With PREEMPT_RT enabled, some of the calls to put_task_struct() coming
from rt_mutex_adjust_prio_chain() could happen in preemptible context and
with a mutex enqueued. That could lead to this sequence:

        rt_mutex_adjust_prio_chain()
          put_task_struct()
            __put_task_struct()
              sched_ext_free()
                spin_lock_irqsave()
                  rtlock_lock() --->  TRIGGERS
                                      lockdep_assert(!current->pi_blocked_on);

Fix that by unconditionally resorting to the deferred call to
__put_task_struct() if PREEMPT_RT is enabled.

Suggested-by: Crystal Wood <crwood@redhat.com>
Fixes: 893cdaaa3977 ("sched: avoid false lockdep splat in put_task_struct()")
Signed-off-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>
---

v2: (Rostedt) remove the #ifdef from put_task_struct() and create
    tsk_is_pi_blocked_on() in sched.h to make the change cleaner.
v3: (Sebastian, PeterZ) always call the deferred __put_task_struct() on RT.
v4: Fix the implementation of what was requested on v3.
v5: Add the "Fixes:" tag.

 include/linux/sched/task.h |   17 ++++++++---------
 1 file changed, 8 insertions(+), 9 deletions(-)

diff --git a/include/linux/sched/task.h b/include/linux/sched/task.h
index 0f2aeb37bbb04..51678a541477a 100644
--- a/include/linux/sched/task.h
+++ b/include/linux/sched/task.h
@@ -134,11 +134,8 @@ static inline void put_task_struct(struct task_struct *t)
 	if (!refcount_dec_and_test(&t->usage))
 		return;
 
-	/*
-	 * In !RT, it is always safe to call __put_task_struct().
-	 * Under RT, we can only call it in preemptible context.
-	 */
-	if (!IS_ENABLED(CONFIG_PREEMPT_RT) || preemptible()) {
+	/* In !RT, it is always safe to call __put_task_struct(). */
+	if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
 		static DEFINE_WAIT_OVERRIDE_MAP(put_task_map, LD_WAIT_SLEEP);
 
 		lock_map_acquire_try(&put_task_map);
@@ -148,11 +145,13 @@ static inline void put_task_struct(struct task_struct *t)
 	}
 
 	/*
-	 * under PREEMPT_RT, we can't call put_task_struct
+	 * Under PREEMPT_RT, we can't call __put_task_struct
 	 * in atomic context because it will indirectly
-	 * acquire sleeping locks.
+	 * acquire sleeping locks. The same is true if the
+	 * current process has a mutex enqueued (blocked on
+	 * a PI chain).
 	 *
-	 * call_rcu() will schedule delayed_put_task_struct_rcu()
+	 * call_rcu() will schedule __put_task_struct_rcu_cb()
 	 * to be called in process context.
 	 *
 	 * __put_task_struct() is called when
@@ -165,7 +164,7 @@ static inline void put_task_struct(struct task_struct *t)
 	 *
 	 * delayed_free_task() also uses ->rcu, but it is only called
 	 * when it fails to fork a process. Therefore, there is no
-	 * way it can conflict with put_task_struct().
+	 * way it can conflict with __put_task_struct().
 	 */
 	call_rcu(&t->rcu, __put_task_struct_rcu_cb);
 }

On 2025-06-17 11:00:36 [-0300], Luis Claudio R. Goncalves wrote:
> With PREEMPT_RT enabled, some of the calls to put_task_struct() coming
> from rt_mutex_adjust_prio_chain() could happen in preemptible context and
> with a mutex enqueued. That could lead to this sequence:
> 
>         rt_mutex_adjust_prio_chain()
>           put_task_struct()
>             __put_task_struct()
>               sched_ext_free()
>                 spin_lock_irqsave()
>                   rtlock_lock() --->  TRIGGERS
>                                       lockdep_assert(!current->pi_blocked_on);

Maybe with the addition of

| The first case was observed with sched_ext_free(). 
| Crystal Wood was able to reproduce the problem to __put_task_struct()
| being called during rt_mutex_adjust_prio_chain().

The first sentence will imply a Fixes: with the introduction of
sched_ext. The second implies that the original fix was not complete and
nobody managed to trigger it until now.

> Fix that by unconditionally resorting to the deferred call to
> __put_task_struct() if PREEMPT_RT is enabled.
> 
> Suggested-by: Crystal Wood <crwood@redhat.com>
> Fixes: 893cdaaa3977 ("sched: avoid false lockdep splat in put_task_struct()")
> Signed-off-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>

Sebastian

On Wed, 2025-06-18 at 09:03 +0200, Sebastian Andrzej Siewior wrote:
> On 2025-06-17 11:00:36 [-0300], Luis Claudio R. Goncalves wrote:
> > With PREEMPT_RT enabled, some of the calls to put_task_struct() coming
> > from rt_mutex_adjust_prio_chain() could happen in preemptible context and
> > with a mutex enqueued. That could lead to this sequence:
> > 
> >         rt_mutex_adjust_prio_chain()
> >           put_task_struct()
> >             __put_task_struct()
> >               sched_ext_free()
> >                 spin_lock_irqsave()
> >                   rtlock_lock() --->  TRIGGERS
> >                                       lockdep_assert(!current->pi_blocked_on);
> 
> Maybe with the addition of
> 
> > The first case was observed with sched_ext_free(). 
> > Crystal Wood was able to reproduce the problem to __put_task_struct()
> > being called during rt_mutex_adjust_prio_chain().
> 
> The first sentence will imply a Fixes: with the introduction of
> sched_ext. The second implies that the original fix was not complete and
> nobody managed to trigger it until now.

sched_ext_free() just happens to be the first cleanup function called,
so that's where the blowup happens.  I think the "nobody managed to
trigger it" was because we didn't have the pi_blocked_on assert until
recently -- and my "other cases with a similar cause" was probably older
kernels with the assert backported, but not sched_ext, so the backtrace
was different.

-Crystal

On Tue, Jun 17, 2025 at 11:00:36AM -0300, Luis Claudio R. Goncalves wrote:
> With PREEMPT_RT enabled, some of the calls to put_task_struct() coming
> from rt_mutex_adjust_prio_chain() could happen in preemptible context and
> with a mutex enqueued. That could lead to this sequence:
> 
>         rt_mutex_adjust_prio_chain()
>           put_task_struct()
>             __put_task_struct()
>               sched_ext_free()
>                 spin_lock_irqsave()
>                   rtlock_lock() --->  TRIGGERS
>                                       lockdep_assert(!current->pi_blocked_on);
> 
> Fix that by unconditionally resorting to the deferred call to
> __put_task_struct() if PREEMPT_RT is enabled.
> 
> Suggested-by: Crystal Wood <crwood@redhat.com>
> Fixes: 893cdaaa3977 ("sched: avoid false lockdep splat in put_task_struct()")
> Signed-off-by: Luis Claudio R. Goncalves <lgoncalv@redhat.com>
> ---
> 
> v2: (Rostedt) remove the #ifdef from put_task_struct() and create
>     tsk_is_pi_blocked_on() in sched.h to make the change cleaner.
> v3: (Sebastian, PeterZ) always call the deferred __put_task_struct() on RT.
> v4: Fix the implementation of what was requested on v3.
> v5: Add the "Fixes:" tag.
> 
>  include/linux/sched/task.h |   17 ++++++++---------
>  1 file changed, 8 insertions(+), 9 deletions(-)
> 
> diff --git a/include/linux/sched/task.h b/include/linux/sched/task.h
> index 0f2aeb37bbb04..51678a541477a 100644
> --- a/include/linux/sched/task.h
> +++ b/include/linux/sched/task.h
> @@ -134,11 +134,8 @@ static inline void put_task_struct(struct task_struct *t)
>  	if (!refcount_dec_and_test(&t->usage))
>  		return;
>  
> -	/*
> -	 * In !RT, it is always safe to call __put_task_struct().
> -	 * Under RT, we can only call it in preemptible context.
> -	 */
> -	if (!IS_ENABLED(CONFIG_PREEMPT_RT) || preemptible()) {
> +	/* In !RT, it is always safe to call __put_task_struct(). */
> +	if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
>  		static DEFINE_WAIT_OVERRIDE_MAP(put_task_map, LD_WAIT_SLEEP);
>  
>  		lock_map_acquire_try(&put_task_map);
> @@ -148,11 +145,13 @@ static inline void put_task_struct(struct task_struct *t)
>  	}
>  
>  	/*
> -	 * under PREEMPT_RT, we can't call put_task_struct
> +	 * Under PREEMPT_RT, we can't call __put_task_struct
>  	 * in atomic context because it will indirectly
> -	 * acquire sleeping locks.
> +	 * acquire sleeping locks. The same is true if the
> +	 * current process has a mutex enqueued (blocked on
> +	 * a PI chain).
>  	 *
> -	 * call_rcu() will schedule delayed_put_task_struct_rcu()
> +	 * call_rcu() will schedule __put_task_struct_rcu_cb()
>  	 * to be called in process context.
>  	 *
>  	 * __put_task_struct() is called when
> @@ -165,7 +164,7 @@ static inline void put_task_struct(struct task_struct *t)
>  	 *
>  	 * delayed_free_task() also uses ->rcu, but it is only called
>  	 * when it fails to fork a process. Therefore, there is no
> -	 * way it can conflict with put_task_struct().
> +	 * way it can conflict with __put_task_struct().
>  	 */
>  	call_rcu(&t->rcu, __put_task_struct_rcu_cb);
>  }
> 

Reviewed-by: Wander Lairson Costa <wander@redhat.com>