[ntp:hackers] Bug in frequency calculations?

Danny Mayer mayer at ntp.isc.org
Thu Jan 31 13:47:50 UTC 2008


Hal Murray wrote:
> There is a bug in the TSC calibration routine in the Linux 2.6 kernel.
> 
> 
> Run
>   grep Det /var/log/messages | grep MHz
> You will probably find things like this
> Jan 25 10:10:00 shuksan kernel: Detected 2793.049 MHz processor.
> Jan 25 22:45:41 shuksan kernel: Detected 2793.112 MHz processor.
> Jan 25 22:47:44 shuksan kernel: Detected 2793.051 MHz processor.
> Jan 25 22:49:34 shuksan kernel: Detected 2793.050 MHz processor.
> Jan 25 22:51:07 shuksan kernel: Detected 2793.147 MHz processor.
> 
> That's basically jerking the drift file around.  It takes ntpd a
> while to recover, maybe a half day for big jumps.

That's probably also affecting Bill Unruh's experiments.

Danny


More information about the hackers mailing list