mm/slub.c: replace kick_all_cpus_sync() with synchronize_sched() in kmem_cache_shrink()
authorVladimir Davydov <vdavydov@virtuozzo.com>
Fri, 20 May 2016 00:10:34 +0000 (17:10 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 20 May 2016 02:12:14 +0000 (19:12 -0700)
When we call __kmem_cache_shrink on memory cgroup removal, we need to
synchronize kmem_cache->cpu_partial update with put_cpu_partial that
might be running on other cpus.  Currently, we achieve that by using
kick_all_cpus_sync, which works as a system wide memory barrier.  Though
fast it is, this method has a flaw - it issues a lot of IPIs, which
might hurt high performance or real-time workloads.

To fix this, let's replace kick_all_cpus_sync with synchronize_sched.
Although the latter one may take much longer to finish, it shouldn't be
a problem in this particular case, because memory cgroups are destroyed
asynchronously from a workqueue so that no user visible effects should
be introduced.  OTOH, it will save us from excessive IPIs when someone
removes a cgroup.

Anyway, even if using synchronize_sched turns out to take too long, we
can always introduce a kind of __kmem_cache_shrink batching so that this
method would only be called once per one cgroup destruction (not per
each per memcg kmem cache as it is now).

Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com>
Reported-by: Peter Zijlstra <peterz@infradead.org>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/slub.c

index 4dbb109eb8cd1f5eb03bd82e90285be1090eacd7..ba81cf672f08d083b0a4908c530ad28b3c228a45 100644 (file)
--- a/mm/slub.c
+++ b/mm/slub.c
@@ -3697,7 +3697,7 @@ int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate)
                 * s->cpu_partial is checked locklessly (see put_cpu_partial),
                 * so we have to make sure the change is visible.
                 */
-               kick_all_cpus_sync();
+               synchronize_sched();
        }
 
        flush_all(s);