At present our slab debugging tells us that it detected a double-free or
corruption - it does not distinguish between them.  Sometimes it's useful
to be able to differentiate between these two types of information.

Add double-free detection to redzone verification when freeing an object.
As explained by Manfred, when we are freeing an object, both redzones
should be RED_ACTIVE.  However, if both are RED_INACTIVE, we are trying to
free an object that was already free'd.
Signed-off-by: NManfred Spraul <manfred@colorfullife.com>
Signed-off-by: NPekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

With likely/unlikely profiling on my not-so-busy-typical-developmentsystem
there are 5k misses vs 2k hits.  So I guess we should remove the unlikely.
Signed-off-by: NHua Zhong <hzhong@gmail.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus 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: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Rework the swsusp's memory shrinker in the following way:

- Simplify balance_pgdat() by removing all of the swsusp-related code
  from it.

- Make shrink_all_memory() use shrink_slab() and a new function
  shrink_all_zones() which calls shrink_active_list() and
  shrink_inactive_list() directly for each zone in a way that's optimized
  for suspend.

In shrink_all_memory() we try to free exactly as many pages as the caller
asks for, preferably in one shot, starting from easier targets.   If slab
caches are huge, they are most likely to have enough pages to reclaim.
 The inactive lists are next (the zones with more inactive pages go first)
etc.

Each time shrink_all_memory() attempts to shrink the active and inactive
lists for each zone in 5 passes.   In the first pass, only the inactive
lists are taken into consideration.   In the next two passes the active
lists are also shrunk, but mapped pages are not reclaimed.   In the last
two passes the active and inactive lists are shrunk and mapped pages are
reclaimed as well.  The aim of this is to alter the reclaim logic to choose
the best pages to keep on resume and improve the responsiveness of the
resumed system.
Signed-off-by: NRafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: NCon Kolivas <kernel@kolivas.org>
Signed-off-by: NAdrian Bunk <bunk@stusta.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Use the _entry variant everywhere to clean the code up a tiny bit.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The last ifdef addition hit the ugliness treshold on this functions, so:

 - rename the variable i to nr_pages so it's somewhat descriptive
 - remove the addr variable and do the page_address call at the very end
 - instead of ifdef'ing the whole alloc_pages_node call just make the
   __GFP_COMP addition to flags conditional
 - rewrite the __GFP_COMP comment to make sense
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Current hugetlb strict accounting for shared mapping always assume mapping
starts at zero file offset and reserves pages between zero and size of the
file.  This assumption often reserves (or lock down) a lot more pages then
necessary if application maps at none zero file offset.  libhugetlbfs is
one example that requires proper reservation on shared mapping starts at
none zero offset.

This patch extends the reservation and hugetlb strict accounting to support
any arbitrary pair of (offset, len), resulting a much more robust and
accurate scheme.  More importantly, it won't lock down any hugetlb pages
outside file mapping.
Signed-off-by: NKen Chen <kenneth.w.chen@intel.com>
Acked-by: NAdam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: William Lee Irwin III <wli@holomorphy.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This fixes a few typos in the comments in mm/oom_kill.c.
Signed-off-by: NDavid S. Peterson <dsp@llnl.gov>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This patch adds panic_on_oom sysctl under sys.vm.

When sysctl vm.panic_on_oom = 1, the kernel panics intead of killing rogue
processes.  And if vm.panic_on_oom is 0 the kernel will do oom_kill() in
the same way as it does today.  Of course, the default value is 0 and only
root can modifies it.

In general, oom_killer works well and kill rogue processes.  So the whole
system can survive.  But there are environments where panic is preferable
rather than kill some processes.
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

We have architectures where the size of page_to_pfn and pfn_to_page are
significant enough to overall image size that they wish to push them out of
line.  However, in the process we have grown a second copy of the
implementation of each of these routines for each memory model.  Share the
implmentation exposing it either inline or out-of-line as required.
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

In current code, zonelist is considered to be build once, no modification.
But MemoryHotplug can add new zone/pgdat.  It must be updated.

This patch modifies build_all_zonelists().  By this, build_all_zonelist() can
reconfig pgdat's zonelists.

To update them safety, this patch use stop_machine_run().  Other cpus don't
touch among updating them by using it.

In old version (V2 of node hotadd), kernel updated them after zone
initialization.  But present_page of its new zone is still 0, because
online_page() is not called yet at this time.  Build_zonelists() checks
present_pages to find present zone.  It was too early.  So, I changed it after
online_pages().
Signed-off-by: NYasunori Goto     <y-goto@jp.fujitsu.com>
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Wait_table is initialized according to zone size at boot time.  But, we cannot
know the maixmum zone size when memory hotplug is enabled.  It can be
changed....  And resizing of wait_table is hard.

So kernel allocate and initialzie wait_table as its maximum size.
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NYasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

[PATCH] wait_table and zonelist initializing for memory hotadd: add return code for init_current_empty_zone

When add_zone() is called against empty zone (not populated zone), we have to
initialize the zone which didn't initialize at boot time.  But,
init_currently_empty_zone() may fail due to allocation of wait table.  So,
this patch is to catch its error code.

Changes against wait_table is in the next patch.
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NYasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

[PATCH] wait_table and zonelist initializing for memory hotadd: change to meminit for build_zonelist

Change definitions of some functions and data from __init to __meminit.

These functions and data can be used after bootup by this patch to be used for
hot-add codes.
Signed-off-by: NYasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This is just to rename from wait_table_size() to wait_table_hash_nr_entries().
Signed-off-by: NYasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Remove two unnecessary PageSwapCache checks.  The page refcount is raised
and therefore page migration cannot occur in both functions.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Clean up slab allocator page mapping a bit.  The memory allocated for a
slab is physically contiguous so it is okay to assume struct pages are too
so kill the long-standing comment.  Furthermore, rename set_slab_attr to
slab_map_pages and add a comment explaining why its needed.
Signed-off-by: NPekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Move alien object freeing to cache_free_alien() to reduce #ifdef clutter in
__cache_free().
Signed-off-by: NPekka Enberg <penberg@cs.helsinki.fi>
Acked-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

It is better to redo the complete fault if do_swap_page() finds that the
page is not in PageSwapCache() because the page migration code may have
replaced the swap pte already with a pte pointing to valid memory.

do_swap_page() may interpret an invalid swap entry without this patch
because we do not reload the pte if we are looping back.  The page
migration code may already have reused the swap entry referenced by our
local swp_entry.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The buddy allocator has a requirement that boundaries between contigious
zones occur aligned with the the MAX_ORDER ranges.  Where they do not we
will incorrectly merge pages cross zone boundaries.  This can lead to pages
from the wrong zone being handed out.

Originally the buddy allocator would check that buddies were in the same
zone by referencing the zone start and end page frame numbers.  This was
removed as it became very expensive and the buddy allocator already made
the assumption that zones boundaries were aligned.

It is clear that not all configurations and architectures are honouring
this alignment requirement.  Therefore it seems safest to reintroduce
support for non-aligned zone boundaries.  This patch introduces a new check
when considering a page a buddy it compares the zone_table index for the
two pages and refuses to merge the pages where they do not match.  The
zone_table index is unique for each node/zone combination when
FLATMEM/DISCONTIGMEM is enabled and for each section/zone combination when
SPARSEMEM is enabled (a SPARSEMEM section is at least a MAX_ORDER size).
Signed-off-by: NAndy Whitcroft <apw@shadowen.org>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Yasunori Goto <y-goto@jp.fujitsu.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Give the statfs superblock operation a dentry pointer rather than a superblock
pointer.

This complements the get_sb() patch.  That reduced the significance of
sb->s_root, allowing NFS to place a fake root there.  However, NFS does
require a dentry to use as a target for the statfs operation.  This permits
the root in the vfsmount to be used instead.

linux/mount.h has been added where necessary to make allyesconfig build
successfully.

Interest has also been expressed for use with the FUSE and XFS filesystems.
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NAl Viro <viro@zeniv.linux.org.uk>
Cc: Nathan Scott <nathans@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Extend the get_sb() filesystem operation to take an extra argument that
permits the VFS to pass in the target vfsmount that defines the mountpoint.

The filesystem is then required to manually set the superblock and root dentry
pointers.  For most filesystems, this should be done with simple_set_mnt()
which will set the superblock pointer and then set the root dentry to the
superblock's s_root (as per the old default behaviour).

The get_sb() op now returns an integer as there's now no need to return the
superblock pointer.

This patch permits a superblock to be implicitly shared amongst several mount
points, such as can be done with NFS to avoid potential inode aliasing.  In
such a case, simple_set_mnt() would not be called, and instead the mnt_root
and mnt_sb would be set directly.

The patch also makes the following changes:

 (*) the get_sb_*() convenience functions in the core kernel now take a vfsmount
     pointer argument and return an integer, so most filesystems have to change
     very little.

 (*) If one of the convenience function is not used, then get_sb() should
     normally call simple_set_mnt() to instantiate the vfsmount. This will
     always return 0, and so can be tail-called from get_sb().

 (*) generic_shutdown_super() now calls shrink_dcache_sb() to clean up the
     dcache upon superblock destruction rather than shrink_dcache_anon().

     This is required because the superblock may now have multiple trees that
     aren't actually bound to s_root, but that still need to be cleaned up. The
     currently called functions assume that the whole tree is rooted at s_root,
     and that anonymous dentries are not the roots of trees which results in
     dentries being left unculled.

     However, with the way NFS superblock sharing are currently set to be
     implemented, these assumptions are violated: the root of the filesystem is
     simply a dummy dentry and inode (the real inode for '/' may well be
     inaccessible), and all the vfsmounts are rooted on anonymous[*] dentries
     with child trees.

     [*] Anonymous until discovered from another tree.

 (*) The documentation has been adjusted, including the additional bit of
     changing ext2_* into foo_* in the documentation.

[akpm@osdl.org: convert ipath_fs, do other stuff]
Signed-off-by: NDavid Howells <dhowells@redhat.com>
Acked-by: NAl Viro <viro@zeniv.linux.org.uk>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Roland Dreier <rolandd@cisco.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

shmem_rmdir() must undo the increment of i_nlink done in
shmem_get_inode() for directories, otherwise at least
IN_DELETE_SELF inotify event generation is broken.
Signed-off-by: NSergey Vlasov <vsu@altlinux.ru>
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

I noticed a strange behavior in a tmpfs file system the other day, while
building packages - occasionally, and seemingly at random, make decided to
rebuild a target. However, only on tmpfs.

A file would be created, and if checked, it had a sub-second timestamp.
However, after an utimes related call where sub-seconds should be set, they
were zeroed instead. In the case that a file was created, and utimes(...,NULL)
was used on it in the same second, the timestamp on the file moved backwards.

After some digging, I found that this was being caused by tmpfs not having a
time granularity set, thus inheriting the default 1 second granularity.

Hugh adds: yes, we missed tmpfs when the s_time_gran mods went into 2.6.11.
Unfortunately, the granularity of CURRENT_TIME, often used in filesystems,
does not match the default granularity set by alloc_super.  A few more such
discrepancies have been found, but this is the most important to fix now.
Signed-off-by: NRobin H. Johnson <robbat2@gentoo.org>
Acked-by: NAndi Kleen <ak@suse.de>
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

From: Christoph Lameter <clameter@sgi.com>

Looks like a comma was left from the conversion from a struct to an
assignment.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

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: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

From: Yasunori Goto <y-goto@jp.fujitsu.com>

If hot-added memory's address is smaller than old area, spanned_pages will
not be updated.  It must be fixed.

example) Old zone_start_pfn = 0x60000, and spanned_pages = 0x10000
         Added new memory's start_pfn = 0x50000, and end_pfn = 0x60000

  new spanned_pages will be still 0x10000 by old code.
  (It should be updated to 0x20000.) Because old_zone_end_pfn will be
  0x70000, and end_pfn smaller than it. So, spanned_pages will not be
  updated.

In current code, spanned_pages is updated only when end_pfn is updated.
But, it should be updated by subtraction between bigger end_pfn and new
zone_start_pfn.
Signed-off-by: NYasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: NDave Hansen <haveblue@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Andy added code to buddy allocator which does not require the zone's
endpoints to be aligned to MAX_ORDER.  An issue is that the buddy allocator
requires the node_mem_map's endpoints to be MAX_ORDER aligned.  Otherwise
__page_find_buddy could compute a buddy not in node_mem_map for partial
MAX_ORDER regions at zone's endpoints.  page_is_buddy will detect that
these pages at endpoints are not PG_buddy (they were zeroed out by bootmem
allocator and not part of zone).  Of course the negative here is we could
waste a little memory but the positive is eliminating all the old checks
for zone boundary conditions.

SPARSEMEM won't encounter this issue because of MAX_ORDER size constraint
when SPARSEMEM is configured.  ia64 VIRTUAL_MEM_MAP doesn't need the logic
either because the holes and endpoints are handled differently.  This
leaves checking alloc_remap and other arches which privately allocate for
node_mem_map.
Signed-off-by: NBob Picco <bob.picco@hp.com>
Acked-by: NMel Gorman <mel@csn.ul.ie>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Fix a couple of infrequently encountered 'sleeping function called from
invalid context' in the cpuset hooks in __alloc_pages.  Could sleep while
interrupts disabled.

The routine cpuset_zone_allowed() is called by code in mm/page_alloc.c
__alloc_pages() to determine if a zone is allowed in the current tasks
cpuset.  This routine can sleep, for certain GFP_KERNEL allocations, if the
zone is on a memory node not allowed in the current cpuset, but might be
allowed in a parent cpuset.

But we can't sleep in __alloc_pages() if in interrupt, nor if called for a
GFP_ATOMIC request (__GFP_WAIT not set in gfp_flags).

The rule was intended to be:
  Don't call cpuset_zone_allowed() if you can't sleep, unless you
  pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
  the code that might scan up ancestor cpusets and sleep.

This rule was being violated in a couple of places, due to a bogus change
made (by myself, pj) to __alloc_pages() as part of the November 2005 effort
to cleanup its logic, and also due to a later fix to constrain which swap
daemons were awoken.

The bogus change can be seen at:
  http://linux.derkeiler.com/Mailing-Lists/Kernel/2005-11/4691.html
  [PATCH 01/05] mm fix __alloc_pages cpuset ALLOC_* flags

This was first noticed on a tight memory system, in code that was disabling
interrupts and doing allocation requests with __GFP_WAIT not set, which
resulted in __might_sleep() writing complaints to the log "Debug: sleeping
function called ...", when the code in cpuset_zone_allowed() tried to take
the callback_sem cpuset semaphore.

We haven't seen a system hang on this 'might_sleep' yet, but we are at
decent risk of seeing it fairly soon, especially since the additional
cpuset_zone_allowed() check was added, conditioning wakeup_kswapd(), in
March 2006.

Special thanks to Dave Chinner, for figuring this out, and a tip of the hat
to Nick Piggin who warned me of this back in Nov 2005, before I was ready
to listen.
Signed-off-by: NPaul Jackson <pj@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

A bad calculation/loop in __section_nr() could result in incorrect section
information being put into sysfs memory entries.  This primarily impacts
memory add operations as the sysfs information is used while onlining new
memory.

Fix suggested by Dave Hansen.

Note that the bug may not be obvious from the patch.  It actually occurs in
the function's return statement:

	return (root_nr * SECTIONS_PER_ROOT) + (ms - root);

In the existing code, root_nr has already been multiplied by
SECTIONS_PER_ROOT.
Signed-off-by: NMike Kravetz <kravetz@us.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

With CONFIG_NUMA set, kmem_cache_destroy() may fail and say "Can't
free all objects."  The problem is caused by sequences such as the
following (suppose we are on a NUMA machine with two nodes, 0 and 1):

 * Allocate an object from cache on node 0.
 * Free the object on node 1.  The object is put into node 1's alien
   array_cache for node 0.
 * Call kmem_cache_destroy(), which ultimately ends up in __cache_shrink().
 * __cache_shrink() does drain_cpu_caches(), which loops through all nodes.
   For each node it drains the shared array_cache and then handles the
   alien array_cache for the other node.

However this means that node 0's shared array_cache will be drained,
and then node 1 will move the contents of its alien[0] array_cache
into that same shared array_cache.  node 0's shared array_cache is
never looked at again, so the objects left there will appear to be in
use when __cache_shrink() calls __node_shrink() for node 0.  So
__node_shrink() will return 1 and kmem_cache_destroy() will fail.

This patch fixes this by having drain_cpu_caches() do
drain_alien_cache() on every node before it does drain_array() on the
nodes' shared array_caches.

The problem was originally reported by Or Gerlitz <ogerlitz@voltaire.com>.
Signed-off-by: NRoland Dreier <rolandd@cisco.com>
Acked-by: NChristoph Lameter <clameter@sgi.com>
Acked-by: NPekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

slab_is_available() indicates slab based allocators are available for use.
SPARSEMEM code needs to know this as it can be called at various times
during the boot process.
Signed-off-by: NMike Kravetz <kravetz@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

As pointed out in http://bugzilla.kernel.org/show_bug.cgi?id=6490, this
function can experience overflows on 32-bit machines, causing our response to
changed values of min_free_kbytes to go whacky.

Fixing it efficiently is all too hard, so fix it with 64-bit math instead.

Cc: Ake Sandgren <ake.sandgren@hpc2n.umu.se>
Cc: Martin Bligh <mbligh@google.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Based on an older patch from  Mike Kravetz <kravetz@us.ibm.com>

We need to have a mem_map for high addresses in order to make fops->no_page
work on spufs mem and register files.  So far, we have used the
memory_present() function during early bootup, but that did not work when
CONFIG_NUMA was enabled.

We now use the __add_pages() function to add the mem_map when loading the
spufs module, which is a lot nicer.
Signed-off-by: NArnd Bergmann <arnd.bergmann@de.ibm.com>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

This patch fixes two bugs with the way sparsemem interacts with memory add.
They are:

- memory leak if memmap for section already exists

- calling alloc_bootmem_node() after boot

These bugs were discovered and a first cut at the fixes were provided by
Arnd Bergmann <arnd@arndb.de> and Joel Schopp <jschopp@us.ibm.com>.
Signed-off-by: NMike Kravetz <kravetz@us.ibm.com>
Signed-off-by: NJoel Schopp <jschopp@austin.ibm.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Currently we check PageDirty() in order to make the decision to swap out
the page.  However, the dirty information may be only be contained in the
ptes pointing to the page.  We need to first unmap the ptes before checking
for PageDirty().  If unmap is successful then the page count of the page
will also be decreased so that pageout() works properly.

This is a fix necessary for 2.6.17.  Without this fix we may migrate dirty
pages for filesystems without migration functions.  Filesystems may keep
pointers to dirty pages.  Migration of dirty pages can result in the
filesystem keeping pointers to freed pages.

Unmapping is currently not be separated out from removing all the
references to a page and moving the mapping.  Therefore try_to_unmap will
be called again in migrate_page() if the writeout is successful.  However,
it wont do anything since the ptes are already removed.

The coming updates to the page migration code will restructure the code
so that this is no longer necessary.
Signed-off-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

transfer_objects should only be called when all of the cpus in the
node are online.  CPU_DEAD notifier callback marks l3->shared to NULL.
Signed-off-by: NJacob Shin <jacob.shin@amd.com>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

find_get_pages_contig() will break out if we hit a hole in the page cache.
From Andrew Morton, small modifications and documentation by me.
Signed-off-by: NJens Axboe <axboe@suse.de>

Few of the notifier_chain_register() callers use __init in the definition
of notifier_call.  It is incorrect as the function definition should be
available after the initializations (they do not unregister them during
initializations).

This patch fixes all such usages to _not_ have the notifier_call __init
section.
Signed-off-by: NChandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

Basic problem: pages of a shared memory segment can only be migrated once.

In 2.6.16 through 2.6.17-rc1, shared memory mappings do not have a
migratepage address space op.  Therefore, migrate_pages() falls back to
default processing.  In this path, it will try to pageout() dirty pages.
Once a shared memory page has been migrated it becomes dirty, so
migrate_pages() will try to page it out.  However, because the page count
is 3 [cache + current + pte], pageout() will return PAGE_KEEP because
is_page_cache_freeable() returns false.  This will abort all subsequent
migrations.

This patch adds a migratepage address space op to shared memory segments to
avoid taking the default path.  We use the "migrate_page()" function
because it knows how to migrate dirty pages.  This allows shared memory
segment pages to migrate, subject to other conditions such as # pte's
referencing the page [page_mapcount(page)], when requested.

I think this is safe.  If we're migrating a shared memory page, then we
found the page via a page table, so it must be in memory.

Can be verified with memtoy and the shmem-mbind-test script, both
available at:  http://free.linux.hp.com/~lts/Tools/Signed-off-by: NLee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: NChristoph Lameter <clameter@sgi.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>