Re: [PATCH v3 08/10] drm/rockchip: dw_dp: Add runtime PM support

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Runtime PM initialized without matching active hardware state, causing clocks to remain permanently enabled.
- [High] Using devm_pm_runtime_enable() in a component bind callback ties PM lifecycle incorrectly, causing Use-After-Free and devres leaks.
--

commit 77c03b774daaa80f329daaa791a8229e9e24ad9e
Author: Sebastian Reichel <sebastian.reichel@collabora.com>

drm/rockchip: dw_dp: Add runtime PM support
    
Add support for runtime PM to the Rockchip RK3576/3588 Synopsys
DesignWare DisplayPort driver.

> diff --git a/drivers/gpu/drm/rockchip/dw_dp-rockchip.c b/drivers/gpu/drm/rockchip/dw_dp-rockchip.c
> index 9c53f1d2c29ab..1f53228e56d9f 100644
> --- a/drivers/gpu/drm/rockchip/dw_dp-rockchip.c
> +++ b/drivers/gpu/drm/rockchip/dw_dp-rockchip.c
[ ... ]
> @@ -213,6 +218,12 @@ static int dw_dp_rockchip_bind(struct device *dev, struct device *master, void *
>  	if (IS_ERR(dp->base))
>  		return PTR_ERR(dp->base);
>  
> +	pm_runtime_use_autosuspend(dev);
> +	pm_runtime_set_autosuspend_delay(dev, 500);

[Severity: High]
Since dw_dp_bind() enables the clocks (via devm_clk_get_enabled() when
setting up the base dw_dp struct), shouldn't we call
pm_runtime_set_active() before enabling runtime PM here?

Without it, the PM core assumes the device starts in RPM_SUSPENDED state.
When the device resumes, dw_dp_runtime_resume() will enable the clocks
again (incrementing the enable count to 2). When autosuspend fires,
dw_dp_runtime_suspend() will disable them (decrementing the count back to
1), meaning the hardware clocks are never physically disabled.

> +	ret = devm_pm_runtime_enable(dev);

[Severity: High]
Does using devm_pm_runtime_enable() here tie the PM lifecycle to the
component platform device instead of the DRM master?

The rockchip_dw_dp structure is allocated using drmm_kzalloc() earlier in
this function, which binds its lifetime to the DRM master device. However,
devm_pm_runtime_enable() attaches the devres action to the component device,
whose lifetime spans the full probe/remove cycle.

If the DRM master is unbound, dw_dp_rockchip_unbind() will execute but it
does not disable PM runtime or flush the autosuspend timer. The DRM core
will then free the drmm-managed memory.

Could an autosuspend timer firing after unbind cause a use-after-free
regression in dw_dp_rockchip_runtime_suspend() when it tries to access the
freed dp pointer? Repeated bind/unbind cycles might also endlessly leak
devres actions on the platform device.

> +	if (ret)
> +		return dev_err_probe(dev, ret, "failed to enable runtime PM\n");
> +
>  	connector = drm_bridge_connector_init(drm_dev, encoder);

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260612-synopsys-dw-dp-improvements-v3-0-dc61e6352508@collabora.com?part=8