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;
MIME-Version: 1.0
References: <20230317063246.1128219-1-zangchunxin@lixiang.com>
In-Reply-To: <20230317063246.1128219-1-zangchunxin@lixiang.com>
From:   Namhyung Kim <namhyung@kernel.org>
Date:   Mon, 27 Mar 2023 12:00:52 -0700
Message-ID: <CAM9d7chc+eJvPMm3ZhcxcbYUjNzVvBG7N0DvVwMzmOOb+-xiMQ@mail.gmail.com>
Subject: Re: [PATCH v3] perf sched: Fix sched latency analysis incorrect
To:     Chunxin Zang <zangchunxin@lixiang.com>
Cc:     peterz@infradead.org, mingo@redhat.com, acme@kernel.org,
        mark.rutland@arm.com, alexander.shishkin@linux.intel.com,
        jolsa@kernel.org, irogers@google.com,
        linux-perf-users@vger.kernel.org, linux-kernel@vger.kernel.org,
        Jerry Zhou <zhouchunhua@lixiang.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
Precedence: bulk

On Thu, Mar 16, 2023 at 11:33 PM Chunxin Zang <zangchunxin@lixiang.com> wrote:
>
> 'perf sched latency' is incorrect to get process schedule latency
> when it used 'sched:sched_wakeup' to analysis perf.data.
>
> Because 'perf record' prefer use 'sched:sched_waking' to
> 'sched:sched_wakeup' since commit d566a9c2d482 ("perf sched: Prefer
> sched_waking event when it exists"). It's very reasonable to
> evaluate process schedule latency.
>
> Similarly, update sched latency/map/replay to use sched_waking events.

Have you checked perf sched replay (add_sched_event_wakeup)?  I think
it'd make unnecessary sched atoms for sched_wakeup.

Can we check the wakeup and waking events and disable the wakeup
if the waking is found like in perf_sched__timehist?

Thanks,
Namhyung


>
> I used 'perf record -e "sched:* "' to record both wakeup and waking events,
> and use fixed perf version to analysis them, the result is correct.
> Because the function "latency_wakeup_event" will change atom->state to
> THREAD_WAIT_CPU at waking event, and judge the 'atom->state !=
> THREAD_SLEEPING' is false then returned at wakeup event.
>
> Signed-off-by: Chunxin Zang <zangchunxin@lixiang.com>
> Signed-off-by: Jerry Zhou <zhouchunhua@lixiang.com>
> ---
>
>         changelogs in v3:
>         1) fix non-ASCII characters in commit log.
>
>         changelogs in v2:
>         1) fix email address disappearing in 'signed off by'
>
>  tools/perf/builtin-sched.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c
> index 86e18575c9be..1af4ec1ac824 100644
> --- a/tools/perf/builtin-sched.c
> +++ b/tools/perf/builtin-sched.c
> @@ -1819,6 +1819,7 @@ static int perf_sched__read_events(struct perf_sched *sched)
>         const struct evsel_str_handler handlers[] = {
>                 { "sched:sched_switch",       process_sched_switch_event, },
>                 { "sched:sched_stat_runtime", process_sched_runtime_event, },
> +               { "sched:sched_waking",       process_sched_wakeup_event, },
>                 { "sched:sched_wakeup",       process_sched_wakeup_event, },
>                 { "sched:sched_wakeup_new",   process_sched_wakeup_event, },
>                 { "sched:sched_migrate_task", process_sched_migrate_task_event, },
> --
> 2.25.1
>
> 声明：这封邮件只允许文件接收者阅读，有很高的机密性要求。禁止其他人使用、打开、复制或转发里面的任何内容。如果本邮件错误地发给了你，请联系邮件发出者并删除这个文件。机密及法律的特权并不因为误发邮件而放弃或丧失。任何提出的观点或意见只属于作者的个人见解，并不一定代表本公司。