The page-boundary detail is the part I found most surprising. It’s a good example of why “this opcode always traps” is not enough if the CPU still has to decode extra bytes. ud2 being parameterless makes its behavior much easier to reason about.