Skip to contentSkip to conversation

System clock frequency offset changes drastically across reboots

DISCUSSION John Sigler 1 participant 3 messages Jul 27, 2007 – Aug 3, 2007

lore
Share & tools
lore
#0 lore raw

Hello everyone,

I'm using 2.6.20.7-rt8 on a P3.

I've noticed that the frequency offset of my system clock (computed either by ntpd, or by hand) changes drastically across reboots.

By drastically, I mean +/- 60 ppm every time I reboot.

Apparently this is caused by some imprecision in the frequency estimation done in calculate_cpu_khz() (arch/i386/kernel/tsc.c)

My brutally crude work-around was to skip the computation altogether and hard-code the "correct" value for cpu_khz.

http://linux.kernel.free.fr/cpu_khz.adlink

Has this issue been investigated more thoroughly by other users?

Regards.

#1 lore raw

[ Expanding recipients list ]

Quoted 1 line · John Sigler

I'm using 2.6.20.7-rt8 on a P3.

But some ntpd users have reported the problem with mainline kernels and different versions.

Quoted 11 lines

I've noticed that the frequency offset of my system clock (computed either by ntpd, or by hand) changes drastically across reboots.

By drastically, I mean +/- 60 ppm every time I reboot.

Apparently this is caused by some imprecision in the frequency estimation done in calculate_cpu_khz() (arch/i386/kernel/tsc.c)

My brutally crude work-around was to skip the computation altogether and hard-code the "correct" value for cpu_khz.

http://linux.kernel.free.fr/cpu_khz.adlink

Has this issue been investigated more thoroughly by other users?

Regards.

#2 lore raw

[ Expanding recipients list to include NTP mailing list ]

Quoted 17 lines
Quoted 1 line · John Sigler

I'm using 2.6.20.7-rt8 on a P3.

(But some ntpd users have reported the problem with mainline kernels and different versions.)

Quoted 11 lines

I've noticed that the frequency offset of my system clock (computed either by ntpd, or by hand) changes drastically across reboots.

By drastically, I mean +/- 60 ppm every time I reboot.

Apparently this is caused by some imprecision in the frequency estimation done in calculate_cpu_khz() (arch/i386/kernel/tsc.c)

My brutally crude work-around was to skip the computation altogether and hard-code the "correct" value for cpu_khz.

http://linux.kernel.free.fr/cpu_khz.adlink

Has this issue been investigated more thoroughly by other users?

Is there, perhaps, more appropriate a forum to discuss such matters?

Regards.