Re: [PATCH v7 12/14] sched/fair: Select an energy-efficient CPU on task wake-up

[Peter Zijlstra]

On Wed, Sep 12, 2018 at 10:13:07AM +0100, Quentin Perret wrote:
> +	while (pd) {
> +		unsigned long cur_energy, spare_cap, max_spare_cap = 0;
> +		int max_spare_cap_cpu = -1;
> +
> +		for_each_cpu_and(cpu, perf_domain_span(pd), sched_domain_span(sd)) {

Which of the two masks do we expect to be the smallest?

> +			if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
> +				continue;
> +
> +			/* Skip CPUs that will be overutilized. */
> +			util = cpu_util_next(cpu, p, cpu);
> +			cpu_cap = capacity_of(cpu);
> +			if (cpu_cap * 1024 < util * capacity_margin)
> +				continue;
> +
> +			/* Always use prev_cpu as a candidate. */
> +			if (cpu == prev_cpu) {
> +				prev_energy = compute_energy(p, prev_cpu, head);
> +				if (prev_energy < best_energy)
> +					best_energy = prev_energy;

				best_energy = min(best_energy, prev_energy);

That's both shorter and clearer.

> +				continue;
> +			}
> +
> +			/*
> +			 * Find the CPU with the maximum spare capacity in
> +			 * the performance domain
> +			 */
> +			spare_cap = cpu_cap - util;
> +			if (spare_cap > max_spare_cap) {
> +				max_spare_cap = spare_cap;
> +				max_spare_cap_cpu = cpu;
> +			}

Sometimes I wonder if something like:

#define min_filter(varp, val)		\
({					\
	typeof(varp) _varp = (varp);	\
	typeof(val)  _val  = (val);	\
	bool f = false;			\
					\
	if (_val < *_varp) {		\
		*_varp = _val;		\
		f = true;		\
	}				\
					\
	f;				\
})

and the corresponding max_filter() are worth the trouble; it would allow
writing:

	if (max_filter(&max_spare_cap, spare_cap))
		max_spare_cap_cpu = cpu;

and:

> +		}
> +
> +		/* Evaluate the energy impact of using this CPU. */
> +		if (max_spare_cap_cpu >= 0) {
> +			cur_energy = compute_energy(p, max_spare_cap_cpu, head);
> +			if (cur_energy < best_energy) {
> +				best_energy = cur_energy;
> +				best_energy_cpu = max_spare_cap_cpu;
> +			}

	if (min_filter(&best_energy, cur_energy))
		best_energy_cpu = max_spare_cap_cpu;

But then I figure, it is not... dunno. We do lots of this stuff.

> +		}
> +		pd = pd->next;
> +	}
> +
> +	/*
> +	 * Pick the best CPU if prev_cpu cannot be used, or if it saves at
> +	 * least 6% of the energy used by prev_cpu.
> +	 */
> +	if (prev_energy == ULONG_MAX ||
> +			(prev_energy - best_energy) > (prev_energy >> 4))
> +		return best_energy_cpu;

Does that become more readable if we split that into two conditions?

	if (prev_energy == ULONG_MAX)
		return best_energy_cpu;

	if ((prev_energy - best_energy) > (prev_energy >> 4))
		return best_energy_cpu;

> +	return prev_cpu;
> +}
> +
>  /*
>   * select_task_rq_fair: Select target runqueue for the waking task in domains
>   * that have the 'sd_flag' flag set. In practice, this is SD_BALANCE_WAKE,
> @@ -6360,13 +6468,37 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_f
>  	int want_affine = 0;
>  	int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
>  
> +	rcu_read_lock();
>  	if (sd_flag & SD_BALANCE_WAKE) {
>  		record_wakee(p);
> +
> +		/*
> +		 * Forkees are not accepted in the energy-aware wake-up path
> +		 * because they don't have any useful utilization data yet and
> +		 * it's not possible to forecast their impact on energy
> +		 * consumption. Consequently, they will be placed by
> +		 * find_idlest_cpu() on the least loaded CPU, which might turn
> +		 * out to be energy-inefficient in some use-cases. The
> +		 * alternative would be to bias new tasks towards specific types
> +		 * of CPUs first, or to try to infer their util_avg from the
> +		 * parent task, but those heuristics could hurt other use-cases
> +		 * too. So, until someone finds a better way to solve this,
> +		 * let's keep things simple by re-using the existing slow path.
> +		 */
> +		if (sched_feat(ENERGY_AWARE)) {
> +			struct root_domain *rd = cpu_rq(cpu)->rd;
> +			struct perf_domain *pd = rcu_dereference(rd->pd);
> +
> +			if (pd && !READ_ONCE(rd->overutilized)) {
> +				new_cpu = find_energy_efficient_cpu(p, prev_cpu, pd);
> +				goto unlock;
> +			}
> +		}
> +
> +		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu) &&
> +			      cpumask_test_cpu(cpu, &p->cpus_allowed);
>  	}

I would much prefer this to be something like:

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index a8f601edd958..5475a885ec9f 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -6299,12 +6299,19 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_f
 {
 	struct sched_domain *tmp, *sd = NULL;
 	int cpu = smp_processor_id();
-	int new_cpu = prev_cpu;
+	unsigned int new_cpu = prev_cpu;
 	int want_affine = 0;
 	int sync = (wake_flags & WF_SYNC) && !(current->flags & PF_EXITING);
 
 	if (sd_flag & SD_BALANCE_WAKE) {
 		record_wakee(p);
+
+		if (static_branch_unlikely(sched_eas_balance)) {
+			new_cpu = select_task_rq_eas(p, prev_cpu, sd_flags, wake_flags);
+			if (new_cpu < nr_cpu_ids)
+				return new_cpu;
+		}
+
 		want_affine = !wake_wide(p) && !wake_cap(p, cpu, prev_cpu)
 			      && cpumask_test_cpu(cpu, &p->cpus_allowed);
 	}


and then hide everything (including that giant comment) in
select_task_rq_eas().