Re: Problems with Compaq DL380G2 running Redhat 7.2

[Jeremy Katz]

On Thu, 2002-05-30 at 14:01, Thierry Laronde wrote:
> On Thu, May 30, 2002 at 05:21:43PM +0800, Yeo Eng Hee wrote:
> > I am encountering a problem with the bootloader.
> > 
> > Everything worked fine until I had to compile and install my own kernel.
> > (I needed to run Linux 2.4.22 kernel to access the hot-plug tape that
> > comes with the DL380G2).
> > 
> > I tried configuring grub using new-kernel-pkg and grubby, but when I
> > reboot, I got a kernel panic with a message saying the my root fs is not
> > mounted, "cciss/c0d0p1" not valid and to set root= properly (I am
> > writing this from memory).
> 
> I think some people from RedHat are listening to the list. Because since
> `new-kernel-pkg' and `grubby' are neither part of the GRUB vanilla or
> standard on other distributions, it will be difficult to debug without
> knowing exactly what these utils do...

For reference --  
  new-kernel-pkg is just a script which gets called on the installation
of our kernel packages to create an initrd if needed, update modules.dep
files, and update the boot loader configuration.  grubby is the program
which actually handled the updating of the boot loader config, basically
by cloning the previous config and changing the file names.  For an idea
of what they did here, more information on exactly what arguments were
passed to each would be needed.
 
> For the second time problems, it seems to me that the informations in
> stage1 have been spoiled by the failed `grub-install'. If, here, only
> the vanilla `grub-install' has spoiled its data (writing something
> without being sure to he able to run), there is a bug. But I'm a bit
> suspicious about this one too. Is the `grub-install' on RedHat modified
> in order to query from the kernel the root partition (invoquing rdev for
> example)?

Nope, the changes in grub-install are all pretty much cosmetic...  we
have Erich's patch to add a sync or three and then move the copying into
a function instead of just being directly there so that there can be a
command line flag just to copy the stage2 bits over from their initial
location.

Cheers,

Jeremy