DISCONTIGMEM on x86-32 implements pfn -> nid mapping similarly to
SPARSEMEM; however, it calls each mapping unit ELEMENT instead of
SECTION. This patch renames it to SECTION so that PAGES_PER_SECTION
is valid for both DISCONTIGMEM and SPARSEMEM. This will be used by
the next patch to implement mapping granularity check.
This patch is trivial constant rename.
Signed-off-by: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20110712074422.GA2872@htj.dyndns.org
Cc: Hans Rosenfeld <hans.rosenfeld@amd.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
During 32/64 NUMA init unification, commit 797390d855 ("x86-32,
NUMA: use sparse_memory_present_with_active_regions()") made
32bit mm init call memory_present() automatically from
active_regions instead of leaving it to each NUMA init path.
This commit description is inaccurate - memory_present() calls
aren't the same for flat and numaq. After the commit,
memory_present() is only called for the intersection of e820 and
NUMA layout. Before, on flatmem, memory_present() would be
called from 0 to max_pfn. After, it would be called only on the
areas that e820 indicates to be populated.
This is how x86_64 works and should be okay as memmap is allowed
to contain holes; however, x86_32 DISCONTIGMEM is missing
early_pfn_valid(), which makes memmap_init_zone() assume that
memmap doesn't contain any hole. This leads to the following
oops if e820 map contains holes as it often does on machine with
near or more 4GiB of memory by calling pfn_to_page() on a pfn
which isn't mapped to a NUMA node, a reported by Conny Seidel:
BUG: unable to handle kernel paging request at 000012b0
IP: [<c1aa13ce>] memmap_init_zone+0x6c/0xf2
*pdpt =3D 0000000000000000 *pde =3D f000eef3f000ee00
Oops: 0000 [#1] SMP
last sysfs file:
Modules linked in:
Pid: 0, comm: swapper Not tainted 2.6.39-rc5-00164-g797390d #1 To Be Filled By O.E.M. To Be Filled By O.E.M./E350M1
EIP: 0060:[<c1aa13ce>] EFLAGS: 00010012 CPU: 0
EIP is at memmap_init_zone+0x6c/0xf2
EAX: 00000000 EBX: 000a8000 ECX: 000a7fff EDX: f2c00b80
ESI: 000a8000 EDI: f2c00800 EBP: c19ffe54 ESP: c19ffe34
DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
Process swapper (pid: 0, ti=3Dc19fe000 task=3Dc1a07f60 task.ti=3Dc19fe000)
Stack:
00000002 00000000 0023f000 00000000 10000000 00000a00 f2c00000 f2c00b58
c19ffeb0 c1a80f24 000375fe 00000000 f2c00800 00000800 00000100 00000030
c1abb768 0000003c 00000000 00000000 00000004 00207a02 f2c00800 000375fe
Call Trace:
[<c1a80f24>] free_area_init_node+0x358/0x385
[<c1a81384>] free_area_init_nodes+0x420/0x487
[<c1a79326>] paging_init+0x114/0x11b
[<c1a6cb13>] setup_arch+0xb37/0xc0a
[<c1a69554>] start_kernel+0x76/0x316
[<c1a690a8>] i386_start_kernel+0xa8/0xb0
This patch fixes the bug by defining early_pfn_valid() to be the
same as pfn_valid() when DISCONTIGMEM.
Reported-bisected-and-tested-by: Conny Seidel <conny.seidel@amd.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: hans.rosenfeld@amd.com
Cc: Christoph Lameter <cl@linux.com>
Cc: Conny Seidel <conny.seidel@amd.com>
Link: http://lkml.kernel.org/r/20110628094107.GB3386@htj.dyndns.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
commit 21a3c96 uses node_start/end_pfn(nid) for detection start/end
of nodes. But, it's not defined in linux/mmzone.h but defined in
/arch/???/include/mmzone.h which is included only under
CONFIG_NEED_MULTIPLE_NODES=y.
Then, we see
mm/page_cgroup.c: In function 'page_cgroup_init':
mm/page_cgroup.c:308: error: implicit declaration of function 'node_start_pfn'
mm/page_cgroup.c:309: error: implicit declaration of function 'node_end_pfn'
So, fixiing page_cgroup.c is an idea...
But node_start_pfn()/node_end_pfn() is a very generic macro and
should be implemented in the same manner for all archs.
(m32r has different implementation...)
This patch removes definitions of node_start/end_pfn() in each archs
and defines a unified one in linux/mmzone.h. It's not under
CONFIG_NEED_MULTIPLE_NODES, now.
A result of macro expansion is here (mm/page_cgroup.c)
for !NUMA
start_pfn = ((&contig_page_data)->node_start_pfn);
end_pfn = ({ pg_data_t *__pgdat = (&contig_page_data); __pgdat->node_start_pfn + __pgdat->node_spanned_pages;});
for NUMA (x86-64)
start_pfn = ((node_data[nid])->node_start_pfn);
end_pfn = ({ pg_data_t *__pgdat = (node_data[nid]); __pgdat->node_start_pfn + __pgdat->node_spanned_pages;});
Changelog:
- fixed to avoid using "nid" twice in node_end_pfn() macro.
Reported-and-acked-by: Randy Dunlap <randy.dunlap@oracle.com>
Reported-and-tested-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
SRAT support implementation in srat_32.c and srat.c are generally
similar; however, there are some differences.
First of all, 64bit implementation supports more types of SRAT
entries. 64bit supports x2apic, affinity, memory and SLIT. 32bit
only supports processor and memory.
Most other differences stem from different initialization protocols
employed by 64bit and 32bit NUMA init paths.
On 64bit,
* Mappings among PXM, node and apicid are directly done in each SRAT
entry callback.
* Memory affinity information is passed to numa_add_memblk() which
takes care of all interfacing with NUMA init.
* Doesn't directly initialize NUMA configurations. All the
information is recorded in numa_nodes_parsed and memblks.
On 32bit,
* Checks numa_off.
* Things go through one more level of indirection via private tables
but eventually end up initializing the same mappings.
* node_start/end_pfn[] are initialized and
memblock_x86_register_active_regions() is called for each memory
chunk.
* node_set_online() is called for each online node.
* sort_node_map() is called.
There are also other minor differences in sanity checking and messages
but taking 64bit version should be good enough.
This patch drops the 32bit specific implementation and makes the 64bit
implementation common for both 32 and 64bit.
The init protocol differences are dealt with in two places - the
numa_add_memblk() shim added in the previous patch and new temporary
numa_32.c:get_memcfg_from_srat() which wraps invocation of
x86_acpi_numa_init().
The shim numa_add_memblk() handles the folowings.
* node_start/end_pfn[] initialization.
* node_set_online() for memory nodes.
* Invocation of memblock_x86_register_active_regions().
The shim get_memcfg_from_srat() handles the followings.
* numa_off check.
* node_set_online() for CPU nodes.
* sort_node_map() invocation.
* Clearing of numa_nodes_parsed and active_ranges on failure.
The shims are temporary and will be removed as the generic NUMA init
path in 32bit is replaced with 64bit one.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
There's no reason get_memcfg_numa() to be implemented inline in
mmzone_32.h. Move it to numa_32.c and also make
get_memcfg_numa_flag() static.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Yinghai Lu <yinghai@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Impact: fix new breakages introduced by previous fix
Commit c132937556 tried to clean up
bootmem arch wrapper but it wasn't quite correct. Before the commit,
the followings were broken.
* Low level interface functions prefixed with __ ignored arch
preference.
* reserve_bootmem(...) can't be mapped into
reserve_bootmem_node(NODE_DATA(0)->bdata, ...) because the node is
not preference here. The region specified MUST fall into the
specified region; otherwise, it will panic.
After the commit,
* If allocation fails for the arch preferred node, it should fallback
to whatever is available. Instead, it simply failed allocation.
There are too many internal details to allow generic wrapping and
still keep things simple for archs. Plus, all that arch wants is a
way to prefer certain node over another.
This patch drops the generic wrapping around alloc_bootmem_core() and
add alloc_bootmem_core() instead. If necessary, arch can define
bootmem_arch_referred_node() macro or function which takes all
allocation information and returns the preferred node. bootmem
generic code will always try the preferred node first and then
fallback to other nodes as usual.
Breakages noted and changes reviewed by Johannes Weiner.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Impact: cleaner and consistent bootmem wrapping
By setting CONFIG_HAVE_ARCH_BOOTMEM_NODE, archs can define
arch-specific wrappers for bootmem allocation. However, this is done
a bit strangely in that only the high level convenience macros can be
changed while lower level, but still exported, interface functions
can't be wrapped. This not only is messy but also leads to strange
situation where alloc_bootmem() does what the arch wants it to do but
the equivalent __alloc_bootmem() call doesn't although they should be
able to be used interchangeably.
This patch updates bootmem such that archs can override / wrap the
backend function - alloc_bootmem_core() instead of the highlevel
interface functions to allow simpler and consistent wrapping. Also,
HAVE_ARCH_BOOTMEM_NODE is renamed to HAVE_ARCH_BOOTMEM.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Johannes Weiner <hannes@saeurebad.de>
What's happening is that the assertion in mm/page_alloc.c:move_freepages()
is triggering:
BUG_ON(page_zone(start_page) != page_zone(end_page));
Once I knew this is what was happening, I added some annotations:
if (unlikely(page_zone(start_page) != page_zone(end_page))) {
printk(KERN_ERR "move_freepages: Bogus zones: "
"start_page[%p] end_page[%p] zone[%p]\n",
start_page, end_page, zone);
printk(KERN_ERR "move_freepages: "
"start_zone[%p] end_zone[%p]\n",
page_zone(start_page), page_zone(end_page));
printk(KERN_ERR "move_freepages: "
"start_pfn[0x%lx] end_pfn[0x%lx]\n",
page_to_pfn(start_page), page_to_pfn(end_page));
printk(KERN_ERR "move_freepages: "
"start_nid[%d] end_nid[%d]\n",
page_to_nid(start_page), page_to_nid(end_page));
...
And here's what I got:
move_freepages: Bogus zones: start_page[2207d0000] end_page[2207dffc0] zone[fffff8103effcb00]
move_freepages: start_zone[fffff8103effcb00] end_zone[fffff8003fffeb00]
move_freepages: start_pfn[0x81f600] end_pfn[0x81f7ff]
move_freepages: start_nid[1] end_nid[0]
My memory layout on this box is:
[ 0.000000] Zone PFN ranges:
[ 0.000000] Normal 0x00000000 -> 0x0081ff5d
[ 0.000000] Movable zone start PFN for each node
[ 0.000000] early_node_map[8] active PFN ranges
[ 0.000000] 0: 0x00000000 -> 0x00020000
[ 0.000000] 1: 0x00800000 -> 0x0081f7ff
[ 0.000000] 1: 0x0081f800 -> 0x0081fe50
[ 0.000000] 1: 0x0081fed1 -> 0x0081fed8
[ 0.000000] 1: 0x0081feda -> 0x0081fedb
[ 0.000000] 1: 0x0081fedd -> 0x0081fee5
[ 0.000000] 1: 0x0081fee7 -> 0x0081ff51
[ 0.000000] 1: 0x0081ff59 -> 0x0081ff5d
So it's a block move in that 0x81f600-->0x81f7ff region which triggers
the problem.
This patch:
Declaration of early_pfn_to_nid() is scattered over per-arch include
files, and it seems it's complicated to know when the declaration is used.
I think it makes fix-for-memmap-init not easy.
This patch moves all declaration to include/linux/mm.h
After this,
if !CONFIG_NODES_POPULATES_NODE_MAP && !CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
-> Use static definition in include/linux/mm.h
else if !CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
-> Use generic definition in mm/page_alloc.c
else
-> per-arch back end function will be called.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Tested-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reported-by: David Miller <davem@davemlloft.net>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: <stable@kernel.org> [2.6.25.x, 2.6.26.x, 2.6.27.x, 2.6.28.x]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Impact: fix crash during hibernation on 32-bit NUMA
The NUMA code on x86_32 creates special memory mapping that allows
each node's pgdat to be located in this node's memory. For this
purpose it allocates a memory area at the end of each node's memory
and maps this area so that it is accessible with virtual addresses
belonging to low memory. As a result, if there is high memory,
these NUMA-allocated areas are physically located in high memory,
although they are mapped to low memory addresses.
Our hibernation code does not take that into account and for this
reason hibernation fails on all x86_32 systems with CONFIG_NUMA=y and
with high memory present. Fix this by adding a special mapping for
the NUMA-allocated memory areas to the temporary page tables created
during the last phase of resume.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Change header guards named "ASM_X86__*" to "_ASM_X86_*" since:
a. the double underscore is ugly and pointless.
b. no leading underscore violates namespace constraints.
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Tejun Heo