Re: [PATCH v5 00/11] rust: bindings: Auto-generate inline static functions

[Alistair Francis]

On Tue, Jan 14, 2025 at 4:02 PM Alistair Francis <alistair23@xxxxxxxxx> wrote:
>
> On Tue, Jan 7, 2025 at 9:48 PM Andreas Hindborg <a.hindborg@xxxxxxxxxx> wrote:
> >
> > "Alistair Francis" <alistair@xxxxxxxxxxxxx> writes:
> >
> > > The kernel includes a large number of static inline functions that are
> > > defined in header files. One example is the crypto_shash_descsize()
> > > function which is defined in hash.h as
> > >
> > > ```
> > > static inline unsigned int crypto_shash_descsize(struct crypto_shash *tfm)
> > > {
> > >         return tfm->descsize;
> > > }
> > > ```
> > >
> > > bindgen is currently unable to generate bindings to these functions as
> > > they are not publically exposed (they are static after all).
> > >
> > > The Rust code currently uses rust_helper_* functions, such as
> > > rust_helper_alloc_pages() for example to call the static inline
> > > functions. But this is a hassle as someone needs to write a C helper
> > > function.
> > >
> > > Instead we can use the bindgen wrap-static-fns feature. The feature
> > > is marked as experimental, but has recently been promoted to
> > > non-experimental (depending on your version of bindgen).
> > >
> > > By supporting wrap-static-fns we automatically generate a C file called
> > > extern.c that exposes the static inline functions, for example like this
> > >
> > > ```
> > > unsigned int crypto_shash_descsize__extern(struct crypto_shash *tfm) { return crypto_shash_descsize(tfm); }
> > > ```
> > >
> > > The nice part is that this is auto-generated.
> > >
> > > We then also get a bindings_generate_static.rs file with the Rust
> > > binding, like this
> > >
> > > ```
> > > extern "C" {
> > >     #[link_name = "crypto_shash_descsize__extern"]
> > >     pub fn crypto_shash_descsize(tfm: *mut crypto_shash) -> core::ffi::c_uint;
> > > }
> > > ```
> > >
> > > So now we can use the static inline functions just like normal
> > > functions.
> > >
> > > There are a bunch of static inline functions that don't work though, because
> > > the C compiler fails to build extern.c:
> > >  * functions with inline asm generate "operand probably does not match constraints"
> > >    errors (rip_rel_ptr() for example)
> > >  * functions with bit masks (u32_encode_bits() and friends) result in
> > >    "call to ‘__bad_mask’ declared with attribute error: bad bitfield mask"
> > >    errors
> > >
> > > As well as that any static inline function that calls a function that has been
> > > kconfig-ed out will fail to link as the function being called isn't built
> > > (mdio45_ethtool_gset_npage for example)
> > >
> > > Due to these failures we use a allow-list system (where functions must
> > > be manually enabled).
> > >
> > > This series adds support for bindgen generating wrappers for inline statics and
> > > then converts the existing helper functions to this new method. This doesn't
> > > work for C macros, so we can't reamove all of the helper functions, but we
> > > can remove most.
> > >
> > > v5:
> > >  - Rebase on latest rust-next on top of v6.13-rc6
> > >  - Allow support for LTO inlining (not in this series, see
> > >    https://github.com/alistair23/linux/commit/e6b847324b4f5e904e007c0e288c88d2483928a8)
> >
> > Thanks! Since Gary just sent v2 of the LTO patch [1], could you rebase
> > on that and list it as a dependency? If you are using b4 there is a nice
>
> I can't get Gary's series to apply on rust-next (it does apply on
> master though).
>
> I might just wait until Gary's series gets picked up to rust-next as
> there is already a lot of manual rebasing going on and my series
> currently applies on rust-next.
>
> Unfortunately there are just constant conflicts as the number of
> manual C helpers are continually growing.
>
> Obviously when/if this series is approved I can do a final rebase, I
> would just like to avoid unnecessary churn in the meantime

Any more thoughts on this?

Alistair