Re: System goes into complete hang (Was: Is amd64 ready for the desktop?)

["Greg A. Woods; Planix, Inc." <woods%planix.ca@localhost>]

On 2-Apr-08, at 11:06 AM, D'Arcy J.M. Cain wrote:
> On Wed, 2 Apr 2008 14:59:57 +0000
> Herb Peyerl <hpeyerl%beer.org@localhost> wrote:
> > While I've totally not been paying attention to your problem or  
> > anything,
> > but if you think the interrupt system is turned off (as opposed to  
> > something
> > at a really high spl spinning), you might try twiddling something  
> > like DTR
> > on a serial port... Com ports are usually pretty high priority so  
> > you might
> > get lucky and end up with some traction on the problem...
>
> Isn't the keyboard pretty high too?  I can't even get the NumLock or
> CapsLock to light up.

The keyboard interrupt never seems high enough up.  :-)

(On BSD/OS I seem to remember being able to get Ctrl-Alt-DEL to work  
even when everything else seemed to be hung solid, but I can't  
remember if it would only work when caps-lock would work or not.)

Anyway I think the first thing you should do is get a serial console  
working on that system.  :-)

Then at least a BREAK condition should get you to DDB if there's any  
chance of it at all, i.e. if it's not a hardware problem or something  
where the hardware has been effectively disabled completely.  Then  
along with DEBUG and DIAGNOSTIC and LOCKDEBUG it might help someone  
find the problem.

I'm wondering if you're now maybe seeing the same hang I've been  
seeing on my Dell server though.  It hangs or crashes almost every  
night during the nightly cron jobs, though not 100% of the time, and  
even though it is otherwise completely idle at the time, yet it can  
run simultaneous build.sh runs (with -j6), big cvs-updates, and also  
big pkgsrc builds (e.g. mozilla, etc.) all at once and all without  
even the slightest hiccup.  I haven't had the courage to boot a - 
current kernel on my Dell though.  There were suggestions that it  
might be the SA stuff in netbsd-4 that was causing problems so I've  
been trying the wrstuden-fixsa branch of netbsd-4, but all that's done  
is change the symptoms of the hang a wee bit (now processes get stuck  
in vmmapva and wedge the whole system).

I guess I should try a -current kernel, but I really really really  
hate to do that on my main production file server and build host.  I  
really need netbsd-4 to be stable and reliable too, so randomly trying  
-current isn't necessarily much help in finding the problem with  
netbsd-4, at least not for me.

--
                                        Greg A. Woods; Planix, Inc.
                                        <woods%planix.ca@localhost>