Re: [PATCH RFC 1/4] dt-bindings: clock: rename mt7986-clk.h to mediatek,mt7986-clk.h

[Krzysztof Kozlowski]

On 24/02/2023 22:06, Daniel Golle wrote:
> On Thu, Feb 23, 2023 at 11:18:37AM +0100, Krzysztof Kozlowski wrote:
>> On 22/02/2023 19:09, Daniel Golle wrote:
>>> Rename dt-bindings header file mt7986-clk.h to mediatek,mt7986-clk.h,
>>> propagate this change also to mt7986a.dtsi which is the only user.
>>>
>>> Signed-off-by: Daniel Golle <daniel@xxxxxxxxxxxxxx>
>>> ---
>>>  arch/arm64/boot/dts/mediatek/mt7986a.dtsi                       | 2 +-
>>>  drivers/clk/mediatek/clk-mt7986-infracfg.c                      | 2 +-
>>>  drivers/clk/mediatek/clk-mt7986-topckgen.c                      | 2 +-
>>>  .../dt-bindings/clock/{mt7986-clk.h => mediatek,mt7986-clk.h}   | 0
>>
>> You cannot have bindings and DTS and drivers mixed together. Which
>> points to fact - you cannot make such change... Also your commit msg
>> does not justify "why" you are doing it.
> 
> The reason is simply consistency with most other SoCs.
> 
> So to make such a change possible, I should do it in three steps, right?
> Step 1: dt-bindings: copy mt7986-clk.h to mediatek,mt7986-clk.h
> Step 2: changes in drivers
> Step 3: dt-bindings: remove mt7986-clk.h
> 
> Would this change be acceptable if carried out in 3 commits?

For new code - it would have to be in three commits. For existing code,
how you are going to achieve it with keeping full bisectability?
Remember that each commits goes to different branch and/or tree! It's
probably doable within two Linux releases with some header wrappers...
but is it worth the effort just for "consistency"?

> And using the same strategy also the other changes suggested in the
> series?

All of them have the same bisectability problems...

Best regards,
Krzysztof