[PATCH] epoll: use RCU protect wakeup_source in epitem

[Eric Wong]

Eric Dumazet <eric.dumazet@xxxxxxxxx> wrote:
> On Sun, 2013-03-10 at 01:11 +0000, Eric Wong wrote:
> >  
> >  static void ep_destroy_wakeup_source(struct epitem *epi)
> >  {
> > -	wakeup_source_unregister(epi->ws);
> > -	epi->ws = NULL;
> > +	struct wakeup_source *ws = epi->ws;
> > +
> > +	rcu_assign_pointer(epi->ws, NULL);
> 
> There is no need to use a memory barrier before setting epi->ws to NULL
> 
> You should use RCU_POINTER_INIT()

Thanks for looking at this patch.  Using RCU_POINTER_INIT in my
updated patch below.

> > +	synchronize_rcu(); /* wait for ep_pm_stay_awake to finish */
> 
> Wont this add a significant slowdown ?

Maybe, but this is a very rare code path (using EPOLL_CTL_MOD to remove
a wakeup source), and synchronize_rcu gets called later, too:

	wakeup_source_unregister
		wakeup_source_remove
			synchronize_rcu

> > +	wakeup_source_unregister(ws);
> >  }
> 
> Please add the following in your .config
> 
> CONFIG_SPARSE_RCU_POINTER=y
> 
> and try :
> 
> make C=2 fs/eventpoll.o
> 
> And fix all warnings ;)

Done.

---------------------------------8<------------------------------