sched: Increase sched_tick_remote timeout

[PATCH] sched: Increase sched_tick_remote timeout

Posted by Phil Auld 3 weeks ago

Increase the sched_tick_remote WARN_ON timeout to remove false
positives due to temporarily busy HK cpus. The suggestion
was 30 seconds to catch really stuck remote tick processing
but not trigger it too easily.

Signed-off-by: Phil Auld <pauld@redhat.com>
Suggested-by: Frederic Weisbecker <frederic@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/sched/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index be00629f0ba4..ef90d358252d 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -5724,7 +5724,7 @@ static void sched_tick_remote(struct work_struct *work)
 				 * reasonable amount of time.
 				 */
 				u64 delta = rq_clock_task(rq) - curr->se.exec_start;
-				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
+				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 30);
 			}
 			curr->sched_class->task_tick(rq, curr, 0);
 
-- 
2.51.0

Re: [PATCH] sched: Increase sched_tick_remote timeout

Posted by Phil Auld 1 week, 2 days ago

Hi,

On Thu, Sep 11, 2025 at 12:13:00PM -0400 Phil Auld wrote:
> Increase the sched_tick_remote WARN_ON timeout to remove false
> positives due to temporarily busy HK cpus. The suggestion
> was 30 seconds to catch really stuck remote tick processing
> but not trigger it too easily.
> 
> Signed-off-by: Phil Auld <pauld@redhat.com>
> Suggested-by: Frederic Weisbecker <frederic@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Frederic Weisbecker <frederic@kernel.org>

Frederic ack'd this. Any other thoughts or opinions on this one
character patch?

Cheers,
Phil



> ---
>  kernel/sched/core.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index be00629f0ba4..ef90d358252d 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -5724,7 +5724,7 @@ static void sched_tick_remote(struct work_struct *work)
>  				 * reasonable amount of time.
>  				 */
>  				u64 delta = rq_clock_task(rq) - curr->se.exec_start;
> -				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
> +				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 30);
>  			}
>  			curr->sched_class->task_tick(rq, curr, 0);
>  
> -- 
> 2.51.0
> 

--

Re: [PATCH] sched: Increase sched_tick_remote timeout

Posted by wangtao (EQ) 2 weeks, 2 days ago

Increasing timeout alerts can reduce the probability of deadlocks. However, in the 'sched_tick_remote' method, there are 'WARN_ON_ONCE(rq->curr!= rq->donor)' and 'assert_clock_updated' in 'rq_clock_task'. Regardless of why these alerts are triggered, once they are triggered, 'printk' is called, which still leaves potential deadlock issues. Is there a better way to address these problems?

在 2025/9/12 0:13, Phil Auld 写道:
> Increase the sched_tick_remote WARN_ON timeout to remove false
> positives due to temporarily busy HK cpus. The suggestion
> was 30 seconds to catch really stuck remote tick processing
> but not trigger it too easily.
>
> Signed-off-by: Phil Auld <pauld@redhat.com>
> Suggested-by: Frederic Weisbecker <frederic@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Frederic Weisbecker <frederic@kernel.org>
> ---
>   kernel/sched/core.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index be00629f0ba4..ef90d358252d 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -5724,7 +5724,7 @@ static void sched_tick_remote(struct work_struct *work)
>   				 * reasonable amount of time.
>   				 */
>   				u64 delta = rq_clock_task(rq) - curr->se.exec_start;
> -				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
> +				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 30);
>   			}
>   			curr->sched_class->task_tick(rq, curr, 0);
>

Re: [PATCH] sched: Increase sched_tick_remote timeout

Posted by Phil Auld 2 weeks, 2 days ago

On Tue, Sep 16, 2025 at 04:44:39PM +0800 wangtao (EQ) wrote:
> Increasing timeout alerts can reduce the probability of deadlocks. However, in the 'sched_tick_remote' method, there are 'WARN_ON_ONCE(rq->curr!= rq->donor)' and 'assert_clock_updated' in 'rq_clock_task'. Regardless of why these alerts are triggered, once they are triggered, 'printk' is called, which still leaves potential deadlock issues. Is there a better way to address these problems?
>

I'm not specically trying to solve the printk deadlock problem. My patch is
to make this particular warning go away by reducing the false positives.
That's tangential to your original posting. 

You can use the new printk mechanism with an atomic console to get around
the printk bug I think.

I think you could also use a serial console instead of a framebuffer based
console.


Cheers,
Phil



> 在 2025/9/12 0:13, Phil Auld 写道:
> > Increase the sched_tick_remote WARN_ON timeout to remove false
> > positives due to temporarily busy HK cpus. The suggestion
> > was 30 seconds to catch really stuck remote tick processing
> > but not trigger it too easily.
> > 
> > Signed-off-by: Phil Auld <pauld@redhat.com>
> > Suggested-by: Frederic Weisbecker <frederic@kernel.org>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Frederic Weisbecker <frederic@kernel.org>
> > ---
> >   kernel/sched/core.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index be00629f0ba4..ef90d358252d 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -5724,7 +5724,7 @@ static void sched_tick_remote(struct work_struct *work)
> >   				 * reasonable amount of time.
> >   				 */
> >   				u64 delta = rq_clock_task(rq) - curr->se.exec_start;
> > -				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
> > +				WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 30);
> >   			}
> >   			curr->sched_class->task_tick(rq, curr, 0);
> 

--

Re: [PATCH] sched: Increase sched_tick_remote timeout

Posted by Frederic Weisbecker 3 weeks ago

Le Thu, Sep 11, 2025 at 12:13:00PM -0400, Phil Auld a écrit :
> Increase the sched_tick_remote WARN_ON timeout to remove false
> positives due to temporarily busy HK cpus. The suggestion
> was 30 seconds to catch really stuck remote tick processing
> but not trigger it too easily.
> 
> Signed-off-by: Phil Auld <pauld@redhat.com>
> Suggested-by: Frederic Weisbecker <frederic@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Frederic Weisbecker <frederic@kernel.org>

Acked-by: Frederic Weisbecker <frederic@kernel.org>

-- 
Frederic Weisbecker
SUSE Labs

Re: [PATCH] sched: Increase sched_tick_remote timeout

Posted by wangtao (EQ) 2 weeks, 1 day ago

Do we have plans to merge this patch into the mainline?

Thanks,

Tao

在 2025/9/12 0:29, Frederic Weisbecker 写道:
> Le Thu, Sep 11, 2025 at 12:13:00PM -0400, Phil Auld a écrit :
>> Increase the sched_tick_remote WARN_ON timeout to remove false
>> positives due to temporarily busy HK cpus. The suggestion
>> was 30 seconds to catch really stuck remote tick processing
>> but not trigger it too easily.
>>
>> Signed-off-by: Phil Auld <pauld@redhat.com>
>> Suggested-by: Frederic Weisbecker <frederic@kernel.org>
>> Cc: Peter Zijlstra <peterz@infradead.org>
>> Cc: Frederic Weisbecker <frederic@kernel.org>
> Acked-by: Frederic Weisbecker <frederic@kernel.org>
>