On 03/04/2023 10:43, AngeloGioacchino Del Regno wrote:
> Il 03/04/23 09:52, Stephen Rothwell ha scritto:
>> Hi all,
>>
>> After merging the mediatek tree, today's linux-next build (arm64
>> defconfig) failed like this:
>>
>> Error: arch/arm64/boot/dts/mediatek/mt6795.dtsi:647.21-22 syntax error
>> FATAL ERROR: Unable to parse input tree
>>
>> Maybe caused by commit
>>
>> a7c7f1fe2fde ("arm64: dts: mediatek: mt6795: Add MMSYS node for multimedia clocks")
>>
>> but I don't know how.
>>
>> I have reverted that commit (and the following 2) for today.
>>
>
> Matthias, this issue happens because you didn't apply 07/17 [1] from the
> 6795/xperia-m5 series...
>
> [1]:
> https://lore.kernel.org/all/20230327083647.22017-8-angelogioacchino.delregno@xxxxxxxxxxxxx/
Bindings are supposed to go via subsystem (driver) tree and only via ARM
SoC if the former is not interested in them.
I don't think the issue happens because Matthias did not apply binding
patch. Nothing in the cover letter or the patch mentioned such
dependency, so why he should pick up unrelated patch?
The submitter should clearly mark such dependencies or make patches
independent. It's nothing new, for clock/reset drivers (driver + binding
+ DTS) uncoupling the dependency is pretty often a requirement or
maintainers need to coordinate with cross-tree pulls (and DTS cannot get
pulled into the driver tree/branch).
>
> That contains the mediatek,mt6795-gce.h header that has the definitions used in
> a7c7f1fe2fde ("arm64: dts: mediatek: mt6795: Add MMSYS node for multimedia clocks")
> ....and this is the reason of the syntax error.
>
> Does this commit have to go through devicetree trees?
Bindings go via subsystem maintainer, so here rather mailbox. Not DT tree.
> I'm adding Krzysztof and Rob to the loop for this.
I guess now, the easiest would be if Matthias takes the binding change
with mailbox Ack. Assuming there are no conflicts there...
Best regards,
Krzysztof
