Re: [PATCH v3] drm/msm/dp: check hpd_state before push idle pattern at dp_bridge_disable()

[Kuogee Hsieh <quic_khsieh@quicinc.com>]

On 8/10/2022 3:22 PM, Stephen Boyd wrote:
> Quoting Kuogee Hsieh (2022-08-10 12:25:51)
>> diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c
>> index b36f8b6..678289a 100644
>> --- a/drivers/gpu/drm/msm/dp/dp_display.c
>> +++ b/drivers/gpu/drm/msm/dp/dp_display.c
>> @@ -1729,10 +1729,20 @@ void dp_bridge_disable(struct drm_bridge *drm_bridge)
>>          struct msm_dp_bridge *dp_bridge = to_dp_bridge(drm_bridge);
>>          struct msm_dp *dp = dp_bridge->dp_display;
>>          struct dp_display_private *dp_display;
>> +       u32 state;
>>
>>          dp_display = container_of(dp, struct dp_display_private, dp_display);
>>
>> +       mutex_lock(&dp_display->event_mutex);
>> +
>> +       state = dp_display->hpd_state;
>> +       if (state != ST_DISCONNECT_PENDING && state != ST_CONNECTED) {
> It's concerning that we have to check this at all. Are we still
> interjecting into the disable path when the cable is disconnected?

yes,

The problem is not from cable disconnected.

There is a corner case that this function is called at drm shutdown 
(drm_release).

At that time, mainlink is not enabled, hence dp_ctrl_push_idle() will 
cause system crash.



>> +               mutex_unlock(&dp_display->event_mutex);
>> +               return;
>> +       }
>> +
>>          dp_ctrl_push_idle(dp_display->ctrl);
>> +       mutex_unlock(&dp_display->event_mutex);