]> git.karo-electronics.de Git - karo-tx-linux.git/commit
vmscan: add barrier to prevent evictable page in unevictable list
authorMinchan Kim <minchan.kim@gmail.com>
Mon, 24 Oct 2011 14:54:30 +0000 (01:54 +1100)
committerStephen Rothwell <sfr@canb.auug.org.au>
Tue, 25 Oct 2011 09:07:51 +0000 (20:07 +1100)
commit87881677372c3b72b1da04a0d3971bad21b8ceea
tree1aa183221e43c428bd4cb58100e5763195a44095
parenta0fe64f73912d0559d820a989257a15362000774
vmscan: add barrier to prevent evictable page in unevictable list

When a race between putback_lru_page() and shmem_lock with lock=0 happens,
progrom execution order is as follows, but clear_bit in processor #1 could
be reordered right before spin_unlock of processor #1.  Then, the page
would be stranded on the unevictable list.

spin_lock
SetPageLRU
spin_unlock
                                clear_bit(AS_UNEVICTABLE)
                                spin_lock
                                if PageLRU()
                                        if !test_bit(AS_UNEVICTABLE)
                                         move evictable list
smp_mb
if !test_bit(AS_UNEVICTABLE)
        move evictable list
                                spin_unlock

But, pagevec_lookup() in scan_mapping_unevictable_pages() has
rcu_read_[un]lock() so it could protect reordering before reaching
test_bit(AS_UNEVICTABLE) on processor #1 so this problem never happens.
But it's a unexpected side effect and we should solve this problem
properly.

This patch adds a barrier after mapping_clear_unevictable.

I didn't meet this problem but just found during review.

Signed-off-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Rik van Riel <riel@redhat.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Johannes Weiner <jweiner@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/shmem.c
mm/vmscan.c