Re: [RFC PATCH 11/16] sched: migration changes for core scheduling(Internet mail)

[Li, Aubrey]

On 2020/7/22 22:32, benbjiang(èå) wrote:
> Hi,
> 
>> On Jul 22, 2020, at 8:13 PM, Li, Aubrey <aubrey.li@xxxxxxxxxxxxxxx> wrote:
>>
>> On 2020/7/22 16:54, benbjiang(èå) wrote:
>>> Hi, Aubrey,
>>>
>>>> On Jul 1, 2020, at 5:32 AM, Vineeth Remanan Pillai <vpillai@xxxxxxxxxxxxxxxx> wrote:
>>>>
>>>> From: Aubrey Li <aubrey.li@xxxxxxxxx>
>>>>
>>>> - Don't migrate if there is a cookie mismatch
>>>>    Load balance tries to move task from busiest CPU to the
>>>>    destination CPU. When core scheduling is enabled, if the
>>>>    task's cookie does not match with the destination CPU's
>>>>    core cookie, this task will be skipped by this CPU. This
>>>>    mitigates the forced idle time on the destination CPU.
>>>>
>>>> - Select cookie matched idle CPU
>>>>    In the fast path of task wakeup, select the first cookie matched
>>>>    idle CPU instead of the first idle CPU.
>>>>
>>>> - Find cookie matched idlest CPU
>>>>    In the slow path of task wakeup, find the idlest CPU whose core
>>>>    cookie matches with task's cookie
>>>>
>>>> - Don't migrate task if cookie not match
>>>>    For the NUMA load balance, don't migrate task to the CPU whose
>>>>    core cookie does not match with task's cookie
>>>>
>>>> Signed-off-by: Aubrey Li <aubrey.li@xxxxxxxxxxxxxxx>
>>>> Signed-off-by: Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx>
>>>> Signed-off-by: Vineeth Remanan Pillai <vpillai@xxxxxxxxxxxxxxxx>
>>>> ---
>>>> kernel/sched/fair.c  | 64 ++++++++++++++++++++++++++++++++++++++++----
>>>> kernel/sched/sched.h | 29 ++++++++++++++++++++
>>>> 2 files changed, 88 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>>>> index d16939766361..33dc4bf01817 100644
>>>> --- a/kernel/sched/fair.c
>>>> +++ b/kernel/sched/fair.c
>>>> @@ -2051,6 +2051,15 @@ static void task_numa_find_cpu(struct task_numa_env *env,
>>>> 		if (!cpumask_test_cpu(cpu, env->p->cpus_ptr))
>>>> 			continue;
>>>>
>>>> +#ifdef CONFIG_SCHED_CORE
>>>> +		/*
>>>> +		 * 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;
>>>> +#endif
>>>> +
>>>> 		env->dst_cpu = cpu;
>>>> 		if (task_numa_compare(env, taskimp, groupimp, maymove))
>>>> 			break;
>>>> @@ -5963,11 +5972,17 @@ find_idlest_group_cpu(struct sched_group *group, struct task_struct *p, int this
>>>>
>>>> 	/* Traverse only the allowed CPUs */
>>>> 	for_each_cpu_and(i, sched_group_span(group), p->cpus_ptr) {
>>>> +		struct rq *rq = cpu_rq(i);
>>>> +
>>>> +#ifdef CONFIG_SCHED_CORE
>>>> +		if (!sched_core_cookie_match(rq, p))
>>>> +			continue;
>>>> +#endif
>>>> +
>>>> 		if (sched_idle_cpu(i))
>>>> 			return i;
>>>>
>>>> 		if (available_idle_cpu(i)) {
>>>> -			struct rq *rq = cpu_rq(i);
>>>> 			struct cpuidle_state *idle = idle_get_state(rq);
>>>> 			if (idle && idle->exit_latency < min_exit_latency) {
>>>> 				/*
>>>> @@ -6224,8 +6239,18 @@ 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))
>>>> -			break;
>>>> +
>>>> +		if (available_idle_cpu(cpu) || sched_idle_cpu(cpu)) {
>>>> +#ifdef CONFIG_SCHED_CORE
>>>> +			/*
>>>> +			 * If Core Scheduling is enabled, select this cpu
>>>> +			 * only if the process cookie matches core cookie.
>>>> +			 */
>>>> +			if (sched_core_enabled(cpu_rq(cpu)) &&
>>>> +			    p->core_cookie == cpu_rq(cpu)->core->core_cookie)
>>> Why not also add similar logic in select_idle_smt to reduce forced-idle? :)
>> We hit select_idle_smt after we scaned the entire LLC domain for idle cores
>> and idle cpus and failed,so IMHO, an idle smt is probably a good choice under
>> this scenario.
> 
> AFAIC, selecting idle sibling with unmatched cookie will cause unnecessary fored-idle, unfairness and latency, compared to choosing *target* cpu.
Choosing target cpu could increase the runnable task number on the target runqueue, this
could trigger busiest->nr_running > 1 logic and makes the idle sibling trying to pull but
not success(due to cookie not match). Putting task to the idle sibling is relatively stable IMHO.

> Besides, choosing *target* cpu may be more cache friendly. So IMHO, *target* cpu may be a better choice if cookie not match, instead of idle sibling.
I'm not sure if it's more cache friendly as the target is busy, and the coming task
is a cookie unmatched task.

> 
>>
>>>
>>>> +#endif
>>>> +				break;
>>>> +		}
>>>> 	}
>>>>
>>>> 	time = cpu_clock(this) - time;
>>>> @@ -7609,8 +7634,9 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
>>>> 	 * We do not migrate tasks that are:
>>>> 	 * 1) throttled_lb_pair, or
>>>> 	 * 2) cannot be migrated to this CPU due to cpus_ptr, or
>>>> -	 * 3) running (obviously), or
>>>> -	 * 4) are cache-hot on their current CPU.
>>>> +	 * 3) task's cookie does not match with this CPU's core cookie
>>>> +	 * 4) running (obviously), or
>>>> +	 * 5) are cache-hot on their current CPU.
>>>> 	 */
>>>> 	if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
>>>> 		return 0;
>>>> @@ -7645,6 +7671,15 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
>>>> 		return 0;
>>>> 	}
>>>>
>>>> +#ifdef CONFIG_SCHED_CORE
>>>> +	/*
>>>> +	 * 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 0;
>>>> +#endif
>>>> +
>>>> 	/* Record that we found atleast one task that could run on dst_cpu */
>>>> 	env->flags &= ~LBF_ALL_PINNED;
>>>>
>>>> @@ -8857,6 +8892,25 @@ find_idlest_group(struct sched_domain *sd, struct task_struct *p,
>>>> 					p->cpus_ptr))
>>>> 			continue;
>>>>
>>>> +#ifdef CONFIG_SCHED_CORE
>>>> +		if (sched_core_enabled(cpu_rq(this_cpu))) {
>>>> +			int i = 0;
>>>> +			bool cookie_match = false;
>>>> +
>>>> +			for_each_cpu(i, sched_group_span(group)) {
>>> Should we consider the p->cpus_ptr here? like,
>>> 			for_each_cpu_and(i, sched_group_span(group), p->cpus_ptr ) {
>>
>> This is already considered just above #ifdef CONFIG_SCHED_CORE, but not included
>> in the patch file.
>>
>> Thanks,
>> -Aubrey
> 
> The above consideration is,
>  8893                 /* Skip over this group if it has no CPUs allowed */
>  8894                 if (!cpumask_intersects(sched_group_span(group),
>  8895                                         p->cpus_ptr))
>  8896                         continue;
>  8897
> It only considers the case of *p is not allowed for the whole group*, which is not enough.
> If( cpumask_subset(p->cpus_ptr, sched_group_span(group)), the following sched_core_cookie_match() may choose a *wrong(not allowed)* cpu to match cookie. In that case, the matching result could be confusing and lead to wrong result.
> On the other hand, considering p->cpus_ptr here could reduce the loop times and cost, if cpumask_and(p->cpus_ptr, sched_group_span(group)) is the subset of sched_group_span(group).

Though find_idlest_group_cpu() will check p->cpus_ptr again, I believe this is a good catch and
should be fixed in the next iteration.

Thanks,
-Aubrey