tegra_pwm_probe() enables runtime PM before requesting the initial resume.
If pm_runtime_resume_and_get() fails, probe returns with runtime PM still
enabled as the managed driver resources are released.

The later probe unwind and remove paths use pm_runtime_force_suspend(),
which disables runtime PM on success but enables it again if the suspend
callback fails. Ignoring that error can therefore also leave runtime PM
enabled during resource cleanup.

Disable runtime PM on the initial resume failure and when force-suspend
fails. Do not disable it a second time after successful force-suspend, and
preserve the original probe error. The failed resume_and_get call has
already balanced its usage count, so it needs no additional put.

This issue was identified during our ongoing static-analysis research while
reviewing kernel code.

Fixes: 3da9b0feaa16 ("pwm: tegra: Add runtime PM and OPP support")
Cc: stable@vger.kernel.org
Assisted-by: OpenAI:GPT-5.6
Co-developed-by: Ijae Kim <ae878000@gmail.com>
Signed-off-by: Ijae Kim <ae878000@gmail.com>
Signed-off-by: Myeonghun Pak <mhun512@gmail.com>
---
 drivers/pwm/pwm-tegra.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/drivers/pwm/pwm-tegra.c b/drivers/pwm/pwm-tegra.c
index 5cdbe120b..55287afcb 100644
--- a/drivers/pwm/pwm-tegra.c
+++ b/drivers/pwm/pwm-tegra.c
@@ -346,8 +346,10 @@ static int tegra_pwm_probe(struct platform_device *pdev)
 
 	pm_runtime_enable(&pdev->dev);
 	ret = pm_runtime_resume_and_get(&pdev->dev);
-	if (ret)
+	if (ret) {
+		pm_runtime_disable(&pdev->dev);
 		return ret;
+	}
 
 	/* Set maximum frequency of the IP */
 	ret = dev_pm_opp_set_rate(&pdev->dev, ULONG_MAX);
@@ -395,7 +397,8 @@ static int tegra_pwm_probe(struct platform_device *pdev)
 	return 0;
 put_pm:
 	pm_runtime_put_sync_suspend(&pdev->dev);
-	pm_runtime_force_suspend(&pdev->dev);
+	if (pm_runtime_force_suspend(&pdev->dev))
+		pm_runtime_disable(&pdev->dev);
 	return ret;
 }
 
@@ -408,7 +411,8 @@ static void tegra_pwm_remove(struct platform_device *pdev)
 
 	reset_control_assert(pc->rst);
 
-	pm_runtime_force_suspend(&pdev->dev);
+	if (pm_runtime_force_suspend(&pdev->dev))
+		pm_runtime_disable(&pdev->dev);
 }
 
 static int __maybe_unused tegra_pwm_runtime_suspend(struct device *dev)

On Sunday, September 13, 2026 1:06 PM Myeonghun Pak wrote:
> tegra_pwm_probe() enables runtime PM before requesting the initial resume.
> If pm_runtime_resume_and_get() fails, probe returns with runtime PM still
> enabled as the managed driver resources are released.
> 
> The later probe unwind and remove paths use pm_runtime_force_suspend(),
> which disables runtime PM on success but enables it again if the suspend
> callback fails. Ignoring that error can therefore also leave runtime PM
> enabled during resource cleanup.
> 
> Disable runtime PM on the initial resume failure and when force-suspend
> fails. Do not disable it a second time after successful force-suspend, and
> preserve the original probe error. The failed resume_and_get call has
> already balanced its usage count, so it needs no additional put.
> 
> This issue was identified during our ongoing static-analysis research while
> reviewing kernel code.
> 
> Fixes: 3da9b0feaa16 ("pwm: tegra: Add runtime PM and OPP support")
> Cc: stable@vger.kernel.org
> Assisted-by: OpenAI:GPT-5.6
> Co-developed-by: Ijae Kim <ae878000@gmail.com>
> Signed-off-by: Ijae Kim <ae878000@gmail.com>
> Signed-off-by: Myeonghun Pak <mhun512@gmail.com>
> ---
>  drivers/pwm/pwm-tegra.c | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/pwm/pwm-tegra.c b/drivers/pwm/pwm-tegra.c
> index 5cdbe120b..55287afcb 100644
> --- a/drivers/pwm/pwm-tegra.c
> +++ b/drivers/pwm/pwm-tegra.c
> @@ -346,8 +346,10 @@ static int tegra_pwm_probe(struct platform_device *pdev)
>  
>  	pm_runtime_enable(&pdev->dev);
>  	ret = pm_runtime_resume_and_get(&pdev->dev);
> -	if (ret)
> +	if (ret) {
> +		pm_runtime_disable(&pdev->dev);
>  		return ret;
> +	}
>  
>  	/* Set maximum frequency of the IP */
>  	ret = dev_pm_opp_set_rate(&pdev->dev, ULONG_MAX);
> @@ -395,7 +397,8 @@ static int tegra_pwm_probe(struct platform_device *pdev)
>  	return 0;
>  put_pm:
>  	pm_runtime_put_sync_suspend(&pdev->dev);
> -	pm_runtime_force_suspend(&pdev->dev);
> +	if (pm_runtime_force_suspend(&pdev->dev))
> +		pm_runtime_disable(&pdev->dev);
>  	return ret;
>  }
>  
> @@ -408,7 +411,8 @@ static void tegra_pwm_remove(struct platform_device *pdev)
>  
>  	reset_control_assert(pc->rst);
>  
> -	pm_runtime_force_suspend(&pdev->dev);
> +	if (pm_runtime_force_suspend(&pdev->dev))
> +		pm_runtime_disable(&pdev->dev);
>  }
>  
>  static int __maybe_unused tegra_pwm_runtime_suspend(struct device *dev)
> 

This driver seems to be good candidate for devm_pm_runtime_enable. Using 
that, I think we can avoid any calls to pm_runtime_disable or 
pm_runtime_force_suspend.

Thanks
Mikko

Thanks, Mikko. I'll use devm_pm_runtime_enable() in v2 and retain
cleanup for devices that remain active.

Best regards,
Myeonghun