Tang, Changqing wrote:
Or: Thank you for the description. I have read the spec carefully and got some idea. But here is a case I don't know.I have 1024 QPs on a single port/cable. There is NO receive posted because I use pure RDMA write. And also there is no pending send. At this point I pull the cable out. I will get the port error event(right ?). Do I also get 1024 QP error events ? Because there is no way to report through completion status. Or the QPs are still in good state even though I pull out cable ?
All the QPs for which there is inflight TX (send/rdma) when the cable is plugged out would be moved by the HCA to the ERROR state, and you will detect this by getting completion with error on the associated CQ. As Roland explained (and pointed you to the spec...) you will not QP error (async) event just b/c the cable was plugged out.
Now, as for the QPs which are "idle" before the time of the link down (cable removal) and till the link is up (cable is back) - my IB understanding tells me that they should be live and kicking and you should be able to use them. This b/c the RC QP lives in IB L4 (transport layer, the equivalent of TCP) and the port up/down is IB L2 (link layer, the ~equivalent of Ethernet for this discussion) event.
Or. _______________________________________________ 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
