Re: [PATCH v3 06/10] sched/fair: Use the prefer_sibling flag of the current sched domain

[Ricardo Neri]

On Thu, Feb 09, 2023 at 03:05:03PM -0800, Tim Chen wrote:
> On Thu, 2023-02-09 at 20:00 +0000, Chen, Tim C wrote:
> > > >  static inline void update_sd_lb_stats(struct lb_env *env, struct
> > > > sd_lb_stats *sds)  {
> > > > -       struct sched_domain *child = env->sd->child;
> > > >         struct sched_group *sg = env->sd->groups;
> > > >         struct sg_lb_stats *local = &sds->local_stat;
> > > >         struct sg_lb_stats tmp_sgs;
> > > > @@ -10045,9 +10044,11 @@ static inline void
> > > > update_sd_lb_stats(struct
> > > lb_env *env, struct sd_lb_stats *sd
> > > >                 sg = sg->next;
> > > >         } while (sg != env->sd->groups);
> > > > 
> > > > -       /* Tag domain that child domain prefers tasks go to
> > > > siblings first */
> > > > -       sds->prefer_sibling = child && child->flags &
> > > > SD_PREFER_SIBLING;
> > > > -
> > > > +       /*
> > > > +        * Tag domain that @env::sd prefers to spread excess
> > > > tasks among
> > > > +        * sibling sched groups.
> > > > +        */
> > > > +       sds->prefer_sibling = env->sd->flags & SD_PREFER_SIBLING;
> > > > 
> > > This does help fix the issue that non-SMT core fails to pull task
> > > from busy SMT-
> > > cores.
> > > And it also semantically changes the definination of prefer
> > > sibling. Do we also
> > > need to change this:
> > >        if ((sd->flags & SD_ASYM_CPUCAPACITY) && sd->child)
> > >                sd->child->flags &= ~SD_PREFER_SIBLING; might be:
> > >        if ((sd->flags & SD_ASYM_CPUCAPACITY))
> > >                sd->flags &= ~SD_PREFER_SIBLING;
> > > 
> > 
> > Yu,
> > 
> > I think you are talking about the code in sd_init() 
> > where SD_PREFER_SIBLING is first set
> > to "ON" and updated depending on SD_ASYM_CPUCAPACITY.  The intention
> > of the code
> > is if there are cpus in the scheduler domain that have differing cpu
> > capacities,
> > we do not want to do spreading among the child groups in the sched
> > domain.
> > So the flag is turned off in the child group level and not the parent
> > level. But with your above
> > change, the parent's flag is turned off, leaving the child level flag
> > on. 
> > This moves the level where spreading happens (SD_PREFER_SIBLING on) 
> > up one level which is undesired (see table below).

But my patch moves the level at which we act on prefer_sibling: it now
checks the SD_PREFER_SIBLING flag at the current level, not its child.
Thus, removing SD_PREFER_SIBLING from a level with SD_ASYM_CPUCAPACITY
prevents spreading among CPUs of different CPU capacity, no?

Thanks and BR,
Ricardo