Re: [PATCH v3 05/10] dt-bindings: clock: ipq9574: Rename NSS CC source clocks to drop rate

[Luo Jie]

On 7/18/2025 5:28 PM, Konrad Dybcio wrote:
> On 7/18/25 11:12 AM, Luo Jie wrote:
> > On 7/11/2025 8:15 PM, Konrad Dybcio wrote:
> > > On 7/11/25 12:54 AM, Rob Herring wrote:
> > > > On Thu, Jul 10, 2025 at 08:28:13PM +0800, Luo Jie wrote:
> > > > > Drop the clock rate suffix from the NSS Clock Controller clock names for
> > > > > PPE and NSS clocks. A generic name allows for easier extension of support
> > > > > to additional SoCs that utilize same hardware design.
> > > >
> > > > This is an ABI change. You must state that here and provide a reason the
> > > > change is okay (assuming it is). Otherwise, you are stuck with the name
> > > > even if not optimal.
> > >
> > > The reason here seems to be simplifying the YAML.. which is not a good
> > > reason really..
> > >
> > > I would instead suggest keeping the clocks list as-is for ipq9574 (this
> > > existing case), whereas improving it for any new additions
> > >
> > > Konrad
> >
> > Thanks Rob and Konrad for the comments.
> >
> > "nss_1200" and "nss" refer to the same clock pin on different SoC.
> > As per Krzystof's previous comment on V2, including the frequency
> > as a suffix in the clock name is not required, since only the
> > frequencies vary across different IPQ SoCs, while the source clock
> > pins for 'PPE' and 'NSS' clocks are the same. Hence this ABI change
> > was deemed necessary.
> >
> > By removing the frequency suffix, the device tree bindings becomes
> > more flexible and easier to extend for supporting new hardware
> > variants in the future.
> >
> > Impact due to this ABI change: The NSS clock controller node is only
> > enabled for the IPQ9574 DTS. In this patch series, the corresponding
> > DTS changes for IPQ9574 are also included to align with this ABI
> > change.
>
> The point of an ABI is to keep exposing the same interface without
> any change requirements, i.e. if a customer ships the DT from
> torvalds/master in firmware and is not willing to update it, they
> can no longer update the kernel without a workaround.
>
> > Please let me know if further clarification or adjustments are needed.
>
> What we're asking for is that you don't alter the name on the
> existing platform, but use a no-suffix version for the ones you
> introduce going forward
>
> i.e. (pseudo-YAML)
>
> if:
>    properties:
>      compatible:
>        - const: qcom,ipq9574-nsscc
> then:
>    properties:
>      clock-names:
>        items:
>          - clockname_1200
> else:
>    properties:
>      clock-names:
>        items:
>          - clockname # no suffix
>
> Konrad

We had adopted this proposal in version 2 previously, but as noted in
the discussion linked below, Krzysztof had suggested to avoid using the
clock rate in the clock names when defining the constraints for them.
However I do agree that we should keep the interface for IPQ9574
unchanged and instead use a generic clock name to support the newer
SoCs.

https://lore.kernel.org/all/20250701-optimistic-esoteric-swallow-d93fc6@krzk-bin/

Request Krzysztof to provide his comments as well, on whether we can
follow your suggested approach to avoid breaking ABI for IPQ9574.