kexec+kdump + vmcoreinfo patch

[Neil Horman]

On Mon, Aug 27, 2007 at 10:09:52AM -0700, Randy Dunlap wrote:
> On Mon, 27 Aug 2007 12:55:38 -0400 Neil Horman wrote:
> 
> > On Sun, Aug 26, 2007 at 10:04:31AM -0700, Randy Dunlap wrote:
> > > On Sun, 26 Aug 2007 09:23:05 -0400 Neil Horman wrote:
> > > 
> > > > On Fri, Aug 24, 2007 at 05:01:39PM -0700, Randy Dunlap wrote:
> > > > > On Fri, 24 Aug 2007 14:41:47 -0700 Randy Dunlap wrote:
> > > > > 
> > > > > > Hi,
> > > > > > 
> > > > > > I'm trying to use the recent vmcoreinfo patch...
> > > > > 
> > > > > > c.  My 2.6.23-rc3 dump-capture kernel currently faults a few times,
> > > > > > then it panics.  I'll add a serial console and capture its output
> > > > > > to see what is going on.
> > > > > 
> > > > > I booted 3 times with no problems (without serial console on the
> > > > > dump-capture kernel), then I added a serial console to that kernel
> > > > > and got lots of problems.  They all seem to be OOM problems, then
> > > > > the kernel gives up and falls over.  (log attached)
> > > > > 
> > > > haven't looked at your log yet, but how much memory are you passing in your
> > > > crashkernel parameter in your first kernel boot?  Nominally 64MB is the minimum
> > > > recommended for most kernel configurations
> > > 
> > > Hi Neil,
> > > 
> > > crashkernel=64M at 16M
> > > 
> > Hmm, well that should be enough.  looking at your log, it appears as though you
> > have a 512k chunk allocatable, which should seem sufficient to go on a little
> > longer.  The fact that your not seems to indicate that you are allocating a
> > suspiciously large single chunk of ram.  I'd configure your initramfs to drop to
> > a shell prompt before it sets up lp0.  Then you can step through the actions of
> > the init script and monitor the contents of /proc/slabinfo to get an idea of
> > whats eating up all your lowmem prior to the oom kill.
> 
> I don't have an initramfs (nor an initrd).
> Is one required for using kexec + kdump?
> 
Are you passing a root device directly to the kernel for mounting then?  that
likely changes things.  As to weather or not you need an initramfs depends
entirely on your kernel configuration.  Strictly speaking, you don't need one,
since you can configure the kernel to find the root device on its own via the
command line, but I would recommend an initramfs, since you can customize it for
use in a particular dump environment, without needing to rely on things like
udevd to create device nodes for you.  You can find the latest kexec-tools srpm
on my home page:
http://people.redhat.com/nhorman
if you want to look at the mkdumprd utility to see how we assemble the
initramfs.

Regards
Neil
 
> Anyway, I'll try to do your suggestion somehow.  Thanks.
> 
> > > > > d.  Documentation/kdump/kdump.txt mentions an option to produce
> > > > > ELF32 headers instead of 64-bit headers:  --elf32-core-headers.
> > > > > Where is this option used?  I.e., what program recognizes it?
> > > > > kdump.txt isn't telling me this info and I can't read it between
> > > > > the lines.
> 
> 
> ---
> ~Randy
> *** Remember to use Documentation/SubmitChecklist when testing your code ***

-- 
/***************************************************
 *Neil Horman
 *Software Engineer
 *Red Hat, Inc.
 *nhorman at redhat.com
 *gpg keyid: 1024D / 0x92A74FA1
 *http://pgp.mit.edu
 ***************************************************/