Re: [PATCH] arm64: dts: qcom: qcs615: disable the CTI device of the camera block

[Konrad Dybcio]

On 6/3/25 5:17 AM, Jie Gan wrote:
> 
> 
> On 5/31/2025 7:05 AM, Konrad Dybcio wrote:
>> On 5/28/25 5:02 AM, Jie Gan wrote:
>>>
>>>
>>> On 5/27/2025 6:41 PM, Konrad Dybcio wrote:
>>>> On 5/27/25 12:32 PM, Jie Gan wrote:
>>>>>
>>>>>
>>>>> On 5/27/2025 6:23 PM, Konrad Dybcio wrote:
>>>>>> On 5/27/25 3:52 AM, Jie Gan wrote:
>>>>>>> Disable the CTI device of the camera block to prevent potential NoC errors
>>>>>>> during AMBA bus device matching.
>>>>>>>
>>>>>>> The clocks for the Qualcomm Debug Subsystem (QDSS) are managed by aoss_qmp
>>>>>>> through a mailbox. However, the camera block resides outside the AP domain,
>>>>>>> meaning its QDSS clock cannot be controlled via aoss_qmp.
>>>>>>
>>>>>> Which clock drives it then?
>>>>>
>>>>> It's qcom,aoss-qmp.
>>>>>
>>>>> clk_prepare->qmp_qdss_clk_prepare
>>>>> https://elixir.bootlin.com/linux/v6.15-rc7/source/drivers/soc/qcom/qcom_aoss.c#L280
>>>>
>>>> I'm confused about this part:
>>>>
>>>>> However, the camera block resides outside the AP domain,
>>>>> meaning its QDSS clock cannot be controlled via aoss_qmp.
>>>>
>>>> Do we need to poke the QMP of another DRV?
>>>
>>> The AOSS has a clock control register for all QDSS clocks. when we vote the qdss clock, the aoss_qmp driver will send a message to AOSS to enable the clock control register, then the clock control register will enable all QDSS clocks.
>>>
>>> The QDSS clock is not a single clock source, it is a term that representing all the clock sources utilized by the QDSS.
>>
>> What I'm trying to ask is, is there any way we could enable that
>> clock from Linux? Can the camera hw turn these on? Maybe we could
>> trick it into enabling them?
> 
> There is a power issue if we keep the debug clock on with a long time.
> 
> We had a discussion with AOP to check if possible to add the debug clock of titan to the QDSS clock list, but they need time to evaluate it.

Changing the firmware is a band-aid solution, as the update will never
reach millions of devices on the market. I'm curious in whether there's
any way (or os-accessible debug register) to manage the necessary clocks
from Linux, as a workaround.

> From Coresight view, what we can do by now is disable it in DT to prevent the unexpected NoC error.

How about something like this:

diff --git a/arch/arm64/boot/dts/qcom/qcs615.dtsi b/arch/arm64/boot/dts/qcom/qcs615.dtsi
index bb8b6c3ebd03..fc2ab750f2cd 100644
--- a/arch/arm64/boot/dts/qcom/qcs615.dtsi
+++ b/arch/arm64/boot/dts/qcom/qcs615.dtsi
@@ -2461,6 +2461,9 @@ cti@6c13000 {
 
                        clocks = <&aoss_qmp>;
                        clock-names = "apb_pclk";
+
+                       /* Not all required clocks can be enabled from the OS */
+                       status = "fail";
                };
 
                cti@6c20000 {

Konrad