[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: kernel upgrade/ kickstart question



On Thu, 13 Feb 2003, William F. Acker WB2FLW +1-303-777-8123 wrote:

> On Thu, 13 Feb 2003, Panu Matilainen wrote:
> 
> > On Wed, 12 Feb 2003, William F. Acker WB2FLW +1-303-777-8123 wrote:
> > > > 
> > > Hi Tony,
> > > 
> > >      The last Valhalla kernel with which the installer could be built was
> > > kernel-BOOT-2.4.18-10.  Just build the installer using that kernel, then
> > > update the tree to the latest kernel, as well as everything else, do the
> > > genhdlist and pkgorder and the rest as normal.  When you install, you'll
> > > get the latest kernel, not 2.4.18-10.  ISO's done this way can be found at
> > > ftp://linux-speakup.org/pub/speakup/disks/redhat/7.3/ although they're a 
> > > bit stale now.
> > 
> > Note that while you can do this, it breaks kudzu to some extent (been
> > there, done that...). At the postinstall stage kudzu probes for USB and
> > sound at least, the problem is it tries to locate kernel modules for its
> > findings from /lib/modules/<version of running kernel>/ path and if the
> > versions differ it fails, resulting in things like USB mouse working at
> > install time but not on the first boot before you manually add "alias
> > usb-controller.." to /etc/modules.conf etc. Kudzu from rawhide can be told
> > where to look for the modules so you could patch anaconda to make kudzu 
> > look for your actual kernel version.
> > 
> > 
> Hi,
> 
>      That explains why, at some point in my series of updates, the system 
> would hang at the "performing post install configuration" stage if my USB 
> floppy was attached.  However, I'm a bit confused about which modules 
> Kudzu is going after.  We know that we're using the BOOT kernel for the 
> install.  Therefore, whatever version of modules found in the /lib/modules 
> directory of the newly installed system would be wrong in one way or 
> another.  I wonder if the kernel-BOOT package is installed and removed at 
> the end.  We could probably prove this by interrupting the installation 
> process at just the right point.

Easier to read the source :) Kudzu has (at least until recent rawhide
versions) special logic in it when 'uname -r' is x.y.z-rBOOT, in that case
it just strips out the BOOT and looks for kernel modules under
/lib/modules/x.y.z-r/ directory of the installed system image.


-- 
	- Panu -





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]