calcru: negative time ... followed by freeze

[Robert Watson]

FYI, this is a Xeon box with two physical processors, each with two
logical processors, and the problem could well be SMP-related.


Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Senior Research Scientist, McAfee Research

On Sun, 20 Jun 2004, Robert Watson wrote:

> 
> I've now seen the following scenario happen several times in the last few
> days while doing testing and benchmarking: I run a multi-threaded test,
> such as super-smack, that causes a moderately high system load.  I then
> hit Ctrl-T or run top, or some other activity that forces calcru() to
> execute.  I will not infrequently get an extremely hard hang -- can't get
> to DDB using serial break, etc.  I don't remember it happening when using
> non-threaded apps, so I'm wondering if there's a poor interaction with
> KSE/scheduler/who knows what.
> 
>  7:55PM  up 6 mins, 2 users, load averages: 1.37, 0.91, 0.43
> USER             TTY      FROM              LOGIN@  IDLE WHAT
> root             d0       -                 7:55PM     - w
> rwatson          p0       cboss             7:50PM     2 super-smack
> select-key
> hippy# top
> calcru: negative time of 1834075 usec for pid 654 (super-smack)
> ca~~
> 
> In this case, I ran super-smack with the following parameters:
> 
> hippy:/usr/tmp/super-smack> super-smack select-key.smack 15 1000
> 
> This generates 15 workers, which should cause mysql to spawn off threads
> as well.  I'm running with stock libpthread on this system (slightly old)
> but an up-to-date kernel from CVS, GENERIC.
> 
> Has anyone else seen this?
> 
> Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
> robert at fledge.watson.org      Senior Research Scientist, McAfee Research
> 
>