Message-ID: <20220713093708.jaa3og2fablqr723@houat>
Date:   Wed, 13 Jul 2022 11:37:08 +0200
From:   Maxime Ripard <maxime@...no.tech>
To:     Geert Uytterhoeven <geert@...ux-m68k.org>
Cc:     Thomas Zimmermann <tzimmermann@...e.de>,
        Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
        David Airlie <airlied@...ux.ie>,
        Daniel Vetter <daniel@...ll.ch>,
        Hans de Goede <hdegoede@...hat.com>,
        DRI Development <dri-devel@...ts.freedesktop.org>,
        Linux Fbdev development list <linux-fbdev@...r.kernel.org>,
        Linux/m68k <linux-m68k@...r.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 4/5] drm/modes: Add support for driver-specific named
 modes

Hi Geert,

On Mon, Jul 11, 2022 at 02:08:06PM +0200, Geert Uytterhoeven wrote:
> On Mon, Jul 11, 2022 at 2:02 PM Maxime Ripard <maxime@...no.tech> wrote:
> > On Mon, Jul 11, 2022 at 01:59:28PM +0200, Geert Uytterhoeven wrote:
> > > On Mon, Jul 11, 2022 at 1:42 PM Maxime Ripard <maxime@...no.tech> wrote:
> > > > On Mon, Jul 11, 2022 at 01:11:14PM +0200, Thomas Zimmermann wrote:
> > > > > Am 11.07.22 um 11:35 schrieb Maxime Ripard:
> > > > > > On Mon, Jul 11, 2022 at 11:03:38AM +0200, Thomas Zimmermann wrote:
> > > > > > > Am 08.07.22 um 20:21 schrieb Geert Uytterhoeven:
> > > > > > > > The mode parsing code recognizes named modes only if they are explicitly
> > > > > > > > listed in the internal whitelist, which is currently limited to "NTSC"
> > > > > > > > and "PAL".
> > > > > > > >
> > > > > > > > Provide a mechanism for drivers to override this list to support custom
> > > > > > > > mode names.
> > > > > > > >
> > > > > > > > Ideally, this list should just come from the driver's actual list of
> > > > > > > > modes, but connector->probed_modes is not yet populated at the time of
> > > > > > > > parsing.
> > > > > > >
> > > > > > > I've looked for code that uses these names, couldn't find any. How is this
> > > > > > > being used in practice? For example, if I say "PAL" on the command line, is
> > > > > > > there DRM code that fills in the PAL mode parameters?
> > > > > >
> > > > > > We have some code to deal with this in sun4i:
> > > > > > https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/sun4i/sun4i_tv.c#L292
> > > > > >
> > > > > > It's a bit off topic, but for TV standards, I'm still not sure what the
> > > > > > best course of action is. There's several interactions that make this a
> > > > > > bit troublesome:
> > > > > >
> > > > > >    * Some TV standards differ by their mode (ie, PAL vs NSTC), but some
> > > > > >      other differ by parameters that are not part of drm_display_mode
> > > > > >      (NTSC vs NSTC-J where the only difference is the black and blanking
> > > > > >      signal levels for example).
> > > > > >
> > > > > >    * The mode names allow to provide a fairly convenient way to add that
> > > > > >      extra information, but the userspace is free to create its own mode
> > > > > >      and might omit the mode name entirely.
> > > > > >
> > > > > > So in the code above, if the name has been preserved we match by name,
> > > > > > but we fall back to matching by mode if it hasn't been, which in this
> > > > > > case means that we have no way to differentiate between NTSC, NTSC-J,
> > > > > > PAL-M in this case.
> > > > > >
> > > > > > We have some patches downstream for the RaspberryPi that has the TV
> > > > > > standard as a property. There's a few extra logic required for the
> > > > > > userspace (like setting the PAL property, with the NTSC mode) so I'm not
> > > > > > sure it's preferable.
> > > > > >
> > > > > > Or we could do something like a property to try that standard, and
> > > > > > another that reports the one we actually chose.
> > > > > >
> > > > > > > And another question I have is whether this whitelist belongs into the
> > > > > > > driver at all. Standard modes exist independent from drivers or hardware.
> > > > > > > Shouldn't there simply be a global list of all possible mode names? Drivers
> > > > > > > would filter out the unsupported modes anyway.
> > > > > >
> > > > > > We should totally do something like that, yeah
> > > > >
> > > > > That sun code already looks like sometihng the DRM core/helpers should be
> > > > > doing. And if we want to support named modes well, there's a long list of
> > > > > modes in Wikipedia.
> > > > >
> > > > > https://en.wikipedia.org/wiki/Video_Graphics_Array#/media/File:Vector_Video_Standards2.svg
> > > >
> > > > Yeah, and NTSC is missing :)
> > >
> > > And that diagram is about the "digital" variant of PAL.
> > > If you go the analog route, the only fixed parts are vfreq/hfreq,
> > > number of lines, and synchronization. Other parameters like overscan
> > > can vary.  The actual dot clock can vary wildly: while there is an
> > > upper limit due to bandwidth limitations, you can come up with an
> > > almost infinite number of video modes that can be called PAL, which
> > > is one of the reasons why I don't want hardware-specific variants to
> > > end up in a global video mode database.
> >
> > Do you have an example of what that would look like?
> 
> You mean a PAL mode that does not use 768x576?

I meant what the almost infinite number of video modes that can be
called PAL and would have to be defined in drivers

> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/video/fbdev/amifb.c#n834

But that works :)

I don't see what really is troublesome if we go with the mode + property
setup here.

We can deal easily with the interlaced vs non-interlaced variants
already with DRM_MODE_FLAG_INTERLACE, and the ff variants can be dealt
with DRM_MODE_FLAG_DBLCLK.

We still need something to differentiate between, say, PAL-M and NTSC-J
where the differences are between things not exposed by the mode itself
(black and blanking levels differ from NSTC for NTSC-J, and the color
carrier frequency is PAL's for PAL-M)

Am I missing something?

> (TAG_HIRES is replaced by the actual dot clock at runtime, as it
>  depends on the crystal present on the mainboard).

If we have the crystal frequency in the kernel somehow, we could filter
them out from the driver (or fill them in) depending on that frequency.

I still think the mode + property is the way to go, possibly with some
generic component that would take the mode name from the command line
and create that initial state depending on the value for backward
compatibility.

What do you think?

Maxime

Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)

Powered by blists - more mailing lists