Re: [PATCH v3 04/10] sched/fair: rework load_balance

[Vincent Guittot]

On Tue, 1 Oct 2019 at 19:47, Valentin Schneider
<valentin.schneider@xxxxxxx> wrote:
>
> On 19/09/2019 08:33, Vincent Guittot wrote:
>
> [...]
>
> > @@ -8283,69 +8363,133 @@ static inline void update_sd_lb_stats(struct lb_env *env, struct sd_lb_stats *sd
> >   */
> >  static inline void calculate_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
> >  {
> > -     unsigned long max_pull, load_above_capacity = ~0UL;
> >       struct sg_lb_stats *local, *busiest;
> >
> >       local = &sds->local_stat;
> >       busiest = &sds->busiest_stat;
> >
> > -     if (busiest->group_asym_packing) {
> > +     if (busiest->group_type == group_misfit_task) {
> > +             /* Set imbalance to allow misfit task to be balanced. */
> > +             env->balance_type = migrate_misfit;
> > +             env->imbalance = busiest->group_misfit_task_load;
> > +             return;
> > +     }
> > +
> > +     if (busiest->group_type == group_asym_packing) {
> > +             /*
> > +              * In case of asym capacity, we will try to migrate all load to
> > +              * the preferred CPU.
> > +              */
> > +             env->balance_type = migrate_load;
> >               env->imbalance = busiest->group_load;
> >               return;
> >       }
> >
> > +     if (busiest->group_type == group_imbalanced) {
> > +             /*
> > +              * In the group_imb case we cannot rely on group-wide averages
> > +              * to ensure CPU-load equilibrium, try to move any task to fix
> > +              * the imbalance. The next load balance will take care of
> > +              * balancing back the system.
> > +              */
> > +             env->balance_type = migrate_task;
> > +             env->imbalance = 1;
> > +             return;
> > +     }
> > +
> >       /*
> > -      * Avg load of busiest sg can be less and avg load of local sg can
> > -      * be greater than avg load across all sgs of sd because avg load
> > -      * factors in sg capacity and sgs with smaller group_type are
> > -      * skipped when updating the busiest sg:
> > +      * Try to use spare capacity of local group without overloading it or
> > +      * emptying busiest
> >        */
> > -     if (busiest->group_type != group_misfit_task &&
> > -         (busiest->avg_load <= sds->avg_load ||
> > -          local->avg_load >= sds->avg_load)) {
> > -             env->imbalance = 0;
> > +     if (local->group_type == group_has_spare) {
> > +             if (busiest->group_type > group_fully_busy) {
> > +                     /*
> > +                      * If busiest is overloaded, try to fill spare
> > +                      * capacity. This might end up creating spare capacity
> > +                      * in busiest or busiest still being overloaded but
> > +                      * there is no simple way to directly compute the
> > +                      * amount of load to migrate in order to balance the
> > +                      * system.
> > +                      */
> > +                     env->balance_type = migrate_util;
> > +                     env->imbalance = max(local->group_capacity, local->group_util) -
> > +                                 local->group_util;
> > +                     return;
> > +             }
> > +
> > +             if (busiest->group_weight == 1 || sds->prefer_sibling) {
> > +                     /*
> > +                      * When prefer sibling, evenly spread running tasks on
> > +                      * groups.
> > +                      */
> > +                     env->balance_type = migrate_task;
> > +                     env->imbalance = (busiest->sum_h_nr_running - local->sum_h_nr_running) >> 1;
>
> Isn't that one somewhat risky?
>
> Say both groups are classified group_has_spare and we do prefer_sibling.
> We'd select busiest as the one with the maximum number of busy CPUs, but it
> could be so that busiest.sum_h_nr_running < local.sum_h_nr_running (because
> pinned tasks or wakeup failed to properly spread stuff).
>
> The thing should be unsigned so at least we save ourselves from right
> shifting a negative value, but we still end up with a gygornous imbalance
> (which we then store into env.imbalance which *is* signed... Urgh).

so it's not clear what happen with a right shift on negative signed
value and this seems to be compiler dependent so even
max_t(long, 0, (local->idle_cpus - busiest->idle_cpus) >> 1) might be wrong

I'm going to update it


>
> [...]