[alsa-devel] [RFC PATCH] ASoC: codec: hdac_hdmi: no checking monitor in hw_params

[Pierre-Louis Bossart]

On 5/6/19 10:41 AM, Takashi Iwai wrote:
> On Mon, 06 May 2019 17:37:32 +0200,
> Pierre-Louis Bossart wrote:
>>
>> On 5/6/19 1:59 AM, libin.yang at intel.com wrote:
>>> From: Libin Yang <libin.yang at intel.com>
>>>
>>> This patch move the check of monitor from hw_params to trigger callback.
>>>
>>> The original code will check the monitor presence in hw_params. If the
>>> monitor doesn't exist, hw_params will return -ENODEV. Mostly this is OK.
>>>
>>> However, pulseaudio will check the pcm devices when kernel is booting up.
>>> It will try to open, set hw_params, prepare such pcm devices. We can't
>>> guarantee that the monitor will be connected when kernel is booting up.
>>> Especially, hdac_hdmi will export 3 pcms at most. It's hard to say users
>>> will connect 3 monitors to the HDMI/DP ports. This will cause pulseaudio
>>> fail in parsing the pcm devices because the driver will return -ENODEV in
>>> hw_params.
>>>
>>> This patch tries to move the check of monitor presence into trigger
>>> callback. This can "trick" the pulseaudio the pcm is ready.
>>>
>>> This bug is found when we try to enable HDMI detection in
>>> gnome-sound-setting for ASoC hdac_hdmi. After we enable the hdmi in UCM,
>>> pulseaudio will try to parse the hdmi pcm devices. It will cause failure if
>>> there are no monitors connected.
>>
>> Out of curiosity, how is this handled in the legacy driver? I haven't
>> done this for a long time but I remember very clearly being able to
>> play on the HDMI:3,7, etc devices without any monitors
>> connected. You'd get of course no sound but there was no error
>> reported to userspace. The hardware is perfectly capable of pushing
>> samples into the display controller using the HDAudio/iDisp link.
> 
> As mentioned in the thread, PA just picks up the stream that is
> connected via a monitor by checking / notified by the corresponding
> Jack control.  On hdac_hdmi driver, the jack control has different
> base name that is irrelevant with the output pins, so PA doesn't know
> how to interpret it, hence it's ignored.

Yes, but do we have any error checks in the hw_params or trigger cases 
with the legacy driver?