[ofa-general] [PATCH] ib_mthca: Pre link receive WQEs

[Eli Cohen]

Pre link receive WQEs

Pre linking of receive WQEs is required in Tavor mode. This is
required for both SRQ and regular QPs.
Remove an always true condition.
For memfree linking to the nxt wqe is moved to mthca_free_srq_wqe()
as in Tavor mode.

Signed-off-by: Eli Cohen <eli at mellanox.co.il>
Reviewed-by: Jack Morgenstein <jackm at mellanox.co.il>
---
 drivers/infiniband/hw/mthca/mthca_qp.c  |   13 ++++++++-----
 drivers/infiniband/hw/mthca/mthca_srq.c |   27 ++++++++++++++-------------
 2 files changed, 22 insertions(+), 18 deletions(-)

diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index 86aa732..7072a29 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -1175,6 +1175,7 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
 {
 	int ret;
 	int i;
+	struct mthca_next_seg *next;
 
 	qp->refcount = 1;
 	init_waitqueue_head(&qp->wait);
@@ -1217,7 +1218,6 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
 	}
 
 	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;
@@ -1240,6 +1240,13 @@ static int mthca_alloc_qp_common(struct mthca_dev *dev,
 						    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 - 1)) <<
+					      qp->rq.wqe_shift) | 1);
+		}
+
 	}
 
 	qp->sq.last = get_send_wqe(qp, qp->sq.max - 1);
@@ -1863,7 +1870,6 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 		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;
@@ -1885,9 +1891,6 @@ int mthca_tavor_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 
 		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);
 
diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c
index 553d681..af8483c 100644
--- a/drivers/infiniband/hw/mthca/mthca_srq.c
+++ b/drivers/infiniband/hw/mthca/mthca_srq.c
@@ -175,9 +175,17 @@ static int mthca_alloc_srq_buf(struct mthca_dev *dev, struct mthca_pd *pd,
 	 * 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);
@@ -470,16 +478,15 @@ out:
 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);
 
-	if (likely(srq->first_free >= 0))
-		*wqe_to_link(get_wqe(srq, srq->last_free)) = ind;
-	else
-		srq->first_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;
 
@@ -528,7 +535,6 @@ int mthca_tavor_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
 		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 */
 
@@ -549,9 +555,6 @@ int mthca_tavor_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
 		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);
 
@@ -633,8 +636,6 @@ int mthca_arbel_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
 			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 */
 
-- 
1.5.3.8