Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
CC:     <yangyicong@hisilicon.com>, Josh Don <joshdon@google.com>,
        Chen Yu <yu.c.chen@intel.com>, <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 1/5] sched/fair: ignore SIS_UTIL when has idle core
To:     Abel Wu <wuyun.abel@bytedance.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Mel Gorman <mgorman@suse.de>,
        Vincent Guittot <vincent.guittot@linaro.org>
References: <20220712082036.5130-1-wuyun.abel@bytedance.com>
 <20220712082036.5130-2-wuyun.abel@bytedance.com>
 <8e7d75d4-613e-f35e-e932-323789666fb1@huawei.com>
 <4dde05be-8470-5984-0a30-ba077b9fe6bd@bytedance.com>
From:   Yicong Yang <yangyicong@huawei.com>
Message-ID: <e8a59a8f-1e0c-2bb6-2d1b-4e76f5c511f5@huawei.com>
Date:   Thu, 14 Jul 2022 15:15:04 +0800
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101
 Thunderbird/78.5.1
MIME-Version: 1.0
In-Reply-To: <4dde05be-8470-5984-0a30-ba077b9fe6bd@bytedance.com>
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 2022/7/14 14:58, Abel Wu wrote:
> 
> On 7/14/22 2:19 PM, Yicong Yang Wrote:
>> On 2022/7/12 16:20, Abel Wu wrote:
>>> When SIS_UTIL is enabled, SIS domain scan will be skipped if
>>> the LLC is overloaded. Since the overloaded status is checked
>>> in the load balancing at LLC level, the interval is llc_size
>>> miliseconds. The duration might be long enough to affect the
>>> overall system throughput if idle cores are out of reach in
>>> SIS domain scan.
>>>
>>> Signed-off-by: Abel Wu <wuyun.abel@bytedance.com>
>>> ---
>>>   kernel/sched/fair.c | 15 +++++++++------
>>>   1 file changed, 9 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>>> index a78d2e3b9d49..cd758b3616bd 100644
>>> --- a/kernel/sched/fair.c
>>> +++ b/kernel/sched/fair.c
>>> @@ -6392,16 +6392,19 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, bool
>>>       struct sched_domain *this_sd;
>>>       u64 time = 0;
>>>   -    this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
>>> -    if (!this_sd)
>>> -        return -1;
>>> -
>>>       cpumask_and(cpus, sched_domain_span(sd), p->cpus_ptr);
>>>   -    if (sched_feat(SIS_PROP) && !has_idle_core) {
>>> +    if (has_idle_core)
>>> +        goto scan;
>>> +
>>> +    if (sched_feat(SIS_PROP)) {
>>>           u64 avg_cost, avg_idle, span_avg;
>>>           unsigned long now = jiffies;
>>>   +        this_sd = rcu_dereference(*this_cpu_ptr(&sd_llc));
>>> +        if (!this_sd)
>>> +            return -1;
>>> +
>>
>> I don't follow the change here. True that this_sd is used only in SIS_PROP, but it seems irrelevant with your
>> commit. Does the position of this make any performance difference?
> 
> No, this change doesn't make much difference to performance. Are
> you suggesting that I should make this a separate patch?
> 

It just makes me think that dereference is unnecessary if this_cpu and target locates in
the same LLC, since it's already been passed. But since you noticed no difference it may
have little effect. :)

> Thanks,
> Abel
> 
>>
>> Thanks.
>>
>>>           /*
>>>            * If we're busy, the assumption that the last idle period
>>>            * predicts the future is flawed; age away the remaining
>>> @@ -6436,7 +6439,7 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, bool
>>>                   return -1;
>>>           }
>>>       }
>>> -
>>> +scan:
>>>       for_each_cpu_wrap(cpu, cpus, target + 1) {
>>>           if (has_idle_core) {
>>>               i = select_idle_core(p, cpu, cpus, &idle_cpu);
>>>
> .