[Intel-gfx] i915 DVI resolution regression (3.13.7+)

[Ville Syrjälä]

On Tue, Apr 15, 2014 at 02:27:41PM +0800, Daniel J Blueman wrote:
> On 9 April 2014 15:08, Jani Nikula <jani.nikula at linux.intel.com> wrote:
> > On Wed, 09 Apr 2014, Dave Airlie <airlied at gmail.com> wrote:
> >> On Wed, Apr 9, 2014 at 4:07 PM, Daniel J Blueman <daniel at quora.org> wrote:
> >>> On 9 April 2014 11:41, Dave Airlie <airlied at gmail.com> wrote:
> >>>> On Tue, Apr 8, 2014 at 5:32 PM, Daniel J Blueman <daniel at quora.org> wrote:
> >>>>> On 8 April 2014 15:14, Jani Nikula <jani.nikula at linux.intel.com> wrote:
> >>>>>> On Tue, 08 Apr 2014, Daniel J Blueman <daniel at quora.org> wrote:
> >>>>>>> Ville et al,
> >>>>>>>
> >>>>>>> It looks like commit e3ea8fa6beaf55fee64bf816f3b8a80ad733b2c2 (or
> >>>>>>> another commit in 3.13.7) broke modes which require DVI-D dual-link,
> >>>>>>> eg 2560x1440 with my panel.
> >>>>>>>
> >>>>>>> I don't see these modelines in 3.13.7 or later (eg 3.14):
> >>>>>>>
> >>>>>>> [     5.582] (II) intel(0): Modeline "2560x1440"x60.0  312.25  2560
> >>>>>>> 2752 3024 3488  1440 1443 1448 1493 -hsync +vsync (89.5 kHz eP)
> >>>>>>> [     5.582] (II) intel(0): Modeline "2560x1440"x60.0  312.25  2560
> >>>>>>> 2752 3024 3488  1440 1443 1448 1493 -hsync +vsync (89.5 kHz eP)
> >>>>>>> [     5.582] (II) intel(0): Modeline "1920x1200"x59.9  193.25  1920
> >>>>>>> 2056 2256 2592  1200 1203 1209 1245 -hsync +vsync (74.6 kHz e)
> >>>>>>>
> >>>>>>> My monitor is a Dell U2713HM; mobo uses an H87 chipset with i5-4670.
> >>>>>>
> >>>>>> By allowing those modes we regressed setups which were not capable of
> >>>>>> displaying them. So you've got an HDMI->DVI converter?
> >>>>>>
> >>>>>> https://bugzilla.kernel.org/show_bug.cgi?id=72961
> >>>>>
> >>>>> I am using a dual-link DVI-D to DVI-D cable to this monitor, since I
> >>>>> previously couldn't get 2560x1440 via HDMI.
> >>>>
> >>>> Intel hw has dual-link DVI-D? I'm not sure I've ever seen that, is
> >>>> this SDVO device or plain DVI-D?
> >>>
> >>> It's the DVI-D connector on: https://www.asus.com/Motherboards/H87IPLUS/
> >>
> >> The link which says "
> >>
> >> Integrated Graphics Processor
> >> - Supports HDMI with max. resolution 4096 x 2304 @ 24 Hz
> >> - Supports DVI with max. resolution 1920 x 1200 @ 60 Hz
> >> - Supports RGB with max. resolution 1920 x 1200 @ 60 Hz
> >>
> >> I'm even wondering electrically how a HDMI->dual-link DVI adapter works.
> >
> > The current assumption is that in the working case, they are really
> > operated in single-link, with a frequency beyond the single-link DVI
> > cable spec.
> 
> I checked with a single link DVI-D cable, and see the same behaviour
> as with a dual link cable. This would have had to be driven out of
> spec to achieve the resolution and refresh rate of my panel. That
> said, the spec is old and the shielding on the dual-link cable
> supplied with the panel is way over-specced.
> 
> If Intel GPUs (such as my Haswell board here) don't support dual-link
> DVI, then this patch will prevent using panels >1920x1200 unless they
> advertise low-refresh rate modes. It looks like the panel
> manufacturers assume dual-link will be available.

I don't think I've ever seen a display that would _require_ dual-link.
You should at least get some kind of picture with single-link.

> 
> The lesser of the evils may be to allow reasonable out-of-spec
> dotclocks, unless I'm missing something?

No, we can't do that. If the display won't accept the out of spec
signal you get no picture. If the display will accept higher
frequencies it needs to tell us that by including the HDMI VSDB
in the EDID. There's no other way for us to know.

What we do now is allow the user to specify a mode manually that's out
of spec, but we won't include such modes in the list automatically. That
should guarantee that everyone gets some kind of picture on the screen
off the bat, while still allowing the user to force any mode he wishes.

-- 
Ville Syrjälä
Intel OTC