Timekeeping hosed by factor 3, high lapic[01] interrupt rates
John Baldwin
jhb at FreeBSD.org
Wed Jun 29 18:30:46 GMT 2005
On Wednesday 29 June 2005 01:42 pm, Jens Schweikhardt wrote:
> John et al,
>
> # Can you get the same ntpdate output for the kernel without the patch?
>
> Sure; boot with kern.smp.disabled=1.
>
> Script started on Wed Jun 29 19:33:39 2005
> # /etc/rc.d/ntpd stop
> Stopping ntpd.
> # ntpdate 192.53.103.103; sleep 10; ntpdate 192.53.103.103
> 29 Jun 19:34:10 ntpdate[602]: adjust time server 192.53.103.103 offset
> -0.393914 sec 29 Jun 19:34:21 ntpdate[604]: adjust time server
> 192.53.103.103 offset -0.391510 sec # sysctl
> kern.timecounter.hardware=i8254
> kern.timecounter.hardware: TSC -> i8254
> # ntpdate 192.53.103.103; sleep 10; ntpdate 192.53.103.103
> 29 Jun 19:35:09 ntpdate[616]: step time server 192.53.103.103 offset
> 7.003573 sec 29 Jun 19:36:02 ntpdate[618]: step time server 192.53.103.103
> offset 42.615163 sec # sysctl kern.timecounter.hardware=TSC
> kern.timecounter.hardware: i8254 -> TSC
> # ntpdate 192.53.103.103; sleep 10; ntpdate 192.53.103.103
> 29 Jun 19:36:21 ntpdate[620]: step time server 192.53.103.103 offset
> 9.605683 sec 29 Jun 19:36:31 ntpdate[622]: adjust time server
> 192.53.103.103 offset 0.002316 sec Script done on Wed Jun 29 19:36:53 2005
>
> Regards,
>
> Jens
Well, that's certainly weird. I got those types of numbers (~40 secs) when I
was using the broken patch. Without any patches at all, my adjustments were
-100 seconds, not +40. There probably should be at least some difference
between unpatched and patched. For the sake of my sanity, can you please
build a GENERIC kernel from scratch without any patches and capture the above
output and then apply http://www.FreeBSD.org/~jhb/patches/i8254_lapic.patch
and rebuild and capture the same output again?
--
John Baldwin <jhb at FreeBSD.org> <>< http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve" = http://www.FreeBSD.org
More information about the freebsd-current
mailing list