vmscan: order evictable rescue in LRU putback
authorJohannes Weiner <hannes@cmpxchg.org>
Mon, 26 Oct 2009 23:50:00 +0000 (16:50 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 29 Oct 2009 14:39:30 +0000 (07:39 -0700)
commit6a7b95481d49f73991d3dbf8c1e696a24684ac05
tree32f80d8839daf939bcae8d7711e64b8e96d89ce1
parentb05ca7385a2848abdc72051f832722641daed8b0
vmscan: order evictable rescue in LRU putback

Isolators putting a page back to the LRU do not hold the page lock, and if
the page is mlocked, another thread might munlock it concurrently.

Expecting this, the putback code re-checks the evictability of a page when
it just moved it to the unevictable list in order to correct its decision.

The problem, however, is that ordering is not garuanteed between setting
PG_lru when moving the page to the list and checking PG_mlocked
afterwards:

#0: #1

spin_lock()
if (TestClearPageMlocked())
  if (PageLRU())
    move to evictable list
SetPageLRU()
spin_unlock()
if (!PageMlocked())
  move to evictable list

The PageMlocked() check may get reordered before SetPageLRU() in #0,
resulting in #0 not moving the still mlocked page, and in #1 failing to
isolate and move the page as well.  The page is now stranded on the
unevictable list.

The race condition is very unlikely.  The consequence currently is one
page falling off the reclaim grid and eventually getting freed with
PG_unevictable set, which triggers a warning in the page allocator.

TestClearPageMlocked() in #1 already provides full memory barrier
semantics.

This patch adds an explicit full barrier to force ordering between
SetPageLRU() and PageMlocked() so that either one of the competitors
rescues the page.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Reviewed-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/vmscan.c