Confusion with drift at the rail

NTP classic does this too. The solution is to delete the drift file and restart ntpd.

ntpd should handle this case. I'm not sure how it gets started. Once there, it keeps stepping the time until a human fixes things.

One guess would be an overflow in some calculation.

Here are a few samples of things to look for in the log file:

02-21T03:27:21 ntpd[27658]: 0.0.0.0 c012 02 freq_set kernel -490.866 PPM
02-21T03:53:05 ntpd[765]: 0.0.0.0 c01c 0c clock_step +0.448230 s

Both ntpd and the kernel expect the magnitude of the drift to be less than 500 PPM.