[PATCH 0/3] ARM: st: use a correct pwr compatible for stm32mp15

[Patrick DELAUNAY]

Hi,

On 4/26/24 14:51, Rob Herring wrote:
> On Fri, Apr 26, 2024 at 6:42 AM Patrick DELAUNAY
> <patrick.delaunay at foss.st.com> wrote:
>> Hi,
>>
>> On 4/25/24 18:30, Rob Herring wrote:
>>> On Thu, Apr 25, 2024 at 09:48:31AM +0200, Patrick Delaunay wrote:
>>>> This patchset removes the unexpected comma in the PWR compatible
>>>> "st,stm32mp1,pwr-reg" and uses a new compatible "st,stm32mp1-pwr-reg"
>>>> in STM3MP15 device trees.
>>> Why? I don't see any warnings from this. Yes, we wouldn't new cases
>>> following this pattern, but I don't think it is worth maintaining
>>> support for both strings. We're stuck with it. And the only way to
>>> maintain forward compatibility is:
>>
>> Yes, no warning because the compatible string are not yet checked by tools.
> What do you mean? There's a schema for it, so it is checked. I ran the
> tools and there's no warning. If there was a warning, I'd fix the
> tools in this case.


Sorry, I am  no clear


the tools (dts check or check patch) don't check the recommendation for 
compatible name:

     vendor specific string in the form|<vendor>,[<device>-]<usage>|

|   => for me: compatible should have only one comma,
               used as separator between vendor prefix end the device 
identifier.|


But it is normal because existing device tree have a already lot a 
strange compatible


>> I propose this patch to avoid the usage of this compatible for other SoC
>> in STM32MP1 family.
>>
>>
>> I see the invalid compatible string when I reviewed the U-Boot patch to
>> add the PWR node for STM32MP13 family:
>>
>> https://patchwork.ozlabs.org/project/uboot/patch/20240319024534.103299-1-marex@denx.de/
>>
> Perhaps you should add SoC specific compatible string instead.


yes it is a solution.


>
>> So I prefer change the PWR binding before to have the same patch applied
>> on Linux device tree
>>
>>> compatible = "st,stm32mp1-pwr-reg", "st,stm32mp1,pwr-reg";
>>
>> Yes, I will update the SoC patch with you proposal.
> NO! We don't want to support that.


Even mark the old binding deprecated is not acceptable:

  properties:
    compatible:
-    const: st,stm32mp1,pwr-reg
+    oneOf:
+    - const: st,stm32mp1-pwr-reg
+    - items:
+      - const: st,stm32mp1-pwr-reg
+      - const: st,stm32mp1,pwr-reg
+      deprecated: true


I understood.


>
> We have *tons* of examples in DT which don't follow recommended
> patterns and we're stuck with them. This is no different. We can get
> away with changing node names, but that's about it.


Ok,  I am stucked with this compatible for STM32MP15 = "st,stm32mp1,pwr-reg"

and I have no elegant solution to solved it.

So I will change my serie to add a new compatible for STM32MP13

"st,stm32mp13-pwr-reg"


>
> Rob


Regards

Patrick