Re: [PATCH v2 2/6] sched/deadline: Optimize dl_bw_cpus()

[Dietmar Eggemann]

On 30/04/2020 12:55, Pavan Kondeti wrote:
> On Mon, Apr 27, 2020 at 10:37:05AM +0200, Dietmar Eggemann wrote:

[..]

>> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
>> index 504d2f51b0d6..4ae22bfc37ae 100644
>> --- a/kernel/sched/deadline.c
>> +++ b/kernel/sched/deadline.c
>> @@ -54,10 +54,16 @@ static inline struct dl_bw *dl_bw_of(int i)
>>  static inline int dl_bw_cpus(int i)
>>  {
>>  	struct root_domain *rd = cpu_rq(i)->rd;
>> -	int cpus = 0;
>> +	int cpus;
>>  
>>  	RCU_LOCKDEP_WARN(!rcu_read_lock_sched_held(),
>>  			 "sched RCU must be held");
>> +
>> +	if (cpumask_subset(rd->span, cpu_active_mask))
>> +		return cpumask_weight(rd->span);
>> +
> 
> Looks good to me. This is a nice optimization.
> 
>> +	cpus = 0;
>> +
>>  	for_each_cpu_and(i, rd->span, cpu_active_mask)
>>  		cpus++;
>>  
> Do you know why this check is in place? Is it only to cover
> the case of cpuset_cpu_inactive()->dl_cpu_busy()?

It should cover:

(1) Preventing CPU hp when DL detects a possible overflow w/o the CPU:

    sched_cpu_deactivate() -> cpuset_cpu_inactive() -> dl_cpu_busy() ->
    dl_bw_cpus() [now replaced by dl_bw_capacity()].

(2) DL Admission Control in CPU HP:

    __sched_setscheduler() -> sched_dl_overflow() -> dl_bw_cpus()
                                           [now + -> dl_bw_capacity()]

(3) In create/destroy exclusive cpusets scenarios (comment in
    set_cpus_allowed_dl(), although I wasn't able to provoke this so
    far:

    do_set_cpus_allowed() -> p->sched_class->set_cpus_allowed() was
    never called when I ran a DL testcase and create/destroy exclusive
    cpusets at the same time?