git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From mgorman@suse.de  Mon Aug  1 11:41:52 2011
   2 From: Mel Gorman <mgorman@suse.de>
   3 Date: Tue, 19 Jul 2011 10:15:49 +0100
   4 Subject: mm: compaction: ensure that the compaction free scanner does not move to the next zone
   5 To: stable@kernel.org
   6 Cc: Andrea Arcangeli <aarcange@redhat.com>, Andrew Morton <akpm@linux-foundation.org>, Thomas Sattler <tsattler@gmx.de>, Mel Gorman <mgorman@suse.de>
   7 Message-ID: <1311066951-25546-2-git-send-email-mgorman@suse.de>
   8
   9 From: Mel Gorman <mgorman@suse.de>
  10
  11 commit: 7454f4ba40b419eb999a3c61a99da662bf1a2bb8 upstream
  12
  13 Compaction works with two scanners, a migration and a free scanner.  When
  14 the scanners crossover, migration within the zone is complete.  The
  15 location of the scanner is recorded on each cycle to avoid excesive
  16 scanning.
  17
  18 When a zone is small and mostly reserved, it's very easy for the migration
  19 scanner to be close to the end of the zone.  Then the following situation
  20 can occurs
  21
  22   o migration scanner isolates some pages near the end of the zone
  23   o free scanner starts at the end of the zone but finds that the
  24     migration scanner is already there
  25   o free scanner gets reinitialised for the next cycle as
  26     cc->migrate_pfn + pageblock_nr_pages
  27     moving the free scanner into the next zone
  28   o migration scanner moves into the next zone
  29
  30 When this happens, NR_ISOLATED accounting goes haywire because some of the
  31 accounting happens against the wrong zone.  One zones counter remains
  32 positive while the other goes negative even though the overall global
  33 count is accurate.  This was reported on X86-32 with !SMP because !SMP
  34 allows the negative counters to be visible.  The fact that it is the bug
  35 should theoritically be possible there.
  36
  37 Signed-off-by: Mel Gorman <mgorman@suse.de>
  38 Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
  39 Reviewed-by: Michal Hocko <mhocko@suse.cz>
  40 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  41 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  42 Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
  43 ---
  44  mm/compaction.c |   13 ++++++++++++-
  45  1 file changed, 12 insertions(+), 1 deletion(-)
  46
  47 --- a/mm/compaction.c
  48 +++ b/mm/compaction.c
  49 @@ -144,9 +144,20 @@ static void isolate_freepages(struct zon
  50         int nr_freepages = cc->nr_freepages;
  51         struct list_head *freelist = &cc->freepages;
  52
  53 +       /*
  54 +        * Initialise the free scanner. The starting point is where we last
  55 +        * scanned from (or the end of the zone if starting). The low point
  56 +        * is the end of the pageblock the migration scanner is using.
  57 +        */
  58         pfn = cc->free_pfn;
  59         low_pfn = cc->migrate_pfn + pageblock_nr_pages;
  60 -       high_pfn = low_pfn;
  61 +
  62 +       /*
  63 +        * Take care that if the migration scanner is at the end of the zone
  64 +        * that the free scanner does not accidentally move to the next zone
  65 +        * in the next isolation cycle.
  66 +        */
  67 +       high_pfn = min(low_pfn, pfn);
  68
  69         /*
  70          * Isolate free pages until enough are available to migrate the