Using the `--with-derive-custom-{struct,union}` option of bindgen, add
`#[derive(MaybeZeroable)]` to every struct & union. This makes those
types implement `Zeroable` if all their fields implement it.
Sadly bindgen doesn't add custom derives to the `__BindgenBitfieldUnit`
struct. So manually implement `Zeroable` for that.
Signed-off-by: Benno Lossin <[email protected]>
Reviewed-by: Alice Ryhl <[email protected]>
[ Formatted comment. - Miguel ]
Signed-off-by: Miguel Ojeda <[email protected]>
For some architectures, such as X86_64, ARCH_KMALLOC_MINALIGN is not
resolvable for bindgen. E.g. due to being defined as
__alignof__(unsigned long long).
Hence, we have to create a rust helper, i.e. let the C compiler evaluate
the expression and store it in a const.
However, if for other architectures, such as arm64,
ARCH_KMALLOC_MINALIGN does evaluate to something that can be directly
processed by bindgen, we end up with multiple definitions of
ARCH_KMALLOC_MINALIGN in the generated bindings.
error[E0428]: the name `ARCH_KMALLOC_MINALIGN` is defined multiple times
--> /builddir/build/BUILD/kernel-6.17.0-build/kernel-next-20250818/linux-6.17.0-0.0.next.20250818.423.vanilla.fc44.aarch64/rust/bindings/bindings_generated.rs:134545:1
|
9622 | pub const ARCH_KMALLOC_MINALIGN: u32 = 8;
| ----------------------------------------- previous definition of the value `ARCH_KMALLOC_MINALIGN` here
...
134545 | pub const ARCH_KMALLOC_MINALIGN: usize = 8;
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `ARCH_KMALLOC_MINALIGN` redefined here
|
= note: `ARCH_KMALLOC_MINALIGN` must be defined only once in the value namespace of this module
To fix this up, add ARCH_KMALLOC_MINALIGN to the blocklist of bindgen,
such that we always only generate the symbol from the rust helper.
Reported-by: Thorsten Leemhuis <[email protected]>
Closes: https://lore.kernel.org/all/[email protected]/
Fixes: 1b1a946dc2 ("rust: alloc: specify the minimum alignment of each allocator")
Tested-by: Thorsten Leemhuis <[email protected]>
Reviewed-by: Alice Ryhl <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Danilo Krummrich <[email protected]>
Currently bindgen has special logic to recognise `size_t` and `ssize_t`
and map them to Rust `usize` and `isize`. Similarly, `ptrdiff_t` is
mapped to `isize`.
However this falls short for `__kernel_size_t`, `__kernel_ssize_t` and
`__kernel_ptrdiff_t`. To ensure that they are mapped to usize/isize
rather than 32/64 integers depending on platform, blocklist them in
bindgen parameters and manually provide their definition.
Signed-off-by: Gary Guo <[email protected]>
Reviewed-by: Alice Ryhl <[email protected]>
Reviewed-by: Trevor Gross <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
[ Formatted comment. - Miguel ]
Signed-off-by: Miguel Ojeda <[email protected]>
We use const helpers in form of
const size_t RUST_CONST_HELPER_ARCH_SLAB_MINALIGN = ARCH_SLAB_MINALIGN;
to aid generation of constants by bindgen because it is otherwise a
macro definition of an expression and bindgen doesn't expand the
constant. The helpers are then have `RUST_CONST_HELPER` prefix stripped
and exposed to Rust code as if `ARCH_SLAB_MISALIGN` is generated
natively by bindgen.
This works well for most constants, but on RISC-V, `ARCH_SLAB_MINALIGN`
is defined directly as literal constant if `!CONFIG_MMU`, and bindgen
would generate `ARCH_SLAB_MINALIGN` directly, thus conflict with the
one generated through the helper.
To fix this, we simply need to block bindgen from generating directly
without going through helper.
Reported-by: kernel test robot <[email protected]>
Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
Signed-off-by: Gary Guo <[email protected]>
Reviewed-by: Christian Schrefl <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Miguel Ojeda <[email protected]>
Neither bindgen nor Rust know about the preserve-most calling
convention, and Clang describes it as unstable. Since we aren't using
functions with this calling convention from Rust, blocklist them.
These functions are only added to the build when list hardening is
enabled, which is likely why others didn't notice this yet.
Signed-off-by: Matthew Maurer <[email protected]>
Reviewed-by: Martin Rodriguez Reboredo <[email protected]>
Reviewed-by: Alice Ryhl <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
[ Used Markdown for consistency with the other comments in the file. ]
Signed-off-by: Miguel Ojeda <[email protected]>
Benno Lossin