Add targets upstream for new ARM Cortex chips

[jamesmunns]

We should consider adding target JSONs and build rust-std compenents for the following targets before the 2018 era lands:

• ARM Cortex-R Target(s) based on thumbv7em (Cortex-R4, R5, R7, R8)
• ARM Cortex-M Target(s) based on thumbv8em (Cortex-M23 and Cortex-M33)
• ARM Cortex-R Target(s) based on thumbv8em (Cortex-R52)

This may also require updates to low level crates like cortex-m, maybe compiler_builtins, though we should probably open separate issues for that.

We likely will need physical chips to verify functionality before adding them upstream, which may be difficult for chips based on thumbv8em, as they are not currently retailing at the moment (as far as I know, the Nordic nRF91 is the only one shipping, and it is only in a limited sample release).

[Lakier15]

I have been working on a R5F for an year and a half with Rust without any problem.

I want to start working on other targets of the Cortex-R family so that I can PR and add official support for them.

thumbv7em, thumbv8em are not the only targets we would need pursuing. Remember that the ARMv7 instruction set differs a lot between their subset ( ARMv7-A, ARMv7-R, ARMv7E-M, ARMv7-M );

In order to support what is now commercially available we would need (for what regards the Cortex-R architecture):

• thumbv7r-none-eabi (Cortex-R4, R5, R7, R8)
• thumbv7r-none-eabihf (Cortex-R4F, R5F, R7F, R8F)
• thumbv7rbe-none-eabihf (Cortex-R5F Big Endian)
• thumbv8r-none-eabi (Cortex-R52)
• thumbv8r-none-eabihf (Cortex-R52F)

I have been working with the TMS570LC4357, which is probably the most unique R5F processor in the market, as it is the only one that has a R5F architecture and is big endian.

Cortex-R chips on the market

Cortex-R4

Texas Instruments: RM41Lx, RM42Lx, TMS570LS02x, TMS570LS03x, TMS570LS04x
Samsung: Artik 05x

Cortex-R4F

Texas Instruments: RM44Lx, RM46Lx, RM48Lx, TMS570LS07x, TMS570LS09x, TMS570LS11x, TMS570LS12x, TMS570LS21x, TMS570LS31x

Cortex-R5

Cypress: S6J31x

Cortex-R5F

Cypress: S6J32x, S6J33x, S6J34x, S6J35x

Cortex-R5F Big Endian

Texas Instruments: TMS570LC4357x, RM57Lx

---

I personally don't know about the R7(F), R8(F), R52(F). Maybe if anybody knows any better they can add to them to this list.

I have a R5F Big Endian (TMS570LC4357) and an R4 (Artik055s), so I could work to add support for those.

A word about the TMS570LC4357:
Because this is the only Big Endian processor I found pretty impossible to have a GCC prebuild toolchain to use as a linker.
My "hack" in this case is to compile everything as a library and link it in with the TI ARM CGT toolchain in order to produce a binary. This means that I cannot have a "pure" rust binary source code.
If the LLD linker can this job instead, that would be the perfect solution!

[paoloteti]

Here https://github.com/paoloteti/ti-hercules-bsp you can find a board support package + JSON for TI TMS570LS3xx (tested on TMS570LS3137) that is a BigEndian Cortex-R4F (in LockStep mode).

Linaro toolchain (armeb-none-eabi) it is perfect and you don't need to use TI ARM CGT.

I already sent patches on compiler_builtins to add some missing features (mainly on VFP) , so it works as it is now!

Only the F021 flash memory need the TI F021 SDK because the algorithm is proprietary and the overall logic it is totally different form other flash memories. Mainly to ensure a set of safety constraints at silicon level. Here you can find a Rust binding.

I don't have time at the moment to add/integrate my work on upstream, but I can test on real hardware on my spare time.

BTW do not compare Cortex-R with Cortex-M: it is a different beast.

[paoloteti]

Cortex R4F/R5F (BE, ARM mode) is now upstream (PR 50813)

[jamesmunns]

It seems like some of the thumbv8em chips are now shipping.

• ATSAML11D15A
• SAML11 Xplained Pro Evaluation Kit

[ithinuel]

I don't know if these devices are already available but I also heard about :

Nuvoton M2351

https://www.cnx-software.com/2017/03/21/numicro-m2351-trustzone-enabled-arm-cortex-m23-mcu-is-designed-for-fingerprint-applications/
http://armdevices.net/2017/03/20/worlds-first-arm-cortex-m23-in-nuvoton-m2351/
ARMmbed/mbed-os#7302

Arm Musca A1

https://developer.arm.com/products/system-design/development-boards/iot-test-chips-and-boards/musca-a-test-chip-board
zephyrproject-rtos/zephyr#7645

[japaric]

I was looking at the LLVM features for the ARMv8-M architecture and there are two ATM: v8m for baseline ARMv8-M (i.e. Cortex-M23) and v8m.main for mainline ARMv8-M (i.e. Cortex-M33). What I'm not sure about is how to call these new targets: thumbv8m-none-eabi for the M23, but then what would the M33 be called? thumbv8m.main-none-eabi? Furthermore the M33 has an optional FPU so we'll want two targets for that one: eabi and eabihf.

[paoloteti]

@japaric

What I'm not sure about is how to call these new targets: thumbv8m-none-eabi for the M23, but then what would the M33 be called? thumbv8m.main-none-eabi? Furthermore the M33 has an optional FPU so we'll want two targets for that one: eabi and eabihf.

Exactly, infact llc accept both thumbv8m.main-none-eabi and thumbv8m.base-none-eabi and with eabihf sets the VFP attribute (as minimum). (something like .eabi_attribute 28, 1 @ Tag_ABI_VFP_args).

[hannobraun]

NXP has a Cortex-M33 MCU in pre-production now: https://www.nxp.com/products/processors-and-microcontrollers/arm-based-processors-and-mcus/lpc-cortex-m-mcus/lpc5500-series/high-efficiency-arm-cortex-m33-based-microcontroller-family:LPC55S6x

[evq]

I opened a PR to add a target for baseline armv8-m rust-lang/rust#55041

[jamesmunns]

Nice! Thanks @evq!

[thejpster]

@evq is there any chance you might be able to share the test code you have for the SAM L11?