> > > Keep a pointer to the local (src) netdevice in struct rdma_dev_addr. In > > a > > > High-Availability scheme this information can be used by the rdma-cm to > > align RDMA > > > sessions to use the same links as the IP stack does under fail-over and > > route change cases.
> > I don't understand how keeping the netdevice is related to this chunk: > The rdma-cm doesn't issue address resolution (route lookup, neigh > lookup, etc) in the passive side of the connection and hence there is > some asymmetry between the code path taken at the client side vs the > server side. This difference had two implications: > > - the passive side network device was not resolved, where now it is > resolved by rdma_translate_ip > > - there was some trivial code duplication between rdma_copy_addr to > cma_new_conn_id, which is now removed as with this patch the chain of > calls becomes cma_new_conn_id --> rdma_translate_ip --> rdma_copy_addr So you need to add the call to rdma_translate_ip() now to get the netdevice set properly? _______________________________________________ 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
