lkml.org 
[lkml]   [2018]   [Feb]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RESEND PATCH] sched/fair: consider RT/IRQ pressure in select_idle_sibling
From
Date
On 02/09/2018 11:05 PM, Rohit Jain wrote:
>
> On 02/09/2018 07:46 AM, Dietmar Eggemann wrote:
>> On 02/09/2018 01:53 PM, Peter Zijlstra wrote:
>>> On Mon, Jan 29, 2018 at 03:27:09PM -0800, Rohit Jain wrote:
>>
>> [...]
>>
>>>> @@ -6173,8 +6183,15 @@ static int select_idle_cpu(struct task_struct
>>>> *p, struct sched_domain *sd, int t
>>>>               return -1;
>>>>           if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
>>>>               continue;
>>>> +        if (idle_cpu(cpu)) {
>>>> +            if (full_capacity(cpu)) {
>>>> +                best_cpu = cpu;
>>>> +                break;
>>>> +            } else if (capacity_of(cpu) > best_cap) {
>>>> +                best_cap = capacity_of(cpu);
>>>> +                best_cpu = cpu;
>>>> +            }
>>>> +        }
>>>
>>> No need for the else. And you'll note you're once again inconsistent
>>> with your previous self.
>>>
>>> But here I worry about big.little a wee bit. I think we're allowed big
>>> and little cores on the same L3 these days, and you can't directly
>>> compare capacity between them.
>>>
>>> Morten / Dietmar, any comments?
>>
>> Yes, for DynamIQ (big.little successor) systems, those cpus can have
>> different capacity_orig_of() values already.
>>
>
> OK, given that there are asymmetric capacities in L3 cores, we would
> probably have something like the below(?) in select_idle_cpu:
>
>           if (!cpumask_test_cpu(cpu, &p->cpus_allowed))
>               continue;
> +        if (idle_cpu(cpu) && !reduced_capacity(cpu))
> +            break;
>

Only returning the first idle cpu w/o reduced capacity will definitely
work better here on future DynamIQ systems than trying to find the
best_cap cpu by taking capacity_of() into consideration.

\
 
 \ /
  Last update: 2018-02-14 10:12    [W:0.055 / U:0.104 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site