Re: [PATCH v5] rust: alloc: satisfy POSIX alignment requirement

[Tamir Duberstein]

On Wed, Feb 12, 2025 at 4:24 PM Tamir Duberstein <tamird@xxxxxxxxx> wrote:
>
> On Wed, Feb 12, 2025 at 3:58 PM Danilo Krummrich <dakr@xxxxxxxxxx> wrote:
> >
> > On Wed, Feb 12, 2025 at 03:47:11PM -0500, Tamir Duberstein wrote:
> > > Looks like I wasn't the only one to fall into the trap (rust/kernel/io.rs):
> > >
> > >     #[inline]
> > >     const fn io_addr_assert<U>(&self, offset: usize) -> usize {
> > >         build_assert!(Self::offset_valid::<U>(offset, SIZE));
> > >
> > >         self.addr() + offset
> > >     }
> > >
> > > since offset isn't known at compile time, this can easily be misused?
> >
> > Well, that's intentional.
> >
> > iomem.readb(0x0)     // succeeds if SIZE >=1
> > iomem.readb(foo)     // fails if foo is not known at compile time
>
> By "fails" here you mean fail to link, right?
>
> > iomem.try_readb(foo) // succeeds if self.maxsize() >= 1
>
> Apologies for being dense throughout this discussion. Could you check
> my understanding?
>
> The trick is that `build_error` is marked `#[export_name =
> "rust_build_error"]` which isn't exported unless
> CONFIG_RUST_BUILD_ASSERT_ALLOW is defined, causing linking to it to
> fail. This even works for doctests, but not for #[test] in the kernel
> crate because they are built as part of the crate. The only to way
> make that work correctly is to put `build_error` in a crate all by
> itself.

Which of course, it is.

I thought maybe this was specific to building on macOS, but it
reproduces on Linux as well.

Gary, can you help me understand how the linker magic works? Is it
possible to make it work on the host as well?