http://www.holm.cc/stress/log/cons141.html

[Peter Holm]

On Thu, Jul 14, 2005 at 11:10:42PM +0400, Gleb Smirnoff wrote:
> On Thu, Jul 14, 2005 at 08:39:32PM +0200, Peter Holm wrote:
> P> > On Wed, Jul 13, 2005 at 10:51:55AM +0200, Peter Holm wrote:
> P> > P> > how often the panic in subject fires? I guess it is related to the problem
> P> > P> > discussed in this thread [1], and may be even the patch I've posted there
> P> > P> > will help.
> P> > P> >
> P> > P> 
> P> > P> The panic is almost instantanious if I run the stress test while doing
> P> > P> an "arp -d xxx".
> P> > 
> P> > The attached WIP patch helps me. The [1] script can't drop my box to panic
> P> > anymore, when running in 15 instances on 4-x CPU box.
> P> 
> P> Yes, works for me! I've been testing for three hours without any
> P> problems and then ran into another unrelated livelock.
> 
> Were you running it with WITNESS or without? How many CPUs does your box have?
> I've got a hard panic without access to ddb when running without WITNESS.
> I think WITNESS slowness can hide some races. I'll continue to work on this
> tomorrow.
> 

Single CPU + WITNESS:
$ diff GENERIC PHO
68a69
> options         BREAK_TO_DEBUGGER
295a297,303
> 
> options         DEBUG_LOCKS
> options         DEBUG_VFS_LOCKS
> options         KTR
> options         KTR_ENTRIES=4096
> options         KTR_MASK=(KTR_BUF|KTR_VFS)
> options         KTR_COMPILE=(KTR_BUF|KTR_VFS)

- Peter

> -- 
> Totus tuus, Glebius.
> GLEBIUS-RIPN GLEB-RIPE