home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
minnie.tuhs.org
/
unixen.tar
/
unixen
/
PDP-11
/
Distributions
/
ucb
/
2.9-derivatives
/
2.9BSD-MSCP
/
Older
/
2.9BSD-MSCP.txt
next >
Wrap
Text File
|
2001-08-24
|
14KB
|
357 lines
Introduction
------------
This describes a port of 2.9BSD that includes support for MSCP hard
disks and VTserver. It will run on a very minimal PDP-11/23.
I developed this using simh on a PC, debugged it under Ersatz-11 and
tested it on a real PDP-11/23 consisting of the following configuration:
- KDF11-AA CPU
- MSV11-L 256 Kbyte RAM
- Andromeda UDC11 MSCP hard drive controller (including boot ROM)
- a custom 3M board including 2 SLU and LTC
- a 20 megabyte Seagate ST-225 hard drive
- a dual-sided 5 1/4" floppy
While I have many years experience in C and assembly programming,
embedded firmware, and operating systems, this is my first experience
with porting or modifying UNIX, or working intimately with a PDP-11. I
seem to have gotten this working, but there is no guarantee that I've
done it right. I welcome any suggestions for improvements, and will make
neccessary corrections. On the other hand, I do not want maintenance of
2.9BSD to become a major continuing project. I have several small
PDP-11/23 systems in my collection. My main motivation for doing this
was to have a decent and interesting operating system for these
machines, a legal operating system for those I wish to dispose of, and
to make UNIX available to others who have or want to build similar
PDP-11/23s. After this project I would like to turn my attention to the
PDP-11/53 and 2.11BSD.
Differences
-----------
This version of 2.9BSD UNIX is the same as the standard distribution on
the PUPS archive, with the following changes:
- the MSCP (ra) hard disk driver from the pro350 release has been ported
in. The block device number is 2, the character number is 6. The
autoconfig probe routine in uprobe1 is hardwired to always return
"device present". If you recompile this system with a different disk
as root, and then try to access a non-existant ra drive, the system
will panic.
- only the ra (4 units), rl (4 units), and xp (3 units) disk drivers
have been compiled in. The other drivers have been commented out in
dtab. One reason is to save space. Another is that simh seemed to get
confused while autoconfiging xp disks if certain other devices were
compiled in. I included the rl mainly for the simulators, for
compatibility with released disk images. I have no idea what an xp is, I
just needed a large hard drive for the simulator.
- The system boots from ra0, and ra0a is the root.
- the partitions are as follows:
ra0a: 3200 1K blocks (/)
ra0b: 1920 1K blocks (swap)
ra0c: the remainder (/usr)
ra0h: the entire disk
There are other partitions defined, that I left as found, but I've
never used them. The pro350 version of the partition table had a bug:
the virtual cylinder size in rareg.h and the partition table disagreed.
It has been consistantly set to 64 disk (512 byte) blocks. 3200 might
be rather smalll for a root partition, but it was neccessary in order
to fit a sysgenable system onto a 20 meg hard drive. I have not tested
this with a second drive.
- The standalone ra driver has been built into the standalones. I had to
modify this slightly to make it work.
- The standalone vt driver (for VTserver) has been built into the
standalones.
- a new ra boot block has been written, based on the 2.11 version of
rauboot.s. The boot program must be named "boot", and must be located
in the root directory. This boot block prints characters on the terminal
to report its progress:
I Initializing the MSCP device
B reading a block of the root directory
-<name> checking a filename. Only the characters checked are printed.
. reading a block of "boot"
S starting the boot program
F failed to find the boot program
- a debug boot block, testboot, has been written. It implements two
terminal commands:
d <addr> <size> -- dump memory
m <addr> <data> ... -- modify memory
Distribution
------------
The distribution is setup to be loaded onto a clean PDP-11/23 using
VTserver. The following files are included:
- boot: the boot program. It can load programs from several devices,
including ra and vt.
- testboot: a standalone debug program that can be loaded from the boot
sector or via boot.
- mkfs: initialize a partition with a file system
- icheck: check a partition
- restor: restore a dump file onto a partition
- root.dump: dump of the root partition
- usrsmall.dump: a dump of /usr that will fit onto a 17 megabyte
partition. Many files and directories have been stripped in order to
get this to fit, including:
/usr/70
/usr/ingres
/usr/src except what is neccessary to rebuild unix
/usr/man/cat*/*
/usr/contrib
/usr/net
/usr/lib/learn
and probably some other stuff
- usrmid.dump: a dump of /usr that will fit onto a 37 megabyte
partition. It is missing only /usr/ingres and the "learn" package.
- usrfull.dump: the entire contents of the original usr.tar in dump
format. A df of a full /usr reports 38398 blocks used.
- vtserver.exe: this is a PC executable of VTserver that has been
modified to work with a PDP-11/23. The regular VTserver for some reason
doesn't work with the 11/23. I had to add initialization of SP, PS, and
PC, then send "p", rather than "140000g" to start the bootstrap. This
version will probably go away once Fred van Kempen releases the next
official version of VTserver.
- .vtrc: a sample Vtserver init file.
- vtserver.c: the source code for vtserver.
- vtserver.dsw and vtserver.dsp: Visual Studio project files.
- vtreadme.txt: Warren Toomey's vtserver documentation.
- 2.9BSD-MSCP.txt: this document.
I have tested usrsmall.dump, but not usrmid and usrfull.
The original pro350 patches have been added under /usr/src/sys/pro350.
How to load 2.9BSD onto your PDP-11/23
--------------------------------------
I assume that you have a PDP-11/23 with a Seagate ST-225 hard drive, and
a PC acting as host. If you have a larger hard drive, the procedure is
much the same, except you will have more space left over, and you might
be able to load a larger /usr image.
It might be best if you can first use some other software, such as
RT-11, to make sure your hardware, including the disk subsystem, is
setup and working. So far I have only used the Andromeda UDC11
controller. The UDC11 stores the hard drive configuration in an onboard
non-volatile memory. The setup and formatting utilities run under RT-11.
I have several RQDX3s that I eventually have to conquer. I understand
that these need to be set up using a standalone diagnostic system called
XXDP.
The UDC11 can partition a hard drive into several logical drives. For
2.9BSD, it is best to setup the disk as a single logical drive, and let
UNIX handle the partitioning. 2.9BSD uses a more primitive hard drive
partitioning scheme than 2.11: the partition table is hard-coded into
UNIX, as ra_sizes in ioconf.c. If you want to change it, you will have
to recompile.
Configure, format, and qualify the hard drive. THe UDC11 utilities setup
the controller to reserve the last track for a transparent bad block
replacement scheme, so UNIX will see an ST-225 as having only 614
cylinders. The ST-255 has a good reputation for robustnes and longevity.
I have found quite a few that are 15 years old and still work well.
After formatting I run the qualify routine overnight (about 200 passes).
Do not use a disk that generates many errors.
If you are using the RQDX3 or another controller, you are on you own
for now.
I assume you have Kermit or another terminal emulator on your PC.
Unpack the distribution into its own directory on the PC.
Run VTserver. Make sure that any other program that might allocate COM1:
is closed. I have found that if you have run Kermit in a DOS box, and
then run Vtserver, VTserver will not be able to allocate COM1:. You will
have to close the DOS box and open a new one.
My flavor of VTserver implements a couple control characters: ^B sends a
break, and ^A resends the VT boot block. Once you have the PDP-11's
attention by sending break hit ^A and return. VTserver should transmit
the boot block, then load "boot" from file 0. If you are not familar
with VTserver, read Warren's document, vtreadme.txt.
Boot will ask you what program you want to load. The standalone programs
(boot, mkfs, icheck, and restor) use device names of the form dd(a,b) or
dd(a,b)name, where dd is a device name, such as "ra" or "vt", a is a unit
number, and b is an offset. For vt, the offset is the index of the file
in .vtrc. For ra, the offset is the number of the first 512 byte disk
block of the partition.
The default .vtrc organizes the distribution files like this:
0: boot
1: testboot
2: mkfs
3: restor
4: icheck
5: root.dump
6: usrsmall.dump
7: usrmid.dump
8: usrfull.dump
First, load mkfs from vt(0,2). Initalize ra0a as ra(0,0) with a size of
3200. Mkfs (and the rest of the standalone programs) use 512 byte disk
blocks to locate partitions, but 1K blocks for the rest of their logic. I
have not experimented with different interleave factors yet, I just
accept the defaults. When mkfs is complete, it should return to the boot
program. If not, hit ^B, ^A, and "return" to reload boot.
Next load restor from vt(0,3). Restore root.dump from vt(0,5) to ra(0,0).
At this point you can boot unix via VTserver. Tell boot to load
"ra(0,0)unix". At this point you may have to quickly quit VTserver and
switch to Kermit, because UNIX will soon set the terminal to 7 bits even
parity. You should see the UNIX banner and the autoconfig report. When
you see a # prompt, you will be at a UNIX shell prompt.
The next task is to copy the boot block using:
"dd if=/mdec/rauboot of=/dev/rra0a count=1"
Sync and restart the system. The PDP-11 should now boot UNIX on it's
own without help from Vtserver.
Next, create a file system on /dev/ra0c using the regular (not
standalone) mkfs. The size calculation is as follows for an ST-225:
cylinders: 615
tracks/cyl: 4
sectors per track: 18
controller reserved area: 1 cyl
total 1K blocks = (615-1)*4*18/2 = 22104
blocks for ra0a: 3200
blocks for ra0b: 1920
blocks for ra0c: 22104-3200-1920 = 16984
standalone offset for ra0c: (3200+1920)*2 = 10240
The command to init ra0c would be:
mkfs /dev/ra0c 16984
Sync and reboot VTserver. Double check the disk formatting by running
icheck from vt(0,4). Check both ra(0,0) a.k.a. ra0a, and ra(0,10240)
a.k.a. ra0c.
Load restor from vt(0,3). Restore usrsmall.dump from vt(0,6) to ra0c at
ra(0,10240). If you have a larger hard drive, you can load usrmid or
usrfull instead. The destination address will be the same. This will
take several hours.
When this restor is complete, reboot. When you get the # prompt, hit ^D.
The next prompt should be "User:". Log in as "root" and you should get a
shell prompt. UNIX installation should now be complete.
To verify a correct installation, you can cd to /usr/src/sys/RA, and
recomplile the system by typing "make unix". This will take a while.
When complete, "diff unix /unix". They should be the same.
To rebuild UNIX
---------------
This unix is built in /usr/src/sys/RA. Do not reconfig from conf: I've
edited files in RA by hand.
cd /usr/sys/src/RA # unix
make unix
cp unix /unix
cd /usr/src/sys/stand # boot program
make
cp boot /boot
cd /usr/src/sys/mdec # boot block
sh -v mkra
cp rauboot /mdec
dd if=/mdec/rauboot of=/dev/ra0a count=1
Note that if you regen boot for VTserver, you will have to strip the
header from the copy of boot that VTserver uses via "dd if=boot
of=boot.vt bs=16 skip=1".
How to set the correct year
---------------------------
This software probably has Y2K bugs in it. UNIX will keep the right year
if you can get it set. The "date" command can't handle dates outside
19xx. To set the year to (for example) 2001, do this:
date 9912312359
wait for a minute until the year rools over to 2000.
date 12312359
wait until it rolls over to 2001
set the correct date, omitting the year.
After that, every time you boot, you will have to set MMDDHHMM again, but
the year will be correct, since UNIX evidently remembers this somehow.
Hints for developing UNIX software using simh and E11
-----------------------------------------------------
If you cannot boot UNIX after installing it, you will have to do some
debug. I debugged the boot block, boot, and unix "main" by pepperring
them with printfs and recompiling. I edited and managed files on the PC
using normal PC tools, compiled on an rl and xp based 2.9BSD system
running under simh, and debugged the ra drivers using E11. I transferred
files between the PC and simh as tar files via /dev/xp2h. Files were
transferred between simh and E11 via disk images.
You can gather files on the PC and send them to a running simh like so:
emacs whatever.c # microemacs
fix whatever.c # utility to strip \r
tar cvf \pdp-11\simh\xfer.tar whatever.c # MKS or Cygwin tar program
# xfer.tar is attached to rp2 and available in UNIX as /dev/xp2h
Then unpack them from UNIX under simh by typing
tar xvf /dev/xp2h
Going from UNIX to PC, under simh:
tar cvf /dev/xp2h whatever.c
sync
and on the PC
tar xvf \pdp-11\simh\xfer.tar
When transfering files from simh to E11, the only sure way to make sure
that the changes are flushed to the hard drive is to sync UNIX, then ^E
to break to the simh command shell, and detach the disk image that will
be transferred to E11: "det rl3".
--
Jonathan Engdahl Rockwell Automation
Principal Research Engineer 24800 Tungsten Road
Advanced Technology Euclid, OH 44117 USA
Euclid Labs engdahl@cle.ab.com
http://users.safeaccess.com/engdahl