FWIW msm8226 has a similar problem but there it is enough to provide <&xo_board> instead of the equivalent &rpmcc reference to gcc, which is what was done in a recent patch for msm8226.dtsi.
But in the end clk and mailbox drivers should probably be split so that this dependency cycle can be broken. But I haven't looked into this further yet.
Regards
Luca
Konrad Dybcio <konradybcio@xxxxxxxxxx> schreef op 5 september 2024 14:38:53 CEST:
>On 4.09.2024 8:26 PM, Fabien Parent wrote:
>> Commit 22e4e43484c4 ("arm64: dts: qcom: msm8939: Use mboxes
>> properties for APCS") broke the boot on msm8939 platforms.
>>
>> The issue comes from the SMD driver failing to request the mbox
>> channel because of circular dependencies:
>> 1. rpm -> apcs1_mbox -> rpmcc (RPM_SMD_XO_CLK_SRC) -> rpm.
>> 2. rpm -> apcs1_mbox -> gcc -> rpmcc (RPM_SMD_XO_CLK_SRC) -> rpm
>> 3. rpm -> apcs1_mbox -> apcs2 -> gcc -> rpmcc (RPM_SMD_XO_CLK_SRC) -> rpm
>>
>> To fix this issue let's switch back to using the deprecated
>> qcom,ipc property for the RPM node.
>>
>> Fixes: 22e4e43484c4 ("arm64: dts: qcom: msm8939: Use mboxes properties for APCS")
>> Signed-off-by: Fabien Parent <fabien.parent@xxxxxxxxxx>
>> ---
>
>This is fine as a stopgap solution, Bjorn can we push it in -fixes?
>
>This platform is admittedly not of top importance.. but fixing this
>eventually would be nice
>
>#regzbot introduced: 22e4e43484c4
>
>Konrad
