[Intel-gfx] [REGRESSION] i915: No HDMI output with 4.4

[Daniel Vetter]

On Mon, Feb 15, 2016 at 04:42:35PM +0200, Ville Syrjälä wrote:
> On Mon, Feb 15, 2016 at 10:55:33AM +0200, Oleksandr Natalenko wrote:
> > 13.02.2016 01:23, Ville Syrjälä wrote:
> > > - Do you have another monitor to try?
> > > - Do you have another cable to try?
> > 
> > More on this.
> > 
> > Computer DVI —— old DVI-HDMI cable —— old monitor HDMI == not working
> > Computer DVI —— another DVI-HDMI cable —— old monitor HDMI == not 
> > working
> > Computer DVI —— DVI-DVI cable —— another monitor DVI == works
> > 
> > So
> > 
> > > Shouldn't really matter. HDMI and DVI are identical at this level.
> > 
> > Not quite, as far as I can see.
> 
> Well, it seems this particular monitor is just somehow funky. It's a bit
> strange that the hpd interrupt still works. It would seem to indicate
> that there's two separate voltage thresholds for detection, one for the
> hpd generation, and another for the live status. I did see something
> similar on another platforms (CHV) where it had two different hpd
> detection registers, and those produced different results when the
> pullup on the hpd pin was misconfigured.
> 
> Anyway, I'm out of ideas now :( Anyone else got something up their
> sleeve?
> 
> I'm starting to think this is going to be our only option:
> -       if (intel_hdmi_set_edid(connector, live_status)) {
> +       if (intel_hdmi_set_edid(connector, true)) {
> 
> It would more or less turn the live status check into a fixed
> msleep(80) for the disconnect case. For the connect case it would
> still break out sooner when live status works.
> 
> The downside is that if the cable is yanked slowly, we'll still succeed
> in the ddc communication during unplug and thus fail to notice that the
> monitor was actually disconnected. But the delay should make that less
> likely.

The other downside is that it'll make us non-compliant, which was the
point of this entire ordeal: HDMI spec forbids us from starting any i2c
transactions when the hpd isn't signalling a present screen.

So maybe we need to buy one of these broken screens.

Oleksandr, what exact model are you using? And any chance that you could
test this on some other machine with intel gfx and latest kernel, just to
make sure this really is some issue with the sink and not with the machine
itself? And I guess you've tested with some other hdmi sink, and that
works?

Thanks, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch