[Clfs-dev] Weirdness.

[Ken Moffat]

On Wed, Dec 26, 2007 at 01:04:27PM -0500, Bryan Kadzban wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: RIPEMD160
> 
> Ken Moffat wrote:
> >  Took a look at the udev rules - the host (running udev-113) had a
> > 70-persistent-net.rules file which associated the mac address with
> > eth0.  To my knowledge, I never deliberately created that, and the
> > timestamp suggests it was generated around the time I had first
> > booted the host.
> 
> Yep, CLFS doesn't generate that file (which is probably good -- udev-118
> doesn't have any way to pre-generate that file anymore; the downside is
> that you can't create config files or directories for your NICs until
> after you reboot, if you have more than one ethX device, because the
> kernel will enumerate them randomly until you have rules to rename
> them).  But the udev rules from upstream do create it, assuming the root
> FS is writable when that rule runs.  (If not, it'll put them into /dev
> and the bootscripts are supposed to copy them to the real root FS
> later.)
> 
 OK.
> > Copied it over to the new system, almost booted -
> > got a dhcp lease, ntp came up, but then mount.nfs failed 'no such
> > device'.  And the keyboard was now inoperative.
> 
> That's ... odd.  Having never done NFS myself, I'd guess that the "no
> such device" is just a generic error from the mount syscall; the NIC
> device obviously does exist if you were able to get a lease on it.  It
> looks like something may be going wrong higher up the NFS stack?
> 
 Possible, but no obvious issues when I copied the kernel and
modules to an older system.

> The keyboard issue is really odd, since the keyboard doesn't require
> udev AFAIK.  (...Well, at least, it doesn't on x86; maybe that's
> different on PPC?)
> 

 Lots of things differ, but that isn't something I would expect.
> > This is my first
> > experience with udev-118, maybe I've missed a change (I got the
> > change in the make and install commands, but maybe there is
> > something else I should have changed ?)
> 
> There are standard udev rule files that get installed now, and there's
> the change mentioned above where it can't pre-write NIC-naming rules,
> but I don't think either of those are causing your issue.
> 
> Without a keyboard, it's hard to do much -- can this box use a serial
> console?  It'd be interesting to see whether there are any NFS related
> errors in the kernel logs after the mount fails.

 So far, I've avoided serial consoles.  Looking round the back (it's
a G4 mac mini) the only practical option might be a net console -
I'll need to read up about that, and I think it might need a
reconfigured kernel.
> 
> Alternately, does it work any differently if you use init=/bin/bash and
> run the bootscripts manually, one at a time?

 Would you believe still no keyboard - even on the original host
system running its earlier 2.6.23 kernel.  The config came
originally from debian - I know there is still a lot of bloat
(unlikely things as modules) in it, maybe something for the keyboard
has been defined as using a module, or maybe it does indeed need a
device.  I've avoided touching that part because the PS/2->usb
keyboard works fine, even on the British extra '\|' key - unlike on
my G5.  My first thought yesterday was that a nic module was no longer
being loaded, so I unmodularised that part but it didn't seem to
help.

 Looks like I'm going to be spending a long time playing about with
different configs, fixed ip, fixed /dev :-(  Thanks for the
suggestions.

ĸen
-- 
das eine Mal als Tragödie, das andere Mal als Farce