[Devel] Re: [PATCH v5 08/18] memcg: infrastructure to match an allocation to the right cache

[Tejun Heo]

Hello, Glauber.

On Thu, Oct 25, 2012 at 03:05:22PM +0400, Glauber Costa wrote:
> > Is there any rmb() pair?
> > As far as I know, without rmb(), wmb() doesn't guarantee anything.
> > 
> 
> There should be. But it seems I missed it. Speaking of which, I should

You probably can use read_barrier_depends().

> wmb() after the NULL assignment in release cache as well.

And you probably don't need that.  dangling pointer is protected by
RCU and there isn't any memory accesses which can paired with that
anyway.

Thanks.

-- 
tejun