macOS: Don't inspect dead threadtls during exception handling. (Julia…

…Lang#50871)

Otherwise we may end up with the state of a dead thread, as the system
IDs can alias. This can lead to an early return from the exception
handler, resulting in e.g. safepoint exceptions being actually delivered
to user code.

---------

Co-authored-by: Jameson Nash <[email protected]>
Co-authored-by: Cody Tapscott <[email protected]>

src/signals-mach.c

@@ -279,12 +279,16 @@ kern_return_t catch_mach_exception_raise(
  int nthreads = jl_atomic_load_acquire(&jl_n_threads);
  for (tid = 0; tid < nthreads; tid++) {
  jl_ptls_t _ptls2 = jl_atomic_load_relaxed(&jl_all_tls_states)[tid];
+ if (jl_atomic_load_relaxed(&_ptls2->current_task) == NULL) {
+ // this thread is dead
+ continue;
+ }
  if (pthread_mach_thread_np(_ptls2->system_id) == thread) {
  ptls2 = _ptls2;
  break;
  }
  }
- if (!ptls2 || ptls2->current_task == NULL) {
+ if (!ptls2) {
  // We don't know about this thread, let the kernel try another handler
  // instead. This shouldn't actually happen since we only register the
  // handler for the threads we know about.