Re: [PATCH] drm/msm/dp: stop event kernel thread when DP unbind

[Abhinav Kumar <quic_abhinavk@quicinc.com>]

On 4/11/2022 5:22 PM, Dmitry Baryshkov wrote:
> On 12/04/2022 03:21, Stephen Boyd wrote:
>> Quoting Kuogee Hsieh (2022-04-11 17:08:49)
>>> Current DP driver implementation, event thread is kept running
>>> after DP display is unbind. This patch fix this problem by disabling
>>> DP irq and stop event thread to exit gracefully at dp_display_unbind().
>>>
>>> Fixes: e91e3065a806 ("drm/msm/dp: Add DP compliance tests on 
>>> Snapdragon Chipsets")
>>> Signed-off-by: Kuogee Hsieh <quic_khsieh@quicinc.com>
>>
>> Should add a Reported-by tag from Dmitry here.
>>
>>> ---
>>>   drivers/gpu/drm/msm/dp/dp_display.c | 24 ++++++++++++++++++++++--
>>>   1 file changed, 22 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/msm/dp/dp_display.c 
>>> b/drivers/gpu/drm/msm/dp/dp_display.c
>>> index 01453db..fa1ef8e 100644
>>> --- a/drivers/gpu/drm/msm/dp/dp_display.c
>>> +++ b/drivers/gpu/drm/msm/dp/dp_display.c
>>> @@ -273,6 +274,8 @@ static int dp_display_bind(struct device *dev, 
>>> struct device *master,
>>>          return rc;
>>>   }
>>>
>>> +static void dp_hpd_event_stop(struct dp_display_private *dp_priv);
>>
>> Why can't the function be defined here?
>>
>>> +
>>>   static void dp_display_unbind(struct device *dev, struct device 
>>> *master,
>>>                                void *data)
>>>   {
>>> @@ -280,6 +283,8 @@ static void dp_display_unbind(struct device *dev, 
>>> struct device *master,
>>>          struct drm_device *drm = dev_get_drvdata(master);
>>>          struct msm_drm_private *priv = drm->dev_private;
>>>
>>> +       disable_irq(dp->irq);
>>> +       dp_hpd_event_stop(dp);
>>>          dp_power_client_deinit(dp->power);
>>>          dp_aux_unregister(dp->aux);
>>>          priv->dp[dp->id] = NULL;
>>> @@ -1054,7 +1059,7 @@ static int hpd_event_thread(void *data)
>>>
>>>          dp_priv = (struct dp_display_private *)data;
>>>
>>> -       while (1) {
>>> +       while (!kthread_should_stop()) {
>>>                  if (timeout_mode) {
>>>                          wait_event_timeout(dp_priv->event_q,
>>>                                  (dp_priv->event_pndx == 
>>> dp_priv->event_gndx),
>>> @@ -1137,7 +1142,22 @@ static void dp_hpd_event_setup(struct 
>>> dp_display_private *dp_priv)
>>>          init_waitqueue_head(&dp_priv->event_q);
>>>          spin_lock_init(&dp_priv->event_lock);
>>>
>>> -       kthread_run(hpd_event_thread, dp_priv, "dp_hpd_handler");
>>> +       dp_priv->ev_tsk = kthread_run(hpd_event_thread, dp_priv, 
>>> "dp_hpd_handler");
>>> +
>>> +       if (IS_ERR(dp_priv->ev_tsk))
>>> +               DRM_ERROR("failed to create DP event thread\n");
>>
>> Why can't we error out? Why can't this kthread be started in probe?
> 
> Just my 2c. I don't think starting it in probe is a good idea. The 
> driver uses components, so, in my opinion, the thread should be started 
> from bind and stopped in unbind.

Yes, I also agree it should be started in bind and stopped in unbind.

> 
>>
>>> +}
>>> +
>>> +static void dp_hpd_event_stop(struct dp_display_private *dp_priv)
>>> +{
>>> +       if (IS_ERR(dp_priv->ev_tsk))
>>> +               return;
>>> +
>>> +       kthread_stop(dp_priv->ev_tsk);
>>> +
>>> +       /* reset event q to empty */
>>> +       dp_priv->event_gndx = 0;
>>> +       dp_priv->event_pndx = 0;
>>>   }
>>>
> 
>