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@flygoat.com/

We can see a complete snapshot here:
https://github.com/chenhuacai/linux/tree/loongarch-next

Cross-compile tool chain to build kernel:
https://github.com/sunhaiyong1978/Yongbao-Embedded/releases/download/0.13-2025.9.3/loongarch32-unknown-linux-gnu-gcc-0.13-20251105.x86_64.tar.xz

A CLFS-based Linux distro (Yongbao):
https://github.com/sunhaiyong1978/Yongbao-Embedded/releases/download/0.13-2025.9.3/loongarch32-Yongbao-Embedded-0.13-20251018-sysroot.tar.xz

Open-source tool chain which is under review (Also in maillist for upstream):
https://github.com/cloudspurs/binutils-gdb/tree/la32/
https://github.com/cloudspurs/gcc/tree/la32/
https://github.com/cloudspurs/glibc/tree/la32/

LA32 QEMU emulator (hardware is not as widely used as LA64):
https://github.com/loongson-community/qemu/tree/la32-user-exp/

Usage:
1. Extract Yongbao to a virtual disk (la32.img);
2. qemu-system-loongarch -kernel vmlinux -append "rw root=/dev/vda console=ttyS0,115200" \
   -m 256M -cpu max32 -machine virt,accel=tcg -drive file=la32.img,format=raw,if=virtio -display none -serial stdio

V1 -> V2:
1, Rebased on 6.18-rc6;
2, Port a full runnable kernel (except irqchip drivers);
3, Remove controversial flush_icache syscall for now.

Jiaxun Yang, Yawei Li and Huacai Chen(14):
 LoongArch: Add atomic operations for 32BIT/64BIT;
 LoongArch: Add adaptive CSR accessors for 32BIT/64BIT;
 LoongArch: Adjust common macro definitions for 32BIT/64BIT;
 LoongArch: Adjust boot & setup for 32BIT/64BIT;
 LoongArch: Adjust memory management for 32BIT/64BIT;
 LoongArch: Adjust process management for 32BIT/64BIT;
 LoongArch: Adjust time routines for 32BIT/64BIT;
 LoongArch: Adjust module loader for 32BIT/64BIT;
 LoongArch: Adjust system call for 32BIT/64BIT;
 LoongArch: Adjust user accessors for 32BIT/64BIT;
 LoongArch: Adjust misc routines for 32BIT/64BIT;
 LoongArch: Adjust VDSO/VSYSCALL for 32BIT/64BIT;
 LoongArch: Adjust default config files for 32BIT/64BIT;
 LoongArch: Adjust build infrastructure for 32BIT/64BIT.

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Signed-off-by: Yawei Li <liyawei@loongson.cn>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
---
 arch/loongarch/Kconfig                             |  115 +-
 arch/loongarch/Makefile                            |   29 +-
 arch/loongarch/boot/Makefile                       |    6 +
 arch/loongarch/configs/loongson32_defconfig        | 1110 ++++++++++++++++++++
 .../{loongson3_defconfig => loongson64_defconfig}  |    0
 arch/loongarch/include/asm/Kbuild                  |    1 +
 arch/loongarch/include/asm/addrspace.h             |   15 +-
 arch/loongarch/include/asm/asm.h                   |   77 +-
 arch/loongarch/include/asm/asmmacro.h              |  117 ++-
 arch/loongarch/include/asm/atomic-amo.h            |  206 ++++
 arch/loongarch/include/asm/atomic-llsc.h           |  100 ++
 arch/loongarch/include/asm/atomic.h                |  197 +---
 arch/loongarch/include/asm/checksum.h              |    4 +
 arch/loongarch/include/asm/cmpxchg.h               |   48 +-
 arch/loongarch/include/asm/cpu-features.h          |    3 -
 arch/loongarch/include/asm/elf.h                   |    1 +
 arch/loongarch/include/asm/inst.h                  |   12 +-
 arch/loongarch/include/asm/irq.h                   |    5 +
 arch/loongarch/include/asm/jump_label.h            |   12 +-
 arch/loongarch/include/asm/local.h                 |   37 +
 arch/loongarch/include/asm/loongarch.h             |  100 +-
 arch/loongarch/include/asm/module.h                |   11 +
 arch/loongarch/include/asm/page.h                  |    2 +-
 arch/loongarch/include/asm/percpu.h                |   44 +-
 arch/loongarch/include/asm/pgtable-bits.h          |   36 +-
 arch/loongarch/include/asm/pgtable.h               |   74 +-
 arch/loongarch/include/asm/stackframe.h            |   34 +-
 arch/loongarch/include/asm/string.h                |    2 +
 arch/loongarch/include/asm/timex.h                 |   29 +-
 arch/loongarch/include/asm/uaccess.h               |   63 +-
 arch/loongarch/include/asm/vdso/gettimeofday.h     |    4 +
 arch/loongarch/include/uapi/asm/Kbuild             |    1 +
 arch/loongarch/include/uapi/asm/unistd.h           |    6 +
 arch/loongarch/kernel/Makefile.syscalls            |    1 +
 arch/loongarch/kernel/cpu-probe.c                  |   23 +-
 arch/loongarch/kernel/efi-header.S                 |    4 +
 arch/loongarch/kernel/efi.c                        |    4 +-
 arch/loongarch/kernel/entry.S                      |   22 +-
 arch/loongarch/kernel/env.c                        |    9 +-
 arch/loongarch/kernel/fpu.S                        |  111 ++
 arch/loongarch/kernel/head.S                       |   39 +-
 arch/loongarch/kernel/module.c                     |   80 +-
 arch/loongarch/kernel/proc.c                       |   13 +-
 arch/loongarch/kernel/process.c                    |    6 +-
 arch/loongarch/kernel/ptrace.c                     |    5 +
 arch/loongarch/kernel/relocate.c                   |    9 +-
 arch/loongarch/kernel/switch.S                     |   25 +-
 arch/loongarch/kernel/syscall.c                    |   15 +-
 arch/loongarch/kernel/time.c                       |   31 +-
 arch/loongarch/kernel/traps.c                      |   15 +-
 arch/loongarch/kernel/unaligned.c                  |   30 +-
 arch/loongarch/kernel/vmlinux.lds.S                |    7 +-
 arch/loongarch/kvm/Kconfig                         |    2 +-
 arch/loongarch/kvm/vcpu.c                          |    5 +-
 arch/loongarch/lib/Makefile                        |    5 +-
 arch/loongarch/lib/clear_user.S                    |   22 +-
 arch/loongarch/lib/copy_user.S                     |   28 +-
 arch/loongarch/lib/dump_tlb.c                      |   14 +-
 arch/loongarch/lib/unaligned.S                     |   72 +-
 arch/loongarch/mm/init.c                           |    4 +-
 arch/loongarch/mm/page.S                           |  118 +--
 arch/loongarch/mm/tlb.c                            |   12 +-
 arch/loongarch/mm/tlbex.S                          |  322 ++++--
 arch/loongarch/pci/pci.c                           |    8 +-
 arch/loongarch/power/hibernate.c                   |    6 +-
 arch/loongarch/power/suspend.c                     |   24 +-
 arch/loongarch/power/suspend_asm.S                 |   72 +-
 arch/loongarch/vdso/Makefile                       |    7 +-
 arch/loongarch/vdso/vdso.lds.S                     |    4 +-
 arch/loongarch/vdso/vgetcpu.c                      |    8 +
 drivers/firmware/efi/libstub/Makefile              |    1 +
 drivers/firmware/efi/libstub/loongarch.c           |    8 +-
 drivers/pci/controller/Kconfig                     |    2 +-
 lib/crc/Kconfig                                    |    2 +-
 74 files changed, 2915 insertions(+), 781 deletions(-)
 create mode 100644 arch/loongarch/configs/loongson32_defconfig
 rename arch/loongarch/configs/{loongson3_defconfig => loongson64_defconfig} (100%)
 create mode 100644 arch/loongarch/include/asm/atomic-amo.h
 create mode 100644 arch/loongarch/include/asm/atomic-llsc.h
--
2.27.0

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@flygoat.com/
>
> 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.

Reviewed-by: Arnd Bergmann <arnd@arndb.de>

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
maintaining another ABI (or two if you count LA32R/S separately)
is substantial.

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?
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

Hi, Arnd,

On Tue, Nov 18, 2025 at 10:08 PM Arnd Bergmann <arnd@arndb.de> 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@flygoat.com/
> >
> > 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@arndb.de>
>
> 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