Recent update to RELENG_6 creates lots of calcru warnings

[James Long]

On Sun, Sep 02, 2007 at 01:18:05PM +0200, Kris Kennaway wrote:
> James Long wrote:
>> I'm running FreeBSD RELENG_6 on a Compaq DL380 G2 dual PIII-1.4GHz with an 
>> SMP kernel.  dmesg below.
>> Prior to about August 12, the system had been quite stable.
>> Since that time, I am finding lots of calcru warnings, which often
>> cause system services to fail when they see unorthodox time shifts.
>> Section 5.19 of the handbook appears not to be current any longer,
>> suggesting setting kern.timecounter.method=1.  
>> kern.timecounter.hardware=i8254 also does not improve things.
>> Other suggestions?
>> Thanks,
>> Jim
>> Sep  1 18:00:30 ns sudo:    james : TTY=ttyp0 ; 
>> PWD=/usr/local/www/sites/com.umpquanet/support ; USER=root ; 
>> COMMAND=/sbin/sysctl -w kern.timecounter.hardware=i8254
>> Sep  1 18:02:44 ns kernel: calcru: runtime went backwards from 1343309 
>> usec to 1178478 usec for pid 3663 (hpasmd)
>> Sep  1 18:02:44 ns kernel: calcru: negative runtime of -1274607 usec for 
>> pid 3660 (hpasmd)

... big snip ...

> Does reverting to the older kernel work?  If so, can you isolate which 
> commit caused the problem by doing a binary search of date ranges?  This 
> should be easy to do since the rate of changes to RELENG_6 is low.
> 
> Kris

Please pardon the rookie question, but the machine is not blazingly 
fast to build (circa 90 minute world, 20 min. kernel).  For the 
purposes of the binary search, may I build just the kernel, and would 
you suggest leaving the up-to-date world and running against a 
slightly older kernel as I test, or should I roll the world back to 
early August, and test against a slightly new kernel -- or will it 
matter?