[RFC PATCH 1/3] dt-bindings: arm64: bcmbca: Merge BCM4908 into BCMBCA

[Rob Herring]

On Wed, Jul 13, 2022 at 11:37:18AM -0700, William Zhang wrote:
> Hi Rafal,
> 
> On 7/13/22 03:58, Rafał Miłecki wrote:
> > On 2022-07-13 12:50, Rafał Miłecki wrote:
> > > On 2022-07-13 02:57, William Zhang wrote:
> > > > On 7/12/22 11:18, Krzysztof Kozlowski wrote:
> > > > > On 12/07/2022 19:37, William Zhang wrote:
> > > > > > > > +      - description: BCM4908 Family based boards
> > > > > > > > +        items:
> > > > > > > > +          - enum:
> > > > > > > > +              # BCM4908 SoC based boards
> > > > > > > > +              - brcm,bcm94908
> > > > > > > > +              - asus,gt-ac5300
> > > > > > > > +              - netgear,raxe500
> > > > > > > > +              # BCM4906 SoC based boards
> > > > > > > > +              - brcm,bcm94906
> > > > > > > > +              - netgear,r8000p
> > > > > > > > +              - tplink,archer-c2300-v1
> > > > > > > > +          - enum:
> > > > > > > > +              - brcm,bcm4908
> > > > > > > > +              - brcm,bcm4906
> > > > > > > > +              - brcm,bcm49408
> > > > > > > 
> > > > > > > This is wrong.  brcm,bcm94908 followed by brcm,bcm4906 does not look
> > > > > > > like valid list of compatibles.
> > > > > > > 
> > > > > > For 4908 board variant, it will need to be followed by
> > > > > > 4908 chip. Sorry
> > > > > > for the basic question but is there any requirement to
> > > > > > enforce this kind
> > > > > > of rule?  I would assume dts writer know what he/she is
> > > > > > doing and select
> > > > > > the right combination.
> > > > > 
> > > > > The entire point of DT schema is to validate DTS.
> > > > > Combination like above
> > > > > prevents that goal.
> > > > > 
> > > > > Best regards,
> > > > > Krzysztof
> > > > Understand the DT schema purpose. But items property allows multiple
> > > > enums in the list which gives a lot of flexibility but make it hard to
> > > > validate. I am not familiar with DT schema, is there any directive to
> > > > specify one enum value depending on another so dts validation tool can
> > > > report error if combination is wrong?
> > > > 
> > > > This is our preferred format of all bcmbca compatible string
> > > > especially when we could have more than 10 chip variants for the same
> > > > chip family and we really want to work on the chip family id.  We will
> > > > make sure they are in the right combination in our own patch and patch
> > > > from other contributors. Would this work? If not, I will probably have
> > > > to revert the change of 4908(maybe append brcm,bcmbca as this chip
> > > > belongs to the same bca group) and use "enum board variant", "const
> > > > main chip id", "brcm,bca" for all other chips as our secondary choice.
> > > 
> > > I'm not sure why I didn't even receive 1/3 and half of discussion
> > > e-mails.
> > > 
> > > You can't just put all strings into a single bag and allow mixing them
> > > in any combos. Please check how it's properly handled in the current
> > > existing binding:
> > > Documentation/devicetree/bindings/arm/bcm/brcm,bcm4908.yaml
> > > 
> > > Above binding enforces that non-matching compatible strings are not used
> > > together.
> > 
> > I just noticed you're actually removing brcm,bcm4908.yaml in the 2/3 so
> > you must be aware of that file.
> > 
> > So you see a cleanly working binding in the brcm,bcm4908.yaml but
> > instead copying it you decided to wrote your own one from scratch.
> > Incorrectly.
> > 
> > This smells of NIH (not invented here). Please just use that binding I
> > wrote and move if it needed.
> 
> Not mean to discredit any of your work and I did copy over your binding and
> combine them into one SoC entry to the new bcmbca.yaml and add you as one of
> the maintainer to this file. As this change would certainly concern you,
> that's why I sent RFC first.  As I explained in the cover letter, the
> purpose of the change is to reduce the number of compatible strings and keep
> one entry for one chip family due to possible large number of chip variants.
> But since there is no way to validate the combination, I will copy the
> existing 4908 bindings as they are now but I would propose to append "brcm,
> bcmbca" as it is part of bcmbca chip. And for the other chips, we would just
> use enum "board variant", const "main chip id", const "brcm,bca".  Does that
> sound good to you?

If you want fewer combinations of compatibles, adding a genericish 
"brcm,bcmbca" is not going to help. Is there much value to adding it? 
What can you do with that information (and nothing else) is the 
question to ask. 

Rob