Rewrite the vmap allocator to use rbtrees and lazy tlb flushing, and
provide a fast, scalable percpu frontend for small vmaps (requires a
slightly different API, though).
The biggest problem with vmap is actually vunmap. Presently this requires
a global kernel TLB flush, which on most architectures is a broadcast IPI
to all CPUs to flush the cache. This is all done under a global lock. As
the number of CPUs increases, so will the number of vunmaps a scaled
workload will want to perform, and so will the cost of a global TLB flush.
This gives terrible quadratic scalability characteristics.
Another problem is that the entire vmap subsystem works under a single
lock. It is a rwlock, but it is actually taken for write in all the fast
paths, and the read locking would likely never be run concurrently anyway,
so it's just pointless.
This is a rewrite of vmap subsystem to solve those problems. The existing
vmalloc API is implemented on top of the rewritten subsystem.
The TLB flushing problem is solved by using lazy TLB unmapping. vmap
addresses do not have to be flushed immediately when they are vunmapped,
because the kernel will not reuse them again (would be a use-after-free)
until they are reallocated. So the addresses aren't allocated again until
a subsequent TLB flush. A single TLB flush then can flush multiple
vunmaps from each CPU.
XEN and PAT and such do not like deferred TLB flushing because they can't
always handle multiple aliasing virtual addresses to a physical address.
They now call vm_unmap_aliases() in order to flush any deferred mappings.
That call is very expensive (well, actually not a lot more expensive than
a single vunmap under the old scheme), however it should be OK if not
called too often.
The virtual memory extent information is stored in an rbtree rather than a
linked list to improve the algorithmic scalability.
There is a per-CPU allocator for small vmaps, which amortizes or avoids
global locking.
To use the per-CPU interface, the vm_map_ram / vm_unmap_ram interfaces
must be used in place of vmap and vunmap. Vmalloc does not use these
interfaces at the moment, so it will not be quite so scalable (although it
will use lazy TLB flushing).
As a quick test of performance, I ran a test that loops in the kernel,
linearly mapping then touching then unmapping 4 pages. Different numbers
of tests were run in parallel on an 4 core, 2 socket opteron. Results are
in nanoseconds per map+touch+unmap.
threads vanilla vmap rewrite
1 14700 2900
2 33600 3000
4 49500 2800
8 70631 2900
So with a 8 cores, the rewritten version is already 25x faster.
In a slightly more realistic test (although with an older and less
scalable version of the patch), I ripped the not-very-good vunmap batching
code out of XFS, and implemented the large buffer mapping with vm_map_ram
and vm_unmap_ram... along with a couple of other tricks, I was able to
speed up a large directory workload by 20x on a 64 CPU system. I believe
vmap/vunmap is actually sped up a lot more than 20x on such a system, but
I'm running into other locks now. vmap is pretty well blown off the
profiles.
Before:
1352059 total 0.1401
798784 _write_lock 8320.6667 <- vmlist_lock
529313 default_idle 1181.5022
15242 smp_call_function 15.8771 <- vmap tlb flushing
2472 __get_vm_area_node 1.9312 <- vmap
1762 remove_vm_area 4.5885 <- vunmap
316 map_vm_area 0.2297 <- vmap
312 kfree 0.1950
300 _spin_lock 3.1250
252 sn_send_IPI_phys 0.4375 <- tlb flushing
238 vmap 0.8264 <- vmap
216 find_lock_page 0.5192
196 find_next_bit 0.3603
136 sn2_send_IPI 0.2024
130 pio_phys_write_mmr 2.0312
118 unmap_kernel_range 0.1229
After:
78406 total 0.0081
40053 default_idle 89.4040
33576 ia64_spinlock_contention 349.7500
1650 _spin_lock 17.1875
319 __reg_op 0.5538
281 _atomic_dec_and_lock 1.0977
153 mutex_unlock 1.5938
123 iget_locked 0.1671
117 xfs_dir_lookup 0.1662
117 dput 0.1406
114 xfs_iget_core 0.0268
92 xfs_da_hashname 0.1917
75 d_alloc 0.0670
68 vmap_page_range 0.0462 <- vmap
58 kmem_cache_alloc 0.0604
57 memset 0.0540
52 rb_next 0.1625
50 __copy_user 0.0208
49 bitmap_find_free_region 0.2188 <- vmap
46 ia64_sn_udelay 0.1106
45 find_inode_fast 0.1406
42 memcmp 0.2188
42 finish_task_switch 0.1094
42 __d_lookup 0.0410
40 radix_tree_lookup_slot 0.1250
37 _spin_unlock_irqrestore 0.3854
36 xfs_bmapi 0.0050
36 kmem_cache_free 0.0256
35 xfs_vn_getattr 0.0322
34 radix_tree_lookup 0.1062
33 __link_path_walk 0.0035
31 xfs_da_do_buf 0.0091
30 _xfs_buf_find 0.0204
28 find_get_page 0.0875
27 xfs_iread 0.0241
27 __strncpy_from_user 0.2812
26 _xfs_buf_initialize 0.0406
24 _xfs_buf_lookup_pages 0.0179
24 vunmap_page_range 0.0250 <- vunmap
23 find_lock_page 0.0799
22 vm_map_ram 0.0087 <- vmap
20 kfree 0.0125
19 put_page 0.0330
18 __kmalloc 0.0176
17 xfs_da_node_lookup_int 0.0086
17 _read_lock 0.0885
17 page_waitqueue 0.0664
vmap has gone from being the top 5 on the profiles and flushing the crap
out of all TLBs, to using less than 1% of kernel time.
[akpm@linux-foundation.org: cleanups, section fix]
[akpm@linux-foundation.org: fix build on alpha]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Krzysztof Helt <krzysztof.h1@poczta.fm>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Try to comment away a little of the confusion between mm's vm_area_struct
vm_flags and vmalloc's vm_struct flags: based on an idea by Ulrich Drepper.
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Implement a new proc file that allows the display of the currently allocated
vmalloc memory.
It allows to see the users of vmalloc. That is important if vmalloc space is
scarce (i386 for example).
And it's going to be important for the compound page fallback to vmalloc.
Many of the current users can be switched to use compound pages with fallback.
This means that the number of users of vmalloc is reduced and page tables no
longer necessary to access the memory. /proc/vmallocinfo allows to review how
that reduction occurs.
If memory becomes fragmented and larger order allocations are no longer
possible then /proc/vmallocinfo allows to see which compound page allocations
fell back to virtual compound pages. That is important for new users of
virtual compound pages. Such as order 1 stack allocation etc that may
fallback to virtual compound pages in the future.
/proc/vmallocinfo permissions are made readable-only-by-root to avoid possible
information leakage.
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: CONFIG_MMU=n build fix]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Arjan van de Ven <arjan@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make vmalloc functions work the same way as kfree() and friends that
take a const void * argument.
[akpm@linux-foundation.org: fix consts, coding-style]
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>
get_vm_area always returns an area with an adjacent guard page. That guard
page is included in vm_struct.size. iounmap uses vm_struct.size to
determine how much address space needs to have change_page_attr applied to
it, which will BUG if applied to the guard page.
This patch adds a helper function - get_vm_area_size() in linux/vmalloc.h -
to return the actual size of a vm area, and uses it to make iounmap do the
right thing. There are probably other places which should be using
get_vm_area_size().
Thanks to Dave Young <hidave.darkstar@gmail.com> for debugging the
problem.
[ Andi, it wasn't clear to me whether x86_64 needs the same fix. ]
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Dave Young <hidave.darkstar@gmail.com>
Cc: Chuck Ebbert <cebbert@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Allocate/release a chunk of vmalloc address space:
alloc_vm_area reserves a chunk of address space, and makes sure all
the pagetables are constructed for that address range - but no pages.
free_vm_area releases the address space range.
Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ian Pratt <ian.pratt@xensource.com>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: "Jan Beulich" <JBeulich@novell.com>
Cc: "Andi Kleen" <ak@muc.de>
This makes unmap_vm_area static and a wrapper around a new
exported unmap_kernel_range that takes an explicit range instead
of a vm_area struct.
This makes it more versatile for code that wants to play with kernel
page tables outside of the standard vmalloc area.
(One example is some rework of the PowerPC PCI IO space mapping
code that depends on that patch and removes some code duplication
and horrible abuse of forged struct vm_struct).
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
This patch moves the die notifier handling to common code. Previous
various architectures had exactly the same code for it. Note that the new
code is compiled unconditionally, this should be understood as an appel to
the other architecture maintainer to implement support for it aswell (aka
sprinkling a notify_die or two in the proper place)
arm had a notifiy_die that did something totally different, I renamed it to
arm_notify_die as part of the patch and made it static to the file it's
declared and used at. avr32 used to pass slightly less information through
this interface and I brought it into line with the other architectures.
[akpm@linux-foundation.org: build fix]
[akpm@linux-foundation.org: fix vmalloc_sync_all bustage]
[bryan.wu@analog.com: fix vmalloc_sync_all in nommu]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: <linux-arch@vger.kernel.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- reorder 'struct vm_struct' to speedup lookups on CPUS with small cache
lines. The fields 'next,addr,size' should be now in the same cache line,
to speedup lookups.
- One minor cleanup in __get_vm_area_node()
- Bugfixes in vmalloc_user() and vmalloc_32_user() NULL returns from
__vmalloc() and __find_vm_area() were not tested.
[akpm@osdl.org: remove redundant BUG_ONs]
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If __vmalloc is called to allocate memory with GFP_ATOMIC in atomic
context, the chain of calls results in __get_vm_area_node allocating memory
for vm_struct with GFP_KERNEL, causing the 'sleeping from invalid context'
warning. This patch fixes it by passing the gfp flags along so
__get_vm_area_node allocates memory for vm_struct with the same flags.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The function is exported but not used from anywhere else. It's also marked as
"not for driver use" so noone out there should really care.
Signed-off-by: Rolf Eike Beer <eike-kernel@sf-tec.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch makes the following needlessly global functions static:
- slab.c: kmem_find_general_cachep()
- swap.c: __page_cache_release()
- vmalloc.c: __vmalloc_node()
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
__vunmap must not rely on area->nr_pages when picking the release methode
for area->pages. It may be too small when __vmalloc_area_node failed early
due to lacking memory. Instead, use a flag in vmstruct to differentiate.
Signed-off-by: Jan Kiszka <jan.kiszka@web.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add remap_vmalloc_range, vmalloc_user, and vmalloc_32_user so that drivers
can have a nice interface for remapping vmalloc memory.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch adds
vmalloc_node(size, node) -> Allocate necessary memory on the specified node
and
get_vm_area_node(size, flags, node)
and the other functions that it depends on.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
- added typedef unsigned int __nocast gfp_t;
- replaced __nocast uses for gfp flags with gfp_t - it gives exactly
the same warnings as far as sparse is concerned, doesn't change
generated code (from gcc point of view we replaced unsigned int with
typedef) and documents what's going on far better.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Version 6 of the ARM architecture introduces the concept of 16MB pages
(supersections) and 36-bit (40-bit actually, but nobody uses this) physical
addresses. 36-bit addressed memory and I/O and ARMv6 can only be mapped
using supersections and the requirement on these is that both virtual and
physical addresses be 16MB aligned. In trying to add support for ioremap()
of 36-bit I/O, we run into the issue that get_vm_area() allows for a
maximum of 512K alignment via the IOREMAP_MAX_ORDER constant. To work
around this, we can:
- Allocate a larger VM area than needed (size + (1ul << IOREMAP_MAX_ORDER))
and then align the pointer ourselves, but this ends up with 512K of
wasted VM per ioremap().
- Provide a new __get_vm_area_aligned() API and make __get_vm_area() sit
on top of this. I did this and it works but I don't like the idea
adding another VM API just for this one case.
- My preferred solution which is to allow the architecture to override
the IOREMAP_MAX_ORDER constant with it's own version.
Signed-off-by: Deepak Saxena <dsaxena@plexity.net>
Cc: Russell King <rmk@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Caused oopses again. Also fix potential mismatch in checking if
change_page_attr was needed.
To do it without races I needed to change mm/vmalloc.c to export a
__remove_vm_area that does not take vmlist lock.
Noticed by Terence Ripperda and based on a patch of his.
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Alexey Dobriyan