Re: [PATCH v4 10/24] sched/fair: Use IPCC scores to select a busiest runqueue

[Ricardo Neri]

On Thu, Jun 22, 2023 at 10:03:17AM +0100, Ionela Voinescu wrote:
> On Monday 12 Jun 2023 at 21:24:08 (-0700), Ricardo Neri wrote:
> > Use IPCC scores to break a tie between two runqueues with the same priority
> > and number of running tasks: select the runqueue of which the task enqueued
> > last would get a higher IPC boost when migrated to the destination CPU.
> > (These tasks are migrated first during load balance.)
> > 
> > For now, restrict the utilization of IPCC scores to scheduling domains
> > marked with the SD_ASYM_PACKING flag.
> > 
> > Cc: Ben Segall <bsegall@xxxxxxxxxx>
> > Cc: Daniel Bristot de Oliveira <bristot@xxxxxxxxxx>
> > Cc: Dietmar Eggemann <dietmar.eggemann@xxxxxxx>
> > Cc: Ionela Voinescu <ionela.voinescu@xxxxxxx>
> > Cc: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > Cc: Len Brown <len.brown@xxxxxxxxx>
> > Cc: Lukasz Luba <lukasz.luba@xxxxxxx>
> > Cc: Mel Gorman <mgorman@xxxxxxx>
> > Cc: Perry Yuan <Perry.Yuan@xxxxxxx>
> > Cc: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > Cc: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
> > Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
> > Cc: Tim C. Chen <tim.c.chen@xxxxxxxxx>
> > Cc: Valentin Schneider <vschneid@xxxxxxxxxx>
> > Cc: Zhao Liu <zhao1.liu@xxxxxxxxxxxxxxx>
> > Cc: x86@xxxxxxxxxx
> > Cc: linux-pm@xxxxxxxxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
> > ---
> > Changes since v3:
> >  * Do not compute the IPCC stats using the current tasks of runqueues.
> >    Instead, use the tasks at the back of the queue. These are the tasks
> >    that will be pulled first during load balance. (Vincent)
> > 
> > Changes since v2:
> >  * Only use IPCC scores to break ties if the sched domain uses
> >    asym_packing. (Ionela)
> >  * Handle errors of arch_get_ipcc_score(). (Ionela)
> > 
> > Changes since v1:
> >  * Fixed a bug when selecting a busiest runqueue: when comparing two
> >    runqueues with equal nr_running, we must compute the IPCC score delta
> >    of both.
> >  * Renamed local variables to improve the layout of the code block.
> >    (PeterZ)
> >  * Used the new interface names.
> > ---
> >  kernel/sched/fair.c | 61 +++++++++++++++++++++++++++++++++++++++++++++
> >  1 file changed, 61 insertions(+)
> > 
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index fcec791ede4f..da3e009eef42 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -9564,6 +9564,41 @@ static bool sched_asym_ipcc_pick(struct sched_group *a,
> >  	return sched_asym_ipcc_prefer(a_stats, b_stats);
> >  }
> >  
> > +/**
> > + * ipcc_score_delta - Get the IPCC score delta wrt the load balance's dst_cpu
> > + * @rq:		A runqueue
> > + * @env:	Load balancing environment
> > + *
> > + * Returns: The IPCC score delta that the last task enqueued in @rq would get
> > + * if placed in the destination CPU of @env. LONG_MIN to indicate that the
> > + * delta should not be used.
> > + */
> > +static long ipcc_score_delta(struct rq *rq, struct lb_env *env)
> > +{
> > +	unsigned long score_src, score_dst;
> > +	unsigned short ipcc;
> > +
> > +	if (!sched_ipcc_enabled())
> > +		return LONG_MIN;
> > +
> > +	/* Only asym_packing uses IPCC scores at the moment. */
> > +	if (!(env->sd->flags & SD_ASYM_PACKING))
> > +		return LONG_MIN;
> > +
> > +	if (rq_last_task_ipcc(env->dst_cpu, rq, &ipcc))
> > +		return LONG_MIN;
> > +
> > +	score_dst = arch_get_ipcc_score(ipcc, env->dst_cpu);
> > +	if (IS_ERR_VALUE(score_dst))
> > +		return LONG_MIN;
> > +
> > +	score_src = arch_get_ipcc_score(ipcc, cpu_of(rq));
> > +	if (IS_ERR_VALUE(score_src))
> > +		return LONG_MIN;
> > +
> > +	return score_dst - score_src;
> > +}
> > +
> >  #else /* CONFIG_IPC_CLASSES */
> >  static void update_sg_lb_ipcc_stats(int dst_cpu, struct sg_lb_stats *sgs,
> >  				    struct rq *rq)
> > @@ -9594,6 +9629,11 @@ static bool sched_asym_ipcc_pick(struct sched_group *a,
> >  	return false;
> >  }
> >  
> > +static long ipcc_score_delta(struct rq *rq, struct lb_env *env)
> > +{
> > +	return LONG_MIN;
> > +}
> > +
> >  #endif /* CONFIG_IPC_CLASSES */
> >  
> >  /**
> > @@ -10769,6 +10809,7 @@ static struct rq *find_busiest_queue(struct lb_env *env,
> >  {
> >  	struct rq *busiest = NULL, *rq;
> >  	unsigned long busiest_util = 0, busiest_load = 0, busiest_capacity = 1;
> > +	long busiest_ipcc_delta = LONG_MIN;
> >  	unsigned int busiest_nr = 0;
> >  	int i;
> >  
> > @@ -10885,6 +10926,26 @@ static struct rq *find_busiest_queue(struct lb_env *env,
> >  			if (busiest_nr < nr_running) {
> >  				busiest_nr = nr_running;
> >  				busiest = rq;
> > +
> > +				/*
> > +				 * Remember the IPCC score of the busiest
> > +				 * runqueue. We may need it to break a tie with
> > +				 * other queues with equal nr_running.
> > +				 */
> > +				busiest_ipcc_delta = ipcc_score_delta(busiest, env);
> > +			/*
> > +			 * For ties, select @rq if doing would give its last
> > +			 * queued task a bigger IPC boost when migrated to
> > +			 * dst_cpu.
> > +			 */
> > +			} else if (busiest_nr == nr_running) {
> > +				long delta = ipcc_score_delta(rq, env);
> > +
> > +				if (busiest_ipcc_delta < delta) {
> > +					busiest_ipcc_delta = delta;
> > +					busiest_nr = nr_running;
> 
> nit: there's no need as busiest_nr is already equal to nr_running.

True! I will remove this pointless assignment.

Thanks and BR,
Ricardo