8.0RC2 amd64 - kernel panic running make buildworld
Alexandre "Sunny" Kovalenko
gaijin.k at gmail.com
Tue Nov 10 14:44:19 UTC 2009
On Tue, 2009-11-10 at 12:41 +0300, S.N.Grigoriev wrote:
> 07.11.09, 13:32, "Alexandre \"Sunny\" Kovalenko" <gaijin.k at gmail.com>
> > Sergey, I think it would be best if you follow
> > http://www.freebsd.org/doc/en_US.ISO8859-1/books/developers-handbook/kerneldebug.html#KERNELDEBUG-OBTAIN
> > and the do an ultimate test:
> > * quiesce your system
> > * switch to the console
> > * sync (few times, if you are really old school ;)
> > * sysctl debug.kdb.panic=1 (this would *panic* the system and, given
> > everything is set-up properly, produce the crash dump)
> > if you do not have debug.kdb.panic sysctl, please, add option KDB to
> > your kernel configuration.
> > If you get crash dump from the kernel-induced panic and your system
> > keeps rebooting without a trace, I would suspect some hardware testing
> > might be in order.
> I followed your tips. The kernel configuration now contains options DDB
> and KDB. The sysctl variable 'debug.debugger_on_panic' is set to '1'.
> After the 'sysctl debug.kdb.panic=1' command the debugger prompt
> To have a crash dump I should type 'panic' on the debugger prompt.
> If I type 'reboot' instead, there are no crash dumps. Is that behaviour
This is correct -- if you use reboot from the debugger prompt, you will
not get crash dump. I don't know whether this is "right" behavior or
not, but I have definitely seen this before.
> Another question: must all panics go to the bebugger prompt?
No, set debug.debugger_on_panic to 0 and system will reboot on panic
producing crash dump at startup, provided you have crash dump device
configured properly. If I understood you correctly, after you typed
'panic' in the debugger prompt, your system restarted and, eventually,
usable dump found its way to /var/crash. Is this correct? If yes, you
can change your system to skip debugger on panic.
> I still have neither crash dumps nor debugger prompt during
> world/kernel compilations. Just reboots.
I am not qualified to tell you that there is absolutely no possibility
that FreeBSD could reboot without leaving any trace on the perfectly
functioning hardware, but I would think exploring other options might
not be out of order now.
Unscientifically, I have observed that 8.0RC2 runs hotter on long builds
than 7-STABLE did before. I have lowered _PSV, because handrest gets
uncomfortably warm and I seem to see it being hit (CPU is being
throttled) more often than I remember from the days of 7-STABLE. Again,
this is *completely* unscientific. It is also possible that I have dust
accumulated in the airduct and has nothing to do with the differences in
the OS level.
If you still dual boot on that machine and have some device to check
power draw during buildworld for both 7.2 and 8.0 it might be a
worthwhile exercise. ANother thing would be to monitor whatever thermal
sensor you have available to you, I think k8temp(8) provides necessary
services for AMD64.
Connecting either serial or firewire console to the machine and
capturing anything sent to the console right before the reboot might be
yet another thing to try.
Alexandre Kovalenko (Олександр Коваленко)
More information about the freebsd-current