Re: [PATCH 3/4] memcg: get rid of percpu_charge_mutex lock

[Michal Hocko]

On Thu 21-07-11 13:47:04, Michal Hocko wrote:
> On Thu 21-07-11 19:30:51, KAMEZAWA Hiroyuki wrote:
> > On Thu, 21 Jul 2011 09:58:24 +0200
> > Michal Hocko <mhocko@xxxxxxx> wrote:
[...]
> > > --- a/mm/memcontrol.c
> > > +++ b/mm/memcontrol.c
> > > @@ -2166,7 +2165,8 @@ static void drain_all_stock(struct mem_cgroup *root_mem, bool sync)
> > >  
> > >  	for_each_online_cpu(cpu) {
> > >  		struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu);
> > > -		if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
> > > +		if (root_mem == stock->cached &&
> > > +				test_bit(FLUSHING_CACHED_CHARGE, &stock->flags))
> > >  			flush_work(&stock->work);
> > 
> > Doesn't this new check handle hierarchy ?
> > css_is_ancestor() will be required if you do this check.
> 
> Yes you are right. Will fix it. I will add a helper for the check.

Here is the patch with the helper. The above will then read 
	if (mem_cgroup_same_or_subtree(root_mem, stock->cached))

---