Re: [PATCH v4 4/5] arm64: dts: qcom: sm8750: fix traceNoC probe issue

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Switching the traceNoC device from an AMBA binding to a platform binding exposes a pre-existing double-disable bug for the `apb_pclk` clock during driver unbind if the device is runtime-suspended.
--

commit c18b0cbf674e991c292d9a2d479c037f1c2eafdf
Author: Jie Gan <jie.gan@oss.qualcomm.com>

arm64: dts: qcom: sm8750: fix traceNoC probe issue

This commit changes the traceNoC device binding from an AMBA compatible
string to a standalone platform compatible string. This allows the device
to probe successfully on hardware that does not expose a readable primecell
ID register.

> diff --git a/arch/arm64/boot/dts/qcom/sm8750.dtsi b/arch/arm64/boot/dts/qcom/sm8750.dtsi
> index fafed417c66fc..d58483f9f93a3 100644
> --- a/arch/arm64/boot/dts/qcom/sm8750.dtsi
> +++ b/arch/arm64/boot/dts/qcom/sm8750.dtsi
> @@ -4687,7 +4687,7 @@ tpdm_rdpm_cmb2_out: endpoint {
>  		};
>  
>  		tn@109ab000 {
> -			compatible = "qcom,coresight-tnoc", "arm,primecell";
> +			compatible = "qcom,coresight-agtnoc";

[Severity: High]
This is a pre-existing issue, but does switching this device to a platform
binding expose a double-disable bug for the apb_pclk clock during driver
unbind if the device is runtime-suspended?

When binding as a non-AMBA device, the clock is acquired and enabled via a
devm-managed API which registers a devres action to disable it on unbind:

drivers/hwtracing/coresight/coresight-core.c:coresight_get_enable_clocks() {
    ...
    *pclk = devm_clk_get_optional_enabled(dev, "apb_pclk");
    ...
}

If the device is unbound while idle, the driver core will suspend it before
removal. The runtime PM suspend callback disables the clock:

drivers/hwtracing/coresight/coresight-tnoc.c:tnoc_runtime_suspend() {
    ...
    clk_disable_unprepare(drvdata->pclk);
    ...
}

However, the remove callback doesn't call pm_runtime_get_sync() to resume the
device before disabling runtime PM:

drivers/hwtracing/coresight/coresight-tnoc.c:tnoc_platform_remove() {
    ...
    coresight_unregister(drvdata->csdev);
    pm_runtime_disable(&pdev->dev);
    ...
}

This leaves the clock disabled by runtime PM. When the devres cleanup action
for devm_clk_get_optional_enabled() runs, it unconditionally executes
clk_disable_unprepare() again.

Could this underflow the enable_count and turn off the clock while other
devices sharing the apb_pclk are still using it?

>  			reg = <0x0 0x109ab000 0x0 0x4200>;
>  
>  			clocks = <&aoss_qmp>;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260701-fix-tracenoc-probe-issue-v4-0-aefab449a470@oss.qualcomm.com?part=4