On Wed, Apr 22, 2020 at 09:51:20AM -0700, Shakeel Butt wrote: > On Mon, Apr 20, 2020 at 3:11 PM Johannes Weiner <hannes@xxxxxxxxxxx> wrote: > > @@ -5426,15 +5420,23 @@ static int mem_cgroup_move_account(struct page *page, > > } > > > > /* > > + * All state has been migrated, let's switch to the new memcg. > > + * > > * It is safe to change page->mem_cgroup here because the page > > - * is referenced, charged, and isolated - we can't race with > > - * uncharging, charging, migration, or LRU putback. > > + * is referenced, charged, isolated, and locked: we can't race > > + * with (un)charging, migration, LRU putback, or anything else > > + * that would rely on a stable page->mem_cgroup. > > + * > > + * Note that lock_page_memcg is a memcg lock, not a page lock, > > + * to save space. As soon as we switch page->mem_cgroup to a > > + * new memcg that isn't locked, the above state can change > > + * concurrently again. Make sure we're truly done with it. > > */ > > + smp_mb(); > > You said theoretical race in the subject but the above comment > convinced me that smp_mb() is required. So, why is the race still > theoretical? Sorry about the confusion. I said theoretical because I spotted it while thinking about the code. I'm not aware of any real users that suffered the consequences of this race condition. But they could exist in theory :-) I think it's a real bug that needs fixing.
