Message-ID: <20240125-bucked-payroll-47f82077b262@wendy>
Date: Thu, 25 Jan 2024 12:30:34 +0000
From: Conor Dooley <conor.dooley@...rochip.com>
To: Miguel Ojeda <miguel.ojeda.sandonis@...il.com>
CC: Conor Dooley <conor@...nel.org>, <linux-riscv@...ts.infradead.org>, Miguel
 Ojeda <ojeda@...nel.org>, Alex Gaynor <alex.gaynor@...il.com>, Wedson Almeida
 Filho <wedsonaf@...il.com>, Boqun Feng <boqun.feng@...il.com>, Gary Guo
	<gary@...yguo.net>, Björn Roy Baron
	<bjorn3_gh@...tonmail.com>, Jonathan Corbet <corbet@....net>, Paul Walmsley
	<paul.walmsley@...ive.com>, Palmer Dabbelt <palmer@...belt.com>, Nathan
 Chancellor <nathan@...nel.org>, Nick Desaulniers <ndesaulniers@...gle.com>,
	Tom Rix <trix@...hat.com>, <rust-for-linux@...r.kernel.org>,
	<linux-doc@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
	<llvm@...ts.linux.dev>
Subject: Re: [PATCH v1 0/2] RISC-V: enable rust

On Thu, Jan 18, 2024 at 05:09:40PM +0100, Miguel Ojeda wrote:
> On Thu, Jan 18, 2024 at 4:49 PM Conor Dooley <conor@...nel.org> wrote:
> >
> > The bit that worries me most is bindgen, and in particular detecting the
> > version of libclang used. I mentioned to Nathan or Nick about needing a
> > buildtime test for the version of LIBCLANG being used.
> > I'm less worried about this for LLVM=1 builds, since while I think it is
> > possible to provide a LIBCLANG path to the build system, I suspect that
> > for LLVM=1 builds it's almost always going to match the LLVM toolchain
> > in use.

I chatted with the clang built linux folks about this yesterday, Nathan
agreed that dealing with incompatibility issues iff they crop up is a
reasonable way to go.

> `scripts/rust_is_available.sh` tests whether `libclang` is at least
> the minimum LLVM supported version; and under `LLVM=1` builds, it also
> tests whether the `bindgen` found one matches the C compiler. Do you
> mean something like that?

If by "the bindgen found one matches the C compiler" you mean that the
version of libclang used by bindgen matches the C compiler, then that
sounds great.

> For `bindgen` under GCC builds, we will eventually want a "proper" way
> to use GCC instead (or possibly other approaches like querying the
> information): https://github.com/rust-lang/rust-bindgen/issues/1949.

> Recently, there has been a thread in our Zulip and a couple people are
> experimenting: https://rust-for-linux.zulipchat.com/#narrow/stream/288089-General/topic/Bindgen.20--.20GCC.20backend.20port

That link for me goes to a message on 22/01, so later than the email you
sent.

> > I'll do another rebase and resend after the merge window closes I
> > suppose :)


That said, I gave things another spin today, in a different environment,
as a final check before sending and found an issue causing kernel
panics. RISC-V (and x86/arm64) supports kcfi (CFI_CLANG) but enabling
sanitisers seems to be a nightly only option for rustc. The kernel I
built today had CFI_CLANG enabled and that caused panics when the rust
samples were loaded.

The CFI_CLANG Kconfig entry has a cc-option test for whether the option
is supported, but from a quick check I don't see a comparable test to
use for rust. Even if a test was added, the current flag is an unstable
one, so I am not sure if testing for it is the right call in the first
place, given the stabilised flag would be entirely different?

The tracking issue seems to be complete:
https://github.com/rust-lang/rust/issues/89653
but the tracking issue for sanitisiers themselves is only 3/5:
https://github.com/rust-lang/rust/issues/39699

The simple thing would be to make them mutually exclusive options in
Kconfig.

What do you think?

Cheers,
Conor.

Download attachment "signature.asc" of type "application/pgp-signature" (229 bytes)

Powered by blists - more mailing lists