Re: [PATCH v8 1/2] dt-bindings: usb: cdns3: Add cdns,cdnsp-no-drd compatible string

[Rob Herring <robh@kernel.org>]

On Tue, May 19, 2026 at 06:49:36PM +0100, Conor Dooley wrote:
> On Tue, May 19, 2026 at 08:47:30AM +0000, Pawel Laszczak wrote:
> > >
> > >On Fri, May 15, 2026 at 01:24:38PM +0200, Pawel Laszczak via B4 Relay wrote:
> > >> From: Pawel Laszczak <pawell@cadence.com>
> > >>
> > >> Introduce a new compatible string 'cdns,cdnsp-no-drd' for Cadence
> > >> USBSS/USBSSP controllers to support hardware configurations where the
> > >> Dual-Role Device (DRD) register block is missing or inaccessible.
> > >>
> > >> This change follows the recommendation to imply hardware limitations
> > >> directly from the compatible string rather than using a dedicated
> > >> boolean property.
> > >>
> > >> When 'cdns,cdnsp-no-drd' is used:
> > >> - The 'otg' register and interrupt resources are not required.
> > >> - The 'reg' and 'interrupts' properties are restricted to 2 items
> > >>   (host and device).
> > >> - 'dr_mode' must be explicitly set to either 'host' or 'peripheral'.
> > >>
> > >> The standard 'cdns,usb3' compatible remains unchanged, maintaining
> > >> backward compatibility by requiring all 3 resource sets (otg, host, dev).
> > >>
> > >> Signed-off-by: Pawel Laszczak <pawell@cadence.com>
> > >> ---
> > >> v8:
> > >> - Update commit message to reflect schema changes.
> > >> - Removed 'cdns,no-drd' boolean property as per Rob Herring's suggestion.
> > >> - Introduced a new compatible string 'cdns,cdnsp-no-drd' for controller
> > >>   variants that lack the DRD/OTG register block.
> > >>
> > >> v7:
> > >> - Rename 'no_drd' to 'cdns,no-drd'.
> > >> - Update commit message to reflect property renaming and schema
> > >changes.
> > >> - Simplify 'reg-names' using a single enum.
> > >> - Revert 'interrupt-names' to a list of constants.
> > >> - Move 'reg' item descriptions to if/else blocks for accuracy.
> > >> - Clean up 'if/then' logic (remove redundant checks).
> > >> - Add explicit 'items' list for 'interrupt-names' in the 'else' block.
> > >>
> > >> v6:
> > >> - Fixed validation error for 'interrupt-names' by correcting
> > >>   the items definition.
> > >> - Adjusted 'minItems'/'maxItems' to properly support the optional
> > >>  'wakeup' interrupt.
> > >> - Fixed 'too long' schema error in examples.
> > >>
> > >> v5:
> > >> - Implemented strict conditional validation using if-then-else logic.
> > >> - Enforced 2 register/interrupt items and required 'dr_mode'
> > >>   (host or peripheral) when 'no_drd' is present.
> > >> - Enforced the standard 3 register/interrupt items (otg, host, dev)
> > >>   when 'no_drd' is absent to ensure backward compatibility.
> > >> - Updated 'reg-names' and 'interrupt-names' to use enums in the main
> > >>   properties section to support flexible resource ordering during
> > >>   validation.
> > >> ---
> > >> ---
> > >>  .../devicetree/bindings/usb/cdns,usb3.yaml         | 60
> > >++++++++++++++++++----
> > >>  1 file changed, 50 insertions(+), 10 deletions(-)
> > >>
> > >> diff --git a/Documentation/devicetree/bindings/usb/cdns,usb3.yaml
> > >b/Documentation/devicetree/bindings/usb/cdns,usb3.yaml
> > >> index 2d95fb7321af..7b0aa9c4a2bd 100644
> > >> --- a/Documentation/devicetree/bindings/usb/cdns,usb3.yaml
> > >> +++ b/Documentation/devicetree/bindings/usb/cdns,usb3.yaml
> > >> @@ -17,22 +17,22 @@ description:
> > >>
> > >>  properties:
> > >>    compatible:
> > >> -    const: cdns,usb3
> > >> +    enum:
> > >> +      - cdns,usb3
> > >> +      - cdns,cdnsp-no-drd
> > >
> > >I doubt this is what Rob meant, he asked for soc-specific compatibles
> > >and this is generic. A soc-specific compatible would be something like
> > >microchip,newsoc-usb3
> > 
> > Hi Conor, Rob
> > 
> > To clarify the hardware situation: I am developing and testing this on
> > an FPGA development board connected via pure PCI, without any Device Tree.
> > I do not have a specific SoC platform, nor do I know what silicon target
> > the end customers will use in the future.
> > Introducing a fake SoC-specific compatible string just to pass the DT
> > validation feels misleading and unnecessary.
> > 
> > Since cdns,usb3 is already an established generic compatible, and we
> > cannot use a generic configuration string like cdns,cdnsp-no-drd,
> > the cleanest way forward is to make this resource-driven instead of
> > compatible-driven.
> > 
> > In v9, I propose to:
> >  - Keep only the existing cdns,usb3 compatible.
> >  - Update cdns,usb3.yaml to allow minItems: 2 for reg and reg-names
> >    if dr_mode is explicitly set to "host" or "peripheral" (indicating
> >    that the OTG register block is absent).
> >  - In the driver code, determine the lack of DRD dynamically if the "otg"
> >    resource/register block is missing.
> > 
> > This elegantly solves the issue for my PCI/FPGA platform (where I just
> > won't pass the "otg" resource), complies with DT practices, and leaves
> > the door open for any future customer SoC to use cdns,usb3 with
> > standard dr_mode constraints.
> > 
> > Does this approach look acceptable to you?
> 
> Not really. I want enforcement of the correct properties in the binding,
> not permission of multiple different combinations.
> 
> Perhaps we could go back to the original cdns,usbssp compatible and
> leave an empty slot for the soc-specific compatibles, like:
>       - items:
>           - {}
>           - const: sifive,clint2        # SiFive CLINT v2 IP block
> 
> That way you can proceed with this effort without a soc-compatible
> but we don't end up permitting a generic compatible.

That sounds good to me.

Rob