> Need to adjust minimum qp capability values prior to size and max resource > calculations.
Is this actually fixing a problem? I don't see how it could make a difference: > + attr->cap.max_recv_wr = attr->cap.max_recv_wr ? attr->cap.max_recv_wr : > 1; align_queue_size() always returns at least 1 so I don't see why this matters. > + attr->cap.max_recv_sge = attr->cap.max_recv_sge ? > attr->cap.max_recv_sge : 1; I don't see anything that uses max_recv_sge before it gets set in the current code. > + attr->cap.max_send_wr = attr->cap.max_send_wr ? attr->cap.max_send_wr : > 1; If max_send_wr is 0 then the call to align_queue_size will always add at least one more WQE because sq_spare_wqes will never be a power of 2. > + attr->cap.max_send_sge = attr->cap.max_send_sge ? > attr->cap.max_send_sge : 1; mlx4_calc_sq_wqe_size() will always end up with at least a 64-byte WQE size so does this matter? Oh, I guess a UD QP could end up with 0 send gather entries, but I'm not sure that's a big deal -- after all, the user gets what he asked for, and the HW shouldn't be bothered, should it? - R. _______________________________________________ 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
