Re: [PATCH v4 1/3] sched/uclamp: Set max_spare_cap_cpu even if max_spare_cap is 0

[Qais Yousef]

On 08/23/23 12:30, Dietmar Eggemann wrote:
> On 22/08/2023 00:45, Qais Yousef wrote:
> > When uclamp_max is being used, the util of the task could be higher than
> > the spare capacity of the CPU, but due to uclamp_max value we force fit
> > it there.
> > 
> > The way the condition for checking for max_spare_cap in
> > find_energy_efficient_cpu() was constructed; it ignored any CPU that has
> > its spare_cap less than or _equal_ to max_spare_cap. Since we initialize
> > max_spare_cap to 0; this lead to never setting max_spare_cap_cpu and
> > hence ending up never performing compute_energy() for this cluster and
> > missing an opportunity for a better energy efficient placement to honour
> > uclamp_max setting.
> > 
> > 	max_spare_cap = 0;
> > 	cpu_cap = capacity_of(cpu) - task_util(p);  // 0 if task_util(p) is high
> 
> Nitpick:
> 
> s/task_util(p)/cpu_util(cpu, p, cpu, ...) which is
> 
> max(cpu_util + task_util, cpu_util_est + task_util_est)
> 
> > 
> > 	...
> > 
> > 	util_fits_cpu(...);		// will return true if uclamp_max forces it to fit
> > 
> > 	...
> > 
> > 	// this logic will fail to update max_spare_cap_cpu if cpu_cap is 0
> > 	if (cpu_cap > max_spare_cap) {
> > 		max_spare_cap = cpu_cap;
> > 		max_spare_cap_cpu = cpu;
> > 	}
> > 
> > prev_spare_cap suffers from a similar problem.
> > 
> > Fix the logic by converting the variables into long and treating -1
> > value as 'not populated' instead of 0 which is a viable and correct
> > spare capacity value. We need to be careful signed comparison is used
> > when comparing with cpu_cap in one of the conditions.
> > 
> > Fixes: 1d42509e475c ("sched/fair: Make EAS wakeup placement consider uclamp restrictions")
> > Reviewed-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> > Signed-off-by: Qais Yousef (Google) <qyousef@xxxxxxxxxxx>
> > ---
> >  kernel/sched/fair.c | 11 +++++------
> >  1 file changed, 5 insertions(+), 6 deletions(-)
> > 
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index 0b7445cd5af9..5da6538ed220 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -7707,11 +7707,10 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> >  	for (; pd; pd = pd->next) {
> >  		unsigned long util_min = p_util_min, util_max = p_util_max;
> >  		unsigned long cpu_cap, cpu_thermal_cap, util;
> > -		unsigned long cur_delta, max_spare_cap = 0;
> > +		long prev_spare_cap = -1, max_spare_cap = -1;
> >  		unsigned long rq_util_min, rq_util_max;
> > -		unsigned long prev_spare_cap = 0;
> > +		unsigned long cur_delta, base_energy;
> >  		int max_spare_cap_cpu = -1;
> > -		unsigned long base_energy;
> >  		int fits, max_fits = -1;
> >  
> >  		cpumask_and(cpus, perf_domain_span(pd), cpu_online_mask);
> > @@ -7774,7 +7773,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> >  				prev_spare_cap = cpu_cap;
> >  				prev_fits = fits;
> >  			} else if ((fits > max_fits) ||
> > -				   ((fits == max_fits) && (cpu_cap > max_spare_cap))) {
> > +				   ((fits == max_fits) && ((long)cpu_cap > max_spare_cap))) {
> >  				/*
> >  				 * Find the CPU with the maximum spare capacity
> >  				 * among the remaining CPUs in the performance
> > @@ -7786,7 +7785,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> >  			}
> >  		}
> >  
> > -		if (max_spare_cap_cpu < 0 && prev_spare_cap == 0)
> > +		if (max_spare_cap_cpu < 0 && prev_spare_cap < 0)
> >  			continue;
> >  
> >  		eenv_pd_busy_time(&eenv, cpus, p);
> > @@ -7794,7 +7793,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> >  		base_energy = compute_energy(&eenv, pd, cpus, p, -1);
> >  
> >  		/* Evaluate the energy impact of using prev_cpu. */
> > -		if (prev_spare_cap > 0) {
> > +		if (prev_spare_cap > -1) {
> >  			prev_delta = compute_energy(&eenv, pd, cpus, p,
> >  						    prev_cpu);
> >  			/* CPU utilization has changed */
> 
> We still need a solution to deal with situations in which `pd + task
> contribution` > `pd_capacity`:
> 
>   compute_energy()
> 
>     if (dst_cpu >= 0)
>      busy_time = min(pd_capacity, pd_busy_time + task_busy_time);
>                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>                                   pd + task contribution
> 
> busy_time is based on util (ENERGY_UTIL), not on the uclamp values
> (FREQUENCY_UTIL) we try to fit into a PD (and finally onto a CPU).
> 
> With that as a reminder for us and the change in the cover letter:

This is not being ignored, but I don't see this as an urgent problem too. There
are more pressing issues that make uclamp_max not effective in practice, and
this ain't a bottleneck yet. Actually it might be doing a good thing as there's
a desire to keep those tasks away on smallest CPU. But we shall visit this
later for sure, don't worry :-) Ultimately we want EAS algorithm to be the
judge of best placement for sure.

I hope to send patches to address load balancer and max aggregation issues in
the coming weeks.

> 
> Reviewed-by: Dietmar Eggemann <dietmar.eggemann@xxxxxxx>

Thanks for the review!

I will wait for the maintainers to see if they would like a v5 to address the
nitpicks or it's actually good enough and happy to pick this up. I think the
commit messages explain the problem clear enough and doesn't warrant sending
a new version. But happy to do so if there's insistence :-)


Thanks!

--
Qais Yousef