Re: [PATCH v2 2/6] x86/rust: support MITIGATION_RETPOLINE

From: Gary Guo
Date: Wed Jul 24 2024 - 15:38:20 EST


On Wed, 24 Jul 2024 18:14:55 +0200
Miguel Ojeda <ojeda@xxxxxxxxxx> wrote:

> Support `MITIGATION_RETPOLINE` by enabling the target features that
> Clang does.
>
> The existing target feature being enabled was a leftover from
> our old `rust` branch, and it is not enough: the target feature
> `retpoline-external-thunk` only implies `retpoline-indirect-calls`, but
> not `retpoline-indirect-branches` (see LLVM's `X86.td`), unlike Clang's
> flag of the same name `-mretpoline-external-thunk` which does imply both
> (see Clang's `lib/Driver/ToolChains/Arch/X86.cpp`).
>
> Without this, `objtool` would complain if enabled for Rust, e.g.:
>
> rust/core.o: warning: objtool:
> _R...escape_default+0x13: indirect jump found in RETPOLINE build
>
> In addition, change the comment to note that LLVM is the one disabling
> jump tables when retpoline is enabled, thus we do not need to use
> `-Zno-jump-tables` for Rust here -- see commit c58f2166ab39 ("Introduce
> the "retpoline" x86 mitigation technique ...") [1]:
>
> The goal is simple: avoid generating code which contains an indirect
> branch that could have its prediction poisoned by an attacker. In
> many cases, the compiler can simply use directed conditional
> branches and a small search tree. LLVM already has support for
> lowering switches in this way and the first step of this patch is
> to disable jump-table lowering of switches and introduce a pass to
> rewrite explicit indirectbr sequences into a switch over integers.
>
> As well as a live example at [2].
>
> These should be eventually enabled via `-Ctarget-feature` when `rustc`
> starts recognizing them (or via a new dedicated flag) [3].
>
> Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
> Link: https://github.com/llvm/llvm-project/commit/c58f2166ab3987f37cb0d7815b561bff5a20a69a [1]
> Link: https://godbolt.org/z/G4YPr58qG [2]
> Link: https://github.com/rust-lang/rust/issues/116852 [3]
> Signed-off-by: Miguel Ojeda <ojeda@xxxxxxxxxx>

Reviewed-by: Gary Guo <gary@xxxxxxxxxxx>

> ---
> arch/x86/Makefile | 2 +-
> scripts/generate_rust_target.rs | 7 +++++++
> 2 files changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/Makefile b/arch/x86/Makefile
> index 801fd85c3ef6..e8214bff1aeb 100644
> --- a/arch/x86/Makefile
> +++ b/arch/x86/Makefile
> @@ -220,7 +220,7 @@ ifdef CONFIG_MITIGATION_RETPOLINE
> KBUILD_CFLAGS += $(RETPOLINE_CFLAGS)
> # Additionally, avoid generating expensive indirect jumps which
> # are subject to retpolines for small number of switch cases.
> - # clang turns off jump table generation by default when under
> + # LLVM turns off jump table generation by default when under
> # retpoline builds, however, gcc does not for x86. This has
> # only been fixed starting from gcc stable version 8.4.0 and
> # onwards, but not for older ones. See gcc bug #86952.
> diff --git a/scripts/generate_rust_target.rs b/scripts/generate_rust_target.rs
> index 641b713a033a..44952f0a3aac 100644
> --- a/scripts/generate_rust_target.rs
> +++ b/scripts/generate_rust_target.rs
> @@ -164,7 +164,14 @@ fn main() {
> );
> let mut features = "-3dnow,-3dnowa,-mmx,+soft-float".to_string();
> if cfg.has("MITIGATION_RETPOLINE") {
> + // The kernel uses `-mretpoline-external-thunk` (for Clang), which Clang maps to the
> + // target feature of the same name plus the other two target features in
> + // `clang/lib/Driver/ToolChains/Arch/X86.cpp`. These should be eventually enabled via
> + // `-Ctarget-feature` when `rustc` starts recognizing them (or via a new dedicated
> + // flag); see https://github.com/rust-lang/rust/issues/116852.
> features += ",+retpoline-external-thunk";
> + features += ",+retpoline-indirect-branches";
> + features += ",+retpoline-indirect-calls";
> }
> ts.push("features", features);
> ts.push("llvm-target", "x86_64-linux-gnu");
> --
> 2.45.2