Linux kernel dumps 386 chip support Linus Torvalds has announced the Linux kernel no longer supports Intel's 80386 processors. Reg readers will doubtless recall that the 386 debuted way back in 1985 and made something of a splash when the chip found its way into PCs made by Compaq before an IBM PC bearing the processor reached the market. 386s screamed along, …
Well not with typists of the one fingered variety perhaps. But the screen couldn't keep up with even my own random 10 fingered crawl across the keys let alone any moderately competent typist. I remember having to talk to quite a number of secretaries who were being driven up the wall by how much time it would take to catch up with them.
I'd imagine the FFS of AmigaOS (unless you have a recent ROM/OS) has some limitations on disc size.
You can get an IDE to CF adaptor that will be cheaper than finding an old 2.5" IDE small enough. Not having virtual memory means it won't wear out too fast.
I don't remember the typing problem. And I replaced an actual typewriter with my first PC, a 25Mhz 386 SX with 2 whole MB of RAM. It's amazing how little you can get on a 40MB hard disk, when it's already half full of DOS and Windows 3.1.
My 8 notes of polyphony Soundblaster was amazing too. Wolfenstein 3D was my game of choice. The sad thing is I saw it somewhere a few months ago and played a bit, and I can still remember the way round all the levels (including where about half of the secret passages are).
"It's amazing how little you can get on a 40MB hard disk, when it's already half full of DOS and Windows 3.1."
HAHA. I remember those days. When Doom came out I had a choice of having Windows or Doom on so I got very good at installing Windows. I actually still have that 40MB Hard drive. I tried killing it once by driving over it in a 38 tonne lorry several times but it lived and worked perfectly. They dont build them like that any more.
"It's amazing how little you can get on a 40MB hard disk, when it's already half full of DOS and Windows 3.1."
40MB, We used to dream of havin' 40 MB...... etc. etc.
The first PC I had had a 20MB HD with a RLL controller to trick it into giving you 32MB, when I got it a friend asked me "what do you need that much space for, are you going to use it as a file server?".
That was in the days when the turbo C compiler and debugger lived quite comfortably on a 5-1/4" floppy and terabyte was something you heard of in SiFi films . I got rid of that PC when the install for MickeySoft 'C' (V6.0 ???) hit 40MB.
The first PC I had had a 20MB HD with a RLL controller to trick it into giving you 32MB
Lucky bastard. My first computer had a pair of 5 1/4 in floppy drives and a floppy labeled "DOS Boot Disk". No internal storage whatsoever. I had at least 10 copies of that boot disk to, just in case something happened to it.
I remember having that problem, and I didn't learn how to type properly until later. My typing speed at that point couldn't have been more that 20 or 25 wpm. That was only in Microsoft Works, however. In a proper word processor (my definition of proper at the time being Q&A.....did anyone else use that?) the computer kept up just fine with my mother, who was an excellent typist at the time, even if she did constantly complain about the differences between the computer keyboard and the one on her typewriter.
I assume the 486 still scrapes by? There are still products using 486-compatible processors on the market, I've used some recently. It would be somewhat annoying if Linux no longer supported those, but I haven't used a 386 CPU for many years (despite having a couple of old DOS 386 machines in the loft).
You run your existing hardware with an existing kernel.
If you just have software which requires a 386, that you could do that in an emulator.
Must I think of everything? ;)
Actually, its getting hard enough to find a distro with a downloadable non-pae installation kernel on a pentium m.
> Isn't 486 a 386+maths coprocessor?
Not necessarily: remember the 486SX?
Plus it made some significant changes to the instruction set that made multi-threading significantly easier. For example the atomic interlocked increment and decrement operations return the new value rather than just a zero or not-zero indication. Hence (from the article:
> which complexity has plagued us with extra work whenever we wanted to change SMP primitives
(And also why the InterlockedIncrement and InterlockedDecrement Win32APIs where different on WinNT vs. Win9x.)
The 486 also introduced some other instructions iirc, bound and bswap. (though for bulk swapping SSE solutions are faster)
Note that this might not only kill 386s, but also 3rd party x86 chips that are not entirely 486 compatible in the sense that matters for the kernel. Cyrix and UMC among others had 486 alikes.
486 is still supported, judging by comments in the changelog. There are some sort-of-486-like processors that are no longer supported, however:
"Note that the 386 is no longer supported, this includes AMD/Cyrix/Intel 386DX/DXL/SL/SLC/SX, Cyrix/TI 486DLC/DLC2, UMC 486SX-S and the NexGen Nx586."
"Note that the 386 is no longer supported, this includes AMD/Cyrix/Intel 386DX/DXL/SL/SLC/SX, Cyrix/TI 486DLC/DLC2, UMC 486SX-S and the NexGen Nx586."
Some chip manufacturers played fast and loose with the numbering so they might have said 486 and supported the instruction but internally they might have had a largely 386 architecture such as bus. Anyway I expect they're dumped for their rarity and disuse as much as anything else - who knows or cares if they work because nobody has the hardware to actually test them.
"The more things change, the more they stay the same.... my current work laptop sometimes cannot keep up with my typing speed (Windows 7 + Word)"
Just this morning, on my 2 year old Core i5 w/6GB RAM, Win 7 and was typing an email in Outlook, it was struggling to keep up and I ended up with a word inserted into the middle of anwordother.
Yet my 486 with Clarisworks (albeit I never ran email on it until it became a Pentium 233) never had this issue, was always snappy. At least until I put Win95 on it and had to DriveSpace the 170mb HDD to get any space.
MS has always acted as if the purpose of an operating system is to suck all the life out of the CPU, so you can't run any non-MS software, unless you invest in extreme hardware, like gaming machines. MS Word on my work 2GHz Core 2 Duo XP machine is slower than Wordstar on my old 12 MHz 286 DOS 3.3 from back in the last millenium.
Ah, but 386 motherboards had a cool feature modern boards don't: an clock circuit that kept accurate time. I think they only disappeared with the advent of the Pentium. My first employer use to look for them because he could pick them up and sell the clock circuits to a client who repurposed them.
Ah, the 386. I remember when the first Compaq 386 machines came out (I believe at 20MHz, not 33MHz, though Wikipedia tells me that slower 386s were available), and my fanboyism disliking the fact that it took the performance crown for desktops back from the 8MHz ARM2 Archimedes machines (probably as measured by dhrystone, although BASIC may have been involved). I'm sure my copy of Structured Computer Organization contains some comment about Intel "finally making a decent CPU", but unfortunately I'm a few thousand miles away and can't check - anyone got the red edition?
The 486 was a bit nicer as designs go, excluding the slight problem of getting everyone to optimize code in a way that was pessimal for Pentiums. I still wish IBM hadn't decided to use the chip from their printers and had gone with the 68000 series from the start (and if we were going to end up with thirty years of compatible machines foisted on us by Windows, we'd ended up with anything less crufty than x86), but at least it meant that near and far pointers weren't always obligatory...
"ended up with anything less crufty than x86"
As always, it's the marketting and "ease" of use that wins, not the better implementation or sophistication.
I used to be surprised at how well x86 coped with the exploding computing market, but (as you said) Windows lock-in pretty much guaranteed that no one really wanted to put money in to alternatives.
I remember doing that on my first home PC (massive steel-framed beige box that it was).
When it was finally retired by my dad's purchase of a 486, I took the thing apart and was slightly disappointed to find that the values shown on the 7-segment display were purely down to the state of the toggle button (being now much wiser, it clearly makes sense).
On the plus side, it meant that I was able to fit it to the 486 case and continue with the placebo "active turbo!" fun ;)
I think that this great ideal which would aim at being able to run Linux on ANY device prevails over these complexities.
I understand that Linux will not suddenly stop running on 386 boxes, but this is a step in the wrong direction, especially when you consider that with Linux , so much can be done with so little CPU.
>but this is a step in the wrong direction
I'm not sure how dumping legacy stuff that causes issues supporting the majority of users is a "step in the wrong direction". I would go further than Linus.. I would remove all of the archs and boards that haven't been touched for years and only allow them back in when someone steps up and offers to support them and can prove that they still work or are fixable.
>especially when you consider that with Linux , so much can be done with so little CPU.
If you turn off everything that makes using a recent kernel worth using..
Anyone who has a 386 processor and wants to install Linux on it now can easily download one of the zillions of distros that support kernel up to 3.3 (or whatever) and it will run just fine. And then they can do the same for all of those other processors that haven't been sold in years.
Just provide a document somewhere that says processor X is supported in kernel version [1, 3.3] or whatever with link to some distro locations that include that.
Then the people maintaining the kernel can get on with supporting new stuff (only)
>Anyone who has a 386 processor and wants to install Linux on it now can easily download one
>of the zillions of distros that support kernel
I don't think that is true.. as someone else said glibc doesn't work on 386 anymore.. IIRC even Debian's i386 arch has only worked on 486+ for at least the last stable release, maybe the one before that.
Erm, they aren't going back in time and deleting previous kernel versions. Just saying future kernel versions won't work.
So just use the ones which do work. No problem.
As for support, they will be supported in the same way as any other linux kernel - you support it in-house yourself, or buy support separately from a company or person with expertise.
There comes a point where you have to cut off support though, otherwise Linux would become bloated and inefficient. If relatively few people are using 386 class machines and supporting them is causing problems, why not drop support? After all, they'll probably not gain any advantage of using a newer kernel anyway as the machine won't be powerful enough to support a lot of new features.
Earlier kernels probably aren't going to stop working because a newer kernel has been released, and, it being open source, there is nothing stopping someone restoring 386 support to newer builds.
Being the head of one of the most successful and innovative open source movements in the world clearly has its freedoms. No paying customers means zero commercial obligation to keep supporting them, and one can just walk away without a care in the world. Bliss!
So there's a little lesson. You really can't count on free software being sustainable in the long run. If your product life cycle isn't going to evolve at the same-ish rate as the free software you're currently depending on then it's not necessarily a good idea to use it in the first place. Many people use Linux as if it will be there forever, and will always be the best thing in town. With enough people using it that will probably become a self fulfilling prophecy (certainly no bad thing!), but it's not written into any contract anywhere that that really will be the case.
In practise I suspect that it won't matter that much to embedded 386 users (I'm not one - PowerPC me). Embedded software is updated far less often than mainstream IT, and if it works now it'll keep working. I can understand Linus's point of view, and it will be quite a long time before it becomes a major headache for the users of these ancient chips.
On a side note, a Linux driver dev I know moans that Linux is a bit Mozilla-ish. Every major release changes everything all over again and drivers have to be re-written, etc. etc. It's a ball ache to keep up. That may indeed be the price of innovation and progress, but on the whole many people would prefer a more considered approach. Device driver devs matter a lot, and if they get too fed up and stop bothering? Well, that sort of thing can do real damage to an OS.
MS (and presumably Apple) long ago learnt the true value of drivers, API stability, etc. Like it or loathe it, the fact that Windows XP has been maintained all this time, still works, and so forth is fairly impressive and has meant that many paying customers have stuck with MS. Windows 7 looks like it will go the same way, and that's good news for many people out there.
By contrast vendors of paid-for software withdraw support for their OSes afor their own commercial advantage, and with the objective of forcing their still-locked-in customers onto another platform controlled by that vendor... ka-ching! Exhibit A - HP, which has notched up - RTE, MPE, Domain/OS, WebOS at least and is probably trying to snuff out OpenVMS as we speak.
We're talking about a part that ceased production 5 years ago (at which time it would have been made for anyone to be using it in new designs) - which embedded device vendor is still releasing major firmware updates for 5 year old products (yes I appreciate that device manufacturers' stocks of the part may take a year or two to be used up)?
Major flaw there. Try running Windows XP, which is still in some form of official support, on a domain with mandatory profiles and Internet Explorer 8 - the latest version to work on it and report back to me how well that works out for you. I know for a fact that it doesn't work as well as anyone would like, as IE creates files that have names that are too long, which messes up the locally cached profile so when a user logs in again it can't do anything with them and therefore doesn't let them log in. They also won't be fixing it, even though MS have admitted it is a problem, with the advice being... upgrade to a newer version of Windows.
Linux has a heck of a longer support cycle than Windows, period. But if you need to support an ancient chip 27 years after it was released, I'd be advising that you make your own custom OS, and not expect anyone else to.
That's stable and supported software too...
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
