[amd64] Instability worse than ever

[Brian Fundakowski Feldman]

On Mon, Jul 26, 2004 at 06:58:23AM -0500, Conrad J. Sabatier wrote:
> 
> On 25-Jul-2004 Brian Fundakowski Feldman wrote:
> > On Sat, Jul 24, 2004 at 08:33:57PM -0500, Conrad J. Sabatier wrote:
> >> 
> >> On 25-Jul-2004 Scott Long wrote:
> >> > Conrad J. Sabatier wrote:
> >> > 
> >> >> For the last couple of days, my amd64 box has taken to
> >> >> spontaneously rebooting.  No panic, no coredump, nothing in
> >> >> logs, just reboots.
> >> >> 
> >> >> These always seem to occur just as I'm clicking on something
> >> >> under GNOME.
> >> >> 
> >> >> Very mysterious!
> 
> Happened again this morning.  Waking the machine up from DPMS sleep
> mode, it went straight into a reboot.  :-(
> 
> >> > Do you have INVARIANTS turned on?
> >> 
> >> No, I really should do that.  Don't know why I hadn't already.  Duh.
> >> :-)
> >> 
> >> > If you do, would you be able to hook up a serial console and see
> >> > what the kernel is saying?
> >> 
> >> Been meaning to ask about that.  Would a null modem connection from
> >> my other machine be useable as the serial console?  I have all the
> >> kernel config options and the sio flags set to enable the serial
> >> console, but wasn't sure if that would work or not.  If so, I could
> >> try that.
> > 
> > Absolutely; I just set this up a few days ago, so I know this works
> > with -CURRENT.
> > First, set the sio.0.flags hint in /boot/device.hints to
> > "0x90" so it is a vailable for use as a serial console.  Then, set up
> > the kernel with DDB and KDB and set /etc/sysctl.conf to turn the
> > debug.debugger_on_panic switch to 1.  You should probably be running
> > with at least WITNESS and INVARIANTS, but not necessarily DIAGNOSTIC.
> 
> OK, I've built a new kernel (after cvsup) with INVARIANTS and WITNESS
> enabled, and double-checked my serial debugging setup, however, it's
> just occurred to me that this probably isn't going to help very much in
> this particular situation, as I already had debug.debugger_on_panic
> enabled, and the system still just went directly to a reboot, bypassing
> the debugger altogether.
> 
> Still, perhaps WITNESS may at least provide a clue.  Let's hope so.  :-)

Were you running X?  It is recognized that the console is "unavailable"
when running X, and if no consoles are available, debug.debugger_on_panic
has no effect.

-- 
Brian Fundakowski Feldman                           \'[ FreeBSD ]''''''''''\
  <> green at FreeBSD.org                               \  The Power to Serve! \
 Opinions expressed are my own.                       \,,,,,,,,,,,,,,,,,,,,,,\