Message-ID: <CAA8EJprxf5RBfuiJVJsfnb7buC9Mbxr=U7VSaPRc1+OMJcBFZg@mail.gmail.com>
Date:   Sat, 16 Sep 2023 04:07:58 +0300
From:   Dmitry Baryshkov <dmitry.baryshkov@...aro.org>
To:     Kuogee Hsieh <quic_khsieh@...cinc.com>
Cc:     dri-devel@...ts.freedesktop.org, robdclark@...il.com,
        sean@...rly.run, swboyd@...omium.org, dianders@...omium.org,
        vkoul@...nel.org, daniel@...ll.ch, airlied@...il.com,
        agross@...nel.org, andersson@...nel.org, quic_abhinavk@...cinc.com,
        quic_jesszhan@...cinc.com, quic_sbillaka@...cinc.com,
        marijn.suijten@...ainline.org, freedreno@...ts.freedesktop.org,
        linux-arm-msm@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v3 4/7] drm/msm/dp: incorporate pm_runtime framework into
 DP driver

On Sat, 16 Sept 2023 at 00:38, Kuogee Hsieh <quic_khsieh@...cinc.com> wrote:
>
> Currently DP driver is executed independent of PM runtime framework.
> This lead DP driver incompatible with others. Incorporating pm runtime

Why is it incompatible? Which others are mentioned here?

> framework into DP driver so that both power and clocks to enable/disable
> host controller fits with PM runtime mechanism. Once pm runtime framework
> is incorporated into DP driver, wake up device from power up path is not
> necessary. Hence remove it. Both EV_POWER_PM_GET and EV_POWER_PM_PUT events
> are introduced to perform pm runtime control for the HPD GPIO routing to a
> display-connector case.
>
> Changes in v3:
> -- incorporate removing pm_runtime_xx() from dp_pwer.c to this patch
> -- use pm_runtime_resume_and_get() instead of pm_runtime_get()
> -- error checking pm_runtime_resume_and_get() return value
> -- add EV_POWER_PM_GET and PM_EV_POWER_PUT to handle HPD_GPIO case

Previous changelog?

>
> Signed-off-by: Kuogee Hsieh <quic_khsieh@...cinc.com>
> ---
>  drivers/gpu/drm/msm/dp/dp_aux.c     |   5 ++
>  drivers/gpu/drm/msm/dp/dp_display.c | 114 +++++++++++++++++++++++++++---------
>  drivers/gpu/drm/msm/dp/dp_power.c   |   9 ---
>  3 files changed, 90 insertions(+), 38 deletions(-)
>
> diff --git a/drivers/gpu/drm/msm/dp/dp_aux.c b/drivers/gpu/drm/msm/dp/dp_aux.c
> index 8e3b677..8fa93c5 100644
> --- a/drivers/gpu/drm/msm/dp/dp_aux.c
> +++ b/drivers/gpu/drm/msm/dp/dp_aux.c
> @@ -291,6 +291,9 @@ static ssize_t dp_aux_transfer(struct drm_dp_aux *dp_aux,
>                 return -EINVAL;
>         }
>
> +       if (pm_runtime_resume_and_get(dp_aux->dev))
> +               return  -EINVAL;

Please propagate error values instead of reinventing them.

> +
>         mutex_lock(&aux->mutex);
>         if (!aux->initted) {
>                 ret = -EIO;
> @@ -364,6 +367,8 @@ static ssize_t dp_aux_transfer(struct drm_dp_aux *dp_aux,
>
>  exit:
>         mutex_unlock(&aux->mutex);
> +       pm_runtime_mark_last_busy(dp_aux->dev);
> +       pm_runtime_put_autosuspend(dp_aux->dev);

What is the reason for using autosuspend? Such design decisions should
be described in the commit message.

>
>         return ret;
>  }
> diff --git a/drivers/gpu/drm/msm/dp/dp_display.c b/drivers/gpu/drm/msm/dp/dp_display.c
> index 59f9d85..e7af7f7 100644
> --- a/drivers/gpu/drm/msm/dp/dp_display.c
> +++ b/drivers/gpu/drm/msm/dp/dp_display.c
> @@ -60,6 +60,8 @@ enum {
>         EV_IRQ_HPD_INT,
>         EV_HPD_UNPLUG_INT,
>         EV_USER_NOTIFICATION,
> +       EV_POWER_PM_GET,
> +       EV_POWER_PM_PUT,
>  };
>
>  #define EVENT_TIMEOUT  (HZ/10) /* 100ms */
> @@ -276,13 +278,6 @@ static int dp_display_bind(struct device *dev, struct device *master,
>         dp->dp_display.drm_dev = drm;
>         priv->dp[dp->id] = &dp->dp_display;
>
> -       rc = dp->parser->parse(dp->parser);
> -       if (rc) {
> -               DRM_ERROR("device tree parsing failed\n");
> -               goto end;
> -       }
> -
> -
>         dp->drm_dev = drm;
>         dp->aux->drm_dev = drm;
>         rc = dp_aux_register(dp->aux);
> @@ -291,12 +286,6 @@ static int dp_display_bind(struct device *dev, struct device *master,
>                 goto end;
>         }
>
> -       rc = dp_power_client_init(dp->power);
> -       if (rc) {
> -               DRM_ERROR("Power client create failed\n");
> -               goto end;
> -       }
> -
>         rc = dp_register_audio_driver(dev, dp->audio);
>         if (rc) {
>                 DRM_ERROR("Audio registration Dp failed\n");
> @@ -320,10 +309,6 @@ static void dp_display_unbind(struct device *dev, struct device *master,
>         struct dp_display_private *dp = dev_get_dp_display_private(dev);
>         struct msm_drm_private *priv = dev_get_drvdata(master);
>
> -       /* disable all HPD interrupts */
> -       if (dp->core_initialized)
> -               dp_catalog_hpd_config_intr(dp->catalog, DP_DP_HPD_INT_MASK, false);
> -
>         kthread_stop(dp->ev_tsk);
>
>         of_dp_aux_depopulate_bus(dp->aux);
> @@ -467,6 +452,18 @@ static void dp_display_host_deinit(struct dp_display_private *dp)
>         dp->core_initialized = false;
>  }
>
> +static void dp_display_pm_get(struct dp_display_private *dp)
> +{
> +       if (pm_runtime_resume_and_get(&dp->pdev->dev))
> +               DRM_ERROR("failed to start power\n");
> +}

Huge NAK here. This means that the error is completely ignored (other
than being dumped to the log). This is a short path to Sync error and
other kinds of reboot.

> +
> +static void dp_display_pm_put(struct dp_display_private *dp)
> +{
> +       pm_runtime_mark_last_busy(&dp->pdev->dev);
> +       pm_runtime_put_autosuspend(&dp->pdev->dev);
> +}
> +
>  static int dp_display_usbpd_configure_cb(struct device *dev)
>  {
>         struct dp_display_private *dp = dev_get_dp_display_private(dev);
> @@ -1096,7 +1093,6 @@ static int hpd_event_thread(void *data)
>
>                 switch (todo->event_id) {
>                 case EV_HPD_INIT_SETUP:
> -                       dp_display_host_init(dp_priv);
>                         break;
>                 case EV_HPD_PLUG_INT:
>                         dp_hpd_plug_handle(dp_priv, todo->data);
> @@ -1111,6 +1107,12 @@ static int hpd_event_thread(void *data)
>                         dp_display_send_hpd_notification(dp_priv,
>                                                 todo->data);
>                         break;
> +               case EV_POWER_PM_GET:
> +                       dp_display_pm_get(dp_priv);
> +                       break;
> +               case EV_POWER_PM_PUT:
> +                       dp_display_pm_put(dp_priv);
> +                       break;

No. runtime_get / runtime_put are not HPD events. They should be
executed directly from the place where the drivers needs the device to
be powered up.

>                 default:
>                         break;
>                 }
> @@ -1251,6 +1253,18 @@ static int dp_display_probe(struct platform_device *pdev)
>                 return -EPROBE_DEFER;
>         }
>
> +       rc = dp->parser->parse(dp->parser);
> +       if (rc) {
> +               DRM_ERROR("device tree parsing failed\n");
> +               return -EPROBE_DEFER;
> +       }
> +
> +       rc = dp_power_client_init(dp->power);
> +       if (rc) {
> +               DRM_ERROR("Power client create failed\n");
> +               return -EPROBE_DEFER;
> +       }

Why? This moves resource allocation to the probe function, which is
irrelevant to the pm_runtime code. If this is required, you can move
these changes to a separate patch.

> +
>         /* setup event q */
>         mutex_init(&dp->event_mutex);
>         init_waitqueue_head(&dp->event_q);
> @@ -1263,6 +1277,10 @@ static int dp_display_probe(struct platform_device *pdev)
>
>         platform_set_drvdata(pdev, &dp->dp_display);
>
> +       devm_pm_runtime_enable(&pdev->dev);

error code handling?

> +       pm_runtime_set_autosuspend_delay(&pdev->dev, 1000);
> +       pm_runtime_use_autosuspend(&pdev->dev);
> +
>         rc = dp_display_request_irq(dp);
>         if (rc)
>                 return rc;
> @@ -1285,6 +1303,36 @@ static int dp_display_remove(struct platform_device *pdev)
>
>         platform_set_drvdata(pdev, NULL);
>
> +       pm_runtime_put_sync_suspend(&pdev->dev);

Why? Who is holding the pm count here?

> +       pm_runtime_dont_use_autosuspend(&pdev->dev);
> +       pm_runtime_disable(&pdev->dev);

Why do you need _disable if you have a devm_pm_runtime_enable()? Not
to mention that pm_runtime_disable_action() already has a call to
pm_runtime_dont_use_autosuspend()

> +
> +       return 0;
> +}
> +
> +static int dp_pm_runtime_suspend(struct device *dev)
> +{
> +       struct dp_display_private *dp = dev_get_dp_display_private(dev);
> +
> +       if (dp->dp_display.is_edp) {
> +               dp_display_host_phy_exit(dp);
> +               dp_catalog_ctrl_hpd_disable(dp->catalog);
> +       }
> +       dp_display_host_deinit(dp);
> +
> +       return 0;
> +}
> +
> +static int dp_pm_runtime_resume(struct device *dev)
> +{
> +       struct dp_display_private *dp = dev_get_dp_display_private(dev);
> +
> +       dp_display_host_init(dp);
> +       if (dp->dp_display.is_edp) {
> +               dp_catalog_ctrl_hpd_enable(dp->catalog);
> +               dp_display_host_phy_init(dp);
> +       }
> +
>         return 0;
>  }
>
> @@ -1389,6 +1437,7 @@ static int dp_pm_suspend(struct device *dev)
>  }
>
>  static const struct dev_pm_ops dp_pm_ops = {
> +       SET_RUNTIME_PM_OPS(dp_pm_runtime_suspend, dp_pm_runtime_resume, NULL)
>         .suspend = dp_pm_suspend,
>         .resume =  dp_pm_resume,
>  };
> @@ -1473,10 +1522,6 @@ static int dp_display_get_next_bridge(struct msm_dp *dp)
>         aux_bus = of_get_child_by_name(dev->of_node, "aux-bus");
>
>         if (aux_bus && dp->is_edp) {
> -               dp_display_host_init(dp_priv);
> -               dp_catalog_ctrl_hpd_enable(dp_priv->catalog);
> -               dp_display_host_phy_init(dp_priv);
> -
>                 /*
>                  * The code below assumes that the panel will finish probing
>                  * by the time devm_of_dp_aux_populate_ep_devices() returns.
> @@ -1578,6 +1623,11 @@ void dp_bridge_atomic_enable(struct drm_bridge *drm_bridge,
>                 dp_hpd_plug_handle(dp_display, 0);
>
>         mutex_lock(&dp_display->event_mutex);
> +       if (pm_runtime_resume_and_get(&dp_display->pdev->dev)) {
> +               DRM_ERROR("failed to start power\n");
> +               mutex_unlock(&dp_display->event_mutex);
> +               return;
> +       }
>
>         state = dp_display->hpd_state;
>         if (state != ST_DISPLAY_OFF && state != ST_MAINLINK_READY) {
> @@ -1658,6 +1708,8 @@ void dp_bridge_atomic_post_disable(struct drm_bridge *drm_bridge,
>         }
>
>         drm_dbg_dp(dp->drm_dev, "type=%d Done\n", dp->connector_type);
> +
> +       pm_runtime_put_sync(&dp_display->pdev->dev);

So, no autosuspend now?

Also, I think we can get an unbalanced runtime status, as there is no
guarantee that atomic_enable / atomic_disable will be paired. Please
correct me if I'm wrong.
And also there is a possible return earlier in this function. The
driver will leak the runtime status again.

>         mutex_unlock(&dp_display->event_mutex);
>  }
>
> @@ -1697,6 +1749,9 @@ void dp_bridge_hpd_enable(struct drm_bridge *bridge)
>         struct dp_display_private *dp = container_of(dp_display, struct dp_display_private, dp_display);
>
>         mutex_lock(&dp->event_mutex);
> +       if (pm_runtime_resume_and_get(&dp->pdev->dev))
> +               DRM_ERROR("failed to start power\n");

Return?

> +
>         dp_catalog_ctrl_hpd_enable(dp->catalog);
>
>         /* enable HDP interrupts */
> @@ -1718,6 +1773,9 @@ void dp_bridge_hpd_disable(struct drm_bridge *bridge)
>         dp_catalog_ctrl_hpd_disable(dp->catalog);
>
>         dp_display->internal_hpd = false;
> +
> +       pm_runtime_mark_last_busy(&dp->pdev->dev);
> +       pm_runtime_put_autosuspend(&dp->pdev->dev);
>         mutex_unlock(&dp->event_mutex);
>  }
>
> @@ -1732,13 +1790,11 @@ void dp_bridge_hpd_notify(struct drm_bridge *bridge,
>         if (dp_display->internal_hpd)
>                 return;
>
> -       if (!dp->core_initialized) {
> -               drm_dbg_dp(dp->drm_dev, "not initialized\n");
> -               return;
> -       }
> -
> -       if (!dp_display->link_ready && status == connector_status_connected)
> +       if (!dp_display->link_ready && status == connector_status_connected) {
> +               dp_add_event(dp, EV_POWER_PM_GET, 0, 0);

Why? What for?

>                 dp_add_event(dp, EV_HPD_PLUG_INT, 0, 0);
> -       else if (dp_display->link_ready && status == connector_status_disconnected)
> +       } else if (dp_display->link_ready && status == connector_status_disconnected) {
>                 dp_add_event(dp, EV_HPD_UNPLUG_INT, 0, 0);
> +               dp_add_event(dp, EV_POWER_PM_PUT, 0, 0);
> +       }
>  }
> diff --git a/drivers/gpu/drm/msm/dp/dp_power.c b/drivers/gpu/drm/msm/dp/dp_power.c
> index 5cb84ca..ed2f62a 100644
> --- a/drivers/gpu/drm/msm/dp/dp_power.c
> +++ b/drivers/gpu/drm/msm/dp/dp_power.c
> @@ -152,8 +152,6 @@ int dp_power_client_init(struct dp_power *dp_power)
>
>         power = container_of(dp_power, struct dp_power_private, dp_power);
>
> -       pm_runtime_enable(power->dev);
> -
>         return dp_power_clk_init(power);
>  }
>
> @@ -162,8 +160,6 @@ void dp_power_client_deinit(struct dp_power *dp_power)
>         struct dp_power_private *power;
>
>         power = container_of(dp_power, struct dp_power_private, dp_power);
> -
> -       pm_runtime_disable(power->dev);
>  }
>
>  int dp_power_init(struct dp_power *dp_power)
> @@ -173,11 +169,7 @@ int dp_power_init(struct dp_power *dp_power)
>
>         power = container_of(dp_power, struct dp_power_private, dp_power);
>
> -       pm_runtime_get_sync(power->dev);
> -
>         rc = dp_power_clk_enable(dp_power, DP_CORE_PM, true);
> -       if (rc)
> -               pm_runtime_put_sync(power->dev);
>
>         return rc;
>  }
> @@ -189,7 +181,6 @@ int dp_power_deinit(struct dp_power *dp_power)
>         power = container_of(dp_power, struct dp_power_private, dp_power);
>
>         dp_power_clk_enable(dp_power, DP_CORE_PM, false);
> -       pm_runtime_put_sync(power->dev);
>         return 0;
>  }
>
> --
> 2.7.4
>


-- 
With best wishes
Dmitry

Powered by blists - more mailing lists