Re: [PATCH 3/9] bio-cgroup controller

[Andrea Righi <righi.andrea@gmail.com>]

On Mon, Apr 20, 2009 at 08:35:40PM +0900, Ryo Tsuruta wrote:
> > > +/*
> > > + * Assign "page" the same owner as "opage."
> > > + */
> > > +void bio_cgroup_copy_owner(struct page *npage, struct page *opage)
> > > +{
> > > +	struct page_cgroup *npc, *opc;
> > > +
> > > +	if (bio_cgroup_disabled())
> > > +		return;
> > > +	npc = lookup_page_cgroup(npage);
> > > +	if (unlikely(!npc))
> > > +		return;
> > > +	opc = lookup_page_cgroup(opage);
> > > +	if (unlikely(!opc))
> > > +		return;
> > > +
> > > +	/*
> > > +	 * Do this without any locks. The reason is the same as
> > > +	 * bio_cgroup_reset_owner().
> > > +	 */
> > > +	npc->bio_cgroup_id = opc->bio_cgroup_id;
> > 
> > What protects npc and opc?
> 
> As the same reason mentioned above, bio_cgroup_id can be updated
> without any locks, and npc and opc always point to page_cgroups.
> An integer variable can be set a new value at once on a system which
> can use RCU lock.

mmmh... I'm not sure about this. Actually you read opc->bio_cgroup_id
first and then write to npc->bio_cgroup_id, so it is not atomic at all.
So, you can read or set a wrong ID, but at least it should be always
consistent (the single read or write itself is atomic).

-Andrea