[Intel-gfx] [PATCH 1/4] drm/i915: Add audio hotplug info struct

[Takashi Iwai]

On Wed, 22 Jul 2015 19:52:23 +0200,
David Henningsson wrote:
> 
> 
> 
> On 2015-07-22 16:13, Vinod Koul wrote:
> > On Wed, Jul 22, 2015 at 10:55:55AM +0200, Takashi Iwai wrote:
> >> On Wed, 22 Jul 2015 10:50:03 +0200,
> >> David Henningsson wrote:
> >>>
> >>>>>    struct i915_audio_component {
> >>>>>    	struct device *dev;
> >>>>> +	struct hdac_bus *hdac_bus;
> >>>>
> >>>> If we want to be more generic, using a struct device would be better,
> >>>> e.g.
> >>>> 	struct device *audio_dev;
> >>>
> >>> Does this work? If we want to have the hdac_bus.dev ptr instead of a
> >>> hdac_bus ptr, there does not seem to be an obvious way to go from the
> >>> audio_dev back to the hdac_bus struct (as snd_hdac_bus_init takes an
> >>> arbitrary dev pointer).
> >>
> >> Hrm, right, currently it's not straightforward.  Scratch the idea,
> >> then.
> >
> > That depends on the device we register this with. Actually this makes more
> > sense to me :)
> >
> > If we register with struct device *audio_dev, which in this case would be
> > the codec device we create while probing the bus. This way you are linking i915
> > ops to the codec device. Ofcourse hdac_device has bus pointer but you can
> > invoke device callback without even searching for the device :)
> 
> It would require some extra setup, so I skipped it (at least for now).
> 
> I e, in order to detect codecs, we need to call hdac_i915 functions to 
> turn the power well on. And in order to connect the codec to the 
> i915_audio_component, we need to have detected a codec.
> 
> Which, now that I think of it, actually gives an interesting potential 
> race condition, in case the following happens:
> 
> 1) Monitor is plugged in at boot time
> 2) i915 initializes
> 3) hda starts initializing and sets up the audio component
> 4) i915 finishes initialization and as part of that, calls the hotplug 
> notify to let hda know that the monitor is plugged in. However, at this 
> point, hda has not finished initialization yet, so there are no codecs 
> that listen for this information.
> 
> Anyhow, this is not a problem really yet, but it might be if we ever 
> decide to not write the ELD to the hardware.

For avoid such a problem, maybe we need two ops, one for notification
and one for getting the assigned data.  At the initialization time,
the audio driver queries the assigned status and data.  When a hotplug
happens, it's just notified.  That is, it simply replaces the current
unsol event and the ELD data read via two ops.


Takashi