Re: [RFC 4/6] Add cpumask to track throughput intensive tasks

[Parth Shah <parth015@linux.vnet.ibm.com>]

On 4/12/19 4:43 PM, Dietmar Eggemann wrote:
> On 3/22/19 7:06 AM, Parth Shah wrote:
> 
> [...]
> 
>> @@ -7038,6 +7046,11 @@ pick_next_task_fair(struct rq *rq, struct task_struct *prev, struct rq_flags *rf
>>       struct task_struct *p;
>>       int new_tasks;
>>   +    if (!core_underutilized(cpu_util(rq->cpu),capacity_of(rq->cpu)))
>> +        cpumask_test_and_set_cpu(rq->cpu, highutil_task_cpu_mask);
>> +    else
>> +        cpumask_test_and_clear_cpu(rq->cpu, highutil_task_cpu_mask);
>> +
> 
> Shouldn't this update of the highutil_task_cpu_mask also be under the hood of turbo_sched_enabled()?
> 
> [...]
> 

So if this is kept inside turbo_sched_enabled, then the highutil_task_cpu_mask will contain irrelevant
information when the turbo_sched is enabled. This will result in selecting wrong CPU unless
pick_next_task_fair gets invoked in most CPUs.

Maybe, that is okay since it impacts only during the time of enabling the feature. Also, since I will
remove CONFIG options, almost every piece of TurboSched code will have to pass through turbo_sched_enabled() condition.

Thanks