mm/vmscan: restore sc->gfp_mask after promoting it to __GFP_HIGHMEM
authorWeijie Yang <weijie.yang@samsung.com>
Mon, 7 Apr 2014 22:36:59 +0000 (15:36 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 7 Apr 2014 23:35:50 +0000 (16:35 -0700)
We promote sc->gfp_mask to __GFP_HIGHMEM to forcibly scan highmem if
there are too many buffer_heads pinning highmem.  See cc715d99e5 ("mm:
vmscan: forcibly scan highmem if there are too many buffer_heads pinning
highmem").

This patch restores sc->gfp_mask to its caller original value after
finishing the scan job, to avoid the impact on other invocations from
its upper caller, such as vmpressure_prio(), shrink_slab().

Signed-off-by: Weijie Yang <weijie.yang@samsung.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/vmscan.c

index 1f56a80..1c51e4f 100644 (file)
@@ -2314,6 +2314,7 @@ static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
        unsigned long lru_pages = 0;
        bool aborted_reclaim = false;
        struct reclaim_state *reclaim_state = current->reclaim_state;
+       gfp_t orig_mask;
        struct shrink_control shrink = {
                .gfp_mask = sc->gfp_mask,
        };
@@ -2323,6 +2324,7 @@ static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
         * allowed level, force direct reclaim to scan the highmem zone as
         * highmem pages could be pinning lowmem pages storing buffer_heads
         */
+       orig_mask = sc->gfp_mask;
        if (buffer_heads_over_limit)
                sc->gfp_mask |= __GFP_HIGHMEM;
 
@@ -2393,6 +2395,12 @@ static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
                }
        }
 
+       /*
+        * Restore to original mask to avoid the impact on the caller if we
+        * promoted it to __GFP_HIGHMEM.
+        */
+       sc->gfp_mask = orig_mask;
+
        return aborted_reclaim;
 }