xacct_add_tsk() relies on do_exit()->update_hiwater_xxx() and uses
mm->hiwater_xxx directly, this leads to 2 problems:

- taskstats_user_cmd() can call fill_pid()->xacct_add_tsk() at any
  moment before the task exits, so we should check the current values of
  rss/vm anyway.

- do_exit()->update_hiwater_xxx() calls are racy.  An exiting thread can
  be preempted right before mm->hiwater_xxx = new_val, and another thread
  can use A_LOT of memory and exit in between.  When the first thread
  resumes it can be the last thread in the thread group, in that case we
  report the wrong hiwater_xxx values which do not take A_LOT into
  account.

Introduce get_mm_hiwater_rss() and get_mm_hiwater_vm() helpers and change
xacct_add_tsk() to use them.  The first helper will also be used by
rusage->ru_maxrss accounting.

Kill do_exit()->update_hiwater_xxx() calls.  Unless we are going to
decrease rss/vm there is no point to update mm->hiwater_xxx, and nobody
can look at this mm_struct when exit_mmap() actually unmaps the memory.
Signed-off-by: NOleg Nesterov <oleg@redhat.com>
Acked-by: NHugh Dickins <hugh@veritas.com>
Reviewed-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: NBalbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Frustratingly, gfp_t is really divided into two classes of flags.  One are
the context dependent ones (can we sleep?  can we enter filesystem?  block
subsystem?  should we use some extra reserves, etc.).  The other ones are
the type of memory required and depend on how the algorithm is implemented
rather than the point at which the memory is allocated (highmem?  dma
memory?  etc).

Some of the functions which allocate a page and add it to page cache take
a gfp_t, but sometimes those functions or their callers aren't really
doing the right thing: when allocating pagecache page, the memory type
should be mapping_gfp_mask(mapping).  When allocating radix tree nodes,
the memory type should be kernel mapped (not highmem) memory.  The gfp_t
argument should only really be needed for context dependent options.

This patch doesn't really solve that tangle in a nice way, but it does
attempt to fix a couple of bugs.

- find_or_create_page changes its radix-tree allocation to only include
  the main context dependent flags in order so the pagecache page may be
  allocated from arbitrary types of memory without affecting the
  radix-tree.  In practice, slab allocations don't come from highmem
  anyway, and radix-tree only uses slab allocations.  So there isn't a
  practical change (unless some fs uses GFP_DMA for pages).

- grab_cache_page_nowait() is changed to allocate radix-tree nodes with
  GFP_NOFS, because it is not supposed to reenter the filesystem.  This
  bug could cause lock recursion if a filesystem is not expecting the
  function to reenter the fs (as-per documentation).

Filesystems should be careful about exactly what semantics they want and
what they get when fiddling with gfp_t masks to allocate pagecache.  One
should be as liberal as possible with the type of memory that can be used,
and same for the the context specific flags.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

s_syncing livelock avoidance was breaking data integrity guarantee of
sys_sync, by allowing sys_sync to skip writing or waiting for superblocks
if there is a concurrent sys_sync happening.

This livelock avoidance is much less important now that we don't have the
get_super_to_sync() call after every sb that we sync.  This was replaced
by __put_super_and_need_restart.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix data integrity semantics required by sys_sync, by iterating over all
inodes and waiting for any writeback pages after the initial writeout.
Comments explain the exact problem.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove WB_SYNC_HOLD.  The primary motiviation is the design of my
anti-starvation code for fsync.  It requires taking an inode lock over the
sync operation, so we could run into lock ordering problems with multiple
inodes.  It is possible to take a single global lock to solve the ordering
problem, but then that would prevent a future nice implementation of "sync
multiple inodes" based on lock order via inode address.

Seems like a backward step to remove this, but actually it is busted
anyway: we can't use the inode lists for data integrity wait: an inode can
be taken off the dirty lists but still be under writeback.  In order to
satisfy data integrity semantics, we should wait for it to finish
writeback, but if we only search the dirty lists, we'll miss it.

It would be possible to have a "writeback" list, for sys_sync, I suppose.
But why complicate things by prematurely optimise?  For unmounting, we
could avoid the "livelock avoidance" code, which would be easier, but
again premature IMO.

Fixing the existing data integrity problem will come next.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

WB_SYNC_HOLD is going to be zapped so we should not use it. Use
%WB_SYNC_NONE instead. Here is what akpm said:

"I think I'll just switch that to WB_SYNC_NONE.  The `wait==0' mode is
just an advisory thing to help the fs shove lots of data into the
queues.  If some gets missed then it'll be picked up on the second
->sync_fs call, with wait==1."

Thanks to Randy Dunlap for catching this.
Signed-off-by: NArtem Bityutskiy <Artem.Bityutskiy@nokia.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Direct IO can invalidate and sync a lot of pagecache pages in the mapping.
 A 4K direct IO will actually try to sync and/or invalidate the pagecache
of the entire file, for example (which might be many GB or TB large).

Improve this by doing range syncs.  Also, memory no longer has to be
unmapped to catch the dirty bits for syncing, as dirty bits would remain
coherent due to dirty mmap accounting.

This fixes the immediate DM deadlocks when doing direct IO reads to block
device with a mounted filesystem, if only by papering over the problem
somewhat rather than addressing the fsync starvation cases.
Signed-off-by: NNick Piggin <npiggin@suse.de>
Reviewed-by: NJeff Moyer <jmoyer@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix a little of the coding style in mm/mmap.c

[akpm@linux-foundation.org: cleanup]
Signed-off-by: NZhenwenXu <helight.xu@gmail.com>
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

tiny-shmem shares most of its 130 lines of code with shmem and tends to
break when particular bits of shmem get modified.  Unifying saves code and
makes keeping these two in sync much easier.

before:
  14367	    392	     24	  14783	   39bf	mm/shmem.o
    396      72       8     476	    1dc	mm/tiny-shmem.o

after:
  14367	    392	     24	  14783	   39bf	mm/shmem.o
    412	     72       8     492	    1ec	mm/shmem.o tiny
Signed-off-by: NMatt Mackall <mpm@selenic.com>
Acked-by: NHugh Dickins <hugh@veritas.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Signed-off-by: NFranck Bui-Huu <fbuihuu@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The initial implementation of checking TIF_MEMDIE covers the cases of OOM
killing.  If the process has been OOM killed, the TIF_MEMDIE is set and it
return immediately.  This patch includes:

1.  add the case that the SIGKILL is sent by user processes.  The
   process can try to get_user_pages() unlimited memory even if a user
   process has sent a SIGKILL to it(maybe a monitor find the process
   exceed its memory limit and try to kill it).  In the old
   implementation, the SIGKILL won't be handled until the get_user_pages()
   returns.

2.  change the return value to be ERESTARTSYS.  It makes no sense to
   return ENOMEM if the get_user_pages returned by getting a SIGKILL
   signal.  Considering the general convention for a system call
   interrupted by a signal is ERESTARTNOSYS, so the current return value
   is consistant to that.

Lee:

An unfortunate side effect of "make-get_user_pages-interruptible" is that
it prevents a SIGKILL'd task from munlock-ing pages that it had mlocked,
resulting in freeing of mlocked pages.  Freeing of mlocked pages, in
itself, is not so bad.  We just count them now--altho' I had hoped to
remove this stat and add PG_MLOCKED to the free pages flags check.

However, consider pages in shared libraries mapped by more than one task
that a task mlocked--e.g., via mlockall().  If the task that mlocked the
pages exits via SIGKILL, these pages would be left mlocked and
unevictable.

Proposed fix:

Add another GUP flag to ignore sigkill when calling get_user_pages from
munlock()--similar to Kosaki Motohiro's 'IGNORE_VMA_PERMISSIONS flag for
the same purpose.  We are not actually allocating memory in this case,
which "make-get_user_pages-interruptible" intends to avoid.  We're just
munlocking pages that are already resident and mapped, and we're reusing
get_user_pages() to access those pages.

??  Maybe we should combine 'IGNORE_VMA_PERMISSIONS and '_IGNORE_SIGKILL
into a single flag: GUP_FLAGS_MUNLOCK ???

[Lee.Schermerhorn@hp.com: ignore sigkill in get_user_pages during munlock]
Signed-off-by: NPaul Menage <menage@google.com>
Signed-off-by: NYing Han <yinghan@google.com>
Reviewed-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: NPekka Enberg <penberg@cs.helsinki.fi>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Rohit Seth <rohitseth@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: NLee Schermerhorn <lee.schermerhorn@hp.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

unsigned long ret cannot be negative, but ret can get -EFAULT.
Signed-off-by: NRoel Kluin <roel.kluin@gmail.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Adam Litke <agl@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Cc: Ken Chen <kenchen@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

These three statements manipulate local variables and do not need the lock
coverage.

Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Rik van Riel <riel@redhat.com
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

bad_page() and rmap Eeek messages have said KERN_EMERG for a few years,
which I've followed in print_bad_pte().  These are serious system errors,
on a par with BUGs, but they're not quite emergencies, and we do our best
to carry on: say KERN_ALERT "BUG: " like the x86 oops does.

And remove the "Trying to fix it up, but a reboot is needed" line: it's
not untrue, but I hope the KERN_ALERT "BUG: " conveys as much.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

print_bad_pte() and bad_page() might each need ratelimiting - especially
for their dump_stacks, almost never of interest, yet not quite
dispensible.  Correlating corruption across neighbouring entries can be
very helpful, so allow a burst of 60 reports before keeping quiet for the
remainder of that minute (or allow a steady drip of one report per
second).
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove page_remove_rmap()'s vma arg, which was only for the Eeek message.
And remove the BUG_ON(page_mapcount(page) == 0) from CONFIG_DEBUG_VM's
page_dup_rmap(): we're trying to be more resilient about that than BUGs.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Complete zap_pte_range()'s coverage of bad pagetable entries by calling
print_bad_pte() on a pte_file in a linear vma and on a bad swap entry.
That needs free_swap_and_cache() to tell it, which will also have shown
one of those "swap_free" errors (but with much less information).

Similar checks in fork's copy_one_pte()?  No, that would be more noisy
than helpful: we'll see them when parent and child exec or exit.

Where do_nonlinear_fault() calls print_bad_pte(): omit !VM_CAN_NONLINEAR
case, that could only be a bug in sys_remap_file_pages(), not a bad pte.
VM_FAULT_OOM rather than VM_FAULT_SIGBUS?  Well, okay, that is consistent
with what happens if do_swap_page() operates a bad swap entry; but don't
we have patches to be more careful about killing when VM_FAULT_OOM?
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

print_bad_pte() is so far being called only when zap_pte_range() finds
negative page_mapcount, or there's a fault on a pte_file where it does not
belong.  That's weak coverage when we suspect pagetable corruption.

Originally, it was called when vm_normal_page() found an invalid pfn: but
pfn_valid is expensive on some architectures and configurations, so 2.6.24
put that under CONFIG_DEBUG_VM (which doesn't help in the field), then
2.6.26 replaced it by a VM_BUG_ON (likewise).

Reinstate the print_bad_pte() in vm_normal_page(), but use a cheaper test
than pfn_valid(): memmap_init_zone() (used in bootup and hotplug) keep a
__read_mostly note of the highest_memmap_pfn, vm_normal_page() then check
pfn against that.  We could call this pfn_plausible() or pfn_sane(), but I
doubt we'll need it elsewhere: of course it's not reliable, but gives much
stronger pagetable validation on many boxes.

Also use print_bad_pte() when the pte_special bit is found outside a
VM_PFNMAP or VM_MIXEDMAP area, instead of VM_BUG_ON.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Now that bad pages are kept out of circulation, there is no need for the
infamous page_remove_rmap() BUG() - once that page is freed, its negative
mapcount will issue a "Bad page state" message and the page won't be
freed.  Removing the BUG() allows more info, on subsequent pages, to be
gathered.

We do have more info about the page at this point than bad_page() can know
- notably, what the pmd is, which might pinpoint something like low 64kB
corruption - but page_remove_rmap() isn't given the address to find that.

In practice, there is only one call to page_remove_rmap() which has ever
reported anything, that from zap_pte_range() (usually on exit, sometimes
on munmap).  It has all the info, so remove page_remove_rmap()'s "Eeek"
message and leave it all to zap_pte_range().

mm/memory.c already has a hardly used print_bad_pte() function, showing
some of the appropriate info: extend it to show what we want for the rmap
case: pte info, page info (when there is a page) and vma info to compare.
zap_pte_range() already knows the pmd, but print_bad_pte() is easier to
use if it works that out for itself.

Some of this info is also shown in bad_page()'s "Bad page state" message.
Keep them separate, but adjust them to match each other as far as
possible.  Say "Bad page map" in print_bad_pte(), and add a TAINT_BAD_PAGE
there too.

print_bad_pte() show current->comm unconditionally (though it should get
repeated in the usually irrelevant stack trace): sorry, I misled Nick
Piggin to make it conditional on vm_mm == current->mm, but current->mm is
already NULL in the exit case.  Usually current->comm is good, though
exceptionally it may not be that of the mm (when "swapoff" for example).
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Until now the bad_page() checkers have special-cased PageReserved, keeping
those pages out of circulation thereafter.  Now extend the special case to
all: we want to keep ANY page with bad state out of circulation - the
"free" page may well be in use by something.

Leave the bad state of those pages untouched, for examination by
debuggers; except for PageBuddy - leaving that set would risk bringing the
page back.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Simplify the PAGE_FLAGS checking and clearing when freeing and allocating
a page: check the same flags as before when freeing, clear ALL the flags
(unless PageReserved) when freeing, check ALL flags off when allocating.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

In case of error extending write may have instantiated a few blocks
outside i_size.  We need to trim these blocks.  We have to do it
*regardless* to blocksize.  At least ext2, ext3 and reiserfs interpret
(i_size < biggest block) condition as error.  Fsck will complain about
wrong i_size.  Then fsck will fix the error by changing i_size according
to the biggest block.  This is bad because this blocks contain garbage
from previous write attempt.  And result in data corruption.

####TESTCASE_BEGIN
$touch /mnt/test/BIG_FILE
## at this moment /mnt/test/BIG_FILE size and blocks equal to zero
open("/mnt/test/BIG_FILE", O_WRONLY|O_CREAT|O_DIRECT, 0666) = 3
write(3, "aaaaaaaaaaaa"..., 104857600) = -1 ENOSPC (No space left on device)
## size and block sould't be changed because write op failed.
$stat /mnt/test/BIG_FILE
File: `/mnt/test/BIG_FILE'
Size: 0 Blocks: 110896 IO Block: 1024 regular empty file
<<<<<<<<^^^^^^^^^^^^^^^^^^^^^^^^^^^^^file size is less than biggest block idx
Device: fe07h/65031d Inode: 14 Links: 1
Access: (0644/-rw-r--r--) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2007-01-24 20:03:38.000000000 +0300
Modify: 2007-01-24 20:03:38.000000000 +0300
Change: 2007-01-24 20:03:39.000000000 +0300

#fsck.ext3 -f /dev/VG/test
e2fsck 1.39 (29-May-2006)
Pass 1: Checking inodes, blocks, and sizes
Inode 14, i_size is 0, should be 56556544. Fix<y>? yes
Pass 2: Checking directory structure
....
#####TESTCASE_ENDdiff --git a/fs/direct-io.c b/fs/direct-io.c
index af0558d..4e88bea 100644

[akpm@linux-foundation.org: use i_size_read()]
Signed-off-by: NDmitri Monakhov <dmonakhov@openvz.org>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Badari Pulavarty <pbadari@us.ibm.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

zone_is_near_oom() is unused.
Signed-off-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The vmscan bail out patch move nr_reclaimed variable to struct
scan_control.  Unfortunately, indirect access can easily happen cache
miss.

if heavy memory pressure happend, that's ok.
cache miss already plenty. it is not observable.

but, if memory pressure is lite, performance degression is obserbable.

I compared following three pattern (it was mesured 10 times each)

hackbench 125 process 3000
hackbench 130 process 3000
hackbench 135 process 3000

            2.6.28-rc6                       bail-out

	125	130	135		125	130	135
      ==============================================================
	71.866	75.86	81.274		93.414	73.254	193.382
	74.145	78.295	77.27		74.897	75.021	80.17
	70.305	77.643	75.855		70.134	77.571	79.896
	74.288	73.986	75.955		77.222	78.48	80.619
	72.029	79.947	78.312		75.128	82.172	79.708
	71.499	77.615	77.042		74.177	76.532	77.306
	76.188	74.471	83.562		73.839	72.43	79.833
	73.236	75.606	78.743		76.001	76.557	82.726
	69.427	77.271	76.691		76.236	79.371	103.189
	72.473	76.978	80.643		69.128	78.932	75.736

avg	72.545	76.767	78.534		76.017	77.03	93.256
std	1.89	1.71	2.41		6.29	2.79	34.16
min	69.427	73.986	75.855		69.128	72.43	75.736
max	76.188	79.947	83.562		93.414	82.172	193.382

about 4-5% degression.

Then, this patch introduces a temporary local variable.

result:

            2.6.28-rc6                       this patch

num	125	130	135		125	130	135
      ==============================================================
	71.866	75.86	81.274		67.302	68.269	77.161
	74.145	78.295	77.27   	72.616	72.712	79.06
	70.305	77.643	75.855  	72.475	75.712	77.735
	74.288	73.986	75.955  	69.229	73.062	78.814
	72.029	79.947	78.312  	71.551	74.392	78.564
	71.499	77.615	77.042  	69.227	74.31	78.837
	76.188	74.471	83.562  	70.759	75.256	76.6
	73.236	75.606	78.743  	69.966	76.001	78.464
	69.427	77.271	76.691  	69.068	75.218	80.321
	72.473	76.978	80.643  	72.057	77.151	79.068

avg	72.545	76.767	78.534 		70.425	74.2083	78.462
std 	1.89	1.71	2.41    	1.66	2.34	1.00
min 	69.427	73.986	75.855  	67.302	68.269	76.6
max 	76.188	79.947	83.562  	72.616	77.151	80.321

OK. the degression is disappeared.
Signed-off-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: NRik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When the VM is under pressure, it can happen that several direct reclaim
processes are in the pageout code simultaneously.  It also happens that
the reclaiming processes run into mostly referenced, mapped and dirty
pages in the first round.

This results in multiple direct reclaim processes having a lower
pageout priority, which corresponds to a higher target of pages to
scan.

This in turn can result in each direct reclaim process freeing
many pages.  Together, they can end up freeing way too many pages.

This kicks useful data out of memory (in some cases more than half
of all memory is swapped out).  It also impacts performance by
keeping tasks stuck in the pageout code for too long.

A 30% improvement in hackbench has been observed with this patch.

The fix is relatively simple: in shrink_zone() we can check how many
pages we have already freed, direct reclaim tasks break out of the
scanning loop if they have already freed enough pages and have reached
a lower priority level.

We do not break out of shrink_zone() when priority == DEF_PRIORITY,
to ensure that equal pressure is applied to every zone in the common
case.

However, in order to do this we do need to know how many pages we already
freed, so move nr_reclaimed into scan_control.

akpm: a historical interlude...

We tried this in 2004:

:commit e468e46a9bea3297011d5918663ce6d19094cf87
:Author: akpm <akpm>
:Date:   Thu Jun 24 15:53:52 2004 +0000
:
:[PATCH] vmscan.c: dont reclaim too many pages
:
:    The shrink_zone() logic can, under some circumstances, cause far too many
:    pages to be reclaimed.  Say, we're scanning at high priority and suddenly hit
:    a large number of reclaimable pages on the LRU.
:    Change things so we bale out when SWAP_CLUSTER_MAX pages have been reclaimed.

And we reverted it in 2006:

:commit 210fe530
:Author: Andrew Morton <akpm@osdl.org>
:Date:   Fri Jan 6 00:11:14 2006 -0800
:
:    [PATCH] vmscan: balancing fix
:
:    Revert a patch which went into 2.6.8-rc1.  The changelog for that patch was:
:
:      The shrink_zone() logic can, under some circumstances, cause far too many
:      pages to be reclaimed.  Say, we're scanning at high priority and suddenly
:      hit a large number of reclaimable pages on the LRU.
:
:      Change things so we bale out when SWAP_CLUSTER_MAX pages have been
:      reclaimed.
:
:    Problem is, this change caused significant imbalance in inter-zone scan
:    balancing by truncating scans of larger zones.
:
:    Suppose, for example, ZONE_HIGHMEM is 10x the size of ZONE_NORMAL.  The zone
:    balancing algorithm would require that if we're scanning 100 pages of
:    ZONE_HIGHMEM, we should scan 10 pages of ZONE_NORMAL.  But this logic will
:    cause the scanning of ZONE_HIGHMEM to bale out after only 32 pages are
:    reclaimed.  Thus effectively causing smaller zones to be scanned relatively
:    harder than large ones.
:
:    Now I need to remember what the workload was which caused me to write this
:    patch originally, then fix it up in a different way...

And we haven't demonstrated that whatever problem caused that reversion is
not being reintroduced by this change in 2008.
Signed-off-by: NRik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix the following sparse warnings:

  mm/hugetlb.c:375:3: warning: returning void-valued expression
  mm/hugetlb.c:408:3: warning: returning void-valued expression
Signed-off-by: NHannes Eder <hannes@hanneseder.net>
Acked-by: NNishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove the srandom32((u32)get_seconds()) from non-rotational swapon:
there's been a coincidental discussion of earlier randomization, assume
that goes ahead, let swapon be a client rather than stirring for itself.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Donjun Shin <djshin90@gmail.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Joern Engel <joern@logfs.org>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Tejun Heo <teheo@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Change pgoff_t nr_blocks in discard_swap() and discard_swap_cluster() to
sector_t: given the constraints on swap offsets (in particular, the 5 bits
of swap type accommodated in the same unsigned long), pgoff_t was actually
safe as is, but it certainly looked worrying when shifted left.

[akpm@linux-foundation.org: fix shift overflow]
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Joern Engel <joern@logfs.org>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Donjun Shin <djshin90@gmail.com>
Cc: Tejun Heo <teheo@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Though attempting to find free clusters (Andrea), swap allocation has
always restarted its searches from the beginning of the swap area (sct),
to reduce seek times between swap pages, by not scattering them all over
the partition.

But on a solidstate swap device, seeks are cheap, and block remapping to
level the wear may be limited by zones: in that case it's better to cycle
around the whole partition.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Joern Engel <joern@logfs.org>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Donjun Shin <djshin90@gmail.com>
Cc: Tejun Heo <teheo@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Swap allocation has always started from the beginning of the swap area;
but if we're dealing with a solidstate swap device which can only remap
blocks within limited zones, that would sooner wear out the first zone.

Therefore sys_swapon() test whether blk_queue is non-rotational, and if so
randomize the cluster_next starting position for allocation.

If blk_queue is nonrot, note SWP_SOLIDSTATE for later use, and report it
with an "SS" at the right end of the kernel's "Adding ...  swap" message
(so that if it's both nonrot and discardable, "SSD" will be shown there).
Perhaps something should be shown in /proc/swaps (swapon -s), but we have
to be more cautious before making any addition to that format.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Joern Engel <joern@logfs.org>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Donjun Shin <djshin90@gmail.com>
Cc: Tejun Heo <teheo@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When scan_swap_map() finds a free cluster of swap pages to allocate,
discard the old contents of the cluster if the device supports discard.
But don't bother when swap is so fragmented that we allocate single pages.

Be careful about racing allocations made while we're scanning for a
cluster; and hold up allocations made while we're discarding.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Joern Engel <joern@logfs.org>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Donjun Shin <djshin90@gmail.com>
Cc: Tejun Heo <teheo@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When adding swap, all the old data on swap can be forgotten: sys_swapon()
discard all but the header page of the swap partition (or every extent but
the header of the swap file), to give a solidstate swap device the
opportunity to optimize its wear-levelling.

If that succeeds, note SWP_DISCARDABLE for later use, and report it with a
"D" at the right end of the kernel's "Adding ...  swap" message.  Perhaps
something should be shown in /proc/swaps (swapon -s), but we have to be
more cautious before making any addition to that format.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Matthew Wilcox <matthew@wil.cx>
Cc: Joern Engel <joern@logfs.org>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Donjun Shin <djshin90@gmail.com>
Cc: Tejun Heo <teheo@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Before making functional changes, rearrange scan_swap_map() to simplify
subsequent diffs.  Actually, there is one functional change in there:
leave cluster_nr negative while scanning for a new cluster - resetting it
early increased the likelihood that when we have difficulty finding a free
cluster, another task may come in and try doing exactly the same - just a
waste of cpu.

Before making functional changes, rearrange struct swap_info_struct
slightly: flags will be needed as an unsigned long (for wait_on_bit), next
is a good int to pair with prio, old_block_size is uninteresting so shift
it to the end.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

The kernel has not supported v0 SWAP-SPACE since 2.5.22: I think we can
now safely drop its "version 0 swap is no longer supported" message - just
say "Unable to find swap-space signature" as usual.  This removes one
level of indentation from a stretch of sys_swapon().

I'd have liked to be specific, saying "Unable to find SWAPSPACE2
signature", but it's just too confusing that the version 1 signature shows
the number 2.

Irrelevant nearby cleanup: kmap(page) already gives page_address(page).
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove trailing whitespace from swapfile.c, and odd swap_show() alignment.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Remove the SWP_ACTIVE mask: it just obscures the SWP_WRITEOK flag.
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

sys_swapon()'s swapfilesize (better renamed swapfilepages) is declared as
an int, but should be an unsigned long like the maxpages it's compared
against: on 64-bit (with 4kB pages) a swapfile of 2^44 bytes was rejected
with "Swap area shorter than signature indicates".
Signed-off-by: NHugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Sparse output following warnings.

mm/vmalloc.c:1436:6: warning: symbol 'vread' was not declared. Should it be static?
mm/vmalloc.c:1474:6: warning: symbol 'vwrite' was not declared. Should it be static?

However, it is used by /dev/kmem. fixed here.
Signed-off-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Sparse output following warning.

mm/page_alloc.c:4301:6: warning: symbol 'setup_per_zone_inactive_ratio' was not declared. Should it be static?

cleanup here.
Signed-off-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

sparse output following warning

	mm/vmscan.c:2507:6: warning: symbol 'scan_zone_unevictable_pages' was not declared. Should it be static?

cleanup here.
Signed-off-by: NKOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>