Re: [PATCH v4 3/5] arm64: dts: qcom: kaanapali: 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 Trace NOC from an AMBA device to a platform device exposes a double clock disable bug during driver unbind due to mixing devm-managed clocks with manual PM runtime callbacks.
--

commit dec424aa5c224444a6c7946ea25210a9094f968f
Author: Jie Gan <jie.gan@oss.qualcomm.com>

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

This commit switches the traceNoC node from an AMBA device to a platform
device to fix an issue where the device is stuck in deferred probe due to
missing CID registers.

> diff --git a/arch/arm64/boot/dts/qcom/kaanapali.dtsi b/arch/arm64/boot/dts/qcom/kaanapali.dtsi
> --- a/arch/arm64/boot/dts/qcom/kaanapali.dtsi
> +++ b/arch/arm64/boot/dts/qcom/kaanapali.dtsi
[ ... ]
> @@ -5004,7 +5004,7 @@ tpdm_pcie_rscc_out: endpoint {
>  		};
>  
>  		tn@111b8000 {
> -			compatible = "qcom,coresight-tnoc", "arm,primecell";
> +			compatible = "qcom,coresight-agtnoc";

[Severity: High]
This isn't a bug introduced by this patch, but does migrating this node to
the platform driver expose a double clock disable issue during unbind?

When the driver binds via the platform device, if a user unbinds the driver
via sysfs while the device is runtime suspended, it seems to trigger an
unbalanced clk_disable_unprepare() call.

Looking at tnoc_platform_remove() in drivers/hwtracing/coresight/coresight-tnoc.c:

static void tnoc_platform_remove(struct platform_device *pdev)
{
	struct trace_noc_drvdata *drvdata = platform_get_drvdata(pdev);

	coresight_unregister(drvdata->csdev);
	pm_runtime_disable(&pdev->dev);
	if (drvdata->atid > 0)
		coresight_trace_id_put_system_id(drvdata->atid);
}

Since tnoc_platform_remove() calls pm_runtime_disable() without first
waking the device (e.g. via pm_runtime_get_sync()), the device remains
suspended and its clock remains disabled by tnoc_runtime_suspend(). 
    
When the driver finishes unbinding, the devres framework executes the
cleanup action for devm_clk_get_optional_enabled(), which calls
clk_disable_unprepare() a second time.

Could this lead to a clock subsystem underflow?

>  			reg = <0x0 0x111b8000 0x0 0x4200>;
>  
>  			clocks = <&aoss_qmp>;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260701-fix-tracenoc-probe-issue-v4-0-aefab449a470@oss.qualcomm.com?part=3