Sun-Spots-Request@RICE.EDU (William LeFebvre) (12/04/87)
SUN-SPOTS DIGEST Thursday, 3 December 1987 Volume 5 : Issue 66
Today's Topics:
The archives (2)
Getting the 4.3 timed running under SunOS 3.x
Re: Sun-3/50 hardware bug causes segmentation faults in user programs
Read-Modify-Write bus errors on VMEbus explained
Inbound mail failures
HP LaserJet printcap and a request
Problem with inverted suntools on 3/110
MicroVaxes and Suns on the same Ethernet?
Cheap Source of Simms for 3/60?
How to identify a Sun console?
UUCP and NFS questions
SunView fonts
----------------------------------------------------------------------
Date: Wed, 25 Nov 87 13:28:14 EST
From: lrr%nascar.Princeton.EDU@princeton.edu (Larry Rogers)
Subject: The archives (1)
Where is the sun-source archive? Much thanks.
Larry Rogers
Princeton University
Computing and Information Technology
------------------------------
Date: Thu, 3 Dec 87 21:01:08 CST
From: William LeFebvre <phil@Rice.edu>
Subject: The archives (2)
I have answered this question on an individual basis often enough to
justify, I think, posting this information in a digest.
In addition to editing and distributing the list Sun-Spots, the computer
science department at Rice University also maintains several Sun related
archives. These archives are directly available to the Internet through
an anonymous FTP service. There is also a mail-response archive service
that provides access for those who do not have direct Internet/ARPANet
access.
There are three areas of interest to the Sun-Spots reader. Of course,
there is the sun-spots archives themselves. Currently, all issues of
volume 5 of the Sun-Spots digest are on-line, and each issue is added to
this area as it is sent to the list. The second area, called sun-icons,
contains a collection of source files for icon and background pattern
bitmaps (as produced by iconedit). All the icons were originally
submitted to Sun-Spots, but their long-term interest nature makes the
archive a useful endeavor. Finally, there is the sun-source area. This
is a collection of source code submitted to the Sun-Spots list.
Typically, source is too large to include in a digest, so it is extracted
and placed in this archive for those who are interested. Source is not
necessarily in C---it could be shell programs, Lisp, MLisp, or even diffs.
To access the archives from the Internet, one need only FTP to the host
"titan.rice.edu" and log in as the user "anonymous" (with any password).
This places one in a special part of the file system with several
subdirectories: "sun-spots", "sun-source", and "sun-icons" (there are
other directories that contain other publically-accessible files). Each
subdirectory contains at least two files: "00directory" and "00index".
The former is the output from a recent "ls -l" in that directory (or, in
some cases, an "ls -lt"). The latter is a more descriptive record of the
directory's contents. These files are usually up to date. Most
everything in the archives (including the icons) is in 7-bit ASCII form
and can be transferred with the default FTP settings. But if there is
ever a need to get a tar file or other 8-bit data, one should use the
"image" transfer type. Once again, please note the full name of the
machine: "titan.rice.edu". The machine named just "rice.edu" is
completely different and, in fact, will not allow anonymous FTP access.
The archive server is an automatic mail response program. It scans mail
messages sent to it for commands to execute. These commands can request
the transfer of indices or programs. The server's address is
"archive-server@rice.edu", but the address "archive-server@titan.rice.edu"
will also work. This server is identical (except for certain installation
dependent items) to the archive server running on "decwrl.dec.com",
written by Brian Reid. Those familiar with his server will be very
comfortable with this one. For additional information about the server,
one should send a mail message containing only the word "help" to the
server. If there is ever a need to talk to a person rather than the
program, the address "archive-management@rice.edu" will suffice.
William LeFebvre
Department of Computer Science
Rice University
<phil@Rice.edu>
------------------------------
Date: Tue, 24 Nov 87 17:07:01 EST
From: dsc@seismo.css.gov (David S. Comay)
Subject: Getting the 4.3 timed running under SunOS 3.x
Since i have seen requests in getting timed running on suns for some time
now, i thought i would pass along the changes i made here to get timed
running on machines not running 4.3. the enclosed shar file, timed.pch,
includes diffs to the timed source distributed with 4.3 along with some
other modules that might come in handy in porting timed to other bsd based
systems.
dsc
[[ The shar file is 24K---too large for the digest. It is in the archives
under the name "sun-source/timed.pch.sh". It can be retrieved via
anonymous FTP from the host "titan.rice.edu" or via the archive server.
For more information about the archive server, send a mail message
containing the word "help" to the address "archive-server@rice.edu".
--wnl ]]
------------------------------
Date: Tue, 24 Nov 87 16:25:33 EST
From: harvard!umb!ileaf!io!marvin!md@sally.utexas.edu (Mark Dionne)
Subject: Re: Sun-3/50 hardware bug causes segmentation faults in user programs
In SUN-SPOTS Volume 5 Issue 60 John Gilmore writes about a Sun 3/50
hardware bug which causes segmentation faults and asks:
> I'm curious how many other people have seen this kind of problem on their
> 3/50's, and why it hasn't come up in Sun-Spots before.
We have been experiencing this (or a similar) problem since early spring.
We have narrowed the problem down to this description:
Sometimes a register gets mysteriously clobbered during instructions
which start a short interval (typically 6 bytes) before a page boundary.
The problem is repeatable with care, but does not necessarily happen the
first time the instruction sequence is executed. It has only happened
with large executables (say 1.5 meg of text, 4 meg total). It only
happens on 3/50's. The total number of occurrences of the problem that
we have detected has been a half dozen cases.
We wrote a "diagnostic" which executes the same sequence of instructions
as one of the known failures, after trying to force out the next page so
as to ensure a page fault. It never failed. Something more complex is
apparently happening.
After many tries, the best we have been able to get out of Sun is that it
may be fixed by ECO 3051, a PAL and a jumper change. It was supposedly
phased into production March 31, 1987. We have heard that it involves an
interaction between video memory access and page faults.
In the last month we received one board with an entirely new rev level
(many new jumpers) which fixed the problem. Several boards which have
gone back to Sun for repairs have NOT come back with the problem fixed.
The good board we received is labeled "501-1162-12-A", whereas a typical
one in house (such as mine) is labeled "501-1162-02-C". Our guess is that
the PAL changed is U-0801 from 1287-01 to 1287-02, with a jumper from
U-0801 pin 16 to U-0803 pin 8. However, there were many changes between
the boards.
Sun bug numbers 1003173 and 1005372 may be interesting. The second was
briefly mentioned on page 117 of the 3.4 release notes. Also note that
this problem was much worse on Beta hardware. Our machines with SN 005,
006 and 007 would exhibit the problem in cases where newer machines would
not.
A suggestion: since there are only 127 page boundaries per megabyte,
relinking your code with a minor change might make you lucky (or less
unlucky) with respect to this problem. This kind of magic is what seems
to be protecting our software in the field. In-house, we occasionally get
a developmental executable that has the problem, and it goes away with the
next round of changes. And we recheck the difficult bugs on a 3/160!
If you think about it, page faults in code are mostly caused by subroutine
calls and branch instructions, rather than "straight-line" execution.
Perhaps this helps to explain why this problem is so rare.
If the problem is only triggered by certain kinds of instructions, it
would be very handy to have a program that would examine an executable and
tell if any of these instructions are in bad locations with respect to
page boundaries. One can also imagine changes to ld(1) that would prevent
it from loading code across page boundaries.
It's funny how you kind of expect a computer to execute its instructions
correctly. I guess we're all just spoiled. ;-)
------------------------------
Date: Wed, 25 Nov 87 18:54:24 PST
From: hoptoad!gnu@cgl.ucsf.edu (John Gilmore)
Subject: Read-Modify-Write bus errors on VMEbus explained
Any time the CPU wants to access the VMEbus at the same time a VMEbus
master wants to access Sun memory, there is a deadlock, since both use
virtual addresses and their paths must cross at the MMU. The Sun-3/160
CPU circuitry breaks the deadlock by issuing a "bus error/retry" response
to the 68020. (This typically happens many times a second when the VMEbus
is busy.) The bus error/retry works all the time EXCEPT when in a
read/modify/write cycle, which only happens in a few instructions
specifically designed for multi-CPU applications.
This problem was reported to Motorola as the 68020 and Sun-3 were being
debugged, and Motorola responded that they were having the exact same
problem with their PMMU's (which use bus error/retry when they need to
grab the bus to fetch page table entries). They planned to change the
68020 so that it would handle bus error/retry on RMW memory cycles. If in
a particular application you don't want to bus-error and retry a RMW
cycle, your circuitry can look at the RMW pin and avoid issuing a bus
error/retry during such a cycle. (Of course, if your system has a
deadlock possible, you'll have to figure out some other way around the
deadlock.)
I left Sun before this issue was clearly resolved. If Motorola fixed the
68020, all you need is a newer CPU chip. If not, I don't know of a
convenient hardware fix. You could install a bus error handler in the
kernel (trap.s) that notices otherwise unexplained bus errors on RMW
cycles to VMEbus memory, and just asks the CPU to rerun them, but this
will involve hundreds of cycles of wasted time when it happens. Then
again, it's better than a core dump or crash.
John Gilmore
------------------------------
Date: Tue, 24 Nov 87 16:39:58 EST
From: Graham Campbell <gc@bnl-ewok.arpa>
Subject: Inbound mail failures
For those of you with limited disk space, we have just encountered a new
(to us) problem that occurs when you run out of swap space. Inbound mail
may be rejected with the error:
554 rewrite: expansion too long
We contacted Sun about this, found that they were aware of it and
explained it. It seems that when a malloc call fails, this will result.
For us this seemed to happen on processing the "MAIL From:..." command and
caused the mail to be bounced. The only cure they could suggest was to
increase swap space, fixes to sendmail won't be available until at least
4.0.
Graham
------------------------------
Date: Tue, 24 Nov 1987 10:31:53 LCL
From: Gareth J. Barker <GJBARKER@UFFSC.BITNET>
Subject: HP LaserJet printcap and a request
I have sent the following printcap entry directly to Adam@bbn.com; I am
also posting it here in case it is of use to anyone else:
[[ This was received with an ESCape ending almost every line. Presumably
they were once backslashes (considering the context). I have taken the
liberty of changing them to backslashes. --wnl ]]
### HP Laserjet used as lineprinter
laserlp|hplaserjet used as lp :\
:br#19200:\
:du#1:\
:fc#0300:\
:lp=/dev/ttyb:\
:mc#2:\
:sd=/usr/spool/lpd:\
:tr=\Ef:\
:xc#040:
1|lp|lasp|laser portrait :\
:pl#66:\
:pw#80:\
:tc=laserlp:
lasl|laser landscape :\
:pl#45:\
:pw#105:\
:tc=laserlp:
The lasp entry seems to work but I've never used the lasl entry, and I'm
not sure that the pl and pw lines are correct. You will have to change
the lp and sd entries if necessary, and also change the baud rate and page
lengths on the HP. (The latter can be done using the 'menu' button as
explained in the HP manual).
****** Now a request of my own ******
I am interested in rasterfile filters for the HP (to use with lpr -v) and
also filters for plot(1) and plot(3X) output to use with lpr -g. Can any
one help?
Thanks for any help
Gareth GJBARKER@UFFSC.BITNET
------------------------------
Date: Tue 24 Nov 87 14:47:16-PST
From: Allan G. Weber <WEBER%BRAND.usc.edu@oberon.usc.edu>
Subject: Problem with inverted suntools on 3/110
How do I get my Sun-3/110G to put up Suntools in inverse video (white
characters on a black background) and do it in the 1-bit per pixel plane
so it will scroll faster? When I run suntools with the -i option, all it
does is make the cursor white and leave little black boxes at the cursor
position whenever it has to refresh the screen where the cursor is
sitting. The rest of the screen is unchanged (black characters on white
background).
I can get it to come up in inverse video in the 8-bit per pixel plane but
that slows scrolling down too much. I can also get suntools running
inverted in the 1-bit plane in such a way that the 8-bit plane can't be
accessed. Actually all I need is for the shelltools to be in inverse, the
rest can stay the way it was. We are running version 3.2, if that makes
any difference.
Thanks.
Allan Weber
Arpa: weber%brand.usc.edu@oberon.usc.edu
UUCP: ...sdcrdcf!oberon!brand!weber
------------------------------
Date: Wed, 25 Nov 87 16:40:03 +0100
From: John Dunnion <mcvax!csvax.ucd.irl!john@uunet.uu.net>
Subject: MicroVaxes and Suns on the same Ethernet?
Our Computer Centre is setting up a network of workstations distributed
over three or four buildings connected via Ethernet. They would like to
know the answers to the following questions and to hear comments from
people who have had experience with this:
1. Can a MicroVax under Ultrix act as a file server for a number of
diskless Suns? Can a Sun act as a file server for a number of
uVaxes? We know that NFS is available for Vaxes; has anybody had
any experience with this under Unix, Ultrix and/or VMS?
2. Can a MicroVax be remotely booted up from a Sun, and vice-versa?
All information gratefully accepted. Mail may be sent to
john@csvax.ucd.irl or t-wade@ccvax.ucd.irl. Thanks in advance.
John Dunnion.
------------------------------
Date: 24 Nov 87 14:47:08 GMT
From: km@emory.UUCP (Ken Mandelberg)
Subject: Cheap Source of Simms for 3/60?
Now that the Sun 3/60 is out and in fairly wide distribution, have any
cheap sources of Simms developed? Earlier there was some talk of using
Macintosh II simms. This seems unlikely since I count 8 1 Megabit chips on
the Mac simms and 9 on the Sun simms. Apparently the Mac doesn't use
parity. Its a pity since universities are getting Mac simms for under $800
per 4 Meg from Apple. There is also a lively third party market.
I did get a quote from a local Clearpoint memory dealer for Sun simms at
$1300 per 4 Meg with a lifetime warantee.
Has anyone seen anything better?
***
Oh yes, I might as well ask the age old question once again. The 4 meg
limit on the 3/50 is a real pity. It thorougly obsoletes a machine that is
really fine in all other ways. Has anyone had any thoughts about
retrofits? I find it interesting that there is a company that specializes
in memory retrofits for the AT&T 7300 which is a real slim marketplace,
but no one seems to have done it for the Sun 3/50.
Ken Mandelberg | {decvax,sun!sunatl,gatech}!emory!km UUCP
Emory University | km@emory BITNET
Dept of Math and CS | km@emory.ARPA ARPA,CSNET
Atlanta, GA 30322 | Phone: (404) 727-7963
------------------------------
Date: 25 Nov 87 13:27 -0100
From: Dominique Petitpierre <petitp%cui.unige.ch@relay.cs.net>
Subject: How to identify a Sun console?
Recently the sources of a program that identifies the type of a terminal
were posted on usenet (qterm.alt in comp.sources.unix volume 12 issue 25).
It works by sending to the terminal a string (usually ESCZ or ESC[c) and
reading the string sent by the terminal. It is very useful in a network,
where a terminal does not always communicate with the computer through the
same port (/dev/tty).
Is it possible to identify in this way a Sun console? It doesn't respond
to ESCZ nor to ESC[c, and I have found nothing like this in the manual
description console(4s) (Sun release 3.2).
Mr. Dominique Petitpierre |ean, BITNET, EARN, MHS, X.400: petitp@cui.unige.ch
ISSCO, University of Geneva |uucp: mcvax!cernvax!cui!petitp , petitp@cui.uucp
54 route des Acacias |JANET: petitp%cui.unige.ch@cs.ucl.ac.uk
CH-1227 GENEVA (Switzerland)|csnet, ARPA: petitp%cui.unige.ch@csnet-relay.csnet
Tel: 0041/22/20 93 33 extension 2117
------------------------------
Date: 24 Nov 87 23:26:29 GMT
From: ll-xn!atexrd!sda@rutgers.edu (Stephen Ayers)
Subject: UUCP and NFS questions
Two quick (well, maybe not...) questions:
1) Has anyone tried mounting /usr/lib/uucp and /usr/spool/uucp via NFS to
make two systems act as one virtual UUCP machine? Since the Suns come
standard with only 2 serial ports, it would be nice to have two machines
share the UUCP info. The two problems major problems I see are: root
permissions across NFS (does uucp need root), and faking out the system
name on the second system.
For example our current UUCP gateway in atexrd. We would like to hang an
addition two modems off a second machine call geech. Geech would mount
/usr/lib/uucp and /usr/spool/uucp from atexrd. The /dev/cua{0,1} on geech
would be renamed cua{2,3} and the L-devices changed to match. Now aside
from any permission problems geech should be able to call out. The next
road block is when geech calls out to a remote site asking for work, it
will get none for geech even though work may be there for atexrd. Thus
the need for "faking" out the system name.
2) From time to time I have seen a patch mentioned that will allow NFS
root permissions (doesn't change to uid 0 to -2). Where can I get this
patch (keep in mind I'm one of those poor slobs with no ftp access) and
which end needs the patch, the server or client?
Thanks in advance!
Steve Ayers, Atex, Inc., A Kodak Company
{ll-xn,genrad,kodak,munsell}!atexrd!sda
+1 617 276-7384
------------------------------
Date: 25 Nov 87 19:03:33 GMT
From: umix!umich!eecs.umich.edu!hsk@uunet.uu.net (Hwant-Soo Kim)
Subject: SunView fonts
What is the font name of the "embolden" font used in the top-strip of
windows when the default font is set to cour.r.16? I think the cour.r.16
is too thin and cour.b.16 is too thick. The "embolden" courier 16 looks
perfect, but I couldn't locate it. Is it made on the fly or there exist a
font file for it ? I'll appreciate if anybody tell me where it is.
------------------------------
End of SUN-Spots Digest
***********************