Is it worth doing that here? We're storing the mbuf in the context of a struct ifnet anyway. It might make more sense to just clear rcvif when we enqueue it and populate it again in epair_sintr().
That's sort of what we do in "real" network drivers as well. We populate rcvif in the receive path.

In D33269#752504, @kp wrote:

Is it worth doing that here? We're storing the mbuf in the context of a struct ifnet anyway. It might make more sense to just clear rcvif when we enqueue it and populate it again in epair_sintr().
That's sort of what we do in "real" network drivers as well. We populate rcvif in the receive path.

This makes a lot of sense. I will abandon this revision in favor of a different one.