On Wed, Mar 22, 2023 at 12:39:48AM +0100, Christian Marangi wrote: > On Wed, Mar 22, 2023 at 12:23:59AM +0100, Andrew Lunn wrote: > > > > Are specific ethernet controllers allowed to add their own properties in > > > > led nodes? If so, this doesn't work. As-is, this allows any other > > > > properties. You need 'unevaluatedProperties: false' here to prevent > > > > that. But then no one can add properties. If you want to support that, > > > > then you need this to be a separate schema that devices can optionally > > > > include if they don't extend the properties, and then devices that > > > > extend the binding would essentially have the above with: > > > > > > > > $ref: /schemas/leds/common.yaml# > > > > unevaluatedProperties: false > > > > properties: > > > > a-custom-device-prop: ... > > > > > > > > > > > > If you wanted to define both common ethernet LED properties and > > > > device specific properties, then you'd need to replace leds/common.yaml > > > > above with the ethernet one. > > > > > > > > This is all the same reasons the DSA/switch stuff and graph bindings are > > > > structured the way they are. > > > > > > > > > > Hi Rob, thanks for the review/questions. > > > > > > The idea of all of this is to keep leds node as standard as possible. > > > It was asked to add unevaluatedProperties: False but I didn't understood > > > it was needed also for the led nodes. > > > > > > leds/common.yaml have additionalProperties set to true but I guess that > > > is not OK for the final schema and we need something more specific. > > > > > > Looking at the common.yaml schema reg binding is missing so an > > > additional schema is needed. > > > > > > Reg is needed for ethernet LEDs and PHY but I think we should also permit > > > to skip that if the device actually have just one LED. (if this wouldn't > > > complicate the implementation. Maybe some hints from Andrew about this > > > decision?) > > > > I would make reg mandatory. > > > > Ok will add a new schema and change the regex. > > > We should not encourage additional properties, but i also think we > > cannot block it. > > > > The problem we have is that there is absolutely no standardisation > > here. Vendors are free to do whatever they want, and they do. So i > > would not be too surprised if some vendor properties are needed > > eventually. > > > > Think that will come later with defining a more specific schema. But I > honestly think most of the special implementation will be handled to the > driver internally and not with special binding in DT. Then encourage no additional properties by letting whomever wants to add them to restructure the schema. ;) Rob
