From: "Szigeti Szabolcs" <szigi@ik.bme.hu>
Subject: [TUHS] Re: Porting Unix v6 to i386
Date: Tue, 4 Jun 2002 13:53:50 +0200

I've just joined your mailing list, and while looking at the archives, i saw
this discussion on porting V6 to Intel.
Well, back in '92, in a university scinence students' competition, I ported
v6 to intel286 in protected mode. (I got second place, the winner was a 3D
animation prog, which is more spectacular, than a # prompt :-)

I used Borland C to compile, with some extra mungling the assembly code ( i
can't remember why).

The hard part was to understand the protected mode, and to write the low
level stuff. Other things, like filesystem, etc. compiled with hardly any
modification. (Had to change =+ to =+, introduce long insted of int[2],
etc.). It has floppy, ide, kbd, parallel, serial and vga drivers.

It got to full multiuser operation, but there are bugs and stupid codings
certainly. The C compiler, nroff, and some other parts were not ported.
(Yes, I used Borland C to compile the programs, and a tool to convert it to
a.out :-)

If anyone wants to experiment with it (some parts are pretty ugly, because i
intedned to rewrite it, but never did, and the parts of the comments are in
Hungarian), drop me an email (though i'm now doing my MBA thesis, so might
not answer immediatey). If there is interest, I can summarize my
experiences.

I've not looked it since several years, so i might not remember every
detail, but there are some interesting point, and it was great fun to do.

Some notes:

The CONF, DMR, KEN and H directory contains the kernel sources, while
INCLUDE and SRC the rest of the system.
Tools in UTIL is used to compile these. c2a processes the kernel assembly,
to change the segment names produced by the c compiler because of the
protected mode (I used the command line version of Borland C++ 3.0, the
makefiles are for this comiler).
exe2v6 converts MSDOS exe to a.out (not all exe, it has to be compiled
specially, see the makefile in SRC/UTIL). exetr is a horrible kludge to
transfer files to v6, via a floppy disk, and mkfs is mkfs under Dos.

The kernel compiles in a funny way, you get a unix.exe, which can be run
from dos, and if all is well, it can retorn to dos, with a magic key
sequence (shift-esc ???). It speeds up development, cause you don't have to
reboot.

The code is really ugly at some places, because i started experimenting with
protected mode first, and never cleanly rewrote what worked. Once I startded
to tidy up the code (move to ANSI C, etc), but never finished. Morover, in
the process, I broke something in the fs code, so right now it won't mount
root. Plus, much of the comments are still in Hungarian :-)

The kernel makes heavy use of the special 286 protected mode features
(Tasks, etc.)

More notes:

First of all, the code is currently broken. Something is wrong, and it will
not be able to mount the root fs or something, but shouldn't be hard to fix,
because it used to work, and i did something wrong when playing with the
code some years ago, with no backup to revert to.

Originally, I used the commnad line tools of Borland C++ 3.0 to compile. You
need Turbo Assembler, so the Borland C++ 1.0, which is freely available from
Borland as "antique software" is not enough, you'll need the professional
version at least, because it has the command line assembler.

There is a top level makefile, which will create UN.EXE, this is the unix
kernel. You run it under plain MSDOS, it will switch to protected mode, and
with shift-esc it will return to DOS (do a sync before !). Probably it
wouldn't be hard to make it work without DOS as a boot loader :-), but i
never did it.

If compiled with -DSTANDSH, the kernel will be built with a built-in shell
(see KEN/UPROC.C for the command), that can be used to do some basic setup.
It will for example create the /dev/tty entries on a blank root disk. (You
still need the mkfs utility under dos to make a blank disk). I suggest to
use a floppy as a root disk for initial trials.
If you omit STANDSH, it will look for /etc/init to run.

On the directory layout:
CONF contains c.c, the device switch table
DMR contains most of the device drivers
H contains the kernel header files
INCLUDE - the user header files
KEN - rest of the kernel

The low level kernel code is very ugly, as it evolved from experimenting
with protected mode. There are lots of things, that i'm not proud of :-)

SRC/CLIB0 - the C library
SRC/SYSC - the syscall user side interface
SRC/UTIL - the utilities and such, most should work
UTIL - MSDOS utils to compile

In the UTIL directory, there are the following utlities:

c2a - used to compile the kernel, it search-and-replaces some stuff in the
assemlby code, i don_t exactly remember why, but it has to do something with
the segmenent names.

exe2v6 - utility to convert MSDOS exe program to a.out. See
SRC/UTIL/Makefile for how to make such an exe. The programs have a built in,
fixed size stack, because Borland C had problems separating code and stack
segment. It could brobably be fixed, but i didn't do it, instead opted for
fixed size stack, adjustable at compile time.

exetr - it can transfer a file to unix via a floppy drive, i don_t remember
how it works, i think you put one file to a floppy, and use the kernel-shell
to write it to a file-system. It is a really ugly idea :-)

dumpfs - it is broken, it was intended to back-up a filesystem

mkfs - MSDOS mode mkfs, answer the questions, you'll get the filesystem.
There is no sanity check whatsoever, so be careful, it can screw your
harddrive, if given wrong sector numbers.

So that's it, if someone wants to experiment with it, i can probably answer
questions, but, i have to do some reverse engineering too!

	Szabolcs Szigeti