[PATCH v8 2/2] sched/fair: Scan cluster before scanning LLC in wake-up path

[Yicong Yang]

On 2023/6/12 13:22, Chen Yu wrote:
> On 2023-06-12 at 10:31:39 +0530, Gautham R. Shenoy wrote:
>> Hello Yicong,
>>
>>
>> On Tue, May 30, 2023 at 03:02:53PM +0800, Yicong Yang wrote:
>>> From: Barry Song <song.bao.hua at hisilicon.com>
>> [..snip..]
>>
>>> @@ -7103,7 +7127,7 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
>>>  	bool has_idle_core = false;
>>>  	struct sched_domain *sd;
>>>  	unsigned long task_util, util_min, util_max;
>>> -	int i, recent_used_cpu;
>>> +	int i, recent_used_cpu, prev_aff = -1;
>>>  
>>>  	/*
>>>  	 * On asymmetric system, update task utilization because we will check
>>> @@ -7130,8 +7154,11 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
>>>  	 */
>>>  	if (prev != target && cpus_share_cache(prev, target) &&
>>>  	    (available_idle_cpu(prev) || sched_idle_cpu(prev)) &&
>>> -	    asym_fits_cpu(task_util, util_min, util_max, prev))
>>> -		return prev;
>>> +	    asym_fits_cpu(task_util, util_min, util_max, prev)) {
>>> +		if (cpus_share_lowest_cache(prev, target))
>>
>> For platforms without the cluster domain, the cpus_share_lowest_cache
>> check is a repetition of the cpus_share_cache(prev, target) check. Can
>> we avoid this using a static branch check for cluster ?
>>
>>
> Sounds good. 
>>> +			return prev;
>>> +		prev_aff = prev;
>>> +	}
>>>  
>>>  	/*
>>>  	 * Allow a per-cpu kthread to stack with the wakee if the
>>> @@ -7158,7 +7185,10 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
>>>  	    (available_idle_cpu(recent_used_cpu) || sched_idle_cpu(recent_used_cpu)) &&
>>>  	    cpumask_test_cpu(p->recent_used_cpu, p->cpus_ptr) &&
>>>  	    asym_fits_cpu(task_util, util_min, util_max, recent_used_cpu)) {
>>> -		return recent_used_cpu;
>>> +		if (cpus_share_lowest_cache(recent_used_cpu, target))
>>
>> Same here.
>>
>>> +			return recent_used_cpu;
>>> +	} else {
>>> +		recent_used_cpu = -1;
>>>  	}
>>>  
>>>  	/*
>>> @@ -7199,6 +7229,17 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
>>>  	if ((unsigned)i < nr_cpumask_bits)
>>>  		return i;
>>>  
>>> +	/*
>>> +	 * For cluster machines which have lower sharing cache like L2 or
>>> +	 * LLC Tag, we tend to find an idle CPU in the target's cluster
>>> +	 * first. But prev_cpu or recent_used_cpu may also be a good candidate,
>>> +	 * use them if possible when no idle CPU found in select_idle_cpu().
>>> +	 */
>>> +	if ((unsigned int)prev_aff < nr_cpumask_bits)
>>> +		return prev_aff;
>>
>> Shouldn't we check if prev_aff (and the recent_used_cpu below) is
>> still idle ?
>>
>>
> When we reach here, the target is non-idle, and the prev_aff is idle.
> Although there is a race condition that prev_aff becomes non-idle
> and target becomes idle after select_idle_cpu(), this window might be
> small IMO.
> 

Yes. Races here but adding a check won't cost much, so it's ok for me
to check it or not.

Thanks.