src: dri2: fix MSC handling

The MSC handling was taken from the xf86-video-intel driver, but it
seems to be lacking in that the wrap-handling is substandard; it can
result in DRI2 applications stalling for a very long time.

The problem is that xf86-video-intel is under the impression that the
vblank counter can wind backwards.  This is not true; the kernel only
ever increments the counter (see functions that call store_vblank().)
Hence, remove this logic, and always assume that a vblank sequence
value that is less than a previously seen one means that it has
wrapped.

Signed-off-by: Russell King <rmk@arm.linux.org.uk>

1 files changed, 13 insertions, 6 deletions

diff --git a/src/common_drm.c b/src/common_drm.c
index b5c7f0b..a3c017e 100644
--- a/src/common_drm.c
+++ b/src/common_drm.c
@@ -119,13 +119,20 @@ static uint64_t common_drm_frame_to_msc(xf86CrtcPtr crtc, uint32_t seq)
 {
 	struct common_crtc_info *drmc = common_crtc(crtc);
 
-	if (seq < drmc->last_seq) {
-		if ((int32_t)(drmc->last_seq - seq) > 0x40000000)
-			drmc->last_msc += 0x100000000ULL;
-		else
-			seq = drmc->last_seq;
-	}
+	/*
+	 * The sequence counter wrapped.  Unlike the misleading comments in
+	 * xf86-video-intel, the vblank counter is never wound backwards: it
+	 * always runs forwards.  However, since we don't monitor it with
+	 * any regularity, it can appear to go backwards if we wait (eg)
+	 * 0xc0000000 vblanks between calls.  Hence, whenever we see the
+	 * frame sequence less than the last sequence, assume that it has
+	 * wrapped.  (It may have wrapped more than once.)
+	 */
+	if (seq < drmc->last_seq)
+		drmc->last_msc += 0x100000000ULL;
+
 	drmc->last_seq = seq;
+
 	return drmc->last_msc + seq;
 }