Stream: git-wasmtime

Topic: wasmtime / PR #15072 cranelift: aarch64: fix BTI landing ...

Wasmtime GitHub notifications bot (Oct 29 2026 at 18:08):

NicolasStigler opened PR #15072 from NicolasStigler:fix/aarch64-bti-indirect-targets to bytecodealliance:main:

Fixes #14938

Why

With BTI and pointer authentication enabled on AArch64, several indirect control-flow paths could trap with an illegal instruction. The affected cases include br_table defaults, exception-handler targets, indirect tail calls, and tail-convention prologues using paciaz/pacibz.

What changed

• Mark br_table defaults and exception-handler targets as indirect branch targets so they receive bti j.
• Emit bti c before paciaz/pacibz where needed.
• Route indirect tail calls through x16 before branching.
• Add and update AArch64 filetests and runtime configurations for BTI and pointer authentication.

Testing

• cargo +1.97.0 fmt --all -- --check
• cargo +1.97.0 clippy -p cranelift-codegen --all-targets
• cargo +1.97.0 test -p cranelift-codegen
• cargo +1.97.0 test -p cranelift-filetests
• cargo +1.97.0 test -p cranelift-tools --test filetests
• cargo +1.97.0 test -p cranelift-tools --test logged-filetests
• AArch64 Wasmtime WAST reproducers under QEMU with BTI and pointer-authentication enabled

Wasmtime GitHub notifications bot (Oct 29 2026 at 18:08):

NicolasStigler requested cfallin for a review on PR #15072.

Wasmtime GitHub notifications bot (Oct 29 2026 at 18:08):

NicolasStigler removed the review request for wasmtime-compiler-reviewers from PR #15072.

Wasmtime GitHub notifications bot (Oct 29 2026 at 18:08):

NicolasStigler requested wasmtime-compiler-reviewers for a review on PR #15072.

Wasmtime GitHub notifications bot (Oct 29 2026 at 19:32):

:memo: cfallin submitted PR review:

Thanks! A few thoughts below but overall looks good.

Wasmtime GitHub notifications bot (Oct 29 2026 at 19:32):

:speech_balloon: cfallin created PR review comment:

Can we do this only when PAC/BTI is enabled? I don't want to pessimize the case when it's not (e.g. default aarch64 Linux).

Wasmtime GitHub notifications bot (Oct 29 2026 at 19:32):

:speech_balloon: cfallin created PR review comment:

Let's make a note that this is the spill-temp register, which by Cranelift convention we are allowed to use as a scratch register within one VCode inst, so this is safe.

Wasmtime GitHub notifications bot (Oct 29 2026 at 21:44):

github-actions[bot] added the label cranelift on PR #15072.

Wasmtime GitHub notifications bot (Oct 29 2026 at 21:44):

github-actions[bot] added the label cranelift:area:machinst on PR #15072.

Wasmtime GitHub notifications bot (Oct 29 2026 at 21:44):

github-actions[bot] added the label cranelift:area:aarch64 on PR #15072.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:16):

NicolasStigler updated PR #15072.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:18):

NicolasStigler commented on PR #15072:

Would it be better if i made a new PR or its fine adding a commit to this branch? If a new PR is preferred I can squash both commits in one.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:22):

:memo: NicolasStigler submitted PR review.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:22):

:speech_balloon: NicolasStigler created PR review comment:

The mov is now gated on use_bti(), so the default aarch64 path doesnt pay an extra instruction.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:25):

NicolasStigler edited a comment on PR #15072:

Would it be better if i made a new PR or its fine adding a commit to this branch? If a new PR is preferred I can squash both commits in one. Im asking because I reverted the unnecesary expectation changes in return-call, call-pauth and tail-call-grow-frame*.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:27):

cfallin commented on PR #15072:

Would it be better if i made a new PR or its fine adding a commit to this branch? If a new PR is preferred I can squash both commits in one. Im asking because I reverted the unnecesary expectation changes in return-call, call-pauth and tail-call-grow-frame*.

No, adding commits here to address review feedback is the expected path -- the intent of a review is to say "add these bits and it'll be good". Thanks!

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:27):

:thumbs_up: cfallin submitted PR review.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:27):

cfallin has enabled auto merge for PR #15072.

Wasmtime GitHub notifications bot (Oct 30 2026 at 00:35):

cfallin added PR #15072 cranelift: aarch64: fix BTI landing pads for indirect control flow to the merge queue.

Wasmtime GitHub notifications bot (Oct 30 2026 at 06:43):

:check: cfallin merged PR #15072.

Wasmtime GitHub notifications bot (Oct 30 2026 at 06:43):

cfallin removed PR #15072 cranelift: aarch64: fix BTI landing pads for indirect control flow from the merge queue.

Last updated: Oct 31 2026 at 23:03 UTC