Skip to contentSkip to conversation

drm/vblank: Fix return type for drm_vblank_count()

PATCH In review Dhinakaran Pandiyan 5 participants 24 messages Jan 12, 2018 – Jan 31, 2018

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/5 drm/vblank: Fix return type for drm_vblank_count()

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +1 -1
index 32d9bcf5be7f..768a8e44d99b 100644
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 
Signature · 6 lines
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
#1 lore raw

PATCH 2/5 drm/vblank: Fix data type width for drm_crtc_arm_vblank_event()

Now that drm_vblank_count() returns all bits of the vblank count, update drm_crtc_arm_vblank_event() so that it queues the correct sequence. Otherwise, this leads to prolonged waits for a vblank sequence when the current count is >=2^32.

Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 4 ++--
 include/drm/drm_vblank.h     | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_vblank.c +2 -2
index 768a8e44d99b..f2bf1f5dbaa5 100644
@@ -292,11 +292,11 @@ static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
  * This is mostly useful for hardware that can obtain the scanout position, but
  * doesn't have a hardware frame counter.
  */
-u32 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc)
+u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc)
 {
 	struct drm_device *dev = crtc->dev;
 	unsigned int pipe = drm_crtc_index(crtc);
-	u32 vblank;
+	u64 vblank;
 	unsigned long flags;
 
 	WARN_ONCE(drm_debug & DRM_UT_VBL && !dev->driver->get_vblank_timestamp,
include/drm/drm_vblank.h +1 -1
index 848b463a0af5..a4c3b0a0a197 100644
@@ -179,7 +179,7 @@ void drm_crtc_wait_one_vblank(struct drm_crtc *crtc);
 void drm_crtc_vblank_off(struct drm_crtc *crtc);
 void drm_crtc_vblank_reset(struct drm_crtc *crtc);
 void drm_crtc_vblank_on(struct drm_crtc *crtc);
-u32 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
+u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
 
 bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 					   unsigned int pipe, int *max_error,
Signature · 6 lines
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
maintainer Reviewed-by on v1 2/5
#2 lore raw

Re: [PATCH 2/5] drm/vblank: Fix data type width for drm_crtc_arm_vblank_event()

Quoted 4 lines · Dhinakaran Pandiyan

Now that drm_vblank_count() returns all bits of the vblank count, update drm_crtc_arm_vblank_event() so that it queues the correct sequence. Otherwise, this leads to prolonged waits for a vblank sequence when the current count is >=2^32.

This could be probably squashed to the previous patch... Specially if you apply the Fixes tag.

Anyways, in case you decide to go with 2 patches:

Quoted 44 lines
Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 4 ++--
 include/drm/drm_vblank.h     | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
drivers/gpu/drm/drm_vblank.c +2 -2
index 768a8e44d99b..f2bf1f5dbaa5 100644
@@ -292,11 +292,11 @@ static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
  * This is mostly useful for hardware that can obtain the scanout position, but
  * doesn't have a hardware frame counter.
  */
-u32 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc)
+u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc)
 {
 	struct drm_device *dev = crtc->dev;
 	unsigned int pipe = drm_crtc_index(crtc);
-	u32 vblank;
+	u64 vblank;
 	unsigned long flags;
 
 	WARN_ONCE(drm_debug & DRM_UT_VBL && !dev->driver->get_vblank_timestamp,
include/drm/drm_vblank.h +1 -1
index 848b463a0af5..a4c3b0a0a197 100644
@@ -179,7 +179,7 @@ void drm_crtc_wait_one_vblank(struct drm_crtc *crtc);
 void drm_crtc_vblank_off(struct drm_crtc *crtc);
 void drm_crtc_vblank_reset(struct drm_crtc *crtc);
 void drm_crtc_vblank_on(struct drm_crtc *crtc);
-u32 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
+u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
 
 bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 					   unsigned int pipe, int *max_error,
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#3 lore raw

Re: [PATCH 2/5] drm/vblank: Fix data type width for drm_crtc_arm_vblank_event()

In reply to Dhinakaran Pandiyan
Quoted 4 lines · Dhinakaran Pandiyan

Now that drm_vblank_count() returns all bits of the vblank count, update drm_crtc_arm_vblank_event() so that it queues the correct sequence. Otherwise, this leads to prolonged waits for a vblank sequence when the current count is >=2^32.

The summary for this patch is incorrect; the function being fixed is drm_crtc_accurate_vblank_event.

And, I'm afraid you've uncovered a rabbit hole here -- there are a couple of users of this function outside of the core, including i915 in a couple of places and nouveau. We should at least review the whole call graph starting here and make sure it does what we think it should.

Thanks for finding these bugs!

Signature · 1 line
-keith

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#4 lore raw

PATCH 3/5 drm/vblank: Do not update vblank count if interrupts are already disabled.

In reply to Dhinakaran Pandiyan

Updating vblank counts requires register reads and these reads may not return meaningful values if the device was in a low power state after vblank interrupts were last disabled. So, update the count only if vblank interrupts are enabled. Secondly, this means the registers should be read before disabling vblank interrupts.

v2: Don't check vblank->enabled outside it's lock (Chris)

Cc Chris Wilson <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Michel Dänzer <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
drivers/gpu/drm/drm_vblank.c +10 -8
index f2bf1f5dbaa5..2559d2d7b907 100644
@@ -347,23 +347,25 @@ void drm_vblank_disable_and_save(struct drm_device *dev, unsigned int pipe)
 	spin_lock_irqsave(&dev->vblank_time_lock, irqflags);
 
 	/*
-	 * Only disable vblank interrupts if they're enabled. This avoids
-	 * calling the ->disable_vblank() operation in atomic context with the
-	 * hardware potentially runtime suspended.
+	 * 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.
 	 */
-	if (vblank->enabled) {
-		__disable_vblank(dev, pipe);
-		vblank->enabled = false;
-	}
+	if (!vblank->enabled)
+		goto out;
 
 	/*
-	 * Always update the count and timestamp to maintain the
+	 * 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
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer Reviewed-by on v1 3/5
#5 lore raw

Re: [PATCH 3/5] drm/vblank: Do not update vblank count if interrupts are already disabled.

Quoted 36 lines · Dhinakaran Pandiyan

Updating vblank counts requires register reads and these reads may not return meaningful values if the device was in a low power state after vblank interrupts were last disabled. So, update the count only if vblank interrupts are enabled. Secondly, this means the registers should be read before disabling vblank interrupts.

v2: Don't check vblank->enabled outside it's lock (Chris)

Cc Chris Wilson <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Michel Dänzer <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
drivers/gpu/drm/drm_vblank.c +4 -7
index f2bf1f5dbaa5..2559d2d7b907 100644
@@ -347,23 +347,25 @@ void drm_vblank_disable_and_save(struct drm_device *dev, unsigned int pipe)
 	spin_lock_irqsave(&dev->vblank_time_lock, irqflags);
 
 	/*
-	 * Only disable vblank interrupts if they're enabled. This avoids
-	 * calling the ->disable_vblank() operation in atomic context with the
-	 * hardware potentially runtime suspended.
+	 * 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.
 	 */
-	if (vblank->enabled) {
-		__disable_vblank(dev, pipe);
-		vblank->enabled = false;
-	}

drivers/gpu/drm/drm_vblank.c void drm_vblank_disable_and_save(struct drm_device *dev, unsigned int pipe) · line 363

Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +3 -1
+	if (!vblank->enabled)
+		goto out;
 
 	/*
-	 * Always update the count and timestamp to maintain the
+	 * 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().
 	 */

I feel that this entire comment can be simply removed now... The approach looks good and right to me so you can use

but please ping Ville to take a look here since he introduced this approach with 4dfd64862ff8 ("drm: Use vblank timestamps to guesstimate how many vblanks were missed")

Quoted 11 lines
drivers/gpu/drm/drm_vblank.c +3 -0
 	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
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#6 lore raw

PATCH 4/5 drm/vblank: Restoring vblank counts after device PM events.

In reply to Dhinakaran Pandiyan

The HW frame counter can get reset if device enters a low power state after vblank interrupts were disabled. This messes up any following vblank count update as a negative diff (huge unsigned diff) is calculated from the HW frame counter change. We cannot ignore negative diffs altogther as there could be legitimate wrap arounds. So, allow drivers to update vblank->count with missed vblanks for the time interrupts were disabled. This is similar to _crtc_vblank_on() except that vblanks interrupts are not enabled at the end as this function is expected to be called from the driver _enable_vblank() vfunc.

v2: drm_crtc_vblank_restore should take crtc as arg. (Chris)
Add docs and sprinkle some asserts.

Cc Daniel Vetter <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Michel Dänzer <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 59 ++++++++++++++++++++++++++++++++++++++++++++
 include/drm/drm_vblank.h     |  2 ++
 2 files changed, 61 insertions(+)
drivers/gpu/drm/drm_vblank.c +59 -0
index 2559d2d7b907..2690966694f0 100644
@@ -1237,6 +1237,65 @@ void drm_crtc_vblank_on(struct drm_crtc *crtc)
 }
 EXPORT_SYMBOL(drm_crtc_vblank_on);
 
+/**
+ * drm_vblank_restore - estimated vblanks using timestamps and update it.
+ *
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ *
+ * This function is the legacy version of drm_crtc_vblank_restore().
+ */
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe)
+{
+	ktime_t t_vblank;
+	struct drm_vblank_crtc *vblank;
+	int framedur_ns;
+	u64 diff_ns;
+	u32 cur_vblank, diff = 1;
+	int count = DRM_TIMESTAMP_MAXRETRIES;
+
+	if (WARN_ON(pipe >= dev->num_crtcs))
+		return;
+
+	assert_spin_locked(&dev->vbl_lock);
+	assert_spin_locked(&dev->vblank_time_lock);
+
+	vblank = &dev->vblank[pipe];
+	WARN_ONCE((drm_debug & DRM_UT_VBL) && !vblank->framedur_ns,
+		  "Cannot compute missed vblanks without frame duration\n");
+	framedur_ns = vblank->framedur_ns;
+
+	do {
+		cur_vblank = __get_vblank_counter(dev, pipe);
+		drm_get_last_vbltimestamp(dev, pipe, &t_vblank, false);
+	} while (cur_vblank != __get_vblank_counter(dev, pipe) && --count > 0);
+
+	diff_ns = ktime_to_ns(ktime_sub(t_vblank, vblank->time));
+	if (framedur_ns)
+		diff = DIV_ROUND_CLOSEST_ULL(diff_ns, framedur_ns);
+
+
+	DRM_DEBUG_VBL("missed %d vblanks in %lld ns, frame duration=%d ns, hw_diff=%d\n",
+		      diff, diff_ns, framedur_ns, cur_vblank - vblank->last);
+	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
+}
+EXPORT_SYMBOL(drm_vblank_restore);
+
+/**
+ * drm_crtc_vblank_restore - estimate vblanks using timestamps and update it.
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ */
+void drm_crtc_vblank_restore(struct drm_crtc *crtc)
+{
+	drm_vblank_restore(crtc->dev, drm_crtc_index(crtc));
+}
+EXPORT_SYMBOL(drm_crtc_vblank_restore);
+
 static void drm_legacy_vblank_pre_modeset(struct drm_device *dev,
 					  unsigned int pipe)
 {
include/drm/drm_vblank.h +2 -0
index a4c3b0a0a197..16d46e2a6854 100644
@@ -180,6 +180,8 @@ void drm_crtc_vblank_off(struct drm_crtc *crtc);
 void drm_crtc_vblank_reset(struct drm_crtc *crtc);
 void drm_crtc_vblank_on(struct drm_crtc *crtc);
 u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe);
+void drm_crtc_vblank_restore(struct drm_crtc *crtc);
 
 bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 					   unsigned int pipe, int *max_error,
Signature · 6 lines
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
maintainer
#7 lore raw

Re: [PATCH 4/5] drm/vblank: Restoring vblank counts after device PM events.

Quoted 47 lines · Dhinakaran Pandiyan

The HW frame counter can get reset if device enters a low power state after vblank interrupts were disabled. This messes up any following vblank count update as a negative diff (huge unsigned diff) is calculated from the HW frame counter change. We cannot ignore negative diffs altogther as there could be legitimate wrap arounds. So, allow drivers to update vblank->count with missed vblanks for the time interrupts were disabled. This is similar to _crtc_vblank_on() except that vblanks interrupts are not enabled at the end as this function is expected to be called from the driver _enable_vblank() vfunc.

v2: drm_crtc_vblank_restore should take crtc as arg. (Chris)
Add docs and sprinkle some asserts.

Cc Daniel Vetter <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Michel Dänzer <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 59 ++++++++++++++++++++++++++++++++++++++++++++
 include/drm/drm_vblank.h     |  2 ++
 2 files changed, 61 insertions(+)
drivers/gpu/drm/drm_vblank.c +17 -0
index 2559d2d7b907..2690966694f0 100644
@@ -1237,6 +1237,65 @@ void drm_crtc_vblank_on(struct drm_crtc *crtc)
 }
 EXPORT_SYMBOL(drm_crtc_vblank_on);
 
+/**
+ * drm_vblank_restore - estimated vblanks using timestamps and update it.
+ *
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ *
+ * This function is the legacy version of drm_crtc_vblank_restore().
+ */
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe)
+{
+	ktime_t t_vblank;
+	struct drm_vblank_crtc *vblank;
+	int framedur_ns;
+	u64 diff_ns;
+	u32 cur_vblank, diff = 1;

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1266

Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10 -0
+	int count = DRM_TIMESTAMP_MAXRETRIES;
+
+	if (WARN_ON(pipe >= dev->num_crtcs))
+		return;
+
+	assert_spin_locked(&dev->vbl_lock);
+	assert_spin_locked(&dev->vblank_time_lock);
+
+	vblank = &dev->vblank[pipe];
+	WARN_ONCE((drm_debug & DRM_UT_VBL) && !vblank->framedur_ns,

do we really only need this warn on debug vlb?

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1267

Quoted 1 line
drivers/gpu/drm/drm_vblank.c +1 -0
+		  "Cannot compute missed vblanks without frame duration\n");

The message seems hard... if we *cannot* why do we move fwd?

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1273

Quoted 6 lines
drivers/gpu/drm/drm_vblank.c +6 -0
+	framedur_ns = vblank->framedur_ns;
+
+	do {
+		cur_vblank = __get_vblank_counter(dev, pipe);
+		drm_get_last_vbltimestamp(dev, pipe, &t_vblank, false);
+	} while (cur_vblank != __get_vblank_counter(dev, pipe) && --count > 0);

Based on the commend of drm_update_vblank_count I don't feel that we have to do the loop here... And if we have maybe we should re-org things to avoid duplication?

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1282

Quoted 9 lines
drivers/gpu/drm/drm_vblank.c +9 -0
+
+	diff_ns = ktime_to_ns(ktime_sub(t_vblank, vblank->time));
+	if (framedur_ns)
+		diff = DIV_ROUND_CLOSEST_ULL(diff_ns, framedur_ns);
+
+
+	DRM_DEBUG_VBL("missed %d vblanks in %lld ns, frame duration=%d ns, hw_diff=%d\n",
+		      diff, diff_ns, framedur_ns, cur_vblank - vblank->last);
+	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);

hm... I wonder if the simple store_vblank(... __get_vblank_counter(dev, pipe)); wouldn't work here...

Quoted 35 lines
drivers/gpu/drm/drm_vblank.c +16 -0
+}
+EXPORT_SYMBOL(drm_vblank_restore);
+
+/**
+ * drm_crtc_vblank_restore - estimate vblanks using timestamps and update it.
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ */
+void drm_crtc_vblank_restore(struct drm_crtc *crtc)
+{
+	drm_vblank_restore(crtc->dev, drm_crtc_index(crtc));
+}
+EXPORT_SYMBOL(drm_crtc_vblank_restore);
+
 static void drm_legacy_vblank_pre_modeset(struct drm_device *dev,
 					  unsigned int pipe)
 {
include/drm/drm_vblank.h +2 -0
index a4c3b0a0a197..16d46e2a6854 100644
@@ -180,6 +180,8 @@ void drm_crtc_vblank_off(struct drm_crtc *crtc);
 void drm_crtc_vblank_reset(struct drm_crtc *crtc);
 void drm_crtc_vblank_on(struct drm_crtc *crtc);
 u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe);
+void drm_crtc_vblank_restore(struct drm_crtc *crtc);
 
 bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 					   unsigned int pipe, int *max_error,
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#8 lore raw

Re: [PATCH 4/5] drm/vblank: Restoring vblank counts after device PM events.

Quoted 61 lines · Rodrigo Vivi
Quoted 47 lines · Dhinakaran Pandiyan

The HW frame counter can get reset if device enters a low power state after vblank interrupts were disabled. This messes up any following vblank count update as a negative diff (huge unsigned diff) is calculated from the HW frame counter change. We cannot ignore negative diffs altogther as there could be legitimate wrap arounds. So, allow drivers to update vblank->count with missed vblanks for the time interrupts were disabled. This is similar to _crtc_vblank_on() except that vblanks interrupts are not enabled at the end as this function is expected to be called from the driver _enable_vblank() vfunc.

v2: drm_crtc_vblank_restore should take crtc as arg. (Chris)
Add docs and sprinkle some asserts.

Cc Daniel Vetter <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Michel Dänzer <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 59 ++++++++++++++++++++++++++++++++++++++++++++
 include/drm/drm_vblank.h     |  2 ++
 2 files changed, 61 insertions(+)
drivers/gpu/drm/drm_vblank.c +17 -0
index 2559d2d7b907..2690966694f0 100644
@@ -1237,6 +1237,65 @@ void drm_crtc_vblank_on(struct drm_crtc *crtc)
 }
 EXPORT_SYMBOL(drm_crtc_vblank_on);
 
+/**
+ * drm_vblank_restore - estimated vblanks using timestamps and update it.
+ *
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ *
+ * This function is the legacy version of drm_crtc_vblank_restore().
+ */
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe)
+{
+	ktime_t t_vblank;
+	struct drm_vblank_crtc *vblank;
+	int framedur_ns;
+	u64 diff_ns;
+	u32 cur_vblank, diff = 1;

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1266

Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10 -0
+	int count = DRM_TIMESTAMP_MAXRETRIES;
+
+	if (WARN_ON(pipe >= dev->num_crtcs))
+		return;
+
+	assert_spin_locked(&dev->vbl_lock);
+	assert_spin_locked(&dev->vblank_time_lock);
+
+	vblank = &dev->vblank[pipe];
+	WARN_ONCE((drm_debug & DRM_UT_VBL) && !vblank->framedur_ns,

do we really only need this warn on debug vlb?

Quoted 3 lines

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1267

Quoted 1 line
drivers/gpu/drm/drm_vblank.c +1 -0
+		  "Cannot compute missed vblanks without frame duration\n");

The message seems hard... if we *cannot* why do we move fwd?

We assume the diff is 1 and make an update, some kind of a default similar to what is implemented in drm_update_vblank_count()

Quoted 9 lines

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1273

Quoted 7 lines
drivers/gpu/drm/drm_vblank.c +6 -0
+	framedur_ns = vblank->framedur_ns;
+
+	do {
+		cur_vblank = __get_vblank_counter(dev, pipe);
+		drm_get_last_vbltimestamp(dev, pipe, &t_vblank, false);
+	} while (cur_vblank != __get_vblank_counter(dev, pipe) && --count > 0);

Based on the commend of drm_update_vblank_count I don't feel that we have to

This one? -
"* We repeat the hardware vblank counter & timestamp query until
* we get consistent results. This to prevent races between gpu
* updating its hardware counter while we are retrieving the
* corresponding vblank timestamp."

I added the loop based on that comment. If the register if partially updated, we want to discard that and loop until it is stable.

Quoted 3 lines

do the loop here... And if we have maybe we should re-org things to avoid duplication?

I considered that, we need to pass at least four args for three lines of code. Felt it was too small to warrant a separate function.

Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/drm_vblank.c +4 -0
+
+	diff_ns = ktime_to_ns(ktime_sub(t_vblank, vblank->time));
+	if (framedur_ns)
+		diff = DIV_ROUND_CLOSEST_ULL(diff_ns, framedur_ns);
Quoted 8 lines

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1282

Quoted 5 lines
drivers/gpu/drm/drm_vblank.c +5 -0
+
+
+	DRM_DEBUG_VBL("missed %d vblanks in %lld ns, frame duration=%d ns, hw_diff=%d\n",
+		      diff, diff_ns, framedur_ns, cur_vblank - vblank->last);
+	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);

hm... I wonder if the simple store_vblank(... __get_vblank_counter(dev, pipe)); wouldn't work here...

We have to store a stable count.

Quoted 31 lines
Quoted 31 lines
drivers/gpu/drm/drm_vblank.c +16 -0
+}
+EXPORT_SYMBOL(drm_vblank_restore);
+
+/**
+ * drm_crtc_vblank_restore - estimate vblanks using timestamps and update it.
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ */
+void drm_crtc_vblank_restore(struct drm_crtc *crtc)
+{
+	drm_vblank_restore(crtc->dev, drm_crtc_index(crtc));
+}
+EXPORT_SYMBOL(drm_crtc_vblank_restore);
+
 static void drm_legacy_vblank_pre_modeset(struct drm_device *dev,
 					  unsigned int pipe)
 {
include/drm/drm_vblank.h +2 -0
index a4c3b0a0a197..16d46e2a6854 100644
@@ -180,6 +180,8 @@ void drm_crtc_vblank_off(struct drm_crtc *crtc);
 void drm_crtc_vblank_reset(struct drm_crtc *crtc);
 void drm_crtc_vblank_on(struct drm_crtc *crtc);
 u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe);
+void drm_crtc_vblank_restore(struct drm_crtc *crtc);
 
Quoted 9 lines
Quoted 5 lines
include/drm/drm_vblank.h +0 -0
 bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 					   unsigned int pipe, int *max_error,
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

maintainer Reviewed-by
#9 lore raw

Re: [Intel-gfx] [PATCH 4/5] drm/vblank: Restoring vblank counts after device PM events.

Quoted 115 lines · Dhinakaran
Quoted 61 lines · Rodrigo Vivi
Quoted 47 lines · Dhinakaran Pandiyan

The HW frame counter can get reset if device enters a low power state after vblank interrupts were disabled. This messes up any following vblank count update as a negative diff (huge unsigned diff) is calculated from the HW frame counter change. We cannot ignore negative diffs altogther as there could be legitimate wrap arounds. So, allow drivers to update vblank->count with missed vblanks for the time interrupts were disabled. This is similar to _crtc_vblank_on() except that vblanks interrupts are not enabled at the end as this function is expected to be called from the driver _enable_vblank() vfunc.

v2: drm_crtc_vblank_restore should take crtc as arg. (Chris)
Add docs and sprinkle some asserts.

Cc Daniel Vetter <[email protected]>
Cc Chris Wilson <[email protected]>
Cc Michel Dänzer <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 59 ++++++++++++++++++++++++++++++++++++++++++++
 include/drm/drm_vblank.h     |  2 ++
 2 files changed, 61 insertions(+)
drivers/gpu/drm/drm_vblank.c +17 -0
index 2559d2d7b907..2690966694f0 100644
@@ -1237,6 +1237,65 @@ void drm_crtc_vblank_on(struct drm_crtc *crtc)
 }
 EXPORT_SYMBOL(drm_crtc_vblank_on);
 
+/**
+ * drm_vblank_restore - estimated vblanks using timestamps and update it.
+ *
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ *
+ * This function is the legacy version of drm_crtc_vblank_restore().
+ */
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe)
+{
+	ktime_t t_vblank;
+	struct drm_vblank_crtc *vblank;
+	int framedur_ns;
+	u64 diff_ns;
+	u32 cur_vblank, diff = 1;

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1266

Quoted 10 lines
drivers/gpu/drm/drm_vblank.c +10 -0
+	int count = DRM_TIMESTAMP_MAXRETRIES;
+
+	if (WARN_ON(pipe >= dev->num_crtcs))
+		return;
+
+	assert_spin_locked(&dev->vbl_lock);
+	assert_spin_locked(&dev->vblank_time_lock);
+
+	vblank = &dev->vblank[pipe];
+	WARN_ONCE((drm_debug & DRM_UT_VBL) && !vblank->framedur_ns,

do we really only need this warn on debug vlb?

Quoted 3 lines

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1267

Quoted 1 line
drivers/gpu/drm/drm_vblank.c +1 -0
+		  "Cannot compute missed vblanks without frame duration\n");

The message seems hard... if we *cannot* why do we move fwd?

We assume the diff is 1 and make an update, some kind of a default similar to what is implemented in drm_update_vblank_count()

Quoted 9 lines

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1273

Quoted 7 lines
drivers/gpu/drm/drm_vblank.c +6 -0
+	framedur_ns = vblank->framedur_ns;
+
+	do {
+		cur_vblank = __get_vblank_counter(dev, pipe);
+		drm_get_last_vbltimestamp(dev, pipe, &t_vblank, false);
+	} while (cur_vblank != __get_vblank_counter(dev, pipe) && --count > 0);

Based on the commend of drm_update_vblank_count I don't feel that we have to

This one? -
"* We repeat the hardware vblank counter & timestamp query until
* we get consistent results. This to prevent races between gpu
* updating its hardware counter while we are retrieving the
* corresponding vblank timestamp."

I added the loop based on that comment. If the register if partially updated, we want to discard that and loop until it is stable.

Quoted 3 lines

do the loop here... And if we have maybe we should re-org things to avoid duplication?

I considered that, we need to pass at least four args for three lines of code. Felt it was too small to warrant a separate function.

Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/drm_vblank.c +4 -0
+
+	diff_ns = ktime_to_ns(ktime_sub(t_vblank, vblank->time));
+	if (framedur_ns)
+		diff = DIV_ROUND_CLOSEST_ULL(diff_ns, framedur_ns);
Quoted 8 lines

drivers/gpu/drm/drm_vblank.c void drm_crtc_vblank_on(struct drm_crtc *crtc) · line 1282

Quoted 5 lines
drivers/gpu/drm/drm_vblank.c +5 -0
+
+
+	DRM_DEBUG_VBL("missed %d vblanks in %lld ns, frame duration=%d ns, hw_diff=%d\n",
+		      diff, diff_ns, framedur_ns, cur_vblank - vblank->last);
+	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);

hm... I wonder if the simple store_vblank(... __get_vblank_counter(dev, pipe)); wouldn't work here...

Quoted 1 line

We have to store a stable count.

I don't see why our counter wouldn't be stable at that point?

Our register is zero or the proper counter. Our frm_counter reg doesn't stop when vblank interrupts are disabled afaik...

But well, I see your point of making something more robust, using the well known and tested methods and preparing for the case where someone start using this call in other platforms as well.... So, whatever you decide..

If you go wit this:

Quoted 32 lines
Quoted 32 lines
Quoted 32 lines
drivers/gpu/drm/drm_vblank.c +16 -0
+}
+EXPORT_SYMBOL(drm_vblank_restore);
+
+/**
+ * drm_crtc_vblank_restore - estimate vblanks using timestamps and update it.
+ * Power manamement features can cause frame counter resets between vblank
+ * disable and enable. Drivers can then use this function in their
+ * &drm_crtc_funcs.enable_vblank implementation to estimate the vblanks since
+ * the last &drm_crtc_funcs.disable_vblank.
+ */
+void drm_crtc_vblank_restore(struct drm_crtc *crtc)
+{
+	drm_vblank_restore(crtc->dev, drm_crtc_index(crtc));
+}
+EXPORT_SYMBOL(drm_crtc_vblank_restore);
+
 static void drm_legacy_vblank_pre_modeset(struct drm_device *dev,
 					  unsigned int pipe)
 {
include/drm/drm_vblank.h +2 -0
index a4c3b0a0a197..16d46e2a6854 100644
@@ -180,6 +180,8 @@ void drm_crtc_vblank_off(struct drm_crtc *crtc);
 void drm_crtc_vblank_reset(struct drm_crtc *crtc);
 void drm_crtc_vblank_on(struct drm_crtc *crtc);
 u64 drm_crtc_accurate_vblank_count(struct drm_crtc *crtc);
+void drm_vblank_restore(struct drm_device *dev, unsigned int pipe);
+void drm_crtc_vblank_restore(struct drm_crtc *crtc);
 
Quoted 9 lines
Quoted 9 lines
Quoted 5 lines
include/drm/drm_vblank.h +0 -0
 bool drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
 					   unsigned int pipe, int *max_error,
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#10 lore raw

PATCH 5/5 drm/i915: Estimate and update missed vblanks.

In reply to Dhinakaran Pandiyan

The frame counter may have got reset between disabling and enabling vblank interrupts due to DMC putting the hardware to DC5/6 state if PSR was active. The frame counter also could have stalled if PSR is active in cases where there is no DMC. The frame counter resetting as a user visible impact of screen freezes. Use drm_vblank_restore() to compute missed vblanks in the duration for which vblank interrupts are disabled. There's no need particularly check if PSR was active in the interrupt disabled duration.

Enabling vblank interrupts wakes up the hardware from DC5/6 and prevents it from going back again as long as the there are pending interrupts. So, we don't have to explicity disallow DC5/6 after enabling vblank interrupts to keep the counter running.

Let's not apply this to CHV for now, as enabling interrupts does not prevent the hardware from activating PSR and thereby stalling the counter.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +6 -0
index 3517c6548e2c..db3466ec6faa 100644
@@ -2956,6 +2956,9 @@ static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	ilk_enable_display_irq(dev_priv, bit);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
@@ -2968,6 +2971,9 @@ static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	bdw_enable_pipe_irq(dev_priv, pipe, GEN8_PIPE_VBLANK);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
Signature · 6 lines
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Acked-by on v1 5/5
#11 lore raw

Re: [PATCH 5/5] drm/i915: Estimate and update missed vblanks.

Quoted 13 lines · Dhinakaran Pandiyan

The frame counter may have got reset between disabling and enabling vblank interrupts due to DMC putting the hardware to DC5/6 state if PSR was active. The frame counter also could have stalled if PSR is active in cases where there is no DMC. The frame counter resetting as a user visible impact of screen freezes. Use drm_vblank_restore() to compute missed vblanks in the duration for which vblank interrupts are disabled. There's no need particularly check if PSR was active in the interrupt disabled duration.

Enabling vblank interrupts wakes up the hardware from DC5/6 and prevents it from going back again as long as the there are pending interrupts. So, we don't have to explicity disallow DC5/6 after enabling vblank interrupts to keep the counter running.

Quoted 5 lines

Let's not apply this to CHV for now, as enabling interrupts does not prevent the hardware from activating PSR and thereby stalling the counter.

Read this series, and I think compared to all the previous attempts to fix this issue this one here seems the cleanest. Not an expert on the vblank code (Michel or Ville are best for that), but on the series:

Cheers, Daniel

Quoted 35 lines
---
 drivers/gpu/drm/i915/i915_irq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +6 -0
index 3517c6548e2c..db3466ec6faa 100644
@@ -2956,6 +2956,9 @@ static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	ilk_enable_display_irq(dev_priv, bit);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
@@ -2968,6 +2971,9 @@ static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	bdw_enable_pipe_irq(dev_priv, pipe, GEN8_PIPE_VBLANK);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
Signature · 6 lines
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
maintainer
#12 lore raw

Re: [PATCH 5/5] drm/i915: Estimate and update missed vblanks.

In reply to Dhinakaran Pandiyan
Quoted 27 lines · Dhinakaran Pandiyan

The frame counter may have got reset between disabling and enabling vblank interrupts due to DMC putting the hardware to DC5/6 state if PSR was active. The frame counter also could have stalled if PSR is active in cases where there is no DMC. The frame counter resetting as a user visible impact of screen freezes. Use drm_vblank_restore() to compute missed vblanks in the duration for which vblank interrupts are disabled. There's no need particularly check if PSR was active in the interrupt disabled duration.

Enabling vblank interrupts wakes up the hardware from DC5/6 and prevents it from going back again as long as the there are pending interrupts. So, we don't have to explicity disallow DC5/6 after enabling vblank interrupts to keep the counter running.

Let's not apply this to CHV for now, as enabling interrupts does not prevent the hardware from activating PSR and thereby stalling the counter.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +0 -0
index 3517c6548e2c..db3466ec6faa 100644

drivers/gpu/drm/i915/i915_irq.c static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe) · line 2961

Quoted 6 lines
drivers/gpu/drm/i915/i915_irq.c +3 -0
@@ -2956,6 +2956,9 @@ static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	ilk_enable_display_irq(dev_priv, bit);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+

I don't believe we need this one here.

pre-gen9 platform has psr but not dmc, so the case where we need to restore the counter doesn't exist.

drivers/gpu/drm/i915/i915_irq.c static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe) · line 2974

Quoted 8 lines
drivers/gpu/drm/i915/i915_irq.c +1 -0
 	return 0;
 }
 
@@ -2968,6 +2971,9 @@ static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	bdw_enable_pipe_irq(dev_priv, pipe, GEN8_PIPE_VBLANK);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))

HAS_PSR(dev_priv) && HAS_CSR(dev_priv)
maybe?
So it gets clear that it is not because PSR that we need to restore the counter, but also don't do it when not needed.

Quoted 8 lines
drivers/gpu/drm/i915/i915_irq.c +2 -0
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
Signature · 1 line
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#13 lore raw

Re: [Intel-gfx] [PATCH 5/5] drm/i915: Estimate and update missed vblanks.

Quoted 31 lines · Rodrigo Vivi
Quoted 30 lines · Dhinakaran Pandiyan

The frame counter may have got reset between disabling and enabling vblank interrupts due to DMC putting the hardware to DC5/6 state if PSR was active. The frame counter also could have stalled if PSR is active in cases where there is no DMC. The frame counter resetting as a user visible impact of screen freezes. Use drm_vblank_restore() to compute missed vblanks in the duration for which vblank interrupts are disabled. There's no need particularly check if PSR was active in the interrupt disabled duration.

Enabling vblank interrupts wakes up the hardware from DC5/6 and prevents it from going back again as long as the there are pending interrupts. So, we don't have to explicity disallow DC5/6 after enabling vblank interrupts to keep the counter running.

Let's not apply this to CHV for now, as enabling interrupts does not prevent the hardware from activating PSR and thereby stalling the counter.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +0 -0
index 3517c6548e2c..db3466ec6faa 100644
@@ -2956,6 +2956,9 @@ static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	ilk_enable_display_irq(dev_priv, bit);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
Quoted 8 lines

drivers/gpu/drm/i915/i915_irq.c static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe) · line 2961

Quoted 3 lines
drivers/gpu/drm/i915/i915_irq.c +3 -0
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+

I don't believe we need this one here.

pre-gen9 platform has psr but not dmc, so the case where we need to restore the counter doesn't exist.

Even without DMC, counter should be stuck when PSR is active as no frames are generated by the pipe. I am using drm_vblank_restore_count() to take care of that.

Quoted 8 lines
Quoted 8 lines
drivers/gpu/drm/i915/i915_irq.c +0 -0
 	return 0;
 }
 
@@ -2968,6 +2971,9 @@ static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	bdw_enable_pipe_irq(dev_priv, pipe, GEN8_PIPE_VBLANK);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
Quoted 6 lines

drivers/gpu/drm/i915/i915_irq.c static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe) · line 2974

Quoted 1 line
drivers/gpu/drm/i915/i915_irq.c +1 -0
+	if (HAS_PSR(dev_priv))

HAS_PSR(dev_priv) && HAS_CSR(dev_priv)
maybe?
So it gets clear that it is not because PSR that we need to restore the counter, but also don't do it when not needed.

Same reason as above, let me test this again by disabling DMC.

Quoted 6 lines
Quoted 6 lines
drivers/gpu/drm/i915/i915_irq.c +2 -0
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
Quoted 7 lines
Quoted 3 lines
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

maintainer Reviewed-by
#14 lore raw

Re: [Intel-gfx] [PATCH 5/5] drm/i915: Estimate and update missed vblanks.

Quoted 38 lines · Dhinakaran
Quoted 37 lines · Rodrigo Vivi
Quoted 27 lines · Dhinakaran Pandiyan

The frame counter may have got reset between disabling and enabling vblank interrupts due to DMC putting the hardware to DC5/6 state if PSR was active. The frame counter also could have stalled if PSR is active in cases where there is no DMC. The frame counter resetting as a user visible impact of screen freezes. Use drm_vblank_restore() to compute missed vblanks in the duration for which vblank interrupts are disabled. There's no need particularly check if PSR was active in the interrupt disabled duration.

Enabling vblank interrupts wakes up the hardware from DC5/6 and prevents it from going back again as long as the there are pending interrupts. So, we don't have to explicity disallow DC5/6 after enabling vblank interrupts to keep the counter running.

Let's not apply this to CHV for now, as enabling interrupts does not prevent the hardware from activating PSR and thereby stalling the counter.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +0 -0
index 3517c6548e2c..db3466ec6faa 100644

drivers/gpu/drm/i915/i915_irq.c static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe) · line 2961

Quoted 6 lines
drivers/gpu/drm/i915/i915_irq.c +3 -0
@@ -2956,6 +2956,9 @@ static int ironlake_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	ilk_enable_display_irq(dev_priv, bit);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
+	if (HAS_PSR(dev_priv))
+		drm_vblank_restore(dev, pipe);
+

I don't believe we need this one here.

Quoted 6 lines
Quoted 2 lines

pre-gen9 platform has psr but not dmc, so the case where we need to restore the counter doesn't exist.

Even without DMC, counter should be stuck when PSR is active as no frames are generated by the pipe. I am using drm_vblank_restore_count() to take care of that.

Oh oh! Indeed. Now I remember you had told me this here. Can you please add a comment with this info somewhere so I don't ask the same question again ;)

anyways:

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
drivers/gpu/drm/i915/i915_irq.c +0 -0
 	return 0;
 }
 
@@ -2968,6 +2971,9 @@ static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe)
 	bdw_enable_pipe_irq(dev_priv, pipe, GEN8_PIPE_VBLANK);
 	spin_unlock_irqrestore(&dev_priv->irq_lock, irqflags);
 
Quoted 22 lines
Quoted 6 lines

drivers/gpu/drm/i915/i915_irq.c static int gen8_enable_vblank(struct drm_device *dev, unsigned int pipe) · line 2974

Quoted 1 line
drivers/gpu/drm/i915/i915_irq.c +1 -0
+	if (HAS_PSR(dev_priv))

HAS_PSR(dev_priv) && HAS_CSR(dev_priv)
maybe?
So it gets clear that it is not because PSR that we need to restore the counter, but also don't do it when not needed.

Same reason as above, let me test this again by disabling DMC.

Quoted 13 lines
Quoted 9 lines
drivers/gpu/drm/i915/i915_irq.c +2 -0
+		drm_vblank_restore(dev, pipe);
+
 	return 0;
 }
 
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#15 lore raw

✓ Fi.CI.BAT: success for series starting with [1/5] drm/vblank: Fix return type for drm_vblank_count()

In reply to Dhinakaran Pandiyan
== Series Details ==

== Series Details ==

Series: series starting with [1/5] drm/vblank: Fix return type for drm_vblank_count()
URL : https://patchwork.freedesktop.org/series/36435/
State : success

== Summary ==

Series 36435v1 series starting with [1/5] drm/vblank: Fix return type for drm_vblank_count()
https://patchwork.freedesktop.org/api/1.0/series/36435/revisions/1/mbox/

Test gem_exec_suspend:

        Subgroup basic-s3:
                pass       -> INCOMPLETE (fi-skl-6700k2) fdo#104108

fdo#104108 https://bugs.freedesktop.org/show_bug.cgi?id=104108

fi-bdw-5557u     total:288  pass:267  dwarn:0   dfail:0   fail:0   skip:21  time:424s
fi-bdw-gvtdvm    total:288  pass:264  dwarn:0   dfail:0   fail:0   skip:24  time:428s
fi-blb-e6850     total:288  pass:223  dwarn:1   dfail:0   fail:0   skip:64  time:370s
fi-bsw-n3050     total:288  pass:242  dwarn:0   dfail:0   fail:0   skip:46  time:487s
fi-bwr-2160      total:288  pass:183  dwarn:0   dfail:0   fail:0   skip:105 time:282s
fi-bxt-dsi       total:288  pass:258  dwarn:0   dfail:0   fail:0   skip:30  time:482s
fi-bxt-j4205     total:288  pass:259  dwarn:0   dfail:0   fail:0   skip:29  time:488s
fi-byt-j1900     total:288  pass:253  dwarn:0   dfail:0   fail:0   skip:35  time:467s
fi-byt-n2820     total:288  pass:249  dwarn:0   dfail:0   fail:0   skip:39  time:461s
fi-elk-e7500     total:224  pass:168  dwarn:10  dfail:0   fail:0   skip:45 
fi-gdg-551       total:288  pass:179  dwarn:0   dfail:0   fail:1   skip:108 time:273s
fi-glk-1         total:288  pass:260  dwarn:0   dfail:0   fail:0   skip:28  time:511s
fi-hsw-4770      total:288  pass:261  dwarn:0   dfail:0   fail:0   skip:27  time:392s
fi-hsw-4770r     total:288  pass:261  dwarn:0   dfail:0   fail:0   skip:27  time:403s
fi-ilk-650       total:288  pass:228  dwarn:0   dfail:0   fail:0   skip:60  time:411s
fi-ivb-3520m     total:288  pass:259  dwarn:0   dfail:0   fail:0   skip:29  time:461s
fi-ivb-3770      total:288  pass:255  dwarn:0   dfail:0   fail:0   skip:33  time:416s
fi-kbl-7500u     total:288  pass:263  dwarn:1   dfail:0   fail:0   skip:24  time:467s
fi-kbl-7560u     total:288  pass:269  dwarn:0   dfail:0   fail:0   skip:19  time:498s
fi-kbl-7567u     total:288  pass:268  dwarn:0   dfail:0   fail:0   skip:20  time:457s
fi-kbl-r         total:288  pass:260  dwarn:1   dfail:0   fail:0   skip:27  time:503s
fi-pnv-d510      total:288  pass:222  dwarn:1   dfail:0   fail:0   skip:65  time:578s
fi-skl-6260u     total:288  pass:268  dwarn:0   dfail:0   fail:0   skip:20  time:431s
fi-skl-6600u     total:288  pass:261  dwarn:0   dfail:0   fail:0   skip:27  time:509s
fi-skl-6700hq    total:288  pass:262  dwarn:0   dfail:0   fail:0   skip:26  time:529s
fi-skl-6700k2    total:108  pass:96   dwarn:0   dfail:0   fail:0   skip:11 
fi-skl-6770hq    total:288  pass:268  dwarn:0   dfail:0   fail:0   skip:20  time:489s
fi-skl-gvtdvm    total:288  pass:265  dwarn:0   dfail:0   fail:0   skip:23  time:431s
fi-snb-2520m     total:288  pass:248  dwarn:0   dfail:0   fail:0   skip:40  time:538s
fi-snb-2600      total:288  pass:248  dwarn:0   dfail:0   fail:0   skip:40  time:395s
Blacklisted hosts:
fi-cfl-s2        total:288  pass:262  dwarn:0   dfail:0   fail:0   skip:26  time:573s
fi-glk-dsi       total:288  pass:258  dwarn:0   dfail:0   fail:0   skip:30  time:472s

fcf7fdfd2db6cfe8e529c79c5822f555a8b38fd4 drm-tip: 2018y-01m-12d-17h-40m-49s UTC integration manifest ef67bfb35c68 drm/i915: Estimate and update missed vblanks.
b1edcc47812a drm/vblank: Restoring vblank counts after device PM events. 365390c83462 drm/vblank: Do not update vblank count if interrupts are already disabled. 18e5209cfde5 drm/vblank: Fix data type width for drm_crtc_arm_vblank_event() 60165b22a0be drm/vblank: Fix return type for drm_vblank_count()

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_7660/issues.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#16 lore raw

✗ Fi.CI.IGT: failure for series starting with [1/5] drm/vblank: Fix return type for drm_vblank_count()

In reply to Dhinakaran Pandiyan
== Series Details ==

== Series Details ==

Series: series starting with [1/5] drm/vblank: Fix return type for drm_vblank_count()
URL : https://patchwork.freedesktop.org/series/36435/
State : failure

== Summary ==

Test gem_caching:

        Subgroup writes:
                pass       -> INCOMPLETE (shard-snb)
Test kms_frontbuffer_tracking:
        Subgroup fbc-1p-pri-indfb-multidraw:
                pass       -> FAIL       (shard-snb) fdo#103167
        Subgroup fbc-1p-offscren-pri-shrfb-draw-blt:
                pass       -> FAIL       (shard-snb) fdo#101623 +1
Test kms_flip:
        Subgroup vblank-vs-suspend-interruptible:
                incomplete -> PASS       (shard-hsw) fdo#100368 +1
        Subgroup vblank-vs-suspend:
                pass       -> SKIP       (shard-snb) fdo#102365

fdo#103167
fdo#101623
fdo#100368
fdo#102365

shard-hsw        total:2713 pass:1537 dwarn:1   dfail:0   fail:10  skip:1165 time:9012s
shard-snb        total:2683 pass:1294 dwarn:1   dfail:0   fail:12  skip:1375 time:7610s
Blacklisted hosts:
shard-apl        total:2713 pass:1687 dwarn:1   dfail:0   fail:24  skip:1001 time:13691s
shard-kbl        total:2708 pass:1804 dwarn:5   dfail:0   fail:24  skip:874 time:10328s

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_7660/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#17 lore raw
In reply to Dhinakaran Pandiyan
Quoted 8 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Quoted 4 lines
Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>

Sounds like the 64bit widening wasn't all that well tested ... do we have an igt for this? Iirc the base igt was merged already. -Daniel

Quoted 20 lines
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +1 -1
index 32d9bcf5be7f..768a8e44d99b 100644
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 
Signature · 1 line
2.11.0
Signature · 7 lines
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#18 lore raw
Quoted 9 lines · Daniel Vetter
Quoted 8 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Quoted 7 lines
Quoted 4 lines
Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>

Sounds like the 64bit widening wasn't all that well tested ... do we have an igt for this? Iirc the base igt was merged already.

I don't see anything that would particularly trigger this condition i.e., vblank->count > 2^32 in the IGTs. We'll need to implement something to force set a very large vblank->count and then request a vblank sequence.

Quoted 14 lines

-Daniel

Quoted 12 lines
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +0 -0
index 32d9bcf5be7f..768a8e44d99b 100644
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
Quoted 9 lines
Quoted 8 lines
drivers/gpu/drm/drm_vblank.c +1 -1
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 
Signature · 1 line
2.11.0

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#19 lore raw
In reply to Dhinakaran Pandiyan

ping for review.

Let me know if there's anything that needs to be done, thanks!

Quoted 21 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +0 -0
index 32d9bcf5be7f..768a8e44d99b 100644

drivers/gpu/drm/drm_vblank.c static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe, · line 277

Quoted 8 lines
drivers/gpu/drm/drm_vblank.c +1 -1
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

maintainer
#20 lore raw
Quoted 1 line · Dhinakaran

ping for review.

sorry for not getting back sooner here.

But yey \o/
I finally have dmc and psr working well on my own laptop! so far so good! :)

Quoted 29 lines

Let me know if there's anything that needs to be done, thanks!

Quoted 24 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +0 -0
index 32d9bcf5be7f..768a8e44d99b 100644
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
Quoted 5 lines

drivers/gpu/drm/drm_vblank.c static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe, · line 277

Quoted 5 lines
drivers/gpu/drm/drm_vblank.c +1 -1
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

maintainer Reviewed-by on v1 1/5
#21 lore raw
In reply to Dhinakaran Pandiyan
Quoted 7 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Worth to mention and probably

Fixes 570e86963a51 ("drm: Widen vblank count to 64-bits [v3]")

btw, I spotted at least one more place even with the series applied. 32 current_vblank; at drm_mode_page_flip_ioctl...

so, probably worth to do a deeper check down to all paths...

anayway, for this patch:

Quoted 25 lines
Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +1 -1
index 32d9bcf5be7f..768a8e44d99b 100644
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#22 lore raw
Quoted 9 lines · Rodrigo Vivi
Quoted 7 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count. The effect of this is when drm_wait_vblank_ioctl() tries to widen the user space requested vblank sequence using this clipped 32-bit count(when the value is >= 2^32) as reference, the requested sequence remains a 32-bit value and gets queued like that. However, the code that checks if the requested sequence has passed compares this against the 64-bit vblank count.

Quoted 6 lines

Worth to mention and probably

Fixes 570e86963a51 ("drm: Widen vblank count to 64-bits [v3]")

btw, I spotted at least one more place even with the series applied. 32 current_vblank; at drm_mode_page_flip_ioctl...

There seem to be several such callers for drm_crtc_vblank_count(). I can fix it up as a follow-up series.

Quoted 27 lines

so, probably worth to do a deeper check down to all paths...

anayway, for this patch:

Quoted 22 lines
Cc Keith Packard <[email protected]>
Cc Michel Dänzer <[email protected]>
Cc Daniel Vetter <[email protected]>
---
 drivers/gpu/drm/drm_vblank.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/drm_vblank.c +1 -1
index 32d9bcf5be7f..768a8e44d99b 100644
@@ -271,7 +271,7 @@ static void drm_update_vblank_count(struct drm_device *dev, unsigned int pipe,
 	store_vblank(dev, pipe, diff, t_vblank, cur_vblank);
 }
 
-static u32 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
+static u64 drm_vblank_count(struct drm_device *dev, unsigned int pipe)
 {
 	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 
Quoted 7 lines
Quoted 3 lines
Signature · 1 line
2.11.0

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#23 lore raw
In reply to Dhinakaran Pandiyan
Quoted 2 lines · Dhinakaran Pandiyan

drm_vblank_count() has a u32 type returning what is a 64-bit vblank count.

It looks like a general review of the 64-bit widening patch is needed.

* drm_crtc_accurate_vblank_count has a 32-bit return, and uses a 32-bit temporary
* drm_wait_one_vblank uses a 32-bit temporary.

I looked at every 'u32' in drm_vblank.c; it would be good to have more eyes check this.

Thanks for finding the first one.

Signature · 1 line
-keith

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel