Re: [PATCH 02/11] sched_ext: Refactor consume_remote_task()
From: David Vernet
Date: Sat Aug 31 2024 - 19:40:15 EST
On Fri, Aug 30, 2024 at 07:33:21PM -1000, Tejun Heo wrote:
> Hello,
>
> On Fri, Aug 30, 2024 at 11:05:16PM -0500, David Vernet wrote:
> ...
> > Not a functional change from the prior patch, but it occurred to me that
> > if we just deactivate like this then we'll also fire the ops.quiescent()
> > callback in dequeue_task_scx(). Should we add a check to skip the
> > dequeue callbacks if p->scx.holding_cpu >= 0?
>
> Right, migrations shouldn't trigger quiescent / runnable events. We should
> be able to suppress based on holding_cpu and sticky_cpu. Will look into
> that.
Ah right, holding_cpu on its own is of course not enough because its
whole point is to avoid racing with actual, non-migratory dequeues.
Thanks for taking a look at that -- in the meantime, this patch LG:
Acked-by: David Vernet <void@xxxxxxxxxxxxx>
>
> Thanks.
>
> --
> tejun
Attachment:
signature.asc
Description: PGP signature