Re: [Experimental][PATCH 19/21] memcg-fix-pre-destroy.patch

[Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>]

On Thu, 4 Dec 2008 18:43:09 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> wrote:
> On Thu, 4 Dec 2008 18:34:28 +0900
> Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote:
> 
> > Added CC: Paul Menage <menage@google.com>
> > 
> > > @@ -2096,7 +2112,7 @@ static void mem_cgroup_get(struct mem_cg
> > >  static void mem_cgroup_put(struct mem_cgroup *mem)
> > >  {
> > >  	if (atomic_dec_and_test(&mem->refcnt)) {
> > > -		if (!mem->obsolete)
> > > +		if (!css_under_removal(&mem->css))
> > >  			return;
> > >  		mem_cgroup_free(mem);
> > >  	}
> > I don't think it's safe to check css_under_removal here w/o cgroup_lock.
> > (It's safe *NOW* just because memcg is the only user of css->refcnt.)
> > 
> 
> > As Li said before, css_under_removal doesn't necessarily mean
> > this this group has been destroyed, but mem_cgroup will be freed.
> > 
> > But adding cgroup_lock/unlock here causes another dead lock,
> > because mem_cgroup_get_next_node calls mem_cgroup_put.
> > 
> > hmm.. hierarchical reclaim code will be re-written completely by [21/21],
> > so would it be better to change patch order or to take another approach ?
> > 
> Hmm, ok.
> 
> How about this ?
> ==
> 	At initlization, mem_cgroup_create(), set memcg->refcnt to be 1.
> 
> 	At destroy(), put this refcnt by 1.
> 
> 	remove css_under_removal(&mem->css) check.
> ==
> 
would be make sence.

Thanks,
Daisuke Nishimura.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>