Comment 35 for bug 1403152

Great progress, thanks Chris. I did some initial instrumentation of dev_put and dev_hold calls earlier, dumping out the call stack and comparing a good run to a bad one. Nothing definitive yet, and it's bit tricky to match up on an SMP machine, but one thing stood out: many of the bad traces had UDP in the mix (for example, calls to __udp4_lib_mcast_deliver and __udp4_lib_rcv). Maybe useful information, maybe not.