Re: [PATCH v12 1/3] rust: io: add resource abstraction

[Daniel Almeida <daniel.almeida@collabora.com>]

Hi Alice,

[…]


>> +
>> +impl Resource {
>> +    /// Creates a reference to a [`Resource`] from a valid pointer.
>> +    ///
>> +    /// # Safety
>> +    ///
>> +    /// The caller must ensure that for the duration of 'a, the pointer will
>> +    /// point at a valid `bindings::resource`.
>> +    ///
>> +    /// The caller must also ensure that the [`Resource`] is only accessed via the
>> +    /// returned reference for the duration of 'a.
>> +    pub(crate) const unsafe fn as_ref<'a>(ptr: *mut bindings::resource) -> &'a Self {
> 
> We usually call this method `from_raw`.

Hmm, pretty sure I have seen the exact opposite being asked. In fact, this was
discussed a bit at length a while ago. See the thread starting at [0] for context.

> 
>> +            )
>> +        };
>> +
>> +        Some(Region(NonNull::new(region)?))
>> +    }
>> +
>> +    /// Returns the size of the resource.
>> +    pub fn size(&self) -> ResourceSize {
>> +        let inner = self.0.get();
>> +        // SAFETY: safe as per the invariants of `Resource`.
>> +        unsafe { bindings::resource_size(inner) }
>> +    }
>> +
>> +    /// Returns the start address of the resource.
>> +    pub fn start(&self) -> u64 {
>> +        let inner = self.0.get();
>> +        // SAFETY: safe as per the invariants of `Resource`.
>> +        unsafe { *inner }.start
> 
> This needs to be
> 
> unsafe { (*inner).start }
> 
> What you wrote is not equivalent. (*inner) is a place expression, but
> when you write `unsafe { <place expr> }` that turns it into a value
> expression by reading the value. So this code copies the entire struct
> to the stack, and then you read `start` from the copy on the stack.

To be honest, I've seen a bug where the opposite was going on, a field was
being written on the value that was copied to the stack, leaving the original unchanged.

In any case, I thought this would be optimized away by the compiler. Also, IMHO
there should be a lint for this, because it does not make sense to copy a struct
just to read a field if we can just read the original location.

Thanks for catching that though, I will fix it on the next iteration :)

[…]

Thanks for the other comments,

— Daniel

[0] https://lore.kernel.org/rust-for-linux/B3564D77-9E26-4019-9B75-B0C53B26B64F@collabora.com/