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:   Javier Martinez Canillas <javierm@redhat.com>
To:     Geert Uytterhoeven <geert@linux-m68k.org>,
        Maxime Ripard <mripard@kernel.org>
Cc:     linux-kernel@vger.kernel.org,
        Thomas Zimmermann <tzimmermann@suse.de>,
        Daniel Vetter <daniel.vetter@ffwll.ch>,
        Daniel Vetter <daniel@ffwll.ch>,
        David Airlie <airlied@gmail.com>,
        dri-devel@lists.freedesktop.org
Subject: Re: [PATCH v4] drm/ssd130x: Allocate buffers in the plane's
 .atomic_check callback
In-Reply-To: <CAMuHMdX54sYfv2HJQZxMD_O2XT_nmuoQEE2iTbWkJ9-QvfAWYQ@mail.gmail.com>
References: <20230721070955.1170974-1-javierm@redhat.com>
 <CAMuHMdVxF80mdTaiXA6Y8Gd59Z7pxkoEphB4ohpVcK1q-+Yy=g@mail.gmail.com>
 <n4fmda4zw4vuezvwva35tgma3yqhdvb253q2tfdyauoxbfqshx@s6fsa7de6g5s>
 <CAMuHMdXtOozswqujA1h2spL8J86n65Q6=+z=5Jbb0nSXaBwqzA@mail.gmail.com>
 <874jlqlv5v.fsf@minerva.mail-host-address-is-not-set>
 <CAMuHMdX+J848ckG2JqsuDkRcWzRypw_Kv=0G+Hc329xstu_nqQ@mail.gmail.com>
 <g6bu3b26evk464x4cn77xnzjiuotsq7pfvyakfvwnir5e3ihnk@2jh4dd56fsza>
 <CAMuHMdX54sYfv2HJQZxMD_O2XT_nmuoQEE2iTbWkJ9-QvfAWYQ@mail.gmail.com>
Date:   Wed, 26 Jul 2023 16:22:45 +0200
Message-ID: <87cz0e4uru.fsf@minerva.mail-host-address-is-not-set>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

Geert Uytterhoeven <geert@linux-m68k.org> writes:

Hello Geert,

> Hi Maxime,
>
> On Wed, Jul 26, 2023 at 3:54=E2=80=AFPM Maxime Ripard <mripard@kernel.org=
> wrote:
>> On Wed, Jul 26, 2023 at 02:33:06PM +0200, Geert Uytterhoeven wrote:
>> > > >> Also, Javier pointed me to a discussion you had on IRC about usin=
g devm
>> > > >> allocation here. We can't really do that. KMS devices are only fr=
eed
>> > > >> once the last userspace application closes its fd to the device f=
ile, so
>> > > >> you have an unbounded window during which the driver is still cal=
lable
>> > > >> by userspace (and thus can still trigger an atomic commit) but the
>> > > >> buffer would have been freed for a while.
>> > > >
>> > > > It should still be safe for (at least) the data_array buffer. That
>> > > > buffer is only used to store pixels in hardware format, and immedi=
ately
>> > > > send them to the hardware.  If this can be called that late, it wi=
ll
>> > > > fail horribly, as you can no longer talk to the hardware at that p=
oint
>> > > > (as ssd130x is an i2c driver, it might actually work; but a DRM dr=
iver
>> > > >  that calls devm_platform_ioremap_resource() will crash when writi=
ng
>> > > >  to its MMIO registers)?!?
>> > >
>> > > At the very least the SPI driver will fail since the GPIO that is us=
ed to
>> > > toggle the D/C pin is allocated with devm_gpiod_get_optional(), but =
also
>> > > the regulator, backlight device, etc.
>> > >
>> > > But in any case, as mentioned it is only relevant if the data_array =
buffer
>> > > is allocated at probe time, and from Maxime's explanation is more co=
rrect
>> > > to do it in the .atomic_check handler.
>> >
>> > You need (at least) data_array for clear_screen, too, which is called
>> > from .atomic_disable().
>>
>> I'm not sure I get what your concern is?
>>
>> Even if we entirely disable the plane, the state will not have been
>> destroyed yet so you still have at least access to the data_array from
>> the old state.
>
> Currently, clearing the screen is done from the plane's .atomic_disable()
> callback, so the plane's state should be fine.
>
> But IIUIC, DRM allows the user to enable/disable the display without
> creating any plane first, so clearing should be handled in the CRTC's

But it's needed to be clared in this case? Currently the power on/off
is done in the encoder's .atomic_{en,dis}able() but the actual panel
clear is only done for the plane disable as you mentioned.

> .atomic_disable() callback instead? Then, would we still have access
> to valid plane state?
>

If the clear has to be moved to the CRTC atomic enable/disable, then
the driver should be reworked to not use the data_array and instead
just allocate a zero'ed buffer and pass that as you proposed.

But again that's something that could be done later as another patch.

> Thanks!
>
> Gr{oetje,eeting}s,
>

--=20
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat