Re: [PATCH 1/4] dt-bindings: arm: qcom: document qcom,msm-id and qcom,board-id

[Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>]

On 10/06/2022 18:33, Rob Herring wrote:
> On Tue, Jun 07, 2022 at 01:15:51PM +0200, Krzysztof Kozlowski wrote:
>> On 05/06/2022 17:07, Rob Herring wrote:
>>> On Sun, May 29, 2022 at 10:26:26PM +0200, Krzysztof Kozlowski wrote:
>>>> The top level qcom,msm-id and qcom,board-id properties are utilized by
>>>> bootloaders on Qualcomm MSM platforms to determine which device tree
>>>> should be used and passed to the kernel.
>>>>
>>>> The commit b32e592d3c28 ("devicetree: bindings: Document qcom board
>>>> compatible format") from 2015 was a consensus during discussion about
>>>> upstreaming qcom,msm-id and qcom,board-id fields.  There are however still
>>>> problems with that consensus:
>>>> 1. It was reached 7 years ago but it turned out its implementation did
>>>>    not reach all possible products.
>>>>
>>>> 2. Initially additional tool (dtbTool) was needed for parsing these
>>>>    fields to create a QCDT image consisting of multiple DTBs, later the
>>>>    bootloaders were improved and they use these qcom,msm-id and
>>>>    qcom,board-id properties directly.
>>>>
>>>> 3. Extracting relevant information from the board compatible requires
>>>>    this additional tool (dtbTool), which makes the build process more
>>>>    complicated and not easily reproducible (DTBs are modified after the
>>>>    kernel build).
>>>>
>>>> 4. Some versions of Qualcomm bootloaders expect these properties even
>>>>    when booting with a single DTB.  The community is stuck with these
>>>>    bootloaders thus they require properties in the DTBs.
>>>>
>>>> Since several upstreamed Qualcomm SoC-based boards require these
>>>> properties to properly boot and the properties are reportedly used by
>>>> bootloaders, document them.
>>>
>>> My primary issue here is accepting this will be an endorsement for 
>>> other vendors doing something similar. I'm not against an ID 
>>> property(ies) in the root node, but would rather see something common 
>>> if we do anything.
>>
>> Hi Rob,
>>
>> A more common approach was merged back in 2015 - encoding this ID
>> information in the board compatibles. If I understood previous
>> discussion correctly, this common method was later used by Qualcomm DTB
>> post-processing tool. At least for some of the cases.
>>
>> Other cases (several Qualcomm boards from different vendors) still use
>> these ID properties. It even turns out they use it differently between
>> vendors (e.g. Xiaomi vs OnePlus).
>>
>> Important arguments for documenting these properties:
>> 1. These ID properties are already on released boards where changing
>> bootloader is non-trivial or even not possible. It will not be possible
>> to remove these properties, without seriously affecting the community
>> working with them.
> 
> Accepting things because they are already in use is also not a path we 
> want to go down. If it's the color of the bike shed, then fine.
> 
>> 2. According to Konrad [1] (second paragraph), newer chipsets (starting
>> with sm8350 released in 2021) do not use these properties. These newer
>> DTS do not have them.
>>
>> Considering 1+2 above, maybe let's document these properties as
>> compatible? Would that solve your point of "endorsement for other vendors"?
> 
> What do you mean? Only allow them for certain root compatible strings? I 
> suppose that would be okay by me. It would also be useful documentation 
> of where they are needed.

Bah, I wrote something else than I had in mind. So one more try:

Considering 1+2 above, maybe let's document these properties as
*deprecated*? Would that solve your point of "endorsement for other
vendors"?

However the idea to restrict them per-compatible, is also nice. Although
I cannot guarantee the list will not grow for older SoCs.


Best regards,
Krzysztof