mm: drain memcg stocks on css offlining

Memcg charge is batched using per-cpu stocks, so an offline memcg can be
pinned by a cached charge up to a moment, when a process belonging to some
other cgroup will charge some memory on the same cpu.  In other words,
cached charges can prevent a memory cgroup from being reclaimed for some
time, without any clear need.

Let's optimize it by explicit draining of all stocks on css offlining.  As
draining is performed asynchronously, and is skipped if any parallel
draining is happening, it's cheap.

Link: http://lkml.kernel.org/r/20180827162621.30187-2-guro@fb.com


Signed-off-by: Roman Gushchin <guro@fb.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Acked-by: Michal Hocko <mhocko@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>