While upgrading RedHat 5.2->6.0 on a headless SS20 dual HyperSPARC, /etc/inittab is replaced with a version that expects to use virtual consoles, and *only* virtual consoles. As a result, the system becomes inaccessible once it enters multiuser mode, if the network in unavailable (as it became for me due to a ypbind startup bug -- separate bug report has been filed.) RedHat 5.2 handled this flawlessly out of the box.
This bug still exists on the recently released images (6/2/99). Workaround: Log in single user mode. Edit /etc/inittab. Add a line that looks like this: t1:2345:respawn:/sbin/getty ttyS0 19200 vt100 Permanent solution: test your software *before* releasing it. There's not way to miss a bug like this.
I have verified this to be true on a sparc in our test lab.
The old inittab should be saved as inittab.rpmsave (if not, that is definitely a bug...); the problem is, that that's a change made to a configuration file, and we can't sanely munge config files with RPM. AFAIK, serial consoles were never set up in inittab by the installer... ------- Email Received From "H. Peter Anvin" <hpa> 06/14/99 15:47 -------
Shouldn't that change to the inittab file be: sc:12345:respawn:/sbin/getty cua0 9600 vt100 (almost verbatim from a 5.2 installation) You forgot run level 1 and the "default" serial port speed is 9600. Yeah higher is nice but the chances are that if you are installing from a default machine the serial ports are at 9600. Also, the virtual terminals are commented out in my 5.2 install - should that be done too when "fixing" a 6.0 install?
> sc:12345:respawn:/sbin/getty cua0 9600 vt100 This is incorrect; the /dev/cua* devices have been deprecated for some time in favor of /dev/ttyS*. Additionally, you do *not* want to start a getty in runlevel 1 (single user mode); Linux dumps you to a shell automatically in this runlevel, and an attempt to start getty would just cause problems. You are correct that the baud rate should probably be set to 9600.
This should be fixed in the 6.1 installer.