mm: vmscan: move logic from balance_pgdat() to kswapd_shrink_zone()
authorMel Gorman <mgorman@suse.de>
Wed, 3 Jul 2013 22:01:54 +0000 (15:01 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 3 Jul 2013 23:07:28 +0000 (16:07 -0700)
balance_pgdat() is very long and some of the logic can and should be
internal to kswapd_shrink_zone().  Move it so the flow of
balance_pgdat() is marginally easier to follow.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Jiri Slaby <jslaby@suse.cz>
Cc: Valdis Kletnieks <Valdis.Kletnieks@vt.edu>
Tested-by: Zlatko Calusic <zcalusic@bitsync.net>
Cc: dormando <dormando@rydia.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/vmscan.c

index a2d0c68426169e59fb5daaeda397d014bdf72d7b..4a43c289b23a21e8bd27c88eae345f8f138465e6 100644 (file)
@@ -2709,18 +2709,53 @@ static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining,
  * This is used to determine if the scanning priority needs to be raised.
  */
 static bool kswapd_shrink_zone(struct zone *zone,
+                              int classzone_idx,
                               struct scan_control *sc,
                               unsigned long lru_pages,
                               unsigned long *nr_attempted)
 {
        unsigned long nr_slab;
+       int testorder = sc->order;
+       unsigned long balance_gap;
        struct reclaim_state *reclaim_state = current->reclaim_state;
        struct shrink_control shrink = {
                .gfp_mask = sc->gfp_mask,
        };
+       bool lowmem_pressure;
 
        /* Reclaim above the high watermark. */
        sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone));
+
+       /*
+        * Kswapd reclaims only single pages with compaction enabled. Trying
+        * too hard to reclaim until contiguous free pages have become
+        * available can hurt performance by evicting too much useful data
+        * from memory. Do not reclaim more than needed for compaction.
+        */
+       if (IS_ENABLED(CONFIG_COMPACTION) && sc->order &&
+                       compaction_suitable(zone, sc->order) !=
+                               COMPACT_SKIPPED)
+               testorder = 0;
+
+       /*
+        * We put equal pressure on every zone, unless one zone has way too
+        * many pages free already. The "too many pages" is defined as the
+        * high wmark plus a "gap" where the gap is either the low
+        * watermark or 1% of the zone, whichever is smaller.
+        */
+       balance_gap = min(low_wmark_pages(zone),
+               (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
+               KSWAPD_ZONE_BALANCE_GAP_RATIO);
+
+       /*
+        * If there is no low memory pressure or the zone is balanced then no
+        * reclaim is necessary
+        */
+       lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone));
+       if (!lowmem_pressure && zone_balanced(zone, testorder,
+                                               balance_gap, classzone_idx))
+               return true;
+
        shrink_zone(zone, sc);
 
        reclaim_state->reclaimed_slab = 0;
@@ -2735,6 +2770,18 @@ static bool kswapd_shrink_zone(struct zone *zone,
 
        zone_clear_flag(zone, ZONE_WRITEBACK);
 
+       /*
+        * If a zone reaches its high watermark, consider it to be no longer
+        * congested. It's possible there are dirty pages backed by congested
+        * BDIs but as pressure is relieved, speculatively avoid congestion
+        * waits.
+        */
+       if (!zone->all_unreclaimable &&
+           zone_balanced(zone, testorder, 0, classzone_idx)) {
+               zone_clear_flag(zone, ZONE_CONGESTED);
+               zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY);
+       }
+
        return sc->nr_scanned >= sc->nr_to_reclaim;
 }
 
@@ -2870,8 +2917,6 @@ static unsigned long balance_pgdat(pg_data_t *pgdat, int order,
                 */
                for (i = 0; i <= end_zone; i++) {
                        struct zone *zone = pgdat->node_zones + i;
-                       int testorder;
-                       unsigned long balance_gap;
 
                        if (!populated_zone(zone))
                                continue;
@@ -2892,61 +2937,14 @@ static unsigned long balance_pgdat(pg_data_t *pgdat, int order,
                        sc.nr_reclaimed += nr_soft_reclaimed;
 
                        /*
-                        * We put equal pressure on every zone, unless
-                        * one zone has way too many pages free
-                        * already. The "too many pages" is defined
-                        * as the high wmark plus a "gap" where the
-                        * gap is either the low watermark or 1%
-                        * of the zone, whichever is smaller.
-                        */
-                       balance_gap = min(low_wmark_pages(zone),
-                               (zone->managed_pages +
-                                       KSWAPD_ZONE_BALANCE_GAP_RATIO-1) /
-                               KSWAPD_ZONE_BALANCE_GAP_RATIO);
-                       /*
-                        * Kswapd reclaims only single pages with compaction
-                        * enabled. Trying too hard to reclaim until contiguous
-                        * free pages have become available can hurt performance
-                        * by evicting too much useful data from memory.
-                        * Do not reclaim more than needed for compaction.
+                        * There should be no need to raise the scanning
+                        * priority if enough pages are already being scanned
+                        * that that high watermark would be met at 100%
+                        * efficiency.
                         */
-                       testorder = order;
-                       if (IS_ENABLED(CONFIG_COMPACTION) && order &&
-                                       compaction_suitable(zone, order) !=
-                                               COMPACT_SKIPPED)
-                               testorder = 0;
-
-                       if ((buffer_heads_over_limit && is_highmem_idx(i)) ||
-                           !zone_balanced(zone, testorder,
-                                          balance_gap, end_zone)) {
-                               /*
-                                * There should be no need to raise the
-                                * scanning priority if enough pages are
-                                * already being scanned that high
-                                * watermark would be met at 100% efficiency.
-                                */
-                               if (kswapd_shrink_zone(zone, &sc, lru_pages,
-                                                      &nr_attempted))
-                                       raise_priority = false;
-                       }
-
-                       if (zone->all_unreclaimable) {
-                               if (end_zone && end_zone == i)
-                                       end_zone--;
-                               continue;
-                       }
-
-                       if (zone_balanced(zone, testorder, 0, end_zone))
-                               /*
-                                * If a zone reaches its high watermark,
-                                * consider it to be no longer congested. It's
-                                * possible there are dirty pages backed by
-                                * congested BDIs but as pressure is relieved,
-                                * speculatively avoid congestion waits
-                                * or writing pages from kswapd context.
-                                */
-                               zone_clear_flag(zone, ZONE_CONGESTED);
-                               zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY);
+                       if (kswapd_shrink_zone(zone, end_zone, &sc,
+                                       lru_pages, &nr_attempted))
+                               raise_priority = false;
                }
 
                /*