BINARY LIBRARY

DEC OSF/1 Resurrection

Project summary

I have had interest in the Mach microkernel for some time, being an interesting historical footnote. I have been searching for an interesting way to play around with it, and especially have been interested in Mach-based Unix systems. I heard about OSF/1 some time back, an (experimental?) operating system from the Open Systems Foundation, ironically not an open source or free software organization, which was a seemingly 100% certified UnixTM, and used the Mach 2.5 kernel. A month or so back while retreading ground, I found that there was a very broken old source release that seemed to contain source for DEC OSF/1 V2.0, which is the following:

• Mach 2.5 (which is a single address space version of Mach + BSD 4.3 (Reno), so technically not a microkernel, but still Mach), which had minor changes for this version of OSF/1 I believe.
• The terminal userland for OSF/1 1.2 with minor changes (the init is from DEC?), which is to say, basically just the BSD 4.3 Reno userland tools.
• A separate repo with X sources, which should have been basically corresponding. The ones for Mips/Alpha were likely closer. i386 was a bit rougher. See below about architectures.
• In the X repo, sources for Motif and various DEC GUI things (DECWindows greeter, DECWindows demo programs and some built in utilities, though not all)
• At least partial architecture support for i386, Mips, and Alpha. DEC had seemingly not worked on i386 in a while since it had some serious code staleness issues and the i386 X server (X386-based) seemingly wasn't exercised.

Don't be fooled, it is not OSF/1 2.0, though I would have liked that. DEC had a separate version numbering scheme. There's a string in the Motif periodic demo that says it is OSF/1 V1.2, which I would agree with, as OSF/1 1.3 moved to Mach 3, a real microkernel.

There being available source piqued my curiosity, so I passed it off to Claude. It determined there were a few missing pieces, but nothing major. Being as it's October, it seemed interesting to try to resurrect it. It's alive!

After a few weeks of work with Claude, iterating in QEMU, then 86Box, then finally real hardware, I have a mostly working kit to build a working disk image of OSF/1 i386. There are a few profiles with levels of historicality, though I doubt DEC ever intended it for i386. The terminal version is the most accurate, followed by X, followed by a custom ease-of-use version where I have added the following:

• PS/2 Mouse support (basic two-button, no scrolling)
• Cirrus ISA/VLB/PCI SVGA chipset drivers (since it's what the system I wanted to test on had...), should work on CL-GD5420 on up.
• NE2000 network driver (there were no PC network drivers)
• Large hard disk support (though, there was no tool to adjust the partitions with, and I just didn't bother with making one, since ultimately 400MB is way more than enough even with all the extras)
• A few extra demo programs (ported NCSA Mosaic, TCC compiler, and smallchess (command is smolchess), and a fully custom gui fractal renderer that uses Motif toolkit)

There is also a custom DOS-based installer that can be used to install to hard disk on a real machine, from any storage medium you can get a DOS driver for. In my own testing I installed from CD-ROM, and over mTCP Netdrive. The installer is a normal .COM binary and needs to be stored next to the hard drive image.

Getting Started

The build kit is available here. I've tested it on Debian Forky with GCC 16, and I am pretty sure it will also work on Ubuntu 24.04 with GCC 14. It should be possible to run on windows this under Ubuntu WSL. Among (many) other things, it fetches and builds GCC 2.95 as a cross compiler, and goes from there. There are dependencies, the script should tell you what they are if you don't have them. If not, check the readme included. Claude wrote it so it's of course very wordy, but it is at least detailed. The wordiness is why we're here now, in Getting Started.

To use it, you will need to get the OSF1 sources from the following link:

You will want to go to each branch (OSC200, OSCX200) and download them as ZIP.

After this, extract the kit into a folder, and run sh at386-start.sh to build the disk image. My usual command line was:

XZIP=./OSF1-OSCX200.zip PROFILE=easy ISO=1 sh at386-start.sh ./OSF1-OSC200.zip ./workdir

This was with the OSF1-OSC200.zip and OSF1-OSCX200.zip located in the same folder the script is extracted to. It is designed to build the CD installer option. With that, any DOS boot diskette with a CD driver can open the CD drive and run the installer. XZIP is only required if you want an image with X.

There are an array of options for the build script but the most relevant options are the profiles:

1. No profile, or base: base system (terminal only), I'd say probably 99% historical, but also the most boring.
2. x: X and historical X programs and demos, maybe 90% historical, though probably never ran on i386
3. easy: X and X programs, plus the quality of life architectural features. Can be overridden to terminal only by other arguments. Check readme for more.

If that command line is used, when the script is finished you will have disk1.img and osf1-the following in your workdir under the OSC200 folder:

Hardware requirements

In principle, the bare minimum requirements would be

• 386 SX or DX. I would suggest a real FPU. FPU shouldn't be required since there is an FPU emulator, though it was only available in binary form, your mileage may vary.
• 500MB or larger hard drive, probably has to be IDE Primary master.
• It may run on 8MB of RAM. I would recommend 16 if using X.
• A supported VGA chipset. I have only tested Cirrus but the others probably work.
  • Tseng ET3000
  • Tseng ET4000
  • Paradise PVGA1 / WDC WD90C00 (Possibly later PVGA chips or WD90Cxx chips might work)
  • Genova GVGA
  • Cirrus CL-GD542x/543x/544x (Should work on 5480 as well, and might work on Laguna, don't know)
  • QEMU VGA (is this a "hardware" spec?)
• If using X: either a serial mouse, or in the easy profile, a PS/2 mouse.
• For networking: Any ol NE2000 clone should be fine.

I would NOT recommend running this with X on the actual minimum requirements. It should work but it will be seriously slow. It is even a bit slow on the real hardware I have tested on, partly due to resolution and redrawing by CPU, and partly due to disk access, which I think is slow despite the CF card:

• Unisys CWD4001 mini 486 system
• 486DX2-66, no L2 cache
• 16MB RAM
• Cirrus CL-GD5424, VLB attached integrated, 1MB VRAM, can run the 1024x768 mode
• 1GB CF card adapted to IDE for hard drive
• Integrated NE2000 clone, IO port 0x300, IRQ 10

Conclusion

I wasn't sure if this project would actually work. I am quite impressed that Claude was able to power through the entire code base. It has been a while since I have last tested any of the AI models for anything nearly as difficult as this, and I was impressed how it did even with the low level parts (bootloader modifications for large disk, i8042 PS/2 mouse interfacing, VGA driver programming and debugging). I was quite shocked when I saw it booting the first time. I am probably done with this project, but it has given me new ideas...

Loading...