Messages in this thread |  | | Date | Sat, 03 Oct 2026 17:03:08 +0100 | | From | Bradley Morgan <> | | Subject | Re: [PATCH 0/2] ARM: rust: Enable Rust support for ARMv5TE |
| |
On 3 October 2026 10:38:25 BST, Karl Mehltretter <kmehltretter@gmail.com> wrote: >Rust is currently limited to ARMv7 on 32-bit ARM. This series extends >it to ARMv5TE. > >Linus Walleij asked for older ARM cores during review of the ARMv7 >support [1]. Christian Schrefl tried ARMv5 with the armv5te-none-eabi >target at the time [2]. He hit missing core atomics, missing >__aeabi_mem* symbols and a floating-point helper reached from >pr_info!() formatting. None of these show up anymore. The kernel crate >now implements atomics through the C helpers, and the >armv5te-unknown-linux-gnueabi target used here keeps the Linux EABI. > >The only remaining problem was a link error. Pre-ARMv6 __arch_xchg() >has no 2-byte case, which the Rust Atomic<i16> helpers need. Patch 1 >adds it the way the other pre-ARMv6 atomics work, with interrupts >disabled. Patch 2 enables Rust for ARMv5TE. > >Bradley Morgan's two-byte cmpxchg() emulation series [3] does not >overlap with this. Its ARMv6 part was dropped in v2 and it never >covered xchg() or pre-ARMv6.
On purpose! It'll be dead code after one of ards changes
> >Tested on v7.3-rc1 with clang 22.1.8, rustc 1.99.0 and bindgen 0.73.2. > >- QEMU versatilepb (ARM926EJ-S). The Rust samples, all Rust KUnit > suites and 355 doctests pass. >- Microchip SAM9X75 Curiosity board (ARM926EJ-S), at91_dt_defconfig > without the ARMv4T AT91RM9200. The samples, all Rust KUnit suites and > 356 doctests pass. The Rust sample modules (rust_minimal, rust_print, > rust_misc_device, rust_driver_faux) load and unload cleanly. >- Build test with every Rust driver, abstraction and sample that can be > selected on ARMv5TE (rnull, binder, the two PHY drivers, cpufreq-dt, > the DRM panic QR code). PWM_TH1520 is the exception. It uses a native > u64 division, which does not link on 32-bit ARM. > >As a further test I ported the Atmel PWM driver to Rust and ran it next >to the C driver on the SAM9X75 board. The port is not part of this >series. The comparison found a prescaler bug in the C driver for >periods of 2^32 clock cycles or more [4]. > >I plan to get a Raspberry Pi 1 Model B+ and then add ARMv6K support as >a separate patch. > >The ARMv7 enablement went in through Russell's patch system as >"ARM: 9441/1". Unless someone prefers another route I would submit >this series there after review. > >[1] >https://lore.kernel.org/rust-for-linux/CACRpkdYF0sVB2-qgy=GzETSR3+2sagVQPGdunDQDJrn8KqJorA@mail.gmail.com/ >[2] >https://lore.kernel.org/rust-for-linux/b13d37bd-ec68-4713-94e5-e9ed4d6a6354@gmail.com/ >[3] >https://lore.kernel.org/lkml/20260922173354.14404-1-brads@mainlining.org/ >[4] >https://lore.kernel.org/linux-pwm/20261003083036.21584-1-kmehltretter@gmail.com/ > >Karl Mehltretter (2): > ARM: cmpxchg: support 2-byte xchg() on pre-ARMv6 CPUs > ARM: rust: Enable Rust support for ARMv5TE > > Documentation/rust/arch-support.rst | 2 +- > arch/arm/Kconfig | 2 +- > arch/arm/Makefile | 4 ++++ > arch/arm/include/asm/cmpxchg.h | 14 +++++++++++--- > 4 files changed, 17 insertions(+), 5 deletions(-) > > >base-commit: cee9395acd8043be0644b25c34bfa86623f2b935 >
--- Thanks! https://lore.kernel.org/all/EE579805-42F2-4C58-B752-F28779EEB717@grrlz.net/
|  |