Bug 1551671

pthread_start.c contains this:

/* We want to make sure that GC_thread_exit_proc() is unconditionally   */
/* invoked, even if the client is not compiled with -fexceptions, but   */
/* the GC is.  The workaround is to put GC_inner_start_routine() in its */
/* own file (pthread_start.c), and undefine __EXCEPTIONS in the GCC     */
/* case at the top of the file.  FIXME: it's still unclear whether this */
/* will actually cause the exit handler to be invoked last when         */
/* thread_exit is called (and if -fexceptions is used).                 */
#if defined(__GNUC__) && defined(__linux__)
  /* We undefine __EXCEPTIONS to avoid using GCC __cleanup__ attribute. */
  /* The current NPTL implementation of pthread_cleanup_push uses       */
  /* __cleanup__ attribute when __EXCEPTIONS is defined (-fexceptions). */
  /* The stack unwinding and cleanup with __cleanup__ attributes work   */
  /* correctly when everything is compiled with -fexceptions, but it is */
  /* not the requirement for this library clients to use -fexceptions   */
  /* everywhere.  With __EXCEPTIONS undefined, the cleanup routines are */
  /* registered with __pthread_register_cancel thus should work anyway. */
# undef __EXCEPTIONS
#endif

Fedora is built with -fexceptions, and our tools assume that unwind tables exist, so gc should rely on this function, to benefit from pthread_cancel hardening (which the above hack bypasses).