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;
Date:   Wed, 29 Mar 2023 19:20:59 +0000
In-Reply-To: <CAJD7tkb-UpKm2QbjYzB=B=oGk6Hyj9cbUviZUPC+7VsvBecH7g@mail.gmail.com>
Mime-Version: 1.0
References: <20230328221644.803272-1-yosryahmed@google.com>
 <20230328221644.803272-5-yosryahmed@google.com> <ZCQfZJFufkJ10o01@dhcp22.suse.cz>
 <CAJD7tkb-UpKm2QbjYzB=B=oGk6Hyj9cbUviZUPC+7VsvBecH7g@mail.gmail.com>
Message-ID: <20230329192059.2nlme5ubshzdbpg6@google.com>
Subject: Re: [PATCH v2 4/9] cgroup: rstat: add WARN_ON_ONCE() if flushing
 outside task context
From:   Shakeel Butt <shakeelb@google.com>
To:     Yosry Ahmed <yosryahmed@google.com>
Cc:     Michal Hocko <mhocko@suse.com>, Tejun Heo <tj@kernel.org>,
        Josef Bacik <josef@toxicpanda.com>,
        Jens Axboe <axboe@kernel.dk>,
        Zefan Li <lizefan.x@bytedance.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Muchun Song <muchun.song@linux.dev>,
        Andrew Morton <akpm@linux-foundation.org>,
        "Michal =?utf-8?Q?Koutn=C3=BD?=" <mkoutny@suse.com>,
        Vasily Averin <vasily.averin@linux.dev>,
        cgroups@vger.kernel.org, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        bpf@vger.kernel.org
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, Mar 29, 2023 at 11:41:39AM -0700, Yosry Ahmed wrote:
> On Wed, Mar 29, 2023 at 4:22=E2=80=AFAM Michal Hocko <mhocko@suse.com> wr=
ote:
> >
> > On Tue 28-03-23 22:16:39, Yosry Ahmed wrote:
> > > rstat flushing is too expensive to perform in irq context.
> > > The previous patch removed the only context that may invoke an rstat
> > > flush from irq context, add a WARN_ON_ONCE() to detect future
> > > violations, or those that we are not aware of.
> > >
> > > Ideally, we wouldn't flush with irqs disabled either, but we have one
> > > context today that does so in mem_cgroup_usage(). Forbid callers from
> > > irq context for now, and hopefully we can also forbid callers with ir=
qs
> > > disabled in the future when we can get rid of this callsite.
> >
> > I am sorry to be late to the discussion. I wanted to follow up on
> > Johannes reply in the previous version but you are too fast ;)
> >
> > I do agree that this looks rather arbitrary. You do not explain how the
> > warning actually helps. Is the intention to be really verbose to the
> > kernel log when somebody uses this interface from the IRQ context and
> > get bug reports? What about configurations with panic on warn? Do we
> > really want to crash their systems for something like that?
>=20
> Thanks for taking a look, Michal!
>=20
> The ultimate goal is not to flush in irq context or with irqs
> disabled, as in some cases it causes irqs to be disabled for a long
> time, as flushing is an expensive operation. The previous patch in the
> series should have removed the only context that flushes in irq
> context, and the purpose of the WARN_ON_ONCE() is to catch future uses
> or uses that we might have missed.
>=20
> There is still one code path that flushes with irqs disabled (also
> mem_cgroup_usage()), and we cannot remove this just yet; we need to
> deprecate usage threshold events for root to do that. So we cannot
> enforce not flushing with irqs disabled yet.
>=20
> So basically the patch is trying to enforce what we have now, not
> flushing in irq context, and hopefully at some point we will also be
> able to enforce not flushing with irqs disabled.
>=20
> If WARN_ON_ONCE() is the wrong tool for this, please let me know.
>=20

If I understand Michal's concern, the question is should be start with
pr_warn_once() instead of WARN_ON_ONCE() and I think yes we should start
with pr_warn_once().