asm! support for the Xtensa architecture

[MabezDev]

This implements the asm! support for Xtensa. We've been using this code for a few years in our fork and it's been working well. I finally found some time to clean it up a bit and start the upstreaming process. This should be one of the final PRs for Xtensa support on the Rust side (minus bug fixes of course). After this, we're mostly just waiting on the LLVM upstreaming which is going well. This PR doesn't cover all possible asm options for Xtensa, but the base ISA plus a few extras that are used in Espressif chips.

r? Amanieu

[rustbot]

Some changes occurred in compiler/rustc_codegen_gcc

cc @antoyo, @GuillaumeGomez

[Amanieu]

This is also missing an update to the unstable book entry for asm_experimental_arch.

View changes since this review

[bors]

☔ The latest upstream changes (presumably #147645) made this pull request unmergeable. Please resolve the merge conflicts.

[MabezDev]

I've made most of the updates here (sorry for the delay!).

It turns out we'll need a patch that landed just after rust branched LLVM: https://github.com/rust-lang/llvm-project/tree/rustc/21.1-2025-08-01/llvm/lib/Target/Xtensa , otherwise the asm tests fail :(.

The good news is that we've since landed FP support in LLVM upstream too, so maybe the next time rust branches LLVM 21 (I guess 21.2?) we can rerun CI on this PR and it will be green.

[Amanieu]

Unfortunately that's not exactly true. LLVM will force the use of a frame pointer if the stack frame has variable-sized objects. While this currently isn't possible in Rust, it is possible when doing cross-language LTO with C code. Therefore we should always reject the use of the frame pointer in asm.

[Amanieu]

Looking at the LLVM code, it seems use a7 instead of a15 as the frame pointer when the windowed ABI is used.

[Amanieu]

This shouldn't be part of the reg class since it doesn't usually make sense for asm to select this regiser for the reg constraint. It should probably be in its own clobber-only register class.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[rust-log-analyzer]

The job pr-check-2 failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

/dev/sda15      105M  6.2M   99M   6% /boot/efi
tmpfs           1.6G   12K  1.6G   1% /run/user/1001
================================================================================

Sufficient disk space available (95818484KB >= 52428800KB). Skipping cleanup.
##[group]Run echo "[CI_PR_NUMBER=$num]"
echo "[CI_PR_NUMBER=$num]"
shell: /usr/bin/bash --noprofile --norc -e -o pipefail {0}
---
##[group]Linting stage2 bootstrap (stage1 -> stage2, x86_64-unknown-linux-gnu)
error: failed to run `rustc` to learn about target-specific information

Caused by:
  process didn't exit successfully: `/checkout/obj/build/bootstrap/debug/rustc /checkout/obj/build/x86_64-unknown-linux-gnu/stage1/bin/clippy-driver /checkout/obj/build/bootstrap/debug/rustc - --crate-name ___ --print=file-names --sysroot /checkout/obj/build/x86_64-unknown-linux-gnu/stage1 --cfg=windows_raw_dylib -Csymbol-mangling-version=v0 -Zannotate-moves -Zunstable-options -Zforce-unstable-if-unmarked -Zmacro-backtrace -Csplit-debuginfo=off -Clink-args=-Wl,-z,origin '-Clink-args=-Wl,-rpath,$ORIGIN/../lib' -Zunstable-options --target x86_64-unknown-linux-gnu --crate-type bin --crate-type rlib --crate-type dylib --crate-type cdylib --crate-type staticlib --crate-type proc-macro --print=sysroot --print=split-debuginfo --print=crate-name --print=cfg -Wwarnings` (exit status: 101)
  --- stderr

  thread 'rustc' (32276) panicked at compiler/rustc_span/src/symbol.rs:2720:13:
  duplicate symbols in the rustc symbol list and the extra symbols added by the driver: ["println_macro", "thread_local_macro", "convert_identity", "print_macro"]
  stack backtrace:
     0: __rustc::rust_begin_unwind
     1: core::panicking::panic_fmt
     2: <rustc_span::symbol::Interner>::with_extra_symbols
     3: <rustc_span::SessionGlobals>::new
---
  warning: the ICE couldn't be written to `/checkout/rustc-ice-2026-03-03T16_38_04-32275.txt`: Read-only file system (os error 30)

  note: rustc 1.96.0-nightly (c69cde14c 2026-03-03) running on x86_64-unknown-linux-gnu

  note: compiler flags: -C symbol-mangling-version=v0 -Z annotate-moves -Z unstable-options -Z force-unstable-if-unmarked -Z macro-backtrace -C split-debuginfo=off -C link-args=-Wl,-z,origin -C link-args=-Wl,-rpath,$ORIGIN/../lib -Z unstable-options --crate-type bin --crate-type rlib --crate-type dylib --crate-type cdylib --crate-type staticlib --crate-type proc-macro -Z on-broken-pipe=kill -Z tls-model=initial-exec -Z allow-features=binary-dep-depinfo,proc_macro_span,proc_macro_span_shrink,proc_macro_diagnostic

  query stack during panic:
  end of query stack
  note: Clippy version: clippy 0.1.95 (c69cde14cd 2026-03-03)