Instances are not wholly cleaned up when they error out due to unexpected signal stack sizes.

[fst-crenshaw]

The moment when the lucet runtime switches into a guest to extract the instance's result is a function called swap_and_return. If the call to self.with_signals_on returns with an Error resulting from an unexpected signal stack size, swap_and_return returns without cleaning up the borrowed instance. As in, this doesn't happen.

        CURRENT_INSTANCE.with(|current_instance| {
            *current_instance.borrow_mut() = None;
        });

That moment is here:

lucet/lucet-runtime/lucet-runtime-internals/src/instance.rs

Line 972 in c765316

})?;

To reproduce the issue:

1. Go to ~/lucet-runtime/lucet-runtime-internal
2. Run cargo test -- --test-threads=1

The result:

---- alloc::tests::reject_sigstack_smaller_than_min stdout ----
thread 'main' panicked at 'no other instance is running on this thread', lucet-runtime/lucet-runtime-internals/src/instance.rs:956:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace.
failures:
  alloc::tests::reject_sigstack_smaller_than_min
test result: FAILED. 65 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out
error: test failed, to rerun pass '--lib'

Tactically, what happens:

This test fails (expectedly) but the cleanup doesn't happen.
alloc::tests::reject_debug_sigstack_smaller_than_12kib

So this test fails (unexpectedly):
alloc::tests::reject_sigstack_smaller_than_min

[acfoltzer]

058cc85