[PATCH V2] drm/vkms: Avoid extra discount in the timestamp value

[Rodrigo Siqueira]

After the commit def35e7c5926 ("drm/vkms: Bugfix extra vblank frame")
some of the crc tests started to fail in the vkms with the following
error:

 [drm:drm_crtc_add_crc_entry [drm]] *ERROR* Overflow of CRC buffer,
    userspace reads too slow.
 [drm] failed to queue vkms_crc_work_handle
 ...

The aforementioned commit fixed the extra vblank added by
`drm_crtc_arm_vblank_event()` which is invoked inside
`vkms_crtc_atomic_flush()` if the vblank event count was zero, otherwise
`drm_crtc_send_vblank_event()` is invoked. The fix was implemented in
`vkms_get_vblank_timestamp()` by subtracting one period from the current
timestamp, as the code snippet below illustrates:

 if (!in_vblank_irq)
  *vblank_time -= output->period_ns;

The above fix works well when `drm_crtc_arm_vblank_event()` is invoked.
However, it does not properly work when `drm_crtc_send_vblank_event()`
executes since it subtracts the correct timestamp, which it shouldn't.
In this case, the `drm_crtc_accurate_vblank_count()` function will
returns the wrong frame number, which generates the aforementioned
error. Such decrease in `get_vblank_timestamp()` produce a negative
number in the following calculation within `drm_update_vblank_count()`:

 u64 diff_ns = ktime_to_ns(ktime_sub(t_vblank, vblank->time));

After this operation, the DIV_ROUND_CLOSEST_ULL macro is invoked using
diff_ns with a negative number, which generates an undefined result;
therefore, the returned frame is a huge and incorrect number. Finally,
the code below is part of the `vkms_crc_work_handle()`, note that the
while loop depends on the returned value from
`drm_crtc_accurate_vblank_count()` which may cause the loop to take a
long time to finish in case of huge value.

 frame_end = drm_crtc_accurate_vblank_count(crtc);
 while (frame_start <= frame_end)
   drm_crtc_add_crc_entry(crtc, true, frame_start++, &crc32);

This commit fixes this issue by checking if the vblank timestamp
corresponding to the current software vblank counter is equal to the
current vblank; if they are equal, it means that
`drm_crtc_send_vblank_event()` was invoked and vkms does not need to
discount the extra vblank, otherwise, `drm_crtc_arm_vblank_event()` was
executed and vkms have to discount the extra vblank. This fix made the
CRC tests work again whereas keep all tests from kms_flip working as
well.

V2: Update commit message

Signed-off-by: Rodrigo Siqueira <rodrigosiqueiramelo at gmail.com>
Signed-off-by: Shayenne Moura <shayenneluzmoura at gmail.com>
---
 drivers/gpu/drm/vkms/vkms_crtc.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/gpu/drm/vkms/vkms_crtc.c b/drivers/gpu/drm/vkms/vkms_crtc.c
index 7508815fac11..3ce60e66673e 100644
--- a/drivers/gpu/drm/vkms/vkms_crtc.c
+++ b/drivers/gpu/drm/vkms/vkms_crtc.c
@@ -74,9 +74,13 @@ bool vkms_get_vblank_timestamp(struct drm_device *dev, unsigned int pipe,
 {
 	struct vkms_device *vkmsdev = drm_device_to_vkms_device(dev);
 	struct vkms_output *output = &vkmsdev->output;
+	struct drm_vblank_crtc *vblank = &dev->vblank[pipe];
 
 	*vblank_time = output->vblank_hrtimer.node.expires;
 
+	if (*vblank_time == vblank->time)
+		return true;
+
 	if (!in_vblank_irq)
 		*vblank_time -= output->period_ns;
 
-- 
2.21.0

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20190604/fb4cb817/attachment-0001.sig>