Received-SPF: pass (google.com: domain of linux-kernel+bounces-21661-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Precedence: bulk
MIME-Version: 1.0
References: <20240102175338.62012-1-ryncsn@gmail.com> <20240102175338.62012-7-ryncsn@gmail.com>
 <87a5pg9qno.fsf@yhuang6-desk2.ccr.corp.intel.com>
In-Reply-To: <87a5pg9qno.fsf@yhuang6-desk2.ccr.corp.intel.com>
From: Kairui Song <ryncsn@gmail.com>
Date: Wed, 10 Jan 2024 10:53:42 +0800
Message-ID: <CAMgjq7Ce6sCSTRd==N3ihrAcvVd2ggszdZuTwxDbFYcBWzcE_g@mail.gmail.com>
Subject: Re: [PATCH v2 6/9] mm/swap: handle swapcache lookup in swapin_entry
To: "Huang, Ying" <ying.huang@intel.com>
Cc: linux-mm@kvack.org, Andrew Morton <akpm@linux-foundation.org>, 
	Chris Li <chrisl@kernel.org>, Hugh Dickins <hughd@google.com>, 
	Johannes Weiner <hannes@cmpxchg.org>, Matthew Wilcox <willy@infradead.org>, Michal Hocko <mhocko@suse.com>, 
	Yosry Ahmed <yosryahmed@google.com>, David Hildenbrand <david@redhat.com>, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

Huang, Ying <ying.huang@intel.com> =E4=BA=8E2024=E5=B9=B41=E6=9C=888=E6=97=
=A5=E5=91=A8=E4=B8=80 16:28=E5=86=99=E9=81=93=EF=BC=9A
>
> Kairui Song <ryncsn@gmail.com> writes:
>
> > From: Kairui Song <kasong@tencent.com>
> >
> > Since all callers of swapin_entry need to check the swap cache first, w=
e
> > can merge this common routine into swapin_entry, so it can be shared an=
d
> > optimized later.
> >
> > Also introduce a enum to better represent possible swap cache usage, an=
d
> > add some comments about it, make the usage of swap cache easier to
> > understand.
>
> I don't find any benefit to do this.  The code line number isn't
> reduced.  The concept of swap cache isn't hided either.

Hi Ying

Thanks for the comments.

Maybe I should squash this with the following commit? The following
commit want to do cache lookup in swapin_entry to avoid a redundant
shadow lookup, it can help to improve the performance by about 4% for
swapin path. So it need to return a enum to represent cache status.

Further more, note the comments added here:

+/*
+ * Caller of swapin_entry may need to know the cache lookup result:
+ *
+ * SWAP_CACHE_HIT: cache hit, cached folio is retured.
+ * SWAP_CACHE_MISS: cache miss, folio is allocated, read from swap device
+ *                  and adde to swap cache, but still may return a cached
+ *                  folio if raced (check __read_swap_cache_async).
+ * SWAP_CACHE_BYPASS: cache miss, folio is new allocated and read
+ *                    from swap device bypassing the cache.
+ */

SWAP_CACHE_MISS might be inaccurate, this is not an issue introduced
by this commit, but better exposed. May worth a fix later. So far I
can see two benefits fixing it:
- More accurate maj/min page fault count.
- Note the PageHWPoison check in do_swap_page, it ignored the race
case, if a page getting poisoned is raced with swapcache then it may
not work as expected.

These are all minor issue indeed, some other optimization might also
be doable, but at least a comment might be helpful.

> --
> Best Regards,
> Huang, Ying
>