On Fri, Jan 22, 2016 at 04:50:42PM +0300, Vladimir Davydov wrote:

From first glance, it looks like the bug was triggered, because
mem_cgroup_css_offline was run for a child cgroup earlier than for its
parent. This couldn't happen for sure before the cgroup was switched to
percpu_ref, because cgroup_destroy_wq has always had max_active == 1.
Now, however, it looks like this is perfectly possible for
css_killed_ref_fn is called from an rcu callback - see kill_css ->
percpu_ref_kill_and_confirm. This breaks kmemcg assumptions.

I'll take a look what can be done about that.

It's an acknowledged problem in the cgroup core then, and not an issue
with kmemcg. Tejun sent a fix to correct the offlining order here:

https://www.mail-archive.com/linux-kernel@vger.kernel.org/msg1056544.html

--
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>