On Tue, Mar 21, 2017 at 08:53:20PM -0400, Johannes Weiner wrote:
> 0a6b76dd23fa ("mm: workingset: make shadow node shrinker memcg aware")
> enabled cgroup-awareness in the shadow node shrinker, but forgot to
> also enable cgroup-awareness in the list_lru the shadow nodes sit on.
>
> Consequently, all shadow nodes are sitting on a global (per-NUMA node)
> list, while the shrinker applies the limits according to the amount of
> cache in the cgroup its shrinking. The result is excessive pressure on
> the shadow nodes from cgroups that have very little cache.
>
> Enable memcg-mode on the shadow node LRUs, such that per-cgroup limits
> are applied to per-cgroup lists.
>
> Fixes: 0a6b76dd23fa ("mm: workingset: make shadow node shrinker memcg aware")
> Signed-off-by: Johannes Weiner <han...@cmpxchg.org>
> Cc: <sta...@vger.kernel.org> # 4.6+
Acked-by: Vladimir Davydov <vdavy...@tarantool.org>
> ---
> mm/workingset.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/workingset.c b/mm/workingset.c
> index ac839fca0e76..eda05c71fa49 100644
> --- a/mm/workingset.c
> +++ b/mm/workingset.c
> @@ -532,7 +532,7 @@ static int __init workingset_init(void)
> pr_info("workingset: timestamp_bits=%d max_order=%d bucket_order=%u\n",
> timestamp_bits, max_order, bucket_order);
>
> - ret = list_lru_init_key(&shadow_nodes, &shadow_nodes_key);
> + ret = __list_lru_init(&shadow_nodes, true, &shadow_nodes_key);
> if (ret)
> goto err;
> ret = register_shrinker(&workingset_shadow_shrinker);
> --
> 2.12.0
>
