Re: [PATCH v4] sched/fair: Skip sched_balance_running cmpxchg when balance is not due

[Peter Zijlstra]

On Wed, Nov 12, 2025 at 11:37:40AM +0100, Peter Zijlstra wrote:
> On Wed, Nov 12, 2025 at 01:32:23PM +0530, Srikar Dronamraju wrote:
> > >  	group = sched_balance_find_src_group(&env);
> > >  	if (!group) {
> > >  		schedstat_inc(sd->lb_nobusyg[idle]);
> > > @@ -11892,6 +11916,9 @@ static int sched_balance_rq(int this_cpu, struct rq *this_rq,
> > >  			if (!cpumask_subset(cpus, env.dst_grpmask)) {
> > >  				env.loop = 0;
> > >  				env.loop_break = SCHED_NR_MIGRATE_BREAK;
> > > +				if (need_unlock)
> > > +					atomic_set_release(&sched_balance_running, 0);
> > > +
> > 
> > One nit:
> > While the current code is good, would conditionally resetting the
> > need_unlock just after resetting the atomic variable better than
> > unconditional reset that we do now?
> 
> Right, I had the same thought when grabbed the patch yesterday, but
> ignored it.
> 

Hmm, should we not redo while keeping the lock? Doesn't make much sense
to drop and try to reacquire things here.

So perhaps this is the better option -- or did I overlook something with
should_we_balance? It doesn't look like that will make a different
decision on the retry.

---
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -11717,26 +11717,25 @@ static int sched_balance_rq(int this_cpu
 		.fbq_type	= all,
 		.tasks		= LIST_HEAD_INIT(env.tasks),
 	};
-	bool need_unlock;
+	bool need_unlock = false;
 
 	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
 
 	schedstat_inc(sd->lb_count[idle]);
 
-redo:
-	need_unlock = false;
 	if (!should_we_balance(&env)) {
 		*continue_balancing = 0;
 		goto out_balanced;
 	}
 
 	if (sd->flags & SD_SERIALIZE) {
-		if (atomic_cmpxchg_acquire(&sched_balance_running, 0, 1)) {
+		if (!atomic_try_cmpxchg_acquire(&sched_balance_running, 0, 1))
 			goto out_balanced;
-		}
+
 		need_unlock = true;
 	}
 
+redo:
 	group = sched_balance_find_src_group(&env);
 	if (!group) {
 		schedstat_inc(sd->lb_nobusyg[idle]);
@@ -11861,9 +11860,6 @@ static int sched_balance_rq(int this_cpu
 			if (!cpumask_subset(cpus, env.dst_grpmask)) {
 				env.loop = 0;
 				env.loop_break = SCHED_NR_MIGRATE_BREAK;
-				if (need_unlock)
-					atomic_set_release(&sched_balance_running, 0);
-
 				goto redo;
 			}
 			goto out_all_pinned;