Received-SPF: softfail (google.com: domain of transitioning linux-kernel-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
MIME-Version: 1.0
References: <20220513005427.2507335-1-joshdon@google.com> <Yn3JTpXhWRTeahyG@slm.duckdns.org>
In-Reply-To: <Yn3JTpXhWRTeahyG@slm.duckdns.org>
From:   Josh Don <joshdon@google.com>
Date:   Fri, 13 May 2022 12:23:16 -0700
Message-ID: <CABk29Nv=J_ZUnDTkRhwdQop=REr_XDGjJxn_zVy4kBqwx8K57w@mail.gmail.com>
Subject: Re: [PATCH] sched/core: add forced idle accounting for cgroups
To:     Tejun Heo <tj@kernel.org>
Cc:     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>,
        Valentin Schneider <vschneid@redhat.com>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        Cruz Zhao <CruzZhao@linux.alibaba.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Thanks Tejun,

On Thu, May 12, 2022 at 7:58 PM Tejun Heo <tj@kernel.org> wrote:
>
> On Thu, May 12, 2022 at 05:54:27PM -0700, Josh Don wrote:
> > 4feee7d1260 previously added per-task forced idle accounting. This patch
> > extends this to also include cgroups.
> >
> > rstat is used for cgroup accounting, except for the root, which uses
> > kcpustat in order to bypass the need for doing an rstat flush when
> > reading root stats.
> >
> > Only cgroup v2 is supported. Similar to the task accounting, the cgroup
> > accounting requires that schedstats is enabled.
>
> We've been collecting scheduler stats in cgroup core so that we always have
> them available whether cpu controller is enabled or not. There's nothing
> actually specific to cpu controller, right? Would it make sense to collect
> the cpu core stats the same way as the rest of scheduler stats?

Yea, that's right, this doesn't require the cpu controller to be
enabled. Are you suggesting to add a new field to cgroup_base_stat?

One other weird artifact of collecting forceidle time is that a cpu
may account it on behalf of its hyperthread sibling. Currently, the
core rstat code always accounts to the current cpu's percpu rstat
field. I can add an accounting function to support writes to a
different cpu's field, in order to make sure that the per-cpu totals
are correct (the forceidle accounting code holds rq->__lock, which
protects all HT siblings of a core). percpu totals aren't currently
exported in cgroup v2, but this is useful information that we'll
consume, so it would be nice to keep it accurate.

Best,
Josh