phd: we should test NICs #1044
Description
kind of a meta issue: NICs don't get tested in Propolis CI yet and we've missed bugs because of it.
- Instance network interface is gone after guest reboot #1030 we should of course test that NICs work across reboots
- propolis-standalone(?) virtio NIC migration seems funky #987 we should also test that NICs work across migration
- ASpace asserts when covering exactly one element #1019 was tickled because of NIC stuff that we probably wouldn't have seen in Propolis (maybe we should test iPXE! idk) but the most basic "poke every PCI register" fuzzing would have tripped it immediately
- virtio device reset without multiqueue kills vnic instead #1045 makes me think in-propolis (mainly: not dependent on guest drivers) tests of virtio device lifecycle would be super neat-o
- viona: do not retain peak nqueues across device reset #1051 the guest configurations that get here are weird and hard to reason about, it would be much more straightforward for a test to request 10 pairs, then 4 pairs, then reset, etc. or in the most complicated guest case: 1 pair at reset, to enable multiqueue, to 4 pairs used from the guest, to 1 at reset, and so on.
- virtio-nic: setting features should be idempotent? #1052 should be very easy to validate in a PHD kind of context
it might not be unreasonable to have a similar kind of fuzzer like #966 that does some series of operations that should at least never get the device to NEEDS_RESET..