Skip to contentSkip to conversation

handle vblank when disabling ctc with interrupt disabled

PATCH Superseded Robert Beckett 2 participants 5 messages Jun 28, 2019

Subsystems (2)

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover handle vblank when disabling ctc with interrupt disabled

Add warning when about to send stale vblank.
Revert change that stops vblank info being updated if interrupts already disabled. This fixes a stale vblank timestamp issue seen on drm/imx.

Changes since v2:
Split up the patch in to smaller pieces.
Add warning when about to send bogus vblank event.
Update vblank to best guess info during drm_vblank_disable_and_save.

Changes since v3:
Update cover letter text to describe remaining actions. Drop drm/imx patches as they have now been merged. Replace vblank info estimation patch with a revert of the issue that caused the need for estimation.

Robert Beckett (2):
drm/vblank: warn on sending stale event
Revert "drm/vblank: Do not update vblank count if interrupts are
already disabled."

 drivers/gpu/drm/drm_vblank.c | 35 +++++++++++++++++++++++++----------
 1 file changed, 25 insertions(+), 10 deletions(-)
Signature · 1 line
2.18.0
#1 lore raw

PATCH v4 1/2 drm/vblank: warn on sending stale event

Warn when about to send stale vblank info and add advice to documentation on how to avoid.

---
 drivers/gpu/drm/drm_vblank.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
drivers/gpu/drm/drm_vblank.c +17 -0
index 603ab105125d..7dabb2bdb733 100644
@@ -918,6 +918,19 @@ EXPORT_SYMBOL(drm_crtc_arm_vblank_event);
  *
  * See drm_crtc_arm_vblank_event() for a helper which can be used in certain
  * situation, especially to send out events for atomic commit operations.
+ *
+ * Care should be taken to avoid stale timestamps. If:
+ *   - your driver has vblank support (i.e. dev->num_crtcs > 0)
+ *   - the vblank irq is off (i.e. no one called drm_crtc_vblank_get)
+ *   - from the vblank code's pov the pipe is still running (i.e. not
+ *     in-between a drm_crtc_vblank_off()/on() pair)
+ * If all of these conditions hold then drm_crtc_send_vblank_event is
+ * going to give you a garbage timestamp and and sequence number (the last
+ * recorded before the irq was disabled). If you call drm_crtc_vblank_get/put
+ * around it, or after vblank_off, then either of those will have rolled things
+ * forward for you.
+ * So, drivers should call drm_crtc_vblank_off() before this function in their
+ * crtc atomic_disable handlers.
  */
 void drm_crtc_send_vblank_event(struct drm_crtc *crtc,
 				struct drm_pending_vblank_event *e)
@@ -925,8 +938,12 @@ void drm_crtc_send_vblank_event(struct drm_crtc *crtc,
 	struct drm_device *dev = crtc->dev;
 	u64 seq;
 	unsigned int pipe = drm_crtc_index(crtc);
+	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 	ktime_t now;
 
+	WARN_ONCE(dev->num_crtcs > 0 && !vblank->enabled && !vblank->inmodeset,
+		  "sending stale vblank info\n");
+
 	if (dev->num_crtcs > 0) {
 		seq = drm_vblank_count_and_time(dev, pipe, &now);
 	} else {
Signature · 1 line
2.18.0
#3 lore raw

PATCH v4 2/2 Revert "drm/vblank: Do not update vblank count if interrupts are already disabled."

In reply to Robert Beckett

If interrupts are already disabled, then the timestamp for the vblank does not get updated, causing a stale timestamp to be reported to userland while disabling crtcs.

This reverts commit 68036b08b91bc491ccc308f902616a570a49227c.

---
 drivers/gpu/drm/drm_vblank.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
drivers/gpu/drm/drm_vblank.c +8 -10
index 7dabb2bdb733..aeb9734d7799 100644
@@ -371,25 +371,23 @@ void drm_vblank_disable_and_save(struct drm_device *dev, unsigned int pipe)
 	spin_lock_irqsave(&dev->vblank_time_lock, irqflags);
 
 	/*
-	 * Update vblank count and disable vblank interrupts only if the
-	 * interrupts were enabled. This avoids calling the ->disable_vblank()
-	 * operation in atomic context with the hardware potentially runtime
-	 * suspended.
+	 * Only disable vblank interrupts if they're enabled. This avoids
+	 * calling the ->disable_vblank() operation in atomic context with the
+	 * hardware potentially runtime suspended.
 	 */
-	if (!vblank->enabled)
-		goto out;
+	if (vblank->enabled) {
+		__disable_vblank(dev, pipe);
+		vblank->enabled = false;
+	}
 
 	/*
-	 * Update the count and timestamp to maintain the
+	 * Always update the count and timestamp to maintain the
 	 * appearance that the counter has been ticking all along until
 	 * this time. This makes the count account for the entire time
 	 * between drm_crtc_vblank_on() and drm_crtc_vblank_off().
 	 */
 	drm_update_vblank_count(dev, pipe, false);
-	__disable_vblank(dev, pipe);
-	vblank->enabled = false;
 
-out:
 	spin_unlock_irqrestore(&dev->vblank_time_lock, irqflags);
 }
 
Signature · 1 line
2.18.0
author NAK
#2 lore raw

Re: [PATCH v4 1/2] drm/vblank: warn on sending stale event

In reply to Robert Beckett

Nak - I forgot the requested doc changes. Ill re-send

Quoted 57 lines · Robert Beckett

Warn when about to send stale vblank info and add advice to documentation on how to avoid.

---
 drivers/gpu/drm/drm_vblank.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
drivers/gpu/drm/drm_vblank.c +0 -0
index 603ab105125d..7dabb2bdb733 100644
@@ -918,6 +918,19 @@ EXPORT_SYMBOL(drm_crtc_arm_vblank_event);
  *
  * See drm_crtc_arm_vblank_event() for a helper which can be used in

certain
* situation, especially to send out events for atomic commit
operations.
+ *
+ * Care should be taken to avoid stale timestamps. If:
+ * - your driver has vblank support (i.e. dev->num_crtcs > 0)
+ * - the vblank irq is off (i.e. no one called
drm_crtc_vblank_get)
+ * - from the vblank code's pov the pipe is still running (i.e.
not
+ * in-between a drm_crtc_vblank_off()/on() pair)
+ * If all of these conditions hold then drm_crtc_send_vblank_event is
+ * going to give you a garbage timestamp and and sequence number (the last
+ * recorded before the irq was disabled). If you call drm_crtc_vblank_get/put
+ * around it, or after vblank_off, then either of those will have rolled things
+ * forward for you.
+ * So, drivers should call drm_crtc_vblank_off() before this function in their
+ * crtc atomic_disable handlers.
*/
void drm_crtc_send_vblank_event(struct drm_crtc *crtc,

drivers/gpu/drm/drm_vblank.c +4 -0
 				struct drm_pending_vblank_event *e)
@@ -925,8 +938,12 @@ void drm_crtc_send_vblank_event(struct drm_crtc
 *crtc,
 	struct drm_device *dev = crtc->dev;
 	u64 seq;
 	unsigned int pipe = drm_crtc_index(crtc);
+	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 	ktime_t now;
 
+	WARN_ONCE(dev->num_crtcs > 0 && !vblank->enabled && !vblank-
 >inmodeset,
+		  "sending stale vblank info\n");
+
 	if (dev->num_crtcs > 0) {
drivers/gpu/drm/drm_vblank.c +0 -0
 		seq = drm_vblank_count_and_time(dev, pipe, &now);
 	} else {
#4 lore raw

Re: [PATCH v4 2/2] Revert "drm/vblank: Do not update vblank count if interrupts are already disabled."

In reply to Robert Beckett
Quoted 5 lines · Robert Beckett

If interrupts are already disabled, then the timestamp for the vblank does not get updated, causing a stale timestamp to be reported to userland while disabling crtcs.

This reverts commit 68036b08b91bc491ccc308f902616a570a49227c.

Please cc the people involved in that. And I'd drop the lkml cc.

Quoted 51 lines
---
 drivers/gpu/drm/drm_vblank.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
drivers/gpu/drm/drm_vblank.c +8 -10
index 7dabb2bdb733..aeb9734d7799 100644
@@ -371,25 +371,23 @@ void drm_vblank_disable_and_save(struct drm_device *dev, unsigned int pipe)
 	spin_lock_irqsave(&dev->vblank_time_lock, irqflags);
 
 	/*
-	 * Update vblank count and disable vblank interrupts only if the
-	 * interrupts were enabled. This avoids calling the ->disable_vblank()
-	 * operation in atomic context with the hardware potentially runtime
-	 * suspended.
+	 * Only disable vblank interrupts if they're enabled. This avoids
+	 * calling the ->disable_vblank() operation in atomic context with the
+	 * hardware potentially runtime suspended.
 	 */
-	if (!vblank->enabled)
-		goto out;
+	if (vblank->enabled) {
+		__disable_vblank(dev, pipe);
+		vblank->enabled = false;
+	}
 
 	/*
-	 * Update the count and timestamp to maintain the
+	 * Always update the count and timestamp to maintain the
 	 * appearance that the counter has been ticking all along until
 	 * this time. This makes the count account for the entire time
 	 * between drm_crtc_vblank_on() and drm_crtc_vblank_off().
 	 */
 	drm_update_vblank_count(dev, pipe, false);
-	__disable_vblank(dev, pipe);
-	vblank->enabled = false;
 
-out:
 	spin_unlock_irqrestore(&dev->vblank_time_lock, irqflags);
 }
 
Signature · 6 lines
2.18.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Signature · 6 lines
Ville Syrjälä
Intel
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel