Re: [PATCH 1/6] sched: migration changes for core scheduling

[Li, Aubrey]

On 2021/3/22 20:56, Peter Zijlstra wrote:
> On Mon, Mar 22, 2021 at 08:31:09PM +0800, Li, Aubrey wrote:
>> Please let me know if I put cookie match check at the right position
>> in task_hot(), if so, I'll obtain some performance data of it.
>>
>> Thanks,
>> -Aubrey
>>
>> =======================================================
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index 7f2fb08..d4bdcf9 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -1912,6 +1912,13 @@ static void task_numa_find_cpu(struct task_numa_env *env,
>>  		if (!cpumask_test_cpu(cpu, env->p->cpus_ptr))
>>  			continue;
>>  
>> +		/*
>> +		 * Skip this cpu if source task's cookie does not match
>> +		 * with CPU's core cookie.
>> +		 */
>> +		if (!sched_core_cookie_match(cpu_rq(cpu), env->p))
>> +			continue;
>> +
>>  		env->dst_cpu = cpu;
>>  		if (task_numa_compare(env, taskimp, groupimp, maymove))
>>  			break;
> 
> This one might need a little help too, I've not fully considered NUMA
> balancing though.
> 
>> @@ -6109,7 +6120,9 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
>>  	for_each_cpu_wrap(cpu, cpus, target) {
>>  		if (!--nr)
>>  			return -1;
>> -		if (available_idle_cpu(cpu) || sched_idle_cpu(cpu))
>> +
>> +		if ((available_idle_cpu(cpu) || sched_idle_cpu(cpu)) &&
>> +		    sched_cpu_cookie_match(cpu_rq(cpu), p))
>>  			break;
>>  	}
>>  
> 
> This doesn't even apply... That code has changed.
> 
>> @@ -7427,6 +7440,14 @@ static int task_hot(struct task_struct *p, struct lb_env *env)
>>  
>>  	if (sysctl_sched_migration_cost == -1)
>>  		return 1;
>> +
>> +	/*
>> +	 * Don't migrate task if the task's cookie does not match
>> +	 * with the destination CPU's core cookie.
>> +	 */
>> +	if (!sched_core_cookie_match(cpu_rq(env->dst_cpu), p))
>> +		return 1;
>> +
>>  	if (sysctl_sched_migration_cost == 0)
>>  		return 0;
>>  
> 
> Should work I think, but you've put it in a weird spot for breaking up
> that sysctl_sched_migration_cost thing. I'd have put it either in front
> or after that other SMT thing we have there.
> 

I did it on purpose.

If migration cost is huge, the task should not migrate, no matter the
cookie is matched or not. So have to after sysctl_sched_migration_cost == -1.

And if migration cost = 0 or delta < migrate cost , the task can be migrated,
but before migrate, We need to check whether cookie is matched or not. So before
sysctl_sched_migration_cost == 0.

Please correct me if I was wrong.

Thanks,
-Aubrey