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

[Ionela Voinescu <ionela.voinescu@arm.com>]

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@google.com>
> Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> Cc: Ionela Voinescu <ionela.voinescu@arm.com>
> Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
> Cc: Len Brown <len.brown@intel.com>
> Cc: Lukasz Luba <lukasz.luba@arm.com>
> Cc: Mel Gorman <mgorman@suse.de>
> Cc: Perry Yuan <Perry.Yuan@amd.com>
> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Tim C. Chen <tim.c.chen@intel.com>
> Cc: Valentin Schneider <vschneid@redhat.com>
> Cc: Zhao Liu <zhao1.liu@linux.intel.com>
> Cc: x86@kernel.org
> Cc: linux-pm@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
> Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
> ---
> 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.

Ionela.

> +					busiest = rq;
> +				}
>  			}
>  			break;
>  
> -- 
> 2.25.1
>