Re: [PATCH v2] drm/msm/dp: stop event kernel thread when DP unbind

[Stephen Boyd]

Quoting Kuogee Hsieh (2022-04-13 14:04:25)
> 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().
>
> Changes in v2:
> -- start event thread at dp_display_bind()
>
> Fixes: e91e3065a806 ("drm/msm/dp: Add DP compliance tests on Snapdragon Chipsets")
> Signed-off-by: Kuogee Hsieh <quic_khsieh@xxxxxxxxxxx>
> Reported-by: Dmitry Baryshkov <dmitry.baryshkov@xxxxxxxxxx>
> ---
>  drivers/gpu/drm/msm/dp/dp_display.c | 40 +++++++++++++++++++++++++++----------
>  1 file changed, 30 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c
> index 01453db..943e4f1 100644
> --- a/drivers/gpu/drm/msm/dp/dp_display.c
> +++ b/drivers/gpu/drm/msm/dp/dp_display.c
> @@ -113,6 +113,7 @@ struct dp_display_private {
>         u32 hpd_state;
>         u32 event_pndx;
>         u32 event_gndx;
> +       struct task_struct *ev_tsk;
>         struct dp_event event_list[DP_EVENT_Q_MAX];
>         spinlock_t event_lock;
>
> @@ -230,6 +231,31 @@ void dp_display_signal_audio_complete(struct msm_dp *dp_display)
>         complete_all(&dp->audio_comp);
>  }
>
> +static int hpd_event_thread(void *data);

Is there a reason why this is needed vs. defining the kthread start
function after hpd_event_thread()?

> +
> +static void dp_hpd_event_setup(struct dp_display_private *dp_priv)

Maybe dp_hpd_event_thread_start()?

> +{
> +       init_waitqueue_head(&dp_priv->event_q);
> +       spin_lock_init(&dp_priv->event_lock);
> +
> +       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");

Can we return an error from this function?

> +}
> +
> +static void dp_hpd_event_stop(struct dp_display_private *dp_priv)

Maybe dp_hpd_event_thread_stop()?

> +{
> +       if (IS_ERR(dp_priv->ev_tsk))
> +               return;

If we handled the error then this check becomes impossible.

> +
> +       kthread_stop(dp_priv->ev_tsk);
> +
> +       /* reset event q to empty */
> +       dp_priv->event_gndx = 0;
> +       dp_priv->event_pndx = 0;
> +}
> +
>  static int dp_display_bind(struct device *dev, struct device *master,
>                            void *data)
>  {
> @@ -269,6 +295,7 @@ static int dp_display_bind(struct device *dev, struct device *master,
>         if (rc)
>                 DRM_ERROR("Audio registration Dp failed\n");
>
> +       dp_hpd_event_setup(dp); /* start event thread */

The comment is useless, please remove.

>  end:
>         return rc;
>  }
> @@ -280,6 +307,8 @@ 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);

Is the disable_irq() necessary? It would be nicer to silence the
hardware and remove the disable_irq() so that we can reason about the
code assuming the irq is always enabled after it is requested.

> +       dp_hpd_event_stop(dp); /* stop event thread */
>         dp_power_client_deinit(dp->power);
>         dp_aux_unregister(dp->aux);
>         priv->dp[dp->id] = NULL;