Jess reported this previously with u-boot, but I ran into this today with UEFI.

s/UEFI/EDK2/ in the description

This is similar to D45049 but also changes the behaviour of amd64. In that revision I mentioned hitting a problem with EDK2, but I can't reproduce it anymore...

Hmm, so I think we probably should do the version with a tunable so we can override it if needed. Today I hit the same issue I think you hit originally with EDK2 where for some reason bhyve assigned resources to pass through devices in higher numbered slots first, but EDK2 started at slot 0 so it started assigning conflicting resources. I do think we should make the tunable default to false when using a bootrom, but it can be forced on for old EDK2 ROMs if really needed. I would do an early return as I do in this version instead of indenting all the code and doing a dummy read/write of the cmd register.