Re: [PATCH v6 08/30] dt-bindings: soc: fsl: cpm_qe: cpm1-scc-qmc: Add support for QMC HDLC

[Herve Codina]

Hi Krzysztof,

On Tue, 26 Sep 2023 22:59:14 +0200
Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx> wrote:

> On 25/09/2023 15:50, Herve Codina wrote:
> >>>>> With these details, do you still think I need to change the child (channel)
> >>>>> compatible ?      
> >>>>
> >>>> From OS point of view, you have a driver binding to this child-level
> >>>> compatible. How do you enforce Linux driver binding based on parent
> >>>> compatible? I looked at your next patch and I did not see it.    
> >>>
> >>> We do not need to have the child driver binding based on parent.    
> >>
> >> Exactly, that's what I said.
> >>  
> >>> We have to ensure that the child handles a QMC channel and the parent provides
> >>> a QMC channel.
> >>>
> >>> A QMC controller (parent) has to implement the QMC API (include/soc/fsl/qe/qmc.h)
> >>> and a QMC channel driver (child) has to use the QMC API.    
> >>
> >> How does this solve my concerns? Sorry, I do not understand. Your driver
> >> is a platform driver and binds to the generic compatible. How do you
> >> solve regular compatibility issues (need for quirks) if parent
> >> compatible is not used?
> >>
> >> How does being QMC compliant affects driver binding and
> >> compatibility/quirks?
> >>
> >> We are back to my original question and I don't think you answered to
> >> any of the concerns.  
> > 
> > Well, to be sure that I understand correctly, do you mean that I should
> > provide a compatible for the child (HDLC) with something like this:
> > --- 8< ---
> >   compatible:
> >     items:
> >       - enum:
> >           - fsl,mpc885-qmc-hdlc
> >           - fsl,mpc866-qmc-hdlc
> >       - const: fsl,cpm1-qmc-hdlc
> >       - const: fsl,qmc-hdlc
> > --- 8< ---  
> 
> Yes, more or less, depending on actual compatibility and SoC-family.
> Maybe "fsl,cpm1-qmc-hdlc" item in the middle is not needed.

Ok,
I will keep "fsl,cpm1-qmc-hdlc". The CPM1 is the co-processor present in these
SoCs and it handles the QMC controller. So, it makes sense to have it in this
binding.

I plan to add support for other SoCs in the future and for these SoCs, the
co-processor is not the CPM1. So, it makes sense to keep "fsl,cpm1-qmc-hdlc"
to identify the co-processor.

> 
> > 
> > If so, I didn't do that because a QMC channel consumer (driver matching
> > fsl,qmc-hdlc) doesn't contains any SoC specific part.  
> 
> Just like hundreds of other drivers. :)
> 
> There is a paragraph about specific compatibles here:
> https://www.kernel.org/doc/html/latest/devicetree/bindings/writing-schema.html
> 
> 
> > It uses the channel as a communication channel to send/receive HDLC frames
> > to/from this communication channel.
> > All the specific SoC part is handled by the QMC controller (parent) itself and
> > not by any consumer (child).  
> 
> OK, so you guarantee in 100% for this hardware and all future (including
> designs unknown currently), that they will be 100% compatible with
> existing QMC channel consumer (child, matching fsl,qmc-hdlc) driver,
> thus there will be no need for any quirk. Specifically, there will be no
> chances that it would be reasonable to re-use the same driver for child
> (currently fsl,qmc-hdlc) in different parent.

Right,
compatible strings with SoC and co-processor will be added in the next iteration.

Thanks for your feedback.

Best regards,
Hervé

> 
> P.S. If you received this email twice, apologies, I have here troubles
> with internet.
> 
> Best regards,
> Krzysztof
>