[PATCH] sched: Fix stop_one_cpu_nowait() vs hotplug

[Peter Zijlstra]

On Wed, Oct 11, 2023 at 03:24:19AM +0000, Kuyo Chang (張建文) wrote:
> On Tue, 2023-10-10 at 22:04 +0200, Peter Zijlstra wrote:
> >  	 
> > External email : Please do not click links or open attachments until
> > you have verified the sender or the content.
> >  On Tue, Oct 10, 2023 at 04:57:47PM +0200, Peter Zijlstra wrote:
> > > On Tue, Oct 10, 2023 at 02:40:22PM +0000, Kuyo Chang (張建文) wrote:
> > 
> > > > It is running good so far(more than a week)on hotplug/set
> > affinity
> > > > stress test. I will keep it testing and report back if it happens
> > > > again.
> > > 
> > > OK, I suppose I should look at writing a coherent Changelog for
> > this
> > > then...
> > 
> > Something like the below... ?
> > 
> Thanks for illustrate the race scenario. It looks good to me.
> But how about RT? Does RT also need this invocations as below?
> 
> ---
>  kernel/sched/rt.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
> index e93b69ef919b..6aaf0a3d6081 100644
> --- a/kernel/sched/rt.c
> +++ b/kernel/sched/rt.c
> @@ -2063,9 +2063,11 @@ static int push_rt_task(struct rq *rq, bool
> pull)
>                  */
>                 push_task = get_push_task(rq);
>                 if (push_task) {
> +                       preempt_disable();
>                         raw_spin_rq_unlock(rq);
>                         stop_one_cpu_nowait(rq->cpu, push_cpu_stop,
>                                             push_task, &rq->push_work);
> +                       preempt_enable();
>                         raw_spin_rq_lock(rq);
>                 }
> 
> @@ -2402,9 +2404,11 @@ static void pull_rt_task(struct rq *this_rq)
>                 double_unlock_balance(this_rq, src_rq);
> 
>                 if (push_task) {
> +                       preempt_disable();
>                         raw_spin_rq_unlock(this_rq);
>                         stop_one_cpu_nowait(src_rq->cpu, push_cpu_stop,
>                                             push_task, &src_rq-
> >push_work);
> +                       preempt_enable();
>                         raw_spin_rq_lock(this_rq);
>                 }
>         }

bah, clearly git-grep didn't work for me last night, I'll go fix up.