Re: [PATCH v1] dt-bindings: dsp: mediatek: add mt8186 dsp document

From: Krzysztof Kozlowski
Date: Thu Jun 02 2022 - 08:26:28 EST


On 02/06/2022 13:53, Tinghan Shen wrote:
> Hi Krzysztof,
>
> On Thu, 2022-06-02 at 12:45 +0200, Krzysztof Kozlowski wrote:
>> On 02/06/2022 12:19, Tinghan Shen wrote:
>>> Hi Krzysztof,
>>>
>>> On Thu, 2022-06-02 at 09:40 +0200, Krzysztof Kozlowski wrote:
>>>> On 02/06/2022 08:44, Tinghan Shen wrote:
>>>>>>> + mbox-names:
>>>>>>> + items:
>>>>>>> + - const: mbox0
>>>>>>> + - const: mbox1
>>>>>>
>>>>>> These should be rather some meaningful names, e.g. "rx" and "tx".
>>>>>
>>>>> The mbox name has to align with the adsp ipc driver.
>>>>> The adsp ipc driver is using 'mbox%d' for mailbox channels.
>>>>>
>>>>>
>>>>>
>>>
>>>
> https://urldefense.com/v3/__https://git.kernel.org/pub/scm/linux/kernel/git/broonie/sound.git/commit/?id=9db69df4bdd37eb1f65b6931ee067fb15b9a4d5c__;!!CTRNKA9wMg0ARbw!1TmempNkQhC5QuLBhyfWo_AC97MoLuWipsGV-LPaW9RKNPheU7Bgc-eboNi1JA1nC5I$
>>>>>
>>>>>
>>>>> chan_name = kasprintf(GFP_KERNEL, "mbox%d", i);
>>>>>
>>>>> /* ...snip... */
>>>>>
>>>>> adsp_chan->ch = mbox_request_channel_byname(cl, chan_name);
>>>>>
>>>>> Is it ok to continue using these names?
>>>>
>>>> It is a bit confusing... how did that driver got merged recently without
>>>> bindings? Why bindings are separate?
>>>>
>>>> The bindings always come together in one patchset with the driver
>>>> implementing them. Bindings are though a separate patch, yet still
>>>> followed by the driver which uses them.
>>>>
>>>> I do not see any compatibles in that driver, which suggests there is no
>>>> other binding using it. If that's correct, then you need to change the
>>>> driver.
>>>>
>>>
>>> The mtk-adsp-ipc driver's sole function is to encapsulate the operations
>>> of mailbox framework from adsp ipc users. The mtk-adsp-ipc is not defined
>>> in the dts file and we don't need it to be defined. The creation of mtk-adsp-ipc
>>> device is requested by adsp ipc users via the use of 'platform_device_register_data'[1].
>>>
>>> the driver implemented the mailbox framework is 'mtk-adsp-mailbox'[2]. it has
>>> corresponding hardwares and a yaml file[3] to describe it.
>>
>> I don't understand how is this related. We talk here about the
>> mbox-names for this bindings file. You replied, that these bindings are
>> already used by something, but now you say that they are not? So why do
>> you need to change anything in any driver?
>>
>> Simple question - do the bindings here "add mt8186 dsp document" are
>> used by any specific Linux driver already?
>
> This bindings, 'add mt8186 dsp document', are used by the SOF sound driver of MT8186[1].
>
> I'm sorry for miss leading you in previous reply. I was thought that you're
> asking why the mtk-adsp-ipc driver got merged without bindings. So, I tried
> to explain why mtk-adsp-ipc doesn't have bindings.

Then my question is kind of still valid:
How did "mt8186 SOF" driver got merged recently without bindings? Why
bindings are separate?

You cannot just sneak in usage of bindings in a driver, then submit
bindings and say "we already have an user!". No, the bindings come with
the driver. Always.

Linked patch [1] brings undocumented compatible mediatek,mt8186-dsp, so
you should see big fat warning when running checkpatch. So this points
that you did not run checkpatch which is another not acceptable
submission. :(

[1]
https://lore.kernel.org/all/20220422055659.8738-2-tinghan.shen@xxxxxxxxxxxx/


Best regards,
Krzysztof