Re: [PATCH v6 5/6] arm64: dts: qcom: Add device tree for Nord Ride board

[Konrad Dybcio]

On 7/20/26 4:00 PM, Shawn Guo wrote:
> On Mon, Jul 20, 2026 at 10:40:09AM +0200, Konrad Dybcio wrote:
>> On 7/20/26 6:20 AM, Shawn Guo wrote:
>>> On Thu, Jul 16, 2026 at 07:14:56PM +0200, Konrad Dybcio wrote:
>>>> On 7/16/26 4:26 PM, Shawn Guo wrote:
>>>>> From: Deepti Jaggi <deepti.jaggi@xxxxxxxxxxxxxxxx>
>>>>>
>>>>> Add initial device tree for the Qualcomm SA8797P Ride reference board,
>>>>> which is built on Nord GearVM variant.
>>
>> [...]
>>
>>>>> +&scmi15 {
>>>>> +	status = "okay";
>>>>> +};
>>>>> +
>>>>> +&scmi23 {
>>>>> +	status = "okay";
>>>>> +};
>>>>
>>>> Why only these ones? Can all of the SCMI servers be enabled?
>>>
>>> These are the ones that have actual/enabled client/consumer devices,
>>> e.g. scmi3 for ufs, scmi11 for uart, scmi23 for thermal sensors.
>>
>> We normally refrain from enabling resources in the DT only because the
>> OS (really just Linux) doesn't have the necessary support to cope with
>> the ""complete"" system description. In this case, there's a firmware
>> layer drives everything and handles on/off requests gracefully
>> (hopefully).
>>
>>> We can enable all the SCMI servers, but I'm afraid that cause kernel
>>> to map shared memory and issue SMC calls to discover protocols (like
>>> power domains, DVFS, resets) that have no consumers, which might waste
>>> a small amount of memory and boot time.
>>
>> Please give it a shot and see if there's any immediate
>> problems.
> 
> Sorry! I have to take "We can enable all the SCMI servers" back. I just
> tested. It seems that NHLOS ABL doesn't like the idea, and it
> fails/refuses to boot Linux Kernel when we enable all SCMI instances in
> DT.
> 
> I can try to figure out the cause by working with Auto/firmware team
> internally, but that likely takes some time. Can we address this later
> as a separate topic, so that the whole series doesn't get blocked
> for v7.3 merge window?

Yes

Konrad