[PATCH 1/3] arm64: rust: Enable Rust support for AArch64

[Jamie Cunliffe]

On Tue, Jan 31, 2023 at 5:19 PM Will Deacon <will at kernel.org> wrote:
> On Thu, Jan 26, 2023 at 06:56:01PM +0100, Miguel Ojeda wrote:
> > On Thu, Jan 26, 2023 at 5:35 PM Will Deacon <will at kernel.org> wrote:
> > >
> > > Why do we need to specify this stuff here? LLVM already knows about AArch64
> > > and can compute the data-layout string in computeDataLayout(). Can we have
> > > the tools figure this out for us instead, please?
> > 
> > I agree -- the reason we have this for the moment is to provide full
> > control of the target spec, in particular for cases where `rustc` may
> > not provide enough flags to customize exiting builtin target specs
> > and/or may emit warnings for unknown target features etc.
> > 
> > If it is already enough for arm64 to use e.g. the
> > `aarch64-unknown-none` target spec plus flags on top, then we should
> > go for that, because that is the end goal: these target spec files are
> > unstable in the Rust compiler (and not intended to become stable).
> 
> Please confirm that this is the case, but I really think we should be
> aiming for that rather than starting off my specifying this stuff manually.

The json that we generate looks to be almost equivalent (that's why patch 3
exists, to answer your question on that patch). The only difference is the
`aarch64-unknown-none` rustc target sets the LLVM target to also be
`aarch64-unknown-none`, whereas the current target.json uses the
`aarch64-linux-gnu` LLVM target as does Makefile.clang for C code. As you have
been talking about the unknown-none target throughout this thread, I just wanted
to double check that's your intention?

Happy to switch over to using the rustc definition though. I just did it this
way to keep it consistent with the other architectures and not add any more
complexity into the build. On a quick test it does seem to be OK, but I'll
prepare a patch and do some proper testing that it's actually fine to use the
rustc definition though.