Skip to contentSkip to conversation

vmwgfx: use monotonic event timestamps

PATCH RESEND In review Arnd Bergmann 4 participants 6 messages Jan 16, 2018 – Jan 17, 2018

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

PATCH RESEND 1/1 vmwgfx: use monotonic event timestamps

DRM_VMW_EVENT_FENCE_SIGNALED (struct drm_vmw_event_fence) and DRM_EVENT_VBLANK (struct drm_event_vblank) pass timestamps in 32-bit seconds/microseconds format.

As of commit c61eef726a78 ("drm: add support for monotonic vblank timestamps"), other DRM drivers use monotonic times for drm_event_vblank, but vmwgfx still uses CLOCK_REALTIME for both events, which suffers from the y2038/y2106 overflow as well as time jumps.

For consistency, this changes vmwgfx to use ktime_get_ts64 as well, which solves those problems and avoids the deprecated do_gettimeofday() function.

This should be transparent to to user space, as long as it doesn't compare the time against the result of gettimeofday().

---
Originally sent on Nov 27. Sinclair Yeh said he'd pick it up for the next pull request, but it's not in linux-next yet.

Resending the unchanged patch, please pick it up when you have time, or feel free to ignore this email in case it's already in some tree that just isn't part of linux-next but will be sent during the next merge window.

---
 drivers/gpu/drm/vmwgfx/vmwgfx_fence.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +5 -4
index 6c5c75cf5e6c..9ed544f8958f 100644
@@ -901,11 +901,12 @@ static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action)
 	spin_lock_irq(&dev->event_lock);
 
 	if (likely(eaction->tv_sec != NULL)) {
-		struct timeval tv;
+		struct timespec64 ts;
 
-		do_gettimeofday(&tv);
-		*eaction->tv_sec = tv.tv_sec;
-		*eaction->tv_usec = tv.tv_usec;
+		ktime_get_ts64(&ts);
+		/* monotonic time, so no y2038 overflow */
+		*eaction->tv_sec = ts.tv_sec;
+		*eaction->tv_usec = ts.tv_nsec / NSEC_PER_USEC;
 	}
 
 	drm_send_event_locked(dev, eaction->event);
Signature · 6 lines
2.9.0

_______________________________________________
Y2038 mailing list
[email protected]
https://lists.linaro.org/mailman/listinfo/y2038
#1 lore raw

Hi, Arnd,

Sinclair's on paternal leave and I thought this patch was already in drm-next. My bad. Dave, is it too late to pull this in for the next merge window?

/Thomas

Quoted 41 lines · Arnd Bergmann

DRM_VMW_EVENT_FENCE_SIGNALED (struct drm_vmw_event_fence) and DRM_EVENT_VBLANK (struct drm_event_vblank) pass timestamps in 32-bit seconds/microseconds format.

As of commit c61eef726a78 ("drm: add support for monotonic vblank timestamps"), other DRM drivers use monotonic times for drm_event_vblank, but vmwgfx still uses CLOCK_REALTIME for both events, which suffers from the y2038/y2106 overflow as well as time jumps.

For consistency, this changes vmwgfx to use ktime_get_ts64 as well, which solves those problems and avoids the deprecated do_gettimeofday() function.

This should be transparent to to user space, as long as it doesn't compare the time against the result of gettimeofday().

---
Originally sent on Nov 27. Sinclair Yeh said he'd pick it up for the next pull request, but it's not in linux-next yet.

Resending the unchanged patch, please pick it up when you have time, or feel free to ignore this email in case it's already in some tree that just isn't part of linux-next but will be sent during the next merge window.

---
  drivers/gpu/drm/vmwgfx/vmwgfx_fence.c | 9 +++++----
  1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +1 -1
index 6c5c75cf5e6c..9ed544f8958f 100644
@@ -901,11 +901,12 @@ static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action)
  	spin_lock_irq(&dev->event_lock);
 
  	if (likely(eaction->tv_sec != NULL)) {
-		struct timeval tv;
+		struct timespec64 ts;
 

drivers/gpu/drm/vmwgfx/vmwgfx_fence.c static void vmw_event_fence_action_seq_passed(struct vmw_fence_action *action) · line 912

Quoted 10 lines
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +4 -3
-		do_gettimeofday(&tv);
-		*eaction->tv_sec = tv.tv_sec;
-		*eaction->tv_usec = tv.tv_usec;
+		ktime_get_ts64(&ts);
+		/* monotonic time, so no y2038 overflow */
+		*eaction->tv_sec = ts.tv_sec;
+		*eaction->tv_usec = ts.tv_nsec / NSEC_PER_USEC;
  	}
 
  	drm_send_event_locked(dev, eaction->event);

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

#2 lore raw
Quoted 50 lines · Thomas Hellstrom

Hi, Arnd,

Sinclair's on paternal leave and I thought this patch was already in drm-next. My bad. Dave, is it too late to pull this in for the next merge window?

/Thomas

Quoted 41 lines · Arnd Bergmann

DRM_VMW_EVENT_FENCE_SIGNALED (struct drm_vmw_event_fence) and DRM_EVENT_VBLANK (struct drm_event_vblank) pass timestamps in 32-bit seconds/microseconds format.

As of commit c61eef726a78 ("drm: add support for monotonic vblank timestamps"), other DRM drivers use monotonic times for drm_event_vblank, but vmwgfx still uses CLOCK_REALTIME for both events, which suffers from the y2038/y2106 overflow as well as time jumps.

For consistency, this changes vmwgfx to use ktime_get_ts64 as well, which solves those problems and avoids the deprecated do_gettimeofday() function.

This should be transparent to to user space, as long as it doesn't compare the time against the result of gettimeofday().

---
Originally sent on Nov 27. Sinclair Yeh said he'd pick it up for the next pull request, but it's not in linux-next yet.

Resending the unchanged patch, please pick it up when you have time, or feel free to ignore this email in case it's already in some tree that just isn't part of linux-next but will be sent during the next merge window.
---
  drivers/gpu/drm/vmwgfx/vmwgfx_fence.c | 9 +++++----

  1 file changed, 5 insertions(+), 4 deletions(-)
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +1 -1
index 6c5c75cf5e6c..9ed544f8958f 100644
@@ -901,11 +901,12 @@ static void 
 vmw_event_fence_action_seq_passed(struct vmw_fence_action *action)
       spin_lock_irq(&dev->event_lock);
         if (likely(eaction->tv_sec != NULL)) {
-        struct timeval tv;
+        struct timespec64 ts;
   -        do_gettimeofday(&tv);
Quoted 10 lines

drivers/gpu/drm/vmwgfx/vmwgfx_fence.c static void · line 911

Quoted 8 lines
drivers/gpu/drm/vmwgfx/vmwgfx_fence.c +4 -2
-        *eaction->tv_sec = tv.tv_sec;
-        *eaction->tv_usec = tv.tv_usec;
+        ktime_get_ts64(&ts);
+        /* monotonic time, so no y2038 overflow */
+        *eaction->tv_sec = ts.tv_sec;
+        *eaction->tv_usec = ts.tv_nsec / NSEC_PER_USEC;
       }
         drm_send_event_locked(dev, eaction->event);

Thomas, the same way my DRM patch has disappeared:
Date

	Tue, 19 Dec 2017 11:50:57 -0800

From Sinclair Yeh <>
Subject Re: [PATCH v.2] 4.15 vmgfx boot warning

This looks okay to me.

Quoted 25 lines · Woody Suwalski

The 4.15 drm_atomic_helper driver shows a warning during boot (both 32 and 64 bit x86)
It is caused by a mismatch between the result of vmw_enable_vblank() and what the drm_atomic_helper expects:
   /...
   ret = drm_crtc_vblank_get(crtc);
   WARN_ONCE(ret != -EINVAL, "driver forgot to call drm_crtc_vblank_off()\n");
   /...

Signed-off by: Woody Suwalski <[email protected]>

--- a/drivers/gpu/drm/drm_atomic_helper.c    2017-12-16 09:55:33.853374561 -0500
+++ b/drivers/gpu/drm/drm_atomic_helper.c    2017-12-16 10:55:56.089090752 -0500

(quoted diff) +1 -1
@@ -889,7 +889,7 @@ disable_outputs(struct drm_device *dev,
              continue;
 
          ret = drm_crtc_vblank_get(crtc);
-        WARN_ONCE(ret != -EINVAL, "driver forgot to call
 drm_crtc_vblank_off()\n");
+        WARN_ONCE((ret != -EINVAL && ret != -ENOSYS), "driver forgot to
 call drm_crtc_vblank_off()\n");
          if (ret == 0)
Quoted 3 lines

             drm_crtc_vblank_put(crtc);
     }

===============================================

The 4.15 drm_atomic_helper driver shows a warning during boot.
It is caused by a mismatch between the result of vmw_enable_vblank() and what the drm_atomic_helper expects:
  /...
   ret = drm_crtc_vblank_get(crtc);
   WARN_ONCE(ret != -EINVAL, "driver forgot to call drm_crtc_vblank_off()\n"); /...

Signed-off by: Woody Suwalski <[email protected]>

drivers/gpu/drm/drm_atomic_helper.c +1 -1
@@ -889,7 +889,7 @@ disable_outputs(struct drm_device *dev,
 			continue;
 
 		ret = drm_crtc_vblank_get(crtc);
-		WARN_ONCE(ret != -EINVAL, "driver forgot to call drm_crtc_vblank_off()\n");
+		WARN_ONCE((ret != -EINVAL && ret != -ENOSYS), "driver forgot to call drm_crtc_vblank_off()\n");
 		if (ret == 0)
 			drm_crtc_vblank_put(crtc);
 	}

_______________________________________________
Y2038 mailing list
[email protected]
https://lists.linaro.org/mailman/listinfo/y2038

#3 lore raw

drm: fix vmwgfx boot warning WAS Re: [PATCH] [RESEND] vmwgfx: use monotonic event timestamps

Hi, Woody,

Quoted 11 lines · Woody Suwalski

Thomas, the same way my DRM patch has disappeared:
Date

    Tue, 19 Dec 2017 11:50:57 -0800
From     Sinclair Yeh <>
Subject     Re: [PATCH v.2] 4.15 vmgfx boot warning

This looks okay to me.

Since this is a core DRM patch I think Sinclair was expecting it to be pulled in using the drm-misc tree. Did you get a comment from Daniel on this patch?

Thanks,
Thomas

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

#4 lore raw

Re: drm: fix vmwgfx boot warning WAS Re: [PATCH] [RESEND] vmwgfx: use monotonic event timestamps

Quoted 13 lines · Thomas Hellstrom

Hi, Woody,

Quoted 10 lines · Woody Suwalski

Thomas, the same way my DRM patch has disappeared:
Date

    Tue, 19 Dec 2017 11:50:57 -0800
From     Sinclair Yeh <>
Subject     Re: [PATCH v.2] 4.15 vmgfx boot warning

Quoted 7 lines
Quoted 1 line

This looks okay to me.

Since this is a core DRM patch I think Sinclair was expecting it to be pulled in using the drm-misc tree. Did you get a comment from Daniel on this patch?

Thanks,
Thomas

Actually, checkpatch.pl complains about -ENOSYS, so I'm sending a pull request for v1 of this patch instead, to make everybody happy. I've rephrased the commit log somewhat. Hope that's OK.

/Thomas

_______________________________________________
Y2038 mailing list
[email protected]
https://lists.linaro.org/mailman/listinfo/y2038

#5 lore raw
In reply to Thomas Hellstrom
Quoted 4 lines · Thomas Hellstrom

Hi, Arnd,

Sinclair's on paternal leave and I thought this patch was already in drm-next. My bad.

No worries.

Quoted 1 line

Dave, is it too late to pull this in for the next merge window?

The patch isn't urgent, it's fine to wait until after the merge window, I just want to avoid having to send it once more as I make my way through the remaining 'timeval' and 'timespec' users.

Arnd