Re: [PATCH 6/7] ASoC: Intel: soc-acpi-intel-tgl-match: add cs42l43 and cs56l56 support

[Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>]

On 11/30/23 04:15, Richard Fitzgerald wrote:
> On 29/11/23 16:39, Pierre-Louis Bossart wrote:
>>
>>>>>>>> +        .name_prefix = "cs35l56-8"
>>>>>>>
>>>>>>> Can these prefixes be "AMPn" to match the CS35L41, CS35L51 and
>>>>>>> CS35L56-hda driver? This prefix is used to find the matching
>>>>>>> firmware
>>>>>>> files and our naming convention for these has been cs35lxx-xxxx-ampn
>>>>>>>
>>>>>>> Is there anything that depends on the prefixes being "cs35l56-n" ?
>>>>>>
>>>>>> IIRC this name_prefix is just used for the codec_conf and hence for
>>>>>> control names/UCM. At some point userspace/driver need to know if
>>>>>> amp5
>>>>>> is left or right.
>>>>>>
>>>>>> We can certainly align on conventions but the values set in this ACPI
>>>>>> match table will not be used for firmware download - different scope.
>>>>>>
>>>>>
>>>>> They are used for our firmware download. Each amp can have its own
>>>>> unique firmware file. The ALSA prefix is used to identify which
>>>>> firmware
>>>>> file to load to which amp.
>>>>
>>>> The prefix will only be used when the card is created, specifically for
>>>> control names.
>>>> The firmware should be selected and downloaded when the device shows up
>>>> on the bus.
>>>> Card creation and device enumeration/initialization happen on different
>>>> timelines, if the machine driver is "blacklisted" or unbound I am not
>>>> sure what happens.
>>>>
>>>> There is a dependency between machine driver probe and codec firmware
>>>> download that I am not able to follow, can you please elaborate?
>>>>
>>>
>>> The codec driver has to choose which firmware to load from under
>>> /lib/firmware. It does this using a combination of SSID (to identify the
>>> target product), the ALSA prefix string (to identify which amp) and
>>> in some systems a GPIO on the motherboard to select between different
>>> models of speaker when they have multiple suppliers. This results in a
>>> firmware name like:
>>>
>>> cs35l56-<silicon rev>-dsp1-misc-<SSID>[-<SPEAKER MODEL>]-<ALSA PREFIX>
>>>
>>> You can see this if you look in the linux-firmware repo under cirrus/
>>> for cs35l41 firmware files (though the ALSA PREFIX section in those
>>> cases is not "AMPn" because they are not SDCA parts with rotation,
>>> they have a fixed left/right assignment.)
>>>
>>> We have to be careful of the length of the prefix. The 44 characters of
>>> an ALSA control name get eaten up very quickly when we start creating
>>> fully-qualified names for controls published by the firmware. So "AMPn"
>>> was nice because it was descriptive enough but only uses 5 characters
>>> of the 44.
>>>
>>> Having said that, I've calculated that we have enough characters (just)
>>> to use a prefix of "cs35l56-n". If there's a reason why that is
>>> necessary/desirable for SOF or SoundWire then we could do that. But we'd
>>> intended to use "AMPn" prefixes.
>>>
>>> We just need to decide whether to go with "AMPn". Or switch to using
>>> "cs35l56-n" for the ALSA prefix (the therefore the qualifier at the end
>>> of the firmware filename).
>>
>> Yes we have similar issues with control names in topology, the limit is
>> hit very quickly.
>>
>> I think you missed my point though that the ALSA prefix is only set when
>> the card is created, which can be sometime after the firmware needs to
>> be downloaded. I guess you could pick the firmware in the component
>> probe, which happens during the card creation, but that could be
>> sub-optimal. Given the download times you want the download to proceed
>> as early as possible.
> 
> We kick off a background task from our component_probe() to do the
> firmware download. We need the ALSA prefix, and also the wm_adsp library
> that actually handles the DSP is ASoC code so it needs a probed
> component. Doing it in a background work means it doesn't block probe().
> And the download to multiple amps can proceed in parallel - obviously
> that's constrained by bus bandwidth but we are seeing that they
> interleave.

ah ok, that makes sense. In practice the delta between codec enumeration
and component probe is probably negligible, and in a multi-amplifier
setup the download times are much larger.

So to circle back to the initial feedback, do you mind submitting a
patch with the exact naming you'd want for the prefix?