On Mon, Aug 28, 2023 at 05:51:09PM -0400, Joel Fernandes wrote:
> I think the issue is the loop later in the function does
> not try to toggle cpus that came online too late.
>
> So it does not test offloading on all CPUs just because max got updated too
> late.
Right, and therefore for_each_possible_cpu() or for_each_present_cpu()
should be fine to iterate since it's ok to try to toggle an offline CPU.
>
> One fix could be to periodically check in the loop if a new cpu at maxcpu + 1
> ever got onlined. If it did, update the maxcpu.
Is it worth the complication though?
Thanks.
>
> Thanks.
>
>
> >
> > Thanks.
On Tue, Aug 29, 2023 at 11:24:24AM +0200, Frederic Weisbecker wrote:
> On Mon, Aug 28, 2023 at 05:51:09PM -0400, Joel Fernandes wrote:
> > I think the issue is the loop later in the function does
> > not try to toggle cpus that came online too late.
> >
> > So it does not test offloading on all CPUs just because max got updated too
> > late.
>
> Right, and therefore for_each_possible_cpu() or for_each_present_cpu()
> should be fine to iterate since it's ok to try to toggle an offline CPU.
OK, so I will accept the original patch which did just that.
Thank you!
I recently got burned by lack of workqueues on a not-yet-onlined CPU,
hence my questions here. ;-)
Thanx, Paul
> > One fix could be to periodically check in the loop if a new cpu at maxcpu + 1
> > ever got onlined. If it did, update the maxcpu.
>
> Is it worth the complication though?
>
> Thanks.
>
> >
> > Thanks.
> >
> >
> > >
> > > Thanks.
On Tue, Aug 29, 2023 at 05:38:54AM -0700, Paul E. McKenney wrote:
> On Tue, Aug 29, 2023 at 11:24:24AM +0200, Frederic Weisbecker wrote:
> > On Mon, Aug 28, 2023 at 05:51:09PM -0400, Joel Fernandes wrote:
> > > I think the issue is the loop later in the function does
> > > not try to toggle cpus that came online too late.
> > >
> > > So it does not test offloading on all CPUs just because max got updated too
> > > late.
> >
> > Right, and therefore for_each_possible_cpu() or for_each_present_cpu()
> > should be fine to iterate since it's ok to try to toggle an offline CPU.
>
> OK, so I will accept the original patch which did just that.
Which I finally did. I took the liberty of adding Frederic's Reviewed-by,
but please let me know if this is in any way inappropriate.
Thanx, Paul
> Thank you!
>
> I recently got burned by lack of workqueues on a not-yet-onlined CPU,
> hence my questions here. ;-)
>
> Thanx, Paul
>
> > > One fix could be to periodically check in the loop if a new cpu at maxcpu + 1
> > > ever got onlined. If it did, update the maxcpu.
> >
> > Is it worth the complication though?
> >
> > Thanks.
> >
> > >
> > > Thanks.
> > >
> > >
> > > >
> > > > Thanks.