[Main] [News] [Compatibility] [Resources] [Downloads (annotated)]

The expression "under construction" doesn't even approximate what happens here. "Perpetually outdated" comes a little closer.

Last update: Jan 28 1999

MCA is now officially supported in the kernel.  I guess this is what we were all working for. 

A great new 2.0.35 patch and Slackware bootdisk. This includes Z.P. Gu's contributed patches (with things like SCSI fixes and some new drivers) and a small number of assorted fixes of my own.

I tend to get a lot of e-mail. If you're sending me information without asking questions, I may not send a reply. If you're asking a question, I'll usually reply, but maybe not that fast. It's also possible that you e-mail may get lost in the mess, so if you don't hear from me in a week or so, try again.

For those wondering why kernels in the 2.1.x might not be working for them, I should point out that the 2.0.x kernels I release here are a lot more current, at least in terms of MCA. I haven't upgraded to 2.1.x yet myself. I prefer working on stable kernels as much as possible. If you run into problems with 2.1.x and they aren't in 2.0.x, I suggest you don't run 2.1.x. Another option is to try Klaus Kudielka's patches, which have SCSI fixes for 2.1.29.

Speaking of 2.1.x, I get a lot of questions about what I'm doing with it. Currently, nothing. I don't have the time or energy right now to be what you'd call a bleeding edge kernel developer. To maintain a core chunk of the Linux kernel like the MCA patches, you really need to be able to follow the various kernel development mailing lists, try out most of the new 2.1.x kernels as Linus releases them, and integrate all the various patches people send you. I, on the other hand, barely have the spare time to play with my cats. Victi^H^H^H^Holunteers are welcome though.

September 15th, 1999

    (Amidst Hurricane Floyd)  Hello all!  This page has been updated quite a bit.  I will be putting a new face on the page, so be patient. There is now a Redhat 6.0 boot disk, contributed by Moyo Vajgel. Mark Van Bogart has also contributed an NCR 53c94 kernel. 

Jan 28, 1999

          Happy New Year!  Well, the site seems to be pretty active.  Please remeber that I am not the maintainer of the port, just the maintainer of the groups. Chris Beauregard (cpb@acm.org)is still the maintainer.   Mike Lang has informed me that he is now working on a "reliable" version of the merged SCSI driver.  As soon as he releases it, it will be put on this ftp site.  His site is at: http://www.uni-mainz.de/~langm000/linux.html.    Also, there are some hardware issues going on with my server. I am trying to speed things up a bit.

Nov 20th, 1998

          As you all have noticed, this site has been moved to http://www.dgmicro.com/mca If anyone has any suggestions or            comments, please feel free to e-mail me at dwright@dca.net

Aug 9th, 1998

Found some time to get a new 2.0.35 patchset and Slackware package together. This includes all of ZP Gu's fixes, a 3c523 tweak, and an attempted fix for the floppy problems people have (hint: use floppy=slow at the LILO prompt). Read step 4 of the Slackware-MCA document for details.

The MCA Linux discussion forum is going strong, and it's probably the best place to ask questions at the moment. Darren's spent a fair bit of time under the hood, so performance has gotten a lot better in the last little while.

I still haven't heard any news on the attempts at getting the NCR stuff working.

May 6th, 1998

New version of ZP Gu's 2.0.33 patch available for download (man, I hope I got that URL right).

Linux is a free UNIX implementation for, mostly, the Intel 80386 (and 486, and Pentium). Being a net.project, it's been ported to just about every Intel platform available. Contrary to what a lot of people on the net will tell you (including a few Linux FAQs), this does include machines with the Micro Channel bus.

Micro Channel is a workstation bus developed by IBM to deal with all the headaches of ISA. It never caught on with anyone besides IBM, but a lot of big corporations buy IBM machines, so there are surprisingly many MCA machines out there. It's used in the PS/2 series of machines (model 50 and up), as well as the RS/6000 and AS/400. NCR and Apricot also make MCA clones, and there may be others (I've heard something about Philips, Tandon, and something by a UK company called Research Machines). Oh, and the Tandy Model 5000 was also MCA.

MCA Linux has been around for a while now (I've seen datestamps back to 1994, or kernel version 0.99). The original project group put together the PS/2 ESDI driver, the Token-ring drivers, and the Debian installation disks that many people with ESDI are using. Unfortunately, the project seems to have died out. This page, for those interested, was started around December 1995 and has been (ahem) growing strong since.

When I finally put together a machine that I felt could run Linux, the first thing I did was try to find information on Linux under MCA. The only thing I could find was a bunch of kernel patches and the occasional snippet of Usenet alluding to it. Being the net.geek that I am, I set out to find out everything there was to know about MCA Linux, and maybe create some knowledge while I was at it. I think it worked, given some of the fan mail I get.

There are now boot disks for both Redhat 5.2 and 6.0.  For installation instructions, click here

Dave Weis has thrown together a RedHat 4.0 install disk for MCA. It's only been tested on glycerine and doesn't won't work on ESDI machines. But it's a start. There've been various efforts to build more recent bootdisks. We occasionally dump a new version onto FTP site.

Debian now has MCA support in the standard distribution. I haven't personally looked at it though. I've also received an e-mail on installing Debian onto an ESDI box.

And of course there's always my Slackware installation stuff. This is the most current and best-supported I have, so unless you're experienced with one of the others or Linux in general (or just gutsy), it's the one I recommend. On the other hand, I might just be biased towards something that I've put so much work into.

Since 3.3, Slackware has been shipping with a ibmmca.s bootdisk. I have no clue what this is, since once again I'm not able to find any documentation on it. It's apparently some of our older stuff though, so you're better off with the latest from this site.

I'm told that S.U.S.E. (a German distribution) will install nicely using my Slackware instructions and bootdisk. There's also been some information about this on the MCA Linux discussion page.

First, you need to make sure you have a 386 or better processor on your machine. Even a 386SX will do. Note that Pentiums may have some trouble (read on). Note that the model 50, which originally had a 286, will usually work with a processor upgrade.

There are a number of different upgrade boards available featuring all kinds of different processors. The one I have the most information about is the Evergreen Rev-2-486 board. You have to do some special magic to get the cache on the Cyrix/TI processor enabled. If you don't do this, performance is, uh, less than optimial.

You want to grab http://sunsite.unc.edu/pub/Linux/system/hardware/cyrix-1.00.tar.gz.

I'm told that the appropriate cache load command for this is cyrix -i1 -f -r -cd-.

This section is a general overview of the various MCA machines. You can also see information about specific configurations. I've also started compiling a list of specific devices that work and don't work.

There have been various problems with the floppy disk drivers in Linux over the years. They were fixed late in the 1.3.x kernel series, but there may still be problems on some machines. This has happened on just about every model but the 55SX.

More recently, I've been hearing about problems on the 57 series with the newer kernels. There have always been sporadic problems with the model 80.

There have been other reports that the 2.88MB floppy has a problem with 1.44MB disks. Makes booting a little difficult.

The most recent patchset and bootdisk supports a floppy=slow LILO option. This may fix the problem.

There have been about 3 major interations of the ESDI driver. The first version worked badly on just about all machines (but it worked). The second version, which I adopted into the 1.3.59 kernel series, worked fine on everything but the 70 and 80. The current version (starting at 1.3.71) works fine on everything but a few model 70 variants, and possibly some model 80s. In particular the P70 is supposed to have timing problems.

ESDI machines also have a number of problems due to the nature of ESDI, and this is discussed elsewhere.

The integrated ESDI on the Thinkpad series has a few major problems. It seems that some (if not all) of the larger drives (over 170MB) don't work with the current ESDI driver. For example, there's a 243MB that fails, while a 117MB on the same machine works fine. Why? Beats me.

Various versions of the SCSI driver have problem detecting integrated SCSI adapters as found on the 56, 57, 76, 77, and some 95s. Typically you can force detection using the ibmmcascsi=pun kernel command line option.

There are also a bunch of SCSI problems that aren't really machine specific.

Model 90 and 95's with Pentium processors have serious problems. They apparently hang when the kernel tests the HLT instruction, and if you get beyond that, they hang just for the hell of it. Specific models known to have this problem are Model 8595, 9595, and MCA PCServers.

Arnaldo H. Viegas de Lima has isolated the problem to the FPU test in the startup code. The problem is now to get it working without removing the FDIV bug test. If you know the state of your math processing on your Pentium, you can just replace the FPU test in linux/asm/bugs.h with constants. The problem hinges around the fact that the FPU test disables the timer interrupt. No idea why.

Strangely enough, the bug doesn't exist in NCR Pentiums, nor is it a problem with the dual-bus PCI/MCA IBM PC 750.

In my most recent patch sets, I've added a mca-pentium flag to disable the FPU test. I've recently received confirmation that id does indeed "fix" the problem. Yeah!

Extremely rare, there seems to be both ESDI and IDE versions of this machine. Some have been known to work.

The most detailed information I have is on the PS/2 Note N51SLC. This is a 386 with ESDI and a variety of standard ports (serial, parallel, etc). Linux can be installed using the MCA Slackware instructions, however all the ESDI warnings apply (i.e. DOS partition first, watch out for the geometry, that sort of thing). For more details, see the original e-mail.

Apparently IBM shipped a 386SLC-40 on a model 8540-060. This is pretty rare (it _might_ be some kind of upgrade board, but it has a IBM FRU), and the 2.88MB floppy has some problems reading rootdisks.

If you have a Model 55SX (and possibly 56, 57 or 65) with one of the SCSI adapters listed as standard in the table below, you can use a regular Linux SCSI kernel.

Everyone else appears to need a MCA kernel.

The main reason for this is the use of level triggered interrupts under the MCA architecture. Level-triggered interrupts, being the intelligent way to do things, are lacking in the ISA bus. The regular Linux kernel assumes that certain devices, such as the timer, are edge-triggered (in fact, the timer seems to be the only such device). The net result is that the first timer interrupt hangs your system.

The second reason is the hard disk device drivers. The typical PS/2 will either have a PS/2 ESDI or an IBM SCSI controller, with the associated drive(s). Neither of these are supported by the standard Linux distribution.

On top of this, we also have some MCA-specific ethernet drivers available which also aren't in the standard kernel.

There's quite a few MCA devices that work under MCA Linux. In general, if a MCA device is just a clone of a ISA device, it may be that an existing Linux driver for the ISA device will be able to use it out of the box. Check the Hardware Compatibility HOWTO for more details.

On the other hand, many MCA devices are just completely different. In this case, you need a driver. There aren't that many driver available for MCA devices yet, so exotic hardware may be useless under Linux.

The bottom line: if your favorite chunk of silicon isn't mentioned on this page, try Linux with it. Don't send me email asking whether or not it works. I only know about the stuff mentioned here, and I only have hands on experience with my own system.

Memory is memory is memory, right? Wrong. Usually something on a memory adapter is properly detected and used. However, there exist a few heinous concoctions such as the 80386 Enhanced Memory Adapter commonly found on the model 70 that require a special driver to use the extra memory. Of course that driver isn't available for Linux. Kiss your extra memory good bye.

Peter De Schrijver was working on a driver for it, and it's available as part of one of the patch sets at hagar, although the patch set is pretty horrible to look at (sorry Peter, it has to be said). It also seems to have a lot of hardcoded stuff in it. It was written from a disassembled version of the DOS driver. This is not recommended for people uncomfortable with low-level hacking.

The only memory card which I'm sure doesn't work is that mentioned above. The 80286/80386SX Memory Adapter found on the 55SX works fine, as do various Orchid cards, the Kingston KTM-3011, and probably others.

Here is the current list of MCA SCSI controllers tested with Linux.

Those listed as using a standard kernel are detected by the corresponding ISA driver in the kernel. You might still need a MCA kernel though, as discussed above. Other controllers might work, but I don't know about them. In particular, different models of the known adapters have a good chance.

The Future Domain MCS-600/700 now has a separate driver.

The IBM SCSI driver mentioned here is Martin Kolinek's contribution. So far, it's confirmed to work on Models 70, 76, 77, 80, 85, 90, 95, 57, 65SX and 56SLC. This driver is used both for controllers on the motherboard and for adapter cards. This used to be distributed as a separate package, but has now (since 1.3.59) been incorporated into the mainstream MCA Linux patchsets.

There are at least five different IBM SCSI adapters which are supported by the driver. Given that it was written with the IBM technical reference in hand, it should support anything that uses the IBM SCSI subsystem (and is recognized by the driver). I've recently received information that the IBM Fast/Wide SCSI controller is also supported by the driver.

There is another IBM SCSI adapter found on many model 76 and 77's that is actually an OEM'ed Future Domain controller. It requires the kernel command line parameters fdomain=0x140,5,ibm, where 0x140 and 5 are the IO address and IRQs from the reference diskette. I've received confirmation for multiple sources that this does the trick.

While the IBM SCSI driver has only been around for the last six months, the ESDI driver has a much longer history. It was originally written around 1994 along with the Linux Token ring driver. How it's managed to hang around this long without making it into the standard distribution, I know not. Possibly because people still think it's alpha code.

There's a few potential sources of confusion you might encounter when you deal with ESDI. The ESDI driver originally used major number 31. Unfortunately, 31 was never officially allocated to the ESDI driver, and by the time someone thought to request a major number, 36 was the next number in the list. My patches use 36, while anything else uses 31. As major 36 has propogated to virtually all the current patch sets, this should not be an issue for new users.

Another consequence of the major number business has to do with LILO, and probably other standard Linux utilities. LILO doesn't know about either major numbers, so you either have to patch it, or use explicit device numbers (i.e. root=0x1f01 for /dev/eda1 in lilo.conf, if you use major number 31, or root=2401 for major number 36.

As of release 20, LILO now has support for ESDI. New distributions based on this release will cut down the installation difficulties.

Then we get to fdisk. ESDI systems are supposed to use /dev/ed* for device files. The standard fdisk only know about /dev/hd* and /dev/sd*, the typical IDE and SCSI devices. This is especially important when it comes to installing Slackware, since it detects useable paritions via fdisk -l. This ignores /dev/ed*, so Slackware will have some minor problems. You can get around this by mounting your disks by hand, or patching it. You can also softlink the /dev/hd* entries to point to the corresponding /dev/ed* entries.

Anything else? Some versions of the ESDI driver had problems with geometry detection. That should be fixed now. However, somewhere along the way, support for multiple ESDI drives was broken. This seems to manifest itself as a timeout error of some kind.

Special magic may be needed for Thinkpad users. See the MCA Slackware installation documentation for details.

If you're installing using my Slackware instructions, this is all explained there.

For the record, ESDI on most PS/2s is pretty slow. For this reason, I recommend either enough memory in order to minimize the use of swap, or a SCSI controller. You might not think slow disks are a big deal, but to Linux, fast disks are usually more important than CPU or memory. It's also possible that the awfull performance is due to a slow driver, since other operating systems get much better (like 4-5 times) the performance.

Over the years, there have been piles of different ESDI packages produced. At the moment though, the only ones I can recommend are those available here.

There's a small problem with the PS/2 ESDI and IBM SCSI drivers being used together. They both seem to require IRQ 14. Presumably, IBM had some idea that anyone using both ESDI and SCSI on one machine would be running an OS that support shared IRQs. Linux now supports shared IRQs, and the ESDI driver is easily modified to handle them. My latest 2.0.7 kernel may fix this problem (keep trying, folks, eventually we'll get it right).

You may be able to run a dual ESDI/SCSI system with another SCSI adapter, but I have no information on what happens. Apprently you'll have to put the boot loader on the ESDI drive. My Slackware bootdisk has mixed ESDI and non-IBM SCSI driver support, so take a shot at it.

Note that if you don't want to use ESDI on your machine, you'll have to remove either the drive or the controller so that you'll boot from the SCSI drive.

There exists an ST-506 MFM driver. I ran across it by accident, and I have no information about it beyond what's in the documentation and source. For the record, it dates back to 1993, around kernel version 0.83.

There are IDE controllers available for the PS/2. I'd expect that the usual Linux IDE drivers would work fine with them.

I have confirmation of precisely two Arco Electronics IDE controllers working. Some fiddling had to be done with the geometry in LILO, but that's typical of a lot of IDE drives. In theory, the driver should auto-detect the card. In practice, if specifying "ide0=IO,IO,IRQ" doesn't work there's not much I can do to help.

I also have reports of the IDE controllers on some upgrade boards working, with sufficient magic and elbow grease.

Note that some of the model 76/77's have built-in IDE controllers. I don't have any information about how to get one working. They may get detected right away with a IDE-enabled kernel (i.e. my latest slackware bootdisk).

One problem with the parallel ZIP drive was discovered:

This is on a model 70.

From another user,

Once you get Linux up and running, it's quite likely that you'll want to connect it to some other machine. Particularly if you have a slower machine, like a 55SX. I generally do my kernel hacking on a 486 and download the kernel over the network for testing. I won't even mention the advantages of NFS for those of us with limited disk space. And then there's X, if you have the memory.

If you want token-ring, the regular Linux token-ring stuff was developed along with the original MCA Linux work (on a Model 70, if I remember correctly). See the Token-ring HOWTO for more information. Keep in mind that Linux Token Ring support is notoriously unstable.

A reader reports that in order to get it working under Slackware, all that's necessary is to replace every instance of eth0 to tr0 in /etc/rc.d/rc.inet1 (the HOWTO is apparently not quite accurate about this).

I've got a 3Com 3c523 driver working. It's been in use since January '96, so there usually won't be any trouble. However, I've received a few reports of unexplained failures (i.e. they did everything I told them to do and it still didn't work), so there might be room for improvement.

Dave Weis has a SMC Ultra MCA driver. This is also in the main MCA patch set. There are a whole bunch of different SMC cards floating around, so there's some confusion about which exactly are supported by the driver. I'm personally not sure myself.

I have another report that the SMC driver dies a horrible screaming death (page fault) with some memory addresses. If this happens, play with the reference disk and change the memory address settings to see what happens.

There's a patch available for the 3c529 Etherlink III. I've included 3c529 support in my last couple of patches, and it seems to be working okay in the latest.

Wim Dumon has a working NE/2 MCA driver. It's included in the last two patch sets and reports are that it works fine for genuine NE/2's, but has trouble with clones (i.e. they aren't detected at all).

Tom Sightler is working on A IBM Ethernet Adapter/A driver. It's actually a modified SMC (WD 8003) adapter and should be available for testing if you mail him. There's also a second type of Adapter/A (based on the WD 8013) that has support in the experimental (2.1.x) kernels. I've put support for all of that into the latest patch set (in the SMC driver), but I it hasn't been tested.

There is now Intel EtherExpress/MC support available as a modification of the standard EtherExpress 16 driver.

The DE210 is supposed to work, although apparently some modifications are needed. I haven't quite figured out how the modified driver works, but if you want, mail me and I'll send you the depca.c (the person responsible is off the net for a little while).

I have a report of an Arcnet adapter working fine with the standard Arcnet drivers. Given that Arcnet is such a standardized hardware design, this probably means that any MCA Arcnet adapter will work fine.

If all else fails, you can use the serial or parallel ports. My 55SX can handle PPP at 38.4k with only a few errors. Your mileage may vary.

PLIP seems to fail on the model 80. If anyone would care to confirm this for other machines, I'd be interested. It might just be related to the model 80s (ahem) interesting architecture (see discussions of the ESDI problems for more details).

X works with standard VGA, XGA-1, and XGA-2 adapters. There are some instructions you can look at here. One thing to note about XGA cards is that it's (as far as I know) impossible to get the standard VGA X-Server to work with the card, even though it's supposed to support the VGA chipset. You're pretty much going to have to bite the bullet and go all the way to get the AGX server working if you want pretty graphics on your Linux box.

I have one report that on a model 80 tower, the only way to get the VGA server working was to explicitly set the video memory to 512k.

Apparently the 8514 X server will work nicely with the 8514/a adapter (once you do the usual obscure X configurations). Beware, however. One person claims to have trashed a 8514 display with some bad settings. The 85xx series of displays are fixed frequency and tend to react badly to experimentation.

I'm collecting XF86Config files for various PS/2 video adapters and IBM displays.

For the record, various Linux video probes will detect the XGA adapter as a S3 or MCGA adapter. This is, ahem, normal behaviour.

There was this ATI Ultra Pro/MC sitting around the office. I had a shot at getting it working, but no luck. The X server probe couldn't find the adapter. I've heard from a number of people how also haven't had any luck.

Some very late model 77's (9577) use the S3 chipset (the one I know about has a S3 928, but IBM is pretty inconsistent about which exact S3 chipset it uses in machines - it seems to be related to the number of chips they have lying about when the machine is built). This machine uses the generic S3 server.

There are plenty of sound cards available for the PS/2 (check the PS/2 FAQ). I have little information at the moment about sound cards under MCA Linux. Presumably, if they're compatible with some existing driver (SB being the most popular), it'll work if the driver can find the card.

Dave Weis is hard at work on an ACPA driver. I'm sure he'd appreciate help.

The SoundBlaster PRO/MCV works fine with the standard sound drivers. This would indicate that most SB-compatible cards will work as well.

4Front Technologies now supports the Audioware Sound card. Dave says it's a ESS1688 that he tested for them, and "it works great"

Other options for sound are the PC speaker or parallel ports. Check the relevant HOWTO's for more information on how this works. I have had some luck with the PC speaker patches, but the sound quality is, to say the least, pathetic.

Recent patches support the Reply Sound Blaster stuff, whatever it is (I don't have one of these, I'm afraid).

Experimental MCA support for NetBSD. Some of the listed features look pretty neat, and I may attempt to steal a few. Actually, I already have.

Keep in mind that most other MCA Linux pages tend to get outdated very quickly. Heck, even this one is pretty far gone if I leave it for a couple months.

The Micro Channel Enthusiasts page offer a wealth of general MCA info, including a large ADF library and some very obscure DIY hardware info. If you want to know something about MCA, this is the place to go.

asterix contains various 1.3.99+ patches. It's also a source for 2.1.x SCSI patches.

The original Linux-MCA HomePage exists, but doesn't seem to be actively maintained. There's also the homepage for the Linux project at kuleuven, but it only points you back here last I checked. However, fans of the kuleuven krew will be happy to know that the PS/2 Linux operation is still going strong, after a little reorganization.

Randall Fisher and the PAPERS people have built on a parallel processing system composed of 8 model 70s and some specialized hardware. The Supercomputing 1996 pictures look very cool.

There are two general PS/2 reference sites that people may want to know about. The IBM PC page contains the all important reference diskettes. There's probably also a whole bunch of other goodies, but I have better things to do with my bandwidth. There's another site with a load of assorted .adfs, if you happen to be looking for one. I'm not sure how complete it is, but it doesn list many rare non-IBM files.

If you're looking for used PS/2 hardware or any other information, comp.sys.ibm.ps2.hardware is the place to go. In many ways, it's a better place to ask MCA Linux related questions than the usual Linux newsgroups since it's lower volume and we all read it. This is also where the PS/2 FAQ is found. The FAQ is also kept at ftp.demon.co.uk/pub/doc/ibmpc

There's a linux-mca channel on the rutgers mailing list server. This is very low volume (a message every month or so). To join, send a message with subscribe linux-mca in the body to Majordomo@vger.rutgers.edu. Send help to find out more about the list server.

General information on the PS/2, including instructions on installing a non-IBM ESDI drive on the PS/2 and some data on standard IBM displays.

If you're interested in doing a driver for a 3Com adapter, check out their CardFacts offer.

I've been calling it MCA Linux since early 1996. I'm not changing it now.

If you've actually read this far, you've probably gathered the impression that there's a lot of speculation happening here. I've learnt a lot over the last while, and this page has progressed an incredible amount, but there's still gaps.

Much of what I've learned about MCA Linux comes from a few Usenet posts and email from people who have tried stuff out (thanks guys!). If you know something, I'd greatly appreciate hearing from you. Even if it only amounts to "Linux doesn't work on my particular PS/2."

Disclaimer: Even though it may sound like I know a lot about MCA or Linux, I'm still very much in the process of learning. As such, anything I say should be taken with a grain of salt. I do my best to make sure the information presented here is accurate, but I can't accept responsibility for errors. That being said, questions are always welcome, and I'll do my best to help you.

IBM has no responsibility for this page, other than producing the architecture it "supports".

The University of Northern Iowa has no idea what goes on here. Let's keep it that way.

Not only am I not associated with U of Northern Iowa, I've never even been to that part of the U.S.A. I'm actually in Toronto, Canada.

Logo by Raine M. Ekman.

This page has been accessed times since Wednesday, January 15th 1997, 19:30 EST

Copyright 1996, 1997 by Chris Beaureguard