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

[Vincent Guittot]

On Thu, 7 Jul 2022 at 18:43, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
>
> On 02/07/2022 06:52, 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 filterd 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 statistics
> > 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@xxxxxxxxxxx>
> > Reported-by: Zhang Qiao <zhangqiao22@xxxxxxxxxx>
> > Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> > ---
> >
> > David, Zhang,
> >
> > I haven't put your tested-by because I have reworked and cleaned the patch to
> > cover more cases.
> >
> > Could you run some tests with this version ?
> >
> > Thanks
> >
> >  kernel/sched/fair.c | 50 ++++++++++++++++++++++++++++++++++++---------
> >  1 file changed, 40 insertions(+), 10 deletions(-)
> >
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index a78d2e3b9d49..126b82ef4279 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -8798,6 +8798,19 @@ sched_asym(struct lb_env *env, struct sd_lb_stats *sds,  struct sg_lb_stats *sgs
> >       return sched_asym_prefer(env->dst_cpu, group->asym_prefer_cpu);
> >  }
> >
> > +static inline bool
> > +sched_reduced_capacity(struct rq *rq, struct sched_domain *sd)
> > +{
> > +     /*
> > +      * When there is more than 1 task, the group_overloaded case already
> > +      * takes care of cpu with reduced capacity
> > +      */
> > +     if (rq->cfs.h_nr_running != 1)
> > +             return false;
> > +
> > +     return check_cpu_capacity(rq, sd);
> > +}
> > +
> >  /**
> >   * update_sg_lb_stats - Update sched_group's statistics for load balancing.
> >   * @env: The load balancing environment.
> > @@ -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) &&
> > +                        (sgs->group_misfit_task_load < load)) {
> > +                     /* Check for a task running on a CPU with reduced capacity */
> > +                     sgs->group_misfit_task_load = load;
> >               }
> >       }
> >
> > @@ -8908,7 +8928,8 @@ static bool update_sd_pick_busiest(struct lb_env *env,
> >        * CPUs in the group should either be possible to resolve
> >        * internally or be covered by avg_load imbalance (eventually).
> >        */
> > -     if (sgs->group_type == group_misfit_task &&
> > +     if ((env->sd->flags & SD_ASYM_CPUCAPACITY) &&
> > +         (sgs->group_type == group_misfit_task) &&
> >           (!capacity_greater(capacity_of(env->dst_cpu), sg->sgc->max_capacity) ||
> >            sds->local_stat.group_type != group_has_spare))
> >               return false;
> > @@ -9517,9 +9538,18 @@ static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *s
> >       busiest = &sds->busiest_stat;
> >
> >       if (busiest->group_type == group_misfit_task) {
> > -             /* Set imbalance to allow misfit tasks to be balanced. */
> > -             env->migration_type = migrate_misfit;
> > -             env->imbalance = 1;
> > +             if (env->sd->flags & SD_ASYM_CPUCAPACITY) {
> > +                     /* Set imbalance to allow misfit tasks to be balanced. */
> > +                     env->migration_type = migrate_misfit;
> > +                     env->imbalance = 1;
> > +             } else {
> > +                     /*
> > +                      * Set load imbalance to allow moving task from cpu
> > +                      * with reduced capacity
> > +                      */
> > +                     env->migration_type = migrate_load;
> > +                     env->imbalance = busiest->group_misfit_task_load;
>
> I'm wondering why you've chosen that hybrid approach `group_misfit_task
> -> migrate_load` and not `group_misfit_task -> migrate_misfit`.

because, it means enabling the tracking of misfit task on rq at each
task enqueue/dequeue/tick ...  Then mistfit for heterogeneous platform
checks max_cpu_capacity what we don't care and will trigger unwanted
misfit migration for smp

>
> It looks like this `rq->cfs.h_nr_running = 1` case almost (since we
> check `busiest->nr_running > 1`) always ends up in the load_balance()
> `if (!ld_moved)` condition and need_active_balance() can return 1 in
> case `if ((env->idle != CPU_NOT_IDLE) && ...` condition. This leads to
> active load_balance and this
>
> IMHO, the same you can achieve when you would stay with
> `group_misfit_task -> migrate_misfit`.
>
> I think cpu_load(rq) can be used instead of `rq->misfit_task_load` in
> the migrate_misfit case of find_busiest_queue() too.

I don't think because you can have a higher cpu_load() but not being misfit

>
> [...]