On Mon, Jan 18, 2021 at 08:53:24PM +0100, Klaus Jensen wrote: > On Jan 18 21:59, Minwoo Im wrote: > > > The spec requires aligned 32 bit accesses (or the size of the register), > > > so maybe it's about time we actually ignore instead of falling through. > > > > Agreed. > > > > On the other hand. > > The spec just allows undefined behavior if the alignment or size > requirement is violated. So falling through is not wrong. > > Keith, any thoughts on this?
If the host's access is unaligned and breaks something, the host gets to keep both pieces.