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

[Daniel Palmer]

Hi Linus,

On Sat, 12 Sept 2026 at 03:15, Linus Walleij <linusw@xxxxxxxxxx> wrote:
> I would say using generic entry would be a prerequisite for
> doing m68k SMP so we don't introduce *more* entry
> oddities, but after that I'm game.

I have started this for one part of the m68k tree (68000), but it's
one of those things that is hard/frustrating to test on real hardware.
For the last few weeks I've had claude create a QEMU that can boot all
of the m68k supported machines to Linux, Netbsd and their original OS.
It already found this:
https://lore.kernel.org/linux-m68k/20260828094940.3882602-1-daniel@xxxxxxxx/

> As SoC maintainer I see that m68k has a lot of technical debt
> like drivers under arch/m68k and no proper hardware description
> (such as device tree)

I have m68k machines using devicetree... :)

> There are things like
> amiints.c, cia.c, ataints.c, q40ints.c etc that should be in
> drivers/irqchip. So many time.c that should be in
> drivers/clocksource/* and many other unmodern things.

Would be easier with devicetree. But is anyone crazy enough to go
through the pain of merging the devicetree support and then arguing
about devicetree bindings for each of those drivers only to be told
m68k should just be removed?

> Albeit I think actually going and fixing that can't be super-hard
> in the age of LLMs. So listing things like this likely will just make
> someone go and do it. :D

Once you have an emulator that the LLM can operate it can do the
change/build/test/debug cycle all on its own.
With a test harness it can even test the changes work on the real hardware.
But I don't think that's enough to convince people the changes are
worth reviewing/merging.

Cheers,

Daniel