Re: [PATCH v11 08/15] drm/panfrost: Move all DRM device initialisation into device_init()

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] pm_runtime: usage_count corruption on early probe failure
--

commit 83a4c1b15f0bb7a999d8fbc11545b9b267d7fbdf
Author: Adrián Larumbe <adrian.larumbe@collabora.com>

drm/panfrost: Move all DRM device initialisation into device_init()

This commit moves DRM device registration and runtime PM setup from
panfrost_probe() into panfrost_device_init(), and consolidates the
cleanup path into panfrost_device_fini().

> diff --git a/drivers/gpu/drm/panfrost/panfrost_device.c b/drivers/gpu/drm/panfrost/panfrost_device.c
> index 9f2b1967a3984..b8094c0d12528 100644
> --- a/drivers/gpu/drm/panfrost/panfrost_device.c
> +++ b/drivers/gpu/drm/panfrost/panfrost_device.c

[ ... ]

> @@ -311,11 +343,22 @@ int panfrost_device_init(struct panfrost_device *pfdev)
>  	panfrost_reset_fini(pfdev);
>  out_pm_domain:
>  	panfrost_pm_domain_fini(pfdev);
> +
> +	if (pm_runtime_active(pfdev->base.dev)) {
> +		pm_runtime_set_suspended(pfdev->base.dev);
> +		pm_runtime_put_noidle(pfdev->base.dev);
> +	}
> +

[Severity: High]
Can this catch-all conditional block cause a PM runtime usage_count corruption
on early probe failures?

If sysfs power/control is set to 'on' (e.g., via udev rules before probe),
pm_runtime_forbid() increments the usage_count to 1.

If the driver then fails early in panfrost_device_init() (such as a probe
deferral during panfrost_clk_init()), pm_runtime_enable() has not been called
yet. This leaves dev->power.disable_depth at 1, which inherently causes
pm_runtime_active() to return true.

This would cause the error path to incorrectly assume the driver acquired the
PM reference, calling pm_runtime_put_noidle() and erroneously decrementing
the usage_count to 0.

When userspace later sets power/control to 'auto', pm_runtime_allow() would
decrement the usage_count again, underflowing it to -1 and permanently breaking
runtime PM for the device.

>  	return err;
>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260925-claude-fixes-v11-0-0dbf5a58e7ce@collabora.com?part=8