Re: [PATCH 03/10] device property: Add fwnode_property_read_int_array()

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Wed, Apr 02, 2025 at 06:04:13PM +0200, Remo Senekowitsch wrote:
> On Thu Mar 27, 2025 at 9:41 AM CET, Andy Shevchenko wrote:
> > On Wed, Mar 26, 2025 at 06:13:42PM +0100, Remo Senekowitsch wrote:
> >> The rust bindings for reading device properties has a single
> >> implementation supporting differing sizes of integers. The fwnode C API
> >> already has a similar interface, but it is not exposed with the
> >> fwnode_property_ API. Add the fwnode_property_read_int_array() wrapper.

...

> >> +EXPORT_SYMBOL_GPL(fwnode_property_read_int_array);
> >
> > I'm not sure about this. We have a lot of assumptions in the code that the
> > arrays beneath are only represented by the selected number of integer types.
> > This opens a Pandora's box, e.g., reading in u24, which is not supported by
> > the upper layers..
> >
> >> +int fwnode_property_read_int_array(const struct fwnode_handle *fwnode, const char *propname,
> >> +				   unsigned int elem_size, void *val, size_t nval);
> 
> Here's an alternative approach using a macro to map each integer type explicitly
> to its corresponding read function. There are some additional changes that will
> be necessary to make the rest work, but this is the gist of it.

I don;'t know Rust to tell anything about this, but at least it feels much
better approach.

> +macro_rules! impl_property_for_int {
> +    ($($int:ty: $f:ident),* $(,)?) => {
> +        $(
> +            impl<const N: usize> Property for [$int; N] {
> +                fn read(fwnode: &FwNode, name: &CStr) -> Result<Self> {
> +                    let mut val: [MaybeUninit<$int>; N] = [const { MaybeUninit::uninit() }; N];
> +
> +                    // SAFETY: `name` is non-null and null-terminated. `fwnode.as_raw` is valid
> +                    // because `fwnode` is valid. `val.as_ptr` is valid because `val` is valid.
> +                    let ret = unsafe {
> +                        bindings::$f(
> +                            fwnode.as_raw(),
> +                            name.as_char_ptr(),
> +                            val.as_mut_ptr().cast(),
> +                            val.len(),
> +                        )
> +                    };
> +                    to_result(ret)?;
> +
> +                    // SAFETY: `val` is always initialized when
> +                    // fwnode_property_read_$t_array is successful.
> +                    Ok(val.map(|v| unsafe { v.assume_init() }))
> +                }
> +            }
> +            impl Property for $int {
> +                fn read(fwnode: &FwNode, name: &CStr) -> Result<Self> {
> +                    let val: [_; 1] = <[$int; 1] as Property>::read(fwnode, name)?;
> +                    Ok(val[0])
> +                }
> +            }
> +        )*
> +    };
> +}
> +impl_property_for_int! {
> +    u8: fwnode_property_read_u8_array,
> +    u16: fwnode_property_read_u16_array,
> +    u32: fwnode_property_read_u32_array,
> +    u64: fwnode_property_read_u64_array,

> +    i8: fwnode_property_read_u8_array,
> +    i16: fwnode_property_read_u16_array,
> +    i32: fwnode_property_read_u32_array,
> +    i64: fwnode_property_read_u64_array,

Dunno what these mean. Are they signed integers? Do you really need this kind
of types to be supported? Properties usually unsigned and I never saw (but it
doesn't mean they are not exist) the negative values there.  Does DT schema
even allow this? Does DT validator complain or not on the negative values
supplied in the DTS?

> +}

-- 
With Best Regards,
Andy Shevchenko