Re: [PATCH 1/2] dt-bindings: watchdog: mediatek,mtk-wdt: add MT7988 watchdog and toprgu

[Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>]

On 10/11/2023 16:12, Daniel Golle wrote:
>>>>>>>>> diff --git a/include/dt-bindings/reset/mediatek,mt7988-resets.h b/include/dt-bindings/reset/mediatek,mt7988-resets.h
>>>>>>>>> new file mode 100644
>>>>>>>>> index 0000000000000..fa7c937505e08
>>>>>>>>> --- /dev/null
>>>>>>>>> +++ b/include/dt-bindings/reset/mediatek,mt7988-resets.h
>>>>>>>>> @@ -0,0 +1,12 @@
>>>>>>>>> +/* SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause) */
>>>>>>>>> +
>>>>>>>>> +/* TOPRGU resets */
>>>>>>>>
>>>>>>>> The first reset is zero, the second reset is one.
>>>>>>>>
>>>>>>>> Where's the zero'th reset? :-)
>>>>>>>
>>>>>>> Currently the reset numbers represent the corresponding bit positions in
>>>>>>> the toprgu register, as this is how the mtk-wdt driver is organized.
>>>>>>>
>>>>>>> So there is probably something at bit 0, and also at bit 3~11 and
>>>>>>> maybe also 17~23, but it's unknown and may be added later once known
>>>>>>> and/or needed.
>>>>>>
>>>>>> There is no need to put register bits, which are not used by the driver,
>>>>>> in the bindings.
>>>>>
>>>>> There aren't. That's why there isn't a zero'th reset (and also not 3~11, 17~24).
>>>>>
>>>>> Or should the driver be reorganized to provide a mapping of logical to
>>>>> physical resets, and then have only the needed once present and start
>>>>> counting logical resets from 0? This is doable, of course, but it's a
>>>>> bit of effort just for the aesthetical goal of starting to count from
>>>>> zero and continous in header file.
>>>>>
>>>>> And, of course, chances are that other currently still unused bits
>>>>> will be needed at a later point which then would mean having to add
>>>>> them in at least 2 places (header file and mapping logical<->physical)
>>>>> where as currently it would just mean adding a line defining it in the
>>>>> header file.
>>>>
>>>> You can do it, but it's not what I wrote here. So bear with me:
>>>>
>>>> "There is no need to put register bits in the bindings."
>>
>> No comments here, so I assume you agree with this.

Here is the answer to...

>>
>>>>
>>>> You replied "There aren't", which I don't understand in this context. I
>>>> can be clearer:
>>>> Drop this hunk.
>>>
>>> So adding the file to include/dt-bindings/reset/ should go into a
>>> seperate patch? Because including it with the driver itself gave me
>>> a checkpath warning telling me that dt-bindings should go seperate,
>>> which is why I included it with the binding docs.
>>
>> No, I said the hunk should be dropped. Removed.
> 
> I guess we are somehow misunderstanding each other.
> Lets go with an example. I can put the header into a commit of its own,
> just like commit
> 5794dda109fc8 dt-bindings: reset: mt7986: Add reset-controller header file
> https://lore.kernel.org/r/20220105100456.7126-2-sam.shih@mediatek.com
> 
> Would that be acceptable? And if not, why?

...this question.

Again, whether this is separate patch - it is still hunk which I think
should be removed. I gave the reason "why" in this mail thread and in
multiple other discussions.



Best regards,
Krzysztof