Re: [PATCH RFC 2/6] pwm: Add Rust driver for T-HEAD TH1520 SoC

["Michal Wilczynski/Kernel (PLT) /SRPOL/Engineer/Samsung Electronics" <m.wilczynski@samsung.com>]

W dniu 25.05.2025 o 14:03, Danilo Krummrich pisze:
> On Sat, May 24, 2025 at 11:14:56PM +0200, Michal Wilczynski wrote:
>> diff --git a/drivers/pwm/pwm_th1520.rs b/drivers/pwm/pwm_th1520.rs
>> new file mode 100644
>> index 0000000000000000000000000000000000000000..4665e293e8d0bdc1a62a4e295cdaf4d47b3dd134
>> --- /dev/null
>> +++ b/drivers/pwm/pwm_th1520.rs
>> @@ -0,0 +1,272 @@
>> +// SPDX-License-Identifier: GPL-2.0
>> +// Copyright (c) 2025 Samsung Electronics Co., Ltd.
>> +// Author: Michal Wilczynski <m.wilczynski@samsung.com>
>> +
>> +//! Rust T-HEAD TH1520 PWM driver
>> +use kernel::{c_st
>> +
>> +struct Th1520PwmChipData {
>> +    clk: Clk,
>> +    iomem: kernel::devres::Devres<IoMem<0>>,
> Why IoMem<0>? If you put the expected memory region size for this chip instead
> all your subsequent accesses can be iomem.write() / iomem.read() rather than the
> fallible try_{read,write}() variants.
The size of the memory region is not known at the compile time. Instead 
it's configured
via Device Tree. I'm not sure why it should work differently in Rust ?
>
>> +impl Th1520PwmChipData {
>> +    fn _config(
>> +        &self,
>> +        hwpwm: u32,
>> +        duty_ns: u64,
>> +        period_ns: u64,
>> +        target_polarity: pwm::Polarity,
>> +    ) -> Result<u32> {
>> +        let regs = self.iomem.try_access().ok_or_else(|| {
>> +            pr_err!("PWM-{}: Failed to access I/O memory in _config\n", hwpwm);
> Here and throughout the whole driver, please use the dev_*!() print macros.
> Drivers have no reason to use the pr_*!() macros.
>
>> +impl pwm::PwmOps for Th1520PwmChipData {
>> +    // This driver implements get_state
>> +    fn apply(
>> +        pwm_chip_ref: &mut pwm::Chip,
>> +        pwm_dev: &mut pwm::Device,
>> +        target_state: &pwm::State,
>> +    ) -> Result {
> I assume those callbacks can't race with pwmchip_remove() called from driver
> remove()? I.e. the callbacks are guaranteed to complete before pwmchip_remove()
> completes?

Yeah this is my understanding as well - this is something that the PWM 
core should
guarantee. Fairly recently there was a commit adding even more locking
1cc2e1faafb3 ("pwm: Add more locking")

>
> If so, this function signature can provide the parent device of the pwm::Chip as
> device::Device<device::Bound> reference.
>
> This would allow you to access iomem more efficiently.
>
> Instead of
>
> 	data.iomem.try_access()
>
> you could do
>
> 	data.iomem.access(parent) // [1]
>
> which does get you rid of the atomic check and the RCU read side critical
> section implied by try_access().
>
> Actually, I should have added this comment and explanation to the abstraction
> patch, but forgot about it. :)

Thanks ! Appreciate your review !
Michał

>
> [1] https://gitlab.freedesktop.org/drm/kernel/-/blob/drm-next/rust/kernel/devres.rs?ref_type=heads#L213
>