Re: [PATCH v4 6/8] drm/msm: dp: handle the IRQ_HPD events reported by USB-C

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:

New issues:
- [High] Initial connection and replug events are incorrectly bypassed if accompanied by an IRQ_HPD status, leaving the display disconnected.

Pre-existing issues:
- [High] `msm_dp_irq_hpd_handle` modifies shared state without acquiring `dp->plugged_lock`, leading to unprotected concurrent modification of DPCD and link state.
--

commit 48a78358ae98020ff3993aab3250dd8c32cc6835
Author: Dmitry Baryshkov <[email protected]>

drm/msm: dp: handle the IRQ_HPD events reported by USB-C

This patch adds support for tracking and handling IRQ_HPD events
delivered over out-of-band events (such as USB-C AltMode handlers)
in the MSM DisplayPort driver.

> diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c
> index c0f6a8ff9b99d..743d8fa40dc68 100644
> --- a/drivers/gpu/drm/msm/dp/dp_display.c
> +++ b/drivers/gpu/drm/msm/dp/dp_display.c
> @@ -1533,11 +1533,12 @@ void msm_dp_bridge_hpd_notify(struct drm_bridge *bridge,
>  		   msm_dp_display->connector_type, hpd_link_status, status);
>  
>  	if (status == connector_status_connected) {
> -		if (hpd_link_status == ISR_HPD_REPLUG_COUNT) {
> +		if (hpd_link_status == ISR_IRQ_HPD_PULSE_COUNT ||
> +		    extra_status == DRM_CONNECTOR_DP_IRQ_HPD) {
> +			msm_dp_irq_hpd_handle(dp);

[Severity: High]
This is a pre-existing issue, but does msm_dp_irq_hpd_handle() need to
acquire the dp->plugged_lock?

Since msm_dp_irq_hpd_handle() can call msm_dp_display_process_hpd_high(),
which reads from DPCD and modifies shared driver state (like
dp->link->lttpr_count, dp->panel->dpcd, and dp->audio_supported), it seems
like it could race. Other HPD handlers like msm_dp_hpd_plug_handle() use
guard(mutex)(&dp->plugged_lock) to safely synchronize access.

With out-of-band events processed via workqueues, can this function
run concurrently with other HPD handlers or msm_dp_bridge_detect() and
corrupt the internal representation of the hardware configuration?

[Severity: High]
If an out-of-band handler reports an initial connection with both hpd_state=1
and hpd_irq=1 simultaneously, does this new ordering bypass the required plug
sequence?

By placing the IRQ_HPD check at the top of the if/else chain, if
extra_status == DRM_CONNECTOR_DP_IRQ_HPD, the code will execute
msm_dp_irq_hpd_handle() and skip msm_dp_hpd_plug_handle().

If msm_dp_hpd_plug_handle() is skipped, dp->plugged remains false, PHY
initialization (msm_dp_display_host_phy_init) never occurs, and
msm_dp_aux_enable_xfers() is never called. Won't the subsequent AUX reads
within msm_dp_irq_hpd_handle() fail with -ENXIO, aborting the initialization
and leaving the display disconnected?

> +		} else if (hpd_link_status == ISR_HPD_REPLUG_COUNT) {
>  			msm_dp_hpd_unplug_handle(dp);
>  			msm_dp_hpd_plug_handle(dp);
> -		} else if (hpd_link_status == ISR_IRQ_HPD_PULSE_COUNT) {
> -			msm_dp_irq_hpd_handle(dp);
>  		} else {
>  			msm_dp_hpd_plug_handle(dp);
>  		}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=6

_______________________________________________
linux-amlogic mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-amlogic