Re: [PATCH v2] locking: Fix __clear_task_blocked_on() warning from __ww_mutex_wound() path

[K Prateek Nayak <kprateek.nayak@amd.com>]

Hello John,

On 8/2/2025 12:51 AM, John Stultz wrote:
> The __clear_task_blocked_on() helper added a number of sanity
> checks ensuring we hold the mutex wait lock and that the task
> we are clearing blocked_on pointer (if set) matches the mutex.
> 
> However, there is an edge case in the _ww_mutex_wound() logic
> where we need to clear the blocked_on pointer for the task that
> owns the mutex, not the task that is waiting on the mutex.
> 
> For this case the sanity checks aren't valid, so handle this
> by allowing a NULL lock to skip the additional checks.
> 
> This was easier to miss, I realized, as the test-ww_mutex
> driver only exercises the wait-die class of ww_mutexes.
> 
> I've got a follow up patch to extend the test so that it
> will exercise both.
> 
> Fixes: a4f0b6fef4b0 ("locking/mutex: Add p->blocked_on wrappers for correctness checks")
> Reported-by: syzbot+602c4720aed62576cd79@syzkaller.appspotmail.com
> Reported-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Closes: https://lore.kernel.org/lkml/68894443.a00a0220.26d0e1.0015.GAE@google.com/
> Signed-off-by: John Stultz <jstultz@google.com>

I've been running this for a while and haven't seen any splats with
syzbot's C reproducer.

> ---
> v2:
> * Rewording of "lock" to "mutex" in commit and comment for
>   clarity
> * Rework __clear_task_blocked_on() to use READ_ONCE and WRITE_ONCE
>   so we don't trip over the WARNING if two instances race, as suggested
>   by K Prateek Nayak and Maarten Lankhorst
> 
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Juri Lelli <juri.lelli@redhat.com>
> Cc: Vincent Guittot <vincent.guittot@linaro.org>
> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> Cc: Valentin Schneider <valentin.schneider@arm.com>
> Cc: K Prateek Nayak <kprateek.nayak@amd.com>
> Cc: Suleiman Souhlal <suleiman@google.com>
> Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> Cc: airlied@gmail.com
> Cc: mripard@kernel.org
> Cc: simona@ffwll.ch
> Cc: tzimmermann@suse.de
> Cc: dri-devel@lists.freedesktop.org
> Cc: kernel-team@android.com
> ---
>  include/linux/sched.h     | 23 +++++++++++++----------
>  kernel/locking/ww_mutex.h |  6 +++++-
>  2 files changed, 18 insertions(+), 11 deletions(-)
> 
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 40d2fa90df425..700b50d29f7fe 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -2166,16 +2166,19 @@ static inline void set_task_blocked_on(struct task_struct *p, struct mutex *m)

Should we consider using WRITE_ONCE() in __set_task_blocked_on() and
use a local copy of "blocked_on" there too?

I think a set_task_blocked_on() on a separate ww_mutex can still race
with a wound on the ww_ctx which indiscriminately writes NULL to
"owner->blocked_on" and can possibly lead to a splat for:

    WARN_ON_ONCE(p->blocked_on && p->blocked_on != m);

                 ^^^^^^^^^^^^^    ^^^^^^^^^^^^^
    Sees p is blocked on "m"      Turns NULL as a result
            already.              of a concurrent wound.

A READ_ONCE() and WRITE_ONCE() in __set_task_blocked_on() should help
solve the splat in this very unlikely case too unless I'm mistaken.

Apart from that, this fix looks good. Feel free to include:

Reviewed-and-tested-by: K Prateek Nayak <kprateek.nayak@amd.com>

-- 
Thanks and Regards,
Prateek

>  
>  static inline void __clear_task_blocked_on(struct task_struct *p, struct mutex *m)
>  {
> -	WARN_ON_ONCE(!m);
> -	/* Currently we serialize blocked_on under the mutex::wait_lock */
> -	lockdep_assert_held_once(&m->wait_lock);
> -	/*
> -	 * There may be cases where we re-clear already cleared
> -	 * blocked_on relationships, but make sure we are not
> -	 * clearing the relationship with a different lock.
> -	 */
> -	WARN_ON_ONCE(m && p->blocked_on && p->blocked_on != m);
> -	p->blocked_on = NULL;
> +	if (m) {
> +		struct mutex *blocked_on = READ_ONCE(p->blocked_on);
> +
> +		/* Currently we serialize blocked_on under the mutex::wait_lock */
> +		lockdep_assert_held_once(&m->wait_lock);
> +		/*
> +		 * There may be cases where we re-clear already cleared
> +		 * blocked_on relationships, but make sure we are not
> +		 * clearing the relationship with a different lock.
> +		 */
> +		WARN_ON_ONCE(blocked_on && blocked_on != m);
> +	}
> +	WRITE_ONCE(p->blocked_on, NULL);
>  }