Why? OpenBSD seems to think execute-only in userland is important. We've had SMAP on x86 for many years, it helped fixed bugs early, these bugs are rare now, so why is everyone concerned about kernel accesses that aren't using copyin(9)?

EPAN is already supported, hardware is now arriving, it's used if available, but the idea that execute-only was less important than PAN was probably misguided.

How does execute only memory disable privileged access never memory? The bigger problem I expect is the overhead of loosing PC-relative loads unless the hardware still allows these as instruction fetching related? Would you have to dedicated one of your 31 GPRs as the table of contents pointer similar to PowerPC’s ABI (e.g. sizeof(void()(void)) == 2sizeof(void *))?