https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70216
--- Comment #16 from Rich Felker <bugdal at aerifal dot cx> --- The kernel build regression is just a gratuitous unresolved symbol; the code path where is happens should not be reachable or the kernel would crash. So I think the patch as-is is fine for fixing that issue. The need for an instruction that generates SIGILL is just for ensuring that apps that reach __builtin_trap() or equivalent actually trap rather than doing something wrong/random. I think -mbuiltin-trap=insn-FFFD and making it default for Linux (and possibly other hosted targets? NetBSD?) would fully solve that problem.
