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: <20230612093815.133504-1-cerasuolodomenico@gmail.com>
 <20230612093815.133504-2-cerasuolodomenico@gmail.com> <20230618043857.GA2417524@google.com>
 <CAJD7tkbRn2txAt94keKTRbAA7--XF3rXSa8vsh9eGZa2CUWP8w@mail.gmail.com> <20230618052526.GB2417524@google.com>
In-Reply-To: <20230618052526.GB2417524@google.com>
From:   Yosry Ahmed <yosryahmed@google.com>
Date:   Sat, 17 Jun 2023 22:28:48 -0700
Message-ID: <CAJD7tkbcf+MEMJdGeruB7XL=J8tJNgkB7_1UQ2S4VEKjob-NQw@mail.gmail.com>
Subject: Re: [PATCH v3 1/7] mm: zswap: add pool shrinking mechanism
To:     Sergey Senozhatsky <senozhatsky@chromium.org>
Cc:     Domenico Cerasuolo <cerasuolodomenico@gmail.com>,
        vitaly.wool@konsulko.com, minchan@kernel.org, linux-mm@kvack.org,
        ddstreet@ieee.org, sjenning@redhat.com, nphamcs@gmail.com,
        hannes@cmpxchg.org, akpm@linux-foundation.org,
        linux-kernel@vger.kernel.org, kernel-team@meta.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Sat, Jun 17, 2023 at 10:25=E2=80=AFPM Sergey Senozhatsky
<senozhatsky@chromium.org> wrote:
>
> On (23/06/17 21:48), Yosry Ahmed wrote:
> > > On (23/06/12 11:38), Domenico Cerasuolo wrote:
> > > > +static int zswap_reclaim_entry(struct zswap_pool *pool)
> > > > +{
> > > > +     struct zswap_header *zhdr;
> > > > +     struct zswap_entry *entry;
> > > > +     struct zswap_tree *tree;
> > > > +     pgoff_t swpoffset;
> > > > +     int ret;
> > > > +
> > > > +     /* Get an entry off the LRU */
> > > > +     spin_lock(&pool->lru_lock);
> > > > +     if (list_empty(&pool->lru)) {
> > > > +             spin_unlock(&pool->lru_lock);
> > > > +             return -EINVAL;
> > > > +     }
> > > > +     entry =3D list_last_entry(&pool->lru, struct zswap_entry, lru=
);
> > > > +     list_del_init(&entry->lru);
> > >
> > > A quick question: should we zswap_entry_get() here?
> >
> > We need to hold the tree lock for that, and the lock ordering is tree
> > lock -> lru lock. If we try to grab the tree lock here we may
> > deadlock.
>
> We can deadlock doing this?
>
>         lock tree_lock
>         lock lru_lock
>         list_del_init
>         unlock lru_lock
>         entry_get
>         unlock tree_lock
>         writeback

We don't know which tree the zswap entry belongs to until we get from
the LRU -- so we can't hold the tree lock before getting the entry
from the lru (and to get the entry from the LRU we need the lru_lock).