rt_mutex_setprio() has only one caller: rt_mutex_adjust_prio(). It
expects that task_struct::pi_lock and rt_mutex_base::wait_lock are held.
Both locks are raw_spinlock_t and are acquired with disabled interrupts.

Nevertheless rt_mutex_setprio() disables preemption while invoking
__balance_callbacks() and raw_spin_rq_unlock(). Even if one of the
balance callbacks unlocks the rq then it must not enable interrupts
because rt_mutex_base::wait_lock is still locked.
Therefore interrupts should remain disabled and disabling preemption is
not needed.

Commit 4c9a4bc89a9cc ("sched: Allow balance callbacks for check_class_changed()")
adds a preempt-disable section to rt_mutex_setprio() and
__sched_setscheduler(). In __sched_setscheduler() the preemption is
disabled before rq is unlocked and interrupts enabled but I don't see
why it makes a difference in rt_mutex_setprio().

Remove the preempt_disable() section from rt_mutex_setprio().

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---

v1…v2: https://lore.kernel.org/all/20250513142954.ZM5QSQNc@linutronix.de/
  - Repost without RFC

 kernel/sched/core.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index dce50fa57471d..362d8ab888748 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7292,14 +7292,10 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
 
 	check_class_changed(rq, p, prev_class, oldprio);
 out_unlock:
-	/* Avoid rq from going away on us: */
-	preempt_disable();
 
 	rq_unpin_lock(rq, &rf);
 	__balance_callbacks(rq);
 	raw_spin_rq_unlock(rq);
-
-	preempt_enable();
 }
 #endif
 
-- 
2.49.0

On Tue, Jun 10, 2025 at 04:47:00PM +0200, Sebastian Andrzej Siewior wrote:
> rt_mutex_setprio() has only one caller: rt_mutex_adjust_prio(). It
> expects that task_struct::pi_lock and rt_mutex_base::wait_lock are held.
> Both locks are raw_spinlock_t and are acquired with disabled interrupts.
> 
> Nevertheless rt_mutex_setprio() disables preemption while invoking
> __balance_callbacks() and raw_spin_rq_unlock(). Even if one of the
> balance callbacks unlocks the rq then it must not enable interrupts
> because rt_mutex_base::wait_lock is still locked.
> Therefore interrupts should remain disabled and disabling preemption is
> not needed.
> 
> Commit 4c9a4bc89a9cc ("sched: Allow balance callbacks for check_class_changed()")
> adds a preempt-disable section to rt_mutex_setprio() and
> __sched_setscheduler(). In __sched_setscheduler() the preemption is
> disabled before rq is unlocked and interrupts enabled but I don't see
> why it makes a difference in rt_mutex_setprio().
> 
> Remove the preempt_disable() section from rt_mutex_setprio().
> 
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> ---
> 
> v1…v2: https://lore.kernel.org/all/20250513142954.ZM5QSQNc@linutronix.de/
>   - Repost without RFC
> 
>  kernel/sched/core.c | 4 ----
>  1 file changed, 4 deletions(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index dce50fa57471d..362d8ab888748 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -7292,14 +7292,10 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
>  
>  	check_class_changed(rq, p, prev_class, oldprio);
>  out_unlock:
> -	/* Avoid rq from going away on us: */
> -	preempt_disable();

Perhaps add:

	/* IRQs are still disabled */

or something to that effect such that it is obvious from reading the
code that dropping the lock will not enable preemption?

>  
>  	rq_unpin_lock(rq, &rf);
>  	__balance_callbacks(rq);
>  	raw_spin_rq_unlock(rq);
> -
> -	preempt_enable();
>  }
>  #endif
>  
> -- 
> 2.49.0
> 

On Wed, 11 Jun 2025 11:03:06 +0200
Peter Zijlstra <peterz@infradead.org> wrote:

> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -7292,14 +7292,10 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
> >  
> >  	check_class_changed(rq, p, prev_class, oldprio);
> >  out_unlock:
> > -	/* Avoid rq from going away on us: */
> > -	preempt_disable();  
> 
> Perhaps add:
> 
> 	/* IRQs are still disabled */
> 
> or something to that effect such that it is obvious from reading the
> code that dropping the lock will not enable preemption?

Hmm, wouldn't lockdep_assert_irqs_disabled() be better than a comment.
It lets people know that interrupts are disabled and when lockdep is
enabled it also makes sure they are.

-- Steve

rt_mutex_setprio() has only one caller: rt_mutex_adjust_prio(). It
expects that task_struct::pi_lock and rt_mutex_base::wait_lock are held.
Both locks are raw_spinlock_t and are acquired with disabled interrupts.

Nevertheless rt_mutex_setprio() disables preemption while invoking
__balance_callbacks() and raw_spin_rq_unlock(). Even if the possible
balance callbacks unlock the rq they must not enable interrupts as I
doubt that they also unlock rt_mutex_base::wait_lock.
Therefore interrupts should remain disabled and disabling preemption is
not needed.

Commit 4c9a4bc89a9cc ("sched: Allow balance callbacks for check_class_changed()")
adds a preempt-disable section to rt_mutex_setprio() and
__sched_setscheduler(). In __sched_setscheduler() the preemption is
disabled before rq is unlocked and interrupts enabled but I don't see
why it makes a difference in rt_mutex_setprio().

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
v2…v3: https://lore.kernel.org/all/20250610144700.uBK6RS95@linutronix.de/
  - Add a comment why IRQs are still disabled.

v1…v2: https://lore.kernel.org/all/20250513142954.ZM5QSQNc@linutronix.de/
  - Repost without RFC

 kernel/sched/core.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index dce50fa57471d..537212b33fd84 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7292,14 +7292,11 @@ void rt_mutex_setprio(struct task_struct *p, struct task_struct *pi_task)
 
 	check_class_changed(rq, p, prev_class, oldprio);
 out_unlock:
-	/* Avoid rq from going away on us: */
-	preempt_disable();
+	/* Caller holds task_struct::pi_lock, IRQs are still disabled */
 
 	rq_unpin_lock(rq, &rf);
 	__balance_callbacks(rq);
 	raw_spin_rq_unlock(rq);
-
-	preempt_enable();
 }
 #endif
 
-- 
2.49.0