Re: [PATCH v2] sched/fair: sanitize vruntime of entity being migrated

[Vincent Guittot]

On Mon, 13 Mar 2023 at 19:17, Dietmar Eggemann <dietmar.eggemann@xxxxxxx> wrote:
>
> On 13/03/2023 10:06, Dietmar Eggemann wrote:
> > On 10/03/2023 15:29, Vincent Guittot wrote:
> >> Le jeudi 09 mars 2023 � 16:14:38 (+0100), Vincent Guittot a �crit :
> >>> On Thu, 9 Mar 2023 at 15:37, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> >>>>
> >>>> On Thu, Mar 09, 2023 at 03:28:25PM +0100, Peter Zijlstra wrote:
> >>>>> On Thu, Mar 09, 2023 at 02:34:05PM +0100, Vincent Guittot wrote:
>
> [...]
>
> > Looks to me that this patch brings back the old numbers:
> >
> > model name    : Intel(R) Xeon(R) Silver 4314 CPU @ 2.40GHz
> >
> > perf stat --null --repeat 10 -- perf bench sched messaging -g 50 -l 5000
> >
> > tip sched/core
> >
> > a2e90611b9f4 - sched/fair: Remove capacity inversion detection
> > (2023-02-11 Vincent Guittot)
> >
> >   5.7295 +- 0.0219 seconds time elapsed  ( +-  0.38% )
> >
> > 829c1651e9c4 - sched/fair: sanitize vruntime of entity being placed
> > (2023-02-11 Zhang Qiao)
> >
> >   6.0961 +- 0.0297 seconds time elapsed  ( +-  0.49% )
> >
> > this patch on top 829c1651e9c4
> >
> >   5.7165 +- 0.0231 seconds time elapsed  ( +-  0.40% )
> >
> > [...]
>
> Couldn't we not just defer setting `se->exec_start = 0` until the end of
> place_entity() for ENQUEUE_MIGRATED instead to avoid this extra se flag
> `migrated`?

Yes, that's a good point.

I'm going to use something a bit different from your proposal below by
merging initial and flag
static void place_entity(struct cfs_rq *cfs_rq, struct sched_entity
*se, int flags)

with flags:
0 for initial placement
ENQUEUE_WAKEUP for wakeup
ENQUEUE_MIGRATED for migrated task

>
> -->8--
>
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 0c70c558b12c..4df2b3e76b30 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -550,7 +550,6 @@ struct sched_entity {
>         struct rb_node                  run_node;
>         struct list_head                group_node;
>         unsigned int                    on_rq;
> -       unsigned int                    migrated;
>
>         u64                             exec_start;
>         u64                             sum_exec_runtime;
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index a8aa8cd3c745..365ee548e9f0 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -1057,7 +1057,6 @@ update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
>         /*
>          * We are starting a new run period:
>          */
> -       se->migrated = 0;
>         se->exec_start = rq_clock_task(rq_of(cfs_rq));
>  }
>
> @@ -4649,8 +4648,8 @@ static void check_spread(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  #endif
>  }
>
> -static void
> -place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
> +static void place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se,
> +                        int flags, int initial)
>  {
>         u64 vruntime = cfs_rq->min_vruntime;
>         u64 sleep_time;
> @@ -4705,6 +4704,9 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
>                 se->vruntime = vruntime;
>         else
>                 se->vruntime = max_vruntime(se->vruntime, vruntime);
> +
> +       if (flags & ENQUEUE_MIGRATED)
> +               se->exec_start = 0;
>  }
>
>  static void check_enqueue_throttle(struct cfs_rq *cfs_rq);
> @@ -4780,7 +4782,7 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
>         account_entity_enqueue(cfs_rq, se);
>
>         if (flags & ENQUEUE_WAKEUP)
> -               place_entity(cfs_rq, se, 0);
> +               place_entity(cfs_rq, se, flags, 0);
>
>         check_schedstat_required();
>         update_stats_enqueue_fair(cfs_rq, se, flags);
> @@ -7668,9 +7670,6 @@ static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
>         /* Tell new CPU we are migrated */
>         se->avg.last_update_time = 0;
>
> -       /* We have migrated, no longer consider this task hot */
> -       se->migrated = 1;
> -
>         update_scan_period(p, new_cpu);
>  }
>
> @@ -8355,9 +8354,6 @@ static int task_hot(struct task_struct *p, struct lb_env *env)
>         if (sysctl_sched_migration_cost == 0)
>                 return 0;
>
> -       if (p->se.migrated)
> -               return 0;
> -
>         delta = rq_clock_task(env->src_rq) - p->se.exec_start;
>
>         return delta < (s64)sysctl_sched_migration_cost;
> @@ -11999,7 +11995,7 @@ static void task_fork_fair(struct task_struct *p)
>                 update_curr(cfs_rq);
>                 se->vruntime = curr->vruntime;
>         }
> -       place_entity(cfs_rq, se, 1);
> +       place_entity(cfs_rq, se, 0, 1);
>
>         if (sysctl_sched_child_runs_first && curr && entity_before(curr, se)) {
>                 /*
> @@ -12144,7 +12140,7 @@ static void detach_task_cfs_rq(struct task_struct *p)
>                  * Fix up our vruntime so that the current sleep doesn't
>                  * cause 'unlimited' sleep bonus.
>                  */
> -               place_entity(cfs_rq, se, 0);
> +               place_entity(cfs_rq, se, 0, 0);
>                 se->vruntime -= cfs_rq->min_vruntime;
>         }
>
>