Re: [PATCH 2/2] rcu/nocb: Fix missed RCU barrier on deoffloading

[Frederic Weisbecker]

Le Mon, Nov 11, 2024 at 01:07:16PM +0530, Neeraj Upadhyay a écrit :
> 
> > diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
> > index 16865475120b..2605dd234a13 100644
> > --- a/kernel/rcu/tree_nocb.h
> > +++ b/kernel/rcu/tree_nocb.h
> > @@ -891,7 +891,18 @@ static void nocb_cb_wait(struct rcu_data *rdp)
> >  	swait_event_interruptible_exclusive(rdp->nocb_cb_wq,
> >  					    nocb_cb_wait_cond(rdp));
> >  	if (kthread_should_park()) {
> > -		kthread_parkme();
> > +		/*
> > +		 * kthread_park() must be preceded by an rcu_barrier().
> > +		 * But yet another rcu_barrier() might have sneaked in between
> > +		 * the barrier callback execution and the callbacks counter
> > +		 * decrement.
> > +		 */
> > +		if (rdp->nocb_cb_sleep) {
> 
> Is READ_ONCE() not required here?

No because it can't be written concurrently at this point. The value observed
here if kthread_should_park() must have been written locally on the previous
call to nocb_cb_wait().

Thanks.


> 
> 
> - Neeraj
> 
> > +			rcu_nocb_lock_irqsave(rdp, flags);
> > +			WARN_ON_ONCE(rcu_segcblist_n_cbs(&rdp->cblist));
> > +			rcu_nocb_unlock_irqrestore(rdp, flags);
> > +			kthread_parkme();
> > +		}
> >  	} else if (READ_ONCE(rdp->nocb_cb_sleep)) {
> >  		WARN_ON(signal_pending(current));
> >  		trace_rcu_nocb_wake(rcu_state.name, rdp->cpu, TPS("WokeEmpty"));
>