Subject: Re: Autosave of panic backtraces (was Re: Getting into the debugger)
To: Chris G. Demetriou <email@example.com>
From: Brian C. Grayson <firstname.lastname@example.org>
Date: 01/17/1999 02:20:36
On Sat, Jan 16, 1999 at 11:15:48PM -0800, Chris G. Demetriou wrote:
> if you print it out, it'll be in the message buffer. if you print it
> out before you do the crashdump (i.e. from the same place the panic
> message is printed, panic()), then it'll end up in the crashdump
> (along with the rest of the message buffer).
> then you just use dmesg w/ -M and -N on the crashdump and kernel image
> to get out the text.
Maybe it's just me/pilot-error, but crashdumps just haven't
gotten along with me:
1. savecore sometimes says "/dev/wd0e busy" even when dumpdev
is wd0b and /var/crash is wd0f.
2. savecore sometimes prints some message about a completely
bogus device number and gives up.
3. I thought at least once savecore didn't notice a crashdump,
but I could be misremembering.
4. When I invoke "gdb netbsd.gdb" and use "target kcore
netbsd.5.core", I couldn't get a backtrace (because it
thought the IP was 0???). I tried using
"proc curproc", but that failed trying to access 0x0
(invalid PDE). Is there some crucial step I'm missing? This
happened on more than one crashdump, and I still have some
of them lying around in /var/crash, so hints could help
me send better info to Bill S.
Hence my desire for backtraces saved to disk. :) My machine
is from -current of August vintage, roughly, if it matters.
I guess if crashdumps work fine for others, then there's no
reason for what I was proposing!