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;
From:   "Fabio M. De Francesco" <fmdefrancesco@gmail.com>
To:     Al Viro <viro@zeniv.linux.org.uk>
Cc:     Evgeniy Dushistov <dushistov@mail.ru>,
        Christian Brauner <brauner@kernel.org>,
        linux-kernel@vger.kernel.org, bpf@vger.kernel.org,
        "Venkataramanan, Anirudh" <anirudh.venkataramanan@intel.com>,
        Ira Weiny <ira.weiny@intel.com>
Subject: Re: [RESEND PATCH v4] fs/ufs: Replace kmap() with kmap_local_page()
Date:   Sat, 26 Nov 2022 15:12:16 +0100
Message-ID: <2188828.irdbgypaU6@suse>
In-Reply-To: <Y4FG0O7VWTTng5yh@ZenIV>
References: <20221016163855.8173-1-fmdefrancesco@gmail.com> <Y4E++JERgUMoqfjG@ZenIV> <Y4FG0O7VWTTng5yh@ZenIV>
MIME-Version: 1.0
Content-Transfer-Encoding: quoted-printable
Content-Type: text/plain; charset="iso-8859-1"
Precedence: bulk

On venerd=EC 25 novembre 2022 23:50:56 CET Al Viro wrote:
> On Fri, Nov 25, 2022 at 10:17:28PM +0000, Al Viro wrote:
> > The bottom line:
> > 	* teach your ufs_put_page() to accept any address within the page
> > 	* flip the ways you return page and address in ufs_get_page()
> > 	* use offset_in_page(addr) instead of these addr -=20
page_address(page)
> >=20
> > and you'll get a much smaller patch, with a lot less noise in it.
> > What's more, offset_in_page() part can be carved out into a separate
> > commit - it's valid on its own, and it makes both halves easier to
> > follow.
> >=20
> > AFAICS, similar observations apply in your sysvfs patch; the point about
> > calling conventions for ufs_get_page() definitely applies there, and
> > stronger than for ufs - those casts are eye-watering...
>=20
> As the matter of fact, I'd probably split it into 3 steps:
> 	1) switch the calling conventions of ufs_get_page() - callers follow
> it with something like kaddr =3D page_address(page); and that change makes
> sense on its own.  Isolated and easy to follow.
> 	2) offset_in_page() changes.  Again, separate, stands on its own and
> is easy to follow.
> 	3) kmap_local() switch itself - pass address to ufs_put_page(), etc.
> Considerably smaller and less cluttered than your current variant.

Al, thanks for taking the time to answer in such detail although I can beli=
eve=20
that, to spot the mistakes in my patch and then write your email, it probab=
ly=20
took you 10 minutes or so :-)

Instead I had to read two or three times to make sense of it all. I will do=
 my=20
best to first rewrite these kmap_local_page() conversions in fs/ufs by=20
separating the work into three patches, as you recommended.

I'm pretty sure that the next attempt won't be applicable yet, because I mi=
ght=20
not fully understand your suggestions or I might not be able to implement t=
hem=20
correctly. I'm afraid I addressed something that, as it stands, is a little=
=20
beyond my knowledge and experience, so I hope you'll want to chime in and=20
comment on the next version as well.

When the changes to fs/ufs will be done I will also rewrite the patch for f=
s/
sysv which (as you pointed out) needs to be worked out according to the sam=
e=20
requirements.

I am sincerely grateful to you also because it seemed to me that I would ne=
ver=20
get any feedback regarding these two old patches.

Regards,

=46abio