On 04/08/09 07:56, Jason Gunthorpe wrote: > On Mon, Aug 03, 2009 at 08:03:14PM +0300, Yossi Etigin wrote: > >> The ARP stuff works this way: Remote LID changes. In some point, either the >> remote >> node will send an ARP reply (gratuitous), or (more likely) the local network >> stack >> will start sending solicited ARPs, unicast, using the invalid path. They >> will fail, >> so the stack will send broadcast ARP. > > Erm.. Maybe a little tighter integration with the ARP/ND layer is in > order. If it knows unicast isn't working thats a pretty damn good clue > to discard the PR. > > Jason
I agree with that. If the network stack told ipoib when the neighbour became unreachable, life would have been a lot easier. Unfortunately, the closest thing now is neigh_cleanup, and this is only called when neighbour entry removed from the table (which may be quite some time after it becomes unreachable). --Yossi _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
