Skip to contentSkip to conversation

drm/nouveau/disp: use drm_kms_helper_connector_hotplug_event()

PATCH In review Simon Ser 2 participants 5 messages Jun 20, 2023 – Jul 10, 2023

Subsystems (3)
lore
Share & tools
lore

Simon Ser

author
#0 lore raw

PATCH 1/1 drm/nouveau/disp: use drm_kms_helper_connector_hotplug_event()

This adds more information to the hotplug uevent and lets user-space know that it's about a particular connector only.

Cc Ben Skeggs <[email protected]>
Cc Lyude Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Karol Herbst <[email protected]>
---
 drivers/gpu/drm/nouveau/nouveau_display.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)
drivers/gpu/drm/nouveau/nouveau_display.c +14 -3
index ec3ffff487fc..99977e5fe716 100644
@@ -465,7 +465,8 @@ nouveau_display_hpd_work(struct work_struct *work)
 	struct drm_connector *connector;
 	struct drm_connector_list_iter conn_iter;
 	u32 pending;
-	bool changed = false;
+	int changed = 0;
+	struct drm_connector *first_changed_connector = NULL;
 
 	pm_runtime_get_sync(dev->dev);
 
@@ -509,7 +510,12 @@ nouveau_display_hpd_work(struct work_struct *work)
 		if (old_epoch_counter == connector->epoch_counter)
 			continue;
 
-		changed = true;
+		changed++;
+		if (!first_changed_connector) {
+			drm_connector_get(connector);
+			first_changed_connector = connector;
+		}
+
 		drm_dbg_kms(dev, "[CONNECTOR:%d:%s] status updated from %s to %s (epoch counter %llu->%llu)\n",
 			    connector->base.id, connector->name,
 			    drm_get_connector_status_name(old_status),
@@ -520,9 +526,14 @@ nouveau_display_hpd_work(struct work_struct *work)
 	drm_connector_list_iter_end(&conn_iter);
 	mutex_unlock(&dev->mode_config.mutex);
 
-	if (changed)
+	if (changed == 1)
+		drm_kms_helper_connector_hotplug_event(first_changed_connector);
+	else if (changed > 0)
 		drm_kms_helper_hotplug_event(dev);
 
+	if (first_changed_connector)
+		drm_connector_put(first_changed_connector);
+
 	pm_runtime_mark_last_busy(drm->dev->dev);
 noop:
 	pm_runtime_put_autosuspend(dev->dev);
Signature · 1 line
2.41.0
maintainer
#1 lore raw

Looks alright! Some comments below

Quoted 41 lines · Simon Ser

This adds more information to the hotplug uevent and lets user-space know that it's about a particular connector only.

Cc Ben Skeggs <[email protected]>
Cc Lyude Paul <[email protected]>
Cc David Airlie <[email protected]>
Cc Karol Herbst <[email protected]>
---
 drivers/gpu/drm/nouveau/nouveau_display.c | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)
drivers/gpu/drm/nouveau/nouveau_display.c +8 -2
index ec3ffff487fc..99977e5fe716 100644
@@ -465,7 +465,8 @@ nouveau_display_hpd_work(struct work_struct *work)
 	struct drm_connector *connector;
 	struct drm_connector_list_iter conn_iter;
 	u32 pending;
-	bool changed = false;
+	int changed = 0;
+	struct drm_connector *first_changed_connector = NULL;
 
 	pm_runtime_get_sync(dev->dev);
 
@@ -509,7 +510,12 @@ nouveau_display_hpd_work(struct work_struct *work)
 		if (old_epoch_counter == connector->epoch_counter)
 			continue;
 
-		changed = true;
+		changed++;
+		if (!first_changed_connector) {
+			drm_connector_get(connector);
+			first_changed_connector = connector;
+		}
+
 		drm_dbg_kms(dev, "[CONNECTOR:%d:%s] status updated from %s to %s (epoch counter %llu->%llu)\n",
 			    connector->base.id, connector->name,
 			    drm_get_connector_status_name(old_status),

drivers/gpu/drm/nouveau/nouveau_display.c nouveau_display_hpd_work(struct work_struct *work) · line 532

Quoted 8 lines
drivers/gpu/drm/nouveau/nouveau_display.c +3 -1
@@ -520,9 +526,14 @@ nouveau_display_hpd_work(struct work_struct *work)
 	drm_connector_list_iter_end(&conn_iter);
 	mutex_unlock(&dev->mode_config.mutex);
 
-	if (changed)
+	if (changed == 1)
+		drm_kms_helper_connector_hotplug_event(first_changed_connector);
+	else if (changed > 0)
 		drm_kms_helper_hotplug_event(dev);

I'm curious if you think there might be an advantage to doing this per- connector even with multiple connectors? Seems like we could do that if we stored changed connectors as a bitmask.

Quoted 7 lines
drivers/gpu/drm/nouveau/nouveau_display.c +3 -0
+	if (first_changed_connector)
+		drm_connector_put(first_changed_connector);
+
 	pm_runtime_mark_last_busy(drm->dev->dev);
 noop:
 	pm_runtime_put_autosuspend(dev->dev);
Signature · 3 lines
Cheers,
 Lyude Paul (she/her)
 Software Engineer at Red Hat
author
#2 lore raw

Hi Lyude!

Quoted 9 lines · Lyude Paul

drivers/gpu/drm/nouveau/nouveau_display.c nouveau_display_hpd_work(struct work_struct *work) · line 532

Quoted 5 lines
drivers/gpu/drm/nouveau/nouveau_display.c +3 -1
-	if (changed)
+	if (changed == 1)
+		drm_kms_helper_connector_hotplug_event(first_changed_connector);
+	else if (changed > 0)
 		drm_kms_helper_hotplug_event(dev);

I'm curious if you think there might be an advantage to doing this per- connector even with multiple connectors? Seems like we could do that if we stored changed connectors as a bitmask.

I believe firing off multiple per-connector uevents would be worse than a single per-device uevent because user-space would see intermediate states instead of an atomic change.

Let's say I have 3 connectors connected, and different settings set in my compositor when 1, 2 or 3 screens are connected. Let's say an unplug disconnects 2 monitors at the same time. With a single uevent, the compositor refreshes the whole device state, and then applies the 1-screen settings. With 2 uevents (one for each connector), the compositor refreshes the first disconnected connector state, then applies the 2-screen settings, then refreshes the second disconnected connector state, then applies the 1-screen settings. Applying the 2-screen settings might cause an unnecessarily glitchy intermediate state, where windows move around, modesets are performed, user configured commands are run, just to un-do everything right after.

I hope this makes sense,

Simon

maintainer Reviewed-by
#3 lore raw
Quoted 29 lines · Simon Ser

Hi Lyude!

Quoted 9 lines · Lyude Paul

drivers/gpu/drm/nouveau/nouveau_display.c nouveau_display_hpd_work(struct work_struct *work) · line 532

Quoted 5 lines
drivers/gpu/drm/nouveau/nouveau_display.c +3 -1
-	if (changed)
+	if (changed == 1)
+		drm_kms_helper_connector_hotplug_event(first_changed_connector);
+	else if (changed > 0)
 		drm_kms_helper_hotplug_event(dev);

I'm curious if you think there might be an advantage to doing this per- connector even with multiple connectors? Seems like we could do that if we stored changed connectors as a bitmask.

I believe firing off multiple per-connector uevents would be worse than a single per-device uevent because user-space would see intermediate states instead of an atomic change.

Let's say I have 3 connectors connected, and different settings set in my compositor when 1, 2 or 3 screens are connected. Let's say an unplug disconnects 2 monitors at the same time. With a single uevent, the compositor refreshes the whole device state, and then applies the 1-screen settings. With 2 uevents (one for each connector), the compositor refreshes the first disconnected connector state, then applies the 2-screen settings, then refreshes the second disconnected connector state, then applies the 1-screen settings. Applying the 2-screen settings might cause an unnecessarily glitchy intermediate state, where windows move around, modesets are performed, user configured commands are run, just to un-do everything right after.

Quoted 2 lines

I hope this makes sense,

Hi! Seems fine to me :)

Let me know if you need me to push it to drm-misc

Quoted 2 lines

Simon

Signature · 3 lines
Cheers,
 Lyude Paul (she/her)
 Software Engineer at Red Hat
author
#4 lore raw

Thanks for the review, I've pushed the patch.