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: <20230323040037.2389095-1-yosryahmed@google.com>
 <20230323040037.2389095-3-yosryahmed@google.com> <CALvZod5MnM8UJ0pj44QYb4sVwgFZ1B2KpSL6oqBQbJU3wH6eNA@mail.gmail.com>
In-Reply-To: <CALvZod5MnM8UJ0pj44QYb4sVwgFZ1B2KpSL6oqBQbJU3wH6eNA@mail.gmail.com>
From:   Yosry Ahmed <yosryahmed@google.com>
Date:   Wed, 22 Mar 2023 22:16:40 -0700
Message-ID: <CAJD7tkZJWXDinusUeYNBf_qov0+4ug2hG75Ge8NuP=6jG7+byA@mail.gmail.com>
Subject: Re: [RFC PATCH 2/7] memcg: do not disable interrupts when holding stats_flush_lock
To:     Shakeel Butt <shakeelb@google.com>
Cc:     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>,
        Michal Hocko <mhocko@kernel.org>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Muchun Song <muchun.song@linux.dev>,
        Andrew Morton <akpm@linux-foundation.org>,
        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 22, 2023 at 9:32=E2=80=AFPM Shakeel Butt <shakeelb@google.com> =
wrote:
>
> On Wed, Mar 22, 2023 at 9:00=E2=80=AFPM Yosry Ahmed <yosryahmed@google.co=
m> wrote:
> >
> > The rstat flushing code was modified so that we do not disable interrup=
ts
> > when we hold the global rstat lock. Do the same for stats_flush_lock on
> > the memcg side to avoid unnecessarily disabling interrupts throughout
> > flushing.
> >
> > Since the code exclusively uses trylock to acquire this lock, it should
> > be fine to hold from interrupt contexts or normal contexts without
> > disabling interrupts as a deadlock cannot occur. For interrupt contexts
> > we will return immediately without flushing anyway.
> >
> > Signed-off-by: Yosry Ahmed <yosryahmed@google.com>
> > ---
> >  mm/memcontrol.c | 10 ++++++----
> >  1 file changed, 6 insertions(+), 4 deletions(-)
> >
> > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > index 5abffe6f8389..e0e92b38fa51 100644
> > --- a/mm/memcontrol.c
> > +++ b/mm/memcontrol.c
> > @@ -636,15 +636,17 @@ static inline void memcg_rstat_updated(struct mem=
_cgroup *memcg, int val)
> >
> >  static void __mem_cgroup_flush_stats(void)
> >  {
> > -       unsigned long flag;
> > -
> > -       if (!spin_trylock_irqsave(&stats_flush_lock, flag))
> > +       /*
> > +        * This lock can be acquired from interrupt context,
>
> How? What's the code path?

I believe through the charge/uncharge path we can do
memcg_check_events()->mem_cgroup_threshold()->mem_cgroup_usage()->mem_cgrou=
p_flush_stats(),
right? I am assuming we can charge/uncharge memory in an interrupt
context.

Also the current code always disables interrupts before calling
memcg_check_events(), which made me suspect the percpu variables that
are modified by that call can also be modified in interrupt context.

>
> > but we only acquire
> > +        * using trylock so it should be fine as we cannot cause a dead=
lock.
> > +        */
> > +       if (!spin_trylock(&stats_flush_lock))
> >                 return;
> >
> >         flush_next_time =3D jiffies_64 + 2*FLUSH_TIME;
> >         cgroup_rstat_flush_irqsafe(root_mem_cgroup->css.cgroup);
> >         atomic_set(&stats_flush_threshold, 0);
> > -       spin_unlock_irqrestore(&stats_flush_lock, flag);
> > +       spin_unlock(&stats_flush_lock);
> >  }
> >
> >  void mem_cgroup_flush_stats(void)
> > --
> > 2.40.0.rc1.284.g88254d51c5-goog
> >