Re: [PATCH v5 2/9] rust: pwm: Add core 'Device' and 'Chip' object wrappers

[Michal Wilczynski <m.wilczynski@samsung.com>]

On 6/27/25 14:12, Danilo Krummrich wrote:
> On Mon, Jun 23, 2025 at 08:08:50PM +0200, Michal Wilczynski wrote:
>> +    /// Gets the *typed* driver-specific data associated with this chip's embedded device.
>> +    pub fn drvdata<T: 'static>(&self) -> &T {
>> +        // SAFETY: `self.as_raw()` gives a valid pwm_chip pointer.
>> +        // `bindings::pwmchip_get_drvdata` is the C function to retrieve driver data.
>> +        let ptr = unsafe { bindings::pwmchip_get_drvdata(self.as_raw()) };
>> +
>> +        // SAFETY: The only way to create a chip is through Chip::new, which initializes
>> +        // this pointer.
>> +        unsafe { &*ptr.cast::<T>() }
>> +    }
>> +
>> +    /// Sets the *typed* driver-specific data associated with this chip's embedded device.
>> +    pub fn set_drvdata<T: 'static + ForeignOwnable>(&self, data: T) {
>> +        // SAFETY: `self.as_raw()` gives a valid pwm_chip pointer.
>> +        // `bindings::pwmchip_set_drvdata` is the C function to set driver data.
>> +        // `data.into_foreign()` provides a valid `*mut c_void`.
>> +        unsafe { bindings::pwmchip_set_drvdata(self.as_raw(), data.into_foreign().cast()) }
>> +    }
> 
> I think this is unsound, e.g. what happens if someone calls set_drvdata() twice?
> Then you leak the ForeignOwnable from the first call.
> 
> Anyways, this does not need to be public, you should just call
> bindings::pwmchip_set_drvdata() once in Self::new().
> 
> Please also see [1], where I introduce generic accessors for drvdata for Device.

Thanks, it would be a great idea to update the code after below patchset
is merged.

> 
> [1] https://lore.kernel.org/lkml/20250621195118.124245-3-dakr@kernel.org/
> 
>> +    /// Allocates and wraps a PWM chip using `bindings::pwmchip_alloc`.
>> +    ///
>> +    /// Returns an [`ARef<Chip>`] managing the chip's lifetime via refcounting
>> +    /// on its embedded `struct device`.
>> +    pub fn new<T: 'static + ForeignOwnable>(
>> +        parent_dev: &device::Device,
>> +        npwm: u32,
>> +        sizeof_priv: usize,
>> +	drvdata: T,
>> +    ) -> Result<ARef<Self>> {
>> +        // SAFETY: `parent_device_for_dev_field.as_raw()` is valid.
>> +        // `bindings::pwmchip_alloc` returns a valid `*mut bindings::pwm_chip` (refcount 1)
>> +        // or an ERR_PTR.
>> +        let c_chip_ptr_raw =
>> +            unsafe { bindings::pwmchip_alloc(parent_dev.as_raw(), npwm, sizeof_priv) };
>> +
>> +        let c_chip_ptr: *mut bindings::pwm_chip = error::from_err_ptr(c_chip_ptr_raw)?;
>> +
>> +        // Cast the `*mut bindings::pwm_chip` to `*mut Chip`. This is valid because
>> +        // `Chip` is `repr(transparent)` over `Opaque<bindings::pwm_chip>`, and
>> +        // `Opaque<T>` is `repr(transparent)` over `T`.
>> +        let chip_ptr_as_self = c_chip_ptr.cast::<Self>();
>> +
>> +	// SAFETY: The pointer is valid, so we can create a temporary ref to set data.
>> +        let chip_ref = unsafe { &*chip_ptr_as_self };
>> +        chip_ref.set_drvdata(drvdata);
>> +
>> +        // SAFETY: `chip_ptr_as_self` points to a valid `Chip` (layout-compatible with
>> +        // `bindings::pwm_chip`) whose embedded device has refcount 1.
>> +        // `ARef::from_raw` takes this pointer and manages it via `AlwaysRefCounted`.
>> +        Ok(unsafe { ARef::from_raw(NonNull::new_unchecked(chip_ptr_as_self)) })
>> +    }
>> +}
> 

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