Re: [PATCH v2 1/3] dt-bindings: pci: Add devicetree binding for Qualcomm PCIe EP controller

[Rob Herring <robh@kernel.org>]

On Sat, Jun 05, 2021 at 10:13:57PM -0500, Bjorn Andersson wrote:
> On Thu 03 Jun 05:38 CDT 2021, Manivannan Sadhasivam wrote:
> 
> > Add devicetree binding for Qualcomm PCIe EP controller used in platforms
> > like SDX55. The EP controller is based on the Designware core with
> > Qualcomm specific wrappers.
> > 
> > Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
> > ---
> >  .../devicetree/bindings/pci/qcom,pcie-ep.yaml | 144 ++++++++++++++++++
> >  1 file changed, 144 insertions(+)
> >  create mode 100644 Documentation/devicetree/bindings/pci/qcom,pcie-ep.yaml
> > 
> > diff --git a/Documentation/devicetree/bindings/pci/qcom,pcie-ep.yaml b/Documentation/devicetree/bindings/pci/qcom,pcie-ep.yaml
> > new file mode 100644
> > index 000000000000..3e357cb03a5c
> > --- /dev/null
> > +++ b/Documentation/devicetree/bindings/pci/qcom,pcie-ep.yaml
> > @@ -0,0 +1,144 @@
> > +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
> > +%YAML 1.2
> > +---
> > +$id: http://devicetree.org/schemas/pci/qcom,pcie-ep.yaml#
> > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > +
> > +title: Qualcomm PCIe Endpoint Controller binding
> > +
> > +maintainers:
> > +  - Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
> > +
> > +allOf:
> > +  - $ref: "pci-ep.yaml#"
> > +
> > +properties:
> > +  compatible:
> > +    const: qcom,sdx55-pcie-ep
> 
> The binding looks good, but this is going to cause us an inevitable
> warning as we'd have to describe the controller twice (rc + ep) in the
> sdx55.dtsi.
> 
> @Rob, what do you suggest we do about this, because it's the same
> problem currently responsible for hundreds of warnings in the case of
> GENI (where each node is duplicated for different functions).

What determines the mode? Assuming it is fixed for a platform, can't you 
just have 2 .dtsi files and include the right one. The SoC file could 
have the common h/w specific parts (clks, resets, etc.) as those 
shouldn't really be different depending on the mode. IIRC, some PCI 
bindings do this by design (meaning there's only one definition).

Rob