crashdumps not working

[Robert Watson]

On Wed, 8 Dec 2004, Michael Nottebrock wrote:

> On Wednesday, 8. December 2004 12:20, Robert Watson wrote:
> > On Tue, 7 Dec 2004, Michael Nottebrock wrote:
> > > I recently enabled SW_WATCHDOG in my kernel, but when watchdog triggers
> > > a panic, no crashdump is taken although dumps are enabled. What could be
> > > causing this?
> >
> > If you drop to the debugger by using the debug.kdb.enter sysctl, and do
> > "call doadump", followed by a reset, does a dump get generated
> > successfully?
> 
> I don't have kdb enabled in my kernel configuration at all... 

I'm guessing it might be useful at this point, if possible :-).  Do you
have a serial console on the box, or could you set one up?

> > I.e., are they completely broken on your system, or is this 
> > somehow a property of the particular hang you're seeing.
> 
> See my other mail, a different (non-watchdog) panic didn't trigger a dump 
> either. I even had the panic message in dmesg:
> 
> kernel trap 12 with interrupts disabled
> 
> Fatal trap 12: page fault while in kernel mode
> fault virtual address   = 0x14c
> fault code              = supervisor write, page not present
> instruction pointer     = 0x8:0xc0521397
> stack pointer           = 0x10:0xe9794b84
> frame pointer           = 0x10:0xe9794b90
> code segment            = base 0x0, limit 0xfffff, type 0x1b
>                         = DPL 0, pres 1, def32 1, gran 1
> processor eflags        = resume, IOPL = 0
> current process         = 1281 (beep-media-player)
> trap number             = 12
> panic: page fault

This is a NULL pointer dereference; you can use addr2line or gdb on your
kernel.debug to turn it into a line number even without a core.  That
might well be worth doing, as we might be able to debug that even without
getting dumping working on the box.

> Syncing disks, vnodes remaining...4 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
> 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
> 0 0 0 0 0 0 0 (I enabled kern.sync_on_panic only very recently for my
> experimenting with watchdog, dumping didn't work without it either). 

Syncing on panic is, in general, probably not going to make it work better
than not.  I guess there's no chance the box has an NMI button?

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Principal Research Scientist, McAfee Research