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: <85d5087b-450c-351f-270d-c61303cf3187@huawei.com>
Date:   Wed, 27 Jul 2022 12:02:51 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:91.0) Gecko/20100101
 Thunderbird/91.3.2
To:     Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Juri Lelli <juri.lelli@redhat.com>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        "open list:SCHEDULER" <linux-kernel@vger.kernel.org>
From:   "Lihua (lihua, ran)" <hucool.lihua@huawei.com>
Subject: [Question] Reading /proc/stat has a time backward issue
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi all,

I found a problem that the statistical time goes backward, the value read first is 319, and the value read again is 318. As follows：
first：
cat /proc/stat |  grep cpu1
cpu1    319    0    496    41665    0    0    0    0    0    0
then：
cat /proc/stat |  grep cpu1
cpu1    318    0    497    41674    0    0    0    0    0    0

Time goes back, which is counterintuitive.

After debug this, I found that the problem is caused by the implementation of kcpustat_cpu_fetch_vtime. As follows：

                               CPU0                                                                          CPU1
First:
show_stat():
     ->kcpustat_cpu_fetch()
         ->kcpustat_cpu_fetch_vtime()
             ->cpustat[CPUTIME_USER] = kcpustat_cpu(cpu) + vtime->utime + delta;              rq->curr is in user mod
              ---> When CPU1 rq->curr running on userspace, need add utime and delta
                                                                                              --->  rq->curr->vtime->utime is less than 1 tick
Then:
show_stat():
     ->kcpustat_cpu_fetch()
         ->kcpustat_cpu_fetch_vtime()
             ->cpustat[CPUTIME_USER] = kcpustat_cpu(cpu);                                     rq->curr is in kernel mod
             ---> When CPU1 rq->curr running on kernel space, just got kcpustat

Because the values ​​of utime、 stime and delta are temporarily written to cpustat. Therefore, there are two problems  read from /proc/stat:
1. There may be a regression phenomenon;
2. When there are many tasks, the statistics are not accurate enough when utime and stime do not exceed one TICK.
The time goes back is counterintuitive, and I want to discuss whether there is a good solution without compromising performance.

Thanks a lot.