Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210212170159.32153-1-songmuchun@bytedance.com>
 <20210212170159.32153-3-songmuchun@bytedance.com> <YCpBUm2N4Bqm5PM5@dhcp22.suse.cz>
 <CAMZfGtVkh-DeYLLo8Nn7kHMCq055RSvL03eON1iqmhydYiQ-iQ@mail.gmail.com> <YCpMo6gLFyqANsgd@dhcp22.suse.cz>
In-Reply-To: <YCpMo6gLFyqANsgd@dhcp22.suse.cz>
From:   Muchun Song <songmuchun@bytedance.com>
Date:   Mon, 15 Feb 2021 19:34:13 +0800
Message-ID: <CAMZfGtUD3H6e+s_n+2q9aE3ABKJaooRj_vyELBaTTVUssSK-NA@mail.gmail.com>
Subject: Re: [External] Re: [PATCH 3/4] mm: memcontrol: bail out early when id
 is zero
To:     Michal Hocko <mhocko@suse.com>
Cc:     Johannes Weiner <hannes@cmpxchg.org>,
        Vladimir Davydov <vdavydov.dev@gmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Cgroups <cgroups@vger.kernel.org>,
        Linux Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Mon, Feb 15, 2021 at 6:27 PM Michal Hocko <mhocko@suse.com> wrote:
>
> On Mon 15-02-21 18:09:44, Muchun Song wrote:
> > On Mon, Feb 15, 2021 at 5:39 PM Michal Hocko <mhocko@suse.com> wrote:
> > >
> > > On Sat 13-02-21 01:01:58, Muchun Song wrote:
> > > > The memcg ID cannot be zero, but we can pass zero to mem_cgroup_from_id,
> > > > so idr_find() is pointless and wastes CPU cycles.
> > >
> > > Is this possible at all to happen? If not why should we add a test for
> > > _all_ invocations?
> >
> > Yeah, this indeed can happen. If we allocate a new swap cache page
> > and charge it via mem_cgroup_charge, then the page will uncharge
> > the swap counter via mem_cgroup_uncharge_swap. When the swap
> > entry is indeed freed, we will call mem_cgroup_uncharge_swap again,
> > In this routine, we can pass zero to mem_cgroup_from_id. Right?
>
> If the above claim is correct, which I would need to double check then
> it should have been part of the changelog! Please think of your poor
> reviewers and the time they have to invest into the review.

The easy way may be adding a printk to mem_cgroup_from_id when
the parameter is zero.

>
> I would also like to see your waste of CPU cycles argument to be backed
> by something. Are we talking about cycles due to an additional function

Yeah, when the parameter is already zero, idr_find() must return zero.
So I think that the additional function call is unnecessary. I have added
a printk to mem_cgroup_from_id, I found the parameter can be zero
several times.

> call? Is this really something we should even care about?

Maybe not. Just my thoughts.

Thanks.

>
> > > >
> > > > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > > > ---
> > > >  mm/memcontrol.c | 3 +++
> > > >  1 file changed, 3 insertions(+)
> > > >
> > > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > > > index a3f26522765a..68ed4b297c13 100644
> > > > --- a/mm/memcontrol.c
> > > > +++ b/mm/memcontrol.c
> > > > @@ -5173,6 +5173,9 @@ static inline void mem_cgroup_id_put(struct mem_cgroup *memcg)
> > > >  struct mem_cgroup *mem_cgroup_from_id(unsigned short id)
> > > >  {
> > > >       WARN_ON_ONCE(!rcu_read_lock_held());
> > > > +     /* The memcg ID cannot be zero. */
> > > > +     if (id == 0)
> > > > +             return NULL;
> > > >       return idr_find(&mem_cgroup_idr, id);
> > > >  }
> > > >
> > > > --
> > > > 2.11.0
> > >
> > > --
> > > Michal Hocko
> > > SUSE Labs
>
> --
> Michal Hocko
> SUSE Labs