kswapd: avoid unnecessary rebalance after an unsuccessful balancing
authorAlex,Shi <alex.shi@intel.com>
Tue, 1 Nov 2011 00:08:39 +0000 (17:08 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 1 Nov 2011 00:30:48 +0000 (17:30 -0700)
In commit 215ddd66 ("mm: vmscan: only read new_classzone_idx from pgdat
when reclaiming successfully") , Mel Gorman said kswapd is better to sleep
after a unsuccessful balancing if there is tighter reclaim request pending
in the balancing.  But in the following scenario, kswapd do something that
is not matched our expectation.  The patch fixes this issue.

1, Read pgdat request A (classzone_idx, order = 3)
2, balance_pgdat()
3, During pgdat, a new pgdat request B (classzone_idx, order = 5) is placed
4, balance_pgdat() returns but failed since returned order = 0
5, pgdat of request A assigned to balance_pgdat(), and do balancing again.
   While the expectation behavior of kswapd should try to sleep.

Signed-off-by: Alex Shi <alex.shi@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Acked-by: Mel Gorman <mgorman@suse.de>
Tested-by: Pádraig Brady <P@draigBrady.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/vmscan.c

index 77ee24fc891aed4b362e6a728bb36172583e1ecd..dd5fc86dbb82efd2f6280e48a7bcf51222e5780b 100644 (file)
@@ -2823,7 +2823,9 @@ static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx)
 static int kswapd(void *p)
 {
        unsigned long order, new_order;
+       unsigned balanced_order;
        int classzone_idx, new_classzone_idx;
+       int balanced_classzone_idx;
        pg_data_t *pgdat = (pg_data_t*)p;
        struct task_struct *tsk = current;
 
@@ -2854,7 +2856,9 @@ static int kswapd(void *p)
        set_freezable();
 
        order = new_order = 0;
+       balanced_order = 0;
        classzone_idx = new_classzone_idx = pgdat->nr_zones - 1;
+       balanced_classzone_idx = classzone_idx;
        for ( ; ; ) {
                int ret;
 
@@ -2863,7 +2867,8 @@ static int kswapd(void *p)
                 * new request of a similar or harder type will succeed soon
                 * so consider going to sleep on the basis we reclaimed at
                 */
-               if (classzone_idx >= new_classzone_idx && order == new_order) {
+               if (balanced_classzone_idx >= new_classzone_idx &&
+                                       balanced_order == new_order) {
                        new_order = pgdat->kswapd_max_order;
                        new_classzone_idx = pgdat->classzone_idx;
                        pgdat->kswapd_max_order =  0;
@@ -2878,7 +2883,8 @@ static int kswapd(void *p)
                        order = new_order;
                        classzone_idx = new_classzone_idx;
                } else {
-                       kswapd_try_to_sleep(pgdat, order, classzone_idx);
+                       kswapd_try_to_sleep(pgdat, balanced_order,
+                                               balanced_classzone_idx);
                        order = pgdat->kswapd_max_order;
                        classzone_idx = pgdat->classzone_idx;
                        pgdat->kswapd_max_order = 0;
@@ -2895,7 +2901,9 @@ static int kswapd(void *p)
                 */
                if (!ret) {
                        trace_mm_vmscan_kswapd_wake(pgdat->node_id, order);
-                       order = balance_pgdat(pgdat, order, &classzone_idx);
+                       balanced_classzone_idx = classzone_idx;
+                       balanced_order = balance_pgdat(pgdat, order,
+                                               &balanced_classzone_idx);
                }
        }
        return 0;