IB/ib_mthca: Pre-link receive WQEs in Tavor mode

We have recently discovered that Tavor mode requires each WQE in a
posted list of receive WQEs to have a valid NDA field at all times.
This requirement holds true for regular QPs as well as for SRQs. This
patch prelinks the receive queue in a regular QP and keeps the free
list in SRQ always properly linked.

Signed-off-by: Eli Cohen <[e-mail]>
Reviewed-by: Jack Morgenstein <[e-mail]>
Signed-off-by: Roland Dreier <[e-mail]>

drivers/infiniband/hw/mthca/mthca_qp.c

···
 {
 	int ret;
 	int i;
+	struct mthca_next_seg *next;
 
 	qp->refcount = 1;
 	init_waitqueue_head(&qp->wait);
···
 	}
 
 	if (mthca_is_memfree(dev)) {
-		struct mthca_next_seg *next;
 		struct mthca_data_seg *scatter;
 		int size = (sizeof (struct mthca_next_seg) +
 			    qp->rq.max_gs * sizeof (struct mthca_data_seg)) / 16;
···
 						    qp->sq.wqe_shift) +
 						   qp->send_wqe_offset);
 		}
+	} else {
+		for (i = 0; i < qp->rq.max; ++i) {
+			next = get_recv_wqe(qp, i);
+			next->nda_op = htonl((((i + 1) % qp->rq.max) <<
+					      qp->rq.wqe_shift) | 1);
+		}
+
 	}
 
 	qp->sq.last = get_send_wqe(qp, qp->sq.max - 1);
···
 		prev_wqe = qp->rq.last;
 		qp->rq.last = wqe;
 
-		((struct mthca_next_seg *) wqe)->nda_op = 0;
 		((struct mthca_next_seg *) wqe)->ee_nds =
 			cpu_to_be32(MTHCA_NEXT_DBD);
 		((struct mthca_next_seg *) wqe)->flags = 0;
···
 
 		qp->wrid[ind] = wr->wr_id;
 
-		((struct mthca_next_seg *) prev_wqe)->nda_op =
-			cpu_to_be32((ind << qp->rq.wqe_shift) | 1);
-		wmb();
 		((struct mthca_next_seg *) prev_wqe)->ee_nds =
 			cpu_to_be32(MTHCA_NEXT_DBD | size);
 

drivers/infiniband/hw/mthca/mthca_srq.c

···
 	 * scatter list L_Keys to the sentry value of 0x100.
 	 */
 	for (i = 0; i < srq->max; ++i) {
-		wqe = get_wqe(srq, i);
+		struct mthca_next_seg *next;
 
-		*wqe_to_link(wqe) = i < srq->max - 1 ? i + 1 : -1;
+		next = wqe = get_wqe(srq, i);
+
+		if (i < srq->max - 1) {
+			*wqe_to_link(wqe) = i + 1;
+			next->nda_op = htonl(((i + 1) << srq->wqe_shift) | 1);
+		} else {
+			*wqe_to_link(wqe) = -1;
+			next->nda_op = 0;
+		}
 
 		for (scatter = wqe + sizeof (struct mthca_next_seg);
 		     (void *) scatter < wqe + (1 << srq->wqe_shift);
···
 void mthca_free_srq_wqe(struct mthca_srq *srq, u32 wqe_addr)
 {
 	int ind;
+	struct mthca_next_seg *last_free;
 
 	ind = wqe_addr >> srq->wqe_shift;
 
 	spin_lock(&srq->lock);
 
-	*wqe_to_link(get_wqe(srq, srq->last_free)) = ind;
+	last_free = get_wqe(srq, srq->last_free);
+	*wqe_to_link(last_free) = ind;
+	last_free->nda_op = htonl((ind << srq->wqe_shift) | 1);
 	*wqe_to_link(get_wqe(srq, ind)) = -1;
 	srq->last_free = ind;
 
···
 		prev_wqe  = srq->last;
 		srq->last = wqe;
 
-		((struct mthca_next_seg *) wqe)->nda_op = 0;
 		((struct mthca_next_seg *) wqe)->ee_nds = 0;
 		/* flags field will always remain 0 */
 
···
 		if (i < srq->max_gs)
 			mthca_set_data_seg_inval(wqe);
 
-		((struct mthca_next_seg *) prev_wqe)->nda_op =
-			cpu_to_be32((ind << srq->wqe_shift) | 1);
-		wmb();
 		((struct mthca_next_seg *) prev_wqe)->ee_nds =
 			cpu_to_be32(MTHCA_NEXT_DBD);
 
···
 			break;
 		}
 
-		((struct mthca_next_seg *) wqe)->nda_op =
-			cpu_to_be32((next_ind << srq->wqe_shift) | 1);
 		((struct mthca_next_seg *) wqe)->ee_nds = 0;
 		/* flags field will always remain 0 */