Re: [PATCH] sched/fair: Check a task has a fitting cpu when updating misfit

[Dietmar Eggemann <dietmar.eggemann@arm.com>]

On 29/08/2023 17:35, Qais Yousef wrote:
> On 08/29/23 16:10, Vincent Guittot wrote:
>> On Sun, 20 Aug 2023 at 22:34, Qais Yousef <qyousef@layalina.io> wrote:

[...]

>>>  kernel/sched/fair.c | 50 ++++++++++++++++++++++++++++++++++++++-------
>>>  1 file changed, 43 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>>> index 0b7445cd5af9..f08c5f3bf895 100644
>>> --- a/kernel/sched/fair.c
>>> +++ b/kernel/sched/fair.c
>>> @@ -4853,17 +4853,50 @@ static inline int task_fits_cpu(struct task_struct *p, int cpu)
>>>
>>>  static inline void update_misfit_status(struct task_struct *p, struct rq *rq)
>>>  {
>>> +       unsigned long uclamp_min, uclamp_max;
>>> +       unsigned long util, cap_level;
>>> +       bool has_fitting_cpu = false;
>>> +       int cpu = cpu_of(rq);
>>> +
>>>         if (!sched_asym_cpucap_active())
>>>                 return;
>>>
>>> -       if (!p || p->nr_cpus_allowed == 1) {
>>> -               rq->misfit_task_load = 0;
>>> -               return;
>>> -       }
>>> +       if (!p || p->nr_cpus_allowed == 1)
>>> +               goto out;
>>>
>>> -       if (task_fits_cpu(p, cpu_of(rq))) {
>>> -               rq->misfit_task_load = 0;
>>> -               return;
>>> +       uclamp_min = uclamp_eff_value(p, UCLAMP_MIN);
>>> +       uclamp_max = uclamp_eff_value(p, UCLAMP_MAX);
>>> +       util = task_util_est(p);
>>> +
>>> +       if (util_fits_cpu(util, uclamp_min, uclamp_max, cpu) > 0)
>>> +               goto out;

util_fits_cpu() checks fits_capacity(util, capacity_of(cpu)) but the
capacity pressure could change between update_misfit_status() and CFS lb?

>>> +
>>> +       cap_level = capacity_orig_of(cpu);
>>> +
>>> +       /* If we can't fit the biggest CPU, that's the best we can ever get. */
>>> +       if (cap_level == SCHED_CAPACITY_SCALE)
>>> +               goto out;
>>> +
>>> +       /*
>>> +        * If the task affinity is not set to default, make sure it is not
>>> +        * restricted to a subset where no CPU can ever fit it. Triggering
>>> +        * misfit in this case is pointless as it has no where better to move
>>> +        * to. And it can lead to balance_interval to grow too high as we'll
>>> +        * continuously fail to move it anywhere.
>>> +        */
>>> +       if (!cpumask_equal(p->cpus_ptr, cpu_possible_mask)) {
>>> +               for_each_cpu(cpu, p->cpus_ptr) {
>>
>> I haven't looked at the problem in detail and at other possibilities
>> so far but for_each_cpu doesn't scale and update_misfit_status() being
>> called in pick_next_task_fair() so you must find another way to detect
>> this
> 
> Okay, will do.

We have LIST_HEAD(asym_cap_list) (list of cpumasks according to
cpu_capacity_orig CPU groups) in kernel/sched/topology.c to set
SD_ASYM_CPUCAPACITY{,_FULL} for asymmetric CPU capacity systems.
Maybe this could be made usable in fair.c as well?

But checking via util_fits_cpu() wouldn't work then since it's per-CPU.
The check of p's CPU affinity, its uclamped util_avg value and the
cpu_capacity_orig is sufficient here. Then using those cpumasks could work.