[5054] in testers
Re: "krtld: load_exec: fail to expand cpu/$CPU" on reboot
daemon@ATHENA.MIT.EDU (Garry Zacheiss)
Mon Jul 23 17:09:30 2001
Message-Id: <200107232103.RAA00888@test-sun2.mit.edu>
To: ccount@mit.edu
cc: testers@mit.edu
In-Reply-To: Your message of "Sun, 22 Jul 2001 18:54:37 EDT."
<200107222254.SAA00699@birdland.mit.edu>
Date: Mon, 23 Jul 2001 17:03:21 -0400
From: Garry Zacheiss <zacheiss@MIT.EDU>
Ok, I understand what's going on here now. In brief:
When 9.0.10 was released, there were unreleased changed to the
os volume in the dev cell that were carried forward to the athena cell;
however, the stat files used by the public workstation os verification
knew about the unreleased changes.
The particular file that concerns us is
/platform/sun4u/kernel/cpu/sparcv9/SUNW,UltraSPARC. This file appears
to have been added between the -03 and -08 revisions of the Solaris 8
kernel patch, but doesn't exist in the os volume associated with 9.0.10
(it started appearing in the dev cell when 9.0.11 went out).
PUBLIC=true machines in either the dev or athena cells that
updated to 9.0.10 attempted to update SUNW,UltraSPARC at boot time when
oscheck ran, failed, and deleted the file (this is probably a bug). On
the next reboot, (say, when the machine took 9.0.11) it failed to boot.
The machine still was able to boot the 32 bit kernel, which is why "boot
kernel/unix" would work. For Kevin's machine, which had dev cell packs,
the problem then fixed itself because oscheck ran, the file existed in
/os, and it was copied into place. This didn't happen for Craig's
machine because 9.0.11 doesn't exist in the athena cell yet, so the file
couldn't be copied into place.
The good news is that when 9.0.11 is in place, this won't
affect any more machines (although we should be mindful of similar
failure modes in the future) so this isn't a showstopper for going
public tomorrow night. The downside is that this will kill off
PUBLIC=true early machines that haven't rebooted since taking 9.0.10
unless we do something. Greg is working on something now.
Garry
