Implement posting of RDMA and atomic operations

Clean up the definitions of remote address and atomic operations WQE
segments.  Fill in the missing code that fills in these segments when
posting RDMA or atomic operations to a send queue.

Signed-off-by: Roland Dreier <rolandd@cisco.com>

2 files changed, 54 insertions, 25 deletions

diff --git a/src/qp.c b/src/qp.c
index 07e1875..81fe191 100644
--- a/src/qp.c
+++ b/src/qp.c
@@ -150,13 +150,43 @@ int mlx4_post_send(struct ibv_qp *ibqp, struct ibv_send_wr *wr,
 			switch (wr->opcode) {
 			case IBV_WR_ATOMIC_CMP_AND_SWP:
 			case IBV_WR_ATOMIC_FETCH_AND_ADD:
-				/*XXX*/
+				((struct mlx4_wqe_raddr_seg *) wqe)->raddr =
+					htonll(wr->wr.atomic.remote_addr);
+				((struct mlx4_wqe_raddr_seg *) wqe)->rkey =
+					htonl(wr->wr.atomic.rkey);
+				((struct mlx4_wqe_raddr_seg *) wqe)->reserved = 0;
+
+				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
+
+				if (wr->opcode == IBV_WR_ATOMIC_CMP_AND_SWP) {
+					((struct mlx4_wqe_atomic_seg *) wqe)->swap_add =
+						htonll(wr->wr.atomic.swap);
+					((struct mlx4_wqe_atomic_seg *) wqe)->compare =
+						htonll(wr->wr.atomic.compare_add);
+				} else {
+					((struct mlx4_wqe_atomic_seg *) wqe)->swap_add =
+						htonll(wr->wr.atomic.compare_add);
+					((struct mlx4_wqe_atomic_seg *) wqe)->compare = 0;
+				}
+
+				wqe  += sizeof (struct mlx4_wqe_atomic_seg);
+				size += (sizeof (struct mlx4_wqe_raddr_seg) +
+					 sizeof (struct mlx4_wqe_atomic_seg)) / 16;
+
 				break;
 
 			case IBV_WR_RDMA_WRITE:
 			case IBV_WR_RDMA_WRITE_WITH_IMM:
 			case IBV_WR_RDMA_READ:
-				/*XXX*/
+				((struct mlx4_wqe_raddr_seg *) wqe)->raddr =
+					htonll(wr->wr.rdma.remote_addr);
+				((struct mlx4_wqe_raddr_seg *) wqe)->rkey =
+					htonl(wr->wr.rdma.rkey);
+				((struct mlx4_wqe_raddr_seg *) wqe)->reserved = 0;
+
+				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
+				size += sizeof (struct mlx4_wqe_raddr_seg) / 16;
+
 				break;
 
 			default:
@@ -334,20 +364,20 @@ int mlx4_alloc_qp_buf(struct ibv_pd *pd, struct ibv_qp_cap *cap,
 		break;
 
 	case IBV_QPT_UC:
-		size += sizeof (struct mlx4_raddr_seg);
+		size += sizeof (struct mlx4_wqe_raddr_seg);
 		break;
 
 	case IBV_QPT_RC:
-		size += sizeof (struct mlx4_raddr_seg);
+		size += sizeof (struct mlx4_wqe_raddr_seg);
 		/*
 		 * An atomic op will require an atomic segment, a
 		 * remote address segment and one scatter entry.
 		 */
-		if (size < (sizeof (struct mlx4_atomic_seg) +
-			    sizeof (struct mlx4_raddr_seg) +
+		if (size < (sizeof (struct mlx4_wqe_atomic_seg) +
+			    sizeof (struct mlx4_wqe_raddr_seg) +
 			    sizeof (struct mlx4_wqe_data_seg)))
-			size = (sizeof (struct mlx4_atomic_seg) +
-				sizeof (struct mlx4_raddr_seg) +
+			size = (sizeof (struct mlx4_wqe_atomic_seg) +
+				sizeof (struct mlx4_wqe_raddr_seg) +
 				sizeof (struct mlx4_wqe_data_seg));
 		break;
 
@@ -356,8 +386,8 @@ int mlx4_alloc_qp_buf(struct ibv_pd *pd, struct ibv_qp_cap *cap,
 	}
 
 	/* Make sure that we have enough space for a bind request */
-	if (size < sizeof (struct mlx4_bind_seg))
-		size = sizeof (struct mlx4_bind_seg);
+	if (size < sizeof (struct mlx4_wqe_bind_seg))
+		size = sizeof (struct mlx4_wqe_bind_seg);
 
 	size += sizeof (struct mlx4_wqe_ctrl_seg);
 
diff --git a/src/wqe.h b/src/wqe.h
index b19a31a..ee649b9 100644
--- a/src/wqe.h
+++ b/src/wqe.h
@@ -96,25 +96,24 @@ struct mlx4_wqe_srq_next_seg {
 	uint32_t		reserved2[3];
 };
 
-/* XXX the rest of these are still old WQE formats... */
-struct mlx4_bind_seg {
-	uint32_t	flags;	/* [31] Atomic [30] rem write [29] rem read */
-	uint32_t	reserved;
-	uint32_t	new_rkey;
-	uint32_t	lkey;
-	uint64_t	addr;
-	uint64_t	length;
+struct mlx4_wqe_raddr_seg {
+	uint64_t		raddr;
+	uint32_t		rkey;
+	uint32_t		reserved;
 };
 
-struct mlx4_raddr_seg {
-	uint64_t	raddr;
-	uint32_t	rkey;
-	uint32_t	reserved;
+struct mlx4_wqe_atomic_seg {
+	uint64_t		swap_add;
+	uint64_t		compare;
 };
 
-struct mlx4_atomic_seg {
-	uint64_t	swap_add;
-	uint64_t	compare;
+struct mlx4_wqe_bind_seg {
+	uint32_t		flags1;
+	uint32_t		flags2;
+	uint32_t		new_rkey;
+	uint32_t		lkey;
+	uint64_t		addr;
+	uint64_t		length;
 };
 
 #endif /* WQE_H */