Hard crash on 6.x -- reproducible, multiple people affected

[Peter Thoenen]

--- Kris Kennaway <kris at obsecurity.org> wrote:
> On Sun, Apr 02, 2006 at 04:09:30PM -0400, Anish Mistry wrote:
> > > It's not clear what he means by "hard crash", but he also says
> > > power off", which is the part that is most confusing to me.
> > Hmmm...I missed that part.  Yeah, if it powers off then I'm not
> sure what you can do to debug it.

By hard crash I mean the box just instantly powers off.  No shutdown
syncs, no error, panic's, or dump generated, nothing logged in syslog,
and I have a serial cable hooked up to another another dumb terminal
hoping to maybe catch a panic or some debug message dumped out over
serial as it goes down but nothing there also.  It just stops. 

Also note that this didn't happen on 5.x.  And when not running those
particular ports the box stays up just fine for weeks on end.  Just
tried again with 6.1 PRERELEASE and same issue.

> About all I can think of is to try to reproduce the problem with
> enough logging enabled (e.g. running the binary under ktrace), with
> sync-mounted filesystem, and hope that the last few entries give the
> software developer enough of a clue what the process was doing at the
> time of shutdown (the operation itself probably will not be logged)
> that he can focus the investigation further.

Want to walk me through this and will do.  As I said, reproducible.

NOTE: I doubt it a linux syscall issue as i2p / freenet / and tor are
all build on FBSD, not linux binaries.  Freenet and I2P do use
linux-java (as the new diablo ones aren't avaiable for amd64) BUT tor
does not.  Its all C.  Also as noted, other people (using i386, not
just amd64) have emailed me (as the tor maintainer) with the exact same
problem on entirely different hardware (different nics, motherboards,
proc's etc etc).  Somebody suggested maybe a ACPI issue but I just
don't see how that could me it.