Re: [PATCH V3 2/2] sched/fair: Fallback to sched-idle CPU if idle CPU isn't found

[Subhra Mazumdar]

On 7/1/19 1:03 AM, Viresh Kumar wrote:
> On 28-06-19, 18:16, Subhra Mazumdar wrote:
> > On 6/25/19 10:06 PM, Viresh Kumar wrote:
> > > We try to find an idle CPU to run the next task, but in case we don't
> > > find an idle CPU it is better to pick a CPU which will run the task the
> > > soonest, for performance reason.
> > >
> > > A CPU which isn't idle but has only SCHED_IDLE activity queued on it
> > > should be a good target based on this criteria as any normal fair task
> > > will most likely preempt the currently running SCHED_IDLE task
> > > immediately. In fact, choosing a SCHED_IDLE CPU over a fully idle one
> > > shall give better results as it should be able to run the task sooner
> > > than an idle CPU (which requires to be woken up from an idle state).
> > >
> > > This patch updates both fast and slow paths with this optimization.
> > >
> > > Signed-off-by: Viresh Kumar <viresh.kumar@xxxxxxxxxx>
> > > ---
> > >    kernel/sched/fair.c | 43 +++++++++++++++++++++++++++++++++----------
> > >    1 file changed, 33 insertions(+), 10 deletions(-)
> > >
> > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > index 1277adc3e7ed..2e0527fd468c 100644
> > > --- a/kernel/sched/fair.c
> > > +++ b/kernel/sched/fair.c
> > > @@ -5376,6 +5376,15 @@ static struct {
> > >    #endif /* CONFIG_NO_HZ_COMMON */
> > > +/* CPU only has SCHED_IDLE tasks enqueued */
> > > +static int sched_idle_cpu(int cpu)
> > > +{
> > > +	struct rq *rq = cpu_rq(cpu);
> > > +
> > > +	return unlikely(rq->nr_running == rq->cfs.idle_h_nr_running &&
> > > +			rq->nr_running);
> > > +}
> > > +
> > Shouldn't this check if rq->curr is also sched idle?
> Why wouldn't the current set of checks be enough to guarantee that ?
I thought nr_running does not include the on-cpu thread.
> > And why not drop the rq->nr_running non zero check?
> Because CPU isn't sched-idle if nr_running and idle_h_nr_running are both 0,
> i.e. it is an IDLE cpu in that case. And so I thought it is important to have
> this check as well.
idle_cpu() not only checks nr_running is 0 but also rq->curr == rq->idle