Binary package hint: linux-image-2.6.20-13-generic
This may or may not be a duplicate of bug 26064, reported for Breezy (now closed).
We have four Core 2 duo machines with an Intel Q965 chipset that are all losing ticks at about three seconds per minute (three hours per day). This appears to be an upstream kernel bug with both 2.6.19 and 2.6.20, since this occurs for us with both Feisty and Fedora Core 6 (with either 2.6.19 or 2.6.20 kernel). This occurs both with and without the proprietary nvidia kernel module.
The full dmesg log is attached. The relevant lines seem to be:
[ 92.801333] Losing some ticks... checking if CPU frequency changed.
...
[ 2240.211414] warning: many lost ticks.
[ 2240.211416] Your time source seems to be instable or some driver is hogging i
nterupts
[ 2240.211425] rip _spin_unlock_irqrestore+0x8/0x10
As a stab in the dark, I've tried a few boot options, such as hpet=off, clocksource=acpi_pm, clock=jiffies, and acpi=off, but these seem to have no effect.
The problem seems to be with the upstream kernel sources, rather than the Ubuntu-specific patches, since it also occurs under Fedora Core 6 (with either 2.6.19 or 2.6.20). But the problem appears have been resolved in the mainline 2.6.21-rc5 kernel; the timer works normally on the same system with this kernel (self-compiled) but otherwise running Feisty.
I should add that this is a huge problem for these machines (HP Compaq dc7700p desktops). The clock drift is far too severe for ntpd to correct for it. The only workaround I know is to turn off the ntp daemon and run ntpd -q every couple of minutes.