Re: [PATCH v11 3/7] dt-bindings: PCI: qcom: Use sdx55 reg description for ipq9574

[Varadarajan Narayanan <quic_varada@quicinc.com>]

On Mon, Mar 10, 2025 at 12:37:28PM +0100, Krzysztof Kozlowski wrote:
> On 10/03/2025 08:44, Varadarajan Narayanan wrote:
> > On Thu, Mar 06, 2025 at 01:06:13PM +0100, Krzysztof Kozlowski wrote:
> >> On 06/03/2025 12:52, Krzysztof Kozlowski wrote:
> >>> On 20/02/2025 10:42, Varadarajan Narayanan wrote:
> >>>> All DT entries except "reg" is similar between ipq5332 and ipq9574. ipq9574
> >>>> has 5 registers while ipq5332 has 6. MHI is the additional (i.e. sixth
> >>>> entry). Since this matches with the sdx55's "reg" definition which allows
> >>>> for 5 or 6 registers, combine ipq9574 with sdx55.
> >>>>
> >>>> This change is to prepare ipq9574 to be used as ipq5332's fallback
> >>>> compatible.
> >>>>
> >>>> Acked-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
> >>>> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> >>>
> >>> Unreviewed.
> >>>
> >>>> Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
> >>>> ---
> >>>> v8: Add 'Reviewed-by: Krzysztof Kozlowski'
> >>>> ---
> >>>>  Documentation/devicetree/bindings/pci/qcom,pcie.yaml | 2 +-
> >>>>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/Documentation/devicetree/bindings/pci/qcom,pcie.yaml b/Documentation/devicetree/bindings/pci/qcom,pcie.yaml
> >>>> index 7235d6554cfb..4b4927178abc 100644
> >>>> --- a/Documentation/devicetree/bindings/pci/qcom,pcie.yaml
> >>>> +++ b/Documentation/devicetree/bindings/pci/qcom,pcie.yaml
> >>>> @@ -169,7 +169,6 @@ allOf:
> >>>>              enum:
> >>>>                - qcom,pcie-ipq6018
> >>>>                - qcom,pcie-ipq8074-gen3
> >>>> -              - qcom,pcie-ipq9574
> >>>
> >>> Why you did not explain that you are going to affect users of DTS?
> >>>
> >>> NAK
> >
> > Sorry for not explicitly calling this out. I thought that would be seen from the
> > following DTS related patches.
> >
> >> I did not connect the dots, but I pointed out that you break users and
> >> your DTS is wrong:
> >> https://lore.kernel.org/all/f7551daa-cce5-47b3-873f-21b9c5026ed2@kernel.org/
> >>
> >> so you should come back with questions to clarify what to do, not keep
> >> pushing this incorrect patchset.
> >>
> >> My bad, I should really have zero trust.
> >
> > It looks like it is not possible to have ipq9574 as fallback (for ipq5332)
> > without making changes to ipq9574 since the "reg" constraint is different
> > between the two. And this in turn would break the ABI w.r.t. ipq9574.
>
> I don't get why this is not possible. You have one list for ipq9574 and
> existing compatible devices, and you add second list for new device.
>
> ... or you just keep existing order. Why you need to keep changing order
> every time you add new device?

Presently, sdx55 and ipq9574 have the following reg/reg-names constraints.

	compatible	| qcom,pcie-sdx55	| qcom,pcie-ipq9574
	----------------+-----------------------+------------------
        reg	minItems| 5			| 5
		maxItems| 6			| 5
	----------------+-----------------------+------------------
        reg-names	|			|
		minItems| 5			| 5
	----------------+-----------------------+------------------
		maxItems|			| 5 (6 for ipq5332)
	----------------+-----------------------+------------------
		items	|			|
			| parf			| dbi
			| dbi			| elbi
			| elbi			| atu
			| atu			| parf
			| config		| config
			| mhi			| (add mhi for ipq5332)
	----------------+-----------------------+------------------

To make ipq9574 as fallback for ipq5332, have to add "mhi" to reg-names of
ipq9574. Once I add that, the sdx55 and ipq9574 is the same list but in
different order.

If this would not be considered as duplication of the same constraint, then I
can club ipq5332 with ipq9574.

If this would be considered as duplication, then sdx55 and ipq9574 would have to
use the same reg-names list and sdx55 or ipq9574 reg-names order would change.

> > To overcome this, two approaches seem to be availabe
> >
> > 	1. Document that ipq9574 is impacted and rework these patches to
> > 	   minimize the impact as much as possible
>
> What impact? What is the reason to impact ipq9574? What is the actual issue?

By impact, I meant the change in the reg-names order as mentioned above (for
considered as duplication).

> > 		(or)
> >
> > 	2. Handle ipq5332 as a separate compatible (without fallback) and reuse
> > 	   the constraints of sdx55 for "reg" and ipq9574 for the others (like
> > 	   clock etc.). This approach will also have to revert [1], as it
> > 	   assumes ipq9574 as fallback.
> >
> > Please advice which of the above would be appropriate. If there is a better 3rd
> > alternative please let me know, will align with that approach.
>
> Keep existing order. Why every time we see new device, it comes up with
> a different order?

Will be able to do that based on the answer to 'duplication' question and how to
handle that.

	if (adding mhi to ipq9574 reg-names != duplication)

		/* Keep existing order */

		* Append "mhi" to ipq9574
		* use ipq9574 reg-names order for ipq5332

	else
		* combine ipq9574 & sdx55 reg-names

		if (use sdx55 reg-names order)

			/* patchset v11 is using this approach */

			* change ipq9574
			* follow the same for ipq5332

		else if (use ipq9574 order)

			* change sdx55
			* follow the same for ipq5332

Please advice.

Thanks
Varada