On 02/08/2022 15:10, Kunihiko Hayashi wrote:
> On 2022/08/02 17:33, Krzysztof Kozlowski wrote:
>> On 30/07/2022 13:58, Arnd Bergmann wrote:
>>> On Mon, Jul 4, 2022 at 2:20 AM Kunihiko Hayashi
>>> <hayashi.kunihiko@xxxxxxxxxxxxx> wrote:
>>>>
>>>> UniPhier PCIe endpoint controller doesn't use "snps,dw-pcie-ep"
>>>> compatible,
>>>> so this is no longer needed. Remove the compatible string from the
>>>> pcie-ep
>>>> node to fix the following warning.
>>>>
>>>> uniphier-pro5-epcore.dtb: pcie@66000000: compatible:
>>>> ['socionext,uniphier-pro5-pcie-ep', 'snps,dw-pcie-ep'] is too long
>>>> From schema:
>>>> Documentation/devicetree/bindings/pci/socionext,uniphier-pcie-ep.yaml
>>>>
>>>
>>> This sounds like a problem with the binding rather than the dt file. Is
>>> this not
>>> a designware pci endpoint? Should it be documented in that binding
>>> instead?
>
> In term of the binding, it seems that the current binding doesn't allow descriptions
> that list two compatibles. There is something wrong with the binding.
>
>> Depends. We had one or two similar cases, where we dropped the snps/dw
>> generic compatible, because device was actually quite different and
>> could not match against snps/dw compatible. IOW, if device bound/matched
>> via generic compatible it would be entirely non-operational. Logically I
>> think it is okay to drop the generic compatible. Different question is
>> any ABI break.
>
> In term of the controller, we can add dw general compatible if the more generic
> driver (pcie-designware-plat) works on the controller.
>
> However, the generic driver can't do the initialization what the controller
> needs, so we can add controller-specific compatible only.
> The commit bf2942a8b7c3 ("arm64: tegra: Fix Tegra194 PCIe EP compatible string")
> removes the generic compatible for the same reason.
>
> This patch suggests removing the generic compatible for the former reason,
> though, I might suggest it for the controller reason.
The patch does not explain this, though.
Best regards,
Krzysztof
