa_mem_id();
flags &= gfp_allowed_mask;
cachep = slab_pre_alloc_hook(cachep, flags);
if (unlikely(!cachep))
return NULL;
cache_alloc_debugcheck_before(cachep, flags);
local_irq_save(save_flags);
if (nodeid == NUMA_NO_NODE)
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
nodeid = slab_node;
if (unlikely(!get_node(cachep, nodeid))) {
/* Node not bootstrapped yet */
ptr = fallback_alloc(cachep, flags);
goto out;
}
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
if (nodeid == slab_node) {
/*
* Use the locally cached objects if possible.
* However ____cache_alloc does not allow fallback
* to other nodes. It may fail while we still have
* objects on other nodes available.
*/
ptr = ____cache_alloc(cachep, flags);
if (ptr)
goto out;
}
/* ___cache_alloc_node can fall back to other nodes */
ptr = ____cache_alloc_node(cachep, flags, nodeid);
out:
local_irq_restore(save_flags);
ptr = cache_alloc_debugcheck_after(cachep, flags, ptr, caller);
mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options Patch series "add init_on_alloc/init_on_free boot options", v10. Provide init_on_alloc and init_on_free boot options. These are aimed at preventing possible information leaks and making the control-flow bugs that depend on uninitialized values more deterministic. Enabling either of the options guarantees that the memory returned by the page allocator and SL[AU]B is initialized with zeroes. SLOB allocator isn't supported at the moment, as its emulation of kmem caches complicates handling of SLAB_TYPESAFE_BY_RCU caches correctly. Enabling init_on_free also guarantees that pages and heap objects are initialized right after they're freed, so it won't be possible to access stale data by using a dangling pointer. As suggested by Michal Hocko, right now we don't let the heap users to disable initialization for certain allocations. There's not enough evidence that doing so can speed up real-life cases, and introducing ways to opt-out may result in things going out of control. This patch (of 2): The new options are needed to prevent possible information leaks and make control-flow bugs that depend on uninitialized values more deterministic. This is expected to be on-by-default on Android and Chrome OS. And it gives the opportunity for anyone else to use it under distros too via the boot args. (The init_on_free feature is regularly requested by folks where memory forensics is included in their threat models.) init_on_alloc=1 makes the kernel initialize newly allocated pages and heap objects with zeroes. Initialization is done at allocation time at the places where checks for __GFP_ZERO are performed. init_on_free=1 makes the kernel initialize freed pages and heap objects with zeroes upon their deletion. This helps to ensure sensitive data doesn't leak via use-after-free accesses. Both init_on_alloc=1 and init_on_free=1 guarantee that the allocator returns zeroed memory. The two exceptions are slab caches with constructors and SLAB_TYPESAFE_BY_RCU flag. Those are never zero-initialized to preserve their semantics. Both init_on_alloc and init_on_free default to zero, but those defaults can be overridden with CONFIG_INIT_ON_ALLOC_DEFAULT_ON and CONFIG_INIT_ON_FREE_DEFAULT_ON. If either SLUB poisoning or page poisoning is enabled, those options take precedence over init_on_alloc and init_on_free: initialization is only applied to unpoisoned allocations. Slowdown for the new features compared to init_on_free=0, init_on_alloc=0: hackbench, init_on_free=1: +7.62% sys time (st.err 0.74%) hackbench, init_on_alloc=1: +7.75% sys time (st.err 2.14%) Linux build with -j12, init_on_free=1: +8.38% wall time (st.err 0.39%) Linux build with -j12, init_on_free=1: +24.42% sys time (st.err 0.52%) Linux build with -j12, init_on_alloc=1: -0.13% wall time (st.err 0.42%) Linux build with -j12, init_on_alloc=1: +0.57% sys time (st.err 0.40%) The slowdown for init_on_free=0, init_on_alloc=0 compared to the baseline is within the standard error. The new features are also going to pave the way for hardware memory tagging (e.g. arm64's MTE), which will require both on_alloc and on_free hooks to set the tags for heap objects. With MTE, tagging will have the same cost as memory initialization. Although init_on_free is rather costly, there are paranoid use-cases where in-memory data lifetime is desired to be minimized. There are various arguments for/against the realism of the associated threat models, but given that we'll need the infrastructure for MTE anyway, and there are people who want wipe-on-free behavior no matter what the performance cost, it seems reasonable to include it in this series. [glider@google.com: v8] Link: http://lkml.kernel.org/r/20190626121943.131390-2-glider@google.com [glider@google.com: v9] Link: http://lkml.kernel.org/r/20190627130316.254309-2-glider@google.com [glider@google.com: v10] Link: http://lkml.kernel.org/r/20190628093131.199499-2-glider@google.com Link: http://lkml.kernel.org/r/20190617151050.92663-2-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Acked-by: Kees Cook <keescook@chromium.org> Acked-by: Michal Hocko <mhocko@suse.cz> [page and dmapool parts Acked-by: James Morris <jamorris@linux.microsoft.com>] Cc: Christoph Lameter <cl@linux.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: "Serge E. Hallyn" <serge@hallyn.com> Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Kostya Serebryany <kcc@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Sandeep Patil <sspatil@android.com> Cc: Laura Abbott <labbott@redhat.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Jann Horn <jannh@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marco Elver <elver@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:59:19 +08:00
if (unlikely(slab_want_init_on_alloc(flags, cachep)) && ptr)
memset(ptr, 0, cachep->object_size);
slab_post_alloc_hook(cachep, flags, 1, &ptr);
return ptr;
}
static __always_inline void *
__do_cache_alloc(struct kmem_cache *cache, gfp_t flags)
{
void *objp;
if (current->mempolicy || cpuset_do_slab_mem_spread()) {
objp = alternate_node_alloc(cache, flags);
if (objp)
goto out;
}
objp = ____cache_alloc(cache, flags);
/*
* We may just have run out of memory on the local node.
* ____cache_alloc_node() knows how to locate memory on other nodes
*/
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
if (!objp)
objp = ____cache_alloc_node(cache, flags, numa_mem_id());
out:
return objp;
}
#else
static __always_inline void *
__do_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
{
return ____cache_alloc(cachep, flags);
}
#endif /* CONFIG_NUMA */
static __always_inline void *
slab_alloc(struct kmem_cache *cachep, gfp_t flags, unsigned long caller)
{
unsigned long save_flags;
void *objp;
flags &= gfp_allowed_mask;
cachep = slab_pre_alloc_hook(cachep, flags);
if (unlikely(!cachep))
return NULL;
cache_alloc_debugcheck_before(cachep, flags);
local_irq_save(save_flags);
objp = __do_cache_alloc(cachep, flags);
local_irq_restore(save_flags);
objp = cache_alloc_debugcheck_after(cachep, flags, objp, caller);
prefetchw(objp);
mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options Patch series "add init_on_alloc/init_on_free boot options", v10. Provide init_on_alloc and init_on_free boot options. These are aimed at preventing possible information leaks and making the control-flow bugs that depend on uninitialized values more deterministic. Enabling either of the options guarantees that the memory returned by the page allocator and SL[AU]B is initialized with zeroes. SLOB allocator isn't supported at the moment, as its emulation of kmem caches complicates handling of SLAB_TYPESAFE_BY_RCU caches correctly. Enabling init_on_free also guarantees that pages and heap objects are initialized right after they're freed, so it won't be possible to access stale data by using a dangling pointer. As suggested by Michal Hocko, right now we don't let the heap users to disable initialization for certain allocations. There's not enough evidence that doing so can speed up real-life cases, and introducing ways to opt-out may result in things going out of control. This patch (of 2): The new options are needed to prevent possible information leaks and make control-flow bugs that depend on uninitialized values more deterministic. This is expected to be on-by-default on Android and Chrome OS. And it gives the opportunity for anyone else to use it under distros too via the boot args. (The init_on_free feature is regularly requested by folks where memory forensics is included in their threat models.) init_on_alloc=1 makes the kernel initialize newly allocated pages and heap objects with zeroes. Initialization is done at allocation time at the places where checks for __GFP_ZERO are performed. init_on_free=1 makes the kernel initialize freed pages and heap objects with zeroes upon their deletion. This helps to ensure sensitive data doesn't leak via use-after-free accesses. Both init_on_alloc=1 and init_on_free=1 guarantee that the allocator returns zeroed memory. The two exceptions are slab caches with constructors and SLAB_TYPESAFE_BY_RCU flag. Those are never zero-initialized to preserve their semantics. Both init_on_alloc and init_on_free default to zero, but those defaults can be overridden with CONFIG_INIT_ON_ALLOC_DEFAULT_ON and CONFIG_INIT_ON_FREE_DEFAULT_ON. If either SLUB poisoning or page poisoning is enabled, those options take precedence over init_on_alloc and init_on_free: initialization is only applied to unpoisoned allocations. Slowdown for the new features compared to init_on_free=0, init_on_alloc=0: hackbench, init_on_free=1: +7.62% sys time (st.err 0.74%) hackbench, init_on_alloc=1: +7.75% sys time (st.err 2.14%) Linux build with -j12, init_on_free=1: +8.38% wall time (st.err 0.39%) Linux build with -j12, init_on_free=1: +24.42% sys time (st.err 0.52%) Linux build with -j12, init_on_alloc=1: -0.13% wall time (st.err 0.42%) Linux build with -j12, init_on_alloc=1: +0.57% sys time (st.err 0.40%) The slowdown for init_on_free=0, init_on_alloc=0 compared to the baseline is within the standard error. The new features are also going to pave the way for hardware memory tagging (e.g. arm64's MTE), which will require both on_alloc and on_free hooks to set the tags for heap objects. With MTE, tagging will have the same cost as memory initialization. Although init_on_free is rather costly, there are paranoid use-cases where in-memory data lifetime is desired to be minimized. There are various arguments for/against the realism of the associated threat models, but given that we'll need the infrastructure for MTE anyway, and there are people who want wipe-on-free behavior no matter what the performance cost, it seems reasonable to include it in this series. [glider@google.com: v8] Link: http://lkml.kernel.org/r/20190626121943.131390-2-glider@google.com [glider@google.com: v9] Link: http://lkml.kernel.org/r/20190627130316.254309-2-glider@google.com [glider@google.com: v10] Link: http://lkml.kernel.org/r/20190628093131.199499-2-glider@google.com Link: http://lkml.kernel.org/r/20190617151050.92663-2-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Acked-by: Kees Cook <keescook@chromium.org> Acked-by: Michal Hocko <mhocko@suse.cz> [page and dmapool parts Acked-by: James Morris <jamorris@linux.microsoft.com>] Cc: Christoph Lameter <cl@linux.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: "Serge E. Hallyn" <serge@hallyn.com> Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Kostya Serebryany <kcc@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Sandeep Patil <sspatil@android.com> Cc: Laura Abbott <labbott@redhat.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Jann Horn <jannh@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marco Elver <elver@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:59:19 +08:00
if (unlikely(slab_want_init_on_alloc(flags, cachep)) && objp)
memset(objp, 0, cachep->object_size);
slab_post_alloc_hook(cachep, flags, 1, &objp);
return objp;
}
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
/*
* Caller needs to acquire correct kmem_cache_node's list_lock
* @list: List of detached free slabs should be freed by caller
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
*/
static void free_block(struct kmem_cache *cachep, void **objpp,
int nr_objects, int node, struct list_head *list)
{
int i;
struct kmem_cache_node *n = get_node(cachep, node);
struct page *page;
n->free_objects += nr_objects;
for (i = 0; i < nr_objects; i++) {
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
void *objp;
struct page *page;
mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages When a user or administrator requires swap for their application, they create a swap partition and file, format it with mkswap and activate it with swapon. Swap over the network is considered as an option in diskless systems. The two likely scenarios are when blade servers are used as part of a cluster where the form factor or maintenance costs do not allow the use of disks and thin clients. The Linux Terminal Server Project recommends the use of the Network Block Device (NBD) for swap according to the manual at https://sourceforge.net/projects/ltsp/files/Docs-Admin-Guide/LTSPManual.pdf/download There is also documentation and tutorials on how to setup swap over NBD at places like https://help.ubuntu.com/community/UbuntuLTSP/EnableNBDSWAP The nbd-client also documents the use of NBD as swap. Despite this, the fact is that a machine using NBD for swap can deadlock within minutes if swap is used intensively. This patch series addresses the problem. The core issue is that network block devices do not use mempools like normal block devices do. As the host cannot control where they receive packets from, they cannot reliably work out in advance how much memory they might need. Some years ago, Peter Zijlstra developed a series of patches that supported swap over an NFS that at least one distribution is carrying within their kernels. This patch series borrows very heavily from Peter's work to support swapping over NBD as a pre-requisite to supporting swap-over-NFS. The bulk of the complexity is concerned with preserving memory that is allocated from the PFMEMALLOC reserves for use by the network layer which is needed for both NBD and NFS. Patch 1 adds knowledge of the PFMEMALLOC reserves to SLAB and SLUB to preserve access to pages allocated under low memory situations to callers that are freeing memory. Patch 2 optimises the SLUB fast path to avoid pfmemalloc checks Patch 3 introduces __GFP_MEMALLOC to allow access to the PFMEMALLOC reserves without setting PFMEMALLOC. Patch 4 opens the possibility for softirqs to use PFMEMALLOC reserves for later use by network packet processing. Patch 5 only sets page->pfmemalloc when ALLOC_NO_WATERMARKS was required Patch 6 ignores memory policies when ALLOC_NO_WATERMARKS is set. Patches 7-12 allows network processing to use PFMEMALLOC reserves when the socket has been marked as being used by the VM to clean pages. If packets are received and stored in pages that were allocated under low-memory situations and are unrelated to the VM, the packets are dropped. Patch 11 reintroduces __skb_alloc_page which the networking folk may object to but is needed in some cases to propogate pfmemalloc from a newly allocated page to an skb. If there is a strong objection, this patch can be dropped with the impact being that swap-over-network will be slower in some cases but it should not fail. Patch 13 is a micro-optimisation to avoid a function call in the common case. Patch 14 tags NBD sockets as being SOCK_MEMALLOC so they can use PFMEMALLOC if necessary. Patch 15 notes that it is still possible for the PFMEMALLOC reserve to be depleted. To prevent this, direct reclaimers get throttled on a waitqueue if 50% of the PFMEMALLOC reserves are depleted. It is expected that kswapd and the direct reclaimers already running will clean enough pages for the low watermark to be reached and the throttled processes are woken up. Patch 16 adds a statistic to track how often processes get throttled Some basic performance testing was run using kernel builds, netperf on loopback for UDP and TCP, hackbench (pipes and sockets), iozone and sysbench. Each of them were expected to use the sl*b allocators reasonably heavily but there did not appear to be significant performance variances. For testing swap-over-NBD, a machine was booted with 2G of RAM with a swapfile backed by NBD. 8*NUM_CPU processes were started that create anonymous memory mappings and read them linearly in a loop. The total size of the mappings were 4*PHYSICAL_MEMORY to use swap heavily under memory pressure. Without the patches and using SLUB, the machine locks up within minutes and runs to completion with them applied. With SLAB, the story is different as an unpatched kernel run to completion. However, the patched kernel completed the test 45% faster. MICRO 3.5.0-rc2 3.5.0-rc2 vanilla swapnbd Unrecognised test vmscan-anon-mmap-write MMTests Statistics: duration Sys Time Running Test (seconds) 197.80 173.07 User+Sys Time Running Test (seconds) 206.96 182.03 Total Elapsed Time (seconds) 3240.70 1762.09 This patch: mm: sl[au]b: add knowledge of PFMEMALLOC reserve pages Allocations of pages below the min watermark run a risk of the machine hanging due to a lack of memory. To prevent this, only callers who have PF_MEMALLOC or TIF_MEMDIE set and are not processing an interrupt are allowed to allocate with ALLOC_NO_WATERMARKS. Once they are allocated to a slab though, nothing prevents other callers consuming free objects within those slabs. This patch limits access to slab pages that were alloced from the PFMEMALLOC reserves. When this patch is applied, pages allocated from below the low watermark are returned with page->pfmemalloc set and it is up to the caller to determine how the page should be protected. SLAB restricts access to any page with page->pfmemalloc set to callers which are known to able to access the PFMEMALLOC reserve. If one is not available, an attempt is made to allocate a new page rather than use a reserve. SLUB is a bit more relaxed in that it only records if the current per-CPU page was allocated from PFMEMALLOC reserve and uses another partial slab if the caller does not have the necessary GFP or process flags. This was found to be sufficient in tests to avoid hangs due to SLUB generally maintaining smaller lists than SLAB. In low-memory conditions it does mean that !PFMEMALLOC allocators can fail a slab allocation even though free objects are available because they are being preserved for callers that are freeing pages. [a.p.zijlstra@chello.nl: Original implementation] [sebastian@breakpoint.cc: Correct order of page flag clearing] Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-08-01 07:43:58 +08:00
objp = objpp[i];
page = virt_to_head_page(objp);
list_del(&page->slab_list);
check_spinlock_acquired_node(cachep, node);
slab_put_obj(cachep, page, objp);
STATS_DEC_ACTIVE(cachep);
/* fixup slab chains */
if (page->active == 0) {
list_add(&page->slab_list, &n->slabs_free);
n->free_slabs++;
} else {
/* Unconditionally move a slab to the end of the
* partial list on free - maximum time for the
* other objects to be freed, too.
*/
list_add_tail(&page->slab_list, &n->slabs_partial);
}
}
while (n->free_objects > n->free_limit && !list_empty(&n->slabs_free)) {
n->free_objects -= cachep->num;
page = list_last_entry(&n->slabs_free, struct page, slab_list);
list_move(&page->slab_list, list);
n->free_slabs--;
n->total_slabs--;
}
}
static void cache_flusharray(struct kmem_cache *cachep, struct array_cache *ac)
{
int batchcount;
struct kmem_cache_node *n;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
int node = numa_mem_id();
LIST_HEAD(list);
batchcount = ac->batchcount;
check_irq_off();
n = get_node(cachep, node);
spin_lock(&n->list_lock);
if (n->shared) {
struct array_cache *shared_array = n->shared;
int max = shared_array->limit - shared_array->avail;
if (max) {
if (batchcount > max)
batchcount = max;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
memcpy(&(shared_array->entry[shared_array->avail]),
ac->entry, sizeof(void *) * batchcount);
shared_array->avail += batchcount;
goto free_done;
}
}
free_block(cachep, ac->entry, batchcount, node, &list);
free_done:
#if STATS
{
int i = 0;
struct page *page;
list_for_each_entry(page, &n->slabs_free, slab_list) {
BUG_ON(page->active);
i++;
}
STATS_SET_FREEABLE(cachep, i);
}
#endif
spin_unlock(&n->list_lock);
slabs_destroy(cachep, &list);
ac->avail -= batchcount;
memmove(ac->entry, &(ac->entry[batchcount]), sizeof(void *)*ac->avail);
}
/*
* Release an obj back to its cache. If the obj has a constructed state, it must
* be in this state _before_ it is released. Called with disabled ints.
*/
static __always_inline void __cache_free(struct kmem_cache *cachep, void *objp,
unsigned long caller)
{
mm: kasan: initial memory quarantine implementation Quarantine isolates freed objects in a separate queue. The objects are returned to the allocator later, which helps to detect use-after-free errors. When the object is freed, its state changes from KASAN_STATE_ALLOC to KASAN_STATE_QUARANTINE. The object is poisoned and put into quarantine instead of being returned to the allocator, therefore every subsequent access to that object triggers a KASAN error, and the error handler is able to say where the object has been allocated and deallocated. When it's time for the object to leave quarantine, its state becomes KASAN_STATE_FREE and it's returned to the allocator. From now on the allocator may reuse it for another allocation. Before that happens, it's still possible to detect a use-after free on that object (it retains the allocation/deallocation stacks). When the allocator reuses this object, the shadow is unpoisoned and old allocation/deallocation stacks are wiped. Therefore a use of this object, even an incorrect one, won't trigger ASan warning. Without the quarantine, it's not guaranteed that the objects aren't reused immediately, that's why the probability of catching a use-after-free is lower than with quarantine in place. Quarantine isolates freed objects in a separate queue. The objects are returned to the allocator later, which helps to detect use-after-free errors. Freed objects are first added to per-cpu quarantine queues. When a cache is destroyed or memory shrinking is requested, the objects are moved into the global quarantine queue. Whenever a kmalloc call allows memory reclaiming, the oldest objects are popped out of the global queue until the total size of objects in quarantine is less than 3/4 of the maximum quarantine size (which is a fraction of installed physical memory). As long as an object remains in the quarantine, KASAN is able to report accesses to it, so the chance of reporting a use-after-free is increased. Once the object leaves quarantine, the allocator may reuse it, in which case the object is unpoisoned and KASAN can't detect incorrect accesses to it. Right now quarantine support is only enabled in SLAB allocator. Unification of KASAN features in SLAB and SLUB will be done later. This patch is based on the "mm: kasan: quarantine" patch originally prepared by Dmitry Chernenkov. A number of improvements have been suggested by Andrey Ryabinin. [glider@google.com: v9] Link: http://lkml.kernel.org/r/1462987130-144092-1-git-send-email-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Andrey Konovalov <adech.fo@gmail.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Konstantin Serebryany <kcc@google.com> Cc: Dmitry Chernenkov <dmitryc@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-21 07:59:11 +08:00
/* Put the object into the quarantine, don't touch it for now. */
if (kasan_slab_free(cachep, objp, _RET_IP_))
mm: kasan: initial memory quarantine implementation Quarantine isolates freed objects in a separate queue. The objects are returned to the allocator later, which helps to detect use-after-free errors. When the object is freed, its state changes from KASAN_STATE_ALLOC to KASAN_STATE_QUARANTINE. The object is poisoned and put into quarantine instead of being returned to the allocator, therefore every subsequent access to that object triggers a KASAN error, and the error handler is able to say where the object has been allocated and deallocated. When it's time for the object to leave quarantine, its state becomes KASAN_STATE_FREE and it's returned to the allocator. From now on the allocator may reuse it for another allocation. Before that happens, it's still possible to detect a use-after free on that object (it retains the allocation/deallocation stacks). When the allocator reuses this object, the shadow is unpoisoned and old allocation/deallocation stacks are wiped. Therefore a use of this object, even an incorrect one, won't trigger ASan warning. Without the quarantine, it's not guaranteed that the objects aren't reused immediately, that's why the probability of catching a use-after-free is lower than with quarantine in place. Quarantine isolates freed objects in a separate queue. The objects are returned to the allocator later, which helps to detect use-after-free errors. Freed objects are first added to per-cpu quarantine queues. When a cache is destroyed or memory shrinking is requested, the objects are moved into the global quarantine queue. Whenever a kmalloc call allows memory reclaiming, the oldest objects are popped out of the global queue until the total size of objects in quarantine is less than 3/4 of the maximum quarantine size (which is a fraction of installed physical memory). As long as an object remains in the quarantine, KASAN is able to report accesses to it, so the chance of reporting a use-after-free is increased. Once the object leaves quarantine, the allocator may reuse it, in which case the object is unpoisoned and KASAN can't detect incorrect accesses to it. Right now quarantine support is only enabled in SLAB allocator. Unification of KASAN features in SLAB and SLUB will be done later. This patch is based on the "mm: kasan: quarantine" patch originally prepared by Dmitry Chernenkov. A number of improvements have been suggested by Andrey Ryabinin. [glider@google.com: v9] Link: http://lkml.kernel.org/r/1462987130-144092-1-git-send-email-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Andrey Konovalov <adech.fo@gmail.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Konstantin Serebryany <kcc@google.com> Cc: Dmitry Chernenkov <dmitryc@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-21 07:59:11 +08:00
return;
___cache_free(cachep, objp, caller);
}
mm: kasan: initial memory quarantine implementation Quarantine isolates freed objects in a separate queue. The objects are returned to the allocator later, which helps to detect use-after-free errors. When the object is freed, its state changes from KASAN_STATE_ALLOC to KASAN_STATE_QUARANTINE. The object is poisoned and put into quarantine instead of being returned to the allocator, therefore every subsequent access to that object triggers a KASAN error, and the error handler is able to say where the object has been allocated and deallocated. When it's time for the object to leave quarantine, its state becomes KASAN_STATE_FREE and it's returned to the allocator. From now on the allocator may reuse it for another allocation. Before that happens, it's still possible to detect a use-after free on that object (it retains the allocation/deallocation stacks). When the allocator reuses this object, the shadow is unpoisoned and old allocation/deallocation stacks are wiped. Therefore a use of this object, even an incorrect one, won't trigger ASan warning. Without the quarantine, it's not guaranteed that the objects aren't reused immediately, that's why the probability of catching a use-after-free is lower than with quarantine in place. Quarantine isolates freed objects in a separate queue. The objects are returned to the allocator later, which helps to detect use-after-free errors. Freed objects are first added to per-cpu quarantine queues. When a cache is destroyed or memory shrinking is requested, the objects are moved into the global quarantine queue. Whenever a kmalloc call allows memory reclaiming, the oldest objects are popped out of the global queue until the total size of objects in quarantine is less than 3/4 of the maximum quarantine size (which is a fraction of installed physical memory). As long as an object remains in the quarantine, KASAN is able to report accesses to it, so the chance of reporting a use-after-free is increased. Once the object leaves quarantine, the allocator may reuse it, in which case the object is unpoisoned and KASAN can't detect incorrect accesses to it. Right now quarantine support is only enabled in SLAB allocator. Unification of KASAN features in SLAB and SLUB will be done later. This patch is based on the "mm: kasan: quarantine" patch originally prepared by Dmitry Chernenkov. A number of improvements have been suggested by Andrey Ryabinin. [glider@google.com: v9] Link: http://lkml.kernel.org/r/1462987130-144092-1-git-send-email-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Andrey Konovalov <adech.fo@gmail.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Konstantin Serebryany <kcc@google.com> Cc: Dmitry Chernenkov <dmitryc@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-21 07:59:11 +08:00
void ___cache_free(struct kmem_cache *cachep, void *objp,
unsigned long caller)
{
struct array_cache *ac = cpu_cache_get(cachep);
check_irq_off();
mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options Patch series "add init_on_alloc/init_on_free boot options", v10. Provide init_on_alloc and init_on_free boot options. These are aimed at preventing possible information leaks and making the control-flow bugs that depend on uninitialized values more deterministic. Enabling either of the options guarantees that the memory returned by the page allocator and SL[AU]B is initialized with zeroes. SLOB allocator isn't supported at the moment, as its emulation of kmem caches complicates handling of SLAB_TYPESAFE_BY_RCU caches correctly. Enabling init_on_free also guarantees that pages and heap objects are initialized right after they're freed, so it won't be possible to access stale data by using a dangling pointer. As suggested by Michal Hocko, right now we don't let the heap users to disable initialization for certain allocations. There's not enough evidence that doing so can speed up real-life cases, and introducing ways to opt-out may result in things going out of control. This patch (of 2): The new options are needed to prevent possible information leaks and make control-flow bugs that depend on uninitialized values more deterministic. This is expected to be on-by-default on Android and Chrome OS. And it gives the opportunity for anyone else to use it under distros too via the boot args. (The init_on_free feature is regularly requested by folks where memory forensics is included in their threat models.) init_on_alloc=1 makes the kernel initialize newly allocated pages and heap objects with zeroes. Initialization is done at allocation time at the places where checks for __GFP_ZERO are performed. init_on_free=1 makes the kernel initialize freed pages and heap objects with zeroes upon their deletion. This helps to ensure sensitive data doesn't leak via use-after-free accesses. Both init_on_alloc=1 and init_on_free=1 guarantee that the allocator returns zeroed memory. The two exceptions are slab caches with constructors and SLAB_TYPESAFE_BY_RCU flag. Those are never zero-initialized to preserve their semantics. Both init_on_alloc and init_on_free default to zero, but those defaults can be overridden with CONFIG_INIT_ON_ALLOC_DEFAULT_ON and CONFIG_INIT_ON_FREE_DEFAULT_ON. If either SLUB poisoning or page poisoning is enabled, those options take precedence over init_on_alloc and init_on_free: initialization is only applied to unpoisoned allocations. Slowdown for the new features compared to init_on_free=0, init_on_alloc=0: hackbench, init_on_free=1: +7.62% sys time (st.err 0.74%) hackbench, init_on_alloc=1: +7.75% sys time (st.err 2.14%) Linux build with -j12, init_on_free=1: +8.38% wall time (st.err 0.39%) Linux build with -j12, init_on_free=1: +24.42% sys time (st.err 0.52%) Linux build with -j12, init_on_alloc=1: -0.13% wall time (st.err 0.42%) Linux build with -j12, init_on_alloc=1: +0.57% sys time (st.err 0.40%) The slowdown for init_on_free=0, init_on_alloc=0 compared to the baseline is within the standard error. The new features are also going to pave the way for hardware memory tagging (e.g. arm64's MTE), which will require both on_alloc and on_free hooks to set the tags for heap objects. With MTE, tagging will have the same cost as memory initialization. Although init_on_free is rather costly, there are paranoid use-cases where in-memory data lifetime is desired to be minimized. There are various arguments for/against the realism of the associated threat models, but given that we'll need the infrastructure for MTE anyway, and there are people who want wipe-on-free behavior no matter what the performance cost, it seems reasonable to include it in this series. [glider@google.com: v8] Link: http://lkml.kernel.org/r/20190626121943.131390-2-glider@google.com [glider@google.com: v9] Link: http://lkml.kernel.org/r/20190627130316.254309-2-glider@google.com [glider@google.com: v10] Link: http://lkml.kernel.org/r/20190628093131.199499-2-glider@google.com Link: http://lkml.kernel.org/r/20190617151050.92663-2-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Acked-by: Kees Cook <keescook@chromium.org> Acked-by: Michal Hocko <mhocko@suse.cz> [page and dmapool parts Acked-by: James Morris <jamorris@linux.microsoft.com>] Cc: Christoph Lameter <cl@linux.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: "Serge E. Hallyn" <serge@hallyn.com> Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Kostya Serebryany <kcc@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Sandeep Patil <sspatil@android.com> Cc: Laura Abbott <labbott@redhat.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Jann Horn <jannh@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marco Elver <elver@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:59:19 +08:00
if (unlikely(slab_want_init_on_free(cachep)))
memset(objp, 0, cachep->object_size);
kmemleak_free_recursive(objp, cachep->flags);
objp = cache_free_debugcheck(cachep, objp, caller);
/*
* Skip calling cache_free_alien() when the platform is not numa.
* This will avoid cache misses that happen while accessing slabp (which
* is per page memory reference) to get nodeid. Instead use a global
* variable to skip the call, which is mostly likely to be present in
* the cache.
*/
if (nr_online_nodes > 1 && cache_free_alien(cachep, objp))
return;
if (ac->avail < ac->limit) {
STATS_INC_FREEHIT(cachep);
} else {
STATS_INC_FREEMISS(cachep);
cache_flusharray(cachep, ac);
}
mm/slab: re-implement pfmemalloc support Current implementation of pfmemalloc handling in SLAB has some problems. 1) pfmemalloc_active is set to true when there is just one or more pfmemalloc slabs in the system, but it is cleared when there is no pfmemalloc slab in one arbitrary kmem_cache. So, pfmemalloc_active could be wrongly cleared. 2) Search to partial and free list doesn't happen when non-pfmemalloc object are not found in cpu cache. Instead, allocating new slab happens and it is not optimal. 3) Even after sk_memalloc_socks() is disabled, cpu cache would keep pfmemalloc objects tagged with SLAB_OBJ_PFMEMALLOC. It isn't cleared if sk_memalloc_socks() is disabled so it could cause problem. 4) If cpu cache is filled with pfmemalloc objects, it would cause slow down non-pfmemalloc allocation. To me, current pointer tagging approach looks complex and fragile so this patch re-implement whole thing instead of fixing problems one by one. Design principle for new implementation is that 1) Don't disrupt non-pfmemalloc allocation in fast path even if sk_memalloc_socks() is enabled. It's more likely case than pfmemalloc allocation. 2) Ensure that pfmemalloc slab is used only for pfmemalloc allocation. 3) Don't consider performance of pfmemalloc allocation in memory deficiency state. As a result, all pfmemalloc alloc/free in memory tight state will be handled in slow-path. If there is non-pfmemalloc free object, it will be returned first even for pfmemalloc user in fast-path so that performance of pfmemalloc user isn't affected in normal case and pfmemalloc objects will be kept as long as possible. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Tested-by: Mel Gorman <mgorman@techsingularity.net> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:56 +08:00
if (sk_memalloc_socks()) {
struct page *page = virt_to_head_page(objp);
if (unlikely(PageSlabPfmemalloc(page))) {
cache_free_pfmemalloc(cachep, page, objp);
return;
}
}
ac->entry[ac->avail++] = objp;
}
/**
* kmem_cache_alloc - Allocate an object
* @cachep: The cache to allocate from.
* @flags: See kmalloc().
*
* Allocate an object from this cache. The flags are only relevant
* if the cache has no available objects.
*
* Return: pointer to the new object or %NULL in case of error
*/
void *kmem_cache_alloc(struct kmem_cache *cachep, gfp_t flags)
{
void *ret = slab_alloc(cachep, flags, _RET_IP_);
trace_kmem_cache_alloc(_RET_IP_, ret,
cachep->object_size, cachep->size, flags);
return ret;
}
EXPORT_SYMBOL(kmem_cache_alloc);
static __always_inline void
cache_alloc_debugcheck_after_bulk(struct kmem_cache *s, gfp_t flags,
size_t size, void **p, unsigned long caller)
{
size_t i;
for (i = 0; i < size; i++)
p[i] = cache_alloc_debugcheck_after(s, flags, p[i], caller);
}
int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
void **p)
{
size_t i;
s = slab_pre_alloc_hook(s, flags);
if (!s)
return 0;
cache_alloc_debugcheck_before(s, flags);
local_irq_disable();
for (i = 0; i < size; i++) {
void *objp = __do_cache_alloc(s, flags);
if (unlikely(!objp))
goto error;
p[i] = objp;
}
local_irq_enable();
cache_alloc_debugcheck_after_bulk(s, flags, size, p, _RET_IP_);
/* Clear memory outside IRQ disabled section */
mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options Patch series "add init_on_alloc/init_on_free boot options", v10. Provide init_on_alloc and init_on_free boot options. These are aimed at preventing possible information leaks and making the control-flow bugs that depend on uninitialized values more deterministic. Enabling either of the options guarantees that the memory returned by the page allocator and SL[AU]B is initialized with zeroes. SLOB allocator isn't supported at the moment, as its emulation of kmem caches complicates handling of SLAB_TYPESAFE_BY_RCU caches correctly. Enabling init_on_free also guarantees that pages and heap objects are initialized right after they're freed, so it won't be possible to access stale data by using a dangling pointer. As suggested by Michal Hocko, right now we don't let the heap users to disable initialization for certain allocations. There's not enough evidence that doing so can speed up real-life cases, and introducing ways to opt-out may result in things going out of control. This patch (of 2): The new options are needed to prevent possible information leaks and make control-flow bugs that depend on uninitialized values more deterministic. This is expected to be on-by-default on Android and Chrome OS. And it gives the opportunity for anyone else to use it under distros too via the boot args. (The init_on_free feature is regularly requested by folks where memory forensics is included in their threat models.) init_on_alloc=1 makes the kernel initialize newly allocated pages and heap objects with zeroes. Initialization is done at allocation time at the places where checks for __GFP_ZERO are performed. init_on_free=1 makes the kernel initialize freed pages and heap objects with zeroes upon their deletion. This helps to ensure sensitive data doesn't leak via use-after-free accesses. Both init_on_alloc=1 and init_on_free=1 guarantee that the allocator returns zeroed memory. The two exceptions are slab caches with constructors and SLAB_TYPESAFE_BY_RCU flag. Those are never zero-initialized to preserve their semantics. Both init_on_alloc and init_on_free default to zero, but those defaults can be overridden with CONFIG_INIT_ON_ALLOC_DEFAULT_ON and CONFIG_INIT_ON_FREE_DEFAULT_ON. If either SLUB poisoning or page poisoning is enabled, those options take precedence over init_on_alloc and init_on_free: initialization is only applied to unpoisoned allocations. Slowdown for the new features compared to init_on_free=0, init_on_alloc=0: hackbench, init_on_free=1: +7.62% sys time (st.err 0.74%) hackbench, init_on_alloc=1: +7.75% sys time (st.err 2.14%) Linux build with -j12, init_on_free=1: +8.38% wall time (st.err 0.39%) Linux build with -j12, init_on_free=1: +24.42% sys time (st.err 0.52%) Linux build with -j12, init_on_alloc=1: -0.13% wall time (st.err 0.42%) Linux build with -j12, init_on_alloc=1: +0.57% sys time (st.err 0.40%) The slowdown for init_on_free=0, init_on_alloc=0 compared to the baseline is within the standard error. The new features are also going to pave the way for hardware memory tagging (e.g. arm64's MTE), which will require both on_alloc and on_free hooks to set the tags for heap objects. With MTE, tagging will have the same cost as memory initialization. Although init_on_free is rather costly, there are paranoid use-cases where in-memory data lifetime is desired to be minimized. There are various arguments for/against the realism of the associated threat models, but given that we'll need the infrastructure for MTE anyway, and there are people who want wipe-on-free behavior no matter what the performance cost, it seems reasonable to include it in this series. [glider@google.com: v8] Link: http://lkml.kernel.org/r/20190626121943.131390-2-glider@google.com [glider@google.com: v9] Link: http://lkml.kernel.org/r/20190627130316.254309-2-glider@google.com [glider@google.com: v10] Link: http://lkml.kernel.org/r/20190628093131.199499-2-glider@google.com Link: http://lkml.kernel.org/r/20190617151050.92663-2-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Acked-by: Kees Cook <keescook@chromium.org> Acked-by: Michal Hocko <mhocko@suse.cz> [page and dmapool parts Acked-by: James Morris <jamorris@linux.microsoft.com>] Cc: Christoph Lameter <cl@linux.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: "Serge E. Hallyn" <serge@hallyn.com> Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Kostya Serebryany <kcc@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Sandeep Patil <sspatil@android.com> Cc: Laura Abbott <labbott@redhat.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Jann Horn <jannh@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marco Elver <elver@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:59:19 +08:00
if (unlikely(slab_want_init_on_alloc(flags, s)))
for (i = 0; i < size; i++)
memset(p[i], 0, s->object_size);
slab_post_alloc_hook(s, flags, size, p);
/* FIXME: Trace call missing. Christoph would like a bulk variant */
return size;
error:
local_irq_enable();
cache_alloc_debugcheck_after_bulk(s, flags, i, p, _RET_IP_);
slab_post_alloc_hook(s, flags, i, p);
__kmem_cache_free_bulk(s, i, p);
return 0;
}
EXPORT_SYMBOL(kmem_cache_alloc_bulk);
#ifdef CONFIG_TRACING
void *
kmem_cache_alloc_trace(struct kmem_cache *cachep, gfp_t flags, size_t size)
{
void *ret;
ret = slab_alloc(cachep, flags, _RET_IP_);
kasan, mm: change hooks signatures Patch series "kasan: add software tag-based mode for arm64", v13. This patchset adds a new software tag-based mode to KASAN [1]. (Initially this mode was called KHWASAN, but it got renamed, see the naming rationale at the end of this section). The plan is to implement HWASan [2] for the kernel with the incentive, that it's going to have comparable to KASAN performance, but in the same time consume much less memory, trading that off for somewhat imprecise bug detection and being supported only for arm64. The underlying ideas of the approach used by software tag-based KASAN are: 1. By using the Top Byte Ignore (TBI) arm64 CPU feature, we can store pointer tags in the top byte of each kernel pointer. 2. Using shadow memory, we can store memory tags for each chunk of kernel memory. 3. On each memory allocation, we can generate a random tag, embed it into the returned pointer and set the memory tags that correspond to this chunk of memory to the same value. 4. By using compiler instrumentation, before each memory access we can add a check that the pointer tag matches the tag of the memory that is being accessed. 5. On a tag mismatch we report an error. With this patchset the existing KASAN mode gets renamed to generic KASAN, with the word "generic" meaning that the implementation can be supported by any architecture as it is purely software. The new mode this patchset adds is called software tag-based KASAN. The word "tag-based" refers to the fact that this mode uses tags embedded into the top byte of kernel pointers and the TBI arm64 CPU feature that allows to dereference such pointers. The word "software" here means that shadow memory manipulation and tag checking on pointer dereference is done in software. As it is the only tag-based implementation right now, "software tag-based" KASAN is sometimes referred to as simply "tag-based" in this patchset. A potential expansion of this mode is a hardware tag-based mode, which would use hardware memory tagging support (announced by Arm [3]) instead of compiler instrumentation and manual shadow memory manipulation. Same as generic KASAN, software tag-based KASAN is strictly a debugging feature. [1] https://www.kernel.org/doc/html/latest/dev-tools/kasan.html [2] http://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html [3] https://community.arm.com/processors/b/blog/posts/arm-a-profile-architecture-2018-developments-armv85a ====== Rationale On mobile devices generic KASAN's memory usage is significant problem. One of the main reasons to have tag-based KASAN is to be able to perform a similar set of checks as the generic one does, but with lower memory requirements. Comment from Vishwath Mohan <vishwath@google.com>: I don't have data on-hand, but anecdotally both ASAN and KASAN have proven problematic to enable for environments that don't tolerate the increased memory pressure well. This includes (a) Low-memory form factors - Wear, TV, Things, lower-tier phones like Go, (c) Connected components like Pixel's visual core [1]. These are both places I'd love to have a low(er) memory footprint option at my disposal. Comment from Evgenii Stepanov <eugenis@google.com>: Looking at a live Android device under load, slab (according to /proc/meminfo) + kernel stack take 8-10% available RAM (~350MB). KASAN's overhead of 2x - 3x on top of it is not insignificant. Not having this overhead enables near-production use - ex. running KASAN/KHWASAN kernel on a personal, daily-use device to catch bugs that do not reproduce in test configuration. These are the ones that often cost the most engineering time to track down. CPU overhead is bad, but generally tolerable. RAM is critical, in our experience. Once it gets low enough, OOM-killer makes your life miserable. [1] https://www.blog.google/products/pixel/pixel-visual-core-image-processing-and-machine-learning-pixel-2/ ====== Technical details Software tag-based KASAN mode is implemented in a very similar way to the generic one. This patchset essentially does the following: 1. TCR_TBI1 is set to enable Top Byte Ignore. 2. Shadow memory is used (with a different scale, 1:16, so each shadow byte corresponds to 16 bytes of kernel memory) to store memory tags. 3. All slab objects are aligned to shadow scale, which is 16 bytes. 4. All pointers returned from the slab allocator are tagged with a random tag and the corresponding shadow memory is poisoned with the same value. 5. Compiler instrumentation is used to insert tag checks. Either by calling callbacks or by inlining them (CONFIG_KASAN_OUTLINE and CONFIG_KASAN_INLINE flags are reused). 6. When a tag mismatch is detected in callback instrumentation mode KASAN simply prints a bug report. In case of inline instrumentation, clang inserts a brk instruction, and KASAN has it's own brk handler, which reports the bug. 7. The memory in between slab objects is marked with a reserved tag, and acts as a redzone. 8. When a slab object is freed it's marked with a reserved tag. Bug detection is imprecise for two reasons: 1. We won't catch some small out-of-bounds accesses, that fall into the same shadow cell, as the last byte of a slab object. 2. We only have 1 byte to store tags, which means we have a 1/256 probability of a tag match for an incorrect access (actually even slightly less due to reserved tag values). Despite that there's a particular type of bugs that tag-based KASAN can detect compared to generic KASAN: use-after-free after the object has been allocated by someone else. ====== Testing Some kernel developers voiced a concern that changing the top byte of kernel pointers may lead to subtle bugs that are difficult to discover. To address this concern deliberate testing has been performed. It doesn't seem feasible to do some kind of static checking to find potential issues with pointer tagging, so a dynamic approach was taken. All pointer comparisons/subtractions have been instrumented in an LLVM compiler pass and a kernel module that would print a bug report whenever two pointers with different tags are being compared/subtracted (ignoring comparisons with NULL pointers and with pointers obtained by casting an error code to a pointer type) has been used. Then the kernel has been booted in QEMU and on an Odroid C2 board and syzkaller has been run. This yielded the following results. The two places that look interesting are: is_vmalloc_addr in include/linux/mm.h is_kernel_rodata in mm/util.c Here we compare a pointer with some fixed untagged values to make sure that the pointer lies in a particular part of the kernel address space. Since tag-based KASAN doesn't add tags to pointers that belong to rodata or vmalloc regions, this should work as is. To make sure debug checks to those two functions that check that the result doesn't change whether we operate on pointers with or without untagging has been added. A few other cases that don't look that interesting: Comparing pointers to achieve unique sorting order of pointee objects (e.g. sorting locks addresses before performing a double lock): tty_ldisc_lock_pair_timeout in drivers/tty/tty_ldisc.c pipe_double_lock in fs/pipe.c unix_state_double_lock in net/unix/af_unix.c lock_two_nondirectories in fs/inode.c mutex_lock_double in kernel/events/core.c ep_cmp_ffd in fs/eventpoll.c fsnotify_compare_groups fs/notify/mark.c Nothing needs to be done here, since the tags embedded into pointers don't change, so the sorting order would still be unique. Checks that a pointer belongs to some particular allocation: is_sibling_entry in lib/radix-tree.c object_is_on_stack in include/linux/sched/task_stack.h Nothing needs to be done here either, since two pointers can only belong to the same allocation if they have the same tag. Overall, since the kernel boots and works, there are no critical bugs. As for the rest, the traditional kernel testing way (use until fails) is the only one that looks feasible. Another point here is that tag-based KASAN is available under a separate config option that needs to be deliberately enabled. Even though it might be used in a "near-production" environment to find bugs that are not found during fuzzing or running tests, it is still a debug tool. ====== Benchmarks The following numbers were collected on Odroid C2 board. Both generic and tag-based KASAN were used in inline instrumentation mode. Boot time [1]: * ~1.7 sec for clean kernel * ~5.0 sec for generic KASAN * ~5.0 sec for tag-based KASAN Network performance [2]: * 8.33 Gbits/sec for clean kernel * 3.17 Gbits/sec for generic KASAN * 2.85 Gbits/sec for tag-based KASAN Slab memory usage after boot [3]: * ~40 kb for clean kernel * ~105 kb (~260% overhead) for generic KASAN * ~47 kb (~20% overhead) for tag-based KASAN KASAN memory overhead consists of three main parts: 1. Increased slab memory usage due to redzones. 2. Shadow memory (the whole reserved once during boot). 3. Quaratine (grows gradually until some preset limit; the more the limit, the more the chance to detect a use-after-free). Comparing tag-based vs generic KASAN for each of these points: 1. 20% vs 260% overhead. 2. 1/16th vs 1/8th of physical memory. 3. Tag-based KASAN doesn't require quarantine. [1] Time before the ext4 driver is initialized. [2] Measured as `iperf -s & iperf -c 127.0.0.1 -t 30`. [3] Measured as `cat /proc/meminfo | grep Slab`. ====== Some notes A few notes: 1. The patchset can be found here: https://github.com/xairy/kasan-prototype/tree/khwasan 2. Building requires a recent Clang version (7.0.0 or later). 3. Stack instrumentation is not supported yet and will be added later. This patch (of 25): Tag-based KASAN changes the value of the top byte of pointers returned from the kernel allocation functions (such as kmalloc). This patch updates KASAN hooks signatures and their usage in SLAB and SLUB code to reflect that. Link: http://lkml.kernel.org/r/aec2b5e3973781ff8a6bb6760f8543643202c451.1544099024.git.andreyknvl@google.com Signed-off-by: Andrey Konovalov <andreyknvl@google.com> Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Dmitry Vyukov <dvyukov@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Will Deacon <will.deacon@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-12-28 16:29:37 +08:00
ret = kasan_kmalloc(cachep, ret, size, flags);
trace_kmalloc(_RET_IP_, ret,
size, cachep->size, flags);
return ret;
}
EXPORT_SYMBOL(kmem_cache_alloc_trace);
#endif
#ifdef CONFIG_NUMA
/**
* kmem_cache_alloc_node - Allocate an object on the specified node
* @cachep: The cache to allocate from.
* @flags: See kmalloc().
* @nodeid: node number of the target node.
*
* Identical to kmem_cache_alloc but it will allocate memory on the given
* node, which can improve the performance for cpu bound structures.
*
* Fallback to other node is possible if __GFP_THISNODE is not set.
*
* Return: pointer to the new object or %NULL in case of error
*/
void *kmem_cache_alloc_node(struct kmem_cache *cachep, gfp_t flags, int nodeid)
{
void *ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
trace_kmem_cache_alloc_node(_RET_IP_, ret,
cachep->object_size, cachep->size,
flags, nodeid);
return ret;
}
EXPORT_SYMBOL(kmem_cache_alloc_node);
#ifdef CONFIG_TRACING
void *kmem_cache_alloc_node_trace(struct kmem_cache *cachep,
gfp_t flags,
int nodeid,
size_t size)
{
void *ret;
ret = slab_alloc_node(cachep, flags, nodeid, _RET_IP_);
kasan, mm: change hooks signatures Patch series "kasan: add software tag-based mode for arm64", v13. This patchset adds a new software tag-based mode to KASAN [1]. (Initially this mode was called KHWASAN, but it got renamed, see the naming rationale at the end of this section). The plan is to implement HWASan [2] for the kernel with the incentive, that it's going to have comparable to KASAN performance, but in the same time consume much less memory, trading that off for somewhat imprecise bug detection and being supported only for arm64. The underlying ideas of the approach used by software tag-based KASAN are: 1. By using the Top Byte Ignore (TBI) arm64 CPU feature, we can store pointer tags in the top byte of each kernel pointer. 2. Using shadow memory, we can store memory tags for each chunk of kernel memory. 3. On each memory allocation, we can generate a random tag, embed it into the returned pointer and set the memory tags that correspond to this chunk of memory to the same value. 4. By using compiler instrumentation, before each memory access we can add a check that the pointer tag matches the tag of the memory that is being accessed. 5. On a tag mismatch we report an error. With this patchset the existing KASAN mode gets renamed to generic KASAN, with the word "generic" meaning that the implementation can be supported by any architecture as it is purely software. The new mode this patchset adds is called software tag-based KASAN. The word "tag-based" refers to the fact that this mode uses tags embedded into the top byte of kernel pointers and the TBI arm64 CPU feature that allows to dereference such pointers. The word "software" here means that shadow memory manipulation and tag checking on pointer dereference is done in software. As it is the only tag-based implementation right now, "software tag-based" KASAN is sometimes referred to as simply "tag-based" in this patchset. A potential expansion of this mode is a hardware tag-based mode, which would use hardware memory tagging support (announced by Arm [3]) instead of compiler instrumentation and manual shadow memory manipulation. Same as generic KASAN, software tag-based KASAN is strictly a debugging feature. [1] https://www.kernel.org/doc/html/latest/dev-tools/kasan.html [2] http://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html [3] https://community.arm.com/processors/b/blog/posts/arm-a-profile-architecture-2018-developments-armv85a ====== Rationale On mobile devices generic KASAN's memory usage is significant problem. One of the main reasons to have tag-based KASAN is to be able to perform a similar set of checks as the generic one does, but with lower memory requirements. Comment from Vishwath Mohan <vishwath@google.com>: I don't have data on-hand, but anecdotally both ASAN and KASAN have proven problematic to enable for environments that don't tolerate the increased memory pressure well. This includes (a) Low-memory form factors - Wear, TV, Things, lower-tier phones like Go, (c) Connected components like Pixel's visual core [1]. These are both places I'd love to have a low(er) memory footprint option at my disposal. Comment from Evgenii Stepanov <eugenis@google.com>: Looking at a live Android device under load, slab (according to /proc/meminfo) + kernel stack take 8-10% available RAM (~350MB). KASAN's overhead of 2x - 3x on top of it is not insignificant. Not having this overhead enables near-production use - ex. running KASAN/KHWASAN kernel on a personal, daily-use device to catch bugs that do not reproduce in test configuration. These are the ones that often cost the most engineering time to track down. CPU overhead is bad, but generally tolerable. RAM is critical, in our experience. Once it gets low enough, OOM-killer makes your life miserable. [1] https://www.blog.google/products/pixel/pixel-visual-core-image-processing-and-machine-learning-pixel-2/ ====== Technical details Software tag-based KASAN mode is implemented in a very similar way to the generic one. This patchset essentially does the following: 1. TCR_TBI1 is set to enable Top Byte Ignore. 2. Shadow memory is used (with a different scale, 1:16, so each shadow byte corresponds to 16 bytes of kernel memory) to store memory tags. 3. All slab objects are aligned to shadow scale, which is 16 bytes. 4. All pointers returned from the slab allocator are tagged with a random tag and the corresponding shadow memory is poisoned with the same value. 5. Compiler instrumentation is used to insert tag checks. Either by calling callbacks or by inlining them (CONFIG_KASAN_OUTLINE and CONFIG_KASAN_INLINE flags are reused). 6. When a tag mismatch is detected in callback instrumentation mode KASAN simply prints a bug report. In case of inline instrumentation, clang inserts a brk instruction, and KASAN has it's own brk handler, which reports the bug. 7. The memory in between slab objects is marked with a reserved tag, and acts as a redzone. 8. When a slab object is freed it's marked with a reserved tag. Bug detection is imprecise for two reasons: 1. We won't catch some small out-of-bounds accesses, that fall into the same shadow cell, as the last byte of a slab object. 2. We only have 1 byte to store tags, which means we have a 1/256 probability of a tag match for an incorrect access (actually even slightly less due to reserved tag values). Despite that there's a particular type of bugs that tag-based KASAN can detect compared to generic KASAN: use-after-free after the object has been allocated by someone else. ====== Testing Some kernel developers voiced a concern that changing the top byte of kernel pointers may lead to subtle bugs that are difficult to discover. To address this concern deliberate testing has been performed. It doesn't seem feasible to do some kind of static checking to find potential issues with pointer tagging, so a dynamic approach was taken. All pointer comparisons/subtractions have been instrumented in an LLVM compiler pass and a kernel module that would print a bug report whenever two pointers with different tags are being compared/subtracted (ignoring comparisons with NULL pointers and with pointers obtained by casting an error code to a pointer type) has been used. Then the kernel has been booted in QEMU and on an Odroid C2 board and syzkaller has been run. This yielded the following results. The two places that look interesting are: is_vmalloc_addr in include/linux/mm.h is_kernel_rodata in mm/util.c Here we compare a pointer with some fixed untagged values to make sure that the pointer lies in a particular part of the kernel address space. Since tag-based KASAN doesn't add tags to pointers that belong to rodata or vmalloc regions, this should work as is. To make sure debug checks to those two functions that check that the result doesn't change whether we operate on pointers with or without untagging has been added. A few other cases that don't look that interesting: Comparing pointers to achieve unique sorting order of pointee objects (e.g. sorting locks addresses before performing a double lock): tty_ldisc_lock_pair_timeout in drivers/tty/tty_ldisc.c pipe_double_lock in fs/pipe.c unix_state_double_lock in net/unix/af_unix.c lock_two_nondirectories in fs/inode.c mutex_lock_double in kernel/events/core.c ep_cmp_ffd in fs/eventpoll.c fsnotify_compare_groups fs/notify/mark.c Nothing needs to be done here, since the tags embedded into pointers don't change, so the sorting order would still be unique. Checks that a pointer belongs to some particular allocation: is_sibling_entry in lib/radix-tree.c object_is_on_stack in include/linux/sched/task_stack.h Nothing needs to be done here either, since two pointers can only belong to the same allocation if they have the same tag. Overall, since the kernel boots and works, there are no critical bugs. As for the rest, the traditional kernel testing way (use until fails) is the only one that looks feasible. Another point here is that tag-based KASAN is available under a separate config option that needs to be deliberately enabled. Even though it might be used in a "near-production" environment to find bugs that are not found during fuzzing or running tests, it is still a debug tool. ====== Benchmarks The following numbers were collected on Odroid C2 board. Both generic and tag-based KASAN were used in inline instrumentation mode. Boot time [1]: * ~1.7 sec for clean kernel * ~5.0 sec for generic KASAN * ~5.0 sec for tag-based KASAN Network performance [2]: * 8.33 Gbits/sec for clean kernel * 3.17 Gbits/sec for generic KASAN * 2.85 Gbits/sec for tag-based KASAN Slab memory usage after boot [3]: * ~40 kb for clean kernel * ~105 kb (~260% overhead) for generic KASAN * ~47 kb (~20% overhead) for tag-based KASAN KASAN memory overhead consists of three main parts: 1. Increased slab memory usage due to redzones. 2. Shadow memory (the whole reserved once during boot). 3. Quaratine (grows gradually until some preset limit; the more the limit, the more the chance to detect a use-after-free). Comparing tag-based vs generic KASAN for each of these points: 1. 20% vs 260% overhead. 2. 1/16th vs 1/8th of physical memory. 3. Tag-based KASAN doesn't require quarantine. [1] Time before the ext4 driver is initialized. [2] Measured as `iperf -s & iperf -c 127.0.0.1 -t 30`. [3] Measured as `cat /proc/meminfo | grep Slab`. ====== Some notes A few notes: 1. The patchset can be found here: https://github.com/xairy/kasan-prototype/tree/khwasan 2. Building requires a recent Clang version (7.0.0 or later). 3. Stack instrumentation is not supported yet and will be added later. This patch (of 25): Tag-based KASAN changes the value of the top byte of pointers returned from the kernel allocation functions (such as kmalloc). This patch updates KASAN hooks signatures and their usage in SLAB and SLUB code to reflect that. Link: http://lkml.kernel.org/r/aec2b5e3973781ff8a6bb6760f8543643202c451.1544099024.git.andreyknvl@google.com Signed-off-by: Andrey Konovalov <andreyknvl@google.com> Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Dmitry Vyukov <dvyukov@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Will Deacon <will.deacon@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-12-28 16:29:37 +08:00
ret = kasan_kmalloc(cachep, ret, size, flags);
trace_kmalloc_node(_RET_IP_, ret,
size, cachep->size,
flags, nodeid);
return ret;
}
EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
#endif
static __always_inline void *
__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller)
[PATCH] add kmalloc_node, inline cleanup The patch makes the following function calls available to allocate memory on a specific node without changing the basic operation of the slab allocator: kmem_cache_alloc_node(kmem_cache_t *cachep, unsigned int flags, int node); kmalloc_node(size_t size, unsigned int flags, int node); in a similar way to the existing node-blind functions: kmem_cache_alloc(kmem_cache_t *cachep, unsigned int flags); kmalloc(size, flags); kmem_cache_alloc_node was changed to pass flags and the node information through the existing layers of the slab allocator (which lead to some minor rearrangements). The functions at the lowest layer (kmem_getpages, cache_grow) are already node aware. Also __alloc_percpu can call kmalloc_node now. Performance measurements (using the pageset localization patch) yields: w/o patches: Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.97 Wed Mar 30 20:50:43 2005 100 25170.83 91 251.7083 23.12 150.10 Wed Mar 30 20:51:06 2005 200 34601.66 84 173.0083 33.64 294.14 Wed Mar 30 20:51:40 2005 300 37154.47 86 123.8482 46.99 436.56 Wed Mar 30 20:52:28 2005 400 39839.82 80 99.5995 58.43 580.46 Wed Mar 30 20:53:27 2005 500 40036.32 79 80.0726 72.68 728.60 Wed Mar 30 20:54:40 2005 600 44074.21 79 73.4570 79.23 872.10 Wed Mar 30 20:55:59 2005 700 44016.60 78 62.8809 92.56 1015.84 Wed Mar 30 20:57:32 2005 800 40411.05 80 50.5138 115.22 1161.13 Wed Mar 30 20:59:28 2005 900 42298.56 79 46.9984 123.83 1303.42 Wed Mar 30 21:01:33 2005 1000 40955.05 80 40.9551 142.11 1441.92 Wed Mar 30 21:03:55 2005 with pageset localization and slab API patches: Tasks jobs/min jti jobs/min/task real cpu 1 484.19 100 484.1930 12.02 1.98 Wed Mar 30 21:10:18 2005 100 27428.25 92 274.2825 21.22 149.79 Wed Mar 30 21:10:40 2005 200 37228.94 86 186.1447 31.27 293.49 Wed Mar 30 21:11:12 2005 300 41725.42 85 139.0847 41.84 434.10 Wed Mar 30 21:11:54 2005 400 43032.22 82 107.5805 54.10 582.06 Wed Mar 30 21:12:48 2005 500 42211.23 83 84.4225 68.94 722.61 Wed Mar 30 21:13:58 2005 600 40084.49 82 66.8075 87.12 873.11 Wed Mar 30 21:15:25 2005 700 44169.30 79 63.0990 92.24 1008.77 Wed Mar 30 21:16:58 2005 800 43097.94 79 53.8724 108.03 1155.88 Wed Mar 30 21:18:47 2005 900 41846.75 79 46.4964 125.17 1303.38 Wed Mar 30 21:20:52 2005 1000 40247.85 79 40.2478 144.60 1442.21 Wed Mar 30 21:23:17 2005 Signed-off-by: Christoph Lameter <christoph@lameter.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:38 +08:00
{
struct kmem_cache *cachep;
void *ret;
[PATCH] add kmalloc_node, inline cleanup The patch makes the following function calls available to allocate memory on a specific node without changing the basic operation of the slab allocator: kmem_cache_alloc_node(kmem_cache_t *cachep, unsigned int flags, int node); kmalloc_node(size_t size, unsigned int flags, int node); in a similar way to the existing node-blind functions: kmem_cache_alloc(kmem_cache_t *cachep, unsigned int flags); kmalloc(size, flags); kmem_cache_alloc_node was changed to pass flags and the node information through the existing layers of the slab allocator (which lead to some minor rearrangements). The functions at the lowest layer (kmem_getpages, cache_grow) are already node aware. Also __alloc_percpu can call kmalloc_node now. Performance measurements (using the pageset localization patch) yields: w/o patches: Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.97 Wed Mar 30 20:50:43 2005 100 25170.83 91 251.7083 23.12 150.10 Wed Mar 30 20:51:06 2005 200 34601.66 84 173.0083 33.64 294.14 Wed Mar 30 20:51:40 2005 300 37154.47 86 123.8482 46.99 436.56 Wed Mar 30 20:52:28 2005 400 39839.82 80 99.5995 58.43 580.46 Wed Mar 30 20:53:27 2005 500 40036.32 79 80.0726 72.68 728.60 Wed Mar 30 20:54:40 2005 600 44074.21 79 73.4570 79.23 872.10 Wed Mar 30 20:55:59 2005 700 44016.60 78 62.8809 92.56 1015.84 Wed Mar 30 20:57:32 2005 800 40411.05 80 50.5138 115.22 1161.13 Wed Mar 30 20:59:28 2005 900 42298.56 79 46.9984 123.83 1303.42 Wed Mar 30 21:01:33 2005 1000 40955.05 80 40.9551 142.11 1441.92 Wed Mar 30 21:03:55 2005 with pageset localization and slab API patches: Tasks jobs/min jti jobs/min/task real cpu 1 484.19 100 484.1930 12.02 1.98 Wed Mar 30 21:10:18 2005 100 27428.25 92 274.2825 21.22 149.79 Wed Mar 30 21:10:40 2005 200 37228.94 86 186.1447 31.27 293.49 Wed Mar 30 21:11:12 2005 300 41725.42 85 139.0847 41.84 434.10 Wed Mar 30 21:11:54 2005 400 43032.22 82 107.5805 54.10 582.06 Wed Mar 30 21:12:48 2005 500 42211.23 83 84.4225 68.94 722.61 Wed Mar 30 21:13:58 2005 600 40084.49 82 66.8075 87.12 873.11 Wed Mar 30 21:15:25 2005 700 44169.30 79 63.0990 92.24 1008.77 Wed Mar 30 21:16:58 2005 800 43097.94 79 53.8724 108.03 1155.88 Wed Mar 30 21:18:47 2005 900 41846.75 79 46.4964 125.17 1303.38 Wed Mar 30 21:20:52 2005 1000 40247.85 79 40.2478 144.60 1442.21 Wed Mar 30 21:23:17 2005 Signed-off-by: Christoph Lameter <christoph@lameter.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:38 +08:00
mm: don't warn about large allocations for slab Slub does not call kmalloc_slab() for sizes > KMALLOC_MAX_CACHE_SIZE, instead it falls back to kmalloc_large(). For slab KMALLOC_MAX_CACHE_SIZE == KMALLOC_MAX_SIZE and it calls kmalloc_slab() for all allocations relying on NULL return value for over-sized allocations. This inconsistency leads to unwanted warnings from kmalloc_slab() for over-sized allocations for slab. Returning NULL for failed allocations is the expected behavior. Make slub and slab code consistent by checking size > KMALLOC_MAX_CACHE_SIZE in slab before calling kmalloc_slab(). While we are here also fix the check in kmalloc_slab(). We should check against KMALLOC_MAX_CACHE_SIZE rather than KMALLOC_MAX_SIZE. It all kinda worked because for slab the constants are the same, and slub always checks the size against KMALLOC_MAX_CACHE_SIZE before kmalloc_slab(). But if we get there with size > KMALLOC_MAX_CACHE_SIZE anyhow bad things will happen. For example, in case of a newly introduced bug in slub code. Also move the check in kmalloc_slab() from function entry to the size > 192 case. This partially compensates for the additional check in slab code and makes slub code a bit faster (at least theoretically). Also drop __GFP_NOWARN in the warning check. This warning means a bug in slab code itself, user-passed flags have nothing to do with it. Nothing of this affects slob. Link: http://lkml.kernel.org/r/20180927171502.226522-1-dvyukov@gmail.com Signed-off-by: Dmitry Vyukov <dvyukov@google.com> Reported-by: syzbot+87829a10073277282ad1@syzkaller.appspotmail.com Reported-by: syzbot+ef4e8fc3a06e9019bb40@syzkaller.appspotmail.com Reported-by: syzbot+6e438f4036df52cbb863@syzkaller.appspotmail.com Reported-by: syzbot+8574471d8734457d98aa@syzkaller.appspotmail.com Reported-by: syzbot+af1504df0807a083dbd9@syzkaller.appspotmail.com Acked-by: Christoph Lameter <cl@linux.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-27 06:03:12 +08:00
if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
return NULL;
cachep = kmalloc_slab(size, flags);
if (unlikely(ZERO_OR_NULL_PTR(cachep)))
return cachep;
ret = kmem_cache_alloc_node_trace(cachep, flags, node, size);
kasan, mm: change hooks signatures Patch series "kasan: add software tag-based mode for arm64", v13. This patchset adds a new software tag-based mode to KASAN [1]. (Initially this mode was called KHWASAN, but it got renamed, see the naming rationale at the end of this section). The plan is to implement HWASan [2] for the kernel with the incentive, that it's going to have comparable to KASAN performance, but in the same time consume much less memory, trading that off for somewhat imprecise bug detection and being supported only for arm64. The underlying ideas of the approach used by software tag-based KASAN are: 1. By using the Top Byte Ignore (TBI) arm64 CPU feature, we can store pointer tags in the top byte of each kernel pointer. 2. Using shadow memory, we can store memory tags for each chunk of kernel memory. 3. On each memory allocation, we can generate a random tag, embed it into the returned pointer and set the memory tags that correspond to this chunk of memory to the same value. 4. By using compiler instrumentation, before each memory access we can add a check that the pointer tag matches the tag of the memory that is being accessed. 5. On a tag mismatch we report an error. With this patchset the existing KASAN mode gets renamed to generic KASAN, with the word "generic" meaning that the implementation can be supported by any architecture as it is purely software. The new mode this patchset adds is called software tag-based KASAN. The word "tag-based" refers to the fact that this mode uses tags embedded into the top byte of kernel pointers and the TBI arm64 CPU feature that allows to dereference such pointers. The word "software" here means that shadow memory manipulation and tag checking on pointer dereference is done in software. As it is the only tag-based implementation right now, "software tag-based" KASAN is sometimes referred to as simply "tag-based" in this patchset. A potential expansion of this mode is a hardware tag-based mode, which would use hardware memory tagging support (announced by Arm [3]) instead of compiler instrumentation and manual shadow memory manipulation. Same as generic KASAN, software tag-based KASAN is strictly a debugging feature. [1] https://www.kernel.org/doc/html/latest/dev-tools/kasan.html [2] http://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html [3] https://community.arm.com/processors/b/blog/posts/arm-a-profile-architecture-2018-developments-armv85a ====== Rationale On mobile devices generic KASAN's memory usage is significant problem. One of the main reasons to have tag-based KASAN is to be able to perform a similar set of checks as the generic one does, but with lower memory requirements. Comment from Vishwath Mohan <vishwath@google.com>: I don't have data on-hand, but anecdotally both ASAN and KASAN have proven problematic to enable for environments that don't tolerate the increased memory pressure well. This includes (a) Low-memory form factors - Wear, TV, Things, lower-tier phones like Go, (c) Connected components like Pixel's visual core [1]. These are both places I'd love to have a low(er) memory footprint option at my disposal. Comment from Evgenii Stepanov <eugenis@google.com>: Looking at a live Android device under load, slab (according to /proc/meminfo) + kernel stack take 8-10% available RAM (~350MB). KASAN's overhead of 2x - 3x on top of it is not insignificant. Not having this overhead enables near-production use - ex. running KASAN/KHWASAN kernel on a personal, daily-use device to catch bugs that do not reproduce in test configuration. These are the ones that often cost the most engineering time to track down. CPU overhead is bad, but generally tolerable. RAM is critical, in our experience. Once it gets low enough, OOM-killer makes your life miserable. [1] https://www.blog.google/products/pixel/pixel-visual-core-image-processing-and-machine-learning-pixel-2/ ====== Technical details Software tag-based KASAN mode is implemented in a very similar way to the generic one. This patchset essentially does the following: 1. TCR_TBI1 is set to enable Top Byte Ignore. 2. Shadow memory is used (with a different scale, 1:16, so each shadow byte corresponds to 16 bytes of kernel memory) to store memory tags. 3. All slab objects are aligned to shadow scale, which is 16 bytes. 4. All pointers returned from the slab allocator are tagged with a random tag and the corresponding shadow memory is poisoned with the same value. 5. Compiler instrumentation is used to insert tag checks. Either by calling callbacks or by inlining them (CONFIG_KASAN_OUTLINE and CONFIG_KASAN_INLINE flags are reused). 6. When a tag mismatch is detected in callback instrumentation mode KASAN simply prints a bug report. In case of inline instrumentation, clang inserts a brk instruction, and KASAN has it's own brk handler, which reports the bug. 7. The memory in between slab objects is marked with a reserved tag, and acts as a redzone. 8. When a slab object is freed it's marked with a reserved tag. Bug detection is imprecise for two reasons: 1. We won't catch some small out-of-bounds accesses, that fall into the same shadow cell, as the last byte of a slab object. 2. We only have 1 byte to store tags, which means we have a 1/256 probability of a tag match for an incorrect access (actually even slightly less due to reserved tag values). Despite that there's a particular type of bugs that tag-based KASAN can detect compared to generic KASAN: use-after-free after the object has been allocated by someone else. ====== Testing Some kernel developers voiced a concern that changing the top byte of kernel pointers may lead to subtle bugs that are difficult to discover. To address this concern deliberate testing has been performed. It doesn't seem feasible to do some kind of static checking to find potential issues with pointer tagging, so a dynamic approach was taken. All pointer comparisons/subtractions have been instrumented in an LLVM compiler pass and a kernel module that would print a bug report whenever two pointers with different tags are being compared/subtracted (ignoring comparisons with NULL pointers and with pointers obtained by casting an error code to a pointer type) has been used. Then the kernel has been booted in QEMU and on an Odroid C2 board and syzkaller has been run. This yielded the following results. The two places that look interesting are: is_vmalloc_addr in include/linux/mm.h is_kernel_rodata in mm/util.c Here we compare a pointer with some fixed untagged values to make sure that the pointer lies in a particular part of the kernel address space. Since tag-based KASAN doesn't add tags to pointers that belong to rodata or vmalloc regions, this should work as is. To make sure debug checks to those two functions that check that the result doesn't change whether we operate on pointers with or without untagging has been added. A few other cases that don't look that interesting: Comparing pointers to achieve unique sorting order of pointee objects (e.g. sorting locks addresses before performing a double lock): tty_ldisc_lock_pair_timeout in drivers/tty/tty_ldisc.c pipe_double_lock in fs/pipe.c unix_state_double_lock in net/unix/af_unix.c lock_two_nondirectories in fs/inode.c mutex_lock_double in kernel/events/core.c ep_cmp_ffd in fs/eventpoll.c fsnotify_compare_groups fs/notify/mark.c Nothing needs to be done here, since the tags embedded into pointers don't change, so the sorting order would still be unique. Checks that a pointer belongs to some particular allocation: is_sibling_entry in lib/radix-tree.c object_is_on_stack in include/linux/sched/task_stack.h Nothing needs to be done here either, since two pointers can only belong to the same allocation if they have the same tag. Overall, since the kernel boots and works, there are no critical bugs. As for the rest, the traditional kernel testing way (use until fails) is the only one that looks feasible. Another point here is that tag-based KASAN is available under a separate config option that needs to be deliberately enabled. Even though it might be used in a "near-production" environment to find bugs that are not found during fuzzing or running tests, it is still a debug tool. ====== Benchmarks The following numbers were collected on Odroid C2 board. Both generic and tag-based KASAN were used in inline instrumentation mode. Boot time [1]: * ~1.7 sec for clean kernel * ~5.0 sec for generic KASAN * ~5.0 sec for tag-based KASAN Network performance [2]: * 8.33 Gbits/sec for clean kernel * 3.17 Gbits/sec for generic KASAN * 2.85 Gbits/sec for tag-based KASAN Slab memory usage after boot [3]: * ~40 kb for clean kernel * ~105 kb (~260% overhead) for generic KASAN * ~47 kb (~20% overhead) for tag-based KASAN KASAN memory overhead consists of three main parts: 1. Increased slab memory usage due to redzones. 2. Shadow memory (the whole reserved once during boot). 3. Quaratine (grows gradually until some preset limit; the more the limit, the more the chance to detect a use-after-free). Comparing tag-based vs generic KASAN for each of these points: 1. 20% vs 260% overhead. 2. 1/16th vs 1/8th of physical memory. 3. Tag-based KASAN doesn't require quarantine. [1] Time before the ext4 driver is initialized. [2] Measured as `iperf -s & iperf -c 127.0.0.1 -t 30`. [3] Measured as `cat /proc/meminfo | grep Slab`. ====== Some notes A few notes: 1. The patchset can be found here: https://github.com/xairy/kasan-prototype/tree/khwasan 2. Building requires a recent Clang version (7.0.0 or later). 3. Stack instrumentation is not supported yet and will be added later. This patch (of 25): Tag-based KASAN changes the value of the top byte of pointers returned from the kernel allocation functions (such as kmalloc). This patch updates KASAN hooks signatures and their usage in SLAB and SLUB code to reflect that. Link: http://lkml.kernel.org/r/aec2b5e3973781ff8a6bb6760f8543643202c451.1544099024.git.andreyknvl@google.com Signed-off-by: Andrey Konovalov <andreyknvl@google.com> Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Dmitry Vyukov <dvyukov@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Will Deacon <will.deacon@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-12-28 16:29:37 +08:00
ret = kasan_kmalloc(cachep, ret, size, flags);
return ret;
[PATCH] add kmalloc_node, inline cleanup The patch makes the following function calls available to allocate memory on a specific node without changing the basic operation of the slab allocator: kmem_cache_alloc_node(kmem_cache_t *cachep, unsigned int flags, int node); kmalloc_node(size_t size, unsigned int flags, int node); in a similar way to the existing node-blind functions: kmem_cache_alloc(kmem_cache_t *cachep, unsigned int flags); kmalloc(size, flags); kmem_cache_alloc_node was changed to pass flags and the node information through the existing layers of the slab allocator (which lead to some minor rearrangements). The functions at the lowest layer (kmem_getpages, cache_grow) are already node aware. Also __alloc_percpu can call kmalloc_node now. Performance measurements (using the pageset localization patch) yields: w/o patches: Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.97 Wed Mar 30 20:50:43 2005 100 25170.83 91 251.7083 23.12 150.10 Wed Mar 30 20:51:06 2005 200 34601.66 84 173.0083 33.64 294.14 Wed Mar 30 20:51:40 2005 300 37154.47 86 123.8482 46.99 436.56 Wed Mar 30 20:52:28 2005 400 39839.82 80 99.5995 58.43 580.46 Wed Mar 30 20:53:27 2005 500 40036.32 79 80.0726 72.68 728.60 Wed Mar 30 20:54:40 2005 600 44074.21 79 73.4570 79.23 872.10 Wed Mar 30 20:55:59 2005 700 44016.60 78 62.8809 92.56 1015.84 Wed Mar 30 20:57:32 2005 800 40411.05 80 50.5138 115.22 1161.13 Wed Mar 30 20:59:28 2005 900 42298.56 79 46.9984 123.83 1303.42 Wed Mar 30 21:01:33 2005 1000 40955.05 80 40.9551 142.11 1441.92 Wed Mar 30 21:03:55 2005 with pageset localization and slab API patches: Tasks jobs/min jti jobs/min/task real cpu 1 484.19 100 484.1930 12.02 1.98 Wed Mar 30 21:10:18 2005 100 27428.25 92 274.2825 21.22 149.79 Wed Mar 30 21:10:40 2005 200 37228.94 86 186.1447 31.27 293.49 Wed Mar 30 21:11:12 2005 300 41725.42 85 139.0847 41.84 434.10 Wed Mar 30 21:11:54 2005 400 43032.22 82 107.5805 54.10 582.06 Wed Mar 30 21:12:48 2005 500 42211.23 83 84.4225 68.94 722.61 Wed Mar 30 21:13:58 2005 600 40084.49 82 66.8075 87.12 873.11 Wed Mar 30 21:15:25 2005 700 44169.30 79 63.0990 92.24 1008.77 Wed Mar 30 21:16:58 2005 800 43097.94 79 53.8724 108.03 1155.88 Wed Mar 30 21:18:47 2005 900 41846.75 79 46.4964 125.17 1303.38 Wed Mar 30 21:20:52 2005 1000 40247.85 79 40.2478 144.60 1442.21 Wed Mar 30 21:23:17 2005 Signed-off-by: Christoph Lameter <christoph@lameter.com> Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-05-01 23:58:38 +08:00
}
void *__kmalloc_node(size_t size, gfp_t flags, int node)
{
return __do_kmalloc_node(size, flags, node, _RET_IP_);
}
EXPORT_SYMBOL(__kmalloc_node);
void *__kmalloc_node_track_caller(size_t size, gfp_t flags,
int node, unsigned long caller)
{
return __do_kmalloc_node(size, flags, node, caller);
}
EXPORT_SYMBOL(__kmalloc_node_track_caller);
#endif /* CONFIG_NUMA */
/**
* __do_kmalloc - allocate memory
* @size: how many bytes of memory are required.
* @flags: the type of memory to allocate (see kmalloc).
* @caller: function caller for debug tracking of the caller
*
* Return: pointer to the allocated memory or %NULL in case of error
*/
static __always_inline void *__do_kmalloc(size_t size, gfp_t flags,
unsigned long caller)
{
struct kmem_cache *cachep;
void *ret;
mm: don't warn about large allocations for slab Slub does not call kmalloc_slab() for sizes > KMALLOC_MAX_CACHE_SIZE, instead it falls back to kmalloc_large(). For slab KMALLOC_MAX_CACHE_SIZE == KMALLOC_MAX_SIZE and it calls kmalloc_slab() for all allocations relying on NULL return value for over-sized allocations. This inconsistency leads to unwanted warnings from kmalloc_slab() for over-sized allocations for slab. Returning NULL for failed allocations is the expected behavior. Make slub and slab code consistent by checking size > KMALLOC_MAX_CACHE_SIZE in slab before calling kmalloc_slab(). While we are here also fix the check in kmalloc_slab(). We should check against KMALLOC_MAX_CACHE_SIZE rather than KMALLOC_MAX_SIZE. It all kinda worked because for slab the constants are the same, and slub always checks the size against KMALLOC_MAX_CACHE_SIZE before kmalloc_slab(). But if we get there with size > KMALLOC_MAX_CACHE_SIZE anyhow bad things will happen. For example, in case of a newly introduced bug in slub code. Also move the check in kmalloc_slab() from function entry to the size > 192 case. This partially compensates for the additional check in slab code and makes slub code a bit faster (at least theoretically). Also drop __GFP_NOWARN in the warning check. This warning means a bug in slab code itself, user-passed flags have nothing to do with it. Nothing of this affects slob. Link: http://lkml.kernel.org/r/20180927171502.226522-1-dvyukov@gmail.com Signed-off-by: Dmitry Vyukov <dvyukov@google.com> Reported-by: syzbot+87829a10073277282ad1@syzkaller.appspotmail.com Reported-by: syzbot+ef4e8fc3a06e9019bb40@syzkaller.appspotmail.com Reported-by: syzbot+6e438f4036df52cbb863@syzkaller.appspotmail.com Reported-by: syzbot+8574471d8734457d98aa@syzkaller.appspotmail.com Reported-by: syzbot+af1504df0807a083dbd9@syzkaller.appspotmail.com Acked-by: Christoph Lameter <cl@linux.com> Acked-by: Vlastimil Babka <vbabka@suse.cz> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-27 06:03:12 +08:00
if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
return NULL;
cachep = kmalloc_slab(size, flags);
if (unlikely(ZERO_OR_NULL_PTR(cachep)))
return cachep;
ret = slab_alloc(cachep, flags, caller);
kasan, mm: change hooks signatures Patch series "kasan: add software tag-based mode for arm64", v13. This patchset adds a new software tag-based mode to KASAN [1]. (Initially this mode was called KHWASAN, but it got renamed, see the naming rationale at the end of this section). The plan is to implement HWASan [2] for the kernel with the incentive, that it's going to have comparable to KASAN performance, but in the same time consume much less memory, trading that off for somewhat imprecise bug detection and being supported only for arm64. The underlying ideas of the approach used by software tag-based KASAN are: 1. By using the Top Byte Ignore (TBI) arm64 CPU feature, we can store pointer tags in the top byte of each kernel pointer. 2. Using shadow memory, we can store memory tags for each chunk of kernel memory. 3. On each memory allocation, we can generate a random tag, embed it into the returned pointer and set the memory tags that correspond to this chunk of memory to the same value. 4. By using compiler instrumentation, before each memory access we can add a check that the pointer tag matches the tag of the memory that is being accessed. 5. On a tag mismatch we report an error. With this patchset the existing KASAN mode gets renamed to generic KASAN, with the word "generic" meaning that the implementation can be supported by any architecture as it is purely software. The new mode this patchset adds is called software tag-based KASAN. The word "tag-based" refers to the fact that this mode uses tags embedded into the top byte of kernel pointers and the TBI arm64 CPU feature that allows to dereference such pointers. The word "software" here means that shadow memory manipulation and tag checking on pointer dereference is done in software. As it is the only tag-based implementation right now, "software tag-based" KASAN is sometimes referred to as simply "tag-based" in this patchset. A potential expansion of this mode is a hardware tag-based mode, which would use hardware memory tagging support (announced by Arm [3]) instead of compiler instrumentation and manual shadow memory manipulation. Same as generic KASAN, software tag-based KASAN is strictly a debugging feature. [1] https://www.kernel.org/doc/html/latest/dev-tools/kasan.html [2] http://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html [3] https://community.arm.com/processors/b/blog/posts/arm-a-profile-architecture-2018-developments-armv85a ====== Rationale On mobile devices generic KASAN's memory usage is significant problem. One of the main reasons to have tag-based KASAN is to be able to perform a similar set of checks as the generic one does, but with lower memory requirements. Comment from Vishwath Mohan <vishwath@google.com>: I don't have data on-hand, but anecdotally both ASAN and KASAN have proven problematic to enable for environments that don't tolerate the increased memory pressure well. This includes (a) Low-memory form factors - Wear, TV, Things, lower-tier phones like Go, (c) Connected components like Pixel's visual core [1]. These are both places I'd love to have a low(er) memory footprint option at my disposal. Comment from Evgenii Stepanov <eugenis@google.com>: Looking at a live Android device under load, slab (according to /proc/meminfo) + kernel stack take 8-10% available RAM (~350MB). KASAN's overhead of 2x - 3x on top of it is not insignificant. Not having this overhead enables near-production use - ex. running KASAN/KHWASAN kernel on a personal, daily-use device to catch bugs that do not reproduce in test configuration. These are the ones that often cost the most engineering time to track down. CPU overhead is bad, but generally tolerable. RAM is critical, in our experience. Once it gets low enough, OOM-killer makes your life miserable. [1] https://www.blog.google/products/pixel/pixel-visual-core-image-processing-and-machine-learning-pixel-2/ ====== Technical details Software tag-based KASAN mode is implemented in a very similar way to the generic one. This patchset essentially does the following: 1. TCR_TBI1 is set to enable Top Byte Ignore. 2. Shadow memory is used (with a different scale, 1:16, so each shadow byte corresponds to 16 bytes of kernel memory) to store memory tags. 3. All slab objects are aligned to shadow scale, which is 16 bytes. 4. All pointers returned from the slab allocator are tagged with a random tag and the corresponding shadow memory is poisoned with the same value. 5. Compiler instrumentation is used to insert tag checks. Either by calling callbacks or by inlining them (CONFIG_KASAN_OUTLINE and CONFIG_KASAN_INLINE flags are reused). 6. When a tag mismatch is detected in callback instrumentation mode KASAN simply prints a bug report. In case of inline instrumentation, clang inserts a brk instruction, and KASAN has it's own brk handler, which reports the bug. 7. The memory in between slab objects is marked with a reserved tag, and acts as a redzone. 8. When a slab object is freed it's marked with a reserved tag. Bug detection is imprecise for two reasons: 1. We won't catch some small out-of-bounds accesses, that fall into the same shadow cell, as the last byte of a slab object. 2. We only have 1 byte to store tags, which means we have a 1/256 probability of a tag match for an incorrect access (actually even slightly less due to reserved tag values). Despite that there's a particular type of bugs that tag-based KASAN can detect compared to generic KASAN: use-after-free after the object has been allocated by someone else. ====== Testing Some kernel developers voiced a concern that changing the top byte of kernel pointers may lead to subtle bugs that are difficult to discover. To address this concern deliberate testing has been performed. It doesn't seem feasible to do some kind of static checking to find potential issues with pointer tagging, so a dynamic approach was taken. All pointer comparisons/subtractions have been instrumented in an LLVM compiler pass and a kernel module that would print a bug report whenever two pointers with different tags are being compared/subtracted (ignoring comparisons with NULL pointers and with pointers obtained by casting an error code to a pointer type) has been used. Then the kernel has been booted in QEMU and on an Odroid C2 board and syzkaller has been run. This yielded the following results. The two places that look interesting are: is_vmalloc_addr in include/linux/mm.h is_kernel_rodata in mm/util.c Here we compare a pointer with some fixed untagged values to make sure that the pointer lies in a particular part of the kernel address space. Since tag-based KASAN doesn't add tags to pointers that belong to rodata or vmalloc regions, this should work as is. To make sure debug checks to those two functions that check that the result doesn't change whether we operate on pointers with or without untagging has been added. A few other cases that don't look that interesting: Comparing pointers to achieve unique sorting order of pointee objects (e.g. sorting locks addresses before performing a double lock): tty_ldisc_lock_pair_timeout in drivers/tty/tty_ldisc.c pipe_double_lock in fs/pipe.c unix_state_double_lock in net/unix/af_unix.c lock_two_nondirectories in fs/inode.c mutex_lock_double in kernel/events/core.c ep_cmp_ffd in fs/eventpoll.c fsnotify_compare_groups fs/notify/mark.c Nothing needs to be done here, since the tags embedded into pointers don't change, so the sorting order would still be unique. Checks that a pointer belongs to some particular allocation: is_sibling_entry in lib/radix-tree.c object_is_on_stack in include/linux/sched/task_stack.h Nothing needs to be done here either, since two pointers can only belong to the same allocation if they have the same tag. Overall, since the kernel boots and works, there are no critical bugs. As for the rest, the traditional kernel testing way (use until fails) is the only one that looks feasible. Another point here is that tag-based KASAN is available under a separate config option that needs to be deliberately enabled. Even though it might be used in a "near-production" environment to find bugs that are not found during fuzzing or running tests, it is still a debug tool. ====== Benchmarks The following numbers were collected on Odroid C2 board. Both generic and tag-based KASAN were used in inline instrumentation mode. Boot time [1]: * ~1.7 sec for clean kernel * ~5.0 sec for generic KASAN * ~5.0 sec for tag-based KASAN Network performance [2]: * 8.33 Gbits/sec for clean kernel * 3.17 Gbits/sec for generic KASAN * 2.85 Gbits/sec for tag-based KASAN Slab memory usage after boot [3]: * ~40 kb for clean kernel * ~105 kb (~260% overhead) for generic KASAN * ~47 kb (~20% overhead) for tag-based KASAN KASAN memory overhead consists of three main parts: 1. Increased slab memory usage due to redzones. 2. Shadow memory (the whole reserved once during boot). 3. Quaratine (grows gradually until some preset limit; the more the limit, the more the chance to detect a use-after-free). Comparing tag-based vs generic KASAN for each of these points: 1. 20% vs 260% overhead. 2. 1/16th vs 1/8th of physical memory. 3. Tag-based KASAN doesn't require quarantine. [1] Time before the ext4 driver is initialized. [2] Measured as `iperf -s & iperf -c 127.0.0.1 -t 30`. [3] Measured as `cat /proc/meminfo | grep Slab`. ====== Some notes A few notes: 1. The patchset can be found here: https://github.com/xairy/kasan-prototype/tree/khwasan 2. Building requires a recent Clang version (7.0.0 or later). 3. Stack instrumentation is not supported yet and will be added later. This patch (of 25): Tag-based KASAN changes the value of the top byte of pointers returned from the kernel allocation functions (such as kmalloc). This patch updates KASAN hooks signatures and their usage in SLAB and SLUB code to reflect that. Link: http://lkml.kernel.org/r/aec2b5e3973781ff8a6bb6760f8543643202c451.1544099024.git.andreyknvl@google.com Signed-off-by: Andrey Konovalov <andreyknvl@google.com> Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com> Reviewed-by: Dmitry Vyukov <dvyukov@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Will Deacon <will.deacon@arm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-12-28 16:29:37 +08:00
ret = kasan_kmalloc(cachep, ret, size, flags);
trace_kmalloc(caller, ret,
size, cachep->size, flags);
return ret;
}
void *__kmalloc(size_t size, gfp_t flags)
{
return __do_kmalloc(size, flags, _RET_IP_);
}
EXPORT_SYMBOL(__kmalloc);
void *__kmalloc_track_caller(size_t size, gfp_t flags, unsigned long caller)
{
return __do_kmalloc(size, flags, caller);
}
EXPORT_SYMBOL(__kmalloc_track_caller);
/**
* kmem_cache_free - Deallocate an object
* @cachep: The cache the allocation was from.
* @objp: The previously allocated object.
*
* Free an object which was previously allocated from this
* cache.
*/
void kmem_cache_free(struct kmem_cache *cachep, void *objp)
{
unsigned long flags;
cachep = cache_from_obj(cachep, objp);
if (!cachep)
return;
local_irq_save(flags);
debug_check_no_locks_freed(objp, cachep->object_size);
infrastructure to debug (dynamic) objects We can see an ever repeating problem pattern with objects of any kind in the kernel: 1) freeing of active objects 2) reinitialization of active objects Both problems can be hard to debug because the crash happens at a point where we have no chance to decode the root cause anymore. One problem spot are kernel timers, where the detection of the problem often happens in interrupt context and usually causes the machine to panic. While working on a timer related bug report I had to hack specialized code into the timer subsystem to get a reasonable hint for the root cause. This debug hack was fine for temporary use, but far from a mergeable solution due to the intrusiveness into the timer code. The code further lacked the ability to detect and report the root cause instantly and keep the system operational. Keeping the system operational is important to get hold of the debug information without special debugging aids like serial consoles and special knowledge of the bug reporter. The problems described above are not restricted to timers, but timers tend to expose it usually in a full system crash. Other objects are less explosive, but the symptoms caused by such mistakes can be even harder to debug. Instead of creating specialized debugging code for the timer subsystem a generic infrastructure is created which allows developers to verify their code and provides an easy to enable debug facility for users in case of trouble. The debugobjects core code keeps track of operations on static and dynamic objects by inserting them into a hashed list and sanity checking them on object operations and provides additional checks whenever kernel memory is freed. The tracked object operations are: - initializing an object - adding an object to a subsystem list - deleting an object from a subsystem list Each operation is sanity checked before the operation is executed and the subsystem specific code can provide a fixup function which allows to prevent the damage of the operation. When the sanity check triggers a warning message and a stack trace is printed. The list of operations can be extended if the need arises. For now it's limited to the requirements of the first user (timers). The core code enqueues the objects into hash buckets. The hash index is generated from the address of the object to simplify the lookup for the check on kfree/vfree. Each bucket has it's own spinlock to avoid contention on a global lock. The debug code can be compiled in without being active. The runtime overhead is minimal and could be optimized by asm alternatives. A kernel command line option enables the debugging code. Thanks to Ingo Molnar for review, suggestions and cleanup patches. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Ingo Molnar <mingo@elte.hu> Cc: Greg KH <greg@kroah.com> Cc: Randy Dunlap <randy.dunlap@oracle.com> Cc: Kay Sievers <kay.sievers@vrfy.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-04-30 15:55:01 +08:00
if (!(cachep->flags & SLAB_DEBUG_OBJECTS))
debug_check_no_obj_freed(objp, cachep->object_size);
__cache_free(cachep, objp, _RET_IP_);
local_irq_restore(flags);
trace_kmem_cache_free(_RET_IP_, objp);
}
EXPORT_SYMBOL(kmem_cache_free);
slab: implement bulk free in SLAB allocator This patch implements the free side of bulk API for the SLAB allocator kmem_cache_free_bulk(), and concludes the implementation of optimized bulk API for SLAB allocator. Benchmarked[1] cost of alloc+free (obj size 256 bytes) on CPU i7-4790K @ 4.00GHz, with no debug options, no PREEMPT and CONFIG_MEMCG_KMEM=y but no active user of kmemcg. SLAB single alloc+free cost: 87 cycles(tsc) 21.814 ns with this optimized config. bulk- Current fallback - optimized SLAB bulk 1 - 102 cycles(tsc) 25.747 ns - 41 cycles(tsc) 10.490 ns - improved 59.8% 2 - 94 cycles(tsc) 23.546 ns - 26 cycles(tsc) 6.567 ns - improved 72.3% 3 - 92 cycles(tsc) 23.127 ns - 20 cycles(tsc) 5.244 ns - improved 78.3% 4 - 90 cycles(tsc) 22.663 ns - 18 cycles(tsc) 4.588 ns - improved 80.0% 8 - 88 cycles(tsc) 22.242 ns - 14 cycles(tsc) 3.656 ns - improved 84.1% 16 - 88 cycles(tsc) 22.010 ns - 13 cycles(tsc) 3.480 ns - improved 85.2% 30 - 89 cycles(tsc) 22.305 ns - 13 cycles(tsc) 3.303 ns - improved 85.4% 32 - 89 cycles(tsc) 22.277 ns - 13 cycles(tsc) 3.309 ns - improved 85.4% 34 - 88 cycles(tsc) 22.246 ns - 13 cycles(tsc) 3.294 ns - improved 85.2% 48 - 88 cycles(tsc) 22.121 ns - 13 cycles(tsc) 3.492 ns - improved 85.2% 64 - 88 cycles(tsc) 22.052 ns - 13 cycles(tsc) 3.411 ns - improved 85.2% 128 - 89 cycles(tsc) 22.452 ns - 15 cycles(tsc) 3.841 ns - improved 83.1% 158 - 89 cycles(tsc) 22.403 ns - 14 cycles(tsc) 3.746 ns - improved 84.3% 250 - 91 cycles(tsc) 22.775 ns - 16 cycles(tsc) 4.111 ns - improved 82.4% Notice it is not recommended to do very large bulk operation with this bulk API, because local IRQs are disabled in this period. [1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/mm/slab_bulk_test01.c Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Vladimir Davydov <vdavydov@virtuozzo.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:53:56 +08:00
void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p)
{
struct kmem_cache *s;
size_t i;
local_irq_disable();
for (i = 0; i < size; i++) {
void *objp = p[i];
mm: new API kfree_bulk() for SLAB+SLUB allocators This patch introduce a new API call kfree_bulk() for bulk freeing memory objects not bound to a single kmem_cache. Christoph pointed out that it is possible to implement freeing of objects, without knowing the kmem_cache pointer as that information is available from the object's page->slab_cache. Proposing to remove the kmem_cache argument from the bulk free API. Jesper demonstrated that these extra steps per object comes at a performance cost. It is only in the case CONFIG_MEMCG_KMEM is compiled in and activated runtime that these steps are done anyhow. The extra cost is most visible for SLAB allocator, because the SLUB allocator does the page lookup (virt_to_head_page()) anyhow. Thus, the conclusion was to keep the kmem_cache free bulk API with a kmem_cache pointer, but we can still implement a kfree_bulk() API fairly easily. Simply by handling if kmem_cache_free_bulk() gets called with a kmem_cache NULL pointer. This does increase the code size a bit, but implementing a separate kfree_bulk() call would likely increase code size even more. Below benchmarks cost of alloc+free (obj size 256 bytes) on CPU i7-4790K @ 4.00GHz, no PREEMPT and CONFIG_MEMCG_KMEM=y. Code size increase for SLAB: add/remove: 0/0 grow/shrink: 1/0 up/down: 74/0 (74) function old new delta kmem_cache_free_bulk 660 734 +74 SLAB fastpath: 87 cycles(tsc) 21.814 sz - fallback - kmem_cache_free_bulk - kfree_bulk 1 - 103 cycles 25.878 ns - 41 cycles 10.498 ns - 81 cycles 20.312 ns 2 - 94 cycles 23.673 ns - 26 cycles 6.682 ns - 42 cycles 10.649 ns 3 - 92 cycles 23.181 ns - 21 cycles 5.325 ns - 39 cycles 9.950 ns 4 - 90 cycles 22.727 ns - 18 cycles 4.673 ns - 26 cycles 6.693 ns 8 - 89 cycles 22.270 ns - 14 cycles 3.664 ns - 23 cycles 5.835 ns 16 - 88 cycles 22.038 ns - 14 cycles 3.503 ns - 22 cycles 5.543 ns 30 - 89 cycles 22.284 ns - 13 cycles 3.310 ns - 20 cycles 5.197 ns 32 - 88 cycles 22.249 ns - 13 cycles 3.420 ns - 20 cycles 5.166 ns 34 - 88 cycles 22.224 ns - 14 cycles 3.643 ns - 20 cycles 5.170 ns 48 - 88 cycles 22.088 ns - 14 cycles 3.507 ns - 20 cycles 5.203 ns 64 - 88 cycles 22.063 ns - 13 cycles 3.428 ns - 20 cycles 5.152 ns 128 - 89 cycles 22.483 ns - 15 cycles 3.891 ns - 23 cycles 5.885 ns 158 - 89 cycles 22.381 ns - 15 cycles 3.779 ns - 22 cycles 5.548 ns 250 - 91 cycles 22.798 ns - 16 cycles 4.152 ns - 23 cycles 5.967 ns SLAB when enabling MEMCG_KMEM runtime: - kmemcg fastpath: 130 cycles(tsc) 32.684 ns (step:0) 1 - 148 cycles 37.220 ns - 66 cycles 16.622 ns - 66 cycles 16.583 ns 2 - 141 cycles 35.510 ns - 51 cycles 12.820 ns - 58 cycles 14.625 ns 3 - 140 cycles 35.017 ns - 37 cycles 9.326 ns - 33 cycles 8.474 ns 4 - 137 cycles 34.507 ns - 31 cycles 7.888 ns - 33 cycles 8.300 ns 8 - 140 cycles 35.069 ns - 25 cycles 6.461 ns - 25 cycles 6.436 ns 16 - 138 cycles 34.542 ns - 23 cycles 5.945 ns - 22 cycles 5.670 ns 30 - 136 cycles 34.227 ns - 22 cycles 5.502 ns - 22 cycles 5.587 ns 32 - 136 cycles 34.253 ns - 21 cycles 5.475 ns - 21 cycles 5.324 ns 34 - 136 cycles 34.254 ns - 21 cycles 5.448 ns - 20 cycles 5.194 ns 48 - 136 cycles 34.075 ns - 21 cycles 5.458 ns - 21 cycles 5.367 ns 64 - 135 cycles 33.994 ns - 21 cycles 5.350 ns - 21 cycles 5.259 ns 128 - 137 cycles 34.446 ns - 23 cycles 5.816 ns - 22 cycles 5.688 ns 158 - 137 cycles 34.379 ns - 22 cycles 5.727 ns - 22 cycles 5.602 ns 250 - 138 cycles 34.755 ns - 24 cycles 6.093 ns - 23 cycles 5.986 ns Code size increase for SLUB: function old new delta kmem_cache_free_bulk 717 799 +82 SLUB benchmark: SLUB fastpath: 46 cycles(tsc) 11.691 ns (step:0) sz - fallback - kmem_cache_free_bulk - kfree_bulk 1 - 61 cycles 15.486 ns - 53 cycles 13.364 ns - 57 cycles 14.464 ns 2 - 54 cycles 13.703 ns - 32 cycles 8.110 ns - 33 cycles 8.482 ns 3 - 53 cycles 13.272 ns - 25 cycles 6.362 ns - 27 cycles 6.947 ns 4 - 51 cycles 12.994 ns - 24 cycles 6.087 ns - 24 cycles 6.078 ns 8 - 50 cycles 12.576 ns - 21 cycles 5.354 ns - 22 cycles 5.513 ns 16 - 49 cycles 12.368 ns - 20 cycles 5.054 ns - 20 cycles 5.042 ns 30 - 49 cycles 12.273 ns - 18 cycles 4.748 ns - 19 cycles 4.758 ns 32 - 49 cycles 12.401 ns - 19 cycles 4.821 ns - 19 cycles 4.810 ns 34 - 98 cycles 24.519 ns - 24 cycles 6.154 ns - 24 cycles 6.157 ns 48 - 83 cycles 20.833 ns - 21 cycles 5.446 ns - 21 cycles 5.429 ns 64 - 75 cycles 18.891 ns - 20 cycles 5.247 ns - 20 cycles 5.238 ns 128 - 93 cycles 23.271 ns - 27 cycles 6.856 ns - 27 cycles 6.823 ns 158 - 102 cycles 25.581 ns - 30 cycles 7.714 ns - 30 cycles 7.695 ns 250 - 107 cycles 26.917 ns - 38 cycles 9.514 ns - 38 cycles 9.506 ns SLUB when enabling MEMCG_KMEM runtime: - kmemcg fastpath: 71 cycles(tsc) 17.897 ns (step:0) 1 - 85 cycles 21.484 ns - 78 cycles 19.569 ns - 75 cycles 18.938 ns 2 - 81 cycles 20.363 ns - 45 cycles 11.258 ns - 44 cycles 11.076 ns 3 - 78 cycles 19.709 ns - 33 cycles 8.354 ns - 32 cycles 8.044 ns 4 - 77 cycles 19.430 ns - 28 cycles 7.216 ns - 28 cycles 7.003 ns 8 - 101 cycles 25.288 ns - 23 cycles 5.849 ns - 23 cycles 5.787 ns 16 - 76 cycles 19.148 ns - 20 cycles 5.162 ns - 20 cycles 5.081 ns 30 - 76 cycles 19.067 ns - 19 cycles 4.868 ns - 19 cycles 4.821 ns 32 - 76 cycles 19.052 ns - 19 cycles 4.857 ns - 19 cycles 4.815 ns 34 - 121 cycles 30.291 ns - 25 cycles 6.333 ns - 25 cycles 6.268 ns 48 - 108 cycles 27.111 ns - 21 cycles 5.498 ns - 21 cycles 5.458 ns 64 - 100 cycles 25.164 ns - 20 cycles 5.242 ns - 20 cycles 5.229 ns 128 - 155 cycles 38.976 ns - 27 cycles 6.886 ns - 27 cycles 6.892 ns 158 - 132 cycles 33.034 ns - 30 cycles 7.711 ns - 30 cycles 7.728 ns 250 - 130 cycles 32.612 ns - 38 cycles 9.560 ns - 38 cycles 9.549 ns Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Vladimir Davydov <vdavydov@virtuozzo.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:54:00 +08:00
if (!orig_s) /* called via kfree_bulk */
s = virt_to_cache(objp);
else
s = cache_from_obj(orig_s, objp);
mm/slab: sanity-check page type when looking up cache This avoids any possible type confusion when looking up an object. For example, if a non-slab were to be passed to kfree(), the invalid slab_cache pointer (i.e. overlapped with some other value from the struct page union) would be used for subsequent slab manipulations that could lead to further memory corruption. Since the page is already in cache, adding the PageSlab() check will have nearly zero cost, so add a check and WARN() to virt_to_cache(). Additionally replaces an open-coded virt_to_cache(). To support the failure mode this also updates all callers of virt_to_cache() and cache_from_obj() to handle a NULL cache pointer return value (though note that several already handle this case gracefully). [dan.carpenter@oracle.com: restore IRQs in kfree()] Link: http://lkml.kernel.org/r/20190613065637.GE16334@mwanda Link: http://lkml.kernel.org/r/20190530045017.15252-3-keescook@chromium.org Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Cc: Alexander Popov <alex.popov@linux.com> Cc: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:53:26 +08:00
if (!s)
continue;
slab: implement bulk free in SLAB allocator This patch implements the free side of bulk API for the SLAB allocator kmem_cache_free_bulk(), and concludes the implementation of optimized bulk API for SLAB allocator. Benchmarked[1] cost of alloc+free (obj size 256 bytes) on CPU i7-4790K @ 4.00GHz, with no debug options, no PREEMPT and CONFIG_MEMCG_KMEM=y but no active user of kmemcg. SLAB single alloc+free cost: 87 cycles(tsc) 21.814 ns with this optimized config. bulk- Current fallback - optimized SLAB bulk 1 - 102 cycles(tsc) 25.747 ns - 41 cycles(tsc) 10.490 ns - improved 59.8% 2 - 94 cycles(tsc) 23.546 ns - 26 cycles(tsc) 6.567 ns - improved 72.3% 3 - 92 cycles(tsc) 23.127 ns - 20 cycles(tsc) 5.244 ns - improved 78.3% 4 - 90 cycles(tsc) 22.663 ns - 18 cycles(tsc) 4.588 ns - improved 80.0% 8 - 88 cycles(tsc) 22.242 ns - 14 cycles(tsc) 3.656 ns - improved 84.1% 16 - 88 cycles(tsc) 22.010 ns - 13 cycles(tsc) 3.480 ns - improved 85.2% 30 - 89 cycles(tsc) 22.305 ns - 13 cycles(tsc) 3.303 ns - improved 85.4% 32 - 89 cycles(tsc) 22.277 ns - 13 cycles(tsc) 3.309 ns - improved 85.4% 34 - 88 cycles(tsc) 22.246 ns - 13 cycles(tsc) 3.294 ns - improved 85.2% 48 - 88 cycles(tsc) 22.121 ns - 13 cycles(tsc) 3.492 ns - improved 85.2% 64 - 88 cycles(tsc) 22.052 ns - 13 cycles(tsc) 3.411 ns - improved 85.2% 128 - 89 cycles(tsc) 22.452 ns - 15 cycles(tsc) 3.841 ns - improved 83.1% 158 - 89 cycles(tsc) 22.403 ns - 14 cycles(tsc) 3.746 ns - improved 84.3% 250 - 91 cycles(tsc) 22.775 ns - 16 cycles(tsc) 4.111 ns - improved 82.4% Notice it is not recommended to do very large bulk operation with this bulk API, because local IRQs are disabled in this period. [1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/mm/slab_bulk_test01.c Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Vladimir Davydov <vdavydov@virtuozzo.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-03-16 05:53:56 +08:00
debug_check_no_locks_freed(objp, s->object_size);
if (!(s->flags & SLAB_DEBUG_OBJECTS))
debug_check_no_obj_freed(objp, s->object_size);
__cache_free(s, objp, _RET_IP_);
}
local_irq_enable();
/* FIXME: add tracing */
}
EXPORT_SYMBOL(kmem_cache_free_bulk);
/**
* kfree - free previously allocated memory
* @objp: pointer returned by kmalloc.
*
* If @objp is NULL, no operation is performed.
*
* Don't free memory not originally allocated by kmalloc()
* or you will run into trouble.
*/
void kfree(const void *objp)
{
struct kmem_cache *c;
unsigned long flags;
trace_kfree(_RET_IP_, objp);
if (unlikely(ZERO_OR_NULL_PTR(objp)))
return;
local_irq_save(flags);
kfree_debugcheck(objp);
c = virt_to_cache(objp);
mm/slab: sanity-check page type when looking up cache This avoids any possible type confusion when looking up an object. For example, if a non-slab were to be passed to kfree(), the invalid slab_cache pointer (i.e. overlapped with some other value from the struct page union) would be used for subsequent slab manipulations that could lead to further memory corruption. Since the page is already in cache, adding the PageSlab() check will have nearly zero cost, so add a check and WARN() to virt_to_cache(). Additionally replaces an open-coded virt_to_cache(). To support the failure mode this also updates all callers of virt_to_cache() and cache_from_obj() to handle a NULL cache pointer return value (though note that several already handle this case gracefully). [dan.carpenter@oracle.com: restore IRQs in kfree()] Link: http://lkml.kernel.org/r/20190613065637.GE16334@mwanda Link: http://lkml.kernel.org/r/20190530045017.15252-3-keescook@chromium.org Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Cc: Alexander Popov <alex.popov@linux.com> Cc: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:53:26 +08:00
if (!c) {
local_irq_restore(flags);
return;
}
debug_check_no_locks_freed(objp, c->object_size);
debug_check_no_obj_freed(objp, c->object_size);
__cache_free(c, (void *)objp, _RET_IP_);
local_irq_restore(flags);
}
EXPORT_SYMBOL(kfree);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
/*
* This initializes kmem_cache_node or resizes various caches for all nodes.
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
*/
static int setup_kmem_cache_nodes(struct kmem_cache *cachep, gfp_t gfp)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
{
int ret;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
int node;
struct kmem_cache_node *n;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
for_each_online_node(node) {
ret = setup_kmem_cache_node(cachep, node, gfp, true);
if (ret)
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
goto fail;
}
return 0;
fail:
if (!cachep->list.next) {
/* Cache is not active yet. Roll back what we did */
node--;
while (node >= 0) {
n = get_node(cachep, node);
if (n) {
kfree(n->shared);
free_alien_cache(n->alien);
kfree(n);
cachep->node[node] = NULL;
}
node--;
}
}
return -ENOMEM;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
}
/* Always called with the slab_mutex held */
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
static int __do_tune_cpucache(struct kmem_cache *cachep, int limit,
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
int batchcount, int shared, gfp_t gfp)
{
struct array_cache __percpu *cpu_cache, *prev;
int cpu;
cpu_cache = alloc_kmem_cache_cpus(cachep, limit, batchcount);
if (!cpu_cache)
return -ENOMEM;
prev = cachep->cpu_cache;
cachep->cpu_cache = cpu_cache;
slab: avoid IPIs when creating kmem caches Each slab kmem cache has per cpu array caches. The array caches are created when the kmem_cache is created, either via kmem_cache_create() or lazily when the first object is allocated in context of a kmem enabled memcg. Array caches are replaced by writing to /proc/slabinfo. Array caches are protected by holding slab_mutex or disabling interrupts. Array cache allocation and replacement is done by __do_tune_cpucache() which holds slab_mutex and calls kick_all_cpus_sync() to interrupt all remote processors which confirms there are no references to the old array caches. IPIs are needed when replacing array caches. But when creating a new array cache, there's no need to send IPIs because there cannot be any references to the new cache. Outside of memcg kmem accounting these IPIs occur at boot time, so they're not a problem. But with memcg kmem accounting each container can create kmem caches, so the IPIs are wasteful. Avoid unnecessary IPIs when creating array caches. Test which reports the IPI count of allocating slab in 10000 memcg: import os def ipi_count(): with open("/proc/interrupts") as f: for l in f: if 'Function call interrupts' in l: return int(l.split()[1]) def echo(val, path): with open(path, "w") as f: f.write(val) n = 10000 os.chdir("/mnt/cgroup/memory") pid = str(os.getpid()) a = ipi_count() for i in range(n): os.mkdir(str(i)) echo("1G\n", "%d/memory.limit_in_bytes" % i) echo("1G\n", "%d/memory.kmem.limit_in_bytes" % i) echo(pid, "%d/cgroup.procs" % i) open("/tmp/x", "w").close() os.unlink("/tmp/x") b = ipi_count() print "%d loops: %d => %d (+%d ipis)" % (n, a, b, b-a) echo(pid, "cgroup.procs") for i in range(n): os.rmdir(str(i)) patched: 10000 loops: 1069 => 1170 (+101 ipis) unpatched: 10000 loops: 1192 => 48933 (+47741 ipis) Link: http://lkml.kernel.org/r/20170416214544.109476-1-gthelen@google.com Signed-off-by: Greg Thelen <gthelen@google.com> Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Acked-by: David Rientjes <rientjes@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-05-04 05:51:47 +08:00
/*
* Without a previous cpu_cache there's no need to synchronize remote
* cpus, so skip the IPIs.
*/
if (prev)
kick_all_cpus_sync();
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
check_irq_on();
cachep->batchcount = batchcount;
cachep->limit = limit;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
cachep->shared = shared;
if (!prev)
goto setup_node;
for_each_online_cpu(cpu) {
LIST_HEAD(list);
int node;
struct kmem_cache_node *n;
struct array_cache *ac = per_cpu_ptr(prev, cpu);
node = cpu_to_mem(cpu);
n = get_node(cachep, node);
spin_lock_irq(&n->list_lock);
free_block(cachep, ac->entry, ac->avail, node, &list);
spin_unlock_irq(&n->list_lock);
slabs_destroy(cachep, &list);
}
free_percpu(prev);
setup_node:
return setup_kmem_cache_nodes(cachep, gfp);
}
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
static int do_tune_cpucache(struct kmem_cache *cachep, int limit,
int batchcount, int shared, gfp_t gfp)
{
int ret;
struct kmem_cache *c;
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
ret = __do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
if (slab_state < FULL)
return ret;
if ((ret < 0) || !is_root_cache(cachep))
return ret;
lockdep_assert_held(&slab_mutex);
for_each_memcg_cache(c, cachep) {
/* return value determined by the root cache only */
__do_tune_cpucache(c, limit, batchcount, shared, gfp);
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
}
return ret;
}
/* Called with slab_mutex held always */
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
static int enable_cpucache(struct kmem_cache *cachep, gfp_t gfp)
{
int err;
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
int limit = 0;
int shared = 0;
int batchcount = 0;
mm: reorganize SLAB freelist randomization The kernel heap allocators are using a sequential freelist making their allocation predictable. This predictability makes kernel heap overflow easier to exploit. An attacker can careful prepare the kernel heap to control the following chunk overflowed. For example these attacks exploit the predictability of the heap: - Linux Kernel CAN SLUB overflow (https://goo.gl/oMNWkU) - Exploiting Linux Kernel Heap corruptions (http://goo.gl/EXLn95) ***Problems that needed solving: - Randomize the Freelist (singled linked) used in the SLUB allocator. - Ensure good performance to encourage usage. - Get best entropy in early boot stage. ***Parts: - 01/02 Reorganize the SLAB Freelist randomization to share elements with the SLUB implementation. - 02/02 The SLUB Freelist randomization implementation. Similar approach than the SLAB but tailored to the singled freelist used in SLUB. ***Performance data: slab_test impact is between 3% to 4% on average for 100000 attempts without smp. It is a very focused testing, kernbench show the overall impact on the system is way lower. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 49 cycles kfree -> 77 cycles 100000 times kmalloc(16) -> 51 cycles kfree -> 79 cycles 100000 times kmalloc(32) -> 53 cycles kfree -> 83 cycles 100000 times kmalloc(64) -> 62 cycles kfree -> 90 cycles 100000 times kmalloc(128) -> 81 cycles kfree -> 97 cycles 100000 times kmalloc(256) -> 98 cycles kfree -> 121 cycles 100000 times kmalloc(512) -> 95 cycles kfree -> 122 cycles 100000 times kmalloc(1024) -> 96 cycles kfree -> 126 cycles 100000 times kmalloc(2048) -> 115 cycles kfree -> 140 cycles 100000 times kmalloc(4096) -> 149 cycles kfree -> 171 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 70 cycles 100000 times kmalloc(16)/kfree -> 70 cycles 100000 times kmalloc(32)/kfree -> 70 cycles 100000 times kmalloc(64)/kfree -> 70 cycles 100000 times kmalloc(128)/kfree -> 70 cycles 100000 times kmalloc(256)/kfree -> 69 cycles 100000 times kmalloc(512)/kfree -> 70 cycles 100000 times kmalloc(1024)/kfree -> 73 cycles 100000 times kmalloc(2048)/kfree -> 72 cycles 100000 times kmalloc(4096)/kfree -> 71 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 100000 times kmalloc(8) -> 57 cycles kfree -> 78 cycles 100000 times kmalloc(16) -> 61 cycles kfree -> 81 cycles 100000 times kmalloc(32) -> 76 cycles kfree -> 93 cycles 100000 times kmalloc(64) -> 83 cycles kfree -> 94 cycles 100000 times kmalloc(128) -> 106 cycles kfree -> 107 cycles 100000 times kmalloc(256) -> 118 cycles kfree -> 117 cycles 100000 times kmalloc(512) -> 114 cycles kfree -> 116 cycles 100000 times kmalloc(1024) -> 115 cycles kfree -> 118 cycles 100000 times kmalloc(2048) -> 147 cycles kfree -> 131 cycles 100000 times kmalloc(4096) -> 214 cycles kfree -> 161 cycles 2. Kmalloc: alloc/free test 100000 times kmalloc(8)/kfree -> 66 cycles 100000 times kmalloc(16)/kfree -> 66 cycles 100000 times kmalloc(32)/kfree -> 66 cycles 100000 times kmalloc(64)/kfree -> 66 cycles 100000 times kmalloc(128)/kfree -> 65 cycles 100000 times kmalloc(256)/kfree -> 67 cycles 100000 times kmalloc(512)/kfree -> 67 cycles 100000 times kmalloc(1024)/kfree -> 64 cycles 100000 times kmalloc(2048)/kfree -> 67 cycles 100000 times kmalloc(4096)/kfree -> 67 cycles Kernbench, before: Average Optimal load -j 12 Run (std deviation): Elapsed Time 101.873 (1.16069) User Time 1045.22 (1.60447) System Time 88.969 (0.559195) Percent CPU 1112.9 (13.8279) Context Switches 189140 (2282.15) Sleeps 99008.6 (768.091) After: Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.47 (0.562732) User Time 1045.3 (1.34263) System Time 88.311 (0.342554) Percent CPU 1105.8 (6.49444) Context Switches 189081 (2355.78) Sleeps 99231.5 (800.358) This patch (of 2): This commit reorganizes the previous SLAB freelist randomization to prepare for the SLUB implementation. It moves functions that will be shared to slab_common. The entropy functions are changed to align with the SLUB implementation, now using get_random_(int|long) functions. These functions were chosen because they provide a bit more entropy early on boot and better performance when specific arch instructions are not available. [akpm@linux-foundation.org: fix build] Link: http://lkml.kernel.org/r/1464295031-26375-2-git-send-email-thgarnie@google.com Signed-off-by: Thomas Garnier <thgarnie@google.com> Reviewed-by: Kees Cook <keescook@chromium.org> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:21:56 +08:00
err = cache_random_seq_create(cachep, cachep->num, gfp);
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
if (err)
goto end;
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
if (!is_root_cache(cachep)) {
struct kmem_cache *root = memcg_root_cache(cachep);
limit = root->limit;
shared = root->shared;
batchcount = root->batchcount;
}
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
if (limit && shared && batchcount)
goto skip_setup;
/*
* The head array serves three purposes:
* - create a LIFO ordering, i.e. return objects that are cache-warm
* - reduce the number of spinlock operations.
* - reduce the number of linked list operations on the slab and
* bufctl chains: array operations are cheaper.
* The numbers are guessed, we should auto-tune as described by
* Bonwick.
*/
if (cachep->size > 131072)
limit = 1;
else if (cachep->size > PAGE_SIZE)
limit = 8;
else if (cachep->size > 1024)
limit = 24;
else if (cachep->size > 256)
limit = 54;
else
limit = 120;
/*
* CPU bound tasks (e.g. network routing) can exhibit cpu bound
* allocation behaviour: Most allocs on one cpu, most free operations
* on another cpu. For these cases, an efficient object passing between
* cpus is necessary. This is provided by a shared array. The array
* replaces Bonwick's magazine layer.
* On uniprocessor, it's functionally equivalent (but less efficient)
* to a larger limit. Thus disabled by default.
*/
shared = 0;
if (cachep->size <= PAGE_SIZE && num_possible_cpus() > 1)
shared = 8;
#if DEBUG
/*
* With debugging enabled, large batchcount lead to excessively long
* periods with disabled local interrupts. Limit the batchcount
*/
if (limit > 32)
limit = 32;
#endif
slab: propagate tunable values SLAB allows us to tune a particular cache behavior with tunables. When creating a new memcg cache copy, we'd like to preserve any tunables the parent cache already had. This could be done by an explicit call to do_tune_cpucache() after the cache is created. But this is not very convenient now that the caches are created from common code, since this function is SLAB-specific. Another method of doing that is taking advantage of the fact that do_tune_cpucache() is always called from enable_cpucache(), which is called at cache initialization. We can just preset the values, and then things work as expected. It can also happen that a root cache has its tunables updated during normal system operation. In this case, we will propagate the change to all caches that are already active. This change will require us to move the assignment of root_cache in memcg_params a bit earlier. We need this to be already set - which memcg_kmem_register_cache will do - when we reach __kmem_cache_create() Signed-off-by: Glauber Costa <glommer@parallels.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Frederic Weisbecker <fweisbec@redhat.com> Cc: Greg Thelen <gthelen@google.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: JoonSoo Kim <js1304@gmail.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Michal Hocko <mhocko@suse.cz> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Rik van Riel <riel@redhat.com> Cc: Suleiman Souhlal <suleiman@google.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-19 06:23:03 +08:00
batchcount = (limit + 1) / 2;
skip_setup:
err = do_tune_cpucache(cachep, limit, batchcount, shared, gfp);
mm: SLAB freelist randomization Provides an optional config (CONFIG_SLAB_FREELIST_RANDOM) to randomize the SLAB freelist. The list is randomized during initialization of a new set of pages. The order on different freelist sizes is pre-computed at boot for performance. Each kmem_cache has its own randomized freelist. Before pre-computed lists are available freelists are generated dynamically. This security feature reduces the predictability of the kernel SLAB allocator against heap overflows rendering attacks much less stable. For example this attack against SLUB (also applicable against SLAB) would be affected: https://jon.oberheide.org/blog/2010/09/10/linux-kernel-can-slub-overflow/ Also, since v4.6 the freelist was moved at the end of the SLAB. It means a controllable heap is opened to new attacks not yet publicly discussed. A kernel heap overflow can be transformed to multiple use-after-free. This feature makes this type of attack harder too. To generate entropy, we use get_random_bytes_arch because 0 bits of entropy is available in the boot stage. In the worse case this function will fallback to the get_random_bytes sub API. We also generate a shift random number to shift pre-computed freelist for each new set of pages. The config option name is not specific to the SLAB as this approach will be extended to other allocators like SLUB. Performance results highlighted no major changes: Hackbench (running 90 10 times): Before average: 0.0698 After average: 0.0663 (-5.01%) slab_test 1 run on boot. Difference only seen on the 2048 size test being the worse case scenario covered by freelist randomization. New slab pages are constantly being created on the 10000 allocations. Variance should be mainly due to getting new pages every few allocations. Before: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 99 cycles kfree -> 112 cycles 10000 times kmalloc(16) -> 109 cycles kfree -> 140 cycles 10000 times kmalloc(32) -> 129 cycles kfree -> 137 cycles 10000 times kmalloc(64) -> 141 cycles kfree -> 141 cycles 10000 times kmalloc(128) -> 152 cycles kfree -> 148 cycles 10000 times kmalloc(256) -> 195 cycles kfree -> 167 cycles 10000 times kmalloc(512) -> 257 cycles kfree -> 199 cycles 10000 times kmalloc(1024) -> 393 cycles kfree -> 251 cycles 10000 times kmalloc(2048) -> 649 cycles kfree -> 228 cycles 10000 times kmalloc(4096) -> 806 cycles kfree -> 370 cycles 10000 times kmalloc(8192) -> 814 cycles kfree -> 411 cycles 10000 times kmalloc(16384) -> 892 cycles kfree -> 455 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 121 cycles 10000 times kmalloc(64)/kfree -> 121 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 121 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles After: Single thread testing ===================== 1. Kmalloc: Repeatedly allocate then free test 10000 times kmalloc(8) -> 130 cycles kfree -> 86 cycles 10000 times kmalloc(16) -> 118 cycles kfree -> 86 cycles 10000 times kmalloc(32) -> 121 cycles kfree -> 85 cycles 10000 times kmalloc(64) -> 176 cycles kfree -> 102 cycles 10000 times kmalloc(128) -> 178 cycles kfree -> 100 cycles 10000 times kmalloc(256) -> 205 cycles kfree -> 109 cycles 10000 times kmalloc(512) -> 262 cycles kfree -> 136 cycles 10000 times kmalloc(1024) -> 342 cycles kfree -> 157 cycles 10000 times kmalloc(2048) -> 701 cycles kfree -> 238 cycles 10000 times kmalloc(4096) -> 803 cycles kfree -> 364 cycles 10000 times kmalloc(8192) -> 835 cycles kfree -> 404 cycles 10000 times kmalloc(16384) -> 896 cycles kfree -> 441 cycles 2. Kmalloc: alloc/free test 10000 times kmalloc(8)/kfree -> 121 cycles 10000 times kmalloc(16)/kfree -> 121 cycles 10000 times kmalloc(32)/kfree -> 123 cycles 10000 times kmalloc(64)/kfree -> 142 cycles 10000 times kmalloc(128)/kfree -> 121 cycles 10000 times kmalloc(256)/kfree -> 119 cycles 10000 times kmalloc(512)/kfree -> 119 cycles 10000 times kmalloc(1024)/kfree -> 119 cycles 10000 times kmalloc(2048)/kfree -> 119 cycles 10000 times kmalloc(4096)/kfree -> 119 cycles 10000 times kmalloc(8192)/kfree -> 119 cycles 10000 times kmalloc(16384)/kfree -> 119 cycles [akpm@linux-foundation.org: propagate gfp_t into cache_random_seq_create()] Signed-off-by: Thomas Garnier <thgarnie@google.com> Acked-by: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Kees Cook <keescook@chromium.org> Cc: Greg Thelen <gthelen@google.com> Cc: Laura Abbott <labbott@fedoraproject.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:37 +08:00
end:
if (err)
pr_err("enable_cpucache failed for %s, error %d\n",
cachep->name, -err);
return err;
}
/*
* Drain an array if it contains any elements taking the node lock only if
* necessary. Note that the node listlock also protects the array_cache
* if drain_array() is used on the shared array.
*/
static void drain_array(struct kmem_cache *cachep, struct kmem_cache_node *n,
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
struct array_cache *ac, int node)
{
LIST_HEAD(list);
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
/* ac from n->shared can be freed if we don't hold the slab_mutex. */
check_mutex_acquired();
if (!ac || !ac->avail)
return;
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
if (ac->touched) {
ac->touched = 0;
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
return;
}
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
spin_lock_irq(&n->list_lock);
drain_array_locked(cachep, ac, node, false, &list);
spin_unlock_irq(&n->list_lock);
slabs_destroy(cachep, &list);
}
/**
* cache_reap - Reclaim memory from caches.
* @w: work descriptor
*
* Called from workqueue/eventd every few seconds.
* Purpose:
* - clear the per-cpu caches for this CPU.
* - return freeable pages to the main free memory pool.
*
* If we cannot acquire the cache chain mutex then just give up - we'll try
* again on the next iteration.
*/
static void cache_reap(struct work_struct *w)
{
struct kmem_cache *searchp;
struct kmem_cache_node *n;
numa: slab: use numa_mem_id() for slab local memory node Example usage of generic "numa_mem_id()": The mainline slab code, since ~ 2.6.19, does not handle memoryless nodes well. Specifically, the "fast path"--____cache_alloc()--will never succeed as slab doesn't cache offnode object on the per cpu queues, and for memoryless nodes, all memory will be "off node" relative to numa_node_id(). This adds significant overhead to all kmem cache allocations, incurring a significant regression relative to earlier kernels [from before slab.c was reorganized]. This patch uses the generic topology function "numa_mem_id()" to return the "effective local memory node" for the calling context. This is the first node in the local node's generic fallback zonelist-- the same node that "local" mempolicy-based allocations would use. This lets slab cache these "local" allocations and avoid fallback/refill on every allocation. N.B.: Slab will need to handle node and memory hotplug events that could change the value returned by numa_mem_id() for any given node if recent changes to address memory hotplug don't already address this. E.g., flush all per cpu slab queues before rebuilding the zonelists while the "machine" is held in the stopped state. Performance impact on "hackbench 400 process 200" 2.6.34-rc3-mmotm-100405-1609 no-patch this-patch ia64 no memoryless nodes [avg of 10]: 11.713 11.637 ~0.65 diff ia64 cpus all on memless nodes [10]: 228.259 26.484 ~8.6x speedup The slowdown of the patched kernel from ~12 sec to ~28 seconds when configured with memoryless nodes is the result of all cpus allocating from a single node's mm pagepool. The cache lines of the single node are distributed/interleaved over the memory of the real physical nodes, but the zone lock, list heads, ... of the single node with memory still each live in a single cache line that is accessed from all processors. x86_64 [8x6 AMD] [avg of 40]: 2.883 2.845 Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Tejun Heo <tj@kernel.org> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Nick Piggin <npiggin@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Eric Whitney <eric.whitney@hp.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-27 05:45:03 +08:00
int node = numa_mem_id();
struct delayed_work *work = to_delayed_work(w);
if (!mutex_trylock(&slab_mutex))
/* Give up. Setup the next iteration. */
goto out;
list_for_each_entry(searchp, &slab_caches, list) {
check_irq_on();
/*
* We only take the node lock if absolutely necessary and we
* have established with reasonable certainty that
* we can do some work if the lock was obtained.
*/
n = get_node(searchp, node);
reap_alien(searchp, n);
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
drain_array(searchp, n, cpu_cache_get(searchp), node);
/*
* These are racy checks but it does not matter
* if we skip one check or scan twice.
*/
if (time_after(n->next_reap, jiffies))
goto next;
n->next_reap = jiffies + REAPTIMEOUT_NODE;
mm/slab: fix the theoretical race by holding proper lock While processing concurrent allocation, SLAB could be contended a lot because it did a lots of work with holding a lock. This patchset try to reduce the number of critical section to reduce lock contention. Major changes are lockless decision to allocate more slab and lockless cpu cache refill from the newly allocated slab. Below is the result of concurrent allocation/free in slab allocation benchmark made by Christoph a long time ago. I make the output simpler. The number shows cycle count during alloc/free respectively so less is better. * Before Kmalloc N*alloc N*free(32): Average=365/806 Kmalloc N*alloc N*free(64): Average=452/690 Kmalloc N*alloc N*free(128): Average=736/886 Kmalloc N*alloc N*free(256): Average=1167/985 Kmalloc N*alloc N*free(512): Average=2088/1125 Kmalloc N*alloc N*free(1024): Average=4115/1184 Kmalloc N*alloc N*free(2048): Average=8451/1748 Kmalloc N*alloc N*free(4096): Average=16024/2048 * After Kmalloc N*alloc N*free(32): Average=344/792 Kmalloc N*alloc N*free(64): Average=347/882 Kmalloc N*alloc N*free(128): Average=390/959 Kmalloc N*alloc N*free(256): Average=393/1067 Kmalloc N*alloc N*free(512): Average=683/1229 Kmalloc N*alloc N*free(1024): Average=1295/1325 Kmalloc N*alloc N*free(2048): Average=2513/1664 Kmalloc N*alloc N*free(4096): Average=4742/2172 It shows that performance improves greatly (roughly more than 50%) for the object class whose size is more than 128 bytes. This patch (of 11): If we don't hold neither the slab_mutex nor the node lock, node's shared array cache could be freed and re-populated. If __kmem_cache_shrink() is called at the same time, it will call drain_array() with n->shared without holding node lock so problem can happen. This patch fix the situation by holding the node lock before trying to drain the shared array. In addition, add a debug check to confirm that n->shared access race doesn't exist. Signed-off-by: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-05-20 08:10:02 +08:00
drain_array(searchp, n, n->shared, node);
if (n->free_touched)
n->free_touched = 0;
else {
int freed;
freed = drain_freelist(searchp, n, (n->free_limit +
5 * searchp->num - 1) / (5 * searchp->num));
STATS_ADD_REAPED(searchp, freed);
}
next:
cond_resched();
}
check_irq_on();
mutex_unlock(&slab_mutex);
[PATCH] slab: Node rotor for freeing alien caches and remote per cpu pages. The cache reaper currently tries to free all alien caches and all remote per cpu pages in each pass of cache_reap. For a machines with large number of nodes (such as Altix) this may lead to sporadic delays of around ~10ms. Interrupts are disabled while reclaiming creating unacceptable delays. This patch changes that behavior by adding a per cpu reap_node variable. Instead of attempting to free all caches, we free only one alien cache and the per cpu pages from one remote node. That reduces the time spend in cache_reap. However, doing so will lengthen the time it takes to completely drain all remote per cpu pagesets and all alien caches. The time needed will grow with the number of nodes in the system. All caches are drained when they overflow their respective capacity. So the drawback here is only that a bit of memory may be wasted for awhile longer. Details: 1. Rename drain_remote_pages to drain_node_pages to allow the specification of the node to drain of pcp pages. 2. Add additional functions init_reap_node, next_reap_node for NUMA that manage a per cpu reap_node counter. 3. Add a reap_alien function that reaps only from the current reap_node. For us this seems to be a critical issue. Holdoffs of an average of ~7ms cause some HPC benchmarks to slow down significantly. F.e. NAS parallel slows down dramatically. NAS parallel has a 12-16 seconds runtime w/o rotor compared to 5.8 secs with the rotor patches. It gets down to 5.05 secs with the additional interrupt holdoff reductions. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-10 09:33:54 +08:00
next_reap_node();
out:
/* Set up the next iteration */
mm, slab: reschedule cache_reap() on the same CPU cache_reap() is initially scheduled in start_cpu_timer() via schedule_delayed_work_on(). But then the next iterations are scheduled via schedule_delayed_work(), i.e. using WORK_CPU_UNBOUND. Thus since commit ef557180447f ("workqueue: schedule WORK_CPU_UNBOUND work on wq_unbound_cpumask CPUs") there is no guarantee the future iterations will run on the originally intended cpu, although it's still preferred. I was able to demonstrate this with /sys/module/workqueue/parameters/debug_force_rr_cpu. IIUC, it may also happen due to migrating timers in nohz context. As a result, some cpu's would be calling cache_reap() more frequently and others never. This patch uses schedule_delayed_work_on() with the current cpu when scheduling the next iteration. Link: http://lkml.kernel.org/r/20180411070007.32225-1-vbabka@suse.cz Fixes: ef557180447f ("workqueue: schedule WORK_CPU_UNBOUND work on wq_unbound_cpumask CPUs") Signed-off-by: Vlastimil Babka <vbabka@suse.cz> Acked-by: Pekka Enberg <penberg@kernel.org> Acked-by: Christoph Lameter <cl@linux.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: David Rientjes <rientjes@google.com> Cc: Tejun Heo <tj@kernel.org> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: John Stultz <john.stultz@linaro.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Stephen Boyd <sboyd@kernel.org> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-14 06:35:38 +08:00
schedule_delayed_work_on(smp_processor_id(), work,
round_jiffies_relative(REAPTIMEOUT_AC));
}
void get_slabinfo(struct kmem_cache *cachep, struct slabinfo *sinfo)
{
unsigned long active_objs, num_objs, active_slabs;
unsigned long total_slabs = 0, free_objs = 0, shared_avail = 0;
unsigned long free_slabs = 0;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
int node;
struct kmem_cache_node *n;
for_each_kmem_cache_node(cachep, node, n) {
check_irq_on();
spin_lock_irq(&n->list_lock);
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
total_slabs += n->total_slabs;
free_slabs += n->free_slabs;
free_objs += n->free_objects;
mm/slab: improve performance of gathering slabinfo stats On large systems, when some slab caches grow to millions of objects (and many gigabytes), running 'cat /proc/slabinfo' can take up to 1-2 seconds. During this time, interrupts are disabled while walking the slab lists (slabs_full, slabs_partial, and slabs_free) for each node, and this sometimes causes timeouts in other drivers (for instance, Infiniband). This patch optimizes 'cat /proc/slabinfo' by maintaining a counter for total number of allocated slabs per node, per cache. This counter is updated when a slab is created or destroyed. This enables us to skip traversing the slabs_full list while gathering slabinfo statistics, and since slabs_full tends to be the biggest list when the cache is large, it results in a dramatic performance improvement. Getting slabinfo statistics now only requires walking the slabs_free and slabs_partial lists, and those lists are usually much smaller than slabs_full. We tested this after growing the dentry cache to 70GB, and the performance improved from 2s to 5ms. Link: http://lkml.kernel.org/r/1472517876-26814-1-git-send-email-aruna.ramakrishna@oracle.com Signed-off-by: Aruna Ramakrishna <aruna.ramakrishna@oracle.com> Acked-by: David Rientjes <rientjes@google.com> Cc: Mike Kravetz <mike.kravetz@oracle.com> Cc: Christoph Lameter <cl@linux.com> Cc: Pekka Enberg <penberg@kernel.org> Cc: David Rientjes <rientjes@google.com> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-28 08:46:32 +08:00
if (n->shared)
shared_avail += n->shared->avail;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
spin_unlock_irq(&n->list_lock);
}
num_objs = total_slabs * cachep->num;
active_slabs = total_slabs - free_slabs;
active_objs = num_objs - free_objs;
sinfo->active_objs = active_objs;
sinfo->num_objs = num_objs;
sinfo->active_slabs = active_slabs;
sinfo->num_slabs = total_slabs;
sinfo->shared_avail = shared_avail;
sinfo->limit = cachep->limit;
sinfo->batchcount = cachep->batchcount;
sinfo->shared = cachep->shared;
sinfo->objects_per_slab = cachep->num;
sinfo->cache_order = cachep->gfporder;
}
void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *cachep)
{
#if STATS
{ /* node stats */
unsigned long high = cachep->high_mark;
unsigned long allocs = cachep->num_allocations;
unsigned long grown = cachep->grown;
unsigned long reaped = cachep->reaped;
unsigned long errors = cachep->errors;
unsigned long max_freeable = cachep->max_freeable;
unsigned long node_allocs = cachep->node_allocs;
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
unsigned long node_frees = cachep->node_frees;
unsigned long overflows = cachep->node_overflow;
seq_printf(m, " : globalstat %7lu %6lu %5lu %4lu %4lu %4lu %4lu %4lu %4lu",
allocs, high, grown,
reaped, errors, max_freeable, node_allocs,
node_frees, overflows);
}
/* cpu stats */
{
unsigned long allochit = atomic_read(&cachep->allochit);
unsigned long allocmiss = atomic_read(&cachep->allocmiss);
unsigned long freehit = atomic_read(&cachep->freehit);
unsigned long freemiss = atomic_read(&cachep->freemiss);
seq_printf(m, " : cpustat %6lu %6lu %6lu %6lu",
allochit, allocmiss, freehit, freemiss);
}
#endif
}
#define MAX_SLABINFO_WRITE 128
/**
* slabinfo_write - Tuning for the slab allocator
* @file: unused
* @buffer: user buffer
* @count: data length
* @ppos: unused
*
* Return: %0 on success, negative error code otherwise.
*/
ssize_t slabinfo_write(struct file *file, const char __user *buffer,
size_t count, loff_t *ppos)
{
char kbuf[MAX_SLABINFO_WRITE + 1], *tmp;
int limit, batchcount, shared, res;
struct kmem_cache *cachep;
if (count > MAX_SLABINFO_WRITE)
return -EINVAL;
if (copy_from_user(&kbuf, buffer, count))
return -EFAULT;
kbuf[MAX_SLABINFO_WRITE] = '\0';
tmp = strchr(kbuf, ' ');
if (!tmp)
return -EINVAL;
*tmp = '\0';
tmp++;
if (sscanf(tmp, " %d %d %d", &limit, &batchcount, &shared) != 3)
return -EINVAL;
/* Find the cache in the chain of caches. */
mutex_lock(&slab_mutex);
res = -EINVAL;
list_for_each_entry(cachep, &slab_caches, list) {
if (!strcmp(cachep->name, kbuf)) {
if (limit < 1 || batchcount < 1 ||
batchcount > limit || shared < 0) {
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
res = 0;
} else {
[PATCH] Numa-aware slab allocator V5 The NUMA API change that introduced kmalloc_node was accepted for 2.6.12-rc3. Now it is possible to do slab allocations on a node to localize memory structures. This API was used by the pageset localization patch and the block layer localization patch now in mm. The existing kmalloc_node is slow since it simply searches through all pages of the slab to find a page that is on the node requested. The two patches do a one time allocation of slab structures at initialization and therefore the speed of kmalloc node does not matter. This patch allows kmalloc_node to be as fast as kmalloc by introducing node specific page lists for partial, free and full slabs. Slab allocation improves in a NUMA system so that we are seeing a performance gain in AIM7 of about 5% with this patch alone. More NUMA localizations are possible if kmalloc_node operates in an fast way like kmalloc. Test run on a 32p systems with 32G Ram. w/o patch Tasks jobs/min jti jobs/min/task real cpu 1 485.36 100 485.3640 11.99 1.91 Sat Apr 30 14:01:51 2005 100 26582.63 88 265.8263 21.89 144.96 Sat Apr 30 14:02:14 2005 200 29866.83 81 149.3342 38.97 286.08 Sat Apr 30 14:02:53 2005 300 33127.16 78 110.4239 52.71 426.54 Sat Apr 30 14:03:46 2005 400 34889.47 80 87.2237 66.72 568.90 Sat Apr 30 14:04:53 2005 500 35654.34 76 71.3087 81.62 714.55 Sat Apr 30 14:06:15 2005 600 36460.83 75 60.7681 95.77 853.42 Sat Apr 30 14:07:51 2005 700 35957.00 75 51.3671 113.30 990.67 Sat Apr 30 14:09:45 2005 800 33380.65 73 41.7258 139.48 1140.86 Sat Apr 30 14:12:05 2005 900 35095.01 76 38.9945 149.25 1281.30 Sat Apr 30 14:14:35 2005 1000 36094.37 74 36.0944 161.24 1419.66 Sat Apr 30 14:17:17 2005 w/patch Tasks jobs/min jti jobs/min/task real cpu 1 484.27 100 484.2736 12.02 1.93 Sat Apr 30 15:59:45 2005 100 28262.03 90 282.6203 20.59 143.57 Sat Apr 30 16:00:06 2005 200 32246.45 82 161.2322 36.10 282.89 Sat Apr 30 16:00:42 2005 300 37945.80 83 126.4860 46.01 418.75 Sat Apr 30 16:01:28 2005 400 40000.69 81 100.0017 58.20 561.48 Sat Apr 30 16:02:27 2005 500 40976.10 78 81.9522 71.02 696.95 Sat Apr 30 16:03:38 2005 600 41121.54 78 68.5359 84.92 834.86 Sat Apr 30 16:05:04 2005 700 44052.77 78 62.9325 92.48 971.53 Sat Apr 30 16:06:37 2005 800 41066.89 79 51.3336 113.38 1111.15 Sat Apr 30 16:08:31 2005 900 38918.77 79 43.2431 134.59 1252.57 Sat Apr 30 16:10:46 2005 1000 41842.21 76 41.8422 139.09 1392.33 Sat Apr 30 16:13:05 2005 These are measurement taken directly after boot and show a greater improvement than 5%. However, the performance improvements become less over time if the AIM7 runs are repeated and settle down at around 5%. Links to earlier discussions: http://marc.theaimsgroup.com/?t=111094594500003&r=1&w=2 http://marc.theaimsgroup.com/?t=111603406600002&r=1&w=2 Changelog V4-V5: - alloc_arraycache and alloc_aliencache take node parameter instead of cpu - fix initialization so that nodes without cpus are properly handled. - simplify code in kmem_cache_init - patch against Andrews temp mm3 release - Add Shai to credits - fallback to __cache_alloc from __cache_alloc_node if the node's cache is not available yet. Changelog V3-V4: - Patch against 2.6.12-rc5-mm1 - Cleanup patch integrated - More and better use of for_each_node and for_each_cpu - GCC 2.95 fix (do not use [] use [0]) - Correct determination of INDEX_AC - Remove hack to cause an error on platforms that have no CONFIG_NUMA but nodes. - Remove list3_data and list3_data_ptr macros for better readability Changelog V2-V3: - Made to patch against 2.6.12-rc4-mm1 - Revised bootstrap mechanism so that larger size kmem_list3 structs can be supported. Do a generic solution so that the right slab can be found for the internal structs. - use for_each_online_node Changelog V1-V2: - Batching for freeing of wrong-node objects (alien caches) - Locking changes and NUMA #ifdefs as requested by Manfred Signed-off-by: Alok N Kataria <alokk@calsoftinc.com> Signed-off-by: Shobhit Dayal <shobhit@calsoftinc.com> Signed-off-by: Shai Fultheim <Shai@Scalex86.org> Signed-off-by: Christoph Lameter <clameter@sgi.com> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-10 04:03:32 +08:00
res = do_tune_cpucache(cachep, limit,
slab: setup allocators earlier in the boot sequence This patch makes kmalloc() available earlier in the boot sequence so we can get rid of some bootmem allocations. The bulk of the changes are due to kmem_cache_init() being called with interrupts disabled which requires some changes to allocator boostrap code. Note: 32-bit x86 does WP protect test in mem_init() so we must setup traps before we call mem_init() during boot as reported by Ingo Molnar: We have a hard crash in the WP-protect code: [ 0.000000] Checking if this processor honours the WP bit even in supervisor mode...BUG: Int 14: CR2 ffcff000 [ 0.000000] EDI 00000188 ESI 00000ac7 EBP c17eaf9c ESP c17eaf8c [ 0.000000] EBX 000014e0 EDX 0000000e ECX 01856067 EAX 00000001 [ 0.000000] err 00000003 EIP c10135b1 CS 00000060 flg 00010002 [ 0.000000] Stack: c17eafa8 c17fd410 c16747bc c17eafc4 c17fd7e5 000011fd f8616000 c18237cc [ 0.000000] 00099800 c17bb000 c17eafec c17f1668 000001c5 c17f1322 c166e039 c1822bf0 [ 0.000000] c166e033 c153a014 c18237cc 00020800 c17eaff8 c17f106a 00020800 01ba5003 [ 0.000000] Pid: 0, comm: swapper Not tainted 2.6.30-tip-02161-g7a74539-dirty #52203 [ 0.000000] Call Trace: [ 0.000000] [<c15357c2>] ? printk+0x14/0x16 [ 0.000000] [<c10135b1>] ? do_test_wp_bit+0x19/0x23 [ 0.000000] [<c17fd410>] ? test_wp_bit+0x26/0x64 [ 0.000000] [<c17fd7e5>] ? mem_init+0x1ba/0x1d8 [ 0.000000] [<c17f1668>] ? start_kernel+0x164/0x2f7 [ 0.000000] [<c17f1322>] ? unknown_bootoption+0x0/0x19c [ 0.000000] [<c17f106a>] ? __init_begin+0x6a/0x6f Acked-by: Johannes Weiner <hannes@cmpxchg.org> Acked-by Linus Torvalds <torvalds@linux-foundation.org> Cc: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Matt Mackall <mpm@selenic.com> Cc: Nick Piggin <npiggin@suse.de> Cc: Yinghai Lu <yinghai@kernel.org> Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
2009-06-11 00:40:04 +08:00
batchcount, shared,
GFP_KERNEL);
}
break;
}
}
mutex_unlock(&slab_mutex);
if (res >= 0)
res = count;
return res;
}
[PATCH] slab: implement /proc/slab_allocators Implement /proc/slab_allocators. It produces output like: idr_layer_cache: 80 idr_pre_get+0x33/0x4e buffer_head: 2555 alloc_buffer_head+0x20/0x75 mm_struct: 9 mm_alloc+0x1e/0x42 mm_struct: 20 dup_mm+0x36/0x370 vm_area_struct: 384 dup_mm+0x18f/0x370 vm_area_struct: 151 do_mmap_pgoff+0x2e0/0x7c3 vm_area_struct: 1 split_vma+0x5a/0x10e vm_area_struct: 11 do_brk+0x206/0x2e2 vm_area_struct: 2 copy_vma+0xda/0x142 vm_area_struct: 9 setup_arg_pages+0x99/0x214 fs_cache: 8 copy_fs_struct+0x21/0x133 fs_cache: 29 copy_process+0xf38/0x10e3 files_cache: 30 alloc_files+0x1b/0xcf signal_cache: 81 copy_process+0xbaa/0x10e3 sighand_cache: 77 copy_process+0xe65/0x10e3 sighand_cache: 1 de_thread+0x4d/0x5f8 anon_vma: 241 anon_vma_prepare+0xd9/0xf3 size-2048: 1 add_sect_attrs+0x5f/0x145 size-2048: 2 journal_init_revoke+0x99/0x302 size-2048: 2 journal_init_revoke+0x137/0x302 size-2048: 2 journal_init_inode+0xf9/0x1c4 Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexander Nyberg <alexn@telia.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Cc: Ravikiran Thirumalai <kiran@scalex86.org> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> DESC slab-leaks3-locking-fix EDESC From: Andrew Morton <akpm@osdl.org> Update for slab-remove-cachep-spinlock.patch Cc: Al Viro <viro@ftp.linux.org.uk> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexander Nyberg <alexn@telia.com> Cc: Pekka Enberg <penberg@cs.helsinki.fi> Cc: Christoph Lameter <clameter@engr.sgi.com> Cc: Ravikiran Thirumalai <kiran@scalex86.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-25 19:06:39 +08:00
#ifdef CONFIG_HARDENED_USERCOPY
/*
* Rejects incorrectly sized objects and objects that are to be copied
* to/from userspace but do not fall entirely within the containing slab
* cache's usercopy region.
*
* Returns NULL if check passes, otherwise const char * to name of cache
* to indicate an error.
*/
void __check_heap_object(const void *ptr, unsigned long n, struct page *page,
bool to_user)
{
struct kmem_cache *cachep;
unsigned int objnr;
unsigned long offset;
ptr = kasan_reset_tag(ptr);
/* Find and validate object. */
cachep = page->slab_cache;
objnr = obj_to_index(cachep, page, (void *)ptr);
BUG_ON(objnr >= cachep->num);
/* Find offset within object. */
offset = ptr - index_to_obj(cachep, page, objnr) - obj_offset(cachep);
/* Allow address range falling entirely within usercopy region. */
if (offset >= cachep->useroffset &&
offset - cachep->useroffset <= cachep->usersize &&
n <= cachep->useroffset - offset + cachep->usersize)
return;
/*
* If the copy is still within the allocated object, produce
* a warning instead of rejecting the copy. This is intended
* to be a temporary method to find any missing usercopy
* whitelists.
*/
if (usercopy_fallback &&
offset <= cachep->object_size &&
n <= cachep->object_size - offset) {
usercopy_warn("SLAB object", cachep->name, to_user, offset, n);
return;
}
usercopy_abort("SLAB object", cachep->name, to_user, offset, n);
}
#endif /* CONFIG_HARDENED_USERCOPY */
/**
* __ksize -- Uninstrumented ksize.
* @objp: pointer to the object
*
* Unlike ksize(), __ksize() is uninstrumented, and does not provide the same
* safety checks as ksize() with KASAN instrumentation enabled.
*
* Return: size of the actual memory used by @objp in bytes
*/
size_t __ksize(const void *objp)
{
mm/slab: sanity-check page type when looking up cache This avoids any possible type confusion when looking up an object. For example, if a non-slab were to be passed to kfree(), the invalid slab_cache pointer (i.e. overlapped with some other value from the struct page union) would be used for subsequent slab manipulations that could lead to further memory corruption. Since the page is already in cache, adding the PageSlab() check will have nearly zero cost, so add a check and WARN() to virt_to_cache(). Additionally replaces an open-coded virt_to_cache(). To support the failure mode this also updates all callers of virt_to_cache() and cache_from_obj() to handle a NULL cache pointer return value (though note that several already handle this case gracefully). [dan.carpenter@oracle.com: restore IRQs in kfree()] Link: http://lkml.kernel.org/r/20190613065637.GE16334@mwanda Link: http://lkml.kernel.org/r/20190530045017.15252-3-keescook@chromium.org Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Cc: Alexander Popov <alex.popov@linux.com> Cc: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:53:26 +08:00
struct kmem_cache *c;
size_t size;
BUG_ON(!objp);
if (unlikely(objp == ZERO_SIZE_PTR))
return 0;
mm/slab: sanity-check page type when looking up cache This avoids any possible type confusion when looking up an object. For example, if a non-slab were to be passed to kfree(), the invalid slab_cache pointer (i.e. overlapped with some other value from the struct page union) would be used for subsequent slab manipulations that could lead to further memory corruption. Since the page is already in cache, adding the PageSlab() check will have nearly zero cost, so add a check and WARN() to virt_to_cache(). Additionally replaces an open-coded virt_to_cache(). To support the failure mode this also updates all callers of virt_to_cache() and cache_from_obj() to handle a NULL cache pointer return value (though note that several already handle this case gracefully). [dan.carpenter@oracle.com: restore IRQs in kfree()] Link: http://lkml.kernel.org/r/20190613065637.GE16334@mwanda Link: http://lkml.kernel.org/r/20190530045017.15252-3-keescook@chromium.org Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Cc: Alexander Popov <alex.popov@linux.com> Cc: Alexander Potapenko <glider@google.com> Cc: Christoph Lameter <cl@linux.com> Cc: David Rientjes <rientjes@google.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Pekka Enberg <penberg@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:53:26 +08:00
c = virt_to_cache(objp);
size = c ? c->object_size : 0;
return size;
}
EXPORT_SYMBOL(__ksize);