Re: Some observations regarding the ongoing x86-32 CPU configuration cleanups

[Peter Zijlstra]

On Mon, Nov 09, 2026 at 10:34:54AM +0000, Luís Mendes wrote:
> Hi everyone,
> 
> First of all i would like to say that i both admire and appreciate all
> your hard work with this lovely OS. I do understand that maintaining
> some systems can cause additional burden on maintenance, but there are
> some aspects that i would like you to reconsider.
> 
> I am writing because I  believe that some aspects have not been taking
> into consideration regarding the removal of the 80486 32-bit support
> from the kernel, namely:
> - There is this active Linux distribution:https://tinycorelinnux.net/
> and the Damn Smal Linux DSL (https://damnsmalllinux.org).
> - There are also these CPUs: https://www.vortex86.com/products/Vortex86EX
> - There may be only a few vintage 486 systems still working, but there
> are also other new x86 CPUs systems, especially in FPGAs like the
> https://github.com/MiSTer-devel/ao486_MiSTer which is a 486SX core
> that operates at 90MHz and can use up to 256MB RAM memory. There are
> plans to update this CPU IP core to become a full 486DX core, by
> adding a FPU unit to it and would turn this into a usable Desktop
> system with tineycorelinux. There is a considerable retro-computing
> community that utilizes the AO486 core on the MiSTer FPGA platform.
> There is even a Voodoo 3dfx accelerator card:
> https://github.com/victor-fisyuk/voodoo-fpga-public.
> 
> Overall i just want to inform you that some new x86 32-bit systems are
> still being sold brand new and can be utilized with Linux. I would
> like that you consider this with regard to how much 80486 support
> should be removed from Linux.
> 
> Personally i do have plans to extend the AO486 core to have the FPU
> and would like to be able to run a modern Linux on it. It would have
> entertainment and educational applications. With tinycoreinux it can
> have a usable desktop.

I don't realistically see 486 support coming back -- the thing doesn't
even have TSC. Getting rid of all the horrible (and completely untested)
code for no-tsc alone was worth deleting 486 support for.

The other option is you actually implementing RDTSC, CPUID (and possibly
CMPXCHG8B) on your FPGA.  I'm sure that's actually a *LOT* simpler than
doing i487 support.

Just add whatever instructions are required to get the bare minimal
32bit build working.