I had to dig through our extensive archive – OSNews was founded in 1997, after all – to see if we reported on it at the time, but it turns out we didn’t: in 2006, Intel announced that in 2007, it would cease production of a range of old chips, including the 386 and 486. In Product Change Notification 106013-01, Intel proclaimed these chips dead.
Intel Corporation has been manufacturing its MCS 51, MCS 251 and MCS 96 Microcontroller Product Lines for over 25 years now, and the Intel 186 Processor Families, the Intel 386 Processor Families and the Intel 486 Processor Families for over 15 years now. Additionally, we have been manufacturing the i960 32 Bit RISC Processor Families for over 15 years. However, at this time, the forecasted volumes for these product lines are now too low to continue production of these products beyond the year 2007. Therefore, Intel will cease manufacturing silicon wafers for our 6″ based processes in 2007. Affected products include Intel’s MCS 51, MCS 251, MCS 96, 80X18X, 80X38X, 80X486DXX, the i960 Family of Microcomputers, in addition to the 82371SB, 82439TX and the 82439HX Chipsets. Intel has no choice but to issue a Product Discontinuance Notice (PDN) effective 3/30/06. Last time orders will be accepted till 3/30/07 with last time ship dates of 9/28/07.
↫ Intel Product Change Notification 106013-01
Considering the 386, 486, and i960 families of processors were only used for niche embedded at very low volumes at that point in time, it made sense to call it quits. We’re 18 years down the line now, and I don’t think anyone really mourns the end of production for these processors. Windows ended support for these chips well before the 2007 end of production date, with Windows 2000 being the last Windows version that would run on a 486, albeit only barely, since it officially required a Pentium processor.
Linux, though, continued to support the 486, but that, too, is now coming to an end. In a patch submitted to the LKML, Ingo Molnár, support for a variety of “complicated hardware emulation facilities” for x86-32 will be removed, effectively ending support for 486 and very early 586 processors, by increasing the minimum kernel support features to include TSC and CX8 (CMPXCHG8B) hardware support. Linus Torvalds has expressed interest in removing support for the 486 back in 2022, so this move doesn’t come as a huge surprise.
While most tech news outlets leave it at that, as I was reading this news, I immediately thought of the Vortex86 line of processors and what this would mean for Linux support for those processors. In case you’re unaware, the Vortex86 is a line of x86-32-compatible processors, originating at SiS, but now developed and produced by DMP Electronics in Taiwain. The last two variants were the Vortex86DX3, a dual-core chip running at 1Ghz, and the Vortex86EX2, a chip with two asymmetrical cores that can run two operating systems at once.
Their platform support documents for Windows and Linux are from 2021, so we can’t rely on those for more information. Digging through some of the documentation from ICOP, who sell industrial PCs based on the latest Vortex86DX3, I think support in modern kernels is very much hit and miss even before this news. All Vortex86 processors are supposedly i586 (with later variants being i686, even), but some of the earlier versions were compatible with the 486SX. On top of that, Linux 4.14 seems to be the last kernel that supports any of these chips out-of-the-box based on the documentation by DMP – but then, if you go back to ICOP, you’ll find news items about Linux 5.16 adding better support for Vortex86, so I’m definitely confused.
My uneducated guess is that the DX3 and EX2 will probably work even after these changes to the Linux kernel, but earlier models might have more problems. Even on the LKML I can find messages from the kind of people who know their stuff who don’t know all the ins and outs of these Vortex86 processors, and which instructions they actually support.
It won’t matter much for people relying on Vortex86 processors in industrial and commercial settings, though, since they tend to use custom stacks built by the vendor, so they’re going to be just fine. What’s more interesting is the what I assume is a small enthusiast market using Vortex86 processors who might want to run modern Linux kernels on them. I have a feeling these code removals might lead to some issues on especially the earlier models, meaning you’ll have to use older kernels.
I’ve always been fascinated by the Vortex86 line of processors, and on numerous occasions I’ve hovered over the buy button on some industrial PC using the VortexDX3 (or earlier) processor. Let me know if you’re interested in seeing what this chip can do, and if there’s enough interest, I can see if I can set a Ko-Fi goal to buy one these and mess around with Windows Embedded/CE, Linux, and god knows what else these things can be made to run.
That was exactly my first thought when I read about it: What about Vortex86? Like you I’ve been tempted more than once to pick one up for fun and to see if Pentium era OSes like BeOS and QNX with Photon will run on them. Having new-ish hardware that can run classic software without the risk of leaky 25 year old capacitors is a great thing.
I kind of already do this with some older thin clients running AMD Geode and early Kabini SoCs, but they are just new enough to not support anything before the Windows XP era. Haiku runs on them but I specifically want BeOS compatibility. Linux and the BSDs are great and all, but my nostalgia is for BeOS and Windows 98.
>”Vortex86EX2, a chip with two asymmetrical cores that can run two operating systems at once”
Wait — WHAT???
Must have one …
Unfortunately, this feature is not available in the most commercial products. I have one system, I can’t access or use the second core. I asked ICOP and they said they only support that in custom designs.
Time to write our own custom OS I guess. 🙂
I am sure some developer will be happy to get paid for maintenance of upstream code, so I don’t see any issue there.
That wouldn’t work. Linux doesn’t remove support for old hardware simply because it’s old, it removes it to simplify the code in the kernel.
In particular lack of cmpxchg8b means that certain lock-free algorithms couldn’t be implemented on 486. When these algorithms start being used in kernel your ability to support old hardware is lost.
It’s like some companies bravely promised that they would support Flash for some time after Google removed it’s support from Chrome.
Ha! Of course there were lots of cleanups held up by the need to support flash – and once Flash support was killed code was immediately changed in a way that made Flash support impossible.
zde,
I don’t argue that linux needs to continue supporting 486. But at the same time I don’t think it’s that big a deal to keep 486 support inside of conditional functions. The cmpxchg instruction did exist in it’s 32bit form and most algorithms that use cmpxchg8 can be implemented on top of cmpxchg (albeit a bit slower).
Realistically most 486 systems didn’t even support SMP, which makes things a lot easier because spin-lock algorithms can be replaced with cli/sti. or even no-ops.
IMHO supporting 486 isn’t particularly difficult (as long as the kernel still supports 32bit in general). Even though it’s not hard to do, I don’t know that anybody is motivated to do it. Few people even own a 486 – it’s a relic that most have discarded decades ago. They have historic value sure, but running modern software on them it doesn’t seem that important.
Arch Linux 32 says it supports old stuff.
https://archlinux32.org/architecture/
But will they do that when Linux will drop support for 486? Will they do patches for it? I don`t think so.
Well, they no longer support 386. That said, I do not think you can use modern GCC or Clang to create 386 binaries as they will emit unsupported instructions (like cmpxchg). So continuing to support 386 is a bigger problem.
For 486, they could choose continue to ship an older kernel while continuing to update the rest of the packages. That would be cool.
I guess kernel 6.14 will be the last mainline kernel to support 486. This kernel supports Docker. This kernel supports DRM / KMS for Xorg and Wayland. It even has NTSYNC! You can still build 486 binaries with GCC. It may be a while before anything you would want to run requires a newer kernel. As you point out though, they may need to backport any security patches themselves.
An alternative to the “last” kernel would be kernel 6.12 because, as detailed elsewhere, this kernel will continue to be supported by the Linux Foundation until at least 2035. For 6.12, they will not need to patch things themselves.
The support for WES09 (the last XP based embedded system) ended Jan 8, 2019 and Vortex chips works just fine with it.
Linus told us this was happening back in 2022
https://www.zdnet.com/article/linus-torvalds-bids-486-linux-adieu/
Yet, the latest Linux kernel still supports the 486 (for now).
This GitHub repo provides a Linux kernel targeting the 486 along with not only kernel 6.13 but also musl 1.25 and the latest BusyBox. This is a newer version of musl than found in Void Linux and Adelie! I am not sure about Alpine.
https://github.com/marmolak/gray486linux
I recently learned about the Linux Foundation’s Civil Infrastructure Platform (CIP). CIP extends support for Debian LTS kernels for at least 10 years. Debian Trixie will become Debian Stable this summer running version 6.12 of the kernel. As kernel 6.12 can be built for the 486, there will be a supported version of the kernel that can be built for the 486 until 2035 or later.
My understanding is the the Vortex86 chips are supported by current kernels as well. The reason that kernel 4.14 is given is that the Linux support list was last updated 2019/04/24 and, on that page, Tiny Core Linux 9 is the compatible distro with the most recent kernel (4.14). So, this is the latest kernel available through installing a Linux distro “out-of-the-box”. Debian 9 is also listed with kernel 4.9.0 as are many older distros.
I cannot remember when Debian change the definition of i386. It is currently Pentium Pro (Pentium is no longer supported) which I think came in Debian 12. I am not sure if it was Debian 11 or Debian 10 that dropped 486.
Tiny Core Linux 16 lists 486DX as its minimum CPU so I imagine it would still work (kernel 6.12 — same as Debian 13).
Latest Vortex86 models indeed should not be affected (at least DX3, EX2, probably DX, EX), they do have required instructions to boot (CX8, TSC, but only EX2 has SSE2 and SSE3 and some software is expecting that (e.g. sudo was failing for me on DX3 on Debian based SparkyLinux if I remember correctly, DX3 has SSE only). Thus, you need either to recompile yourself or look for alternatives. 32-bit linux is in general more limited now from my experience, even if you have supported CPU.
Yes. Digging into the Vortex86 patch in 5.15, gives some insight into this.
As far as I can tell, the Vortex EX and EX2 identify as i686 generation and have proper CPU detection. So, pretty much any modern Linux distro with a 32 bit edition should work on these chips I would think. At least the CPU should be supported. Not sure about video. There are many 32 bit Debian based distros.
The Vortex DX series identify as i585 generation. Most modern distros including those that are Arch and Debian derived will likely not work out-of-the-box with these chips as they require i686 (SSE2). The Linux kernel still does though if you want to compile your own. Or you can use a distro like Adelie Linux or Tiny Core that still support Pentium.
It is difficult to tell but the last Vortex86 that used the 486 instruction set may have been the SX. As discussed elsewhere, even these chips will be targeted by a supported kernel for the next 10 years or more. As long as GCC and Clang still support 486, I would not be surprised to see distros like Tiny Core continue to offer these kernels as options.
So no panic in Vortex land I would not think.
Interestingly, the Vortex86 patch was not needed to add any support. The chips worked fine with mainline Linux before. The patch is to prevent the kernel trying to use Spectre and Meltdown mitigations on Vortex chips as this broke things. The Vortex chips are not vulnerable to these attacks and so Linux just skips the mitigations for Vortex.
While it i486 code is not yet deleted, when the lInux kernel matinees delete it eventually, I’ll revert and restore most of it in T2 SDE Linux, just like we maintain ia64, reiserfs, the orinoco airport driver, and all the other RISCV architectures support and fixes. This is so little code, I don’t understand the maintenance burden argument for this thousands of lines of code for the very simple CPU, history and education and just overall vintage and retro fun and of course the Vortex86 customers (repost due 1st stuck in moderation :-/
Cool!
t2 SDE is so cool. hope the forks do not deviate too much so we can have happy vortex computing for many years in userland.
OSNews was founded in 2007?
Yes, as a BeOS news site, run by the long gone (Thankfully!) Eugenia.
Yeah, she had a habit of removing posts and accounts (including mine from back then) when someone said something she didn’t like – even if they weren’t being snarky and were entirely polite and factual.
My original account was removed by her. Good times.
2001 I think. https://web.archive.org/web/20011202230026/http://www.osnews.com/
I thought it was more like 1997 or so. I see Thom has updated the posting to reflect this as much.
Oh good because thinking about it, 2007 is way too recent.
Thom was not very old in 1997.
Yep, I was gonna say, my OSNews habit started no later than 2002. I distinctly remember spending a lot of time there in 2003 arguing for using Java instead of Mono in the desktop Linux stack.
I know I started browsing it when BeNews was still a thing, probably around early 2000 when I discovered BeOS. For a while I had a group of haters who would send me veiled threats via email if I didn’t stop posting comments on the site, I think this was in the late 2000s or so. I called them out in a comment and it abruptly stopped. Fun times!
I miss that gorgeous, tightly packed site design.
It has a certain charm.