I included that copy to avoid the logic of releasing/allocating ipoib neighbour for every packet xmit'ed before the patch query completes. I thought that it's good enough to do it just once, for the first time. Therefore, to have the mgid test pass for the second xmit, I copied the mgid even if path query fails. It turns out that it's not a good thing to do that, because if a path query fails nothing will trigger it, but ARP refresh, and it takes too much time. In case of SM failover, the first path query indeed fails. So, the best thing is probably to remove this "optimization". Besides that, the patch works same as before.
Roland Dreier wrote:
> The patch below is identical to Yossi's patch but without the copy of HA in neigh_add_path. Why did Yossi include that copy? Does this patch still fix everything? - R. _______________________________________________ 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
_______________________________________________ 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
