New Linux Kernel Code Works On APIC "Decrapification", Suggests Dropping x86 32-bit

[phoronix]

Phoronix: New Linux Kernel Code Works On APIC "Decrapification", Suggests Dropping x86 32-bit

There's a lovely new Linux kernel patch series out that's big in working on a major clean-up of the x86 APIC code (or "decrapification" as it's called in the patches) and also bringing up for discussion the idea of killing off x86 32-bit support. It's unlikely the x86 32-bit support will be removed right now, which is "just museum pieces", but as an alternative would be making it SMP-only to at least remove the uni-processor code paths...

https://www.phoronix.com/news/Linux-Decrapify-x86-APIC

[chromer]

It's too soon for death of 32 bit...

[geearf]

It's too soon for death of 32 bit...

Can't it just be moved to maintenance mode with only security fixes. I doubt such old machines need the latest and greatest anymore.

[dlq84]

Can't it just be moved to maintenance mode with only security fixes. I doubt such old machines need the latest and greatest anymore.

Are you suggesting a branch just for that? Since LTS kernels seems to have 5 years of support these days. And I don't think 5 years from now is a good time to remove the support either.

The last 32 bit x86 CPU is from 2015 as far as I know.

[avis]

It's too soon for death of 32 bit...

They want to deprecate running on pure 32bit CPUs. I don't think there are more than a dozen people in the entire world using Linux 6.4.x on them.

We are talking about Pentium 4 Northwood's released in 2002 or Athlon Barton/Thorton's released in 2003. Net gen Pentiums and Athlons were already 64bit.

You cannot support old shit indefinitely or otherwise there will be people requiring Linux to fully support Intel 8086. Most modern Linux distros are x86-64 only and require at the very least 256MB of RAM just to boot in text mode. What's the point?

Unlike Michael however I don't think these changes will result in any performance improvements. On x86-64 systems the code in question is never used since it's behind #ifdef.

TBO, there's nothing to discuss here. If you have old junk and you feel nostalgic, use something from the bygone era. Modern stuff will run like molasses on it it it runs at all which is questionable.

Edit: I've totally forgotten about Intel Atoms. There are two Atoms, N270 and N280 from 2008 and 2009 respectively, that are 32bit only. Damn it, Intel. If Linus remembers about them, this patchset could be a no go for at least 5 more years.

[flower]

It's too soon for death of 32 bit...

Why does an old machine needs the newest kernel?

[geearf]

Are you suggesting a branch just for that? Since LTS kernels seems to have 5 years of support these days. And I don't think 5 years from now is a good time to remove the support either.

Yeah a branch or LTS, whichever makes sense, didn't Mesa do kind of the same with the old drivers they moved to a separate branch to not have it block future work?

The last 32 bit x86 CPU is from 2015 as far as I know.

Yeah the Atom I believe, how many of these do run latest Linux though? I'd assume most is like Android, stuck to the kernel it was released with.

[jeisom]

They want to deprecate running on pure 32bit CPUs. I don't think there are more than a dozen people in the entire world using Linux 6.4.x on them.

We are talking about Pentium 4 Northwood's released in 2002 or Athlon Barton/Thorton's released in 2003. Net gen Pentiums and Athlons were already 64bit.

You cannot support old shit indefinitely or otherwise there will be people requiring Linux to fully support Intel 8086. Most modern Linux distros are x86-64 only and require at the very least 256MB of RAM just to boot in text mode. What's the point?

Unlike Michael however I don't think these changes will result in any performance improvements. On x86-64 systems the code in question is never used since it's behind #ifdef.

TBO, there's nothing to discuss here. If you have old junk and you feel nostalgic, use something from the bygone era. Modern stuff will run like molasses on it it it runs at all which is questionable.

I have to agree. We are talking 20 year old processors and the kernel is open source. If someone needs a linux kernel for a system that old the old releases are still available. Performance wise with regard to the cleanup I doubt it would be anything noticable. Code would probably just be more readable. There is probably an argument that could be made to drop any arch that stopped being made 15+ or 20+ years ago as well, but those probably have a lower impact on current/future development.

[kpedersen]

It seems strange to damage support for 32-bit x86 in the kernel but keeping old 32-bit sparc and powerpc intact? Whilst fairly rare, there are still *many* more old x86 processors in the wild than 32-bit sparc and ppc.

Is this a case of too many x86 cooks spoiling the x86 broth?

[billyswong]

This reminds me of https://www.vortex86.com/

Okay, that's too niche. But does the mention of SMP means that future Linux will be incompatible with single core VM?