era@scdpyr.ucar.edu (Ed Arnold) (11/23/88)
We recently upgraded a 3/280 running 3.5 to a 4/280 running 4.0. The
hardware includes two of the Systech MTI serial port devices (mti0, mti1
in the config file). Before and after the upgrade, the 32 serial ports
were hooked to a Gandalf PACX terminal switch, and we had had no problems
with serial ports while running 3.x.
As soon as the upgrade had been completed, it was obvious we had serious
problems. The console was cluttered with "getty failing, sleeping"
messages, and the system would crash every 1 to 2 hours. Sun supplied us
with something called the "general tty patch" (we were told that it's not
on the 4.0.1 tape), and this seemed to fix the worst of the problems. At
least, the system no longer crashed every 1-2 hours.
However, we're still having some problems. The console is now cluttered
with "getty: /dev/ttyXX: I/O error" messages, and the system still crashes
occasionally with a data fault in getty. There is some evidence that some
of the ports go into limbo, and users trying to login experience
occasional weirdnesses. (We are running getty with the new "Auto-baud"
clause that's in the 4.0 gettytab.)
Is there anyone out there with a similar configuration who has experienced
these problems? If so, please e-mail me, I'd like to discuss what you've
found out about this.
Ed Arnold * NCAR (Nat'l Center for Atmospheric Research) * Mesa Lab
PO Box 3000 * Boulder, CO 80307-3000 * 303-497-1253
era@ncar.ucar.edu [128.117.64.4] * {ames,gatech,noao,...}!ncar!eraera@scdpyr.ucar.edu (Ed Arnold) (12/03/88)
In v7n23, I requested info from anybody who has had the same type of
problems we experienced with the MTI serial ports on our 4/280 running
4.0.
Sun responded with a solution to this one fairly quickly. If you should
experience the same set of problems:
1) Install the "general tty patch" which I discussed in my earlier note.
You'll have to call Sun to get it.
2) Configure your streams buffer pool so that "netstat -m" doesn't show
any allocation failures. (This is not discussed, so far as I could
find, in the advanced kernel config guide.) Don't waste your time
trying to put "options NBLK4096=8,NBLK..." into your config file,
it doesn't work; edit the NBLK* parameters in param.c.
3) If, after this, you still find "getty: /dev/ttyXX: I/O error"
messages on your console, then you may have a user who has
left behind programs in your process table that are trying
to write to tty(s) on which there is no carrier.
Incidentally, "#define MAXUPRC (NPROC-5)" is still in server param.c under
4.0, allowing ordinary users to (smack gulp munch) your server's process
table. Caveat emptor.
Ed Arnold * NCAR (Nat'l Center for Atmospheric Research) * Mesa Lab
PO Box 3000 * Boulder, CO 80307-3000 * 303-497-1253 * 303-494-6949 (home)
era@ncar.ucar.edu [128.117.64.4] * {ames,gatech,noao,...}!ncar!era