Re: [PATCH v2] sched/fair: fix case with reduced capacity CPU

[Qais Yousef <qais.yousef@arm.com>]

On 07/11/22 18:42, Vincent Guittot wrote:
> On Mon, 11 Jul 2022 at 18:03, Qais Yousef <qais.yousef@arm.com> wrote:
> >
> > Hi Vincent
> >
> > On 07/08/22 17:44, Vincent Guittot wrote:
> > > The capacity of the CPU available for CFS tasks can be reduced because of
> > > other activities running on the latter. In such case, it's worth trying to
> > > move CFS tasks on a CPU with more available capacity.
> > >
> > > The rework of the load balance has filtered the case when the CPU is
> > > classified to be fully busy but its capacity is reduced.
> > >
> > > Check if CPU's capacity is reduced while gathering load balance statistic
> > > and classify it group_misfit_task instead of group_fully_busy so we can
> > > try to move the load on another CPU.
> > >
> > > Reported-by: David Chen <david.chen@nutanix.com>
> > > Reported-by: Zhang Qiao <zhangqiao22@huawei.com>
> > > Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
> > > Tested-by: David Chen <david.chen@nutanix.com>
> > > Tested-by: Zhang Qiao <zhangqiao22@huawei.com>
> > > ---
> >
> > [...]
> >
> > > @@ -8820,8 +8833,9 @@ static inline void update_sg_lb_stats(struct lb_env *env,
> > >
> > >       for_each_cpu_and(i, sched_group_span(group), env->cpus) {
> > >               struct rq *rq = cpu_rq(i);
> > > +             unsigned long load = cpu_load(rq);
> > >
> > > -             sgs->group_load += cpu_load(rq);
> > > +             sgs->group_load += load;
> > >               sgs->group_util += cpu_util_cfs(i);
> > >               sgs->group_runnable += cpu_runnable(rq);
> > >               sgs->sum_h_nr_running += rq->cfs.h_nr_running;
> > > @@ -8851,11 +8865,17 @@ static inline void update_sg_lb_stats(struct lb_env *env,
> > >               if (local_group)
> > >                       continue;
> > >
> > > -             /* Check for a misfit task on the cpu */
> > > -             if (env->sd->flags & SD_ASYM_CPUCAPACITY &&
> > > -                 sgs->group_misfit_task_load < rq->misfit_task_load) {
> > > -                     sgs->group_misfit_task_load = rq->misfit_task_load;
> > > -                     *sg_status |= SG_OVERLOAD;
> > > +             if (env->sd->flags & SD_ASYM_CPUCAPACITY) {
> > > +                     /* Check for a misfit task on the cpu */
> > > +                     if (sgs->group_misfit_task_load < rq->misfit_task_load) {
> > > +                             sgs->group_misfit_task_load = rq->misfit_task_load;
> > > +                             *sg_status |= SG_OVERLOAD;
> > > +                     }
> > > +             } else if ((env->idle != CPU_NOT_IDLE) &&
> > > +                        sched_reduced_capacity(rq, env->sd)) {
> > > +                     /* Check for a task running on a CPU with reduced capacity */
> > > +                     if (sgs->group_misfit_task_load < load)
> > > +                             sgs->group_misfit_task_load = load;
> > >               }
> > >       }
> >
> > Small questions mostly for my education purposes.
> >
> > The new condition only applies for SMP systems. The reason asym systems don't
> > care is because misfit check already considers capacity pressure when checking
> > that the task fits_capacity()?
> 
> Yes
> 
> >
> > It **seems** to me that the migration margin in fits_capacity() acts like the
> > sd->imbalance_pct when check_cpu_capacity() is called by
> > sched_reduced_capacity(), did I get it right?
> 
> Yes
> 
> >
> > If I got it right, if the migration margin ever tweaked, could we potentially
> > start seeing this kind of reported issue on asym systems then? I guess not. It
> > just seems to me for asym systems tweaking the migration margin is similar to
> > tweaking imbalance_pct for smp ones. But the subtlety is greater as
> > imbalance_pct is still used in asym systems.
> 
> You should not because the task will end up being misfit whatever the
> margin. The only change would be how fast you will detect and migrate

Yes. It's just this control/knob of how fast is different for HMP vs SMP.

check_misfit_status() does rely on rq->misfit_task_load AND
check_cpu_capacity(). So for that case the 2 knobs will impact how fast this
check will converge.

I want to say we should document this with a comment, but tbh I have no clue
how this can be explained clearly without being way too verbose.

All looks good to me anyway. Thanks for clarifying.


Cheers

--
Qais Yousef