Re: current kernels hang on i386 and amd64

[Kurt Schreiner <ks%ub.uni-mainz.de@localhost>]

On Thu, May 22, 2008 at 10:25:23PM +0200, Michael Lorenz wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Hello,
> 
> On May 22, 2008, at 15:52, Kurt Schreiner wrote:
> 
> > kernels compiled from -current source cvs updated May 22 20:18 MEST
> > hang under heavy load. The i386 machine is a P4 2.8GHz, HT enabled
> > w/ 2GB Ram and hangs after working some minutes on "build.sh -j8 ...
> > release". "Hang" means no answer to ping, no drop to ddb, no answer
> > to (PS/2-)keyboard only reset or power button helps.
> > The amd64 machine is a Q6600 w/ 8GB Ram and hangs after working
> > some minutes on "build.sh -j31 ... release". "Hang" is the same
> > as above. I've seen no problems w/ lighter loads (-j2 resp. -j12).
> > Last working kernels on both machines are from May 16...
> 
> Sounds like the same scheduler deadlock I'm seeing on macppc - please
> try a kernel with options LOCKDEBUG, if it panics with 'locking
> against myself' on a spinlock in sched_nextlwp() it's the same problem.
> There's a PR about it - kern/38707
Hm, a LOCKDEBUG kernel panics with (amd64 on Q6600):

Mutex error: lockdebug_wantlock 'locking against myself'
lock address : 0xffff800053f46d80 type: spin
[...]
panic: LOCKDEBUG

and thats it, no reaction anymore.

Looks very similar on i386, but drops to dbb. bt shows sched_nextlwp()
somewhere... Couldn't get a dump as "reboot" hangs in syncing disks...

Kurt