[PATCH] zone reclaim with slab: avoid unecessary off node allocations
authorChristoph Lameter <clameter@sgi.com>
Tue, 26 Sep 2006 06:31:53 +0000 (23:31 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Tue, 26 Sep 2006 15:48:52 +0000 (08:48 -0700)
Minor performance fix.

If we reclaimed enough slab pages from a zone then we can avoid going off
node with the current allocation.  Take care of updating nr_reclaimed when
reclaiming from the slab.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
mm/vmscan.c

index 089e943c4d38f56c75309386dde288ab35eed7e8..b950f193816e7ddbd7633a5a7e7924545088bee7 100644 (file)
@@ -1566,6 +1566,7 @@ static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
                .gfp_mask = gfp_mask,
                .swappiness = vm_swappiness,
        };
+       unsigned long slab_reclaimable;
 
        disable_swap_token();
        cond_resched();
@@ -1592,7 +1593,8 @@ static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order)
                } while (priority >= 0 && nr_reclaimed < nr_pages);
        }
 
-       if (zone_page_state(zone, NR_SLAB_RECLAIMABLE) > zone->min_slab_pages) {
+       slab_reclaimable = zone_page_state(zone, NR_SLAB_RECLAIMABLE);
+       if (slab_reclaimable > zone->min_slab_pages) {
                /*
                 * shrink_slab() does not currently allow us to determine how
                 * many pages were freed in this zone. So we take the current
@@ -1603,12 +1605,17 @@ 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.
                 */
-               unsigned long limit = zone_page_state(zone,
-                               NR_SLAB_RECLAIMABLE) - nr_pages;
-
                while (shrink_slab(sc.nr_scanned, gfp_mask, order) &&
-                       zone_page_state(zone, NR_SLAB_RECLAIMABLE) > limit)
+                       zone_page_state(zone, NR_SLAB_RECLAIMABLE) >
+                               slab_reclaimable - nr_pages)
                        ;
+
+               /*
+                * Update nr_reclaimed by the number of slab pages we
+                * reclaimed from this zone.
+                */
+               nr_reclaimed += slab_reclaimable -
+                       zone_page_state(zone, NR_SLAB_RECLAIMABLE);
        }
 
        p->reclaim_state = NULL;