Time going backwards 2005-12-04 - By Ed Wilts
Back
On Sun, Dec 04, 2005 at 11:29:03AM +0100, coyote@(protected) wrote:
> On Sat, 3 Dec 2005, Ed Wilts wrote:
>
> >Thanks, Arjan, I'll give that a try. What should the vendor be using
> >for time keeping if not the tsc counter? I'm not a developer, but I'd
> >like to tell him what he should have done instead.
> >
> >What are the drawbacks to adding notsc? I assume there would be some or
> >it would be the default, wouldn't it?
>
> Just a shot in the dark but since these machines are of course multi-cpu
> it comes to mind.. Are the cpu stepping/revision identical (or not) on the
> systems that exhibit these problems? I don't think this can be checked by
> using /proc/cpuinfo and you might have to look it up on the entire cpu
> module or on the chip itself..
Actually, /proc/cpuinfo does report it. Sure enough, they're different
on the 2 systems that exhibit the problem. Both of these systems were
originally ordered as dual-cpu systems so the vendor needs to be
slapped...
When I was doing the research yesterday on notsc, I ran across an old
SMP Howto from Alan Cox. One of the topics:
Can I run mixed speeds of Intel processors ?
According to Intel probably not. According to real world experience if
you mix two similar Celeron/PII/PIII processors with the same FSB but
different multipliers Linux will work. You must specify the "notsc" boot
option for this configuration
So, it apears that you've hit the nail on the head as to the cause and
Arjan gave the software fix. The hardware fix, of course, is to swap
the 2nd CPU on each of these 2 buggered systems.
Thanks!
.../Ed
--
Ed Wilts, RHCE
Mounds View, MN, USA
mailto:ewilts@(protected)
Member #1, Red Hat Community Ambassador Program
--
Taroon-list mailing list
Taroon-list@(protected)
https://www.redhat.com/mailman/listinfo/taroon-list
|
|
