Re: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi: set additionalProperties to true

[Cristian Marussi <cristian.marussi@arm.com>]

On Tue, Apr 09, 2024 at 09:09:46AM -0500, Rob Herring wrote:
> On Tue, Apr 9, 2024 at 7:01 AM Cristian Marussi
> <cristian.marussi@arm.com> wrote:
> >
> > On Tue, Apr 09, 2024 at 09:25:10AM +0000, Peng Fan wrote:
> > > Hi Krzysztof,
> > >
> > > > Subject: RE: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi: set
> > > > additionalProperties to true
> > > >
> > > > > Subject: Re: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi: set
> > > > > additionalProperties to true
> > > > >
> > > > > On 08/04/2024 08:08, Peng Fan wrote:
> > > > > >> Subject: Re: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi: set
> > > > > >> additionalProperties to true
> > > > > >>
> > > > > >> On 08/04/2024 01:50, Peng Fan wrote:
> > > > > >>>> Subject: Re: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi: set
> > > > > >>>> additionalProperties to true
> > > > > >>>>
> > > > > >>>> On 07/04/2024 12:04, Peng Fan wrote:
> > > > > >>>>>> Subject: Re: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi:
> > > > > >>>>>> set additionalProperties to true
> > > > > >>>>>>
> > > > > >>>>>> On 07/04/2024 02:37, Peng Fan wrote:
> > > > > >>>>>>>> Subject: Re: [PATCH v2 1/6] dt-bindings: firmware: arm,scmi:
> > > > > >>>>>>>> set additionalProperties to true
> > > > > >>>>>>>>
> > > > > >>>>>>>> On 05/04/2024 14:39, Peng Fan (OSS) wrote:
> > > > > >>>>>>>>> From: Peng Fan <peng.fan@nxp.com>
> > > > > >>>>>>>>>
> > > > > >>>>>>>>> When adding vendor extension protocols, there is dt-schema
> > > > > >> warning:
> > > > > >>>>>>>>> "
> > > > > >>>>>>>>> imx,scmi.example.dtb: scmi: 'protocol@81', 'protocol@84' do
> > > > > >>>>>>>>> not match any of the regexes: 'pinctrl-[0-9]+'
> > > > > >>>>>>>>> "
> > > > > >>>>>>>>>
> > > > > >>>>>>>>> Set additionalProperties to true to address the issue.
> > > > > >>>>>>>>
> > > > > >>>>>>>> I do not see anything addressed here, except making the
> > > > > >>>>>>>> binding accepting anything anywhere...
> > > > > >>>>>>>
> > > > > >>>>>>> I not wanna add vendor protocols in arm,scmi.yaml, so will
> > > > > >>>>>>> introduce a new yaml imx.scmi.yaml which add i.MX SCMI
> > > > > >>>>>>> protocol
> > > > > >> extension.
> > > > > >>>>>>>
> > > > > >>>>>>> With additionalProperties set to false, I not know how, please
> > > > > suggest.
> > > > > >>>>>>
> > > > > >>>>>> First of all, you cannot affect negatively existing devices
> > > > > >>>>>> (their
> > > > > >>>>>> bindings) and your patch does exactly that. This should make
> > > > > >>>>>> you thing what is the correct approach...
> > > > > >>>>>>
> > > > > >>>>>> Rob gave you the comment about missing compatible - you still
> > > > > >>>>>> did not address that.
> > > > > >>>>>
> > > > > >>>>> I added the compatible in patch 2/6 in the examples "compatible
> > > > > >>>>> =
> > > > > >>>> "arm,scmi";"
> > > > > >>>>
> > > > > >>>> So you claim that your vendor extensions are the same or fully
> > > > > >>>> compatible with arm,scmi and you add nothing... Are your
> > > > > >>>> extensions/protocol valid for arm,scmi?
> > > > > >>>
> > > > > >>> Yes. They are valid for arm,scmi.
> > > > > >>>
> > > > > >>>  If yes, why is this in separate binding. If no, why you use
> > > > > >>> someone
> > > > > >>>> else's compatible?
> > > > > >>>
> > > > > >>> Per SCMI Spec
> > > > > >>> 0x80-0xFF: Reserved for vendor or platform-specific extensions to
> > > > > >>> this interface
> > > > > >>>
> > > > > >>> i.MX use 0x81 for BBM, 0x84 for MISC. But other vendors will use
> > > > > >>> the id for their own protocol.
> > > > > >>
> > > > > >> So how are they valid for arm,scmi? I don't understand.
> > > > > >
> > > > > > arm,scmi is a firmware compatible string. The protocol node is a sub-node.
> > > > > > I think the arm,scmi is that saying the firmware following SCMI spec
> > > > > > to implement the protocols.
> > > > > >
> > > > > > For vendor reserved ID, firmware also follow the SCMI spec to
> > > > > > implement their own usage, so from firmware level, it is ARM SCMI
> > > > > > spec
> > > > > compatible.
> > > > >
> > > > > That's not the point. It is obvious that your firmware is compatible
> > > > > with arm,scmi, but what you try to say in this and revised patch is
> > > > > that every arm,scmi is compatible with your implementation. What you
> > > > > are saying is that 0x84 is MISC protocol for every firmware, Qualcomm,
> > > > > NXP, Samsung, TI, Mediatek etc.
> > > > >
> > > > > I claim it is not true. 0x84 is not MISC for Qualcomm, for example.
> > > >
> > > > You are right. I am lost now on how to add vendor ID support, using
> > > > arm,scmi.yaml or adding a new imx,scmi.yaml or else.
> > >
> >
> > Hi Peng,
> >
> > I dont think in the following you will find the solution to the problem,
> > it is just to recap the situation and constraints around vendor protocol
> > bindings.
> >
> > Describing SCMI vendors protocols is tricky because while on one side
> > the protocol node has to be rooted under the main scmi fw DT node (like
> > all the standard protocols) and be 'derived' from the arm,scmi.yaml
> > protocol-node definition, the optional additional properties of the a specific
> > vendor protocol nodes can be customized by each single vendor, and since,
> > as you said, you can have multiple protocols from different vendors sharing the
> > same protocol number, you could have multiple disjoint sets of valid properties
> > allowed under that same protocol node number; so on one side you have to stick
> > to some basic protocol-node defs and be rooted under the SCMI node, while on
> > the other side you will have multiple possibly allowed sets of additional
> > properties to check against, so IOW you cannot anyway just set
> > additionalProperties to false since that will result in no checks at all.
> >
> > As a consequence, at runtime, in the final DTB shipped with a specific
> > platform you should have only one of the possible vendor nodes for each
> > of these overlapping protocols, and the SCMI core at probe time will
> > pick the proper protocol implementation based on the vendor/sub_vendor
> > IDs gathered from the running SCMI fw platform at init: this way you
> > can just build the usual "all-inclusive" defconfig without worrying
> > about vendor protocol clashes since the SCMI core can pick the right
> > protocol implementation, you should just had taken care to provide the
> > proper DTB for your protocol; BUT this also means that it is not possible
> > to add multiple DT bindings based on a 'if vendor' condition since the
> > vendor itself is NOT defined and not needed in the bindings since it is
> > discoverable at runtime.
> >
> > So, after all of this blabbing of mine about this, I am wondering if it
> > is not possible that the solution is to handle each and every vendor
> > protocol node that appears with a block of addtional properties that
> > are picked via a oneOf statement from some external vendor specific
> > yaml.
> > (...in a similar way to how pinctrl additional properties are added...)
> >
> >
> > NOTE THAT the following is just an example of what I mean, it is certainly
> > wrong, incomplete annd maybe just not acceptable (and could cause DT
> > maintainers eyes to bleed :P)...
> >
> > ...so it is just fr the sake of explaining what I mean...
> >
> > diff --git a/Documentation/devicetree/bindings/firmware/arm,scmi.yaml b/Documentation/devicetree/bindings/firmware/arm,scmi.yaml
> > index e9d3f043c4ed..3c38a1e3ffed 100644
> > --- a/Documentation/devicetree/bindings/firmware/arm,scmi.yaml
> > +++ b/Documentation/devicetree/bindings/firmware/arm,scmi.yaml
> > @@ -278,6 +278,22 @@ properties:
> >      required:
> >        - reg
> >
> > +  protocol@81:
> > +    $ref: '#/$defs/protocol-node'
> > +    unevaluatedProperties: false
> > +
> > +    properties:
> > +      reg:
> > +        const: 0x81
> > +
> > +    patternProperties:
> > +      '$':
> > +        type: object
> 
> Did you mean to have child nodes under the protocol node rather than in it?

... nope ... it is just as bad as my yaml-fu is :P ... but not sure if
vendors has also this needs or plain props will suffice...

> 
> > +        oneOf:
> > +          - $ref: /schemas/vendor-A/scmi-protos.yaml#
> > +          - $ref: /schemas/vendor-B/protos.yaml#
> 
> Moved up one level, this would work, but it would have to be an
> 'anyOf' because it is possible that 2 vendors have the exact same set
> of properties.
> 

ok

> I can think of 2 other ways to structure this.
> 
> First, is a specific vendor protocol discoverable? Not that is 0x81
> protocol present, but that 0x81 is vendor Foo's extra special
> value-add protocol? If not, I think we should require a compatible
> string on vendor protocols. Then the base SCMI schema can require just
> that, and each vendor protocol defines its node with a $ref to
> '#/$defs/protocol-node'.

Basically yes it is discoverable, since at runtime the SCMI core, early on,
normally discovers the vendor_id/sub_vendor_id by querying the platform via
Base protocol and then later only loads/initializes (by closest match) the
vendor protocols that are present in the DT AND that has been 'tagged' at
compile time with the same vendor_id/sub_vendor_id tuple (in the vendor
module code, struct scmi_protocol)

Of course you should take care to put the proper protocol@81 node in your
vendor_A DTB for the vendor_A SCMI driver to make use of the additional
vendor_A properties that you have defined under your node as referred
in your vendor-protos.yaml...if you botch that up I will load a protocol
and call your vendor_A driver with a vendor_X DT node.

DT is currrently vendor-agnostic.

> 
> The 2nd way is just a variation of the oneOf above, but do we do 1
> file per vendor protocol or 1 file per vendor. Either should be
> doable, just a matter of where 'protocol@81', etc. are defined.
> 

Oh, yes mine was just an ill example...one file per vendor will do just
fine: the important thing is that the list and the yaml itself can be
extended as new vendors appears (in a backward compatble way of course)

Thanks,
Cristian