Re: 2.6.39-rc6-mmotm0506 - lockdep splat in RCU code on page fault

[Valdis . Kletnieks]

On Tue, 10 May 2011 01:20:29 PDT, "Paul E. McKenney" said:

Would test, but it doesn't apply cleanly to my -mmotm0506 tree:

> diff --git a/kernel/rcutree.c b/kernel/rcutree.c
> index 5616b17..20c22c5 100644
> --- a/kernel/rcutree.c
> +++ b/kernel/rcutree.c
> @@ -1525,13 +1525,15 @@ static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
>   */
>  static void rcu_cpu_kthread_timer(unsigned long arg)
>  {
> -	unsigned long flags;
> +	unsigned long old;
> +	unsigned long new;
>  	struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg);
>  	struct rcu_node *rnp = rdp->mynode;
>  
> -	raw_spin_lock_irqsave(&rnp->lock, flags);
> -	rnp->wakemask |= rdp->grpmask;
> -	raw_spin_unlock_irqrestore(&rnp->lock, flags);
> +	do {
> +		old = rnp->wakemask;
> +		new = old | rdp->grpmask;
> +	} while (cmpxchg(&rnp->wakemask, old, new) != old);
>  	invoke_rcu_node_kthread(rnp);
>  }

My source has this:

        raw_spin_lock_irqsave(&rnp->lock, flags);
        rnp->wakemask |= rdp->grpmask;
        invoke_rcu_node_kthread(rnp);
        raw_spin_unlock_irqrestore(&rnp->lock, flags);

the last 2 lines swapped from what you diffed against.  I can easily work around
that, except it's unclear what the implications of the invoke_rcu moving outside
of the irq save/restore pair (or if it being inside is the actual root cause)...

Attachment: pgp00000.pgp
Description: PGP signature