So, if you’re running KVM/QEMU guest domains in Fedora Server 22, it seems that adding a network device to a guest while it is operational may result in a subsequent failure to boot the guest domain (the qemu process crashes as soon as GRUB launches the Fedora kernel, if the guest OS is Fedora) with an error message in /var/log/libvirt/qemu/guestname.log that looks something like this:
2016-01-28T03:46:05.695626Z qemu-system-x86_64: Guest moved used index from 19 to 65533
I have a fairly complex KVM/QEMU arrangement with PCI device passthrough (Intel VT-d IOMMU) for multiple guest domains, and it took me a good hour to figure this one out. In fact, the only reason it took that little time was that I recalled at the end of that hour my resolution for the issue when it happened the first time! I don’t know why I didn’t post it on here the first time around (since I recall it taking an hour or two to understand), so here it is:
Root Cause Theory:
It seems this message refers to the virtio driver’s ability to signal to the hypervisor that the associated hardware is no longer being used by the guest domain. This might be the result of my particular PCIe Ethernet adapter – more research is needed.
Stripping the virtio NICs from your guest domain or changing them over to the less-appealing but non-index-moving emulated hardware drivers and restarting the guest resolves the issue.
You’ll find a number of related RHN bugzilla reports if you Google for “Guest moved used index” (which is a suggested search when you provide just part of it to Google), but the one that directed me to the NICs as the source of the problem (or which at least reminded me of that being the source of the problem as determined the last time I undertook this investigation) was this one. Seems whatever they did didn’t resolve this bug, unless I’m mistaken about its nature.