Received-SPF: pass (google.com: domain of linux-kernel+bounces-31589-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Precedence: bulk
MIME-Version: 1.0
References: <20240117-zswap-xarray-v1-0-6daa86c08fae@kernel.org>
 <20240117-zswap-xarray-v1-2-6daa86c08fae@kernel.org> <CAJD7tkZF102x_8LKAX+sxAttgYD_LNT3cRqeOr7_euwPfNdCFA@mail.gmail.com>
 <CAF8kJuMwF=s-28OPFbCfJf-f3jsfYmyiP6pSBjj8ZgkGmbT9ZA@mail.gmail.com>
 <CAJD7tkbdT-Bdi4YkOS_Px73dkhthq2qn6Yvy7CobzTBji0WMog@mail.gmail.com> <CAF8kJuOOwyzGGz21PKsGBdJBU3B+gk1+o78MwGqG5FDNZK9LSA@mail.gmail.com>
In-Reply-To: <CAF8kJuOOwyzGGz21PKsGBdJBU3B+gk1+o78MwGqG5FDNZK9LSA@mail.gmail.com>
From: Yosry Ahmed <yosryahmed@google.com>
Date: Fri, 19 Jan 2024 13:44:25 -0800
Message-ID: <CAJD7tkbKvJtLNDZn7G=yJKH+tbdmhF7sbkWru-V7vrAVwGb2sg@mail.gmail.com>
Subject: Re: [PATCH 2/2] mm: zswap.c: remove RB tree
To: Chris Li <chrisl@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>, linux-kernel@vger.kernel.org, 
	linux-mm@kvack.org, =?UTF-8?B?V2VpIFh177+8?= <weixugc@google.com>, 
	Yu Zhao <yuzhao@google.com>, Greg Thelen <gthelen@google.com>, 
	Chun-Tse Shao <ctshao@google.com>, =?UTF-8?Q?Suren_Baghdasaryan=EF=BF=BC?= <surenb@google.com>, 
	Brain Geffon <bgeffon@google.com>, Minchan Kim <minchan@kernel.org>, Michal Hocko <mhocko@suse.com>, 
	Mel Gorman <mgorman@techsingularity.net>, Huang Ying <ying.huang@intel.com>, 
	Nhat Pham <nphamcs@gmail.com>, Johannes Weiner <hannes@cmpxchg.org>, Kairui Song <kasong@tencent.com>, 
	Zhongkun He <hezhongkun.hzk@bytedance.com>, Kemeng Shi <shikemeng@huaweicloud.com>, 
	Barry Song <v-songbaohua@oppo.com>, "Matthew Wilcox (Oracle)" <willy@infradead.org>, 
	"Liam R. Howlett" <Liam.Howlett@oracle.com>, Joel Fernandes <joel@joelfernandes.org>, 
	Chengming Zhou <zhouchengming@bytedance.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Fri, Jan 19, 2024 at 1:32=E2=80=AFPM Chris Li <chrisl@kernel.org> wrote:
>
> On Fri, Jan 19, 2024 at 11:37=E2=80=AFAM Yosry Ahmed <yosryahmed@google.c=
om> wrote:
> > > > I think using the xas_* APIs can be avoided here. The only reason w=
e
> > > > need it is that we want to check if there's an existing entry first=
,
> > > > and return -EEXIST. However, in that case, the caller will replace =
it
> > > > anyway (and do some operations on the dupentry):
> > >
> > > We might be able to for the insert case if we don't mind changing the
> > > code behavior a bit. My original intent is to keep close to the
> > > original zswap code and not stir the pot too much for the xarray
> > > replacement. We can always make more adjustment once the RB tree is
> > > gone.
> >
> > I don't see how this changes code behavior though. The current code in
> > zswap_store() will do the following:
>
> I am referring to the log and update counter happening after the zswap
> mapping was updated. Maybe nobody actually cares about that behavior
> difference. In my mind, there is a difference.

I don't think it matters tbh, certainly not worth the more complicated
implementation.

> > > > >  static bool zswap_erase(struct zswap_tree *tree, struct zswap_en=
try *entry)
> > > > >  {
> > > > > +       struct zswap_entry *e;
> > > > >         pgoff_t offset =3D swp_offset(entry->swpentry);
> > > > > -       if (!RB_EMPTY_NODE(&entry->rbnode)) {
> > > > > -               struct zswap_entry *old;
> > > > > -               old =3D xa_erase(&tree->xarray, offset);
> > > > > -               BUG_ON(old !=3D entry);
> > > > > -               rb_erase(&entry->rbnode, &tree->rbroot);
> > > > > -               RB_CLEAR_NODE(&entry->rbnode);
> > > > > -               return true;
> > > > > -       }
> > > > > -       return false;
> > > > > +       XA_STATE(xas, &tree->xarray, offset);
> > > > > +
> > > > > +       do {
> > > > > +               xas_lock_irq(&xas);
> > > > > +               do {
> > > > > +                       e =3D xas_load(&xas);
> > > > > +               } while (xas_retry(&xas, e));
> > > > > +               if (xas_valid(&xas) && e !=3D entry) {
> > > > > +                       xas_unlock_irq(&xas);
> > > > > +                       return false;
> > > > > +               }
> > > > > +               xas_store(&xas, NULL);
> > > > > +               xas_unlock_irq(&xas);
> > > > > +       } while (xas_nomem(&xas, GFP_KERNEL));
> > > > > +       return !xas_error(&xas);
> > > > >  }
> > > >
> > > > Same here, I think we just want:
> > > >
> > > > return !!xa_erase(..);
> > >
> > > For the erase case it is tricky.
> > > The current zswap code does not erase an entry if the tree entry at
> > > the same offset has been changed. It should be fine if the new entry
> > > is NULL. Basically some race to remove the entry already. However, if
> > > the entry is not NULL, then force resetting it to NULL will change
> > > behavior compared to the current.
> >
> > I see, very good point. I think we can use xa_cmpxchg() and pass in NUL=
L?
> >
> That is certainly possible. Thanks for bringing it up.
> Let me try to combine the tree->lock with xarray lock first. If
> xa_cmpxchg() can simplify the result there, I will use it.

SGTM.

> > Handling large folios in zswap is a much larger topic that involves a
> > lot more than this xa_* vs. xas_* apis dispute. Let's not worry about
> > this for now.
>
> Ack. One more reason to use the XAS interface is that zswap currently
> does multiple lookups on typical zswap_load(). It finds entries by
> offset, for the entry (lookup one). Then after folio install to swap
> cache, it deletes the entry, it will performan another lookup to
> delete the entry (look up two). Using XAS might be able to cache the
> node location for the second lookup to avoid the full node walk. That
> is not in my current patch and can be a later improvement patch as
> well.

One more straightforward optimization we can do with the xas_* API is
to cache the lookup done in zswap_load() and reuse it when doing
invalidations for exclusive loads.

For the initial implementation, let's keep it simple and try to use
the xa_* APIs where possible.