Re: [PATCH v2 1/5] media: dt-bindings: ov5645: Convert OV5645 binding to a schema

[Krzysztof Kozlowski]

On 15/10/2022 01:54, Laurent Pinchart wrote:
> Hi Rob,
> 
> On Fri, Oct 14, 2022 at 04:40:29PM -0500, Rob Herring wrote:
>> On Fri, Oct 14, 2022 at 10:27:53PM +0100, Lad, Prabhakar wrote:
>>> On Fri, Oct 14, 2022 at 10:05 PM Rob Herring <robh+dt@xxxxxxxxxx> wrote:
>>>> On Fri, Oct 14, 2022 at 1:35 PM Prabhakar <prabhakar.csengg@xxxxxxxxx> wrote:
>>>>>
>>>>> From: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
>>>>>
>>>>> Convert the simple OV5645 Device Tree binding to json-schema.
>>>>>
>>>>> The previous binding marked the below properties as required which was a
>>>>> driver requirement and not the device requirement so just drop them from
>>>>> the required list during the conversion.
>>>>> - clock-frequency
>>>>> - enable-gpios
>>>>> - reset-gpios
>>>>>
>>>>> Also drop the "clock-names" property as we have a single clock source for
>>>>> the sensor and the driver has been updated to drop the clk referencing by
>>>>> name.
>>>>
>>>> Driver requirements are the ABI!
>>>>
>>>> This breaks a kernel without the driver change and a DTB that has
>>>> dropped the properties.
>>>>
>>> I already have a patch for the driver [0] which I missed to include
>>> along with the series.
>>
>> You completely miss the point. Read the first sentence again. Changing 
>> driver requirements changes the ABI.
>>
>> This breaks the ABI. The driver patch does not help that.
> 
> I'm not following you here. If the DT binding makes a mandatory property
> optional, it doesn't break any existing platform. The only thing that
> would not work is a new DT that doesn't contain the now optional
> property combined with an older driver that makes it required. That's
> not a regression, as it would be a *new* DT.

You're right although in-tree DTS are now not compatible with older
kernels. So it is not only about new DTS, it is about our kernel DTS
which requires new kernel to work.

DTS are exported and used by other systems, thus if someone blindly
takes this new DTS without clock-names, his kernel/OS/bootloader might
stop working.

That is however a more relaxed requirement than kernel ABI against old DTS.

> 
>>>> Also, with 'clock-names' dropped, you've just introduced a bunch of
>>>> warnings on other people's platforms. Are you going to 'fix' all of
>>>> them?
>>>>
>>> Yes I will fix them, once the patch driver patch [0] is merged in.
>>
>> Why? You are just making extra work. We have enough warnings as-is to 
>> fix.
> 
> I agree that a DT binding change should patch all in-tree DTS to avoid
> introducing new warnings.

Yes.

Best regards,
Krzysztof