On Sat, Nov 14, 2015 at 03:36:50PM +0300, Vladimir Davydov wrote:

On Thu, Nov 12, 2015 at 06:41:21PM -0500, Johannes Weiner wrote:

quoted

@@ -2432,20 +2447,6 @@ static bool shrink_zone(struct zone *zone, struct scan_control *sc,
 			}
 		} while ((memcg = mem_cgroup_iter(root, memcg, &reclaim)));
 
-		/*
-		 * Shrink the slab caches in the same proportion that
-		 * the eligible LRU pages were scanned.
-		 */
-		if (global_reclaim(sc) && is_classzone)
-			shrink_slab(sc->gfp_mask, zone_to_nid(zone), NULL,
-				    sc->nr_scanned - nr_scanned,
-				    zone_lru_pages);
-
-		if (reclaim_state) {
-			sc->nr_reclaimed += reclaim_state->reclaimed_slab;
-			reclaim_state->reclaimed_slab = 0;
-		}
-

AFAICS this patch deadly breaks memcg-unaware shrinkers vs LRU balance:
currently we scan (*total* LRU scanned / *total* LRU pages) of all such
objects; with this patch we'd use the numbers from the root cgroup
instead. If most processes reside in memory cgroups, the root cgroup
will have only a few LRU pages and hence the pressure exerted upon such
objects will be unfairly severe.

You're absolutely right, good catch.

Please disregard this patch. It's not necessary for this series after
v2, I just kept it because I thought it's a nice simplification that's
possible after making root_mem_cgroup public.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>