Re: [PATCH v3 0/2] sched: Minor changes for rd->overload access

[Ingo Molnar]

* Shrikanth Hegde <sshegde@xxxxxxxxxxxxx> wrote:

> 
> 
> On 3/28/24 4:37 PM, Ingo Molnar wrote:
> > 
> > * Ingo Molnar <mingo@xxxxxxxxxx> wrote:
> > 
> >> Plus I've applied a patch to rename ::overload to ::overloaded. It is 
> >> silly to use an ambiguous noun instead of a clear adjective when naming 
> >> such a flag ...
> > 
> > Plus SG_OVERLOAD should be SG_OVERLOADED as well - it now looks in line 
> > with SG_OVERUTILIZED:
> > 
> >  /* Scheduling group status flags */
> >  #define SG_OVERLOADED           0x1 /* More than one runnable task on a CPU. */
> >  #define SG_OVERUTILIZED         0x2 /* One or more CPUs are over-utilized. */
> > 
> > My followup question is: why are these a bitmask, why not separate 
> > flags?
> > 
> > AFAICS we only ever set them separately:
> > 
> >  thule:~/tip> git grep SG_OVER kernel/sched/
> >  kernel/sched/fair.c:            set_rd_overutilized_status(rq->rd, SG_OVERUTILIZED);
> >  kernel/sched/fair.c:                    *sg_status |= SG_OVERLOADED;
> >  kernel/sched/fair.c:                    *sg_status |= SG_OVERUTILIZED;
> >  kernel/sched/fair.c:                            *sg_status |= SG_OVERLOADED;
> >  kernel/sched/fair.c:            set_rd_overloaded(env->dst_rq->rd, sg_status & SG_OVERLOADED);
> >  kernel/sched/fair.c:                                       sg_status & SG_OVERUTILIZED);
> >  kernel/sched/fair.c:    } else if (sg_status & SG_OVERUTILIZED) {
> >  kernel/sched/fair.c:            set_rd_overutilized_status(env->dst_rq->rd, SG_OVERUTILIZED);
> >  kernel/sched/sched.h:#define SG_OVERLOADED              0x1 /* More than one runnable task on a CPU. */
> >  kernel/sched/sched.h:#define SG_OVERUTILIZED            0x2 /* One or more CPUs are over-utilized. */
> >  kernel/sched/sched.h:           set_rd_overloaded(rq->rd, SG_OVERLOADED);
> > 
> > In fact this results in suboptimal code:
> > 
> >                 /* update overload indicator if we are at root domain */
> >                 set_rd_overloaded(env->dst_rq->rd, sg_status & SG_OVERLOADED);
> >                         
> >                 /* Update over-utilization (tipping point, U >= 0) indicator */
> >                 set_rd_overutilized_status(env->dst_rq->rd,
> >                                            sg_status & SG_OVERUTILIZED);
> > 
> > Note how the bits that got mixed together in sg_status now have to be 
> > masked out individually.
> > 
> > The sg_status bitmask appears to make no sense at all to me.
> > 
> > By turning these into individual bool flags we could also do away with 
> > all the extra SG_OVERLOADED/SG_OVERUTILIZED abstraction.
> > 
> > Ie. something like the patch below? Untested.
> 
> Looks good. I see it is merged to sched/core. 
> Did a boot with that patch and hackbench is showing same results 320 CPU system.

Thanks, I've added:

    Acked-by: Shrikanth Hegde <sshegde@xxxxxxxxxxxxx>
    Tested-by: Shrikanth Hegde <sshegde@xxxxxxxxxxxxx>

And applied the additional docbook fix below on top as well.

Thaks,

	Ingo

=================>
 kernel/sched/fair.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ebc8d5f855de..1dd37168da50 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9933,7 +9933,8 @@ sched_reduced_capacity(struct rq *rq, struct sched_domain *sd)
  * @sds: Load-balancing data with statistics of the local group.
  * @group: sched_group whose statistics are to be updated.
  * @sgs: variable to hold the statistics for this group.
- * @sg_status: Holds flag indicating the status of the sched_group
+ * @sg_overloaded: sched_group is overloaded
+ * @sg_overutilized: sched_group is overutilized
  */
 static inline void update_sg_lb_stats(struct lb_env *env,
 				      struct sd_lb_stats *sds,