Michael> Does this analysis make sense? If yes, what would be the
Michael> best way to fix this?
I'm not sure if this could really happen or not. But could we add a
header_cache_update() method to the IPoIB struct net_device to handle
this situation if it does occur?
- R.