Solaris 10 11/06 x86_64: almost impossible to shutdown the system

Hi,

After upgrade to 11/06 it became almost impossible to shutdown the
system. There are multiple things that doesn't work.

Things to exclude right away as a source of problem:
- I do not have LD_LIBRARY_PATH set
- I do not have broken package dependencies
- System is upgraded without errors
- I have excluded (off) my own init and smf sctipts/methods for testing
- System boot works without single error or warning on the console or in
the syslog.
- I usualy know how to deal with unix systems for a 10 years and 1 month
from now on almost daily basis.

When I try to do normal shutdown (shutdown -g0 -i5 -y) I get almost all
services stopped, but shutdown fails to stop to the end, to sync the
disks and turn off power. I do not have console login avaiable when this
happens, so the only two things I can do is manually shut down machine
with half-live Solaris - it will do partial SVM mirror resync on the
next boot, or if I have been logged with ssh from other machine, ssh
connection is still alive, so I can type "poweroff" from there - machine
will be powered off normaly.

This is /var/svc/log/svc.startd.log :

pro 25 13:22:39/106: application/graphical-login/cde-login:default
failed: transitioned to maintenance (see 'svcs -xv' for details)
pro 25 13:28:17/1: restarting after interruption
pro 25 13:28:29/6: couldn't get psinfo data for
svc:/system/console-login:default (No such file or dir
ectory); assuming failed

I have tried "svcadm disable -t cde-login" before shutdown, and find out
that this error probably doesn't have anything with initial problem,
because symptoms are the same.

The next method I tried is to kill manually critical services (innd,
postgres, audit, logout from X window system) and then type poweroff
while still in runlevel 3 and have shell prompt. System types on the
console "Syncing disks..." done, but after next boot, boot is taking
longer (journal commit?) and "metastat -c" shows that some partitions
are in partial resync mode. This doesn't happen if I do poweroff from
failed shutdown (runlevel 5) or remote in single user mode.

Single user mode (init S) also doesn't work. It never prompts me for
root password, it kills system services and hangs system console,
leaving unusable system (if I was not logged remote with ssh before
doing this action).

I have written ad-hoc script to shutdown some services, shutdown system
to runlevel 5, sleep 60 and poweroff, altrough I trap every possible
signal in script, script gets killed and poweroff never executed after
sleep 60.

In short:
- if I poweroff system while in multiuser mode, I get inconsistent soft
RAID, and unclear filesystems (unmounted properly), thus risking fs
corruption sooner or later.
- if I do shutdown -g0 -i5 -y, system doesn't poweroff, leaving me
hanged console - same thing with for single user mode

Currently, only way to safely shutdown system is to log in with ssh from
laptop, su to root, then logout from windows system on Solaris machine,
go to console login, type init S or shutdown ... after console hangs, go
to laptop where ssh connection is luckily still alive and type poweroff.

Does anybody has this problem with update 3 on x86/amd64?

I don't have anything interesting in /var/adm/messages, and neither
/var/svc/log/* is helpfull.

Please help! I do not know how or what to try anymore. Google doesn't
find many similar cases like mine, and if it finds, there is no solution
or indices what can be wrong.


--
Man is something that shall be overcome.
-- Friedrich Nietzsche
.