Re: [PATCH v14 02/11] rust: add dma coherent allocator abstraction.

[Boqun Feng]

On Tue, Mar 11, 2025 at 09:34:19PM +0000, Benno Lossin wrote:
> On Tue Mar 11, 2025 at 7:12 PM CET, Boqun Feng wrote:
> > On Tue, Mar 11, 2025 at 07:47:58PM +0200, Abdiel Janulgue wrote:
> > [...]
> >> +    /// Reads the value of `field` and ensures that its type is [`FromBytes`].
> >> +    ///
> >> +    /// # Safety
> >> +    ///
> >> +    /// This must be called from the [`dma_read`] macro which ensures that the `field` pointer is
> >> +    /// validated beforehand.
> >> +    ///
> >> +    /// Public but hidden since it should only be used from [`dma_read`] macro.
> >> +    #[doc(hidden)]
> >> +    pub unsafe fn field_read<F: FromBytes>(&self, field: *const F) -> F {
> >> +        // SAFETY: By the safety requirements field is valid.
> >> +        unsafe { field.read_volatile() }
> >
> > I agree with Andreas that we should document the exception of usage on
> > {read,write}_volatile() here. How about:
> >
> > When dealing with a potential race from a hardware or code outside
> > kernel (e.g. user-space program), we need that read and write on a valid
> > memory are not UBs. Currently {read,write}_volatile() are used for this,
> 
> I would use the singular `UB` here and below.
> 
> > and the rationale behind is that they should generate the same code as
> > READ_ONCE() and WRITE_ONCE() which kernel already relies on to avoid UBs
> 
> s/kernel/the kernel/
> 
> > on data races. Note that the usage of {read,write}_volatile() is limited
> > to this particular case, they cannot be used to emit the UBs caused by
> 
> s/emit/prevent/
> 

These above all looks reasonable to me.

> > racing between two kernel functions nor do they provide atomicity.
> >
> > Thoughts? One problem is that I don't know where to put this document
> > :-( Any suggestion?
> 
> I am a bit out of the loop on this one, but why not put into the safety
> comment? I.e. explicitly state that this is *not* sound as per the usual
> rules and it is a special exception?
> 

We may end up with multiple uses of {read,write}_volatile(), and IIUC,
Andreas wanted [1] some clear documentation on this. Also if we have
some document it'll be easier to sync with Rust language people on the
"rules" we following in the kernel.

[1]: https://lore.kernel.org/lkml/87mse2hrd8.fsf@xxxxxxxxxx/

Regards,
Boqun