Fair point about ifunc, but we're using our own table of function pointers, which can be invalidated/re-initialized. Someone also mentioned that the OS could catch SIGILL.. indeed seems doable to then reset thread affinity to the P cores?