The interrupt handler for HPD is useful only if a display is actually
supposed to be hotpluggable, as that manages the machinery to perform
cable (un)plug detection, debouncing and setup for re-training.

Since eDP panels are not supposed to be hotpluggable we can avoid
using the HPD interrupts altogether and rely on HPD polling only
for the suspend/resume case, saving us some spinlocking action and
the overhead of interrupts firing at every suspend/resume cycle,
achieving a faster (even if just slightly) display resume.

Signed-off-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
---
 drivers/gpu/drm/mediatek/mtk_dp.c | 81 ++++++++++++++++++-------------
 1 file changed, 46 insertions(+), 35 deletions(-)

diff --git a/drivers/gpu/drm/mediatek/mtk_dp.c b/drivers/gpu/drm/mediatek/mtk_dp.c
index e74295ba9707..c06fcc7318e7 100644
--- a/drivers/gpu/drm/mediatek/mtk_dp.c
+++ b/drivers/gpu/drm/mediatek/mtk_dp.c
@@ -2182,9 +2182,11 @@ static int mtk_dp_bridge_attach(struct drm_bridge *bridge,
 
 	mtk_dp->drm_dev = bridge->dev;
 
-	irq_clear_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
-	enable_irq(mtk_dp->irq);
-	mtk_dp_hwirq_enable(mtk_dp, true);
+	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP) {
+		irq_clear_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
+		enable_irq(mtk_dp->irq);
+		mtk_dp_hwirq_enable(mtk_dp, true);
+	}
 
 	return 0;
 
@@ -2199,8 +2201,10 @@ static void mtk_dp_bridge_detach(struct drm_bridge *bridge)
 {
 	struct mtk_dp *mtk_dp = mtk_dp_from_bridge(bridge);
 
-	mtk_dp_hwirq_enable(mtk_dp, false);
-	disable_irq(mtk_dp->irq);
+	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP) {
+		mtk_dp_hwirq_enable(mtk_dp, false);
+		disable_irq(mtk_dp->irq);
+	}
 	mtk_dp->drm_dev = NULL;
 	mtk_dp_poweroff(mtk_dp);
 	drm_dp_aux_unregister(&mtk_dp->aux);
@@ -2579,40 +2583,44 @@ static int mtk_dp_probe(struct platform_device *pdev)
 	mtk_dp->dev = dev;
 	mtk_dp->data = (struct mtk_dp_data *)of_device_get_match_data(dev);
 
-	mtk_dp->irq = platform_get_irq(pdev, 0);
-	if (mtk_dp->irq < 0)
-		return dev_err_probe(dev, mtk_dp->irq,
-				     "failed to request dp irq resource\n");
-
-	mtk_dp->next_bridge = devm_drm_of_get_bridge(dev, dev->of_node, 1, 0);
-	if (IS_ERR(mtk_dp->next_bridge) &&
-	    PTR_ERR(mtk_dp->next_bridge) == -ENODEV)
-		mtk_dp->next_bridge = NULL;
-	else if (IS_ERR(mtk_dp->next_bridge))
-		return dev_err_probe(dev, PTR_ERR(mtk_dp->next_bridge),
-				     "Failed to get bridge\n");
-
 	ret = mtk_dp_dt_parse(mtk_dp, pdev);
 	if (ret)
 		return dev_err_probe(dev, ret, "Failed to parse dt\n");
 
+	/*
+	 * Request the interrupt and install service routine only if we are
+	 * on full DisplayPort.
+	 * For eDP, polling the HPD instead is more convenient because we
+	 * don't expect any (un)plug events during runtime, hence we can
+	 * avoid some locking.
+	 */
+	if (mtk_dp->data->bridge_type != DRM_MODE_CONNECTOR_eDP) {
+		mtk_dp->irq = platform_get_irq(pdev, 0);
+		if (mtk_dp->irq < 0)
+			return dev_err_probe(dev, mtk_dp->irq,
+					     "failed to request dp irq resource\n");
+
+		spin_lock_init(&mtk_dp->irq_thread_lock);
+
+		irq_set_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
+		ret = devm_request_threaded_irq(dev, mtk_dp->irq, mtk_dp_hpd_event,
+						mtk_dp_hpd_event_thread,
+						IRQ_TYPE_LEVEL_HIGH, dev_name(dev),
+						mtk_dp);
+		if (ret)
+			return dev_err_probe(dev, ret,
+					     "failed to request mediatek dptx irq\n");
+
+		mtk_dp->need_debounce = true;
+		timer_setup(&mtk_dp->debounce_timer, mtk_dp_debounce_timer, 0);
+	}
+
 	mtk_dp->aux.name = "aux_mtk_dp";
 	mtk_dp->aux.dev = dev;
 	mtk_dp->aux.transfer = mtk_dp_aux_transfer;
 	mtk_dp->aux.wait_hpd_asserted = mtk_dp_wait_hpd_asserted;
 	drm_dp_aux_init(&mtk_dp->aux);
 
-	spin_lock_init(&mtk_dp->irq_thread_lock);
-
-	irq_set_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
-	ret = devm_request_threaded_irq(dev, mtk_dp->irq, mtk_dp_hpd_event,
-					mtk_dp_hpd_event_thread,
-					IRQ_TYPE_LEVEL_HIGH, dev_name(dev),
-					mtk_dp);
-	if (ret)
-		return dev_err_probe(dev, ret,
-				     "failed to request mediatek dptx irq\n");
-
 	platform_set_drvdata(pdev, mtk_dp);
 
 	if (mtk_dp->data->audio_supported) {
@@ -2634,9 +2642,6 @@ static int mtk_dp_probe(struct platform_device *pdev)
 	mtk_dp->bridge.of_node = dev->of_node;
 	mtk_dp->bridge.type = mtk_dp->data->bridge_type;
 
-	mtk_dp->need_debounce = true;
-	timer_setup(&mtk_dp->debounce_timer, mtk_dp_debounce_timer, 0);
-
 	if (mtk_dp->bridge.type == DRM_MODE_CONNECTOR_eDP) {
 		/*
 		 * Set the data lanes to idle in case the bootloader didn't
@@ -2647,6 +2652,9 @@ static int mtk_dp_probe(struct platform_device *pdev)
 		mtk_dp_initialize_aux_settings(mtk_dp);
 		mtk_dp_power_enable(mtk_dp);
 
+		/* Disable HW interrupts: we don't need any for eDP */
+		mtk_dp_hwirq_enable(mtk_dp, false);
+
 		/*
 		 * Power on the AUX to allow reading the EDID from aux-bus:
 		 * please note that it is necessary to call power off in the
@@ -2692,7 +2700,8 @@ static int mtk_dp_remove(struct platform_device *pdev)
 
 	pm_runtime_put(&pdev->dev);
 	pm_runtime_disable(&pdev->dev);
-	del_timer_sync(&mtk_dp->debounce_timer);
+	if (mtk_dp->data->bridge_type != DRM_MODE_CONNECTOR_eDP)
+		del_timer_sync(&mtk_dp->debounce_timer);
 	platform_device_unregister(mtk_dp->phy_dev);
 	if (mtk_dp->audio_pdev)
 		platform_device_unregister(mtk_dp->audio_pdev);
@@ -2706,7 +2715,8 @@ static int mtk_dp_suspend(struct device *dev)
 	struct mtk_dp *mtk_dp = dev_get_drvdata(dev);
 
 	mtk_dp_power_disable(mtk_dp);
-	mtk_dp_hwirq_enable(mtk_dp, false);
+	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP)
+		mtk_dp_hwirq_enable(mtk_dp, false);
 	pm_runtime_put_sync(dev);
 
 	return 0;
@@ -2718,7 +2728,8 @@ static int mtk_dp_resume(struct device *dev)
 
 	pm_runtime_get_sync(dev);
 	mtk_dp_init_port(mtk_dp);
-	mtk_dp_hwirq_enable(mtk_dp, true);
+	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP)
+		mtk_dp_hwirq_enable(mtk_dp, true);
 	mtk_dp_power_enable(mtk_dp);
 
 	return 0;
-- 
2.40.1

Hi, Angelo:

On Mon, 2023-07-17 at 16:14 +0200, AngeloGioacchino Del Regno wrote:
>  	 
> External email : Please do not click links or open attachments until
> you have verified the sender or the content.
>  The interrupt handler for HPD is useful only if a display is
> actually
> supposed to be hotpluggable, as that manages the machinery to perform
> cable (un)plug detection, debouncing and setup for re-training.
> 
> Since eDP panels are not supposed to be hotpluggable we can avoid
> using the HPD interrupts altogether and rely on HPD polling only
> for the suspend/resume case, saving us some spinlocking action and
> the overhead of interrupts firing at every suspend/resume cycle,
> achieving a faster (even if just slightly) display resume.
> 
> Signed-off-by: AngeloGioacchino Del Regno <
> angelogioacchino.delregno@collabora.com>
> ---
>  drivers/gpu/drm/mediatek/mtk_dp.c | 81 ++++++++++++++++++-----------
> --
>  1 file changed, 46 insertions(+), 35 deletions(-)
> 
> diff --git a/drivers/gpu/drm/mediatek/mtk_dp.c
> b/drivers/gpu/drm/mediatek/mtk_dp.c
> index e74295ba9707..c06fcc7318e7 100644
> --- a/drivers/gpu/drm/mediatek/mtk_dp.c
> +++ b/drivers/gpu/drm/mediatek/mtk_dp.c
> @@ -2182,9 +2182,11 @@ static int mtk_dp_bridge_attach(struct
> drm_bridge *bridge,
>  
>  	mtk_dp->drm_dev = bridge->dev;
>  
> -	irq_clear_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
> -	enable_irq(mtk_dp->irq);
> -	mtk_dp_hwirq_enable(mtk_dp, true);
> +	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP) {
> +		irq_clear_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
> +		enable_irq(mtk_dp->irq);
> +		mtk_dp_hwirq_enable(mtk_dp, true);
> +	}
>  
>  	return 0;
>  
> @@ -2199,8 +2201,10 @@ static void mtk_dp_bridge_detach(struct
> drm_bridge *bridge)
>  {
>  	struct mtk_dp *mtk_dp = mtk_dp_from_bridge(bridge);
>  
> -	mtk_dp_hwirq_enable(mtk_dp, false);
> -	disable_irq(mtk_dp->irq);
> +	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP) {
> +		mtk_dp_hwirq_enable(mtk_dp, false);
> +		disable_irq(mtk_dp->irq);
> +	}
>  	mtk_dp->drm_dev = NULL;
>  	mtk_dp_poweroff(mtk_dp);
>  	drm_dp_aux_unregister(&mtk_dp->aux);
> @@ -2579,40 +2583,44 @@ static int mtk_dp_probe(struct
> platform_device *pdev)
>  	mtk_dp->dev = dev;
>  	mtk_dp->data = (struct mtk_dp_data
> *)of_device_get_match_data(dev);
>  
> -	mtk_dp->irq = platform_get_irq(pdev, 0);
> -	if (mtk_dp->irq < 0)
> -		return dev_err_probe(dev, mtk_dp->irq,
> -				     "failed to request dp irq
> resource\n");
> -
> -	mtk_dp->next_bridge = devm_drm_of_get_bridge(dev, dev->of_node, 
> 1, 0);
> -	if (IS_ERR(mtk_dp->next_bridge) &&
> -	    PTR_ERR(mtk_dp->next_bridge) == -ENODEV)
> -		mtk_dp->next_bridge = NULL;
> -	else if (IS_ERR(mtk_dp->next_bridge))
> -		return dev_err_probe(dev, PTR_ERR(mtk_dp->next_bridge),
> -				     "Failed to get bridge\n");
> -

Why remove next_bridge setting?

Regards,
CK

>  	ret = mtk_dp_dt_parse(mtk_dp, pdev);
>  	if (ret)
>  		return dev_err_probe(dev, ret, "Failed to parse dt\n");
>  
> +	/*
> +	 * Request the interrupt and install service routine only if we
> are
> +	 * on full DisplayPort.
> +	 * For eDP, polling the HPD instead is more convenient because
> we
> +	 * don't expect any (un)plug events during runtime, hence we
> can
> +	 * avoid some locking.
> +	 */
> +	if (mtk_dp->data->bridge_type != DRM_MODE_CONNECTOR_eDP) {
> +		mtk_dp->irq = platform_get_irq(pdev, 0);
> +		if (mtk_dp->irq < 0)
> +			return dev_err_probe(dev, mtk_dp->irq,
> +					     "failed to request dp irq
> resource\n");
> +
> +		spin_lock_init(&mtk_dp->irq_thread_lock);
> +
> +		irq_set_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
> +		ret = devm_request_threaded_irq(dev, mtk_dp->irq,
> mtk_dp_hpd_event,
> +						mtk_dp_hpd_event_thread
> ,
> +						IRQ_TYPE_LEVEL_HIGH,
> dev_name(dev),
> +						mtk_dp);
> +		if (ret)
> +			return dev_err_probe(dev, ret,
> +					     "failed to request
> mediatek dptx irq\n");
> +
> +		mtk_dp->need_debounce = true;
> +		timer_setup(&mtk_dp->debounce_timer,
> mtk_dp_debounce_timer, 0);
> +	}
> +
>  	mtk_dp->aux.name = "aux_mtk_dp";
>  	mtk_dp->aux.dev = dev;
>  	mtk_dp->aux.transfer = mtk_dp_aux_transfer;
>  	mtk_dp->aux.wait_hpd_asserted = mtk_dp_wait_hpd_asserted;
>  	drm_dp_aux_init(&mtk_dp->aux);
>  
> -	spin_lock_init(&mtk_dp->irq_thread_lock);
> -
> -	irq_set_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
> -	ret = devm_request_threaded_irq(dev, mtk_dp->irq,
> mtk_dp_hpd_event,
> -					mtk_dp_hpd_event_thread,
> -					IRQ_TYPE_LEVEL_HIGH,
> dev_name(dev),
> -					mtk_dp);
> -	if (ret)
> -		return dev_err_probe(dev, ret,
> -				     "failed to request mediatek dptx
> irq\n");
> -
>  	platform_set_drvdata(pdev, mtk_dp);
>  
>  	if (mtk_dp->data->audio_supported) {
> @@ -2634,9 +2642,6 @@ static int mtk_dp_probe(struct platform_device
> *pdev)
>  	mtk_dp->bridge.of_node = dev->of_node;
>  	mtk_dp->bridge.type = mtk_dp->data->bridge_type;
>  
> -	mtk_dp->need_debounce = true;
> -	timer_setup(&mtk_dp->debounce_timer, mtk_dp_debounce_timer, 0);
> -
>  	if (mtk_dp->bridge.type == DRM_MODE_CONNECTOR_eDP) {
>  		/*
>  		 * Set the data lanes to idle in case the bootloader
> didn't
> @@ -2647,6 +2652,9 @@ static int mtk_dp_probe(struct platform_device
> *pdev)
>  		mtk_dp_initialize_aux_settings(mtk_dp);
>  		mtk_dp_power_enable(mtk_dp);
>  
> +		/* Disable HW interrupts: we don't need any for eDP */
> +		mtk_dp_hwirq_enable(mtk_dp, false);
> +
>  		/*
>  		 * Power on the AUX to allow reading the EDID from aux-
> bus:
>  		 * please note that it is necessary to call power off
> in the
> @@ -2692,7 +2700,8 @@ static int mtk_dp_remove(struct platform_device
> *pdev)
>  
>  	pm_runtime_put(&pdev->dev);
>  	pm_runtime_disable(&pdev->dev);
> -	del_timer_sync(&mtk_dp->debounce_timer);
> +	if (mtk_dp->data->bridge_type != DRM_MODE_CONNECTOR_eDP)
> +		del_timer_sync(&mtk_dp->debounce_timer);
>  	platform_device_unregister(mtk_dp->phy_dev);
>  	if (mtk_dp->audio_pdev)
>  		platform_device_unregister(mtk_dp->audio_pdev);
> @@ -2706,7 +2715,8 @@ static int mtk_dp_suspend(struct device *dev)
>  	struct mtk_dp *mtk_dp = dev_get_drvdata(dev);
>  
>  	mtk_dp_power_disable(mtk_dp);
> -	mtk_dp_hwirq_enable(mtk_dp, false);
> +	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP)
> +		mtk_dp_hwirq_enable(mtk_dp, false);
>  	pm_runtime_put_sync(dev);
>  
>  	return 0;
> @@ -2718,7 +2728,8 @@ static int mtk_dp_resume(struct device *dev)
>  
>  	pm_runtime_get_sync(dev);
>  	mtk_dp_init_port(mtk_dp);
> -	mtk_dp_hwirq_enable(mtk_dp, true);
> +	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP)
> +		mtk_dp_hwirq_enable(mtk_dp, true);
>  	mtk_dp_power_enable(mtk_dp);
>  
>  	return 0;
> -- 
> 2.40.1

Il 25/07/23 08:37, CK Hu (胡俊光) ha scritto:
> Hi, Angelo:
> 
> On Mon, 2023-07-17 at 16:14 +0200, AngeloGioacchino Del Regno wrote:
>>   	
>> External email : Please do not click links or open attachments until
>> you have verified the sender or the content.
>>   The interrupt handler for HPD is useful only if a display is
>> actually
>> supposed to be hotpluggable, as that manages the machinery to perform
>> cable (un)plug detection, debouncing and setup for re-training.
>>
>> Since eDP panels are not supposed to be hotpluggable we can avoid
>> using the HPD interrupts altogether and rely on HPD polling only
>> for the suspend/resume case, saving us some spinlocking action and
>> the overhead of interrupts firing at every suspend/resume cycle,
>> achieving a faster (even if just slightly) display resume.
>>
>> Signed-off-by: AngeloGioacchino Del Regno <
>> angelogioacchino.delregno@collabora.com>
>> ---
>>   drivers/gpu/drm/mediatek/mtk_dp.c | 81 ++++++++++++++++++-----------
>> --
>>   1 file changed, 46 insertions(+), 35 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/mediatek/mtk_dp.c
>> b/drivers/gpu/drm/mediatek/mtk_dp.c
>> index e74295ba9707..c06fcc7318e7 100644
>> --- a/drivers/gpu/drm/mediatek/mtk_dp.c
>> +++ b/drivers/gpu/drm/mediatek/mtk_dp.c
>> @@ -2182,9 +2182,11 @@ static int mtk_dp_bridge_attach(struct
>> drm_bridge *bridge,
>>   
>>   	mtk_dp->drm_dev = bridge->dev;
>>   
>> -	irq_clear_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
>> -	enable_irq(mtk_dp->irq);
>> -	mtk_dp_hwirq_enable(mtk_dp, true);
>> +	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP) {
>> +		irq_clear_status_flags(mtk_dp->irq, IRQ_NOAUTOEN);
>> +		enable_irq(mtk_dp->irq);
>> +		mtk_dp_hwirq_enable(mtk_dp, true);
>> +	}
>>   
>>   	return 0;
>>   
>> @@ -2199,8 +2201,10 @@ static void mtk_dp_bridge_detach(struct
>> drm_bridge *bridge)
>>   {
>>   	struct mtk_dp *mtk_dp = mtk_dp_from_bridge(bridge);
>>   
>> -	mtk_dp_hwirq_enable(mtk_dp, false);
>> -	disable_irq(mtk_dp->irq);
>> +	if (mtk_dp->bridge.type != DRM_MODE_CONNECTOR_eDP) {
>> +		mtk_dp_hwirq_enable(mtk_dp, false);
>> +		disable_irq(mtk_dp->irq);
>> +	}
>>   	mtk_dp->drm_dev = NULL;
>>   	mtk_dp_poweroff(mtk_dp);
>>   	drm_dp_aux_unregister(&mtk_dp->aux);
>> @@ -2579,40 +2583,44 @@ static int mtk_dp_probe(struct
>> platform_device *pdev)
>>   	mtk_dp->dev = dev;
>>   	mtk_dp->data = (struct mtk_dp_data
>> *)of_device_get_match_data(dev);
>>   
>> -	mtk_dp->irq = platform_get_irq(pdev, 0);
>> -	if (mtk_dp->irq < 0)
>> -		return dev_err_probe(dev, mtk_dp->irq,
>> -				     "failed to request dp irq
>> resource\n");
>> -
>> -	mtk_dp->next_bridge = devm_drm_of_get_bridge(dev, dev->of_node,
>> 1, 0);
>> -	if (IS_ERR(mtk_dp->next_bridge) &&
>> -	    PTR_ERR(mtk_dp->next_bridge) == -ENODEV)
>> -		mtk_dp->next_bridge = NULL;
>> -	else if (IS_ERR(mtk_dp->next_bridge))
>> -		return dev_err_probe(dev, PTR_ERR(mtk_dp->next_bridge),
>> -				     "Failed to get bridge\n");
>> -
> 
> Why remove next_bridge setting?
> 

Hello CK,

Thanks for making me notice that! The removal of this code snippet belongs
to patch [9/11] `drm/mediatek: dp: Add support for embedded DisplayPort aux-bus`
because in that patch I am moving the call to devm_drm_of_get_bridge() to function
mtk_dp_edp_link_panel(), which is called as a .done_probing() callback if the eDP
panel is on aux-bus, or "manually" if not on aux-bus.

I'll send a v7 with the removal of this snippet in the right patch.

Thanks again,
Angelo

Reviewed-by: Alexandre Mergnat <amergnat@baylibre.com>

On 17/07/2023 16:14, AngeloGioacchino Del Regno wrote:
> The interrupt handler for HPD is useful only if a display is actually
> supposed to be hotpluggable, as that manages the machinery to perform
> cable (un)plug detection, debouncing and setup for re-training.
> 
> Since eDP panels are not supposed to be hotpluggable we can avoid
> using the HPD interrupts altogether and rely on HPD polling only
> for the suspend/resume case, saving us some spinlocking action and
> the overhead of interrupts firing at every suspend/resume cycle,
> achieving a faster (even if just slightly) display resume.

-- 
Regards,
Alexandre