[PATCH] dt-bindings: power: Add SiFive Domain Management controllers

[Nick Hu]

On Fri, May 9, 2025 at 11:57 PM Conor Dooley <conor at kernel.org> wrote:
>
> On Fri, May 09, 2025 at 08:40:28AM +0200, Krzysztof Kozlowski wrote:
> > On Fri, May 09, 2025 at 10:16:04AM GMT, Nick Hu wrote:
> > > SiFive Domain Management controller includes the following components
> > > - SiFive Tile Management Controller
> > > - SiFive Cluster Management Controller
> > > - SiFive Core Complex Management Controller
> > >
> > > These controllers control the clock and power domain of the
> > > corresponding domain.
> > >
> > > Signed-off-by: Nick Hu <nick.hu at sifive.com>
> > > Reviewed-by: Samuel Holland <samuel.holland at sifive.com>
> > > ---
> > >  .../devicetree/bindings/power/sifive,tmc.yaml | 89 +++++++++++++++++++
> >
> > Where is a patch with the driver (user of the binding)?
> >
> > >  1 file changed, 89 insertions(+)
> > >  create mode 100644 Documentation/devicetree/bindings/power/sifive,tmc.yaml
> > >
> > > diff --git a/Documentation/devicetree/bindings/power/sifive,tmc.yaml b/Documentation/devicetree/bindings/power/sifive,tmc.yaml
> > > new file mode 100644
> > > index 000000000000..7ed4f290b94b
> > > --- /dev/null
> > > +++ b/Documentation/devicetree/bindings/power/sifive,tmc.yaml
> > > @@ -0,0 +1,89 @@
> > > +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
> > > +%YAML 1.2
> > > +---
> > > +$id: http://devicetree.org/schemas/power/sifive,tmc.yaml#
> > > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > +
> > > +title: SiFive Domain Management Controller
> > > +
> > > +maintainers:
> > > +  - Cyan Yang <cyan.yang at sifive.com>
> > > +  - Nick Hu <nick.hu at sifive.com>
> > > +  - Samuel Holland <samuel.holland at sifive.com>
> > > +
> > > +description: |
> > > +  This is the device tree binding for the following SiFive Domain Management Controllers.
> >
> > Explain the hardware, not that "binding is a binding for ...".
> >
> > Also, wrap according to Linux coding style.
> >
> >
> > > +  - Tile Management Controller
> > > +      - TMC0
> > > +      - TMC1
> > > +      - TMC2
> > > +      - TMC3
> > > +  - Subsystem Management Controller
> > > +      - SMC0
> > > +      - SMC1
> > > +      - SMC2
> > > +      - SMC3
> > > +  - Cluster Management Controller
> > > +      - CMC2
> > > +      - CMC3
> > > +  SiFive Domain Management Controllers support the SiFive Quiet Interface
> > > +  Protocol (SQIP) starting from the Version 1. The control method is
> > > +  different from the Version 0, making them incompatible.
> > > +
> > > +allOf:
> > > +  - $ref: power-domain.yaml#
> > > +
> > > +properties:
> > > +  compatible:
> > > +    oneOf:
> > > +      - items:
> > > +          - {}
> > > +          - pattern: "^sifive,[ts]mc0$"
> > > +      - items:
> > > +          - {}
> > > +          - pattern: "^sifive,[ts]mc3$"
> > > +          - pattern: "^sifive,[ts]mc2$"
> > > +          - pattern: "^sifive,[ts]mc1$"
> > > +      - items:
> > > +          - {}
> > > +          - pattern: "^sifive,[ts]mc2$"
> > > +          - pattern: "^sifive,[ts]mc1$"
> > > +      - items:
> > > +          - {}
> > > +          - pattern: "^sifive,[ts]mc1$"
> > > +      - items:
> > > +          - {}
> > > +          - const: sifive,cmc3
> > > +          - const: sifive,cmc2
> > > +      - items:
> > > +          - {}
> > > +          - const: sifive,cmc2
> >
> > All of this is just unexpected. Why any compatible should come with
> > these?
>
> It's also not quite correct either, right? Or may not be correct at
> least. It permits "xxx", "tmc2", "smc1" and "xxx", "smc2", "tmc1"
> which mean that the smc and tmc must be identical in terms of
> programming model.
>
You are right, I hadn't considered that case. I'll address it in the
next version.

> > You need to use SoC specific compatibles.
>
> I think there's some slack to provide here, sifive are upstreaming it in
> advance of there being customers (or customers ready to upstream) and this
> format allows us to accept bindings/drivers and the customer will have
> to add a soc-specific compatible in order to actually use these in a
> dts. I think it's better to accept something along these lines than
> stall out until a customer decides to upstream their user. That said, I
> would expect this to come (as you mentioned above) with the driver.
>
Thanks for the understanding =)
I'm hoping the driver can be submitted at a later stage.
The driver that handles the MMIO is implemented in OpenSBI and depends
on some prerequisite patches [1], so it will follow afterward.

Links:
- [1] https://lore.kernel.org/opensbi/CAKddAkD00gLgpzOCXY9mXaebr2qZcU9mkUGOZ4278w0bmiLuBQ@mail.gmail.com/T/#t

> > > +
> > > +  reg:
> > > +    maxItems: 1
> > > +
> > > +  sifive,feature-level:
> > > +    description: |
> > > +      Supported power features. This property is absent if the full set of features
> > > +      is supported
> >
> > Compatible defines this. Drop.
> >
> >
> > > +    $ref: /schemas/types.yaml#/definitions/string
> > > +    enum: ["nopg", "ceasepg", "runonlypg"]
> > > +
> > > +  "#power-domain-cells":
> > > +    const: 0
> > > +
> > > +if:
> > > +  not:
> > > +    properties:
> > > +      compatible:
> > > +        contains:
> > > +          pattern: "^sifive,[tsc]mc3$"
> > > +then:
> > > +  properties:
> > > +    sifive,feature-level: false
> > > +
> > > +required:
> > > +  - compatible
> > > +  - reg
> > > +
> > > +additionalProperties: false
> >
> > Missing example.
>
> You can't actually make an example that passes validation when the
> soc-specific compatibles are not added, so this would require adding
> some.
I can add a comment here to note that the example should be included
once a SoC-specific compatible string is available.

Best Regards,
Nick