Function bbr_calc_time

freebsd/netinet/tcp_stacks/bbr.c:1108–1131

* Considers the two time values now (cts) and earlier. * If cts is smaller than earlier, we could have * had a sequence wrap (our counter wraps every * 70 min or so) or it could be just clock skew * getting us two differnt time values. Clock skew * will show up within 10ms or so. So in such * a case (where cts is behind earlier time by * less than 10ms) we return 0. Otherwise we * return t

 * return the true difference between them.
 */
static inline uint32_t
bbr_calc_time(uint32_t cts, uint32_t earlier_time) {
	/*
	 * Given two timestamps, the current time stamp cts, and some other
	 * time-stamp taken in theory earlier return the difference. The
	 * trick is here sometimes locking will get the other timestamp
	 * after the cts. If this occurs we need to return 0.
	 */
	if (TSTMP_GEQ(cts, earlier_time))
		return (cts - earlier_time);
	/*
	 * cts is behind earlier_time if its less than 10ms consider it 0.
	 * If its more than 10ms difference then we had a time wrap. Else
	 * its just the normal locking foo. I wonder if we should not go to
	 * 64bit TS and get rid of this issue.
	 */
	if (TSTMP_GEQ((cts + 10000), earlier_time))
		return (0);
	/*
	 * Ok the time must have wrapped. So we need to answer a large
	 * amount of time, which the normal subtraction should do.
	 */
	return (cts - earlier_time);
}

static int
sysctl_bbr_clear_lost(SYSCTL_HANDLER_ARGS)