Michael Tsirkin wrote: > > Not necessarily. CMA consumer at this point is a rare beast, > but since the issue is not specific to CMA, let's look at > some IB protocols: IPoIB connected mode can simply drop a > packet. So can SRP. SDP (potential CMA consumer!) simply > never needs to send any application data before communication > established event. >
I agree, the simplest and transport neutral rule is that the application is not guaranteed what will happen if they post to the send queue before the communication established event, They MAY get an error, or it MAY be queued, it MAY even be sent because the QP *is* in RTS mode but the consumer just hasn't been told yet, or it may cause a packet to be placed on the wire prematurely with possible adverse effects on the connection (iWARP on the passive side sending a DDP Segment before the active side has received the MPA Response can indeed result in a broken connection). So the best rule for the application is to simply always wait for the communication established event. That always works, and complying with it is hardly a burden for the application. _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
