vmscan: shrink_active_list(): reduce lru_lock hold time

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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: KERN_ALERT BUG instead of KERN_EMERG

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: ratelimit print_bad_pte and bad_page

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: remove vma from page_remove_rmap

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: zap print_bad_pte on swap and file

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: vm_normal_page use print_bad_pte

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: replace page_remove_rmap Eeek and BUG

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: keep any bad page out of circulation

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

badpage: simplify page_alloc flag check+clear

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: truncate blocks outside i_size after O_DIRECT write error

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: Dmitri 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: kill zone_is_near_oom()

zone_is_near_oom() is unused.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: improve reclaim throughput to bail out patch

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: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

vmscan: bail out of direct reclaim after swap_cluster_max pages

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 210fe530305ee50cd889fe9250168228b2994f32
: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: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

hugetlb: fix sparse warnings

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: Hannes Eder <hannes@hanneseder.net>
Acked-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: let others seed random

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: change discard pgoff_t to sector_t

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: swap allocation cycle if nonrot

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: swapon randomize if nonrot

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: swap allocation use discard

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: swapon use discard (trim)

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: Hugh 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: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: rearrange scan and swap_info

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: Hugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: remove v0 SWAP-SPACE message

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: Hugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: remove surplus whitespace

Remove trailing whitespace from swapfile.c, and odd swap_show() alignment.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: remove SWP_ACTIVE mask

Remove the SWP_ACTIVE mask: it just obscures the SWP_WRITEOK flag.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

swapfile: swapon needs larger size type

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: Hugh Dickins <hugh@veritas.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make vread() and vwrite() declaration

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: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make setup_per_zone_inactive_ratio() static

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: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make scan_zone_unevictable_pages() static

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: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make scan_all_zones_unevictable_pages() static

sparse output following warning.

mm/vmscan.c:2549:6: warning: symbol 'scan_all_zones_unevictable_pages' was not declared. Should it be static?

cleanup here.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make mem_cgroup_resize_limit() static

Sparse output following warnings.

mm/memcontrol.c:782:5: warning: symbol 'mem_cgroup_resize_limit' was not
declared.  Should it be static?

cleanup here.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make maddr __iomem

sparse output following warnings.

mm/memory.c:2936:8: warning: incorrect type in assignment (different address spaces)
mm/memory.c:2936:8:    expected void *maddr
mm/memory.c:2936:8:    got void [noderef] <asn:2>

cleanup here.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make init_section_page_cgroup() static

Sparse output following warning.

mm/page_cgroup.c:100:15: warning: symbol 'init_section_page_cgroup' was
not declared.  Should it be static?

cleanup here.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

memcg: reclaim shouldn't change zone->recent_rotated statistics

memcg reclaim shouldn't change zone->recent_rotated statistics.  If
memcgroup reclaim changes zone statistics, global reclaim can get a bit
confused.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: optimize get_scan_ratio for no swap

Rik suggests a simplified get_scan_ratio() for !CONFIG_SWAP.  Yes, the gcc
optimizer gives us that, when nr_swap_pages is #defined as 0L.  Move usual
declaration to swapfile.c: it never belonged in page_alloc.c.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add add_to_swap stub

If we add a failing stub for add_to_swap(), then we can remove the #ifdef
CONFIG_SWAP from mm/vmscan.c.

This was intended as a source cleanup, but looking more closely, it turns
out that the !CONFIG_SWAP case was going to keep_locked for an anonymous
page, whereas now it goes to the more suitable activate_locked, like the
CONFIG_SWAP nr_swap_pages 0 case.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove gfp_mask from add_to_swap

Remove gfp_mask argument from add_to_swap(): it's misleading because its
only caller, shrink_page_list(), is not atomic at that point; and in due
course (implementing discard) we'll sometimes want to allocate some memory
with GFP_NOIO (as is used in swap_writepage) when allocating swap.

No change to the gfp_mask passed down to add_to_swap_cache(): still use
__GFP_HIGH without __GFP_WAIT (with nomemalloc and nowarn as before):
though it's not obvious if that's the best combination to ask for here.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove try_to_munlock from vmscan

An unfortunate feature of the Unevictable LRU work was that reclaiming an
anonymous page involved an extra scan through the anon_vma: to check that
the page is evictable before allocating swap, because the swap could not
be freed reliably soon afterwards.

Now try_to_free_swap() has replaced remove_exclusive_swap_page(), that's
not an issue any more: remove try_to_munlock() call from
shrink_page_list(), leaving it to try_to_munmap() to discover if the page
is one to be culled to the unevictable list - in which case then
try_to_free_swap().

Update unevictable-lru.txt to remove comments on the try_to_munlock() in
shrink_page_list(), and shorten some lines over 80 columns.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: try_to_unuse check removing right swap

There's a possible race in try_to_unuse() which Nick Piggin led me to two
years ago.  Where it does lock_page() after read_swap_cache_async(), what
if another task removed that page from swapcache just before we locked it?

It would sail though the (*swap_map > 1) tests doing nothing (because it
could not have been removed from swapcache before its swap references were
gone), until it reaches the delete_from_swap_cache(page) near the bottom.

Now imagine that this page has been allocated to swap on a different swap
area while we dropped page lock (perhaps at the top, perhaps in unuse_mm):
we could wrongly remove from swap cache before the page has been written
to swap, so a subsequent do_swap_page() would read in stale data from
swap.

I think this case could not happen before: remove_exclusive_swap_page()
refused while page count was raised.  But now with reuse_swap_page() and
try_to_free_swap() removing from swap cache without minding page count, I
think it could happen - the previous patch argued that it was safe because
try_to_unuse() already ignored page count, but overlooked that it might be
breaking the assumptions in try_to_unuse() itself.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: try_to_free_swap replaces remove_exclusive_swap_page

remove_exclusive_swap_page(): its problem is in living up to its name.

It doesn't matter if someone else has a reference to the page (raised
page_count); it doesn't matter if the page is mapped into userspace
(raised page_mapcount - though that hints it may be worth keeping the
swap): all that matters is that there be no more references to the swap
(and no writeback in progress).

swapoff (try_to_unuse) has been removing pages from swapcache for years,
with no concern for page count or page mapcount, and we used to have a
comment in lookup_swap_cache() recognizing that: if you go for a page of
swapcache, you'll get the right page, but it could have been removed from
swapcache by the time you get page lock.

So, give up asking for exclusivity: get rid of
remove_exclusive_swap_page(), and remove_exclusive_swap_page_ref() and
remove_exclusive_swap_page_count() which were spawned for the recent LRU
work: replace them by the simpler try_to_free_swap() which just checks
page_swapcount().

Similarly, remove the page_count limitation from free_swap_and_count(),
but assume that it's worth holding on to the swap if page is mapped and
swap nowhere near full.  Add a vm_swap_full() test in free_swap_cache()?
It would be consistent, but I think we probably have enough for now.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: reuse_swap_page replaces can_share_swap_page

A good place to free up old swap is where do_wp_page(), or do_swap_page(),
is about to redirty the page: the data on disk is then stale and won't be
read again; and if we do decide to write the page out later, using the
previous swap location makes an unnecessary disk seek very likely.

So give can_share_swap_page() the side-effect of delete_from_swap_cache()
when it safely can.  And can_share_swap_page() was always a misleading
name, the more so if it has a side-effect: rename it reuse_swap_page().

Irrelevant cleanup nearby: remove swap_token_default_timeout definition
from swap.h: it's used nowhere.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: wp lock page before deciding cow

An application may rely on get_user_pages() to give it pages writable from
userspace and shared with a driver, GUP breaking COW if necessary.  It may
mprotect() the pages' writability, off and on, from time to time.

Normally this works fine (so long as the app does not fork); but just
occasionally, under memory pressure, a readonly pte in a newly writable
area is COWed unnecessarily, breaking the link with the driver: because
do_wp_page() does trylock_page, and falls back to COW whenever that fails.

For reliable behaviour in the unshared case, when the trylock_page fails,
now unlock pagetable, lock page and relock pagetable, before deciding
whether Copy-On-Write is really necessary.

Reported-by: Zhou Yingchao
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: gup persist for write permission

do_wp_page()'s VM_FAULT_WRITE return value tells __get_user_pages() that
COW has been done if necessary, though it may be leaving the pte without
write permission - for the odd case of forced writing to a readonly vma
for ptrace.  At present GUP then retries the follow_page() without asking
for write permission, to escape an endless loop when forced.

But an application may be relying on GUP to guarantee a writable page
which won't be COWed again when written from userspace, whereas a race
here might leave a readonly pte in place?  Change the VM_FAULT_WRITE
handling to ask follow_page() for write permission again, except in that
odd case of forced writing to a readonly vma.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add dirty_background_bytes and dirty_bytes sysctls

This change introduces two new sysctls to /proc/sys/vm:
dirty_background_bytes and dirty_bytes.

dirty_background_bytes is the counterpart to dirty_background_ratio and
dirty_bytes is the counterpart to dirty_ratio.

With growing memory capacities of individual machines, it's no longer
sufficient to specify dirty thresholds as a percentage of the amount of
dirtyable memory over the entire system.

dirty_background_bytes and dirty_bytes specify quantities of memory, in
bytes, that represent the dirty limits for the entire system.  If either
of these values is set, its value represents the amount of dirty memory
that is needed to commence either background or direct writeback.

When a `bytes' or `ratio' file is written, its counterpart becomes a
function of the written value.  For example, if dirty_bytes is written to
be 8096, 8K of memory is required to commence direct writeback.
dirty_ratio is then functionally equivalent to 8K / the amount of
dirtyable memory:

dirtyable_memory = free pages + mapped pages + file cache

dirty_background_bytes = dirty_background_ratio * dirtyable_memory
-or-
dirty_background_ratio = dirty_background_bytes / dirtyable_memory

AND

dirty_bytes = dirty_ratio * dirtyable_memory
-or-
dirty_ratio = dirty_bytes / dirtyable_memory

Only one of dirty_background_bytes and dirty_background_ratio may be
specified at a time, and only one of dirty_bytes and dirty_ratio may be
specified.  When one sysctl is written, the other appears as 0 when read.

The `bytes' files operate on a page size granularity since dirty limits
are compared with ZVC values, which are in page units.

Prior to this change, the minimum dirty_ratio was 5 as implemented by
get_dirty_limits() although /proc/sys/vm/dirty_ratio would show any user
written value between 0 and 100.  This restriction is maintained, but
dirty_bytes has a lower limit of only one page.

Also prior to this change, the dirty_background_ratio could not equal or
exceed dirty_ratio.  This restriction is maintained in addition to
restricting dirty_background_bytes.  If either background threshold equals
or exceeds that of the dirty threshold, it is implicitly set to half the
dirty threshold.

Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Andrea Righi <righi.andrea@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: change dirty limit type specifiers to unsigned long

The background dirty and dirty limits are better defined with type
specifiers of unsigned long since negative writeback thresholds are not
possible.

These values, as returned by get_dirty_limits(), are normally compared
with ZVC values to determine whether writeback shall commence or be
throttled.  Such page counts cannot be negative, so declaring the page
limits as signed is unnecessary.

Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Andrea Righi <righi.andrea@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/page_alloc.c: eliminate NULL test and memset after alloc_bootmem

As noted by Akinobu Mita in patch b1fceac2b9e04d278316b2faddf276015fc06e3b,
alloc_bootmem and related functions never return NULL and always return a
zeroed region of memory.  Thus a NULL test or memset after calls to these
functions is unnecessary.

This was fixed using the following semantic patch.
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression E;
statement S;
@@

E = \(alloc_bootmem\|alloc_bootmem_low\|alloc_bootmem_pages\|alloc_bootmem_low_pages\|alloc_bootmem_node\|alloc_bootmem_low_pages_node\|alloc_bootmem_pages_node\)(...)
... when != E
(
- BUG_ON (E == NULL);
|
- if (E == NULL) S
)

@@
expression E,E1;
@@

E = \(alloc_bootmem\|alloc_bootmem_low\|alloc_bootmem_pages\|alloc_bootmem_low_pages\|alloc_bootmem_node\|alloc_bootmem_low_pages_node\|alloc_bootmem_pages_node\)(...)
... when != E
- memset(E,0,E1);
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: further cleanup page_add_new_anon_rmap

Moving lru_cache_add_active_or_unevictable() into page_add_new_anon_rmap()
was good but stupid: we can and should SetPageSwapBacked() there too; and
we know for sure that this anonymous, swap-backed page is not file cache.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: make page_lock_anon_vma() static

page_lock_anon_vma() and page_unlock_anon_vma() were made available to
show_page_path() in vmscan.c; but now that has been removed, make them
static in rmap.c again, they're better kept private if possible.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add_active_or_unevictable into rmap

lru_cache_add_active_or_unevictable() and page_add_new_anon_rmap() always
appear together.  Save some symbol table space and some jumping around by
removing lru_cache_add_active_or_unevictable(), folding its code into
page_add_new_anon_rmap(): like how we add file pages to lru just after
adding them to page cache.

Remove the nearby "TODO: is this safe?" comments (yes, it is safe), and
change page_add_new_anon_rmap()'s address BUG_ON to VM_BUG_ON as
originally intended.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: replace some BUG_ONs by VM_BUG_ONs

The swap code is over-provisioned with BUG_ONs on assorted page flags,
mostly dating back to 2.3.  They're good documentation, and guard against
developer error, but a waste of space on most systems: change them to
VM_BUG_ONs, conditional on CONFIG_DEBUG_VM.  Just delete the PagePrivate
ones: they're later, from 2.5.69, but even less interesting now.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Reviewed-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add Set,ClearPageSwapCache stubs

If we add NOOP stubs for SetPageSwapCache() and ClearPageSwapCache(), then
we can remove the #ifdef CONFIG_SWAPs from mm/migrate.c.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove GFP_HIGHUSER_PAGECACHE

GFP_HIGHUSER_PAGECACHE is just an alias for GFP_HIGHUSER_MOVABLE, making
that harder to track down: remove it, and its out-of-work brothers
GFP_NOFS_PAGECACHE and GFP_USER_PAGECACHE.

Since we're making that improvement to hotremove_migrate_alloc(), I think
we can now also remove one of the "o"s from its comment.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: remove cgroup_mm_owner_callbacks

cgroup_mm_owner_callbacks() was brought in to support the memrlimit
controller, but sneaked into mainline ahead of it.  That controller has
now been shelved, and the mm_owner_changed() args were inadequate for it
anyway (they needed an mm pointer instead of a task pointer).

Remove the dead code, and restore mm_update_next_owner() locking to how it
was before: taking mmap_sem there does nothing for memcontrol.c, now the
only user of mm->owner.

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Paul Menage <menage@google.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

do_mpage_readpage(): remove useless clear_buffer_mapped() call

It is known that buffer_mapped() is false in this code path.

Signed-off-by: Franck Bui-Huu <fbuihuu@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/apply_to_range: call pte function with lazy updates

Make the pte-level function in apply_to_range be called in lazy mmu mode,
so that any pagetable modifications can be batched.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmalloc make lazy unmapping configurable

Lazy unmapping in the vmalloc code has now opened the possibility for use
after free bugs to go undetected.  We can catch those by forcing an unmap
and flush (which is going to be slow, but that's what happens).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmalloc use mutex for purge

The vmalloc purge lock can be a mutex so we can sleep while a purge is
going on (purge involves a global kernel TLB invalidate, so it can take
quite a while).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmalloc improve vmallocinfo

If we do that, output of files like /proc/vmallocinfo will show things
like "vmalloc_32", "vmalloc_user", or whomever the caller was as the
caller.  This info is not as useful as the real caller of the allocation.

So, proposal is to call __vmalloc_node node directly, with matching
parameters to save the caller information

Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: vmalloc tweak failure printk

If we can't service a vmalloc allocation, show size of the allocation that
actually failed.  Useful for debugging.

Signed-off-by: Glauber Costa <glommer@redhat.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: more likely reclaim MADV_SEQUENTIAL mappings

File pages mapped only in sequentially read mappings are perfect reclaim
canditates.

This patch makes these mappings behave like weak references, their pages
will be reclaimed unless they have a strong reference from a normal
mapping as well.

It changes the reclaim and the unmap path where they check if the page has
been referenced.  In both cases, accesses through sequentially read
mappings will be ignored.

Benchmark results from KOSAKI Motohiro:

    http://marc.info/?l=linux-mm&m=122485301925098&w=2

Signed-off-by: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Rik van Riel <riel@redhat.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: cleanup: remove #ifdef CONFIG_MIGRATION

#ifdef in *.c file decrease source readability a bit.  removing is better.

This patch doesn't have any functional change.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: get rid of pagevec_release_nonlru()

speculative page references patch (commit:
e286781d5f2e9c846e012a39653a166e9d31777d) removed last
pagevec_release_nonlru() caller.

So this function can be removed now.

This patch doesn't have any functional change.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: print out memmap number only if it is not zero

Don't print the size of the zone's memmap array if it does not have one.

Impact: cleanup

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: show node to memory section relationship with symlinks in sysfs

Show node to memory section relationship with symlinks in sysfs

Add /sys/devices/system/node/nodeX/memoryY symlinks for all
the memory sections located on nodeX.  For example:
/sys/devices/system/node/node1/memory135 -> ../../memory/memory135
indicates that memory section 135 resides on node1.

Also revises documentation to cover this change as well as updating
Documentation/ABI/testing/sysfs-devices-memory to include descriptions
of memory hotremove files 'phys_device', 'phys_index', and 'state'
that were previously not described there.

In addition to it always being a good policy to provide users with
the maximum possible amount of physical location information for
resources that can be hot-added and/or hot-removed, the following
are some (but likely not all) of the user benefits provided by
this change.
Immediate:
  - Provides information needed to determine the specific node
    on which a defective DIMM is located.  This will reduce system
    downtime when the node or defective DIMM is swapped out.
  - Prevents unintended onlining of a memory section that was
    previously offlined due to a defective DIMM.  This could happen
    during node hot-add when the user or node hot-add assist script
    onlines _all_ offlined sections due to user or script inability
    to identify the specific memory sections located on the hot-added
    node.  The consequences of reintroducing the defective memory
    could be ugly.
  - Provides information needed to vary the amount and distribution
    of memory on specific nodes for testing or debugging purposes.
Future:
  - Will provide information needed to identify the memory
    sections that need to be offlined prior to physical removal
    of a specific node.

Symlink creation during boot was tested on 2-node x86_64, 2-node
ppc64, and 2-node ia64 systems.  Symlink creation during physical
memory hot-add tested on a 2-node x86_64 system.

Signed-off-by: Gary Hade <garyhade@us.ibm.com>
Signed-off-by: Badari Pulavarty <pbadari@us.ibm.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: do_sync_mapping_range integrity fix

Chris Mason notices do_sync_mapping_range didn't actually ask for data
integrity writeout.  Unfortunately, it is advertised as being usable for
data integrity operations.

This is a data integrity bug.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages more terminate quickly

Now that we have the early-termination logic in place, it makes sense to
bail out early in all other cases where done is set to 1.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages terminate quickly

Terminate the write_cache_pages loop upon encountering the first page past
end, without locking the page.  Pages cannot have their index change when
we have a reference on them (truncate, eg truncate_inode_pages_range
performs the same check without the page lock).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages optimise page cleaning

In write_cache_pages, if we get stuck behind another process that is
cleaning pages, we will be forced to wait for them to finish, then perform
our own writeout (if it was redirtied during the long wait), then wait for
that.

If a page under writeout is still clean, we can skip waiting for it (if
we're part of a data integrity sync, we'll be waiting for all writeout
pages afterwards, so we'll still be waiting for the other guy's write
that's cleaned the page).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages cleanups

Get rid of some complex expressions from flow control statements, add a
comment, remove some duplicate code.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages integrity fix

In write_cache_pages, nr_to_write is heeded even for data-integrity syncs,
so the function will return success after writing out nr_to_write pages,
even if that was not sufficient to guarantee data integrity.

The callers tend to set it to values that could break data interity
semantics easily in practice.  For example, nr_to_write can be set to
mapping->nr_pages * 2, however if a file has a single, dirty page, then
fsync is called, subsequent pages might be concurrently added and dirtied,
then write_cache_pages might writeout two of these newly dirty pages,
while not writing out the old page that should have been written out.

Fix this by ignoring nr_to_write if it is a data integrity sync.

This is a data integrity bug.

The reason this has been done in the past is to avoid stalling sync
operations behind page dirtiers.

 "If a file has one dirty page at offset 1000000000000000 then someone
  does an fsync() and someone else gets in first and starts madly writing
  pages at offset 0, we want to write that page at 1000000000000000.
  Somehow."

What we do today is return success after an arbitrary amount of pages are
written, whether or not we have provided the data-integrity semantics that
the caller has asked for.  Even this doesn't actually fix all stall cases
completely: in the above situation, if the file has a huge number of pages
in pagecache (but not dirty), then mapping->nrpages is going to be huge,
even if pages are being dirtied.

This change does indeed make the possibility of long stalls lager, and
that's not a good thing, but lying about data integrity is even worse.  We
have to either perform the sync, or return -ELINUXISLAME so at least the
caller knows what has happened.

There are subsequent competing approaches in the works to solve the stall
problems properly, without compromising data integrity.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages writepage error fix

In write_cache_pages, if ret signals a real error, but we still have some
pages left in the pagevec, done would be set to 1, but the remaining pages
would continue to be processed and ret will be overwritten in the process.

It could easily be overwritten with success, and thus success will be
returned even if there is an error.  Thus the caller is told all writes
succeeded, wheras in reality some did not.

Fix this by bailing immediately if there is an error, and retaining the
first error code.

This is a data integrity bug.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages early loop termination

We'd like to break out of the loop early in many situations, however the
existing code has been setting mapping->writeback_index past the final
page in the pagevec lookup for cyclic writeback.  This is a problem if we
don't process all pages up to the final page.

Currently the code mostly keeps writeback_index reasonable and hacked
around this by not breaking out of the loop or writing pages outside the
range in these cases.  Keep track of a real "done index" that enables us
to terminate the loop in a much more flexible manner.

Needed by the subsequent patch to preserve writepage errors, and then
further patches to break out of the loop early for other reasons.  However
there are no functional changes with this patch alone.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: write_cache_pages cyclic fix

In write_cache_pages, scanned == 1 is supposed to mean that cyclic
writeback has circled through zero, thus we should not circle again.
However it gets set to 1 after the first successful pagevec lookup.  This
leads to cases where not enough data gets written.

Counterexample: file with first 10 pages dirty, writeback_index == 5,
nr_to_write == 10.  Then the 5 last pages will be found, and scanned will
be set to 1, after writing those out, we will not cycle back to get the
first 5.

Rework this logic, now we'll always cycle unless we started off from index
0.  When cycling, only write out as far as 1 page before the start page
from the first cycle (so we don't write parts of the file twice).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

do_mpage_readpage(): don't submit lots of small bios on boundary

While tracing I/O patterns with blktrace (a great tool) a few weeks ago I
identified a minor issue in fs/mpage.c

As the comment above mpage_readpages() says, a fs's get_block function
will set BH_Boundary when it maps a block just before a block for which
extra I/O is required.

Since get_block() can map a range of pages, for all these pages the
BH_Boundary flag will be set.  But we only need to push what I/O we have
accumulated at the last block of this range.

This makes do_mpage_readpage() send out the largest possible bio instead
of a bunch of page-sized ones in the BH_Boundary case.

Signed-off-by: Miquel van Smoorenburg <mikevs@xs4all.net>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: print triggering task's cpuset and mems allowed

When cpusets are enabled, it's necessary to print the triggering task's
set of allowable nodes so the subsequently printed meminfo can be
interpreted correctly.

We also print the task's cpuset name for informational purposes.

[rientjes@google.com: task lock current before dereferencing cpuset]
Cc: Paul Menage <menage@google.com>
Cc: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

oom: fix zone_scan_mutex name

zone_scan_mutex is actually a spinlock, so name it appropriately.

Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: invoke oom-killer from page fault

Rather than have the pagefault handler kill a process directly if it gets
a VM_FAULT_OOM, have it call into the OOM killer.

With increasingly sophisticated oom behaviour (cpusets, memory cgroups,
oom killing throttling, oom priority adjustment or selective disabling,
panic on oom, etc), it's silly to unconditionally kill the faulting
process at page fault time.  Create a hook for pagefault oom path to call
into instead.

Only converted x86 and uml so far.

[akpm@linux-foundation.org: make __out_of_memory() static]
[akpm@linux-foundation.org: fix comment]
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Jeff Dike <jdike@addtoit.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: move_pages: no need to set pp->page to ZERO_PAGE(0) by default

pp->page is never used when not set to the right page, so there is no need
to set it to ZERO_PAGE(0) by default.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: rework do_pages_move() to work on page_sized chunks

Rework do_pages_move() to work by page-sized chunks of struct page_to_node
that are passed to do_move_page_to_node_array().  We now only have to
allocate a single page instead a possibly very large vmalloc area to store
all page_to_node entries.

As a result, new_page_node() will now have a very small lookup, hidding
much of the overall sys_move_pages() overhead.

Signed-off-by: Brice Goglin <Brice.Goglin@inria.fr>
Signed-off-by: Nathalie Furmento <Nathalie.Furmento@labri.fr>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: don't mark_page_accessed in shmem_fault

Following "mm: don't mark_page_accessed in fault path", which now
places a mark_page_accessed() in zap_pte_range(), we should remove
the mark_page_accessed() from shmem_fault().

Signed-off-by: Hugh Dickins <hugh@veritas.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: don't mark_page_accessed in fault path

Doing a mark_page_accessed at fault-time, then doing SetPageReferenced at
unmap-time if the pte is young has a number of problems.

mark_page_accessed is supposed to be roughly the equivalent of a young pte
for unmapped references. Unfortunately it doesn't come with any context:
after being called, reclaim doesn't know who or why the page was touched.

So calling mark_page_accessed not only adds extra lru or PG_referenced
manipulations for pages that are already going to have pte_young ptes anyway,
but it also adds these references which are difficult to work with from the
context of vma specific references (eg. MADV_SEQUENTIAL pte_young may not
wish to contribute to the page being referenced).

Then, simply doing SetPageReferenced when zapping a pte and finding it is
young, is not a really good solution either. SetPageReferenced does not
correctly promote the page to the active list for example. So after removing
mark_page_accessed from the fault path, several mmap()+touch+munmap() would
have a very different result from several read(2) calls for example, which
is not really desirable.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Johannes Weiner <hannes@saeurebad.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: report the MMU pagesize in /proc/pid/smaps

The KernelPageSize entry in /proc/pid/smaps is the pagesize used by the
kernel to back a VMA.  This matches the size used by the MMU in the
majority of cases.  However, one counter-example occurs on PPC64 kernels
whereby a kernel using 64K as a base pagesize may still use 4K pages for
the MMU on older processor.  To distinguish, this patch reports
MMUPageSize as the pagesize used by the MMU in /proc/pid/smaps.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Cc: "KOSAKI Motohiro" <kosaki.motohiro@jp.fujitsu.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: report the pagesize backing a VMA in /proc/pid/smaps

It is useful to verify a hugepage-aware application is using the expected
pagesizes for its memory regions. This patch creates an entry called
KernelPageSize in /proc/pid/smaps that is the size of page used by the
kernel to back a VMA. The entry is not called PageSize as it is possible
the MMU uses a different size. This extension should not break any sensible
parser that skips lines containing unrecognised information.

Signed-off-by: Mel Gorman <mel@csn.ul.ie>
Acked-by: "KOSAKI Motohiro" <kosaki.motohiro@jp.fujitsu.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge git://git./linux/kernel/git/agk/linux-2.6-dm

* git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm:
  dm snapshot: extend exception store functions
  dm snapshot: split out exception store implementations
  dm snapshot: rename struct exception_store
  dm snapshot: separate out exception store interface
  dm mpath: move trigger_event to system workqueue
  dm: add name and uuid to sysfs
  dm table: rework reference counting
  dm: support barriers on simple devices
  dm request: extend target interface
  dm request: add caches
  dm ioctl: allow dm_copy_name_and_uuid to return only one field
  dm log: ensure log bitmap fits on log device
  dm log: move region_size validation
  dm log: avoid reinitialising io_req on every operation
  dm: consolidate target deregistration error handling
  dm raid1: fix error count
  dm log: fix dm_io_client leak on error paths
  dm snapshot: change yield to msleep
  dm table: drop reference at unbind

dm snapshot: extend exception store functions

Supply dm_add_exception as a callback to the read_metadata function.
Add a status function ready for a later patch and name the functions
consistently.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm snapshot: split out exception store implementations

Move the existing snapshot exception store implementations out into
separate files.  Later patches will place these behind a new
interface in preparation for alternative implementations.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm snapshot: rename struct exception_store

Rename struct exception_store to dm_exception_store.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm snapshot: separate out exception store interface

Pull structures that bridge the gap between snapshot and
exception store out of dm-snap.h and put them in a new
.h file - dm-exception-store.h.  This file will define the
API for new exception stores.

Ultimately, dm-snap.h is unnecessary, since only dm-snap.c
should be using it.

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm mpath: move trigger_event to system workqueue

The same workqueue is used both for sending uevents and processing queued I/O.
Deadlock has been reported in RHEL5 when sending a uevent was blocked waiting
for the queued I/O to be processed.  Use scheduled_work() for the asynchronous
uevents instead.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm: add name and uuid to sysfs

Implement simple read-only sysfs entry for device-mapper block device.

This patch adds a simple sysfs directory named "dm" under block device
properties and implements
- name attribute (string containing mapped device name)
- uuid attribute (string containing UUID, or empty string if not set)

The kobject is embedded in mapped_device struct, so no additional
memory allocation is needed for initializing sysfs entry.

During the processing of sysfs attribute we need to lock mapped device
which is done by a new function dm_get_from_kobj, which returns the md
associated with kobject and increases the usage count.

Each 'show attribute' function is responsible for its own locking.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm table: rework reference counting

Rework table reference counting.

The existing code uses a reference counter. When the last reference is
dropped and the counter reaches zero, the table destructor is called.
Table reference counters are acquired/released from upcalls from other
kernel code (dm_any_congested, dm_merge_bvec, dm_unplug_all).
If the reference counter reaches zero in one of the upcalls, the table
destructor is called from almost random kernel code.

This leads to various problems:
* dm_any_congested being called under a spinlock, which calls the
  destructor, which calls some sleeping function.
* the destructor attempting to take a lock that is already taken by the
  same process.
* stale reference from some other kernel code keeps the table
  constructed, which keeps some devices open, even after successful
  return from "dmsetup remove". This can confuse lvm and prevent closing
  of underlying devices or reusing device minor numbers.

The patch changes reference counting so that the table destructor can be
called only at predetermined places.

The table has always exactly one reference from either mapped_device->map
or hash_cell->new_map. After this patch, this reference is not counted
in table->holders.  A pair of dm_create_table/dm_destroy_table functions
is used for table creation/destruction.

Temporary references from the other code increase table->holders. A pair
of dm_table_get/dm_table_put functions is used to manipulate it.

When the table is about to be destroyed, we wait for table->holders to
reach 0. Then, we call the table destructor.  We use active waiting with
msleep(1), because the situation happens rarely (to one user in 5 years)
and removing the device isn't performance-critical task: the user doesn't
care if it takes one tick more or not.

This way, the destructor is called only at specific points
(dm_table_destroy function) and the above problems associated with lazy
destruction can't happen.

Finally remove the temporary protection added to dm_any_congested().

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm: support barriers on simple devices

Implement barrier support for single device DM devices

This patch implements barrier support in DM for the common case of dm linear
just remapping a single underlying device. In this case we can safely
pass the barrier through because there can be no reordering between
devices.

 NB. Any DM device might cease to support barriers if it gets
     reconfigured so code must continue to allow for a possible
     -EOPNOTSUPP on every barrier bio submitted.  - agk

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm request: extend target interface

This patch adds the following target interfaces for request-based dm.

  map_rq    : for mapping a request

  rq_end_io : for finishing a request

  busy      : for avoiding performance regression from bio-based dm.
              Target can tell dm core not to map requests now, and
              that may help requests in the block layer queue to be
              bigger by I/O merging.
              In bio-based dm, this behavior is done by device
              drivers managing the block layer queue.
              But in request-based dm, dm core has to do that
              since dm core manages the block layer queue.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm request: add caches

This patch prepares some kmem_caches for request-based dm.

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm ioctl: allow dm_copy_name_and_uuid to return only one field

Allow NULL buffer in dm_copy_name_and_uuid if you only want to return one of
the fields.

(Required by a following patch that adds these fields to sysfs.)

Signed-off-by: Milan Broz <mbroz@redhat.com>
Reviewed-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm log: ensure log bitmap fits on log device

Check that the log bitmap will fit within the log device.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm log: move region_size validation

Move log size validation from mirror target to log constructor.

Removed PAGE_SIZE restriction we no longer think necessary.

Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm log: avoid reinitialising io_req on every operation

rw_header function updates three members of io_req data every time
when I/O is processed. bi_rw and notify.fn are never modified once
they get initialized, and so they can be set in advance.

header_to_disk() can also be pulled out of write_header() since only one
caller needs it and write_header() can be replaced by rw_header()
directly.

Signed-off-by: Takahiro Yasui <tyasui@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm: consolidate target deregistration error handling

Change dm_unregister_target to return void and use BUG() for error
reporting.

dm_unregister_target can only fail because of programming bug in the
target driver. It can't fail because of user's behavior or disk errors.

This patch changes unregister_target to return void and use BUG if
someone tries to unregister non-registered target or unregister target
that is in use.

This patch removes code duplication (testing of error codes in all dm
targets) and reports bugs in just one place, in dm_unregister_target. In
some target drivers, these return codes were ignored, which could lead
to a situation where bugs could be missed.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm raid1: fix error count

Always increase the error count when I/O on a leg of a mirror fails.

The error count is used to decide whether to select an alternative
mirror leg.  If the target doesn't use the "handle_errors" feature, the
error count is not updated and the bio can get requeued forever by the
read callback.

Fix it by increasing error_count before the handle_errors feature
checking.

Cc: stable@kernel.org
Signed-off-by: Milan Broz <mbroz@redhat.com>
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>

dm log: fix dm_io_client leak on error paths

In create_log_context function, dm_io_client_destroy function needs
to be called, when memory allocation of disk_header, sync_bits and
recovering_bits failed, but dm_io_client_destroy is not called.

Cc: stable@kernel.org
Signed-off-by: Takahiro Yasui <tyasui@redhat.com>
Acked-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>