>>Add 1 byte header containing provider/vendor major revision >>to distinguish between consumer and non-consumer rejects. >>Validate size of consumer reject privated data. > >Not saying this is a bad idea, but doesn't it break the >protocol with existing >DAPL? It also shifts all of the existing private data off by >a byte, which >could result in odd data alignment.
If the cma/cm could guarantee that IB_CM_REJ_CONSUMER_DEFINED is always an indication of true consumer called reject versus abnormal termination then I would not need to add the provider header in reject private data. Anyway, private data delivery in rejects is new for DAT v2 and exposed for the first time with this patch set. There is no compatibility issue with existing DAPL because reject private data has been ignored up until this point. I will adjust for odd data alignment. Thanks for the feedback, -arlin . _______________________________________________ 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
