Re: [PATCH 3/3] tty: serial: samsung: Remove unused artpec-8 specific code

[Krzysztof Kozlowski <krzk@kernel.org>]

On 12/09/2025 07:19, Ravi Patel wrote:
>>>> Ah, no, I mixed up patches with recent DTS for Artpec-8. This serial ABI
>>>> was accepted three years ago (!!!), so you are Geert absolutely right -
>>>> that's ABI break.
>>>
>>> Thank you for your review.
>>>
>>> The DTS patches for ARTPEC-8 is added recently (https://lore.kernel.org/linux-samsung-soc/20250901051926.59970-1-
>> ravi.patel@samsung.com/)
>>> Before that, there was no user (in DT) of "axis,artpec8-uart" compatible.
>>> So I am not convinced of ABI break (considering patch #1 and #2 goes first with review comment fixes)
>>
>>
>> ABI is defined by bindings and implemented by kernel. Having DTS user is
>> irrelevant to fact whether ABI is or is not broken.
>>
>> Having DTS user determines the known impact of known ABI breakage.
> 
> OK. So does that mean if someone adds the ABI then it cannot be reverted,
> because of it breaks backword compatibility (users are using ABI in their local DTB) ?
> 
> Please suggest what should be the proper way.


Three years ago you (as in plural) submitted the Artpec-8 bindings and
driver with purpose. I think it was also claimed that DTS might come or
might not come ever, so the purpose was to supported out of tree users.

This means you made a contract in the kernel for that support.

You cannot change that ABI, because otherwise the contract you made
three years ago meant nothing.

You need to keep full backwards compatibility in the kernel. Look at
mailing list or git log how others dealt with it.

Best regards,
Krzysztof