Dumping kernel log (dmesg) and backtraces after a panic

Dumping kernel log (dmesg) and backtraces after a panic

[Chris Johns]

We're using Red Hat EL3 Linux (2.4.21 base kernel plus 300 or so Red 
Hat and/or community patches) and I'm dearly missing KDB already, since 
we previously used 2.4.21 from kernel.org and applied the appropriate 
KDB patch(es). Now with EL3, I'm not even sure what the right patch for 
KDB would be.

The problem is how to debug a hang or panic without KDB. Specifically, 
I'd like to dump out real backtraces of all (or selected) processes 
instead of the pseudo-backtraces that the panic or Alt-Sysrq-t 
provides, and I'd like to dump out the kernel log buffer (dmesg) after 
a hang or panic.

When I say "pseudo-backtraces", it seems that the oops/sysrq processing 
picks everything that looks like a text address from the stack of each 
thread (or the thread that caused the panic) and formats it, rather 
than walking the stack back correctly like KDB's 'bt' command does. And 
I don't know of any way of getting the 'dmesg' output after a 
hang/panic other than by using KDB.

To put it simply, is there either an alternative to KDB that works with 
RH EL3 and provides what I need (bt and dmesg, or just dmesg), or is 
there a version of KDB that would work with EL3 already?

Thanks,

Chris Johns

Re: Dumping kernel log (dmesg) and backtraces after a panic

[Willy Tarreau]

On Fri, Aug 20, 2004 at 09:21:05PM -0500, Chris Johns wrote:
 
> To put it simply, is there either an alternative to KDB that works with 
> RH EL3 and provides what I need (bt and dmesg, or just dmesg), or is 
> there a version of KDB that would work with EL3 already?

I don't know if there's a specific KDB version, but you might give a try
to kmsgdump, it's not so intrusive and might apply on RHEL kernel. It will
allow you to dump the dmesg log buffer onto a floppy disk or a parallel
printer upon panic or on demand.

Cheers,
Willy

Re: Dumping kernel log (dmesg) and backtraces after a panic

[Arjan van de Ven]

[-- Attachment #1: Type: text/plain, Size: 1188 bytes --]

On Sat, 2004-08-21 at 04:21, Chris Johns wrote:
> We're using Red Hat EL3 Linux (2.4.21 base kernel plus 300 or so Red 
> Hat and/or community patches) and I'm dearly missing KDB already, since 
> we previously used 2.4.21 from kernel.org and applied the appropriate 
> KDB patch(es). Now with EL3, I'm not even sure what the right patch for 
> KDB would be.
> 
> The problem is how to debug a hang or panic without KDB. Specifically, 
> I'd like to dump out real backtraces of all (or selected) processes 
> instead of the pseudo-backtraces that the panic or Alt-Sysrq-t 
> provides, and I'd like to dump out the kernel log buffer (dmesg) after 
> a hang or panic.
> 
> When I say "pseudo-backtraces", it seems that the oops/sysrq processing 
> picks everything that looks like a text address from the stack of each 
> thread (or the thread that caused the panic) and formats it, rather 
> than walking the stack back correctly like KDB's 'bt' command does. And 
> I don't know of any way of getting the 'dmesg' output after a 
> hang/panic other than by using KDB.


why not use netdump and then analyze the dump on another machine with
"crash" (a gdb variant) ?

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]