[Intel-gfx] [PATCH] drm/i915: Return early when pipes are not available

[Mika Kahola]

On Thu, 2017-11-09 at 15:15 +0200, Ville Syrjälä wrote:
> On Thu, Nov 09, 2017 at 01:11:05PM +0200, Mika Kahola wrote:
> > 
> > On Thu, 2017-11-09 at 11:01 +0000, Chris Wilson wrote:
> > > 
> > > Quoting Mika Kahola (2017-11-09 10:49:52)
> > > > 
> > > > 
> > > > At least in Coffee Lake it happens that we start initiliazing
> > > > audio
> > > > when
> > > > no display is connected. This was discovered by CI when running
> > > > IGT
> > > > test
> > > > case
> > > > 
> > > > drv_module_reload --r basic-no-display
> > > > 
> > > > The issue here is that the 'intel_device_info_runtime_init()'
> > > > sets
> > > > num_pipes to 0 but before this happens the audio part has
> > > > already
> > > > started
> > > > to initialize itself. Later on the num_pipes is updated to 0 in
> > > > intel_device_info_runtime_init() and we hit that when audio
> > > > part is
> > > > digging
> > > > out ELD. This causes a warning in dmesg. To fix this issue,
> > > > let's
> > > > check the
> > > > number of available pipes when trying to read out ELD.
> > > dev_info_runtime_init() is too late. It depends on mmio being
> > > enabled
> > > to
> > > probe the HW and nothing else; so move it to
> > > i915_driver_init_mmio()?
> > Ok. I could try that. I was also thinking that is there a way to
> > postpone audio initialization?
> We probably shouldn't be registering the audio thing until we've
> actually set up the outputs.
I tried couple of versions. One with Chris's idea to move
intel_device_info_runtime_init() to i915_driver_init_mmio() didn't turn
out to be a success. 

https://patchwork.freedesktop.org/series/33514/

I also tried to set num_pipes to 0 in case of display disable already
in i915_driver_init_early(). That approach didn't turn out to be a
success either.

https://patchwork.freedesktop.org/series/33522/

I'll give it a go for a patch that doesn't register audio in case of
disabled display.

> 
-- 
Mika Kahola - Intel OTC