When a frame is late we schedule its display right away with
g_timeout_add(0, ...). This can result in display_frame() being called
before g_timeout_add() returns. This would cause the timer_id being
reset before schedule_frame() had set it so that it would then never be
reset. So from that point schedule_frame() would always think a frame
was being displayed and thus would not schedule any more frames
resulting in a video freeze.
display_frame() now takes the queues mutex before resetting timer_id
eliminating the race.
Signed-off-by: Francois Gouget <fgouget@xxxxxxxxxxxxxxx>
---
src/channel-display-gst.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/src/channel-display-gst.c b/src/channel-display-gst.c
index 2b4ef7f9..8be95997 100644
--- a/src/channel-display-gst.c
+++ b/src/channel-display-gst.c
@@ -127,9 +127,8 @@ static gboolean display_frame(gpointer video_decoder)
GstBuffer *buffer;
GstMapInfo mapinfo;
- decoder->timer_id = 0;
-
g_mutex_lock(&decoder->queues_mutex);
+ decoder->timer_id = 0;
frame = g_queue_pop_head(decoder->display_queue);
g_mutex_unlock(&decoder->queues_mutex);
/* If the queue is empty we don't even need to reschedule */
--
2.11.0
_______________________________________________
Spice-devel mailing list
Spice-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/spice-devel