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: <20200614123923.99189-1-songmuchun@bytedance.com>
 <20200614123923.99189-4-songmuchun@bytedance.com> <CAAmzW4N1DRBxsMAX2a_jxUW10vOwW1VsGj8ztuvvWT4NB3Ocnw@mail.gmail.com>
 <CAMZfGtUQ+0w=n9YSQjQc1uwJFJrvtfj=rcTnNYnewCPiHmDLEw@mail.gmail.com> <CAAmzW4N=Rq5qc60DA9-ombrzOxaKEnKUFXM6_DJfq=5bCRdO=g@mail.gmail.com>
In-Reply-To: <CAAmzW4N=Rq5qc60DA9-ombrzOxaKEnKUFXM6_DJfq=5bCRdO=g@mail.gmail.com>
From:   Muchun Song <songmuchun@bytedance.com>
Date:   Mon, 15 Jun 2020 15:50:16 +0800
Message-ID: <CAMZfGtVBtRK=rA2CGGnOuTvHyXffVm7vE7m37MAMSbW8fkzjfg@mail.gmail.com>
Subject: Re: [External] Re: [PATCH 3/3] mm/slub: Fix release all resources
 used by a slab cache
To:     Joonsoo Kim <js1304@gmail.com>
Cc:     Christoph Lameter <cl@linux.com>,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linux Memory Management List <linux-mm@kvack.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Jun 15, 2020 at 3:25 PM Joonsoo Kim <js1304@gmail.com> wrote:
>
> 2020=EB=85=84 6=EC=9B=94 15=EC=9D=BC (=EC=9B=94) =EC=98=A4=ED=9B=84 3:41,=
 Muchun Song <songmuchun@bytedance.com>=EB=8B=98=EC=9D=B4 =EC=9E=91=EC=84=
=B1:
> >
> > On Mon, Jun 15, 2020 at 2:23 PM Joonsoo Kim <js1304@gmail.com> wrote:
> > >
> > > 2020=EB=85=84 6=EC=9B=94 14=EC=9D=BC (=EC=9D=BC) =EC=98=A4=ED=9B=84 9=
:39, Muchun Song <songmuchun@bytedance.com>=EB=8B=98=EC=9D=B4 =EC=9E=91=EC=
=84=B1:
> > > >
> > > > The function of __kmem_cache_shutdown() is that release all resourc=
es
> > > > used by the slab cache, while currently it stop release resources w=
hen
> > > > the preceding node is not empty.
> > > >
> > > > Signed-off-by: Muchun Song <songmuchun@bytedance.com>
> > > > ---
> > > >  mm/slub.c | 7 ++++---
> > > >  1 file changed, 4 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/mm/slub.c b/mm/slub.c
> > > > index b73505df3de2..4e477ef0f2b9 100644
> > > > --- a/mm/slub.c
> > > > +++ b/mm/slub.c
> > > > @@ -3839,6 +3839,7 @@ bool __kmem_cache_empty(struct kmem_cache *s)
> > > >   */
> > > >  int __kmem_cache_shutdown(struct kmem_cache *s)
> > > >  {
> > > > +       int ret =3D 0;
> > > >         int node;
> > > >         struct kmem_cache_node *n;
> > > >
> > > > @@ -3846,11 +3847,11 @@ int __kmem_cache_shutdown(struct kmem_cache=
 *s)
> > > >         /* Attempt to free all objects */
> > > >         for_each_kmem_cache_node(s, node, n) {
> > > >                 free_partial(s, n);
> > > > -               if (node_nr_slabs(n))
> > > > -                       return 1;
> > > > +               if (!ret && node_nr_slabs(n))
> > > > +                       ret =3D 1;
> > > >         }
> > >
> > > I don't think that this is an improvement.
> > >
> > > If the shutdown condition isn't met, we don't need to process further=
.
> > > Just 'return 1' looks okay to me.
> > >
> > > And, with this change, sysfs_slab_remove() is called even if the
> > > shutdown is failed.
> > > It's better not to have side effects when failing.
> >
> > If someone calls __kmem_cache_shutdown, he may want to release
> > resources used by the slab cache as much as possible. If we continue,
> > we may release more pages. From this point, is it an improvement?
>
> My opinion is not strong but I still think that it's not useful enough
> to complicate
> the code.
>
> If shutdown is failed, it implies there are some bugs and someone should =
fix it.

Yeah, I agree with you.

> Releasing more resources would mitigate the resource problem but doesn't
> change the situation that someone should fix it soon.
>
> Anyway, I don't object more if you don't agree with my opinion. In that c=
ase,
> please fix not to call sysfs_slab_remove() when ret is 1.
>

Yeah, we should call sysfs_slab_remove only when ret is zero. Thanks very
much.

--=20
Yours,
Muchun