Re: [PATCH v2] rust: types: add FOREIGN_ALIGN to ForeignOwnable

[Andreas Hindborg <a.hindborg@kernel.org>]

"Benno Lossin" <lossin@kernel.org> writes:

> On Tue Jun 10, 2025 at 4:15 PM CEST, Alice Ryhl wrote:
>> On Tue, Jun 10, 2025 at 4:10 PM Andreas Hindborg <a.hindborg@kernel.org> wrote:
>>>
>>> "Benno Lossin" <lossin@kernel.org> writes:
>>>
>>> > On Tue Jun 10, 2025 at 1:30 PM CEST, Andreas Hindborg wrote:
>>> >> diff --git a/rust/kernel/types.rs b/rust/kernel/types.rs
>>> >> index 22985b6f6982..0ccef6b5a20a 100644
>>> >> --- a/rust/kernel/types.rs
>>> >> +++ b/rust/kernel/types.rs
>>> >> @@ -21,15 +21,11 @@
>>> >>  ///
>>> >>  /// # Safety
>>> >>  ///
>>> >> -/// Implementers must ensure that [`into_foreign`] returns a pointer which meets the alignment
>>> >> -/// requirements of [`PointedTo`].
>>> >> -///
>>> >> -/// [`into_foreign`]: Self::into_foreign
>>> >> -/// [`PointedTo`]: Self::PointedTo
>>> >> +/// Implementers must ensure that [`Self::into_foreign`] returns pointers aligned to
>>> >> +/// [`Self::FOREIGN_ALIGN`].
>>> >>  pub unsafe trait ForeignOwnable: Sized {
>>> >> -    /// Type used when the value is foreign-owned. In practical terms only defines the alignment of
>>> >> -    /// the pointer.
>>> >> -    type PointedTo;
>>> >> +    /// The alignment of pointers returned by `into_foreign`.
>>> >> +    const FOREIGN_ALIGN: usize;
>>> >>
>>> >>      /// Type used to immutably borrow a value that is currently foreign-owned.
>>> >>      type Borrowed<'a>;
>>> >> @@ -39,18 +35,20 @@ pub unsafe trait ForeignOwnable: Sized {
>>> >>
>>> >>      /// Converts a Rust-owned object to a foreign-owned one.
>>> >>      ///
>>> >> +    /// The foreign representation is a pointer to void. Aside from the guarantees listed below,
>>> >
>>> > I feel like this reads better:
>>> >
>>> > s/guarantees/ones/
>>> >
>>> >> +    /// there are no other guarantees for this pointer. For example, it might be invalid, dangling
>>> >
>>> > We should also mention that it could be null. (or is that assumption
>>> > wrong?)
>>>
>>> It is probably not going to be null, but it is allowed to. I can add it.
>>>
>>> The list does not claim to be exhaustive, and a null pointer is just a
>>> special case of an invalid pointer.
>>
>> We probably should not allow null pointers. If we do, then
>> try_from_foreign does not make sense.
>
> That's a good point. Then let's add that as a safety requirement for the
> trait.

I disagree. It does not matter for the safety of the trait.

From the point of the user, the pointer is opaque and can be any value.
In fact, one could do a safe implementation where the returned value is
a key into some mapping structure. Probably not super fast, but the user
should not care.


Best regards,
Andreas Hindborg