[Intel-gfx] [PATCH] drm/i915/sdvo: Fallback to current output timings for LVDS fixed mode

[Ville Syrjälä]

On Tue, Oct 25, 2022 at 08:47:32PM +0300, Jani Nikula wrote:
> On Tue, 25 Oct 2022, Ville Syrjala <ville.syrjala at linux.intel.com> wrote:
> > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> >
> > If we can't dig out a fixed mode for LVDS from the VBT or EDID
> > let's fall back to using the current output timings. This should
> > work as long as the BIOS has (somehow) enabled the output.
> >
> > In this case we are dealing with the some kind of BLB based POS
> > machine (Toshiba SurePOS 500) where neither the OpRegion mailbox
> > nor the vbios ROM contain a valid VBT. And no EDID anywhere we
> > could find either.
> >
> > Cc: <stable at vger.kernel.org> # v5.19+
> > Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/7301
> > Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> 
> Reviewed-by: Jani Nikula <jani.nikula at intel.com>
> 
> But they're saying it's a regression between 4.19 and 5.10...

Yeah. I can't actually figure out how it could have worked even
with 4.19.

Hmm. Actually now that I look at some of the hints in the logs it
does look like it maybe did find an EDID after all. What confused
me was that all the modes very much like the _noedid stuff.

Ah, it looks like we fail to fully initialize the DDC stuff
before setting up the outputs, which I guess explains why the
EDID read fails there. Previously there was this funky feedback
loop in that .get_modes() actually filled in the fixed_mode,
so until you called that (after everything else was fully set
up) you didn't have a fixed mode.

And while looking at this stuff more I can see that the whole
multi output support is still very much snafu :/

I'll see if I can make a fairly minimal fix for now...

-- 
Ville Syrjälä
Intel