>I think this needs to be handled with care, because there is the >obvious window while an rdma_modify_channel() operation is pending >where an application has to be prepared for an event to appear on >either the old channel and the new channel.
Agreed - this may work okay for new connection requests, where pending events end up being suppressed in the kernel until the user accepts the connection, but will be challenging as a generic API. Folding this functionality into rdma_accept() may work better as long as it doesn't break the ABI. - Sean _______________________________________________ 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
