Finally got around to it. Thanks to some tips from a fellow Linode/NetBSDer I could avoid almost all of the pitfalls:

  1. Ahead of migration edit /etc/fstab and change all xbd1.* to wd0.*. They were “1”s because XEN required a boot disk, but since that was going to be deleted with KVM I knew the disk would change to “0”.
  2. Also before migration edit rc.conf and change xennet0 to wm0.
  3. And ideally before migration, but I forgot and only did after migration: Edit /etc/cgd/cgd.conf and change xbd1.* to wd0.* and rename /etc/cgd/xbd1e to /etc/cgd/wd0e.

And that’s it. Was much less painful than I was anticipating. At the moment I don’t have serial bootblocks so I don’t have lish access, only glish, but come the next NetBSD update I’ll correct that.

[EDIT: 2018-01-23] See updates on this previous post regarding device names/prefixes if using para-virtualization instead of full (as per above).