Testing s390x with RUST_BACKTRACE=1 in QEMU crashes

[alexcrichton]

In landing #9702 I was wrestling with a s390x-specific failure on CI. The problem seems to stem from RUST_BACKTRACE=1 and using std::backtrace which an updated version of anyhow uses. This code is now all landed on main so the current main branch of Wasmtime fails with:

$ export RUST_BACKTRACE=1 
$ export CARGO_PROFILE_DEV_OPT_LEVEL=2 
$ cargo test -p wasmtime-wasi --target s390x-unknown-linux-gnu -p wasmtime-wasi
...
    Finished `test` profile [optimized + debuginfo] target(s) in 56.55s
     Running unittests src/lib.rs (target/s390x-unknown-linux-gnu/debug/deps/wasmtime_wasi-e36f6fb1833db8d7)

running 15 tests
test host::filesystem::test::table_readdir_works ... ok
test stdio::test::memory_stdin_stream ... ok
test random::test::deterministic ... ok
test stdio::test::async_stdout_stream_unblocks ... ok
test pipe::test::backpressure_read_stream ... ok
test stdio::test::async_stdin_stream ... ok
test pipe::test::infinite_read_stream ... ok
test pipe::test::finite_read_stream ... ok
test pipe::test::empty_read_stream ... ok
test pipe::test::sink_write_stream ... ok
test pipe::test::closed_write_stream ... ok
test pipe::test::multiple_chunks_write_stream ... ok
test pipe::test::multiple_chunks_read_stream ... ok
test pipe::test::backpressure_write_stream ... ok
test pipe::test::backpressure_write_stream_with_flush ... ok

test result: ok. 15 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.36s

     Running tests/all/main.rs (target/s390x-unknown-linux-gnu/debug/deps/all-d771c2793612b780)

running 196 tests
test async_::preview1_clock_time_get ... ok
test async_::preview1_fd_filestat_get ... ok
test api::api_time ... ok
test api::api_reactor ... ok
test async_::preview1_big_random_buf ... ok
qemu: uncaught target signal 11 (Segmentation fault) - core dumped
error: test failed, to rerun pass `-p wasmtime-wasi --test all`

Caused by:
  process didn't exit successfully: `qemu-s390x -L /usr/s390x-linux-gnu -E LD_LIBRARY_PATH=/usr/s390x-linux-gnu/lib -E WASMTIME_TEST_NO_HOG_MEMORY=1 /home/alex/code/wasmtime/target/s390x-unknown-linux-gnu/debug/deps/all-d771c2793612b780` (signal: 11, SIGSEGV: invalid memory reference)

@uweigand would you be able to help take a closer look at this? I'm not sure if this is a Wasmtime/jit code issue (maybe unwind info?) or something else in rustc perhaps

[uweigand]

Good news is I can reproduce natively, so it's not a qemu issue. The segfault happens in MD_FALLBACK_FRAME_STATE_FOR in libgcc, because of an invalid PC. This typically indicates some problem with unwind info in a lower frame. And indeed GDB also isn't able to unwind fully. I'll need to look where this comes from.

[uweigand]

This is a regression introduced with the tail-call ABI. If we have incoming tail-call stack arguments, the DWARF rule to unwind the caller's SP is incorrect. I'm working on a fix.

[uweigand]

The above PR fixes the issue for me.