Re: [PATCH] sched_ext: idle: small CPU iteration refactoring

[Andrea Righi <arighi@nvidia.com>]

Hi Yury,

On Sat, Jan 04, 2025 at 08:00:10PM -0800, Yury Norov wrote:
> On Sat, Jan 04, 2025 at 10:00:09AM +0100, Andrea Righi wrote:
> > loop to check if all SMT CPUs are idle with
> > cpumask_subset(). This simplifies the code and slightly improves
> > efficiency, while preserving the original behavior.
> 
> In my experience, replacing iterator with a dedicated function makes
> you 400-500 times faster. So 'slightly' sounds like an understatement
> here.

There's also the fact that the smt mask typically will have just 2 bits set
(maybe 4 in some high end servers), so I wasn't expecting big improvements.
But in any case, it's definitely a good change to do.

Thanks!
-Andrea

> 
> > Note that idle_masks.smt handling remains racy, which is acceptable as
> > it serves as an optimization and is self-correcting.
> > 
> > Suggested-by: Yury Norov <yury.norov@gmail.com>
> > Signed-off-by: Andrea Righi <arighi@nvidia.com>
> 
> Reviewed-by: Yury Norov <yury.norov@gmail.com>
> 
> > ---
> >  kernel/sched/ext.c | 6 ++----
> >  1 file changed, 2 insertions(+), 4 deletions(-)
> > 
> > diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
> > index 926579624c41..0ce116e0f67c 100644
> > --- a/kernel/sched/ext.c
> > +++ b/kernel/sched/ext.c
> > @@ -3671,10 +3671,8 @@ void __scx_update_idle(struct rq *rq, bool idle)
> >  			 * idle_masks.smt handling is racy but that's fine as
> >  			 * it's only for optimization and self-correcting.
> >  			 */
> > -			for_each_cpu(cpu, smt) {
> > -				if (!cpumask_test_cpu(cpu, idle_masks.cpu))
> > -					return;
> > -			}
> > +			if (!cpumask_subset(smt, idle_masks.cpu))
> > +				return;
> >  			cpumask_or(idle_masks.smt, idle_masks.smt, smt);
> >  		} else {
> >  			cpumask_andnot(idle_masks.smt, idle_masks.smt, smt);
> > -- 
> > 2.47.1