Re: [PATCH v2 2/4] ASoC: codecs: hda: Skip HDMI/DP registration if i915 is missing

[Cezary Rojewski <cezary.rojewski@intel.com>]

On 2024-02-26 12:09 PM, Takashi Iwai wrote:
> On Fri, 23 Feb 2024 18:09:59 +0100,
> Cezary Rojewski wrote:
>>
>> On 2024-02-23 3:44 PM, Takashi Iwai wrote:
>>> On Fri, 23 Feb 2024 12:46:24 +0100,
>>> Cezary Rojewski wrote:
>>>>
>>>> If i915 does not support given platform but the hardware i.e.: HDAudio
>>>> codec is still there, the codec-probing procedure will succeed for such
>>>> device but the follow up initialization will always end up with -ENODEV.
>>>>
>>>> While bus could filter out address '2' which Intel's HDMI/DP codecs
>>>> always enumerate on, more robust approach is to check for i915 presence
>>>> before registering display codecs.
>>>>
>>>> Signed-off-by: Cezary Rojewski <cezary.rojewski@intel.com>
>>>> ---
>>>>    sound/soc/codecs/hda.c | 5 +++++
>>>>    1 file changed, 5 insertions(+)
>>>>
>>>> diff --git a/sound/soc/codecs/hda.c b/sound/soc/codecs/hda.c
>>>> index d2117e36ddd1..d9e7cd8aada2 100644
>>>> --- a/sound/soc/codecs/hda.c
>>>> +++ b/sound/soc/codecs/hda.c
>>>> @@ -350,6 +350,11 @@ static int hda_hdev_attach(struct hdac_device *hdev)
>>>>    	struct hda_codec *codec = dev_to_hda_codec(&hdev->dev);
>>>>    	struct snd_soc_component_driver *comp_drv;
>>>>    +	if (hda_codec_is_display(codec) &&
>>>> !hdev->bus->audio_component) {
>>>> +		dev_dbg(&hdev->dev, "no i915, skip registration for 0x%08x\n", hdev->vendor_id);
>>>> +		return 0;
>>>
>>> Should we return success here, or would it better with -ENODEV?
>>> IIUC, the code path is from the early hda_codec_driver_probe() hook,
>>> so returning an error can work.
>>
>> Good suggestion. Indeed attach() is called by probe() which treats
>> -ENODEV just fine.
>>
>> There is a consequence to that though. Logs from LKF show:
>>
>> snd_soc_hda_codec:hda_hdev_attach: snd_hda_codec_hdmi hdaudioB0D2: no
>> i915, skip registration for 0x80862811
>> snd_soc_hda_codec:hda_hdev_attach: snd_hda_codec_generic hdaudioB0D2:
>> no i915, skip registration for 0x80862811
>> snd_soc_hda_codec:hda_hdev_attach: snd_hda_codec_generic hdaudioB0D2:
>> no i915, skip registration for 0x80862811
>> snd_hda_codec:snd_hda_codec_configure: hdaudio hdaudioB0D2: Unable to
>> bind the codec
>> snd_soc_avs 0000:00:1f.3: failed to config codec -19
>> snd_soc_avs 0000:00:1f.3: Codec #2 probe error; disabling it...
> 
> Yeah the latter two are basically fallbacks, and I guess we can
> disable them for ASoC case?  An additional patch like below.

Thanks for feedback, Takashi. One of the design philosophies for the 
avs-driver is to avoid any 'if (bus->ext_ops)' statements if possible. 
This is to keep runtime flow of snd_hda_intel and its DSP equivalent as 
close as possible.

What I propose is: address the problem directly in the avs-driver by 
ignoring codecs for which probe_codec() returns -ENODEV. The TLDR of my 
previous message is: we do not want to do any kind of recovery e.g.: 
HDAudio controller reset if there is no supporting driver for given 
codec. Will sent a separate patch as part of v3.

> --- a/sound/pci/hda/hda_bind.c
> +++ b/sound/pci/hda/hda_bind.c
> @@ -279,6 +279,10 @@ static int codec_bind_generic(struct hda_codec *codec)
>   	if (codec->probe_id)
>   		return -ENODEV;
>   
> +	/* no generic fallback for ASoC binding */
> +	if (codec->bus->core.ext_ops)
> +		return -ENODEV;
> +
>   	if (is_likely_hdmi_codec(codec)) {
>   		codec->probe_id = HDA_CODEC_ID_GENERIC_HDMI;
>   		request_codec_module(codec);