Fr�d�ric Grosshans wrote: >Le vendredi 15 avril 2005 � 20:16 +0800, William Kenworthy a �crit : > > >>also worth trying: stop ntp or whatever time sync program you are using, >>then delete /etc/adjtime and /etc/ntp.drift (or whatever the drift is) >>Then run ntp, leaving it to run for a few hours so it will write correct >>drift and adjustment values. >> >> > >ntp doesn't work anyway :-( (I'm behind a firewall) > >And playing with /etc/adjtime didn't help when I tried it: it changes >the hardware clock, which is correct, and not the system clock, which >drifts. > > >
Assuming I understand the probelm correctly, that is that the bios clock keeps correct time, but the linux system time drifts, you could try using the clock= kernel parameter, which specifies what timesource the kernel (and thus the rest of the system) uses for tracking time. I've had to do this when running linux inside of VMware on a laptop, because of cpu frequency scaling. But then the affect wasn't 5min/day, it was more like 2sec/sec. Your choices are "clock=[pit,tsc,cyclone,pmtmr]", although I have no idea what "cyclone" is. pit is the default, but you might try pmtmr. You could also resort to a cron job that does a "hwclock --hctosys" every hour or so, although as the hwclock man page says, you might want to do this with adjtimex instead. -Richard -- [email protected] mailing list
