]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
memcg: reparent charges of children before processing parent
authorFilipe Brandenburger <filbranden@google.com>
Mon, 3 Mar 2014 23:38:25 +0000 (15:38 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 24 Mar 2014 04:44:20 +0000 (21:44 -0700)
commit 4fb1a86fb5e4209a7d4426d4e586c58e9edc74ac upstream.

Sometimes the cleanup after memcg hierarchy testing gets stuck in
mem_cgroup_reparent_charges(), unable to bring non-kmem usage down to 0.

There may turn out to be several causes, but a major cause is this: the
workitem to offline parent can get run before workitem to offline child;
parent's mem_cgroup_reparent_charges() circles around waiting for the
child's pages to be reparented to its lrus, but it's holding
cgroup_mutex which prevents the child from reaching its
mem_cgroup_reparent_charges().

Further testing showed that an ordered workqueue for cgroup_destroy_wq
is not always good enough: percpu_ref_kill_and_confirm's call_rcu_sched
stage on the way can mess up the order before reaching the workqueue.

Instead, when offlining a memcg, call mem_cgroup_reparent_charges() on
all its children (and grandchildren, in the correct order) to have their
charges reparented first.

Fixes: e5fca243abae ("cgroup: use a dedicated workqueue for cgroup destruction")
Signed-off-by: Filipe Brandenburger <filbranden@google.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Reviewed-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: <stable@vger.kernel.org> [v3.10+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
mm/memcontrol.c

index 2d1e5220688cc16b77d0b86b52290e615f0b38db..f50600187f4c1aded2382f93dda141b59427d3a4 100644 (file)
@@ -6360,11 +6360,24 @@ static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg)
 static void mem_cgroup_css_offline(struct cgroup_subsys_state *css)
 {
        struct mem_cgroup *memcg = mem_cgroup_from_css(css);
+       struct cgroup_subsys_state *iter;
 
        kmem_cgroup_css_offline(memcg);
 
        mem_cgroup_invalidate_reclaim_iterators(memcg);
-       mem_cgroup_reparent_charges(memcg);
+
+       /*
+        * This requires that offlining is serialized.  Right now that is
+        * guaranteed because css_killed_work_fn() holds the cgroup_mutex.
+        */
+       rcu_read_lock();
+       css_for_each_descendant_post(iter, css) {
+               rcu_read_unlock();
+               mem_cgroup_reparent_charges(mem_cgroup_from_css(iter));
+               rcu_read_lock();
+       }
+       rcu_read_unlock();
+
        mem_cgroup_destroy_all_caches(memcg);
        vmpressure_cleanup(&memcg->vmpressure);
 }