Re: [PATCH 03/13] ARM: remove riscpc

[Linus Walleij]

On Fri, Sep 11, 2026 at 5:44 AM Ethan Nelson-Moore
<enelsonmoore@xxxxxxxxx> wrote:

> Because I think it's cool that the kernel still supports actual Acorn
> hardware, I patched gcc 16.2 to restore ARMv3m support:
> https://github.com/EthanNelson-Moore/gcc-arm-retro

For the record: I also think this is cool.

One question: is ARMv3m there in Clang/LLVM? I only
know that we have ARMv4t.

To the bigger question.

The community has a built in tension around this:

- One camp that are enthusiast like yourself. That kind of people
  are running the show inside arch/m68k on Amigas and Ataris.

- One camp who get their paycheck from present day companies
  and constantly on top of the latest technology. As can be
  seen Qualcomm just taped out a brand new ARMv7 SoC
  named Kuno, for example.

As you can see this leads to a bit of tension.

What I'm after here is not one camp is doing the right thing
and the other one is not, because that inevitably boils down
to the question of what is the meaning of life. Or "what is fun"
actually (Torvalds wrote a book named "just for fun") and
either case could be argued to be quite fun, or meaningful.

> I am working on a clocksource driver, which uses both timers (to
> hopefully fix the awful time drift), and removes the dependency on
> LEGACY_TIMER_TICK.
>
> I also have patches for several cleanups to Risc PC-related code.

As SoC maintainers I think we want to hear about a plan to
migrate RiscPC to device tree should it stay supported. And I
do not think it is that hard to achieve with the right focus by the
right people.

Actually testing it on actual hardware would be a further
requirement, so we're not doing dreamscapes here.

I want to heat from Russell on this.

The ARM kernel started as a passionate project by him, on that very
machine, and he should decide if we keep it going.

IIUC the Clang/LLVM compiler has essentially solved the compiler
issue.

> I am aware that this has no actual use case, but it was fun. No LLM
> was used for any part of the work.

My position is that using LLMs is good in many ways because
when used right by the right people it allows us to keep a wider
range of hardware supported.

Yours,
Linus Walleij