Re: [PATCH 01/10] dt: bindings: clock: add mtmips SoCs clock device tree binding documentation

[Arınç ÜNAL]

On 25.03.2023 01:10, Rob Herring wrote:
> On Tue, Mar 21, 2023 at 12:02:47PM +0300, Arınç ÜNAL wrote:
> > On 21.03.2023 12:01, Krzysztof Kozlowski wrote:
> > > On 21/03/2023 09:53, Arınç ÜNAL wrote:
> > > > > I do not see how choosing one variant for compatibles having two
> > > > > variants of prefixes, complicates things. Following this argument
> > > > > choosing "ralink" also complicates!
> > > >
> > > > The idea is to make every compatible string of MTMIPS to have the ralink
> > > > prefix so it's not mediatek on some schemas and ralink on others. Simpler.
> > >
> > > Which is an ABI break, so you cannot do it.
> >
> > No, both strings stay on the driver, it's the schemas that will only keep
> > ralink.
>
> But you are adding one of the strings to the driver, right? Still an ABI
> break, but only if you have an old kernel and new DT. That can be
> somewhat mitigated with a stable backport of the new id, but still an
> ABI break.

Ah, that makes sense. Yes, I'd be adding new strings to the driver.

> Whatever is in the driver should be in the schema too. 'make 
> dt_compatible_check' will check this. And some day, I'd like that list 
> to get to 0.

I'll keep this in mind for the schemas I maintain. I will add 
ralink,rt2880-pinmux as deprecated on the pinctrl schemas so it would 
disappear from 'make dt_compatible check'. I believe I'm supposed to do 
it like this?

properties:
  compatible:
    enum:
      - ralink,rt2880-pinctrl
      - ralink,rt2880-pinmux
    deprecated:
      items:
        - const: ralink,rt2880-pinmux
          deprecated: true

Arınç