Ah, this makes a lot more sense now. I thought the "fault" was about the kernel interrupting when a new page needs to be loaded into physical memory, which would also happen for memcpy.