Re: [linus:master] [sched/core] 704069649b: kernel-selftests.kvm.hardware_disable_test.fail

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Feb 18, 2026 at 10:40:53AM +0100, Peter Zijlstra wrote:
> On Fri, Feb 13, 2026 at 02:14:14PM -0800, Sean Christopherson wrote:
> > On Fri, Feb 13, 2026, Peter Zijlstra wrote:
> > > On Thu, Feb 12, 2026 at 10:08:04PM +0800, kernel test robot wrote:
> > > > Hello,
> > > > 
> > > > we found the kernel-selftests.kvm.hardware_disable_test failed consistently upon
> > > > this commit but pass on parent. unfortunately, we didn't find many useful
> > > > information in dmesg. this report is just FYI what we observed in our tests.
> > > > 
> > > > kernel test robot noticed "kernel-selftests.kvm.hardware_disable_test.fail" on:
> > > 
> > > With the caveat of PEBKAC (it is Friday after all); I can't reproduce.
> 
> PEBCAK indeed, I managed to consistently boot into the bad kernel :-(
> So never actually tested the good one.
> 
> > > That is, ./hardware_disable_test as build from cee73b1e840c, doesn't
> > > work for me on 704069649b5b^1 either.
> > > 
> > > Sean; is there a magic trick to operating that test, or is it a known
> > > trouble spot?
> > 
> > Hmm, shouldn't require any magic, and hasn't been known to be flaky.
> > 
> > This very decisively points at 704069649b5b ("sched/core: Rework
> > sched_class::wakeup_preempt() and rq_modified_*()"). on my end as well.  With
> > that commit reverted, the below runs in ~40ms total.  With 704069649b5b present,
> > the test constantly stalls for multiple seconds at sem_timedwait().
> > 
> > AFAICT, the key is to have the busy_loop() pthread affined to the same CPU as
> > its parent.  The KVM pieces of the selftest have nothing to do with the failure.
> > 
> > Here's a minimal reproducer that you can build without selftests goo :-)
> > E.g. `gcc -pthread -o busy busy.c` should work.
> 
> Whee, thanks! OK, let me go prod at this with something sharp, see what
> falls out.

This seems to work. I'll go write a Changelog after dinner.

---
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 759777694c78..b7f77c165a6e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6830,6 +6830,7 @@ static void __sched notrace __schedule(int sched_mode)
 		/* SCX must consult the BPF scheduler to tell if rq is empty */
 		if (!rq->nr_running && !scx_enabled()) {
 			next = prev;
+			rq->next_class = &idle_sched_class;
 			goto picked;
 		}
 	} else if (!preempt && prev_state) {
diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index c18e81e8ef51..3afa96d5d1ba 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -2460,7 +2460,7 @@ do_pick_task_scx(struct rq *rq, struct rq_flags *rf, bool force_scx)
 	/* see kick_cpus_irq_workfn() */
 	smp_store_release(&rq->scx.kick_sync, rq->scx.kick_sync + 1);
 
-	rq->next_class = &ext_sched_class;
+	rq_modified_begin(rq, &ext_sched_class);
 
 	rq_unpin_lock(rq, rf);
 	balance_one(rq, prev);
@@ -2475,7 +2475,7 @@ do_pick_task_scx(struct rq *rq, struct rq_flags *rf, bool force_scx)
 	 * If @force_scx is true, always try to pick a SCHED_EXT task,
 	 * regardless of any higher-priority sched classes activity.
 	 */
-	if (!force_scx && sched_class_above(rq->next_class, &ext_sched_class))
+	if (!force_scx && rq_modified_above(rq, &ext_sched_class))
 		return RETRY_TASK;
 
 	keep_prev = rq->scx.flags & SCX_RQ_BAL_KEEP;
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 1e22b7fadd70..8e5864ec4cf9 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -12908,7 +12908,7 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf)
 	t0 = sched_clock_cpu(this_cpu);
 	__sched_balance_update_blocked_averages(this_rq);
 
-	this_rq->next_class = &fair_sched_class;
+	rq_modified_begin(this_rq, &fair_sched_class);
 	raw_spin_rq_unlock(this_rq);
 
 	for_each_domain(this_cpu, sd) {
@@ -12975,7 +12975,7 @@ static int sched_balance_newidle(struct rq *this_rq, struct rq_flags *rf)
 		pulled_task = 1;
 
 	/* If a higher prio class was modified, restart the pick */
-	if (sched_class_above(this_rq->next_class, &fair_sched_class))
+	if (rq_modified_above(this_rq, &fair_sched_class))
 		pulled_task = -1;
 
 out:
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index b82fb70a9d54..43bbf0693cca 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2748,6 +2748,17 @@ static inline const struct sched_class *next_active_class(const struct sched_cla
 
 #define sched_class_above(_a, _b)	((_a) < (_b))
 
+static inline void rq_modified_begin(struct rq *rq, const struct sched_class *class)
+{
+	if (sched_class_above(rq->next_class, class))
+		rq->next_class = class;
+}
+
+static inline bool rq_modified_above(struct rq *rq, const struct sched_class *class)
+{
+	return sched_class_above(rq->next_class, class);
+}
+
 static inline bool sched_stop_runnable(struct rq *rq)
 {
 	return rq->stop && task_on_rq_queued(rq->stop);