Re: [patch 7/9] mm: memcontrol: do not acquire page_cgroup lock for kmem pages

[Johannes Weiner]

On Fri, May 23, 2014 at 03:39:38PM +0200, Michal Hocko wrote:
> I am adding Vladimir to CC
> 
> On Wed 30-04-14 16:25:41, Johannes Weiner wrote:
> > Kmem page charging and uncharging is serialized by means of exclusive
> > access to the page.  Do not take the page_cgroup lock and don't set
> > pc->flags atomically.
> > 
> > Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx>
> 
> The patch is correct I just have some comments below.
> Anyway
> Acked-by: Michal Hocko <mhocko@xxxxxxx>

Thanks!

> >  mm/memcontrol.c | 16 +++-------------
> >  1 file changed, 3 insertions(+), 13 deletions(-)
> > 
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index c528ae9ac230..d3961fce1d54 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -3535,10 +3535,8 @@ void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg,
> >  	}
> >  
> 
> 	/*
> 	 * given page is newly allocated and invisible to everybody but
> 	 * the caller so there is no need to use page_cgroup lock nor
> 	 * SetPageCgroupUsed
> 	 */
> 
> would be helpful?

That makes sense, I added the following:

+       /*
+        * The page is freshly allocated and not visible to any
+        * outside callers yet.  Set up pc non-atomically.
+        */

> >  	pc = lookup_page_cgroup(page);
> > -	lock_page_cgroup(pc);
> >  	pc->mem_cgroup = memcg;
> > -	SetPageCgroupUsed(pc);
> > -	unlock_page_cgroup(pc);
> > +	pc->flags = PCG_USED;
> >  }
> >  
> >  void __memcg_kmem_uncharge_pages(struct page *page, int order)
> > @@ -3548,19 +3546,11 @@ void __memcg_kmem_uncharge_pages(struct page *page, int order)
> >  
> >  
> >  	pc = lookup_page_cgroup(page);
> > -	/*
> > -	 * Fast unlocked return. Theoretically might have changed, have to
> > -	 * check again after locking.
> > -	 */
> 
> This comment was there since the code has been merged. Maybe it was true
> at the time but after "mm: get rid of __GFP_KMEMCG" it is definitely out
> of date.
> 
> 	/*
> 	 * the pages is going away and will be freed and nobody can see
> 	 * it anymore so no need to take page_cgroup lock.
> 	 */
> >  	if (!PageCgroupUsed(pc))
> >  		return;
> >  
> > -	lock_page_cgroup(pc);
> > -	if (PageCgroupUsed(pc)) {
> > -		memcg = pc->mem_cgroup;
> > -		ClearPageCgroupUsed(pc);
> > -	}
> > -	unlock_page_cgroup(pc);
> 
> maybe add
> 	WARN_ON_ONCE(pc->flags != PCG_USED);
> 
> to check for an unexpected flags usage in the kmem path?

There is no overlap between page types that use PCG_USED and those
that don't.  What would be the value of adding this?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/