> -----Original Message----- > > SoBeBike wrote: > > That is what I am already doing (note the comment, "// loop to > > drain..."). I loop calling ibv_poll_cq until it is empty. I just noted > > that due to the usage model, I only see it pull one CQE and then on > > the 2nd pass through the loop the CQ is empty. > > > > When you get to this scenario (for many times you don't get the CQ > event) did you try to poll the CQ and check if there is any completion > in it? > (maybe the problem is that this WR just didn't create any completion > when it ended). >
My code currently blocks (either a blocking call, or looping with a non-blocking FD) waiting for an event (ibv_get_cq_event) before attempting to dequeue (ibv_poll_cq) any entries from the CQ. I assumed that if I did not get a completion event, there would not be a CQ entry waiting. So, no, I have not tried that. I can try that, but it may be a week or more before I am able to get back to my machines. Can I not rely on getting an event when I request signalled completions? Thanks. _______________________________________________ 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
