[PATCH v2 0/7] Rust Abstractions for PWM subsystem with TH1520 PWM driver

[Michal Wilczynski]

On 6/10/25 23:10, Drew Fustini wrote:
> On Tue, Jun 10, 2025 at 02:52:48PM +0200, Michal Wilczynski wrote:
>> This patch series introduces Rust support for the T-HEAD TH1520 PWM
>> controller and demonstrates its use for fan control on the Sipeed Lichee
>> Pi 4A board.
>>
>> The primary goal of this patch series is to introduce a basic set of
>> Rust abstractions for the Linux PWM subsystem. As a first user and
>> practical demonstration of these abstractions, the series also provides
>> a functional PWM driver for the T-HEAD TH1520 SoC. This allows control
>> of its PWM channels and ultimately enables temperature controlled fan
>> support for the Lichee Pi 4A board. This work aims to explore the use of
>> Rust for PWM drivers and lay a foundation for potential future
>> Rust based PWM drivers.
>>
>> The core of this series is a new rust/kernel/pwm.rs module that provides
>> abstractions for writing PWM chip provider drivers in Rust. This has
>> been significantly reworked from v1 based on extensive feedback. The key
>> features of the new abstraction layer include:
>>
>>  - Ownership and Lifetime Management: The pwm::Chip wrapper is managed
>>    by ARef, correctly tying its lifetime to its embedded struct device
>>    reference counter. Chip registration is handled by a pwm::Registration
>>    RAII guard, which guarantees that pwmchip_add is always paired with
>>    pwmchip_remove, preventing resource leaks.
>>
>>  - Modern and Safe API: The PwmOps trait is now based on the modern
>>    waveform API (round_waveform_tohw, write_waveform, etc.) as recommended
>>    by the subsystem maintainer. It is generic over a driver's
>>    hardware specific data structure, moving all unsafe serialization logic
>>    into the abstraction layer and allowing drivers to be written in 100%
>>    safe Rust.
>>
>>  - Ergonomics: The API provides safe, idiomatic wrappers for other PWM
>>    types (State, Args, Device, etc.) and uses standard kernel error
>>    handling patterns.
>>
>> The series is structured as follows:
>>  - Rust PWM Abstractions: The new safe abstraction layer.
>>  - TH1520 PWM Driver: A new Rust driver for the TH1520 SoC, built on
>>    top of the new abstractions.
>>  - Clock Fix: A necessary fix to the TH1520 clock driver to ensure bus
>>    clocks remain enabled.
>>  - Device Tree Bindings & Nodes: The remaining patches add the necessary
>>    DT bindings and nodes for the TH1520 PWM controller, a thermal
>>    sensor, and the PWM fan configuration for the Lichee Pi 4A board.
>>
>> Testing:
>> Tested on the TH1520 SoC. The fan works correctly. The duty/period
>> calculaties are correct. Fan starts slow when the chip is not hot and
>> gradually increases the speed when PVT reports higher temperatures.
>>
>> The patches are based on mainline, with some dependencies which are not
>> merged yet - platform Io support [1] and math wrapper [2].
>>
>> Reference repository with all the patches together can be found on
>> github [3].
> 
> I'm trying to build your rust-next-pwm-working-fan-for-sending-v4 branch
> but I get this error:
> 
> $ make W=1 LLVM=1 ARCH=riscv -j16
>   CALL    scripts/checksyscalls.sh
> .pylintrc: warning: ignored by one of the .gitignore files
>   UPD     include/generated/utsversion.h
>   CC      init/version-timestamp.o
>   KSYMS   .tmp_vmlinux0.kallsyms.S
>   AS      .tmp_vmlinux0.kallsyms.o
>   LD      .tmp_vmlinux1
> ld.lld: error: undefined symbol: rust_build_error
>     referenced by pwm_th1520.4789668fc0b4e501-cgu.0
>                   drivers/pwm/pwm_th1520.o:(<pwm_th1520::Th1520PwmDriverData as kernel::pwm::PwmOps>::get_state) in archive vmlinux.a
>     referenced by pwm_th1520.4789668fc0b4e501-cgu.0
>                   drivers/pwm/pwm_th1520.o:(<pwm_th1520::Th1520PwmDriverData as kernel::pwm::PwmOps>::write_waveform) in archive vmlinux.a
>     referenced by pwm_th1520.4789668fc0b4e501-cgu.0
>                   drivers/pwm/pwm_th1520.o:(<pwm_th1520::Th1520PwmDriverData as kernel::pwm::PwmOps>::write_waveform) in archive vmlinux.a
> make[2]: *** [scripts/Makefile.vmlinux:91: vmlinux] Error 1
> make[1]: *** [/home/pdp7/linux/Makefile:1241: vmlinux] Error 2
> make: *** [Makefile:248: __sub-make] Error 2

Hi,

Thanks for testing !
I can reproduce the issue with your config.

The root of the problem was a failing compile time assertion
(build_assert!) in the underlying Rust abstracions, I think IoMem since
get_state and write_waveform functions are impacted. My development
configuration was accidentally hiding this issue, but your configuration
correctly exposed it.

The kernel config option that is different on my setup is:
CONFIG_RUST_BUILD_ASSERT_ALLOW=y

Now I have to take a look at the IoMem abstractions, I think there is
a new revision [1]. Will apply it and check why exactly compile
assertions are triggering.

[1] - https://lore.kernel.org/all/20250603-topics-tyr-platform_iomem-v9-0-a27e04157e3e@collabora.com/

> 
> I've uploaded the config to:
> https://protect2.fireeye.com/v1/url?k=0cc1a518-535a9c14-0cc02e57-000babff3563-8df2dfc535042c2a&q=1&e=eaf92127-0b5a-4559-9796-3264da753ae3&u=https%3A%2F%2Fgist.github.com%2Fpdp7%2Fe2c34dd7e4349a54bd67b53254bd3a22
> 
> Thanks,
> Drew
> 

Best regards,
-- 
Michal Wilczynski <m.wilczynski at samsung.com>