Re: [tip:sched/core] [sched] b079d93796: WARNING:possible_recursive_locking_detected_migration_is_trying_to_acquire_lock:at:set_cpus_allowed_force_but_task_is_already_holding_lock:at:cpu_stopper_thread

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Oct 27, 2025 at 01:14:09PM +0800, kernel test robot wrote:

> kernel test robot noticed "WARNING:possible_recursive_locking_detected_migration_is_trying_to_acquire_lock:at:set_cpus_allowed_force_but_task_is_already_holding_lock:at:cpu_stopper_thread" on:
> 
> commit: b079d93796528053cde322f2ca838c2d21c297e7 ("sched: Rename do_set_cpus_allowed()")
> https://git.kernel.org/cgit/linux/kernel/git/tip/tip.git sched/core

Your biscect went sideways, it is, as Jan correctly found:

  abfc01077df6 ("sched: Fix do_set_cpus_allowed() locking")


Anyway, this was helpful:

> [  116.814488][   T21] ============================================
> [  116.815227][   T21] WARNING: possible recursive locking detected
> [  116.815957][   T21] 6.18.0-rc1-00014-gb079d9379652 #1 Tainted: G S                 
> [  116.816878][   T21] --------------------------------------------
> [  116.817602][   T21] migration/1/21 is trying to acquire lock:
> [  116.818301][   T21] ee7f1930 (&rq->__lock){-.-.}-{2:2}, at: set_cpus_allowed_force+0x3c/0xc0
> [  116.820432][   T21] 
> [  116.820432][   T21] but task is already holding lock:
> [  116.821314][   T21] ee7f1930 (&rq->__lock){-.-.}-{2:2}, at: cpu_stopper_thread+0x93/0x170

> [  116.841003][   T21] 
> [  116.842427][   T21] 2 locks held by migration/1/21:
> [  116.843393][   T21]  #0: b92d06dc (&p->pi_lock){-.-.}-{2:2}, at: __balance_push_cpu_stop+0x28/0x2b0
> [  116.845044][   T21]  #1: ee7f1930 (&rq->__lock){-.-.}-{2:2}, at: cpu_stopper_thread+0x93/0x170
> [  116.846669][   T21] 
> [  116.846669][   T21] stack backtrace:
> [  116.847890][   T21] CPU: 1 UID: 0 PID: 21 Comm: migration/1 Tainted: G S                  6.18.0-rc1-00014-gb079d9379652 #1 NONE  6d63d2e836521c1c681a07c673117fb98e4815ab
> [  116.847897][   T21] Tainted: [S]=CPU_OUT_OF_SPEC
> [  116.847898][   T21] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014
> [  116.847901][   T21] Stopper: __balance_push_cpu_stop+0x0/0x2b0 <- finish_lock_switch+0x7d/0xd0
> [  116.847909][   T21] Call Trace:

> [  116.847939][   T21]  ? lock_acquire+0xc3/0x1f0
> [  116.847943][   T21]  ? set_cpus_allowed_force+0x3c/0xc0
> [  116.847947][   T21]  ? lock_acquire+0xc3/0x1f0
> [  116.847952][   T21]  ? __task_rq_lock+0x73/0x1d0
> [  116.847955][   T21]  ? set_cpus_allowed_force+0x3c/0xc0
> [  116.847959][   T21]  ? set_cpus_allowed_force+0x3c/0xc0
> [  116.847962][   T21]  ? __balance_push_cpu_stop+0x136/0x2b0
> [  116.847966][   T21]  ? select_fallback_rq+0x148/0x230
> [  116.847970][   T21]  ? __balance_push_cpu_stop+0x163/0x2b0
> [  116.847974][   T21]  ? cpu_stopper_thread+0x93/0x170

Clearly I missed that case :/

---
Subject: sched: Fix the do_set_cpus_allowed() locking fix

Commit abfc01077df6 ("sched: Fix do_set_cpus_allowed() locking")
overlooked that __balance_push_cpu_stop() calls select_fallback_rq()
with rq->lock held. This makes that set_cpus_allowed_force() will
recursively take rq->lock and the machine locks up.

Run select_fallback_rq() earlier, without holding rq->lock. This opens
up a race window where a task could get migrated out from under us, but
that is harmless, we want the task migrated.

select_fallback_rq() itself will not be subject to concurrency as it
will be fully serialized by p->pi_lock, so there is no chance of
set_cpus_allowed_force() getting called with different arguments and
selecting different fallback CPUs for one task.

Fixes: abfc01077df6 ("sched: Fix do_set_cpus_allowed() locking")
Reported-by: Jan Polensky <japo@linux.ibm.com>
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Closes: https://lore.kernel.org/oe-lkp/202510271206.24495a68-lkp@intel.com
---
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 1842285eac1e..67b5f2faab36 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -8044,18 +8044,15 @@ static int __balance_push_cpu_stop(void *arg)
 	struct rq_flags rf;
 	int cpu;
 
-	raw_spin_lock_irq(&p->pi_lock);
-	rq_lock(rq, &rf);
-
-	update_rq_clock(rq);
-
-	if (task_rq(p) == rq && task_on_rq_queued(p)) {
+	scoped_guard (raw_spinlock_irq, &p->pi_lock) {
 		cpu = select_fallback_rq(rq->cpu, p);
-		rq = __migrate_task(rq, &rf, p, cpu);
-	}
 
-	rq_unlock(rq, &rf);
-	raw_spin_unlock_irq(&p->pi_lock);
+		rq_lock(rq, &rf);
+		update_rq_clock(rq);
+		if (task_rq(p) == rq && task_on_rq_queued(p))
+			rq = __migrate_task(rq, &rf, p, cpu);
+		rq_unlock(rq, &rf);
+	}
 
 	put_task_struct(p);