On 01/10/2019 11:14, Vincent Guittot wrote:
> group_asym_packing
>
> On Tue, 1 Oct 2019 at 10:15, Dietmar Eggemann <dietmar.eggem...@arm.com>
> wrote:
>>
>> On 19/09/2019 09:33, Vincent Guittot wrote:
>>
>>
>> [...]
>>
>>> @@ -8042,14 +8104,24 @@ static inline void update_sg_lb_stats(struct lb_env
>>> *env,
>>> }
>>> }
>>>
>>> - /* Adjust by relative CPU capacity of the group */
>>> + /* Check if dst cpu is idle and preferred to this group */
>>> + if (env->sd->flags & SD_ASYM_PACKING &&
>>> + env->idle != CPU_NOT_IDLE &&
>>> + sgs->sum_h_nr_running &&
>>> + sched_asym_prefer(env->dst_cpu, group->asym_prefer_cpu)) {
>>> + sgs->group_asym_packing = 1;
>>> + }
>>> +
>>
>> Since asym_packing check is done per-sg rather per-CPU (like misfit_task),
>> can you
>> not check for asym_packing in group_classify() directly (like for
>> overloaded) and
>> so get rid of struct sg_lb_stats::group_asym_packing?
>
> asym_packing uses a lot of fields of env to set group_asym_packing but
> env is not statistics and i'd like to remove env from
> group_classify() argument so it will use only statistics.
> At now, env is an arg of group_classify only for imbalance_pct field
> and I have replaced env by imabalnce_pct in a patch that is under
> preparation.
> With this change, I can use group_classify outside load_balance()
OK, I see. This relates to the 'update find_idlest_group() to be more
aligned with load_balance()' point mentioned in the cover letter I
assume. To make sure we can use load instead of runnable_load there as well.
[...]
