Re: [PATCH v11 2/3] rust: add dma coherent allocator abstraction.

[Danilo Krummrich <dakr@kernel.org>]

On Mon, Jan 27, 2025 at 02:25:03PM +0100, Alice Ryhl wrote:
> On Mon, Jan 27, 2025 at 1:14 PM Danilo Krummrich <dakr@kernel.org> wrote:
> >
> > On Mon, Jan 27, 2025 at 11:43:39AM +0100, Alice Ryhl wrote:
> > > On Mon, Jan 27, 2025 at 11:37 AM Danilo Krummrich <dakr@kernel.org> wrote:
> > > >
> > > > On Fri, Jan 24, 2025 at 08:27:36AM +0100, Alice Ryhl wrote:
> > > > > On Thu, Jan 23, 2025 at 11:43 AM Abdiel Janulgue
> > > > > <abdiel.janulgue@gmail.com> wrote:
> > > > > > +    /// Reads data from the region starting from `offset` as a slice.
> > > > > > +    /// `offset` and `count` are in units of `T`, not the number of bytes.
> > > > > > +    ///
> > > > > > +    /// Due to the safety requirements of slice, the data returned should be regarded by the
> > > > > > +    /// caller as a snapshot of the region when this function is called, as the region could
> > > > > > +    /// be modified by the device at anytime. For ringbuffer type of r/w access or use-cases
> > > > > > +    /// where the pointer to the live data is needed, `start_ptr()` or `start_ptr_mut()`
> > > > > > +    /// could be used instead.
> > > > > > +    ///
> > > > > > +    /// # Safety
> > > > > > +    ///
> > > > > > +    /// Callers must ensure that no hardware operations that involve the buffer are currently
> > > > > > +    /// taking place while the returned slice is live.
> > > > > > +    pub unsafe fn as_slice(&self, offset: usize, count: usize) -> Result<&[T]> {
> > > > >
> > > > > You were asked to rename this function because it returns a slice, but
> > > > > I wonder if it's better to take an `&mut [T]` argument and to have
> > > > > this function copy data into that argument. That way, we could make
> > > > > the function itself safe. Perhaps the actual copy needs to be
> > > > > volatile?
> > > >
> > > > Why do we consider the existing one unsafe?
> > > >
> > > > Surely, it's not desirable that the contents of the buffer are modified by the
> > > > HW unexpectedly, but is this a concern in terms of Rust safety requirements?
> > > >
> > > > And if so, how does this go away with the proposed approach?
> > >
> > > In Rust, it is undefined behavior if the value behind an immutable
> > > reference changes (unless the type uses UnsafeCell / Opaque or
> > > similar). That is, any two consecutive reads of the same immutable
> > > reference must return the same byte value no matter what happened in
> > > between those reads.
> >
> > Undefined as in the sense of anything is allowed to happen?
> 
> Yes.
> 
> > I thought undefined
> > as in you might still see the old value on two consecutive reads.
> 
> That is the optimization that motivates this being UB, but it's
> defined as full UB.
> 
> > Do you have a pointer to the corresponding docs?
> 
> Sure, it's on the "behavior considered undefined" page:
> Moreover, the bytes pointed to by a shared reference, including
> transitively through other references (both shared and mutable) and
> Boxes, are immutable; transitivity includes those references stored in
> fields of compound types.
> 
> https://doc.rust-lang.org/reference/behavior-considered-undefined.html#r-undefined.immutable
> 
> > > If we manually perform the read as a volatile read, then it is
> > > arguably allowed for the value to be modified by the hardware while we
> > > read the value.
> >
> > From read_volatile() [1]: "In particular, a race between a read_volatile and any
> > write operation to the same location is undefined behavior."
> 
> I mean, ultimately we are a bit on our own here. In C code you just
> use an ordinary read / write, so we could use the ordinary
> core::ptr::copy_nonoverlapping method to mirror that. We've been told
> from the Rust project that we should just do these kinds of things
> like we do in C - technically these things aren't okay in C either,
> but because LLVM will try to avoid breaking patterns used in the
> kernel, they shouldn't break in Rust either.
> 
> But using an immutable reference should be avoided because that gives
> LLVM optimization hints that we are not giving to LLVM in C code.

Thanks for clarifying.

I think we should add this as a note to the corresponding code.