>A) it seems that other than some error handling code, only the flow of >rdma_disconnect() and RDMA_CM_EVENT_REJECTED move the QP associated >with this ID to the ERROR state. From that I conclude that in order to >get flushes on all the WR posted to the QP one must call >rdma_disconnect() in both sides of the connection. Am I right, is this >what we want? if yes, lets document this.
Correct - both sides must call disconnect. rdma_cm.7 and rdma_disconnect.3 man pages call this out to some degree. >B) will RDMA_CM_EVENT_DISCONNECTED event would --always-- be generated >also for the side that called rdma_disconnect()? in both cases (yes >and no), we need to document this. I need to verify this. - 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
