mm: memcontrol: fix NUMA round-robin reclaim at intermediate level
authorJohannes Weiner <>
Tue, 14 May 2019 22:47:15 +0000 (15:47 -0700)
committerLinus Torvalds <>
Wed, 15 May 2019 02:52:53 +0000 (19:52 -0700)
When a cgroup is reclaimed on behalf of a configured limit, reclaim
needs to round-robin through all NUMA nodes that hold pages of the memcg
in question.  However, when assembling the mask of candidate NUMA nodes,
the code only consults the *local* cgroup LRU counters, not the
recursive counters for the entire subtree.  Cgroup limits are frequently
configured against intermediate cgroups that do not have memory on their
own LRUs.  In this case, the node mask will always come up empty and
reclaim falls back to scanning only the current node.

If a cgroup subtree has some memory on one node but the processes are
bound to another node afterwards, the limit reclaim will never age or
reclaim that memory anymore.

To fix this, use the recursive LRU counts for a cgroup subtree to
determine which nodes hold memory of that cgroup.

The code has been broken like this forever, so it doesn't seem to be a
problem in practice.  I just noticed it while reviewing the way the LRU
counters are used in general.

Signed-off-by: Johannes Weiner <>
Reviewed-by: Shakeel Butt <>
Reviewed-by: Roman Gushchin <>
Cc: Michal Hocko <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>

index 7a764cbf84952f557df968c5a78d0f4e8494f06d..e50a2db5b4ff0dcf4bf17b888ea12cfd1284dd56 100644 (file)
@@ -1507,13 +1507,13 @@ static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg,
        struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg);
-       if (lruvec_page_state_local(lruvec, NR_INACTIVE_FILE) ||
-           lruvec_page_state_local(lruvec, NR_ACTIVE_FILE))
+       if (lruvec_page_state(lruvec, NR_INACTIVE_FILE) ||
+           lruvec_page_state(lruvec, NR_ACTIVE_FILE))
                return true;
        if (noswap || !total_swap_pages)
                return false;
-       if (lruvec_page_state_local(lruvec, NR_INACTIVE_ANON) ||
-           lruvec_page_state_local(lruvec, NR_ACTIVE_ANON))
+       if (lruvec_page_state(lruvec, NR_INACTIVE_ANON) ||
+           lruvec_page_state(lruvec, NR_ACTIVE_ANON))
                return true;
        return false;