Re: [PATCH v2] dt-bindings: riscv: Fix CPU schema errors

[Paul Walmsley]

On Wed, 25 Sep 2019, Rob Herring wrote:

> On Wed, Sep 25, 2019 at 4:24 PM Palmer Dabbelt <palmer@xxxxxxxxxx> wrote:
> >
> > On Wed, 25 Sep 2019 06:12:52 PDT (-0700), robh@xxxxxxxxxx wrote:
> > > Fix the errors in the RiscV CPU DT schema:
> > >
> > > Documentation/devicetree/bindings/riscv/cpus.example.dt.yaml: cpu@0: 'timebase-frequency' is a required property
> > > Documentation/devicetree/bindings/riscv/cpus.example.dt.yaml: cpu@1: 'timebase-frequency' is a required property
> > > Documentation/devicetree/bindings/riscv/cpus.example.dt.yaml: cpu@0: compatible:0: 'riscv' is not one of ['sifive,rocket0', 'sifive,e5', 'sifive,e51', 'sifive,u54-mc', 'sifive,u54', 'sifive,u5']
> > > Documentation/devicetree/bindings/riscv/cpus.example.dt.yaml: cpu@0: compatible: ['riscv'] is too short
> > > Documentation/devicetree/bindings/riscv/cpus.example.dt.yaml: cpu@0: 'timebase-frequency' is a required property
> > >
> > > Fixes: 4fd669a8c487 ("dt-bindings: riscv: convert cpu binding to json-schema")
> > > Cc: Paul Walmsley <paul.walmsley@xxxxxxxxxx>
> > > Cc: Palmer Dabbelt <palmer@xxxxxxxxxx>
> > > Cc: Albert Ou <aou@xxxxxxxxxxxxxxxxx>
> > > Cc: linux-riscv@xxxxxxxxxxxxxxxxxxx
> > > Signed-off-by: Rob Herring <robh@xxxxxxxxxx>
> > > ---
> > > v2:
> > >  - Add timebase-frequency to simulator example.
> > >
> > >  .../devicetree/bindings/riscv/cpus.yaml       | 26 ++++++++++---------
> > >  1 file changed, 14 insertions(+), 12 deletions(-)
> > >
> > > diff --git a/Documentation/devicetree/bindings/riscv/cpus.yaml b/Documentation/devicetree/bindings/riscv/cpus.yaml
> > > index b261a3015f84..eb0ef19829b6 100644
> > > --- a/Documentation/devicetree/bindings/riscv/cpus.yaml
> > > +++ b/Documentation/devicetree/bindings/riscv/cpus.yaml
> > > @@ -24,15 +24,17 @@ description: |
> > >
> > >  properties:
> > >    compatible:
> > > -    items:
> > > -      - enum:
> > > -          - sifive,rocket0
> > > -          - sifive,e5
> > > -          - sifive,e51
> > > -          - sifive,u54-mc
> > > -          - sifive,u54
> > > -          - sifive,u5
> > > -      - const: riscv
> > > +    oneOf:
> > > +      - items:
> > > +          - enum:
> > > +              - sifive,rocket0
> > > +              - sifive,e5
> > > +              - sifive,e51
> > > +              - sifive,u54-mc
> > > +              - sifive,u54
> > > +              - sifive,u5
> > > +          - const: riscv
> > > +      - const: riscv    # Simulator only
> > >      description:
> > >        Identifies that the hart uses the RISC-V instruction set
> > >        and identifies the type of the hart.

The above part of this patch looks fine to me, and please consider that 
portion of this patch acked.

> > > @@ -67,8 +69,6 @@ properties:
> > >        lowercase to simplify parsing.
> > >
> > >    timebase-frequency:
> > > -    type: integer
> > > -    minimum: 1
> > >      description:
> > >        Specifies the clock frequency of the system timer in Hz.
> > >        This value is common to all harts on a single system image.
> > > @@ -102,9 +102,9 @@ examples:
> > >      cpus {
> > >          #address-cells = <1>;
> > >          #size-cells = <0>;
> > > -        timebase-frequency = <1000000>;
> > >          cpu@0 {
> > >                  clock-frequency = <0>;
> > > +                timebase-frequency = <1000000>;
> > >                  compatible = "sifive,rocket0", "riscv";
> > >                  device_type = "cpu";
> > >                  i-cache-block-size = <64>;
> > > @@ -120,6 +120,7 @@ examples:
> > >          };
> > >          cpu@1 {
> > >                  clock-frequency = <0>;
> > > +                timebase-frequency = <1000000>;
> > >                  compatible = "sifive,rocket0", "riscv";
> > >                  d-cache-block-size = <64>;
> > >                  d-cache-sets = <64>;
> > > @@ -153,6 +154,7 @@ examples:
> > >                  device_type = "cpu";
> > >                  reg = <0>;
> > >                  compatible = "riscv";
> > > +                timebase-frequency = <1000000>;
> > >                  riscv,isa = "rv64imafdc";
> > >                  mmu-type = "riscv,sv48";
> > >                  interrupt-controller {
> >
> > Looking at this spec
> >
> >     https://github.com/devicetree-org/devicetree-specification/releases/download/v0.2/devicetree-specification-v0.2.pdf
> >
> > section 3.7 says
> >
> >     Properties that have identical values across cpu nodes may be placed in the
> >     /cpus node instead. A client program must
> >     first examine a specific cpu node, but if an expected property is not found
> >     then it should look at the parent /cpus node.
> >     This results in a less verbose representation of properties which are
> >     identical across all CPUs.
> 
> The cpu sections of the spec are certainly not perfect. They are
> largely from PPC with only the most obviously things wrong fixed...

[ ... ]

> > I just bring this up because we've got an outstanding
> > bug in our port where we're not respecting what section 3.7 says and are only
> > looking at /cpus/timebase-frequency instead of /cpus/cpu@*/timebase-frequency,
> > and I'm wondering if the fix should allow for looking at
> > /cpus/timebase-frequency or just not bother.
> 
> It's perfectly fine for some deviation for each arch or being more
> restrictive. For Arm, we've generally gone the direction of everything
> goes into the cpu nodes. So tell me what you want, I just need the
> warnings gone.

We probably should keep the timebase-frequency at the /cpus level, since 
that's how the current public silicon behaves, and that's how our kernel 
code currently works.  Do you want to patch the schemas for that, or would 
you like us to?


- Paul