[Devel] Re: [PATCH v3 06/16] memcg: infrastructure to match an allocation to the right cache

Tejun Heo tj at kernel.org
Mon Sep 24 10:58:01 PDT 2012


On Mon, Sep 24, 2012 at 12:17:37PM +0400, Glauber Costa wrote:
> On 09/22/2012 12:52 AM, Tejun Heo wrote:
> > Missed some stuff.
> > 
> > On Tue, Sep 18, 2012 at 06:12:00PM +0400, Glauber Costa wrote:
> >> +static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg,
> >> +						  struct kmem_cache *cachep)
> >> +{
> > ...
> >> +	memcg->slabs[idx] = new_cachep;
> > ...
> >> +struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep,
> >> +					  gfp_t gfp)
> >> +{
> > ...
> >> +	return memcg->slabs[idx];
> > 
> > I think you need memory barriers for the above pair.
> > 
> > Thanks.
> > 
> 
> Why is that?
> 
> We'll either see a value, or NULL. If we see NULL, we assume the cache
> is not yet created. Not a big deal.

Because when you see !NULL cache pointer you want to be able to see
the cache fully initialized.  You need wmb between cache creation and
pointer assignment and at least read_barrier_depends() between
fetching the cache pointer and dereferencing it.

Thanks.

-- 
tejun




More information about the Devel mailing list