mm, cma: prevent nr_isolated_* counters from going negative
authorHugh Dickins <hughd@google.com>
Thu, 5 May 2016 23:22:15 +0000 (16:22 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 6 May 2016 00:38:53 +0000 (17:38 -0700)
commit14af4a5e9b26ad251f81c174e8a43f3e179434a5
tree38edf026215bd254a7cbe13bb47aaf23c641f019
parentbc22af74f271ef76b2e6f72f3941f91f0da3f5f8
mm, cma: prevent nr_isolated_* counters from going negative

/proc/sys/vm/stat_refresh warns nr_isolated_anon and nr_isolated_file go
increasingly negative under compaction: which would add delay when
should be none, or no delay when should delay.  The bug in compaction
was due to a recent mmotm patch, but much older instance of the bug was
also noticed in isolate_migratepages_range() which is used for CMA and
gigantic hugepage allocations.

The bug is caused by putback_movable_pages() in an error path
decrementing the isolated counters without them being previously
incremented by acct_isolated().  Fix isolate_migratepages_range() by
removing the error-path putback, thus reaching acct_isolated() with
migratepages still isolated, and leaving putback to caller like most
other places do.

Fixes: edc2ca612496 ("mm, compaction: move pageblock checks up from isolate_migratepages_range()")
[vbabka@suse.cz: expanded the changelog]
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/compaction.c