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;
Message-ID: <ab445d4d-70d2-9e6f-dd1d-9e71f00c1796@bytedance.com>
Date:   Fri, 22 Jul 2022 15:33:14 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:102.0)
 Gecko/20100101 Thunderbird/102.0.1
Subject: Re: [PATCH 9/9] sched/psi: add PSI_IRQ to track IRQ/SOFTIRQ pressure
Content-Language: en-US
To:     Abel Wu <wuyun.abel@bytedance.com>, hannes@cmpxchg.org,
        surenb@google.com, mingo@redhat.com, peterz@infradead.org,
        tj@kernel.org, corbet@lwn.net, akpm@linux-foundation.org,
        rdunlap@infradead.org
Cc:     linux-doc@vger.kernel.org, linux-kernel@vger.kernel.org,
        songmuchun@bytedance.com, cgroups@vger.kernel.org
References: <20220721040439.2651-1-zhouchengming@bytedance.com>
 <20220721040439.2651-10-zhouchengming@bytedance.com>
 <65d9f79b-be9b-e21e-0624-5c9f2cc0c0b2@bytedance.com>
 <ce22fa9d-aad0-fc23-d304-14fdd27130f4@bytedance.com>
 <5e5d41e2-5f89-8c52-11e5-0c55c5595a88@bytedance.com>
From:   Chengming Zhou <zhouchengming@bytedance.com>
In-Reply-To: <5e5d41e2-5f89-8c52-11e5-0c55c5595a88@bytedance.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2022/7/22 15:14, Abel Wu wrote:
> On 7/22/22 2:13 PM, Chengming Zhou Wrote:
>> On 2022/7/22 11:30, Abel Wu wrote:
>>> Hi Chengming,
>>>
>>> On 7/21/22 12:04 PM, Chengming Zhou Wrote:
>>>> Now PSI already tracked workload pressure stall information for
>>>> CPU, memory and IO. Apart from these, IRQ/SOFTIRQ could have
>>>> obvious impact on some workload productivity, such as web service
>>>> workload.
>>>>
>>>> When CONFIG_IRQ_TIME_ACCOUNTING, we can get IRQ/SOFTIRQ delta time
>>>> from update_rq_clock_task(), in which we can record that delta
>>>> to CPU curr task's cgroups as PSI_IRQ_FULL status.
>>>
>>> The {soft,}irq affection should be equal to all the runnable tasks
>>> on that cpu, not only rq->curr. Further I think irqstall is per-cpu
>>> rather than per-cgroup.
>>
>> Although IRQ/SOFTIRQ is per-cpu, it's the rq->curr who own the CPU at the time
>> and pay for it, meanwhile other groups would be thought as PSI_CPU_FULL.
> 
> I don't think rq->curr pays for it if you mean consuming quota here.

Yes, it makes rq->curr's groups look more productive than it actually is,
which are clearly different from other groups.

> And it doesn't seem appropriate to let other groups treat it as cpu
> stall because the rq->curr is also the victim rather than the one
> causes stall (so it's different from rq->curr causing memstall and
> observed as cpustall by others).

IMHO, we don't care who causes stall, instead we care about what affects
workload productivity.


> 
>>
>> So I think it's reasonable to account this IRQ/SOFTIRQ delta to rq->curr's groups
>> as PSI_IRQ_FULL pressure stall. And per-cpu IRQ stall can also get from psi_system.
>>
>