mm: fix negative nr_isolated counts
authorHugh Dickins <hughd@google.com>
Thu, 12 Feb 2015 23:00:28 +0000 (15:00 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 13 Feb 2015 02:54:11 +0000 (18:54 -0800)
The vmstat interfaces are good at hiding negative counts (at least when
CONFIG_SMP); but if you peer behind the curtain, you find that
nr_isolated_anon and nr_isolated_file soon go negative, and grow ever
more negative: so they can absorb larger and larger numbers of isolated
pages, yet still appear to be zero.

I'm happy to avoid a congestion_wait() when too_many_isolated() myself;
but I guess it's there for a good reason, in which case we ought to get
too_many_isolated() working again.

The imbalance comes from isolate_migratepages()'s ISOLATE_ABORT case:
putback_movable_pages() decrements the NR_ISOLATED counts, but we forgot
to call acct_isolated() to increment them.

It is possible that the bug whcih this patch fixes could cause OOM kills
when the system still has a lot of reclaimable page cache.

Fixes: edc2ca612496 ("mm, compaction: move pageblock checks up from isolate_migratepages_range()")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: <stable@vger.kernel.org> [3.18+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/compaction.c

index 782772d..d50d6de 100644 (file)
@@ -1103,8 +1103,10 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
                low_pfn = isolate_migratepages_block(cc, low_pfn, end_pfn,
                                                                isolate_mode);
 
-               if (!low_pfn || cc->contended)
+               if (!low_pfn || cc->contended) {
+                       acct_isolated(zone, cc);
                        return ISOLATE_ABORT;
+               }
 
                /*
                 * Either we isolated something and proceed with migration. Or