On 8/2/06, Sean Hefty <[EMAIL PROTECTED]> wrote: > I think the open question is whether a call like rdma_establish() needs to be > added to the API. If it is, then I'm not sure that we need to route events > directly to the IB CM. Given the current choices, adding rdma_establish() may > actually be easier for the ULP than event forwarding, or maybe we just do > both...
Let me see that i follow: rdma_established is to be used by app which is a CMA consumer and now polls from a CQ a completion associated with a QP whose state is RTR (ie no RTU == ESTABLISHED event) was received yet. This may solve the need of apps who do not want to wait for the synthesized ESTABLISHED event that would be delivered at some future point with the approach taken by your patch. Moreover, the solution is one of: - the patch you sent - enforcing the ULP to call rdma_establish (or cm_establish for direct CM consumers) else a repeatedly lost RTU case is not handled. what do you mean by event forwarding, is it to fwd the COMM_EST event from the app QP handler to the CM/CMA? Or. _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
