vmscan: shrink_slab() requires the number of lru_pages, not the page order
KOSAKI Motohiro [Tue, 10 Aug 2010 00:19:54 +0000 (17:19 -0700)]
Presently shrink_slab() has the following scanning equation.

                            lru_scanned        max_pass
  basic_scan_objects = 4 x -------------  x -----------------------------
                            lru_pages        shrinker->seeks (default:2)

  scan_objects = min(basic_scan_objects, max_pass * 2)

If we pass very small value as lru_pages instead real number of lru pages,
shrink_slab() drop much objects rather than necessary.  And now,
__zone_reclaim() pass 'order' as lru_pages by mistake.  That produces a
bad result.

For example, if we receive very low memory pressure (scan = 32, order =
0), shrink_slab() via zone_reclaim() always drop _all_ icache/dcache
objects.  (see above equation, very small lru_pages make very big
scan_objects result).

This patch fixes it.

[akpm@linux-foundation.org: fix layout, typos]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Acked-by: Christoph Lameter <cl@linux-foundation.org>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c

index a3d669f..9789a2c 100644 (file)
@@ -2635,10 +2635,19 @@ static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
                 * Note that shrink_slab will free memory on all zones and may
                 * take a long time.
                 */
-               while (shrink_slab(sc.nr_scanned, gfp_mask, order) &&
-                      (zone_page_state(zone, NR_SLAB_RECLAIMABLE) + nr_pages >
-                               nr_slab_pages0))
-                       ;
+               for (;;) {
+                       unsigned long lru_pages = zone_reclaimable_pages(zone);
+
+                       /* No reclaimable slab or very low memory pressure */
+                       if (!shrink_slab(sc.nr_scanned, gfp_mask, lru_pages))
+                               break;
+
+                       /* Freed enough memory */
+                       nr_slab_pages1 = zone_page_state(zone,
+                                                       NR_SLAB_RECLAIMABLE);
+                       if (nr_slab_pages1 + nr_pages <= nr_slab_pages0)
+                               break;
+               }
 
                /*
                 * Update nr_reclaimed by the number of slab pages we