Re: [PATCH RFC v2 2/6] ASoC: dt-bindings: qcom,sm8250: Add clocks properties for I2S

[Srinivas Kandagatla <srini@kernel.org>]

On 10/9/25 4:30 PM, Neil Armstrong wrote:
> On 10/9/25 16:29, Srinivas Kandagatla wrote:
>>
>>
>> On 10/9/25 3:25 PM, Neil Armstrong wrote:
>>> On 10/9/25 16:06, Srinivas Kandagatla wrote:
>>>>
>>>>
>>>> On 10/9/25 3:03 PM, Neil Armstrong wrote:
>>>>> On 10/9/25 15:36, Srinivas Kandagatla wrote:
>>>>>>
>>>>>>
>>>>>> On 10/8/25 7:56 PM, Neil Armstrong wrote:
>>>>>>> In order to describe the block and master clock of each I2S bus, add
>>>>>>> the first 5 I2S busses clock entries.
>>>>>>>
>>>>>>> The names (primary, secondary, tertiarty, quaternary, quinary,
>>>>>>> senary)
>>>>>>> uses the LPASS clock naming which were used for a long time on
>>>>>>> Qualcomm
>>>>>>> LPASS firmware interfaces.
>>>>>>>
>>>>>>> Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
>>>>>>> ---
>>>>>>>     .../devicetree/bindings/sound/qcom,sm8250.yaml      | 21 ++++
>>>>>>> ++++++
>>>>>>> +++++++++++
>>>>>>>     1 file changed, 21 insertions(+)
>>>>>>>
>>>>>>> diff --git a/Documentation/devicetree/bindings/sound/
>>>>>>> qcom,sm8250.yaml
>>>>>>> b/Documentation/devicetree/bindings/sound/qcom,sm8250.yaml
>>>>>>> index
>>>>>>> 8ac91625dce5ccba5c5f31748c36296b12fac1a6..d1420d138b7ed8152aa53769c4d495e1674275e6 100644
>>>>>>> --- a/Documentation/devicetree/bindings/sound/qcom,sm8250.yaml
>>>>>>> +++ b/Documentation/devicetree/bindings/sound/qcom,sm8250.yaml
>>>>>>> @@ -64,6 +64,27 @@ properties:
>>>>>>>         $ref: /schemas/types.yaml#/definitions/string
>>>>>>>         description: User visible long sound card name
>>>>>>>     +  clocks:
>>>>>>> +    minItems: 2
>>>>>>> +    maxItems: 12
>>>>>>> +
>>>>>>> +  clock-names:
>>>>>>> +    minItems: 2
>>>>>>> +    items:
>>>>>>> +      # mclk is the I2S Master Clock, mi2s the I2S Bit Clock
>>>>>>> +      - const: primary-mi2s
>>>>>>> +      - const: primary-mclk
>>>>>>> +      - const: secondary-mi2s
>>>>>>> +      - const: secondary-mclk
>>>>>>> +      - const: tertiary-mi2s
>>>>>>> +      - const: tertiary-mclk
>>>>>>> +      - const: quaternary-mi2s
>>>>>>> +      - const: quaternary-mclk
>>>>>>> +      - const: quinary-mi2s
>>>>>>> +      - const: quinary-mclk
>>>>>>> +      - const: senary-mi2s
>>>>>>> +      - const: senary-mclk
>>>>>>> +
>>>>>>
>>>>>> I don't this is correct way to handling bitclk and mclks for I2S,
>>>>>> these
>>>>>> are normally handled as part of snd_soc_dai_set_sysclk()
>>>>>> transparently
>>>>>> without need of any device tree description.
>>>>>>
>>>>>> Also doing this way is an issue as this is going to break existing
>>>>>> Elite
>>>>>> based platforms, and the device description should not change across
>>>>>> these both audio firmwares.
>>>>>
>>>>> This is only for AudioReach platforms, on those platforms the
>>>>> clocks are registered in DT and are not accessible by the card.
>>>>>
>>>> Clocks will be acessable via snd_soc_dai_set_sysclk ->
>>>> q6prm_set_lpass_clock once set_sysclk support is added to q6apm-lpass
>>>> i2s dai ops.
>>>>
>>>>
>>>>> Device description is obviously different for the AudioReach
>>>>> platforms.
>>>>
>>>> Why should it be different, its same device.
>>>> We have platforms that use both Elite and Audioreach.
>>>
>>> I'm perfectly aware of that, it's the case for sc7280/qcm6490. And I
>>> agree
>>> the card bindings is the same, but it doesn't mean the DSP elements
>>> are the
>>> same and uses in the same manner.
>>>
>>> So let's forget the bindings and forget those clocks entries, and
>>> imagine
>>> I'll implement those _sys_sysclk calls like for the Elite platforms.
>>> This means I'll bypass the clock framework by directly setting the PRM
>>> clocks, this is clearly a layer violation.
>>
>> You can claim clocks in the dsp layer (q6apm-lpass-dais) instead of
>> claiming it in machine layer, it does not necessarily have to bypass the
>> clk framework.
> 
> The current q6afe implementation totally bypasses the clock framework:
> 
> static int q6afe_set_lpass_clock_v2(struct q6afe_port *port,
>                  struct afe_clk_set *cfg)
> {
>     return q6afe_port_set_param(port, cfg, AFE_PARAM_ID_CLOCK_SET,
>                     AFE_MODULE_CLOCK_SET, sizeof(*cfg));
> }
> 
> I have no time right now to implement all that for q6apm & q6prm in the
> way you propose, so I'll probably not send a new version.

This is not really scalable, you are suggesting that we add this for
every board, when you can do it more generically in the dsp code using
set_sysclk callback.



--srini
> 
> Neil
> 
>>
>> --srini
>>>
>>> Neil
>>>
>>>>
>>>> --srini
>>>>>
>>>>> Neil
>>>>>
>>>>>>
>>>>>> thanks,
>>>>>> Srini
>>>>>>
>>>>>>>     patternProperties:
>>>>>>>       ".*-dai-link$":
>>>>>>>         description:
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>