Re: [PATCH 2/3] rcu/exp: Remove confusing needless full barrier on task unblock

[Joel Fernandes <joelagnelf@nvidia.com>]

On Fri, Feb 14, 2025 at 12:25:58AM +0100, Frederic Weisbecker wrote:
> A full memory barrier in the RCU-PREEMPT task unblock path advertizes
> to order the context switch (or rather the accesses prior to
> rcu_read_unlock()) with the expedited grace period fastpath.
> 
> However the grace period can not complete without the rnp calling into
> rcu_report_exp_rnp() with the node locked. This reports the quiescent
> state in a fully ordered fashion against updater's accesses thanks to:
> 
> 1) The READ-SIDE smp_mb__after_unlock_lock() barrier accross nodes
>    locking while propagating QS up to the root.
> 
> 2) The UPDATE-SIDE smp_mb__after_unlock_lock() barrier while holding the
>    the root rnp to wait/check for the GP completion.
> 
> 3) The (perhaps redundant given step 1) and 2)) smp_mb() in rcu_seq_end()
>    before the grace period completes.
> 
> This makes the explicit barrier in this place superflous. Therefore
> remove it as it is confusing.
> 
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
> ---
>  kernel/rcu/tree_plugin.h | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
> index 3c0bbbbb686f..d51cc7a5dfc7 100644
> --- a/kernel/rcu/tree_plugin.h
> +++ b/kernel/rcu/tree_plugin.h
> @@ -534,7 +534,6 @@ rcu_preempt_deferred_qs_irqrestore(struct task_struct *t, unsigned long flags)
>  		WARN_ON_ONCE(rnp->completedqs == rnp->gp_seq &&
>  			     (!empty_norm || rnp->qsmask));
>  		empty_exp = sync_rcu_exp_done(rnp);
> -		smp_mb(); /* ensure expedited fastpath sees end of RCU c-s. */

I was wondering though, this is a slow path and the smp_mb() has been there
since 2009 or so. Not sure if it is super valuable to remove it at this
point. But we/I should definitely understand it.

I was wondering if you could also point to the fastpath that this is racing
with, it is not immediately clear (to me) what this smp_mb() is pairing with :(

thanks,

 - Joel





>  		np = rcu_next_node_entry(t, rnp);
>  		list_del_init(&t->rcu_node_entry);
>  		t->rcu_blocked_node = NULL;
> -- 
> 2.46.0
>