> on a dual AMD Athlon 1800 running a multiprocessor kernel, the output
> from time(1) frequently has garbage in the user and/or system time
> fields.  Here are two specimens:
> 
>     9051.58 real      4255.13 user -140461532.55 sys
> 
>     3970.16 real -1306333211.13 user -1446799052.29 sys

These are 'usually' caused by microtime going backwards.
The MP kernel (and current 1.6) use a completely different microtime
implementation to the systems Chuck and I did a lot of investigation
on, I think MP systems use a different timer interrupt as well.....

Anything that sets the clock (eg NTP) is very likely to cause
microtime to decrease, unfortunately the scheduler doesn't really
want that!

There are also a few other lurking problems that cause the output from
time to be incorrect.  I think I've already raised PRs for them.

	David

-- 
David Laight: david@l8s.co.uk