On Wed, Jun 26, 2013 at 12:39:25PM -0700, Andrew Morton wrote:

On Wed, 26 Jun 2013 13:39:23 +0100 Mel Gorman [off-list ref] wrote:

quoted

Page reclaim at priority 0 will scan the entire LRU as priority 0 is
considered to be a near OOM condition. Direct reclaim can reach this
priority while still making reclaim progress. This patch avoids
reclaiming at priority 0 unless no reclaim progress was made and
the page allocator would consider firing the OOM killer. The
user-visible impact is that direct reclaim will not easily reach
priority 0 and start swapping prematurely.

That's a bandaid.

Priority 0 should be a pretty darn rare condition.  How often is it
occurring, and do you know why?

There are no flys on you.

The actual rescanning never happens in my experience but priority 0
is reached quickly. Instrumentation showed that it was due to a bug in
reclaim/compaction. Please consider replacing the patch with this.

---8<---
mm: vmscan: Do not continue scanning if reclaim was aborted for compaction

Direct reclaim is not aborting to allow compaction to go ahead properly.
do_try_to_free_pages is told to abort reclaim which is happily ignores
and instead increases priority instead until it reaches 0 and starts
shrinking file/anon equally. This patch corrects the situation by
aborting reclaim when requested instead of raising priority.

Signed-off-by: Mel Gorman <mgorman@suse.de>
---
 mm/vmscan.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 2a5dee2..2f0193c 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c

@@ -2381,8 +2381,10 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
 		aborted_reclaim = shrink_zones(zonelist, sc);
 
 		/*
-		 * Don't shrink slabs when reclaiming memory from
-		 * over limit cgroups
+		 * Don't shrink slabs when reclaiming memory from over limit
+		 * cgroups but do shrink slab at least once when aborting
+		 * reclaim for compaction to avoid unevenly scanning file/anon
+		 * LRU pages over slab pages.
 		 */
 		if (global_reclaim(sc)) {
 			unsigned long lru_pages = 0;

@@ -2428,7 +2430,7 @@ static unsigned long do_try_to_free_pages(struct zonelist *zonelist,
 						WB_REASON_TRY_TO_FREE_PAGES);
 			sc->may_writepage = 1;
 		}
-	} while (--sc->priority >= 0);
+	} while (--sc->priority >= 0 && !aborted_reclaim);
 
 out:
 	delayacct_freepages_end();

--

To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>