Received-SPF: pass (google.com: domain of linux-kernel+bounces-210408-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240608155316.451600-1-flintglass@gmail.com> <20240608155316.451600-3-flintglass@gmail.com>
 <CAJD7tkZAkzUfbXY3C0QOGqCyjQZeiuGzkZac4hmogOoh=yoZsw@mail.gmail.com>
In-Reply-To: <CAJD7tkZAkzUfbXY3C0QOGqCyjQZeiuGzkZac4hmogOoh=yoZsw@mail.gmail.com>
From: Nhat Pham <nphamcs@gmail.com>
Date: Tue, 11 Jun 2024 11:15:05 -0700
Message-ID: <CAKEwX=O4VEfO8f+GFDW_LL_=Ne6xgFEby2E+dxamhXKbFj0VAg@mail.gmail.com>
Subject: Re: [PATCH v1 2/3] mm: zswap: fix global shrinker error handling logic
To: Yosry Ahmed <yosryahmed@google.com>
Cc: Takero Funaki <flintglass@gmail.com>, Johannes Weiner <hannes@cmpxchg.org>, 
	Chengming Zhou <chengming.zhou@linux.dev>, Jonathan Corbet <corbet@lwn.net>, 
	Andrew Morton <akpm@linux-foundation.org>, 
	Domenico Cerasuolo <cerasuolodomenico@gmail.com>, linux-mm@kvack.org, linux-doc@vger.kernel.org, 
	linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Jun 10, 2024 at 1:28=E2=80=AFPM Yosry Ahmed <yosryahmed@google.com>=
 wrote:
>
> On Sat, Jun 8, 2024 at 8:53=E2=80=AFAM Takero Funaki <flintglass@gmail.co=
m> wrote:
> >
> > This patch fixes zswap global shrinker that did not shrink zpool as
> > expected.
> >
> > The issue it addresses is that `shrink_worker()` did not distinguish
> > between unexpected errors and expected error codes that should be
> > skipped, such as when there is no stored page in a memcg. This led to
> > the shrinking process being aborted on the expected error codes.
> >
> > The shrinker should ignore these cases and skip to the next memcg.
> > However,  skipping all memcgs presents another problem. To address this=
,
> > this patch tracks progress while walking the memcg tree and checks for
> > progress once the tree walk is completed.
> >
> > To handle the empty memcg case, the helper function `shrink_memcg()` is
> > modified to check if the memcg is empty and then return -ENOENT.
> >
> > Fixes: a65b0e7607cc ("zswap: make shrinking memcg-aware")
> > Signed-off-by: Takero Funaki <flintglass@gmail.com>
> > ---
> >  mm/zswap.c | 31 ++++++++++++++++++++++++++-----
> >  1 file changed, 26 insertions(+), 5 deletions(-)
> >
> > diff --git a/mm/zswap.c b/mm/zswap.c
> > index d720a42069b6..1a90f434f247 100644
> > --- a/mm/zswap.c
> > +++ b/mm/zswap.c
> > @@ -1393,7 +1393,7 @@ static struct shrinker *zswap_alloc_shrinker(void=
)
> >
> >  static int shrink_memcg(struct mem_cgroup *memcg)
> >  {
> > -       int nid, shrunk =3D 0;
> > +       int nid, shrunk =3D 0, stored =3D 0;
> >
> >         if (!mem_cgroup_zswap_writeback_enabled(memcg))
> >                 return -EINVAL;
> > @@ -1408,9 +1408,16 @@ static int shrink_memcg(struct mem_cgroup *memcg=
)
> >         for_each_node_state(nid, N_NORMAL_MEMORY) {
> >                 unsigned long nr_to_walk =3D 1;
> >
> > +               if (!list_lru_count_one(&zswap_list_lru, nid, memcg))
> > +                       continue;
> > +               ++stored;
> >                 shrunk +=3D list_lru_walk_one(&zswap_list_lru, nid, mem=
cg,
> >                                             &shrink_memcg_cb, NULL, &nr=
_to_walk);
> >         }
> > +
> > +       if (!stored)
> > +               return -ENOENT;
> > +
>
> Can't we just check nr_to_walk here and return -ENOENT if it remains as 1=
?
>
> Something like:
>
> if (nr_to_walk)
>     return -ENOENT;
> if (!shrunk)
>     return -EAGAIN;
> return 0;
>
> >         return shrunk ? 0 : -EAGAIN;
> >  }
> >
> > @@ -1418,12 +1425,18 @@ static void shrink_worker(struct work_struct *w=
)
> >  {
> >         struct mem_cgroup *memcg =3D NULL;
> >         struct mem_cgroup *next_memcg;
> > -       int ret, failures =3D 0;
> > +       int ret, failures =3D 0, progress;
> >         unsigned long thr;
> >
> >         /* Reclaim down to the accept threshold */
> >         thr =3D zswap_accept_thr_pages();
> >
> > +       /*
> > +        * We might start from the last memcg.
> > +        * That is not a failure.
> > +        */
> > +       progress =3D 1;
> > +
> >         /* global reclaim will select cgroup in a round-robin fashion.
> >          *
> >          * We save iteration cursor memcg into zswap_next_shrink,
> > @@ -1461,9 +1474,12 @@ static void shrink_worker(struct work_struct *w)
> >                  */
> >                 if (!memcg) {
> >                         spin_unlock(&zswap_shrink_lock);
> > -                       if (++failures =3D=3D MAX_RECLAIM_RETRIES)
> > +
> > +                       /* tree walk completed but no progress */
> > +                       if (!progress && ++failures =3D=3D MAX_RECLAIM_=
RETRIES)
> >                                 break;
>
> It seems like we may keep iterating the entire hierarchy a lot of
> times as long as we are making any type of progress. This doesn't seem
> right.
>

Ah actually, reading your comment closer then yeah the
shrink-until-accept is the intended-ish behavior (from Domenico's
older patches to re-work zswap reclaim). The one-page-per-attempt is
also "intended" behavior, but not any of our's intentions - just
something that remains from the past implementation as we rework this
codebase one change at a time :)