Re: [PATCH V2 00/14] LoongArch: Add basic LoongArch32 support

[Huacai Chen]

Hi, Arnd,

On Tue, Nov 18, 2025 at 10:08 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Tue, Nov 18, 2025, at 12:27, Huacai Chen wrote:
> > LoongArch includes a reduced 32-bit version (LA32R), a standard 32-bit
> > version (LA32S) and a 64-bit version (LA64). LoongArch32 use FDT as its
> > boot protocol which is already supported in LoongArch64. LoongArch32's
> > ILP32 ABI use the same calling convention as LoongArch64.
> >
> > This patchset is adding basic LoongArch32 support in mainline kernel, it
> > is the successor of Jiaxun Yang's previous work (V1):
> > https://lore.kernel.org/loongarch/20250102-la32-uapi-v1-0-db32aa769b88@xxxxxxxxxxx/
> >
> > We can see a complete snapshot here:
> > https://github.com/chenhuacai/linux/tree/loongarch-next
>
> I looked through all the patches, and this seems completely fine
> implementation-wise. I replied with a few minor comments, but
> found no show-stoppers.
Thank you very much, I will try to solve those problems if possible.

>
> Reviewed-by: Arnd Bergmann <arnd@xxxxxxxx>
>
> I'm still skeptical about the usefulness overall and would warn you
> that you may regret merging this in a few years: 32-bit Linux is
> clearly in decline, and the amount of work in bringing up and
The motivation is explained by Jiaxun Yang, and maybe I should keep it
in the cover letter.

> maintaining another ABI (or two if you count LA32R/S separately)
> is substantial.
The ABI of LA32R/S is the same (I mean they are both ILP32), but
something is not compatible so we can not use a single kernel binary.

>
> In your cover letter, I'm missing information about running LA32
> code on LA64 hardware. Specifically, do you plan to add CONFIG_COMPAT
> support later, and do you plan to support LA32 kernels running
> on LA64-capable hardware?
Yes, Jiaxun is working on CONFIG_COMPAT.


Huacai

> I would suggest supporting COMPAT 32-bit userspace here, but not
> 32-bit kernels: based on the experience with x86, arm, powerpc
> and mips platforms that allow both, the compat mode usually
> results in a much better experience overall. Compat support should
> probably be a follow-up and not part of the initial submission
> though, so what you have here is fine.
>
>        Arnd