Re: [PATCH] rust: io: convert ResourceSize into a transparent newtype

[Lorenzo Delgado <lnsdev@proton.me>]

On Thu Jul 16, 2026 at 11:45 PM CEST, Danilo Krummrich wrote:
> On Sun Jul 12, 2026 at 1:36 PM CEST, Lorenzo Delgado wrote:
> > -        let num_pages = usize::from_safe_cast(sg_entry.dma_len()).div_ceil(GSP_PAGE_SIZE);
> > +        let num_pages = usize::try_from(sg_entry.dma_len())?.div_ceil(GSP_PAGE_SIZE);
>
> I think this is worse, as the conversion becomes fallible.

Agreed, that's a regression for nova-core, where the Kconfig guarantees
the value fits and the conversion should stay infallible. I'll drop the
try_from() there.

> You could implement From<ResourceSize> for u64 and then keep using
> usize::from_safe_cast() in nova-core.

I tried that, but it doesn't build. ResourceSize wraps resource_size_t,
which is u64 on 64-bit (CONFIG_PHYS_ADDR_T_64BIT), so the impl the patch
already has,

    impl From<ResourceSize> for bindings::resource_size_t

is already From<ResourceSize> for u64 there, and a second one conflicts:

    error[E0119]: conflicting implementations of trait
                  `From<ResourceSize>` for type `u64`

There's a simpler way that stays infallible and adds nothing to io.rs.
ResourceSize already has into_raw() (io/resource.rs uses it at the C
boundaries), so nova-core can do:

    let num_pages =
        usize::from_safe_cast(sg_entry.dma_len().into_raw()).div_ceil(GSP_PAGE_SIZE);

into_raw() gives back resource_size_t, and from_safe_cast handles that
as u32 or u64 depending on the config, so it stays infallible. I'll use
that in v2 unless you'd prefer something else.

> Alternatively, we could also consider moving the FromSafeCast trait to
> rust/kernel/num.rs and add FromSafeCast<ResourceSize> impls for usize.
> [...]
> However, by making it commonly availble I do see a risk with the
> cfg-gated impls silently breaking the build.

Agreed on the risk. A FromSafeCast<ResourceSize> for usize impl would
have to be cfg-gated like the u64 one, which is the same
silent-breakage-under-randconfig case you mention.

Since into_raw() keeps this patch self-contained, moving FromSafeCast
into the kernel crate is a separate change from the newtype conversion.
It seems worth doing on its own, and I'm happy to send it as its own
series so it gets reviewed as a new core API, but it doesn't need to
block this patch. I'll post v2 with the into_raw() change.

Thanks for the review.

Lorenzo