Linux kernel devs to drop support for 486 architecture

[Lupino]

Summary

German tech news outlet Heise online reports that Linux kernel devs contemplate dropping support for some very old CPU architectures, specificly the 486 processor architecture. According to developer Ingo Molnar, support for those architectures, which he dubbs "compatibility glue", causes "problems that people spend time to resolve, which time could be spent on other things". Removing the neccessary code for those architectures would amount to some 14.000 deletions to the kernel's source code.

 

Discussion question:

Are you aware of hardware that still uses those kind of processors, regardless on whether or not they are actually running an up-to-date Linux?

 

Sources

https://www.heise.de/news/Linux-Entwickler-erwaegen-Support-fuer-486er-und-andere-CPUs-einzustellen-10363813.html (German)

https://lore.kernel.org/lkml/20250425084216.3913608-1-mingo@kernel.org/ (original message in the Linux kernel mailing list)

[SkipRat]

One of the things I liked about a Linux, was that it supported older computers, extending the life time and slowing down the rate of e-waste.

I would prefer it the did of balancing act.  New version yeah, but still keep some distro's 486 friendly.

[BirdieTX]

I mean, the last 486 processor was released when I was born (30 years ago). The fact that an architecture that old has received any kind of support this long is astonishing at this point.

[RoseTech]

So much for Linux running on anything now

[Kisai]

52 minutes ago, SkipRat said:

One of the things I liked about a Linux, was that it supported older computers, extending the life time and slowing down the rate of e-waste.

I would prefer it the did of balancing act.  New version yeah, but still keep some distro's 486 friendly.

 

There's really two issues that run head first into each other:

a) Who is running this OS on a 386/486? Are they NASA? IBM? Some pacific island nation with only 600 people and it's running one very specific airport?

b) Maintaining legacy code without legacy hardware being widely available is often suicide for the software.

 

Like In the grand scheme of things, Apple, Microsoft, Linus (of Linux), BSD kernel developers, Intel , AMD, Nvidia, ARM, etc would meet up in a room and hammer out a mutual agreement to discontinue all hardware and software support for specific insecure/hard-to-maintain technology not before a certain date, to give room for each other to pull that code out of their kernels and test everything while the hardware still exists and is wide enough available. The biggest chopping block there is "all 32-bit, 16-bit, and 8-bit CPU instructions"

 

In reality, Apple just cuts the cord when it doesn't want to. Microsoft leaves the cord plugged in far too long, thus ensuring perpetual code rot, and Linux/BSD doesn't need to leave the cord plugged in at all if it would just not release binaries for 32-bit platforms. Anyone who wants 32-bit support has to compile the 32-bit version themselves, and ports/package maintainers of those 32-bit OS's will be hard to find shortly after the OS drops it.

 

Hell it's hard enough to get binary packages for BSD AT All, never mind compile the ports source code, and I'm sure the same is of Gentoo if you have a system that is "too old" to be supported by the current distribution. But Windows is just one nightmare after another with coderot at all layers, between stuff Microsoft discontinues for seemingly dumb reasons (eg DirectMusic, DirectX pre DX9 features in general) which leaves trying to make games built on those libraries needing hacky work-arounds. There is no 64-bit support for "DirectX8", nor should there be. Yet you can't just recompile an old thing and make it work on the current OS for this very reason. 

 

There will come a point where we will be running ReactOS or something like CrossOver in a VM on Windows JUST to run any 32-bit game, because Microsoft has made it impossible to "just recompile" an old program because their tools just don't do that. If you open a VS6 project, in VS2022, it will be broken as hell. And assuming you fix the project, where's all the libraries? NuGet is broken, it doesn't work. Cmake is broken, it doesn't work. It's not BSD/Linux stuff doesn't exist where it should be. Permissions don't exist where they need to be, paths don't exist where they are supposed to be. It's quite difficult to get a project to a state where you can even try to recompile it, and then you'll be hitting compile and hoping that you haven't fixed integer rollover conditions the game or program expects as a consequence from 32bit integers being expanded to 64-bits.

 

It's a pretty poor ask for Linux to keep supporting a 486, with the same kernel that runs core i7's and Ryzens. Sometimes you just want to remove kernel bloat, but sometimes you just need to address the elephant in the room, that continuing to support 30+ year old old CPU's holds back gains that could be made with newer processors.

[smcoakley]

Most distros don't offer 486 builds any more, even though the Linux kernel currently supports it. I imagine any distros that have a specific need to continue to support 486 might do so by sticking to the current kernel version indefinitely, or making a kernel fork that maintains 486 support if it means that much to them.

[porina]

Note the difference between Linux the kernel, and and what is generally called Linux the OS. As with all discontinuation of support, it doesn't mean all hardware suddenly stops working. It just means, you don't get any more updates. If you have a 486 embedded in something, that'll still run the same stuff as it ever did. As long as it isn't connected externally then security vulnerabilities wont matter.

 

I do wonder, how many 486s are still in use out there. Where I used to work there were basically embedded PCs in some test equipment. When you turn it on, you saw the POST screen, then DOS loads and finally the test equipment software loaded on top of that. Not even Linux! But that was around quarter century ago, and it wasn't new then either. I think I read somewhere Intel used to use 486 cores as general microcontrollers since they owned the IP anyway, and they're insignificant in size on newer processes. Don't know if they still do or if they moved to new cores by now.

[Thaldor]

3 hours ago, porina said:

I do wonder, how many 486s are still in use out there. Where I used to work there were basically embedded PCs in some test equipment. When you turn it on, you saw the POST screen, then DOS loads and finally the test equipment software loaded on top of that. Not even Linux! But that was around quarter century ago, and it wasn't new then either. I think I read somewhere Intel used to use 486 cores as general microcontrollers since they owned the IP anyway, and they're insignificant in size on newer processes. Don't know if they still do or if they moved to new cores by now.

That's what I was coming to say.

486 is so old that it's miracle something modern even supports it. Some old testing systems are probably the only places where someone would be running them anymore (outside of retro computing but I would separate that side as it's own completely as those also run as old software so they really ain't updating their systems) and with those testing systems, I don't think anyone is updating their OS's because I don't think their testing software could be ran in any newer OS. There's probably some (at this point) ancient military communication gear running 486's and being in some use, but those definedly haven't been updated since they were new and that was decades ago.

 

I would think that anyone still running 486 as modern platform is going to be so deep in the diving training pool of supporting the platform that they alone can support it and probably they alone are supporting it because probably they also need to support everything they are running on it. These could be some radio amateurs and other similar groups but as said, they are deep in their field and knowledge, they have to be to keep it all running and if they aren't, well, their stuff won't work.

 

Third group would be the third world but they probably are already drowning in Core2 and newer platforms that even for them trying to run 486 as computer would be too ancient.

[wanderingfool2]

On 4/28/2025 at 5:49 AM, SkipRat said:

One of the things I liked about a Linux, was that it supported older computers, extending the life time and slowing down the rate of e-waste.

I would prefer it the did of balancing act.  New version yeah, but still keep some distro's 486 friendly.

That is effectively the issue that they would try avoiding though.  Keeping some distro's still 486 friendly would require keeping in code to do so...which means not deleting it in newer versions (albeit less testing maybe) but overall it still means resources are wasted on keeping things around.

 

There comes a point where backwards hardware compatibility really should be toss...specifically I think when there is such a great technological advancement that there isn't any rational behind keeping it.  e.g. Pre modern internet days, those computers don't really have too much use and if you are running a system that needs it, you can always just run the outdated version of it anyways [like I don't see a need for adding new features if the system is already like 30 years old].

 

Overall, I don't mind them dropping support.  Even systems from ~2005 I understand if they start dropping support in favor of reduced code base etc.

[Timme]

I was going to say that the majority of those machines are mostly dead by now. But apparently, 486 was discontinued in 2007. Now, who and why would buy that thing in the 2000s?

[GarlicDeliverySystem]

27 minutes ago, Timme said:

But apparently, 486 was discontinued in 2007. Now, who and why would buy that thing in the 2000s?

Never underestimate the inertia behind large industries when there are safety or production critical embedded systems in play. Plus there were a few things back then that still ran on DOS, requiring replacement hardware. 

 

[Timme]

21 minutes ago, GarlicDeliverySystem said:

Never underestimate the inertia behind large industries when there are safety or production critical embedded systems in play. Plus there were a few things back then that still ran on DOS, requiring replacement hardware.

 

In what way does a derelict hardware contribute to the "safety or production critical embedded systems"? 

[wanderingfool2]

50 minutes ago, Timme said:

In what way does a derelict hardware contribute to the "safety or production critical embedded systems"? 

Some systems where you need reliable equipment that is known.

 

So for example, nuclear missile silos.  You don't want to just rewrite it using the modern stuff when it's already been proven to work correctly.

 

Another example could be some railway systems.  From like 20 years ago it was designed around specific technology, where the safety system was thoroughly vetted and confirmed to be working.  The issue is the more stuff you introduce the harder it is to verify it will be bug free or not have some weird issue with it.  So a lot of those systems stay on the dated technology...as in the literal sense you could be talking about like 20 - 100 mill in just "upgrading" it to the newer systems.  So it's just cheaper to keep on dealing with aging equipment as each year you can get away with it the more you "save".

[igormp]

1 hour ago, GarlicDeliverySystem said:

Never underestimate the inertia behind large industries when there are safety or production critical embedded systems in play. Plus there were a few things back then that still ran on DOS, requiring replacement hardware. 

 

Totally valid, but as a counterpoint in the sense of the actual OP: those systems do not get updated either, so dropping support for those within the kernel is irrelevant.

[Ha-Satan]

1 hour ago, Timme said:

In what way does a derelict hardware contribute to the "safety or production critical embedded systems"? 

 

This isn't a tech-specialist channel so it might not be 100% accurate but it's a good example of how ancient tech or code can persist in critical systems.

[GarlicDeliverySystem]

1 hour ago, Timme said:

In what way does a derelict hardware contribute to the "safety or production critical embedded systems"? 

If you bought machinery for a continuous production line in the mid/late 90s, chances are it had i486 processors on it or PCs that controlled it. If those broke down in the early 2000s you could think about getting a replacement unit for the controller that is a 1:1 replacement, or upgrade to a newer one with a later processor. Often meaning you had to replace the whole machine, since often these parts were not nearly as interchangeable as they are now. Which takes time, money and most importantly: your line is not producing until it is fixed, meaning no money coming in.

 

Small scale example: we had spectrometer that needed an old DOS machine, because newer PCs did not support the drivers and hardware needed to interface (there were special interface cards that used a slot no longer on any mobos for years at that point). When that one broke, vendor told us to buy new spectrometer because there was no way to upgrade to Windows XP or any "modern" platform (emulation wasn't an option and they never bothered to implement any upgrade programs or stuff like that). So we just bought another old DOS machine used and cobbled it together.

37 minutes ago, igormp said:

Totally valid, but as a counterpoint in the sense of the actual OP: those systems do not get updated either, so dropping support for those within the kernel is irrelevant.

For sure, I doubt any of these really need modern linux support. This was more to the point of "who buys decades old hardware".

 

[Timme]

15 minutes ago, Ha-Satan said:

 

This isn't a tech-specialist channel so it might not be 100% accurate but it's a good example of how ancient tech or code can persist in critical systems.

He talks about software, not hardware. Not just that, but the specific case of flight routing, scheduling, and processing enormous data in real-time are the tasks that feed the CPU power. And instead of Wendover, I'd recommend CNBC, Half as Interesting, Modern MBA, Asianometry, Professionally done content, that doesn't waste 15 minutes of your time before even mentioning the main subject of the video.

 

1 hour ago, wanderingfool2 said:

Another example could be some railway systems. From like 20 years ago it was designed around specific technology, where the safety system was thoroughly vetted and confirmed to be working. The issue is the more stuff you introduce the harder it is to verify it will be bug free or not have some weird issue with it. So a lot of those systems stay on the dated technology...as in the literal sense you could be talking about like 20 - 100 mill in just "upgrading" it to the newer systems. So it's just cheaper to keep on dealing with aging equipment as each year you can get away with it the more you "save".

I'd say this only speaks to unwillingness or inability to bring the system up to date. After all, the US railway system is notoriously underdeveloped.

 

1 hour ago, wanderingfool2 said:

Some systems where you need reliable equipment that is known.

 

So for example, nuclear missile silos. You don't want to just rewrite it using the modern stuff when it's already been proven to work correctly.

Again, it is a software matter. Besides, I'm sure these kinds of systems don't rely on commercial CPUs like Intel or AMD.

 

486 is a maximum 120MHz processor..

[Timme]

7 minutes ago, GarlicDeliverySystem said:

So we just bought another old DOS machine used and cobbled it togethe

Or you could buy a newer spectrometer that is more sensitive and sophisticated, with newer software to match? 
 

There is no advantage in using 10-15-year-old technology that is exponentially worth it in every single metric. "Scaling systems" doesn't exist, vulnerable to the Win.CIH inconvenience. It's not that I'm discarding extremely niche use cases.
After all, the CPUs of the 2000s could do everything the 486 did, but 1000 times better.

[wanderingfool2]

32 minutes ago, Timme said:

I'd say this only speaks to unwillingness or inability to bring the system up to date. After all, the US railway system is notoriously underdeveloped.

Not necessarily.  What so many people seem to forget is the general cost of doing business by keeping all systems "up to date", when it's not necessary.

 

A system that is run isolated shouldn't be updated to the modern technology just because they can.  The case I gave of a railway public transit system is a peak example.  To upgrade to the more current technology they would have to switch out a ton of systems, test it and also have downtime.  That isn't something you want to do yearly just to keep "modern".  The fact is the system works, and the upkeep costs of maintaining the older system is cheaper than upgrading; as again if they upgraded tomorrow vs in 5 years time they would have to still be replacing the same components...the only difference is now you have saddled the company with the upgrade costs twice.

 

38 minutes ago, Timme said:

Again, it is a software matter. Besides, I'm sure these kinds of systems don't rely on commercial CPUs like Intel or AMD.

 

486 is a maximum 120MHz processor..

Software and hardware can often run hand in hand.

I even met similar things like this at work.  We kept an isolated NT machine around because we had software that interfaced with it...sure we could have "upgraded" to the new software, but that new software was per user licensing (vs perpetual).  And sure it would offer more features, but ones we didn't require.  Using "older" hardware literally saved us about $10,000 per month in licensing costs [not to mention that it also would have meant reprogramming a bunch of stuff to interface with the newer version].

 

20 minutes ago, Timme said:

Or you could buy a newer spectrometer that is more sensitive and sophisticated, with newer software to match? 
 

There is no advantage in using 10-15-year-old technology that is exponentially worth it in every single metric. "Scaling systems" doesn't exist, vulnerable to the Win.CIH inconvenience. It's not that I'm discarding extremely niche use cases.
After all, the CPUs of the 2000s could do everything the 486 did, but 1000 times better.

That's not how businesses work, and businesses who fall for that kind of trap can often find themselves in financial hardships.  Why go with something that is more "sophisticated" when the current one literally does the job.  The only way you want to go with more sophistication is if it increases things like productivity etc.

 

You won't believe how reliable some of the old technology was.  Some of those older components are workhorses that never die (or the boards can be repaired and the chips are still working).  When you are talking about a company that wants like 99.9% uptime older technology is 100% a thing.

 

It's actually why NASA goes with older hardware as well.  Equipment that has been flight proven, or just chips that are less prone to errors...because going to the smaller architectures has lead to more CPU failures and more things that can go wrong.

 

A lot of old systems, like my NT system, are essentially cordoned off as well not facing the real world threats as well.

[Timme]

15 minutes ago, wanderingfool2 said:

That's not how businesses work, and businesses who fall for that kind of trap can often find themselves in financial hardships.

Investing in obsolete anything, as if this already doesn't sound counterintuitive, is not investing in things that open more opportunities for growth and efficiency in the future of the business. 
Just run a VM.

[wanderingfool2]

22 minutes ago, Timme said:

Investing in obsolete anything, as if this already doesn't sound counterintuitive, is not investing in things that open more opportunities for growth and efficiency in the future of the business. 
Just run a VM.

That is again being ignorant of situations though.  If something on an assembly line breaks, you don't go replacing the entire assembly line with new equipment.  It's a waste of money and how corporate spending can get out of control.

 

If you can spend $10 on old equipment that runs your factor line for example at the same efficiency as the $100,000 replacement then 100% you should be investing in the "obsolete" technology.  And saying "just run a VM" ignores that it's not as simple as just running a VM for a bunch of old stuff sometimes.

[Timme]

26 minutes ago, wanderingfool2 said:

you don't go replacing the entire assembly line with new equipment. It's a waste of money and how corporate spending can get out of control.

Ah, ok. So the equipment tends to wear out over time, which leads to an increase in maintenance, with spare parts being more and more scarce, and the time and money to obtain those parts will add up each time you perform the replacements. The other consideration is that HW is prone to die at around 20-30 years. And then you also have the capacitor ticking bomb. 
It's like owning a Fiat Uno. It works just fine as a car; maintenance seems to be easy and cheap. 2-3 times a year, you drop the "reasonable" repair costs. In two years, you've spent the same amount of money as if you were to sell off your old car and get a newer car. But now you're stuck with a dead-end car that only sucks more money. That's a trap of being a miser and business stagnation. At best.
With a newer system, you'll be able to do more stuff faster, open up expansion and growth of the business. Isn't that the goal of any business venture?

 

[GarlicDeliverySystem]

11 hours ago, Timme said:

Or you could buy a newer spectrometer that is more sensitive and sophisticated, with newer software to match? 
 

There is no advantage in using 10-15-year-old technology that is exponentially worth it in every single metric. "Scaling systems" doesn't exist, vulnerable to the Win.CIH inconvenience. It's not that I'm discarding extremely niche use cases.
After all, the CPUs of the 2000s could do everything the 486 did, but 1000 times better.

Sure... spend 150k EUR instead of 350 EUR, because the PC controlling the thing broke. That is not gonna happen in universities, and also not in most businesses. The thing worked fine and was perfectly adequate for its task, it was mostly used for teaching and routine measurements. No benefit would have been had by spending 150k on a new one, except that it would have come with a USB dongle now.

Postprocessing was done on different machines/hardware anyway, same for software. And the expected lifetime of instruments like this is in the 20 years range, with even a used market for old spare parts. 

 

When it comes to large or specialized hardware you have to understand that more often than not, the PC/processing power required is of no real concern. At least not in that your output etc. would scale with advances in that field. The physical machinery and output matters, the controller has a limited and defined job. Buying a whole new system because a controller died is like building a whole new supermarket because the branch manager is on sick leave.

[Timme]

1 hour ago, GarlicDeliverySystem said:

Sure... spend 150k EUR instead of 1445.40₪ (350 EUR), because the PC controlling the thing broke. That is not gonna happen in universities, and also not in most businesses. The thing worked fine and was perfectly adequate for its task, it was mostly used for teaching and routine measurements. No benefit would have been had by spending 150k on a new one, except that it would have come with a USB dongle now.

Postprocessing was done on different machines/hardware anyway, same for software. And the expected lifetime of instruments like this is in the 20 years range, with even a used market for old spare parts.

 

When it comes to large or specialized hardware you have to understand that more often than not, the PC/processing power required is of no real concern. At least not in that your output etc. would scale with advances in that field. The physical machinery and output matters, the controller has a limited and defined job. Buying a whole new system because a controller died is like building a whole new supermarket because the branch manager is on sick leave.

Ok, so what will you do when there won't be any 486s in a reliable condition? Like in 3-5 years? The last "new" model was almost 20 years ago. Will you quit the job by that point and start another business?
 

Nah, I'm just being a donkey. Got the use cases, still think they should've close the line in 2002, and redirected resources to develop and manufacture next gen thing.
 

[whispous]

Companies that fail to recognise that ALL IT and network infrastructure is an ongoing cost will find themselves left behind, or even seriously harmed when it all comes falling down.