Re: [PATCH] rcu: Check the validity of the boost_kthread_task pointer

["Paul E. McKenney" <paulmck@xxxxxxxxxx>]

On Thu, Mar 24, 2022 at 07:15:15PM +0800, Zqiang wrote:
> The boost kthreads may spawn failed, although the probability of
> creation failure is very low, in consideration of the robustness
> of the code, add boost_kthread_task pointer detection.
> 
> Signed-off-by: Zqiang <qiang1.zhang@xxxxxxxxx>

Good catch!  As you say, rare, but definitely worth fixing.

I could not resist a bit of wordsmithing.  Could you please check
to see if I messed something up?

							Thanx, Paul

------------------------------------------------------------------------

commit ad3c438e248cbc41fe30a41287621c3dbfcdf51c
Author: Zqiang <qiang1.zhang@xxxxxxxxx>
Date:   Thu Mar 24 19:15:15 2022 +0800

    rcu: Check for successful spawn of ->boost_kthread_task
    
    For the spawning of the priority-boost kthreads can fail, improbable
    though this might seem.  This commit therefore refrains from attemoting
    to initiate RCU priority boosting when The ->boost_kthread_task pointer
    is NULL.
    
    Signed-off-by: Zqiang <qiang1.zhang@xxxxxxxxx>
    Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxx>

diff --git a/kernel/rcu/tree_plugin.h b/kernel/rcu/tree_plugin.h
index 6b9bcd45c7b2..d58f6d6073bf 100644
--- a/kernel/rcu/tree_plugin.h
+++ b/kernel/rcu/tree_plugin.h
@@ -1125,7 +1125,8 @@ static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
 	__releases(rnp->lock)
 {
 	raw_lockdep_assert_held_rcu_node(rnp);
-	if (!rcu_preempt_blocked_readers_cgp(rnp) && rnp->exp_tasks == NULL) {
+	if (!rnp->boost_kthread_task ||
+	    (!rcu_preempt_blocked_readers_cgp(rnp) && !rnp->exp_tasks)) {
 		raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
 		return;
 	}