Re: [patch 2/2] mm: memcontrol: fix missed end-writeback page accounting

[Michal Hocko <mhocko-AlSwsSmVLrQ@public.gmane.org>]

On Thu 23-10-14 10:14:43, Johannes Weiner wrote:
> On Thu, Oct 23, 2014 at 03:03:31PM +0200, Michal Hocko wrote:
> > On Wed 22-10-14 14:29:28, Johannes Weiner wrote:
> > > 0a31bc97c80c ("mm: memcontrol: rewrite uncharge API") changed page
> > > migration to uncharge the old page right away.  The page is locked,
> > > unmapped, truncated, and off the LRU, but it could race with writeback
> > > ending, which then doesn't unaccount the page properly:
> > > 
> > > test_clear_page_writeback()              migration
> > >   acquire pc->mem_cgroup->move_lock
> > 
> > I do not think that mentioning move_lock is important/helpful here
> > because the hot path which is taken all the time (except when there is a
> > task move in progress) doesn't take it.
> > Besides that it is not even relevant for the race.
> 
> You're right.  It's not worth mentioning the transaction setup/finish
> at all, because migration does not participate in that protocol.  How
> about this?  Andrew, could you please copy-paste this into the patch?
> 
> test_clear_page_writeback()              migration
>                                            wait_on_page_writeback()
>   TestClearPageWriteback()
>                                            mem_cgroup_migrate()
>                                              clear PCG_USED
>   mem_cgroup_update_page_stat()
>     if (PageCgroupUsed(pc))
>       decrease memcg pages under writeback

Yes, much better! Thanks!

-- 
Michal Hocko
SUSE Labs