Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.36 as permitted sender) client-ip=23.128.96.36;
MIME-Version: 1.0
References: <20231127130754.3affb908@gandalf.local.home> <20231130094130.GA12116@didi-ThinkCentre-M930t-N000>
 <20231130120036.5EpZn4eo@linutronix.de>
In-Reply-To: <20231130120036.5EpZn4eo@linutronix.de>
From:   Yuanhan Zhang <zyhtheonly@gmail.com>
Date:   Fri, 1 Dec 2023 15:27:39 +0800
Message-ID: <CAEQmJ=gh8fdPiBZPo72wwwr8UWo9BDF8cG1aUS5jFsn1itXOAg@mail.gmail.com>
Subject: Re: [PATCH v2] sched/cputime: exclude ktimers threads in irqtime_account_irq
To:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc:     rostedt@goodmis.org, tglx@linutronix.de, mingo@redhat.com,
        peterz@infradead.org, juri.lelli@redhat.com,
        vincent.guittot@linaro.org, linux-kernel@vger.kernel.org,
        dietmar.eggemann@arm.com, bsegall@google.com, mgorman@suse.de,
        bristot@redhat.com, vschneid@redhat.com, zyhtheonly@yeah.net
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Hi Sebastian, thanks for your reply.

Sebastian Andrzej Siewior <bigeasy@linutronix.de> =E4=BA=8E2023=E5=B9=B411=
=E6=9C=8830=E6=97=A5=E5=91=A8=E5=9B=9B 20:00=E5=86=99=E9=81=93=EF=BC=9A
>
> On 2023-11-30 17:41:47 [+0800], tiozhang wrote:
> > In CONFIG_PREEMPT_RT kernel, ktimers also calls __do_softirq,
> > so when accounting CPUTIME_SOFTIRQ, ktimers need to be excluded
> > as well as ksoftirqd.
> > Also add this_cpu_ktimers to keep consistency with this_cpu_ksoftirqd.
>
> I'm still not sure what the benefit here is. It says align with
> ksoftirqd but why? Why don't we account softirq time for ksoftirqd (and
> should continue to do so for ktimersd)?

That is my miss. When CONFIG_IRQ_TIME_ACCOUNTING is enabled,
ksoftirqd is counted elsewhere (where ktimers should also be aligned).
Please review my later patch v3.

>
> ktimers runs almost all the time in softirq context. So does every
> force-threaded interrupt. Should we exclude them, too?

For force-threaded interrupt, it counts on CPUTIME_SYSTEM
instead of CPUTIME_IRQ nor CPUTIME_SOFTIRQ.
To me, it does not quite make sense, I'm also thinking of sending a patch
of this, but IMHO it should not be considered in this patch...

>
> Sebastian