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

[Jani Nikula]

On Fri, 10 Nov 2017, Mika Kahola <mika.kahola at intel.com> wrote:
> 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.

That seems like the way to go. Display disable on the command line
should be no different from num_pipes == 0.

BR,
Jani.

-- 
Jani Nikula, Intel Open Source Technology Center