Checked out #Linux kernel v3.7.10 to see if I can strip it down enough to run on the 386 with 4MB of RAM in the Toshiba T5200. My plan is to do something very similar to what I did in my blog post where I targeted a 486 (https://www.insentricity.com/a.cl/283), but this time use LOADLIN.EXE to load the kernel from DOS since I don't have a working floppy drive.
[#]RetroComputing #VintageComputing.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
No make tinyconfig on v3.7.10. That's going to make stripping it down a lot more work. #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Setup a Debian 7 Wheezy docker environment to build in. But already running into compile problems which seem to be caused by lack of 386 support:
In file included from include/linux/irq_work.h:4:0,
from arch/x86/kernel/irq_work.c:8:
include/linux/llist.h: In function 'llist_add':
include/linux/llist.h:160:11: error: 'boot_cpu_data' undeclared (first use in this function)
[#]Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Looks like v3.3.8 was the last version that could be built for 386. After that there's some change that prevents the kernel from compiling. #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Hey, an error message is progress, right? #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
I’m not sure that’s better. No error messages at all. #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Still can’t get my kernel to boot. Went and got the Slackware 1 kernel and used LOADLIN and that’s at least something! Maybe I should try to fix the floppy drive so I can do a traditional Linux install? #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
If I change the BIOS Memory config so that everything above 640k is Extended instead of some being Expanded then my custom kernel doesn't immediately reboot. Instead it just hangs with a cursor on the screen and doesn't print anything. Getting annoyed with the lack of a reset button on the T5200 too.
Would be nice if there was a way I could get qemu to act exactly like the T5200 so I don't have to keep testing on the real hardware.
[#]Linux #RetroComputing #VintageComputing
=> View attached media | View attached media
=> More informations about this toot | More toots from fozztexx@fozztexx.com
For anyone who wants to play along, I've put the docker environment and my current Linux config here: https://github.com/FozzTexx/old-wheezer
I'll try to keep it updated as I progress.
[#]Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Thanks to @nina_kali_nina I've switched to using PCem instead of qemu since it does a better job replicating the experience I get on the T5200. I've been going through old and older distros trying to find one that has a kernel that boots on the emulated 386DX. So far it seems that no distro with a 2.x kernel will work. Newest so far is Slackware 2.3 with a 1.2.8 kernel. Bumping the RAM up to 32MB doesn't get 2.x kernels working either. Very strange.
[#]Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Newest kernel I can get to boot so far is 2.0.18 from Debian 1.1 from here: https://grumbeer.dyndns.org/ftp/linux/dist/debian/1.1.11/buzz/disks-i386/SpecialKernels/
It will boot using the boot floppy, but if I extract the kernel from the floppy it will NOT boot with LOADLIN and instead the computer reboots.
Also not sure where to get the kernel config that was used for that v2.0.18 kernel. Or where to get the v2.0.18 linux kernel source either, git repo only goes back to v2.6.
[#]Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Found a v2.4.22 kernel built as a bzImage which boots on the emulated 386 here: https://smalllinux2.sourceforge.net/
That proves that it is possible to get a bzImage to boot, so far I've only had luck with older zImage kernels. Unfortunately there is no kernel config to be found anywhere for that distro.
#Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
It seems like my main problem is with the loaders. If I take a kernel which I've had success with booting and move it to a different loader, it fails to boot. If I try to take the 2.4.22 kernel from its floppy and load it with LOADLIN or SYSLINUX it won't boot. If I take my custom 2.4.22 kernel and stick it on the floppy that had 2.4.22 it won't boot.
Until I can find a reliable way of loading the kernel I'll never know if the kernel itself is bad.
[#]Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Found LINLD and I'm able to get my own custom v2.4.37 kernel to load on an emulated 386 with 4MB of RAM! I had to put the root filesystem on /dev/fd0 since there's not enough RAM for an initrd.
Not sure where the source for LINLD is, I found a copy of the .COM file here: https://distro.ibiblio.org/slitaz/boot/
#Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
If I bump up the emulated RAM to 8MB then I can also use LINLD to get gray386 to boot and give me a shell.
[#]Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Got gray386 to load in 4MB! I recompiled the kernel to move PHYS down to 0x100000, extracted the initramfs and put it on an ext2 formatted floppy. Loaded with LINLD and root=/dev/fd0. Can't try this on the Toshiba yet since the Toshiba doesn't have a working floppy drive. #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Got it to boot on the Toshiba! It’s seriously RAM starved and gray386 doesn’t even have swapon. It looks like I’m probably going to have to roll my own distro just for this computer. Many many tools missing and I need to add modules to the kernel. #Linux #RetroComputing #VintageComputing
=> View attached media | View attached media
=> More informations about this toot | More toots from fozztexx@fozztexx.com
What?!? Where's the i386 option, buildroot? #Linux #RetroComputing #VintageComputing
=> More informations about this toot | More toots from fozztexx@fozztexx.com
Got a more complete busybox running on T5200! Started with buildroot commit before 386 support was completely removed, hunted down by @jason_duerstock. Fought getting it to compile because of cross-compile issues and slightly incomplete 386 support. Should give me a base to build on top of. Not sure yet if I want to go full "roll your own distro" w/package manager or manually compile things I want to add. Doubt anyone else wants to run Linux on a 386.
#Linux #RetroComputing #VintageComputing
=> View attached media | View attached media | View attached media
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx @nina_kali_nina PCem has basically discontinued (or at least completely stalled.) You might want to check out the 86Box fork that continues where it left off.
=> More informations about this toot | More toots from nazokiyoubinbou@urusai.social
@nazokiyoubinbou @nina_kali_nina Why when it works for what I need it for?
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx https://archive.org/details/git-history-of-linux
=> More informations about this toot | More toots from eloy@hsnl.social
@fozztexx the kernel source is here: https://www.kernel.org/pub/linux/kernel/v2.0/linux-2.0.18.tar.bz2
=> More informations about this toot | More toots from marcusb@mastodon.sdf.org
@fozztexx for the source, going back that far would probably involve an SVN or CVS repo if Debian even keeps those online (Apache still has an SVN repo out there so there is hope). Also I thought I found a repo from someone else attempting to run a 3.x kernel on a 386. I did not bookmark that unfortunately; however trying to retrace my steps took me to Elks, which is a Linux for ia-16: https://github.com/ghaerr/elks
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@fozztexx ah: https://github.com/marmolak/gray386linux
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes I wasn't able to get that to boot either. If I can get v3.3.8 to work I might come back and see if I can apply those v3.7.10 patches to my tree.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx what apps do you want to wind up running, assuming the goal is something other than "linux system on 386 go brrrr"
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes Right now the main thing I want is Python 3 and support for an ISA MIDI card.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx oof on the Python 3. If you manage that one, that's at least a hackaday write-up.
Apparently someone did a Python 2.4 build for DOS but that may as well be an entirely different language.
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes Well if I can get Linux running on the 386 then all I should need to do is compile Python 3. Right? Right?
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx nods while slowly backing away
I'm wondering how many problems would arise from the lack of a cmpxchg instruction in that build process. Also wonder how much of Python would still be practically operable on hardware that old.
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@fozztexx
I'm guessing /proc doesn't have any config entries?
=> More informations about this toot | More toots from jamie@boothcomputing.social
@jamie nope
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx
:-(
=> More informations about this toot | More toots from jamie@boothcomputing.social
@jamie @fozztexx have you checked:
=> More informations about this toot | More toots from pronoiac@mefi.social
@pronoiac @jamie There's no kernel config anywhere. Not on the small linux repo(s), not in the disk images, not in the kernel.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx Of course the solution to this problem is to write another loader ...
=> More informations about this toot | More toots from mbbrutman@mastodon.sdf.org
@fozztexx Is there any correlation between zImage or bzImage and whether it works or fails?
=> More informations about this toot | More toots from jason_duerstock@oldbytes.space
@jason_duerstock LOADLIN doesn't seem to work with bzImage, even though it happily loads them. It always hangs or resets after loading.
I can't find a pattern yet for why SYSLINUX works with some kernels and not others. Esp. when some SYSLINUX disks work fine but if I change the kernel on the disk then it won't load the kernel anymore.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx gj. I think part of the problem is how the bios transitions to the boot device or memory in 386. It's not like 486 which is part of their routine to check EMS/XMS. 2.4 still use initrd, 2.6 was the introduction for initramfs. Each is loaded differently into memory and its sector size.
=> More informations about this toot | More toots from 7adi@mastodon.social
@7adi gray386 has the phys start set too high which is why it won't load on the 4M system. If I recompile it to start lower then I can get it to load, it just doesn't have enough RAM to decompress the initramfs.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx you could boot to a separate partition, though storage will get cramped unless you are running xtide bios or something similar.
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes It's XT-IDE. But I can't get this kernel to detect a hard drive at all..
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx huh. Is it detecting ATA controllers at all? 3.x would have still needed to support PCI ATA for optical devices at least.
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes It's detecting ATA controllers, but doesn’t seem to probe them for hard drives.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx weird, looks like it's trying to grab other ports/IRQs as disk controllers beyond primary/secondary. The legacy ATA driver may need some configuration (https://github.com/torvalds/linux/blob/master/drivers/ata/pata_legacy.c). I would imagine that in order for it to work on the real hardware (does PCem allow for setting up option ROMs?), the driver should be set to use BIOS calls so that XTIDE can do its thing.
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes I bounced back and forth with the ATA/ATAPI/MFM/RLL and Serial ATA and Parallel ATA drivers (SCSI) drivers and neither would detect the hard drive.
Linux should be able to talk to the drive directly via LBA and not go through the BIOS.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx what controller is the pata_legacy driver reporting? it should be a bit further back in the terminal from your most recent screenshot.
I still think forcing BIOS access might be the starting point just to get the dang thing working from disk.
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes It's not reporting anything else. That's everything ATA related in the messages.
Is there some command line flag to force it to use the BIOS?
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx trying to figure that out now by looking through the source.
based on the comment at line 1197 (https://github.com/torvalds/linux/blob/09a0fa92e5b45e99cf435b2fbf5ebcf889cf8780/drivers/ata/pata_legacy.c#L1197) it kind of sounds like legacy controllers are not probed by default?
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@fozztexx My previous suggestion might be a non-starter. It looks like the driver adds probe entries with a controller type of "unknown" during initialization and from what we see in your screen shot, it's covering primary/secondary plus all potential common legacy IDE port/IRQ combos.
I suppose there's always the possibility of altering the driver source directly but I would think there's a less-invasive way to get it to pick up IDE drives.
Out of curiosity, is it same for 2.4?
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@fozztexx also another dumb question: an emulated disk is present at one of those port/irq combos, right?
=> More informations about this toot | More toots from huronbikes@cyberplace.social
@huronbikes I don't know where else the controller would be. I'm trying to go back through all my posts to see what detected the hard drive and so far it seems like it was only Slackware 1. And that kernel was confused by the drive being impossibly large.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@huronbikes Does seem like PCem might be the problem. Went all the way back to Slack 1.0 and no HD detected. I guess it was detecting when I was using qemu.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx now that's patience!
=> More informations about this toot | More toots from nanochess@mastodon.social
@fozztexx oh cool! Are you able to extend the RAM on the machine, I seem to remember mine having 8 or 12MB. I really should dig it out of storage…
Edit, just checked and nope 4MB is the max, I misremembered.
=> More informations about this toot | More toots from carbontwelve@notacult.social
@carbontwelve It uses proprietary 40pin Toshiba SIMMs. No way I'll ever find any.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx That is insane. You are insane. Please continue.
=> More informations about this toot | More toots from cleverboi@mastodon.social
@fozztexx https://gitlab.com/buildroot.org/buildroot/-/commit/6cb4814c87cb1282b4a1d35e73743e8dd1b12ec5
=> More informations about this toot | More toots from jason_duerstock@oldbytes.space
@jason_duerstock wonder if I can reverse it
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx Worth a shot, or find the last version of buildroot that supported it.
=> More informations about this toot | More toots from jason_duerstock@oldbytes.space
@jason_duerstock reverting sort of worked. But the packages that I want to build don't seem to want to support 386. I might be able to tell buildroot to use older versions of those.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx openwrt is another option than buildroot but it's only network. Old routers are still used so that's best bet. I doubt you can run X11 though, maybe older xfree86.
=> More informations about this toot | More toots from 7adi@mastodon.social
@7adi I have zero interest in pointy-clicky on the 386, so I don't think not being to run X11 matters.
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx @jason_duerstock it isn't complete until it runs Rogue and a C compiler 😛
=> More informations about this toot | More toots from nanochess@mastodon.social
@nanochess @jason_duerstock I'm not sure how useful running the C compiler on it would be. Thing is so RAM starved it will probably swap itself to death just to compile Hello World. Although I guess now I need to find out…
=> More informations about this toot | More toots from fozztexx@fozztexx.com
@fozztexx @nanochess @jason_duerstock I remember compiling kernels on my 486SX 25. :)
=> More informations about this toot | More toots from root42@chaos.social
@fozztexx @nanochess For what it's worth, back in the day I could compile a kernel in 5MB of RAM on a 386sx/16. It took about eight hours or so, but it did work.
=> More informations about this toot | More toots from jason_duerstock@oldbytes.space
@jason_duerstock @nanochess I don't remember how much RAM was on the 386 that I originally used for Linux. I do remember the 8 hour kernel compile time. It was a hand-me-down 386 in 1993 from some other department on campus, so I doubt it had much.
=> More informations about this toot | More toots from fozztexx@fozztexx.com This content has been proxied by September (3851b).Proxy Information
text/gemini