Comment 37 for bug 1403152

Dave,
Working on this now. I tried a bisect between v3.12..v3.13 where it seemed that 3.12 worked better, but it wasn't conclusive. I tried looking at dev_put dev_hold via systemtap, but that seems to modify the timing sufficiently. Looking at comments from Jordan Curzon to see if I can start instrumenting just those sections of the kernel to track down where we're failing to decrement the reference. I'm tracing the NETDEV_UNREGISTER notifier calls now to see if that leads to something conclusive.