On Tue, Nov 12, 2024 at 01:36:51PM +0100, Niklas Cassel wrote:

On Mon, Nov 11, 2024 at 10:25:12AM -0600, Rob Herring wrote:

quoted

quoted

quoted

I don't know the yaml too well, but it is not obvious how adding a few
status = "disabled"; status = "okay"; fixes a "'anyOf' conditional failed".

Maybe you can expand the explanation a bit?

      Andrew

Hi angelo,

I guess the dtbs_check only checks required properties from yaml if the node is enabled.

Yes, that is exactly how it works.

Rob

Hello Rob,

If we look at e.g. this binding:
Documentation/devicetree/bindings/pci/rockchip-dw-pcie.yaml

We can see that it does not define iommu-map in the binding,
likewise the binding does have:
unevaluatedProperties: false


If I apply my patch that adds iommu-map for e.g. the pcie2x1l0 node:
(the patch does not add anything to the binding above):
https://lore.kernel.org/linux-rockchip/20241107123732.1160063-2-cassel@kernel.org/ (local)


If look at the pcie2x1l0 node, it is marked as status = "disabled"
in arch/arm64/boot/dts/rockchip/rk3588-extra.dtsi

but is marked as status = "enabled"
in arch/arm64/boot/dts/rockchip/rk3588-rock-5b.dts

If I run CHECK_DTBS for this dts/dtb:
$ ARCH=arm64 CROSS_COMPILE=aarch64-linux-gnu- make CHECK_DTBS=y rockchip/rk3588-rock-5b.dtb
  DTC [C] arch/arm64/boot/dts/rockchip/rk3588-rock-5b.dtb
$ 

No warnings.

What am I missing?

Considering the warning in this series where the binding also
had unevaluatedProperties: false
I would have expected the same error for the pcie2x1l0 node.

(And if I look at most PCI controler bindings, they actually do define
iommu-map, so it seems a requirement for it to be defined if used.)

Or perhaps the question should have been, if iommu-map is an exception,
why isn't iommus also an exception?


Kind regards,
Niklas