Merge branch 'wr-cleanup' into k.o/for-4.4

63 files changed, 1152 insertions, 988 deletions

diff --git a/drivers/infiniband/core/agent.c b/drivers/infiniband/core/agent.c
index 0429040304fd..4fa524dfb6cf 100644
--- a/drivers/infiniband/core/agent.c
+++ b/drivers/infiniband/core/agent.c
@@ -126,7 +126,7 @@ void agent_send_response(const struct ib_mad_hdr *mad_hdr, const struct ib_grh *
 		mad_send_wr = container_of(send_buf,
 					   struct ib_mad_send_wr_private,
 					   send_buf);
-		mad_send_wr->send_wr.wr.ud.port_num = port_num;
+		mad_send_wr->send_wr.port_num = port_num;
 	}
 
 	if (ib_post_send_mad(send_buf, NULL)) {
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index fa63b89e15aa..8d8af7a41a30 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -752,7 +752,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 	struct ib_device *device = mad_agent_priv->agent.device;
 	u8 port_num;
 	struct ib_wc mad_wc;
-	struct ib_send_wr *send_wr = &mad_send_wr->send_wr;
+	struct ib_ud_wr *send_wr = &mad_send_wr->send_wr;
 	size_t mad_size = port_mad_size(mad_agent_priv->qp_info->port_priv);
 	u16 out_mad_pkey_index = 0;
 	u16 drslid;
@@ -761,7 +761,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 
 	if (rdma_cap_ib_switch(device) &&
 	    smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
-		port_num = send_wr->wr.ud.port_num;
+		port_num = send_wr->port_num;
 	else
 		port_num = mad_agent_priv->agent.port_num;
 
@@ -832,9 +832,9 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 	}
 
 	build_smp_wc(mad_agent_priv->agent.qp,
-		     send_wr->wr_id, drslid,
-		     send_wr->wr.ud.pkey_index,
-		     send_wr->wr.ud.port_num, &mad_wc);
+		     send_wr->wr.wr_id, drslid,
+		     send_wr->pkey_index,
+		     send_wr->port_num, &mad_wc);
 
 	if (opa && smp->base_version == OPA_MGMT_BASE_VERSION) {
 		mad_wc.byte_len = mad_send_wr->send_buf.hdr_len
@@ -894,7 +894,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 
 	local->mad_send_wr = mad_send_wr;
 	if (opa) {
-		local->mad_send_wr->send_wr.wr.ud.pkey_index = out_mad_pkey_index;
+		local->mad_send_wr->send_wr.pkey_index = out_mad_pkey_index;
 		local->return_wc_byte_len = mad_size;
 	}
 	/* Reference MAD agent until send side of local completion handled */
@@ -1039,14 +1039,14 @@ struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
 
 	mad_send_wr->sg_list[1].lkey = mad_agent->qp->pd->local_dma_lkey;
 
-	mad_send_wr->send_wr.wr_id = (unsigned long) mad_send_wr;
-	mad_send_wr->send_wr.sg_list = mad_send_wr->sg_list;
-	mad_send_wr->send_wr.num_sge = 2;
-	mad_send_wr->send_wr.opcode = IB_WR_SEND;
-	mad_send_wr->send_wr.send_flags = IB_SEND_SIGNALED;
-	mad_send_wr->send_wr.wr.ud.remote_qpn = remote_qpn;
-	mad_send_wr->send_wr.wr.ud.remote_qkey = IB_QP_SET_QKEY;
-	mad_send_wr->send_wr.wr.ud.pkey_index = pkey_index;
+	mad_send_wr->send_wr.wr.wr_id = (unsigned long) mad_send_wr;
+	mad_send_wr->send_wr.wr.sg_list = mad_send_wr->sg_list;
+	mad_send_wr->send_wr.wr.num_sge = 2;
+	mad_send_wr->send_wr.wr.opcode = IB_WR_SEND;
+	mad_send_wr->send_wr.wr.send_flags = IB_SEND_SIGNALED;
+	mad_send_wr->send_wr.remote_qpn = remote_qpn;
+	mad_send_wr->send_wr.remote_qkey = IB_QP_SET_QKEY;
+	mad_send_wr->send_wr.pkey_index = pkey_index;
 
 	if (rmpp_active) {
 		ret = alloc_send_rmpp_list(mad_send_wr, mad_size, gfp_mask);
@@ -1151,7 +1151,7 @@ int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
 
 	/* Set WR ID to find mad_send_wr upon completion */
 	qp_info = mad_send_wr->mad_agent_priv->qp_info;
-	mad_send_wr->send_wr.wr_id = (unsigned long)&mad_send_wr->mad_list;
+	mad_send_wr->send_wr.wr.wr_id = (unsigned long)&mad_send_wr->mad_list;
 	mad_send_wr->mad_list.mad_queue = &qp_info->send_queue;
 
 	mad_agent = mad_send_wr->send_buf.mad_agent;
@@ -1179,7 +1179,7 @@ int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
 
 	spin_lock_irqsave(&qp_info->send_queue.lock, flags);
 	if (qp_info->send_queue.count < qp_info->send_queue.max_active) {
-		ret = ib_post_send(mad_agent->qp, &mad_send_wr->send_wr,
+		ret = ib_post_send(mad_agent->qp, &mad_send_wr->send_wr.wr,
 				   &bad_send_wr);
 		list = &qp_info->send_queue.list;
 	} else {
@@ -1244,7 +1244,7 @@ int ib_post_send_mad(struct ib_mad_send_buf *send_buf,
 		 * request associated with the completion
 		 */
 		next_send_buf = send_buf->next;
-		mad_send_wr->send_wr.wr.ud.ah = send_buf->ah;
+		mad_send_wr->send_wr.ah = send_buf->ah;
 
 		if (((struct ib_mad_hdr *) send_buf->mad)->mgmt_class ==
 		    IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
@@ -2457,7 +2457,7 @@ retry:
 	ib_mad_complete_send_wr(mad_send_wr, &mad_send_wc);
 
 	if (queued_send_wr) {
-		ret = ib_post_send(qp_info->qp, &queued_send_wr->send_wr,
+		ret = ib_post_send(qp_info->qp, &queued_send_wr->send_wr.wr,
 				   &bad_send_wr);
 		if (ret) {
 			dev_err(&port_priv->device->dev,
@@ -2515,7 +2515,7 @@ static void mad_error_handler(struct ib_mad_port_private *port_priv,
 			struct ib_send_wr *bad_send_wr;
 
 			mad_send_wr->retry = 0;
-			ret = ib_post_send(qp_info->qp, &mad_send_wr->send_wr,
+			ret = ib_post_send(qp_info->qp, &mad_send_wr->send_wr.wr,
 					&bad_send_wr);
 			if (ret)
 				ib_mad_send_done_handler(port_priv, wc);
@@ -2713,7 +2713,7 @@ static void local_completions(struct work_struct *work)
 			build_smp_wc(recv_mad_agent->agent.qp,
 				     (unsigned long) local->mad_send_wr,
 				     be16_to_cpu(IB_LID_PERMISSIVE),
-				     local->mad_send_wr->send_wr.wr.ud.pkey_index,
+				     local->mad_send_wr->send_wr.pkey_index,
 				     recv_mad_agent->agent.port_num, &wc);
 
 			local->mad_priv->header.recv_wc.wc = &wc;
diff --git a/drivers/infiniband/core/mad_priv.h b/drivers/infiniband/core/mad_priv.h
index 4a4f7aad0978..990698a6ab4b 100644
--- a/drivers/infiniband/core/mad_priv.h
+++ b/drivers/infiniband/core/mad_priv.h
@@ -123,7 +123,7 @@ struct ib_mad_send_wr_private {
 	struct ib_mad_send_buf send_buf;
 	u64 header_mapping;
 	u64 payload_mapping;
-	struct ib_send_wr send_wr;
+	struct ib_ud_wr send_wr;
 	struct ib_sge sg_list[IB_MAD_SEND_REQ_MAX_SG];
 	__be64 tid;
 	unsigned long timeout;
diff --git a/drivers/infiniband/core/uverbs_cmd.c b/drivers/infiniband/core/uverbs_cmd.c
index 8fd081ae9aa9..94816aeb95a0 100644
--- a/drivers/infiniband/core/uverbs_cmd.c
+++ b/drivers/infiniband/core/uverbs_cmd.c
@@ -2427,6 +2427,12 @@ ssize_t ib_uverbs_destroy_qp(struct ib_uverbs_file *file,
 	return in_len;
 }
 
+static void *alloc_wr(size_t wr_size, __u32 num_sge)
+{
+	return kmalloc(ALIGN(wr_size, sizeof (struct ib_sge)) +
+			 num_sge * sizeof (struct ib_sge), GFP_KERNEL);
+};
+
 ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
 			    struct ib_device *ib_dev,
 			    const char __user *buf, int in_len,
@@ -2475,14 +2481,83 @@ ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
 			goto out_put;
 		}
 
-		next = kmalloc(ALIGN(sizeof *next, sizeof (struct ib_sge)) +
-			       user_wr->num_sge * sizeof (struct ib_sge),
-			       GFP_KERNEL);
-		if (!next) {
-			ret = -ENOMEM;
+		if (is_ud) {
+			struct ib_ud_wr *ud;
+
+			if (user_wr->opcode != IB_WR_SEND &&
+			    user_wr->opcode != IB_WR_SEND_WITH_IMM) {
+				ret = -EINVAL;
+				goto out_put;
+			}
+
+			ud = alloc_wr(sizeof(*ud), user_wr->num_sge);
+			if (!ud) {
+				ret = -ENOMEM;
+				goto out_put;
+			}
+
+			ud->ah = idr_read_ah(user_wr->wr.ud.ah, file->ucontext);
+			if (!ud->ah) {
+				kfree(ud);
+				ret = -EINVAL;
+				goto out_put;
+			}
+			ud->remote_qpn = user_wr->wr.ud.remote_qpn;
+			ud->remote_qkey = user_wr->wr.ud.remote_qkey;
+
+			next = &ud->wr;
+		} else if (user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
+			   user_wr->opcode == IB_WR_RDMA_WRITE ||
+			   user_wr->opcode == IB_WR_RDMA_READ) {
+			struct ib_rdma_wr *rdma;
+
+			rdma = alloc_wr(sizeof(*rdma), user_wr->num_sge);
+			if (!rdma) {
+				ret = -ENOMEM;
+				goto out_put;
+			}
+
+			rdma->remote_addr = user_wr->wr.rdma.remote_addr;
+			rdma->rkey = user_wr->wr.rdma.rkey;
+
+			next = &rdma->wr;
+		} else if (user_wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+			   user_wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
+			struct ib_atomic_wr *atomic;
+
+			atomic = alloc_wr(sizeof(*atomic), user_wr->num_sge);
+			if (!atomic) {
+				ret = -ENOMEM;
+				goto out_put;
+			}
+
+			atomic->remote_addr = user_wr->wr.atomic.remote_addr;
+			atomic->compare_add = user_wr->wr.atomic.compare_add;
+			atomic->swap = user_wr->wr.atomic.swap;
+			atomic->rkey = user_wr->wr.atomic.rkey;
+
+			next = &atomic->wr;
+		} else if (user_wr->opcode == IB_WR_SEND ||
+			   user_wr->opcode == IB_WR_SEND_WITH_IMM ||
+			   user_wr->opcode == IB_WR_SEND_WITH_INV) {
+			next = alloc_wr(sizeof(*next), user_wr->num_sge);
+			if (!next) {
+				ret = -ENOMEM;
+				goto out_put;
+			}
+		} else {
+			ret = -EINVAL;
 			goto out_put;
 		}
 
+		if (user_wr->opcode == IB_WR_SEND_WITH_IMM ||
+		    user_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
+			next->ex.imm_data =
+					(__be32 __force) user_wr->ex.imm_data;
+		} else if (user_wr->opcode == IB_WR_SEND_WITH_INV) {
+			next->ex.invalidate_rkey = user_wr->ex.invalidate_rkey;
+		}
+
 		if (!last)
 			wr = next;
 		else
@@ -2495,60 +2570,6 @@ ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
 		next->opcode     = user_wr->opcode;
 		next->send_flags = user_wr->send_flags;
 
-		if (is_ud) {
-			if (next->opcode != IB_WR_SEND &&
-			    next->opcode != IB_WR_SEND_WITH_IMM) {
-				ret = -EINVAL;
-				goto out_put;
-			}
-
-			next->wr.ud.ah = idr_read_ah(user_wr->wr.ud.ah,
-						     file->ucontext);
-			if (!next->wr.ud.ah) {
-				ret = -EINVAL;
-				goto out_put;
-			}
-			next->wr.ud.remote_qpn  = user_wr->wr.ud.remote_qpn;
-			next->wr.ud.remote_qkey = user_wr->wr.ud.remote_qkey;
-			if (next->opcode == IB_WR_SEND_WITH_IMM)
-				next->ex.imm_data =
-					(__be32 __force) user_wr->ex.imm_data;
-		} else {
-			switch (next->opcode) {
-			case IB_WR_RDMA_WRITE_WITH_IMM:
-				next->ex.imm_data =
-					(__be32 __force) user_wr->ex.imm_data;
-			case IB_WR_RDMA_WRITE:
-			case IB_WR_RDMA_READ:
-				next->wr.rdma.remote_addr =
-					user_wr->wr.rdma.remote_addr;
-				next->wr.rdma.rkey        =
-					user_wr->wr.rdma.rkey;
-				break;
-			case IB_WR_SEND_WITH_IMM:
-				next->ex.imm_data =
-					(__be32 __force) user_wr->ex.imm_data;
-				break;
-			case IB_WR_SEND_WITH_INV:
-				next->ex.invalidate_rkey =
-					user_wr->ex.invalidate_rkey;
-				break;
-			case IB_WR_ATOMIC_CMP_AND_SWP:
-			case IB_WR_ATOMIC_FETCH_AND_ADD:
-				next->wr.atomic.remote_addr =
-					user_wr->wr.atomic.remote_addr;
-				next->wr.atomic.compare_add =
-					user_wr->wr.atomic.compare_add;
-				next->wr.atomic.swap = user_wr->wr.atomic.swap;
-				next->wr.atomic.rkey = user_wr->wr.atomic.rkey;
-			case IB_WR_SEND:
-				break;
-			default:
-				ret = -EINVAL;
-				goto out_put;
-			}
-		}
-
 		if (next->num_sge) {
 			next->sg_list = (void *) next +
 				ALIGN(sizeof *next, sizeof (struct ib_sge));
@@ -2582,8 +2603,8 @@ out_put:
 	put_qp_read(qp);
 
 	while (wr) {
-		if (is_ud && wr->wr.ud.ah)
-			put_ah_read(wr->wr.ud.ah);
+		if (is_ud && ud_wr(wr)->ah)
+			put_ah_read(ud_wr(wr)->ah);
 		next = wr->next;
 		kfree(wr);
 		wr = next;
diff --git a/drivers/infiniband/hw/cxgb3/iwch_qp.c b/drivers/infiniband/hw/cxgb3/iwch_qp.c
index b57c0befd962..bac0508fedd9 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_qp.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_qp.c
@@ -95,8 +95,8 @@ static int build_rdma_write(union t3_wr *wqe, struct ib_send_wr *wr,
 	wqe->write.reserved[0] = 0;
 	wqe->write.reserved[1] = 0;
 	wqe->write.reserved[2] = 0;
-	wqe->write.stag_sink = cpu_to_be32(wr->wr.rdma.rkey);
-	wqe->write.to_sink = cpu_to_be64(wr->wr.rdma.remote_addr);
+	wqe->write.stag_sink = cpu_to_be32(rdma_wr(wr)->rkey);
+	wqe->write.to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr);
 
 	if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
 		plen = 4;
@@ -137,8 +137,8 @@ static int build_rdma_read(union t3_wr *wqe, struct ib_send_wr *wr,
 		wqe->read.local_inv = 0;
 	wqe->read.reserved[0] = 0;
 	wqe->read.reserved[1] = 0;
-	wqe->read.rem_stag = cpu_to_be32(wr->wr.rdma.rkey);
-	wqe->read.rem_to = cpu_to_be64(wr->wr.rdma.remote_addr);
+	wqe->read.rem_stag = cpu_to_be32(rdma_wr(wr)->rkey);
+	wqe->read.rem_to = cpu_to_be64(rdma_wr(wr)->remote_addr);
 	wqe->read.local_stag = cpu_to_be32(wr->sg_list[0].lkey);
 	wqe->read.local_len = cpu_to_be32(wr->sg_list[0].length);
 	wqe->read.local_to = cpu_to_be64(wr->sg_list[0].addr);
@@ -146,27 +146,27 @@ static int build_rdma_read(union t3_wr *wqe, struct ib_send_wr *wr,
 	return 0;
 }
 
-static int build_fastreg(union t3_wr *wqe, struct ib_send_wr *wr,
+static int build_fastreg(union t3_wr *wqe, struct ib_send_wr *send_wr,
 				u8 *flit_cnt, int *wr_cnt, struct t3_wq *wq)
 {
+	struct ib_fast_reg_wr *wr = fast_reg_wr(send_wr);
 	int i;
 	__be64 *p;
 
-	if (wr->wr.fast_reg.page_list_len > T3_MAX_FASTREG_DEPTH)
+	if (wr->page_list_len > T3_MAX_FASTREG_DEPTH)
 		return -EINVAL;
 	*wr_cnt = 1;
-	wqe->fastreg.stag = cpu_to_be32(wr->wr.fast_reg.rkey);
-	wqe->fastreg.len = cpu_to_be32(wr->wr.fast_reg.length);
-	wqe->fastreg.va_base_hi = cpu_to_be32(wr->wr.fast_reg.iova_start >> 32);
-	wqe->fastreg.va_base_lo_fbo =
-				cpu_to_be32(wr->wr.fast_reg.iova_start & 0xffffffff);
+	wqe->fastreg.stag = cpu_to_be32(wr->rkey);
+	wqe->fastreg.len = cpu_to_be32(wr->length);
+	wqe->fastreg.va_base_hi = cpu_to_be32(wr->iova_start >> 32);
+	wqe->fastreg.va_base_lo_fbo = cpu_to_be32(wr->iova_start & 0xffffffff);
 	wqe->fastreg.page_type_perms = cpu_to_be32(
-		V_FR_PAGE_COUNT(wr->wr.fast_reg.page_list_len) |
-		V_FR_PAGE_SIZE(wr->wr.fast_reg.page_shift-12) |
+		V_FR_PAGE_COUNT(wr->page_list_len) |
+		V_FR_PAGE_SIZE(wr->page_shift-12) |
 		V_FR_TYPE(TPT_VATO) |
-		V_FR_PERMS(iwch_ib_to_tpt_access(wr->wr.fast_reg.access_flags)));
+		V_FR_PERMS(iwch_ib_to_tpt_access(wr->access_flags)));
 	p = &wqe->fastreg.pbl_addrs[0];
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++, p++) {
+	for (i = 0; i < wr->page_list_len; i++, p++) {
 
 		/* If we need a 2nd WR, then set it up */
 		if (i == T3_MAX_FASTREG_FRAG) {
@@ -175,14 +175,14 @@ static int build_fastreg(union t3_wr *wqe, struct ib_send_wr *wr,
 				Q_PTR2IDX((wq->wptr+1), wq->size_log2));
 			build_fw_riwrh((void *)wqe, T3_WR_FASTREG, 0,
 			       Q_GENBIT(wq->wptr + 1, wq->size_log2),
-			       0, 1 + wr->wr.fast_reg.page_list_len - T3_MAX_FASTREG_FRAG,
+			       0, 1 + wr->page_list_len - T3_MAX_FASTREG_FRAG,
 			       T3_EOP);
 
 			p = &wqe->pbl_frag.pbl_addrs[0];
 		}
-		*p = cpu_to_be64((u64)wr->wr.fast_reg.page_list->page_list[i]);
+		*p = cpu_to_be64((u64)wr->page_list->page_list[i]);
 	}
-	*flit_cnt = 5 + wr->wr.fast_reg.page_list_len;
+	*flit_cnt = 5 + wr->page_list_len;
 	if (*flit_cnt > 15)
 		*flit_cnt = 15;
 	return 0;
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
index acfc2f22b382..1dc9f11a4243 100644
--- a/drivers/infiniband/hw/cxgb4/qp.c
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -528,8 +528,8 @@ static int build_rdma_write(struct t4_sq *sq, union t4_wr *wqe,
 	if (wr->num_sge > T4_MAX_SEND_SGE)
 		return -EINVAL;
 	wqe->write.r2 = 0;
-	wqe->write.stag_sink = cpu_to_be32(wr->wr.rdma.rkey);
-	wqe->write.to_sink = cpu_to_be64(wr->wr.rdma.remote_addr);
+	wqe->write.stag_sink = cpu_to_be32(rdma_wr(wr)->rkey);
+	wqe->write.to_sink = cpu_to_be64(rdma_wr(wr)->remote_addr);
 	if (wr->num_sge) {
 		if (wr->send_flags & IB_SEND_INLINE) {
 			ret = build_immd(sq, wqe->write.u.immd_src, wr,
@@ -566,10 +566,10 @@ static int build_rdma_read(union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16)
 	if (wr->num_sge > 1)
 		return -EINVAL;
 	if (wr->num_sge) {
-		wqe->read.stag_src = cpu_to_be32(wr->wr.rdma.rkey);
-		wqe->read.to_src_hi = cpu_to_be32((u32)(wr->wr.rdma.remote_addr
+		wqe->read.stag_src = cpu_to_be32(rdma_wr(wr)->rkey);
+		wqe->read.to_src_hi = cpu_to_be32((u32)(rdma_wr(wr)->remote_addr
 							>> 32));
-		wqe->read.to_src_lo = cpu_to_be32((u32)wr->wr.rdma.remote_addr);
+		wqe->read.to_src_lo = cpu_to_be32((u32)rdma_wr(wr)->remote_addr);
 		wqe->read.stag_sink = cpu_to_be32(wr->sg_list[0].lkey);
 		wqe->read.plen = cpu_to_be32(wr->sg_list[0].length);
 		wqe->read.to_sink_hi = cpu_to_be32((u32)(wr->sg_list[0].addr
@@ -606,39 +606,36 @@ static int build_rdma_recv(struct c4iw_qp *qhp, union t4_recv_wr *wqe,
 }
 
 static int build_fastreg(struct t4_sq *sq, union t4_wr *wqe,
-			 struct ib_send_wr *wr, u8 *len16, u8 t5dev)
+			 struct ib_send_wr *send_wr, u8 *len16, u8 t5dev)
 {
-
+	struct ib_fast_reg_wr *wr = fast_reg_wr(send_wr);
 	struct fw_ri_immd *imdp;
 	__be64 *p;
 	int i;
-	int pbllen = roundup(wr->wr.fast_reg.page_list_len * sizeof(u64), 32);
+	int pbllen = roundup(wr->page_list_len * sizeof(u64), 32);
 	int rem;
 
-	if (wr->wr.fast_reg.page_list_len >
-	    t4_max_fr_depth(use_dsgl))
+	if (wr->page_list_len > t4_max_fr_depth(use_dsgl))
 		return -EINVAL;
 
 	wqe->fr.qpbinde_to_dcacpu = 0;
-	wqe->fr.pgsz_shift = wr->wr.fast_reg.page_shift - 12;
+	wqe->fr.pgsz_shift = wr->page_shift - 12;
 	wqe->fr.addr_type = FW_RI_VA_BASED_TO;
-	wqe->fr.mem_perms = c4iw_ib_to_tpt_access(wr->wr.fast_reg.access_flags);
+	wqe->fr.mem_perms = c4iw_ib_to_tpt_access(wr->access_flags);
 	wqe->fr.len_hi = 0;
-	wqe->fr.len_lo = cpu_to_be32(wr->wr.fast_reg.length);
-	wqe->fr.stag = cpu_to_be32(wr->wr.fast_reg.rkey);
-	wqe->fr.va_hi = cpu_to_be32(wr->wr.fast_reg.iova_start >> 32);
-	wqe->fr.va_lo_fbo = cpu_to_be32(wr->wr.fast_reg.iova_start &
-					0xffffffff);
+	wqe->fr.len_lo = cpu_to_be32(wr->length);
+	wqe->fr.stag = cpu_to_be32(wr->rkey);
+	wqe->fr.va_hi = cpu_to_be32(wr->iova_start >> 32);
+	wqe->fr.va_lo_fbo = cpu_to_be32(wr->iova_start & 0xffffffff);
 
 	if (t5dev && use_dsgl && (pbllen > max_fr_immd)) {
 		struct c4iw_fr_page_list *c4pl =
-			to_c4iw_fr_page_list(wr->wr.fast_reg.page_list);
+			to_c4iw_fr_page_list(wr->page_list);
 		struct fw_ri_dsgl *sglp;
 
-		for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
-			wr->wr.fast_reg.page_list->page_list[i] = (__force u64)
-				cpu_to_be64((u64)
-				wr->wr.fast_reg.page_list->page_list[i]);
+		for (i = 0; i < wr->page_list_len; i++) {
+			wr->page_list->page_list[i] = (__force u64)
+				cpu_to_be64((u64)wr->page_list->page_list[i]);
 		}
 
 		sglp = (struct fw_ri_dsgl *)(&wqe->fr + 1);
@@ -657,9 +654,8 @@ static int build_fastreg(struct t4_sq *sq, union t4_wr *wqe,
 		imdp->immdlen = cpu_to_be32(pbllen);
 		p = (__be64 *)(imdp + 1);
 		rem = pbllen;
-		for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
-			*p = cpu_to_be64(
-				(u64)wr->wr.fast_reg.page_list->page_list[i]);
+		for (i = 0; i < wr->page_list_len; i++) {
+			*p = cpu_to_be64((u64)wr->page_list->page_list[i]);
 			rem -= sizeof(*p);
 			if (++p == (__be64 *)&sq->queue[sq->size])
 				p = (__be64 *)sq->queue;
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 67c7120f0d52..870e56b6b25f 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -457,7 +457,8 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
 			  struct ib_grh *grh, struct ib_mad *mad)
 {
 	struct ib_sge list;
-	struct ib_send_wr wr, *bad_wr;
+	struct ib_ud_wr wr;
+	struct ib_send_wr *bad_wr;
 	struct mlx4_ib_demux_pv_ctx *tun_ctx;
 	struct mlx4_ib_demux_pv_qp *tun_qp;
 	struct mlx4_rcv_tunnel_mad *tun_mad;
@@ -582,18 +583,18 @@ int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
 	list.length = sizeof (struct mlx4_rcv_tunnel_mad);
 	list.lkey = tun_ctx->pd->local_dma_lkey;
 
-	wr.wr.ud.ah = ah;
-	wr.wr.ud.port_num = port;
-	wr.wr.ud.remote_qkey = IB_QP_SET_QKEY;
-	wr.wr.ud.remote_qpn = dqpn;
-	wr.next = NULL;
-	wr.wr_id = ((u64) tun_tx_ix) | MLX4_TUN_SET_WRID_QPN(dest_qpt);
-	wr.sg_list = &list;
-	wr.num_sge = 1;
-	wr.opcode = IB_WR_SEND;
-	wr.send_flags = IB_SEND_SIGNALED;
-
-	ret = ib_post_send(src_qp, &wr, &bad_wr);
+	wr.ah = ah;
+	wr.port_num = port;
+	wr.remote_qkey = IB_QP_SET_QKEY;
+	wr.remote_qpn = dqpn;
+	wr.wr.next = NULL;
+	wr.wr.wr_id = ((u64) tun_tx_ix) | MLX4_TUN_SET_WRID_QPN(dest_qpt);
+	wr.wr.sg_list = &list;
+	wr.wr.num_sge = 1;
+	wr.wr.opcode = IB_WR_SEND;
+	wr.wr.send_flags = IB_SEND_SIGNALED;
+
+	ret = ib_post_send(src_qp, &wr.wr, &bad_wr);
 out:
 	if (ret)
 		ib_destroy_ah(ah);
@@ -1186,7 +1187,8 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
 			 u8 *s_mac, u16 vlan_id, struct ib_mad *mad)
 {
 	struct ib_sge list;
-	struct ib_send_wr wr, *bad_wr;
+	struct ib_ud_wr wr;
+	struct ib_send_wr *bad_wr;
 	struct mlx4_ib_demux_pv_ctx *sqp_ctx;
 	struct mlx4_ib_demux_pv_qp *sqp;
 	struct mlx4_mad_snd_buf *sqp_mad;
@@ -1257,17 +1259,17 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
 	list.length = sizeof (struct mlx4_mad_snd_buf);
 	list.lkey = sqp_ctx->pd->local_dma_lkey;
 
-	wr.wr.ud.ah = ah;
-	wr.wr.ud.port_num = port;
-	wr.wr.ud.pkey_index = wire_pkey_ix;
-	wr.wr.ud.remote_qkey = qkey;
-	wr.wr.ud.remote_qpn = remote_qpn;
-	wr.next = NULL;
-	wr.wr_id = ((u64) wire_tx_ix) | MLX4_TUN_SET_WRID_QPN(src_qpnum);
-	wr.sg_list = &list;
-	wr.num_sge = 1;
-	wr.opcode = IB_WR_SEND;
-	wr.send_flags = IB_SEND_SIGNALED;
+	wr.ah = ah;
+	wr.port_num = port;
+	wr.pkey_index = wire_pkey_ix;
+	wr.remote_qkey = qkey;
+	wr.remote_qpn = remote_qpn;
+	wr.wr.next = NULL;
+	wr.wr.wr_id = ((u64) wire_tx_ix) | MLX4_TUN_SET_WRID_QPN(src_qpnum);
+	wr.wr.sg_list = &list;
+	wr.wr.num_sge = 1;
+	wr.wr.opcode = IB_WR_SEND;
+	wr.wr.send_flags = IB_SEND_SIGNALED;
 	if (s_mac)
 		memcpy(to_mah(ah)->av.eth.s_mac, s_mac, 6);
 	if (vlan_id < 0x1000)
@@ -1275,7 +1277,7 @@ int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
 	to_mah(ah)->av.eth.vlan = cpu_to_be16(vlan_id);
 
 
-	ret = ib_post_send(send_qp, &wr, &bad_wr);
+	ret = ib_post_send(send_qp, &wr.wr, &bad_wr);
 out:
 	if (ret)
 		ib_destroy_ah(ah);
diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index 2542fd3c1a49..5bba176e9dfa 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -321,21 +321,21 @@ err_free:
 int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
 		    struct ib_mw_bind *mw_bind)
 {
-	struct ib_send_wr  wr;
+	struct ib_bind_mw_wr  wr;
 	struct ib_send_wr *bad_wr;
 	int ret;
 
 	memset(&wr, 0, sizeof(wr));
-	wr.opcode               = IB_WR_BIND_MW;
-	wr.wr_id                = mw_bind->wr_id;
-	wr.send_flags           = mw_bind->send_flags;
-	wr.wr.bind_mw.mw        = mw;
-	wr.wr.bind_mw.bind_info = mw_bind->bind_info;
-	wr.wr.bind_mw.rkey      = ib_inc_rkey(mw->rkey);
-
-	ret = mlx4_ib_post_send(qp, &wr, &bad_wr);
+	wr.wr.opcode		= IB_WR_BIND_MW;
+	wr.wr.wr_id		= mw_bind->wr_id;
+	wr.wr.send_flags	= mw_bind->send_flags;
+	wr.mw			= mw;
+	wr.bind_info		= mw_bind->bind_info;
+	wr.rkey			= ib_inc_rkey(mw->rkey);
+
+	ret = mlx4_ib_post_send(qp, &wr.wr, &bad_wr);
 	if (!ret)
-		mw->rkey = wr.wr.bind_mw.rkey;
+		mw->rkey = wr.rkey;
 
 	return ret;
 }
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 42a051b088e4..f2b2a61898f8 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -2133,14 +2133,14 @@ static int vf_get_qp0_qkey(struct mlx4_dev *dev, int qpn, u32 *qkey)
 }
 
 static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
-				  struct ib_send_wr *wr,
+				  struct ib_ud_wr *wr,
 				  void *wqe, unsigned *mlx_seg_len)
 {
 	struct mlx4_ib_dev *mdev = to_mdev(sqp->qp.ibqp.device);
 	struct ib_device *ib_dev = &mdev->ib_dev;
 	struct mlx4_wqe_mlx_seg *mlx = wqe;
 	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
-	struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+	struct mlx4_ib_ah *ah = to_mah(wr->ah);
 	u16 pkey;
 	u32 qkey;
 	int send_size;
@@ -2148,13 +2148,13 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
 	int spc;
 	int i;
 
-	if (wr->opcode != IB_WR_SEND)
+	if (wr->wr.opcode != IB_WR_SEND)
 		return -EINVAL;
 
 	send_size = 0;
 
-	for (i = 0; i < wr->num_sge; ++i)
-		send_size += wr->sg_list[i].length;
+	for (i = 0; i < wr->wr.num_sge; ++i)
+		send_size += wr->wr.sg_list[i].length;
 
 	/* for proxy-qp0 sends, need to add in size of tunnel header */
 	/* for tunnel-qp0 sends, tunnel header is already in s/g list */
@@ -2179,11 +2179,11 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
 	mlx->rlid = sqp->ud_header.lrh.destination_lid;
 
 	sqp->ud_header.lrh.virtual_lane    = 0;
-	sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED);
+	sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
 	ib_get_cached_pkey(ib_dev, sqp->qp.port, 0, &pkey);
 	sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
 	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER)
-		sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+		sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->remote_qpn);
 	else
 		sqp->ud_header.bth.destination_qpn =
 			cpu_to_be32(mdev->dev->caps.qp0_tunnel[sqp->qp.port - 1]);
@@ -2255,14 +2255,14 @@ static void mlx4_u64_to_smac(u8 *dst_mac, u64 src_mac)
 	}
 }
 
-static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
+static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_ud_wr *wr,
 			    void *wqe, unsigned *mlx_seg_len)
 {
 	struct ib_device *ib_dev = sqp->qp.ibqp.device;
 	struct mlx4_wqe_mlx_seg *mlx = wqe;
 	struct mlx4_wqe_ctrl_seg *ctrl = wqe;
 	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
-	struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+	struct mlx4_ib_ah *ah = to_mah(wr->ah);
 	union ib_gid sgid;
 	u16 pkey;
 	int send_size;
@@ -2276,8 +2276,8 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 	bool is_grh;
 
 	send_size = 0;
-	for (i = 0; i < wr->num_sge; ++i)
-		send_size += wr->sg_list[i].length;
+	for (i = 0; i < wr->wr.num_sge; ++i)
+		send_size += wr->wr.sg_list[i].length;
 
 	is_eth = rdma_port_get_link_layer(sqp->qp.ibqp.device, sqp->qp.port) == IB_LINK_LAYER_ETHERNET;
 	is_grh = mlx4_ib_ah_grh_present(ah);
@@ -2357,7 +2357,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 		mlx->rlid = sqp->ud_header.lrh.destination_lid;
 	}
 
-	switch (wr->opcode) {
+	switch (wr->wr.opcode) {
 	case IB_WR_SEND:
 		sqp->ud_header.bth.opcode	 = IB_OPCODE_UD_SEND_ONLY;
 		sqp->ud_header.immediate_present = 0;
@@ -2365,7 +2365,7 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 	case IB_WR_SEND_WITH_IMM:
 		sqp->ud_header.bth.opcode	 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
 		sqp->ud_header.immediate_present = 1;
-		sqp->ud_header.immediate_data    = wr->ex.imm_data;
+		sqp->ud_header.immediate_data    = wr->wr.ex.imm_data;
 		break;
 	default:
 		return -EINVAL;
@@ -2408,16 +2408,16 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 		if (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE)
 			sqp->ud_header.lrh.source_lid = IB_LID_PERMISSIVE;
 	}
-	sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED);
+	sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
 	if (!sqp->qp.ibqp.qp_num)
 		ib_get_cached_pkey(ib_dev, sqp->qp.port, sqp->pkey_index, &pkey);
 	else
-		ib_get_cached_pkey(ib_dev, sqp->qp.port, wr->wr.ud.pkey_index, &pkey);
+		ib_get_cached_pkey(ib_dev, sqp->qp.port, wr->pkey_index, &pkey);
 	sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
-	sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+	sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->remote_qpn);
 	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
-	sqp->ud_header.deth.qkey = cpu_to_be32(wr->wr.ud.remote_qkey & 0x80000000 ?
-					       sqp->qkey : wr->wr.ud.remote_qkey);
+	sqp->ud_header.deth.qkey = cpu_to_be32(wr->remote_qkey & 0x80000000 ?
+					       sqp->qkey : wr->remote_qkey);
 	sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.ibqp.qp_num);
 
 	header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf);
@@ -2505,43 +2505,45 @@ static __be32 convert_access(int acc)
 		cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_READ);
 }
 
-static void set_fmr_seg(struct mlx4_wqe_fmr_seg *fseg, struct ib_send_wr *wr)
+static void set_fmr_seg(struct mlx4_wqe_fmr_seg *fseg,
+		struct ib_fast_reg_wr *wr)
 {
-	struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(wr->wr.fast_reg.page_list);
+	struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(wr->page_list);
 	int i;
 
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; ++i)
+	for (i = 0; i < wr->page_list_len; ++i)
 		mfrpl->mapped_page_list[i] =
-			cpu_to_be64(wr->wr.fast_reg.page_list->page_list[i] |
+			cpu_to_be64(wr->page_list->page_list[i] |
 				    MLX4_MTT_FLAG_PRESENT);
 
-	fseg->flags		= convert_access(wr->wr.fast_reg.access_flags);
-	fseg->mem_key		= cpu_to_be32(wr->wr.fast_reg.rkey);
+	fseg->flags		= convert_access(wr->access_flags);
+	fseg->mem_key		= cpu_to_be32(wr->rkey);
 	fseg->buf_list		= cpu_to_be64(mfrpl->map);
-	fseg->start_addr	= cpu_to_be64(wr->wr.fast_reg.iova_start);
-	fseg->reg_len		= cpu_to_be64(wr->wr.fast_reg.length);
+	fseg->start_addr	= cpu_to_be64(wr->iova_start);
+	fseg->reg_len		= cpu_to_be64(wr->length);
 	fseg->offset		= 0; /* XXX -- is this just for ZBVA? */
-	fseg->page_size		= cpu_to_be32(wr->wr.fast_reg.page_shift);
+	fseg->page_size		= cpu_to_be32(wr->page_shift);
 	fseg->reserved[0]	= 0;
 	fseg->reserved[1]	= 0;
 }
 
-static void set_bind_seg(struct mlx4_wqe_bind_seg *bseg, struct ib_send_wr *wr)
+static void set_bind_seg(struct mlx4_wqe_bind_seg *bseg,
+		struct ib_bind_mw_wr *wr)
 {
 	bseg->flags1 =
-		convert_access(wr->wr.bind_mw.bind_info.mw_access_flags) &
+		convert_access(wr->bind_info.mw_access_flags) &
 		cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_READ  |
 			    MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_WRITE |
 			    MLX4_WQE_FMR_AND_BIND_PERM_ATOMIC);
 	bseg->flags2 = 0;
-	if (wr->wr.bind_mw.mw->type == IB_MW_TYPE_2)
+	if (wr->mw->type == IB_MW_TYPE_2)
 		bseg->flags2 |= cpu_to_be32(MLX4_WQE_BIND_TYPE_2);
-	if (wr->wr.bind_mw.bind_info.mw_access_flags & IB_ZERO_BASED)
+	if (wr->bind_info.mw_access_flags & IB_ZERO_BASED)
 		bseg->flags2 |= cpu_to_be32(MLX4_WQE_BIND_ZERO_BASED);
-	bseg->new_rkey = cpu_to_be32(wr->wr.bind_mw.rkey);
-	bseg->lkey = cpu_to_be32(wr->wr.bind_mw.bind_info.mr->lkey);
-	bseg->addr = cpu_to_be64(wr->wr.bind_mw.bind_info.addr);
-	bseg->length = cpu_to_be64(wr->wr.bind_mw.bind_info.length);
+	bseg->new_rkey = cpu_to_be32(wr->rkey);
+	bseg->lkey = cpu_to_be32(wr->bind_info.mr->lkey);
+	bseg->addr = cpu_to_be64(wr->bind_info.addr);
+	bseg->length = cpu_to_be64(wr->bind_info.length);
 }
 
 static void set_local_inv_seg(struct mlx4_wqe_local_inval_seg *iseg, u32 rkey)
@@ -2558,46 +2560,47 @@ static __always_inline void set_raddr_seg(struct mlx4_wqe_raddr_seg *rseg,
 	rseg->reserved = 0;
 }
 
-static void set_atomic_seg(struct mlx4_wqe_atomic_seg *aseg, struct ib_send_wr *wr)
+static void set_atomic_seg(struct mlx4_wqe_atomic_seg *aseg,
+		struct ib_atomic_wr *wr)
 {
-	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
-		aseg->swap_add = cpu_to_be64(wr->wr.atomic.swap);
-		aseg->compare  = cpu_to_be64(wr->wr.atomic.compare_add);
-	} else if (wr->opcode == IB_WR_MASKED_ATOMIC_FETCH_AND_ADD) {
-		aseg->swap_add = cpu_to_be64(wr->wr.atomic.compare_add);
-		aseg->compare  = cpu_to_be64(wr->wr.atomic.compare_add_mask);
+	if (wr->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+		aseg->swap_add = cpu_to_be64(wr->swap);
+		aseg->compare  = cpu_to_be64(wr->compare_add);
+	} else if (wr->wr.opcode == IB_WR_MASKED_ATOMIC_FETCH_AND_ADD) {
+		aseg->swap_add = cpu_to_be64(wr->compare_add);
+		aseg->compare  = cpu_to_be64(wr->compare_add_mask);
 	} else {
-		aseg->swap_add = cpu_to_be64(wr->wr.atomic.compare_add);
+		aseg->swap_add = cpu_to_be64(wr->compare_add);
 		aseg->compare  = 0;
 	}
 
 }
 
 static void set_masked_atomic_seg(struct mlx4_wqe_masked_atomic_seg *aseg,
-				  struct ib_send_wr *wr)
+				  struct ib_atomic_wr *wr)
 {
-	aseg->swap_add		= cpu_to_be64(wr->wr.atomic.swap);
-	aseg->swap_add_mask	= cpu_to_be64(wr->wr.atomic.swap_mask);
-	aseg->compare		= cpu_to_be64(wr->wr.atomic.compare_add);
-	aseg->compare_mask	= cpu_to_be64(wr->wr.atomic.compare_add_mask);
+	aseg->swap_add		= cpu_to_be64(wr->swap);
+	aseg->swap_add_mask	= cpu_to_be64(wr->swap_mask);
+	aseg->compare		= cpu_to_be64(wr->compare_add);
+	aseg->compare_mask	= cpu_to_be64(wr->compare_add_mask);
 }
 
 static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg,
-			     struct ib_send_wr *wr)
+			     struct ib_ud_wr *wr)
 {
-	memcpy(dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof (struct mlx4_av));
-	dseg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn);
-	dseg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
-	dseg->vlan = to_mah(wr->wr.ud.ah)->av.eth.vlan;
-	memcpy(dseg->mac, to_mah(wr->wr.ud.ah)->av.eth.mac, 6);
+	memcpy(dseg->av, &to_mah(wr->ah)->av, sizeof (struct mlx4_av));
+	dseg->dqpn = cpu_to_be32(wr->remote_qpn);
+	dseg->qkey = cpu_to_be32(wr->remote_qkey);
+	dseg->vlan = to_mah(wr->ah)->av.eth.vlan;
+	memcpy(dseg->mac, to_mah(wr->ah)->av.eth.mac, 6);
 }
 
 static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev,
 				    struct mlx4_wqe_datagram_seg *dseg,
-				    struct ib_send_wr *wr,
+				    struct ib_ud_wr *wr,
 				    enum mlx4_ib_qp_type qpt)
 {
-	union mlx4_ext_av *av = &to_mah(wr->wr.ud.ah)->av;
+	union mlx4_ext_av *av = &to_mah(wr->ah)->av;
 	struct mlx4_av sqp_av = {0};
 	int port = *((u8 *) &av->ib.port_pd) & 0x3;
 
@@ -2616,18 +2619,18 @@ static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev,
 	dseg->qkey = cpu_to_be32(IB_QP_SET_QKEY);
 }
 
-static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len)
+static void build_tunnel_header(struct ib_ud_wr *wr, void *wqe, unsigned *mlx_seg_len)
 {
 	struct mlx4_wqe_inline_seg *inl = wqe;
 	struct mlx4_ib_tunnel_header hdr;
-	struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+	struct mlx4_ib_ah *ah = to_mah(wr->ah);
 	int spc;
 	int i;
 
 	memcpy(&hdr.av, &ah->av, sizeof hdr.av);
-	hdr.remote_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
-	hdr.pkey_index = cpu_to_be16(wr->wr.ud.pkey_index);
-	hdr.qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+	hdr.remote_qpn = cpu_to_be32(wr->remote_qpn);
+	hdr.pkey_index = cpu_to_be16(wr->pkey_index);
+	hdr.qkey = cpu_to_be32(wr->remote_qkey);
 	memcpy(hdr.mac, ah->av.eth.mac, 6);
 	hdr.vlan = ah->av.eth.vlan;
 
@@ -2699,22 +2702,22 @@ static void __set_data_seg(struct mlx4_wqe_data_seg *dseg, struct ib_sge *sg)
 	dseg->addr       = cpu_to_be64(sg->addr);
 }
 
-static int build_lso_seg(struct mlx4_wqe_lso_seg *wqe, struct ib_send_wr *wr,
+static int build_lso_seg(struct mlx4_wqe_lso_seg *wqe, struct ib_ud_wr *wr,
 			 struct mlx4_ib_qp *qp, unsigned *lso_seg_len,
 			 __be32 *lso_hdr_sz, __be32 *blh)
 {
-	unsigned halign = ALIGN(sizeof *wqe + wr->wr.ud.hlen, 16);
+	unsigned halign = ALIGN(sizeof *wqe + wr->hlen, 16);
 
 	if (unlikely(halign > MLX4_IB_CACHE_LINE_SIZE))
 		*blh = cpu_to_be32(1 << 6);
 
 	if (unlikely(!(qp->flags & MLX4_IB_QP_LSO) &&
-		     wr->num_sge > qp->sq.max_gs - (halign >> 4)))
+		     wr->wr.num_sge > qp->sq.max_gs - (halign >> 4)))
 		return -EINVAL;
 
-	memcpy(wqe->header, wr->wr.ud.header, wr->wr.ud.hlen);
+	memcpy(wqe->header, wr->header, wr->hlen);
 
-	*lso_hdr_sz  = cpu_to_be32(wr->wr.ud.mss << 16 | wr->wr.ud.hlen);
+	*lso_hdr_sz  = cpu_to_be32(wr->mss << 16 | wr->hlen);
 	*lso_seg_len = halign;
 	return 0;
 }
@@ -2813,11 +2816,11 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_ATOMIC_CMP_AND_SWP:
 			case IB_WR_ATOMIC_FETCH_AND_ADD:
 			case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
-				set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
-					      wr->wr.atomic.rkey);
+				set_raddr_seg(wqe, atomic_wr(wr)->remote_addr,
+					      atomic_wr(wr)->rkey);
 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
 
-				set_atomic_seg(wqe, wr);
+				set_atomic_seg(wqe, atomic_wr(wr));
 				wqe  += sizeof (struct mlx4_wqe_atomic_seg);
 
 				size += (sizeof (struct mlx4_wqe_raddr_seg) +
@@ -2826,11 +2829,11 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 				break;
 
 			case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
-				set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
-					      wr->wr.atomic.rkey);
+				set_raddr_seg(wqe, atomic_wr(wr)->remote_addr,
+					      atomic_wr(wr)->rkey);
 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
 
-				set_masked_atomic_seg(wqe, wr);
+				set_masked_atomic_seg(wqe, atomic_wr(wr));
 				wqe  += sizeof (struct mlx4_wqe_masked_atomic_seg);
 
 				size += (sizeof (struct mlx4_wqe_raddr_seg) +
@@ -2841,8 +2844,8 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_RDMA_READ:
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
-				set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(wqe, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
 				size += sizeof (struct mlx4_wqe_raddr_seg) / 16;
 				break;
@@ -2858,7 +2861,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_FAST_REG_MR:
 				ctrl->srcrb_flags |=
 					cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
-				set_fmr_seg(wqe, wr);
+				set_fmr_seg(wqe, fast_reg_wr(wr));
 				wqe  += sizeof (struct mlx4_wqe_fmr_seg);
 				size += sizeof (struct mlx4_wqe_fmr_seg) / 16;
 				break;
@@ -2866,7 +2869,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_BIND_MW:
 				ctrl->srcrb_flags |=
 					cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
-				set_bind_seg(wqe, wr);
+				set_bind_seg(wqe, bind_mw_wr(wr));
 				wqe  += sizeof(struct mlx4_wqe_bind_seg);
 				size += sizeof(struct mlx4_wqe_bind_seg) / 16;
 				break;
@@ -2877,7 +2880,8 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			break;
 
 		case MLX4_IB_QPT_TUN_SMI_OWNER:
-			err =  build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen);
+			err =  build_sriov_qp0_header(to_msqp(qp), ud_wr(wr),
+					ctrl, &seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
@@ -2888,19 +2892,20 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		case MLX4_IB_QPT_TUN_SMI:
 		case MLX4_IB_QPT_TUN_GSI:
 			/* this is a UD qp used in MAD responses to slaves. */
-			set_datagram_seg(wqe, wr);
+			set_datagram_seg(wqe, ud_wr(wr));
 			/* set the forced-loopback bit in the data seg av */
 			*(__be32 *) wqe |= cpu_to_be32(0x80000000);
 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
 			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
 			break;
 		case MLX4_IB_QPT_UD:
-			set_datagram_seg(wqe, wr);
+			set_datagram_seg(wqe, ud_wr(wr));
 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
 			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
 
 			if (wr->opcode == IB_WR_LSO) {
-				err = build_lso_seg(wqe, wr, qp, &seglen, &lso_hdr_sz, &blh);
+				err = build_lso_seg(wqe, ud_wr(wr), qp, &seglen,
+						&lso_hdr_sz, &blh);
 				if (unlikely(err)) {
 					*bad_wr = wr;
 					goto out;
@@ -2912,7 +2917,8 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			break;
 
 		case MLX4_IB_QPT_PROXY_SMI_OWNER:
-			err = build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen);
+			err = build_sriov_qp0_header(to_msqp(qp), ud_wr(wr),
+					ctrl, &seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
@@ -2923,7 +2929,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			add_zero_len_inline(wqe);
 			wqe += 16;
 			size++;
-			build_tunnel_header(wr, wqe, &seglen);
+			build_tunnel_header(ud_wr(wr), wqe, &seglen);
 			wqe  += seglen;
 			size += seglen / 16;
 			break;
@@ -2933,18 +2939,20 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			 * In this case we first add a UD segment targeting
 			 * the tunnel qp, and then add a header with address
 			 * information */
-			set_tunnel_datagram_seg(to_mdev(ibqp->device), wqe, wr,
+			set_tunnel_datagram_seg(to_mdev(ibqp->device), wqe,
+						ud_wr(wr),
 						qp->mlx4_ib_qp_type);
 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
 			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
-			build_tunnel_header(wr, wqe, &seglen);
+			build_tunnel_header(ud_wr(wr), wqe, &seglen);
 			wqe  += seglen;
 			size += seglen / 16;
 			break;
 
 		case MLX4_IB_QPT_SMI:
 		case MLX4_IB_QPT_GSI:
-			err = build_mlx_header(to_msqp(qp), wr, ctrl, &seglen);
+			err = build_mlx_header(to_msqp(qp), ud_wr(wr), ctrl,
+					&seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
 				goto out;
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
index 22123b79d550..29f3ecdbe790 100644
--- a/drivers/infiniband/hw/mlx5/mlx5_ib.h
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -245,6 +245,7 @@ enum mlx5_ib_qp_flags {
 };
 
 struct mlx5_umr_wr {
+	struct ib_send_wr		wr;
 	union {
 		u64			virt_addr;
 		u64			offset;
@@ -257,6 +258,11 @@ struct mlx5_umr_wr {
 	u32				mkey;
 };
 
+static inline struct mlx5_umr_wr *umr_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct mlx5_umr_wr, wr);
+}
+
 struct mlx5_shared_mr_info {
 	int mr_id;
 	struct ib_umem		*umem;
diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index 54a15b5d336d..b30d4ae0fb61 100644
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -687,7 +687,7 @@ static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr,
 			     int access_flags)
 {
 	struct mlx5_ib_dev *dev = to_mdev(pd->device);
-	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+	struct mlx5_umr_wr *umrwr = umr_wr(wr);
 
 	sg->addr = dma;
 	sg->length = ALIGN(sizeof(u64) * n, 64);
@@ -715,7 +715,7 @@ static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr,
 static void prep_umr_unreg_wqe(struct mlx5_ib_dev *dev,
 			       struct ib_send_wr *wr, u32 key)
 {
-	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+	struct mlx5_umr_wr *umrwr = umr_wr(wr);
 
 	wr->send_flags = MLX5_IB_SEND_UMR_UNREG | MLX5_IB_SEND_UMR_FAIL_IF_FREE;
 	wr->opcode = MLX5_IB_WR_UMR;
@@ -752,7 +752,8 @@ static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem,
 	struct device *ddev = dev->ib_dev.dma_device;
 	struct umr_common *umrc = &dev->umrc;
 	struct mlx5_ib_umr_context umr_context;
-	struct ib_send_wr wr, *bad;
+	struct mlx5_umr_wr umrwr;
+	struct ib_send_wr *bad;
 	struct mlx5_ib_mr *mr;
 	struct ib_sge sg;
 	int size;
@@ -798,14 +799,14 @@ static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem,
 		goto free_pas;
 	}
 
-	memset(&wr, 0, sizeof(wr));
-	wr.wr_id = (u64)(unsigned long)&umr_context;
-	prep_umr_reg_wqe(pd, &wr, &sg, dma, npages, mr->mmr.key, page_shift,
-			 virt_addr, len, access_flags);
+	memset(&umrwr, 0, sizeof(umrwr));
+	umrwr.wr.wr_id = (u64)(unsigned long)&umr_context;
+	prep_umr_reg_wqe(pd, &umrwr.wr, &sg, dma, npages, mr->mmr.key,
+			 page_shift, virt_addr, len, access_flags);
 
 	mlx5_ib_init_umr_context(&umr_context);
 	down(&umrc->sem);
-	err = ib_post_send(umrc->qp, &wr, &bad);
+	err = ib_post_send(umrc->qp, &umrwr.wr, &bad);
 	if (err) {
 		mlx5_ib_warn(dev, "post send failed, err %d\n", err);
 		goto unmap_dma;
@@ -851,8 +852,8 @@ int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index, int npages,
 	int size;
 	__be64 *pas;
 	dma_addr_t dma;
-	struct ib_send_wr wr, *bad;
-	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr.wr.fast_reg;
+	struct ib_send_wr *bad;
+	struct mlx5_umr_wr wr;
 	struct ib_sge sg;
 	int err = 0;
 	const int page_index_alignment = MLX5_UMR_MTT_ALIGNMENT / sizeof(u64);
@@ -917,26 +918,26 @@ int mlx5_ib_update_mtt(struct mlx5_ib_mr *mr, u64 start_page_index, int npages,
 		dma_sync_single_for_device(ddev, dma, size, DMA_TO_DEVICE);
 
 		memset(&wr, 0, sizeof(wr));
-		wr.wr_id = (u64)(unsigned long)&umr_context;
+		wr.wr.wr_id = (u64)(unsigned long)&umr_context;
 
 		sg.addr = dma;
 		sg.length = ALIGN(npages * sizeof(u64),
 				MLX5_UMR_MTT_ALIGNMENT);
 		sg.lkey = dev->umrc.pd->local_dma_lkey;
 
-		wr.send_flags = MLX5_IB_SEND_UMR_FAIL_IF_FREE |
+		wr.wr.send_flags = MLX5_IB_SEND_UMR_FAIL_IF_FREE |
 				MLX5_IB_SEND_UMR_UPDATE_MTT;
-		wr.sg_list = &sg;
-		wr.num_sge = 1;
-		wr.opcode = MLX5_IB_WR_UMR;
-		umrwr->npages = sg.length / sizeof(u64);
-		umrwr->page_shift = PAGE_SHIFT;
-		umrwr->mkey = mr->mmr.key;
-		umrwr->target.offset = start_page_index;
+		wr.wr.sg_list = &sg;
+		wr.wr.num_sge = 1;
+		wr.wr.opcode = MLX5_IB_WR_UMR;
+		wr.npages = sg.length / sizeof(u64);
+		wr.page_shift = PAGE_SHIFT;
+		wr.mkey = mr->mmr.key;
+		wr.target.offset = start_page_index;
 
 		mlx5_ib_init_umr_context(&umr_context);
 		down(&umrc->sem);
-		err = ib_post_send(umrc->qp, &wr, &bad);
+		err = ib_post_send(umrc->qp, &wr.wr, &bad);
 		if (err) {
 			mlx5_ib_err(dev, "UMR post send failed, err %d\n", err);
 		} else {
@@ -1122,16 +1123,17 @@ static int unreg_umr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
 {
 	struct umr_common *umrc = &dev->umrc;
 	struct mlx5_ib_umr_context umr_context;
-	struct ib_send_wr wr, *bad;
+	struct mlx5_umr_wr umrwr;
+	struct ib_send_wr *bad;
 	int err;
 
-	memset(&wr, 0, sizeof(wr));
-	wr.wr_id = (u64)(unsigned long)&umr_context;
-	prep_umr_unreg_wqe(dev, &wr, mr->mmr.key);
+	memset(&umrwr.wr, 0, sizeof(umrwr));
+	umrwr.wr.wr_id = (u64)(unsigned long)&umr_context;
+	prep_umr_unreg_wqe(dev, &umrwr.wr, mr->mmr.key);
 
 	mlx5_ib_init_umr_context(&umr_context);
 	down(&umrc->sem);
-	err = ib_post_send(umrc->qp, &wr, &bad);
+	err = ib_post_send(umrc->qp, &umrwr.wr, &bad);
 	if (err) {
 		up(&umrc->sem);
 		mlx5_ib_dbg(dev, "err %d\n", err);
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 6f521a3418e8..9bad68820061 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -1838,9 +1838,9 @@ static __always_inline void set_raddr_seg(struct mlx5_wqe_raddr_seg *rseg,
 static void set_datagram_seg(struct mlx5_wqe_datagram_seg *dseg,
 			     struct ib_send_wr *wr)
 {
-	memcpy(&dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof(struct mlx5_av));
-	dseg->av.dqp_dct = cpu_to_be32(wr->wr.ud.remote_qpn | MLX5_EXTENDED_UD_AV);
-	dseg->av.key.qkey.qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+	memcpy(&dseg->av, &to_mah(ud_wr(wr)->ah)->av, sizeof(struct mlx5_av));
+	dseg->av.dqp_dct = cpu_to_be32(ud_wr(wr)->remote_qpn | MLX5_EXTENDED_UD_AV);
+	dseg->av.key.qkey.qkey = cpu_to_be32(ud_wr(wr)->remote_qkey);
 }
 
 static void set_data_ptr_seg(struct mlx5_wqe_data_seg *dseg, struct ib_sge *sg)
@@ -1908,7 +1908,7 @@ static void set_frwr_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
 	}
 
 	umr->flags = (1 << 5); /* fail if not free */
-	umr->klm_octowords = get_klm_octo(wr->wr.fast_reg.page_list_len);
+	umr->klm_octowords = get_klm_octo(fast_reg_wr(wr)->page_list_len);
 	umr->mkey_mask = frwr_mkey_mask();
 }
 
@@ -1952,7 +1952,7 @@ static __be64 get_umr_update_mtt_mask(void)
 static void set_reg_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
 				struct ib_send_wr *wr)
 {
-	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+	struct mlx5_umr_wr *umrwr = umr_wr(wr);
 
 	memset(umr, 0, sizeof(*umr));
 
@@ -1996,20 +1996,20 @@ static void set_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr,
 		return;
 	}
 
-	seg->flags = get_umr_flags(wr->wr.fast_reg.access_flags) |
+	seg->flags = get_umr_flags(fast_reg_wr(wr)->access_flags) |
 		     MLX5_ACCESS_MODE_MTT;
 	*writ = seg->flags & (MLX5_PERM_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE);
-	seg->qpn_mkey7_0 = cpu_to_be32((wr->wr.fast_reg.rkey & 0xff) | 0xffffff00);
+	seg->qpn_mkey7_0 = cpu_to_be32((fast_reg_wr(wr)->rkey & 0xff) | 0xffffff00);
 	seg->flags_pd = cpu_to_be32(MLX5_MKEY_REMOTE_INVAL);
-	seg->start_addr = cpu_to_be64(wr->wr.fast_reg.iova_start);
-	seg->len = cpu_to_be64(wr->wr.fast_reg.length);
-	seg->xlt_oct_size = cpu_to_be32((wr->wr.fast_reg.page_list_len + 1) / 2);
-	seg->log2_page_size = wr->wr.fast_reg.page_shift;
+	seg->start_addr = cpu_to_be64(fast_reg_wr(wr)->iova_start);
+	seg->len = cpu_to_be64(fast_reg_wr(wr)->length);
+	seg->xlt_oct_size = cpu_to_be32((fast_reg_wr(wr)->page_list_len + 1) / 2);
+	seg->log2_page_size = fast_reg_wr(wr)->page_shift;
 }
 
 static void set_reg_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr)
 {
-	struct mlx5_umr_wr *umrwr = (struct mlx5_umr_wr *)&wr->wr.fast_reg;
+	struct mlx5_umr_wr *umrwr = umr_wr(wr);
 
 	memset(seg, 0, sizeof(*seg));
 	if (wr->send_flags & MLX5_IB_SEND_UMR_UNREG) {
@@ -2034,15 +2034,15 @@ static void set_frwr_pages(struct mlx5_wqe_data_seg *dseg,
 			   struct mlx5_ib_pd *pd,
 			   int writ)
 {
-	struct mlx5_ib_fast_reg_page_list *mfrpl = to_mfrpl(wr->wr.fast_reg.page_list);
-	u64 *page_list = wr->wr.fast_reg.page_list->page_list;
+	struct mlx5_ib_fast_reg_page_list *mfrpl = to_mfrpl(fast_reg_wr(wr)->page_list);
+	u64 *page_list = fast_reg_wr(wr)->page_list->page_list;
 	u64 perm = MLX5_EN_RD | (writ ? MLX5_EN_WR : 0);
 	int i;
 
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++)
+	for (i = 0; i < fast_reg_wr(wr)->page_list_len; i++)
 		mfrpl->mapped_page_list[i] = cpu_to_be64(page_list[i] | perm);
 	dseg->addr = cpu_to_be64(mfrpl->map);
-	dseg->byte_count = cpu_to_be32(ALIGN(sizeof(u64) * wr->wr.fast_reg.page_list_len, 64));
+	dseg->byte_count = cpu_to_be32(ALIGN(sizeof(u64) * fast_reg_wr(wr)->page_list_len, 64));
 	dseg->lkey = cpu_to_be32(pd->ibpd.local_dma_lkey);
 }
 
@@ -2224,22 +2224,22 @@ static int mlx5_set_bsf(struct ib_mr *sig_mr,
 	return 0;
 }
 
-static int set_sig_data_segment(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
-				void **seg, int *size)
+static int set_sig_data_segment(struct ib_sig_handover_wr *wr,
+				struct mlx5_ib_qp *qp, void **seg, int *size)
 {
-	struct ib_sig_attrs *sig_attrs = wr->wr.sig_handover.sig_attrs;
-	struct ib_mr *sig_mr = wr->wr.sig_handover.sig_mr;
+	struct ib_sig_attrs *sig_attrs = wr->sig_attrs;
+	struct ib_mr *sig_mr = wr->sig_mr;
 	struct mlx5_bsf *bsf;
-	u32 data_len = wr->sg_list->length;
-	u32 data_key = wr->sg_list->lkey;
-	u64 data_va = wr->sg_list->addr;
+	u32 data_len = wr->wr.sg_list->length;
+	u32 data_key = wr->wr.sg_list->lkey;
+	u64 data_va = wr->wr.sg_list->addr;
 	int ret;
 	int wqe_size;
 
-	if (!wr->wr.sig_handover.prot ||
-	    (data_key == wr->wr.sig_handover.prot->lkey &&
-	     data_va == wr->wr.sig_handover.prot->addr &&
-	     data_len == wr->wr.sig_handover.prot->length)) {
+	if (!wr->prot ||
+	    (data_key == wr->prot->lkey &&
+	     data_va == wr->prot->addr &&
+	     data_len == wr->prot->length)) {
 		/**
 		 * Source domain doesn't contain signature information
 		 * or data and protection are interleaved in memory.
@@ -2273,8 +2273,8 @@ static int set_sig_data_segment(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
 		struct mlx5_stride_block_ctrl_seg *sblock_ctrl;
 		struct mlx5_stride_block_entry *data_sentry;
 		struct mlx5_stride_block_entry *prot_sentry;
-		u32 prot_key = wr->wr.sig_handover.prot->lkey;
-		u64 prot_va = wr->wr.sig_handover.prot->addr;
+		u32 prot_key = wr->prot->lkey;
+		u64 prot_va = wr->prot->addr;
 		u16 block_size = sig_attrs->mem.sig.dif.pi_interval;
 		int prot_size;
 
@@ -2326,16 +2326,16 @@ static int set_sig_data_segment(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
 }
 
 static void set_sig_mkey_segment(struct mlx5_mkey_seg *seg,
-				 struct ib_send_wr *wr, u32 nelements,
+				 struct ib_sig_handover_wr *wr, u32 nelements,
 				 u32 length, u32 pdn)
 {
-	struct ib_mr *sig_mr = wr->wr.sig_handover.sig_mr;
+	struct ib_mr *sig_mr = wr->sig_mr;
 	u32 sig_key = sig_mr->rkey;
 	u8 sigerr = to_mmr(sig_mr)->sig->sigerr_count & 1;
 
 	memset(seg, 0, sizeof(*seg));
 
-	seg->flags = get_umr_flags(wr->wr.sig_handover.access_flags) |
+	seg->flags = get_umr_flags(wr->access_flags) |
 				   MLX5_ACCESS_MODE_KLM;
 	seg->qpn_mkey7_0 = cpu_to_be32((sig_key & 0xff) | 0xffffff00);
 	seg->flags_pd = cpu_to_be32(MLX5_MKEY_REMOTE_INVAL | sigerr << 26 |
@@ -2346,7 +2346,7 @@ static void set_sig_mkey_segment(struct mlx5_mkey_seg *seg,
 }
 
 static void set_sig_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
-				struct ib_send_wr *wr, u32 nelements)
+				u32 nelements)
 {
 	memset(umr, 0, sizeof(*umr));
 
@@ -2357,37 +2357,37 @@ static void set_sig_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
 }
 
 
-static int set_sig_umr_wr(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
+static int set_sig_umr_wr(struct ib_send_wr *send_wr, struct mlx5_ib_qp *qp,
 			  void **seg, int *size)
 {
-	struct mlx5_ib_mr *sig_mr = to_mmr(wr->wr.sig_handover.sig_mr);
+	struct ib_sig_handover_wr *wr = sig_handover_wr(send_wr);
+	struct mlx5_ib_mr *sig_mr = to_mmr(wr->sig_mr);
 	u32 pdn = get_pd(qp)->pdn;
 	u32 klm_oct_size;
 	int region_len, ret;
 
-	if (unlikely(wr->num_sge != 1) ||
-	    unlikely(wr->wr.sig_handover.access_flags &
-		     IB_ACCESS_REMOTE_ATOMIC) ||
+	if (unlikely(wr->wr.num_sge != 1) ||
+	    unlikely(wr->access_flags & IB_ACCESS_REMOTE_ATOMIC) ||
 	    unlikely(!sig_mr->sig) || unlikely(!qp->signature_en) ||
 	    unlikely(!sig_mr->sig->sig_status_checked))
 		return -EINVAL;
 
 	/* length of the protected region, data + protection */
-	region_len = wr->sg_list->length;
-	if (wr->wr.sig_handover.prot &&
-	    (wr->wr.sig_handover.prot->lkey != wr->sg_list->lkey  ||
-	     wr->wr.sig_handover.prot->addr != wr->sg_list->addr  ||
-	     wr->wr.sig_handover.prot->length != wr->sg_list->length))
-		region_len += wr->wr.sig_handover.prot->length;
+	region_len = wr->wr.sg_list->length;
+	if (wr->prot &&
+	    (wr->prot->lkey != wr->wr.sg_list->lkey  ||
+	     wr->prot->addr != wr->wr.sg_list->addr  ||
+	     wr->prot->length != wr->wr.sg_list->length))
+		region_len += wr->prot->length;
 
 	/**
 	 * KLM octoword size - if protection was provided
 	 * then we use strided block format (3 octowords),
 	 * else we use single KLM (1 octoword)
 	 **/
-	klm_oct_size = wr->wr.sig_handover.prot ? 3 : 1;
+	klm_oct_size = wr->prot ? 3 : 1;
 
-	set_sig_umr_segment(*seg, wr, klm_oct_size);
+	set_sig_umr_segment(*seg, klm_oct_size);
 	*seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
 	*size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
 	if (unlikely((*seg == qp->sq.qend)))
@@ -2454,8 +2454,8 @@ static int set_frwr_li_wr(void **seg, struct ib_send_wr *wr, int *size,
 	if (unlikely((*seg == qp->sq.qend)))
 		*seg = mlx5_get_send_wqe(qp, 0);
 	if (!li) {
-		if (unlikely(wr->wr.fast_reg.page_list_len >
-			     wr->wr.fast_reg.page_list->max_page_list_len))
+		if (unlikely(fast_reg_wr(wr)->page_list_len >
+			     fast_reg_wr(wr)->page_list->max_page_list_len))
 			return	-ENOMEM;
 
 		set_frwr_pages(*seg, wr, mdev, pd, writ);
@@ -2627,7 +2627,6 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		switch (ibqp->qp_type) {
 		case IB_QPT_XRC_INI:
 			xrc = seg;
-			xrc->xrc_srqn = htonl(wr->xrc_remote_srq_num);
 			seg += sizeof(*xrc);
 			size += sizeof(*xrc) / 16;
 			/* fall through */
@@ -2636,8 +2635,8 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_RDMA_READ:
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
-				set_raddr_seg(seg, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(seg, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				seg += sizeof(struct mlx5_wqe_raddr_seg);
 				size += sizeof(struct mlx5_wqe_raddr_seg) / 16;
 				break;
@@ -2666,7 +2665,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_FAST_REG_MR:
 				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
 				qp->sq.wr_data[idx] = IB_WR_FAST_REG_MR;
-				ctrl->imm = cpu_to_be32(wr->wr.fast_reg.rkey);
+				ctrl->imm = cpu_to_be32(fast_reg_wr(wr)->rkey);
 				err = set_frwr_li_wr(&seg, wr, &size, mdev, to_mpd(ibqp->pd), qp);
 				if (err) {
 					mlx5_ib_warn(dev, "\n");
@@ -2678,7 +2677,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 
 			case IB_WR_REG_SIG_MR:
 				qp->sq.wr_data[idx] = IB_WR_REG_SIG_MR;
-				mr = to_mmr(wr->wr.sig_handover.sig_mr);
+				mr = to_mmr(sig_handover_wr(wr)->sig_mr);
 
 				ctrl->imm = cpu_to_be32(mr->ibmr.rkey);
 				err = set_sig_umr_wr(wr, qp, &seg, &size);
@@ -2706,7 +2705,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 					goto out;
 				}
 
-				err = set_psv_wr(&wr->wr.sig_handover.sig_attrs->mem,
+				err = set_psv_wr(&sig_handover_wr(wr)->sig_attrs->mem,
 						 mr->sig->psv_memory.psv_idx, &seg,
 						 &size);
 				if (err) {
@@ -2728,7 +2727,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 				}
 
 				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
-				err = set_psv_wr(&wr->wr.sig_handover.sig_attrs->wire,
+				err = set_psv_wr(&sig_handover_wr(wr)->sig_attrs->wire,
 						 mr->sig->psv_wire.psv_idx, &seg,
 						 &size);
 				if (err) {
@@ -2752,8 +2751,8 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			switch (wr->opcode) {
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
-				set_raddr_seg(seg, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(seg, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				seg  += sizeof(struct mlx5_wqe_raddr_seg);
 				size += sizeof(struct mlx5_wqe_raddr_seg) / 16;
 				break;
@@ -2780,7 +2779,7 @@ int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 				goto out;
 			}
 			qp->sq.wr_data[idx] = MLX5_IB_WR_UMR;
-			ctrl->imm = cpu_to_be32(wr->wr.fast_reg.rkey);
+			ctrl->imm = cpu_to_be32(umr_wr(wr)->mkey);
 			set_reg_umr_segment(seg, wr);
 			seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
 			size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index e354b2f04ad9..35fe506e2cfa 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -1476,7 +1476,7 @@ void mthca_free_qp(struct mthca_dev *dev,
 
 /* Create UD header for an MLX send and build a data segment for it */
 static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
-			    int ind, struct ib_send_wr *wr,
+			    int ind, struct ib_ud_wr *wr,
 			    struct mthca_mlx_seg *mlx,
 			    struct mthca_data_seg *data)
 {
@@ -1485,10 +1485,10 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	u16 pkey;
 
 	ib_ud_header_init(256, /* assume a MAD */ 1, 0, 0,
-			  mthca_ah_grh_present(to_mah(wr->wr.ud.ah)), 0,
+			  mthca_ah_grh_present(to_mah(wr->ah)), 0,
 			  &sqp->ud_header);
 
-	err = mthca_read_ah(dev, to_mah(wr->wr.ud.ah), &sqp->ud_header);
+	err = mthca_read_ah(dev, to_mah(wr->ah), &sqp->ud_header);
 	if (err)
 		return err;
 	mlx->flags &= ~cpu_to_be32(MTHCA_NEXT_SOLICIT | 1);
@@ -1499,7 +1499,7 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	mlx->rlid = sqp->ud_header.lrh.destination_lid;
 	mlx->vcrc = 0;
 
-	switch (wr->opcode) {
+	switch (wr->wr.opcode) {
 	case IB_WR_SEND:
 		sqp->ud_header.bth.opcode = IB_OPCODE_UD_SEND_ONLY;
 		sqp->ud_header.immediate_present = 0;
@@ -1507,7 +1507,7 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	case IB_WR_SEND_WITH_IMM:
 		sqp->ud_header.bth.opcode = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
 		sqp->ud_header.immediate_present = 1;
-		sqp->ud_header.immediate_data = wr->ex.imm_data;
+		sqp->ud_header.immediate_data = wr->wr.ex.imm_data;
 		break;
 	default:
 		return -EINVAL;
@@ -1516,18 +1516,18 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	sqp->ud_header.lrh.virtual_lane    = !sqp->qp.ibqp.qp_num ? 15 : 0;
 	if (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE)
 		sqp->ud_header.lrh.source_lid = IB_LID_PERMISSIVE;
-	sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED);
+	sqp->ud_header.bth.solicited_event = !!(wr->wr.send_flags & IB_SEND_SOLICITED);
 	if (!sqp->qp.ibqp.qp_num)
 		ib_get_cached_pkey(&dev->ib_dev, sqp->qp.port,
 				   sqp->pkey_index, &pkey);
 	else
 		ib_get_cached_pkey(&dev->ib_dev, sqp->qp.port,
-				   wr->wr.ud.pkey_index, &pkey);
+				   wr->pkey_index, &pkey);
 	sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
-	sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+	sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->remote_qpn);
 	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
-	sqp->ud_header.deth.qkey = cpu_to_be32(wr->wr.ud.remote_qkey & 0x80000000 ?
-					       sqp->qkey : wr->wr.ud.remote_qkey);
+	sqp->ud_header.deth.qkey = cpu_to_be32(wr->remote_qkey & 0x80000000 ?
+					       sqp->qkey : wr->remote_qkey);
 	sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.ibqp.qp_num);
 
 	header_size = ib_ud_header_pack(&sqp->ud_header,
@@ -1569,34 +1569,34 @@ static __always_inline void set_raddr_seg(struct mthca_raddr_seg *rseg,
 }
 
 static __always_inline void set_atomic_seg(struct mthca_atomic_seg *aseg,
-					   struct ib_send_wr *wr)
+					   struct ib_atomic_wr *wr)
 {
-	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
-		aseg->swap_add = cpu_to_be64(wr->wr.atomic.swap);
-		aseg->compare  = cpu_to_be64(wr->wr.atomic.compare_add);
+	if (wr->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+		aseg->swap_add = cpu_to_be64(wr->swap);
+		aseg->compare  = cpu_to_be64(wr->compare_add);
 	} else {
-		aseg->swap_add = cpu_to_be64(wr->wr.atomic.compare_add);
+		aseg->swap_add = cpu_to_be64(wr->compare_add);
 		aseg->compare  = 0;
 	}
 
 }
 
 static void set_tavor_ud_seg(struct mthca_tavor_ud_seg *useg,
-			     struct ib_send_wr *wr)
+			     struct ib_ud_wr *wr)
 {
-	useg->lkey    = cpu_to_be32(to_mah(wr->wr.ud.ah)->key);
-	useg->av_addr =	cpu_to_be64(to_mah(wr->wr.ud.ah)->avdma);
-	useg->dqpn    =	cpu_to_be32(wr->wr.ud.remote_qpn);
-	useg->qkey    =	cpu_to_be32(wr->wr.ud.remote_qkey);
+	useg->lkey    = cpu_to_be32(to_mah(wr->ah)->key);
+	useg->av_addr =	cpu_to_be64(to_mah(wr->ah)->avdma);
+	useg->dqpn    =	cpu_to_be32(wr->remote_qpn);
+	useg->qkey    =	cpu_to_be32(wr->remote_qkey);
 
 }
 
 static void set_arbel_ud_seg(struct mthca_arbel_ud_seg *useg,
-			     struct ib_send_wr *wr)
+			     struct ib_ud_wr *wr)
 {
-	memcpy(useg->av, to_mah(wr->wr.ud.ah)->av, MTHCA_AV_SIZE);
-	useg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn);
-	useg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+	memcpy(useg->av, to_mah(wr->ah)->av, MTHCA_AV_SIZE);
+	useg->dqpn = cpu_to_be32(wr->remote_qpn);
+	useg->qkey = cpu_to_be32(wr->remote_qkey);
 }
 
 int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
@@ -1664,11 +1664,11 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			switch (wr->opcode) {
 			case IB_WR_ATOMIC_CMP_AND_SWP:
 			case IB_WR_ATOMIC_FETCH_AND_ADD:
-				set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
-					      wr->wr.atomic.rkey);
+				set_raddr_seg(wqe, atomic_wr(wr)->remote_addr,
+					      atomic_wr(wr)->rkey);
 				wqe += sizeof (struct mthca_raddr_seg);
 
-				set_atomic_seg(wqe, wr);
+				set_atomic_seg(wqe, atomic_wr(wr));
 				wqe += sizeof (struct mthca_atomic_seg);
 				size += (sizeof (struct mthca_raddr_seg) +
 					 sizeof (struct mthca_atomic_seg)) / 16;
@@ -1677,8 +1677,8 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
 			case IB_WR_RDMA_READ:
-				set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(wqe, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				wqe  += sizeof (struct mthca_raddr_seg);
 				size += sizeof (struct mthca_raddr_seg) / 16;
 				break;
@@ -1694,8 +1694,8 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			switch (wr->opcode) {
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
-				set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(wqe, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				wqe  += sizeof (struct mthca_raddr_seg);
 				size += sizeof (struct mthca_raddr_seg) / 16;
 				break;
@@ -1708,13 +1708,13 @@ int mthca_tavor_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			break;
 
 		case UD:
-			set_tavor_ud_seg(wqe, wr);
+			set_tavor_ud_seg(wqe, ud_wr(wr));
 			wqe  += sizeof (struct mthca_tavor_ud_seg);
 			size += sizeof (struct mthca_tavor_ud_seg) / 16;
 			break;
 
 		case MLX:
-			err = build_mlx_header(dev, to_msqp(qp), ind, wr,
+			err = build_mlx_header(dev, to_msqp(qp), ind, ud_wr(wr),
 					       wqe - sizeof (struct mthca_next_seg),
 					       wqe);
 			if (err) {
@@ -2005,11 +2005,11 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			switch (wr->opcode) {
 			case IB_WR_ATOMIC_CMP_AND_SWP:
 			case IB_WR_ATOMIC_FETCH_AND_ADD:
-				set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
-					      wr->wr.atomic.rkey);
+				set_raddr_seg(wqe, atomic_wr(wr)->remote_addr,
+					      atomic_wr(wr)->rkey);
 				wqe += sizeof (struct mthca_raddr_seg);
 
-				set_atomic_seg(wqe, wr);
+				set_atomic_seg(wqe, atomic_wr(wr));
 				wqe  += sizeof (struct mthca_atomic_seg);
 				size += (sizeof (struct mthca_raddr_seg) +
 					 sizeof (struct mthca_atomic_seg)) / 16;
@@ -2018,8 +2018,8 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			case IB_WR_RDMA_READ:
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
-				set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(wqe, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				wqe  += sizeof (struct mthca_raddr_seg);
 				size += sizeof (struct mthca_raddr_seg) / 16;
 				break;
@@ -2035,8 +2035,8 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			switch (wr->opcode) {
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
-				set_raddr_seg(wqe, wr->wr.rdma.remote_addr,
-					      wr->wr.rdma.rkey);
+				set_raddr_seg(wqe, rdma_wr(wr)->remote_addr,
+					      rdma_wr(wr)->rkey);
 				wqe  += sizeof (struct mthca_raddr_seg);
 				size += sizeof (struct mthca_raddr_seg) / 16;
 				break;
@@ -2049,13 +2049,13 @@ int mthca_arbel_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			break;
 
 		case UD:
-			set_arbel_ud_seg(wqe, wr);
+			set_arbel_ud_seg(wqe, ud_wr(wr));
 			wqe  += sizeof (struct mthca_arbel_ud_seg);
 			size += sizeof (struct mthca_arbel_ud_seg) / 16;
 			break;
 
 		case MLX:
-			err = build_mlx_header(dev, to_msqp(qp), ind, wr,
+			err = build_mlx_header(dev, to_msqp(qp), ind, ud_wr(wr),
 					       wqe - sizeof (struct mthca_next_seg),
 					       wqe);
 			if (err) {
diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c
index 44cb513f9a87..f71b37b75f82 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.c
+++ b/drivers/infiniband/hw/nes/nes_verbs.c
@@ -3372,9 +3372,9 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 				wqe_misc |= NES_IWARP_SQ_WQE_LOCAL_FENCE;
 
 			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_STAG_IDX,
-					    ib_wr->wr.rdma.rkey);
+					    rdma_wr(ib_wr)->rkey);
 			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX,
-					    ib_wr->wr.rdma.remote_addr);
+					    rdma_wr(ib_wr)->remote_addr);
 
 			if ((ib_wr->send_flags & IB_SEND_INLINE) &&
 			    ((nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA) == 0) &&
@@ -3409,9 +3409,9 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 			}
 
 			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX,
-					    ib_wr->wr.rdma.remote_addr);
+					    rdma_wr(ib_wr)->remote_addr);
 			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_STAG_IDX,
-					    ib_wr->wr.rdma.rkey);
+					    rdma_wr(ib_wr)->rkey);
 			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX,
 					    ib_wr->sg_list->length);
 			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_FRAG0_LOW_IDX,
@@ -3428,15 +3428,16 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 		case IB_WR_FAST_REG_MR:
 		{
 			int i;
-			int flags = ib_wr->wr.fast_reg.access_flags;
+			struct ib_fast_reg_wr *fwr = fast_reg_wr(ib_wr);
+			int flags = fwr->access_flags;
 			struct nes_ib_fast_reg_page_list *pnesfrpl =
-				container_of(ib_wr->wr.fast_reg.page_list,
+				container_of(fwr->page_list,
 					     struct nes_ib_fast_reg_page_list,
 					     ibfrpl);
 			u64 *src_page_list = pnesfrpl->ibfrpl.page_list;
 			u64 *dst_page_list = pnesfrpl->nes_wqe_pbl.kva;
 
-			if (ib_wr->wr.fast_reg.page_list_len >
+			if (fwr->page_list_len >
 			    (NES_4K_PBL_CHUNK_SIZE / sizeof(u64))) {
 				nes_debug(NES_DBG_IW_TX, "SQ_FMR: bad page_list_len\n");
 				err = -EINVAL;
@@ -3445,19 +3446,19 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 			wqe_misc = NES_IWARP_SQ_OP_FAST_REG;
 			set_wqe_64bit_value(wqe->wqe_words,
 					    NES_IWARP_SQ_FMR_WQE_VA_FBO_LOW_IDX,
-					    ib_wr->wr.fast_reg.iova_start);
+					    fwr->iova_start);
 			set_wqe_32bit_value(wqe->wqe_words,
 					    NES_IWARP_SQ_FMR_WQE_LENGTH_LOW_IDX,
-					    ib_wr->wr.fast_reg.length);
+					    fwr->length);
 			set_wqe_32bit_value(wqe->wqe_words,
 					    NES_IWARP_SQ_FMR_WQE_LENGTH_HIGH_IDX, 0);
 			set_wqe_32bit_value(wqe->wqe_words,
 					    NES_IWARP_SQ_FMR_WQE_MR_STAG_IDX,
-					    ib_wr->wr.fast_reg.rkey);
+					    fwr->rkey);
 			/* Set page size: */
-			if (ib_wr->wr.fast_reg.page_shift == 12) {
+			if (fwr->page_shift == 12) {
 				wqe_misc |= NES_IWARP_SQ_FMR_WQE_PAGE_SIZE_4K;
-			} else if (ib_wr->wr.fast_reg.page_shift == 21) {
+			} else if (fwr->page_shift == 21) {
 				wqe_misc |= NES_IWARP_SQ_FMR_WQE_PAGE_SIZE_2M;
 			} else {
 				nes_debug(NES_DBG_IW_TX, "Invalid page shift,"
@@ -3480,11 +3481,11 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 				wqe_misc |= NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_WINDOW_BIND;
 
 			/* Fill in PBL info: */
-			if (ib_wr->wr.fast_reg.page_list_len >
+			if (fwr->page_list_len >
 			    pnesfrpl->ibfrpl.max_page_list_len) {
 				nes_debug(NES_DBG_IW_TX, "Invalid page list length,"
 					  " ib_wr=%p, value=%u, max=%u\n",
-					  ib_wr, ib_wr->wr.fast_reg.page_list_len,
+					  ib_wr, fwr->page_list_len,
 					  pnesfrpl->ibfrpl.max_page_list_len);
 				err = -EINVAL;
 				break;
@@ -3496,19 +3497,19 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 
 			set_wqe_32bit_value(wqe->wqe_words,
 					    NES_IWARP_SQ_FMR_WQE_PBL_LENGTH_IDX,
-					    ib_wr->wr.fast_reg.page_list_len * 8);
+					    fwr->page_list_len * 8);
 
-			for (i = 0; i < ib_wr->wr.fast_reg.page_list_len; i++)
+			for (i = 0; i < fwr->page_list_len; i++)
 				dst_page_list[i] = cpu_to_le64(src_page_list[i]);
 
 			nes_debug(NES_DBG_IW_TX, "SQ_FMR: iova_start: %llx, "
 				  "length: %d, rkey: %0x, pgl_paddr: %llx, "
 				  "page_list_len: %u, wqe_misc: %x\n",
-				  (unsigned long long) ib_wr->wr.fast_reg.iova_start,
-				  ib_wr->wr.fast_reg.length,
-				  ib_wr->wr.fast_reg.rkey,
+				  (unsigned long long) fwr->iova_start,
+				  fwr->length,
+				  fwr->rkey,
 				  (unsigned long long) pnesfrpl->nes_wqe_pbl.paddr,
-				  ib_wr->wr.fast_reg.page_list_len,
+				  fwr->page_list_len,
 				  wqe_misc);
 			break;
 		}
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
index 9bb710a402cd..dd00219b7d15 100644
--- a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -1997,13 +1997,13 @@ static void ocrdma_build_ud_hdr(struct ocrdma_qp *qp,
 {
 	struct ocrdma_ewqe_ud_hdr *ud_hdr =
 		(struct ocrdma_ewqe_ud_hdr *)(hdr + 1);
-	struct ocrdma_ah *ah = get_ocrdma_ah(wr->wr.ud.ah);
+	struct ocrdma_ah *ah = get_ocrdma_ah(ud_wr(wr)->ah);
 
-	ud_hdr->rsvd_dest_qpn = wr->wr.ud.remote_qpn;
+	ud_hdr->rsvd_dest_qpn = ud_wr(wr)->remote_qpn;
 	if (qp->qp_type == IB_QPT_GSI)
 		ud_hdr->qkey = qp->qkey;
 	else
-		ud_hdr->qkey = wr->wr.ud.remote_qkey;
+		ud_hdr->qkey = ud_wr(wr)->remote_qkey;
 	ud_hdr->rsvd_ahid = ah->id;
 	if (ah->av->valid & OCRDMA_AV_VLAN_VALID)
 		hdr->cw |= (OCRDMA_FLAG_AH_VLAN_PR << OCRDMA_WQE_FLAGS_SHIFT);
@@ -2106,9 +2106,9 @@ static int ocrdma_build_write(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
 	status = ocrdma_build_inline_sges(qp, hdr, sge, wr, wqe_size);
 	if (status)
 		return status;
-	ext_rw->addr_lo = wr->wr.rdma.remote_addr;
-	ext_rw->addr_hi = upper_32_bits(wr->wr.rdma.remote_addr);
-	ext_rw->lrkey = wr->wr.rdma.rkey;
+	ext_rw->addr_lo = rdma_wr(wr)->remote_addr;
+	ext_rw->addr_hi = upper_32_bits(rdma_wr(wr)->remote_addr);
+	ext_rw->lrkey = rdma_wr(wr)->rkey;
 	ext_rw->len = hdr->total_len;
 	return 0;
 }
@@ -2126,13 +2126,14 @@ static void ocrdma_build_read(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
 	hdr->cw |= (OCRDMA_READ << OCRDMA_WQE_OPCODE_SHIFT);
 	hdr->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
 
-	ext_rw->addr_lo = wr->wr.rdma.remote_addr;
-	ext_rw->addr_hi = upper_32_bits(wr->wr.rdma.remote_addr);
-	ext_rw->lrkey = wr->wr.rdma.rkey;
+	ext_rw->addr_lo = rdma_wr(wr)->remote_addr;
+	ext_rw->addr_hi = upper_32_bits(rdma_wr(wr)->remote_addr);
+	ext_rw->lrkey = rdma_wr(wr)->rkey;
 	ext_rw->len = hdr->total_len;
 }
 
-static void build_frmr_pbes(struct ib_send_wr *wr, struct ocrdma_pbl *pbl_tbl,
+static void build_frmr_pbes(struct ib_fast_reg_wr *wr,
+			    struct ocrdma_pbl *pbl_tbl,
 			    struct ocrdma_hw_mr *hwmr)
 {
 	int i;
@@ -2144,12 +2145,12 @@ static void build_frmr_pbes(struct ib_send_wr *wr, struct ocrdma_pbl *pbl_tbl,
 	num_pbes = 0;
 
 	/* go through the OS phy regions & fill hw pbe entries into pbls. */
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+	for (i = 0; i < wr->page_list_len; i++) {
 		/* number of pbes can be more for one OS buf, when
 		 * buffers are of different sizes.
 		 * split the ib_buf to one or more pbes.
 		 */
-		buf_addr = wr->wr.fast_reg.page_list->page_list[i];
+		buf_addr = wr->page_list->page_list[i];
 		pbe->pa_lo = cpu_to_le32((u32) (buf_addr & PAGE_MASK));
 		pbe->pa_hi = cpu_to_le32((u32) upper_32_bits(buf_addr));
 		num_pbes += 1;
@@ -2178,9 +2179,10 @@ static int get_encoded_page_size(int pg_sz)
 
 
 static int ocrdma_build_fr(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
-			   struct ib_send_wr *wr)
+			   struct ib_send_wr *send_wr)
 {
 	u64 fbo;
+	struct ib_fast_reg_wr *wr = fast_reg_wr(send_wr);
 	struct ocrdma_ewqe_fr *fast_reg = (struct ocrdma_ewqe_fr *)(hdr + 1);
 	struct ocrdma_mr *mr;
 	struct ocrdma_dev *dev = get_ocrdma_dev(qp->ibqp.device);
@@ -2188,33 +2190,32 @@ static int ocrdma_build_fr(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
 
 	wqe_size = roundup(wqe_size, OCRDMA_WQE_ALIGN_BYTES);
 
-	if (wr->wr.fast_reg.page_list_len > dev->attr.max_pages_per_frmr)
+	if (wr->page_list_len > dev->attr.max_pages_per_frmr)
 		return -EINVAL;
 
 	hdr->cw |= (OCRDMA_FR_MR << OCRDMA_WQE_OPCODE_SHIFT);
 	hdr->cw |= ((wqe_size / OCRDMA_WQE_STRIDE) << OCRDMA_WQE_SIZE_SHIFT);
 
-	if (wr->wr.fast_reg.page_list_len == 0)
+	if (wr->page_list_len == 0)
 		BUG();
-	if (wr->wr.fast_reg.access_flags & IB_ACCESS_LOCAL_WRITE)
+	if (wr->access_flags & IB_ACCESS_LOCAL_WRITE)
 		hdr->rsvd_lkey_flags |= OCRDMA_LKEY_FLAG_LOCAL_WR;
-	if (wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_WRITE)
+	if (wr->access_flags & IB_ACCESS_REMOTE_WRITE)
 		hdr->rsvd_lkey_flags |= OCRDMA_LKEY_FLAG_REMOTE_WR;
-	if (wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_READ)
+	if (wr->access_flags & IB_ACCESS_REMOTE_READ)
 		hdr->rsvd_lkey_flags |= OCRDMA_LKEY_FLAG_REMOTE_RD;
-	hdr->lkey = wr->wr.fast_reg.rkey;
-	hdr->total_len = wr->wr.fast_reg.length;
+	hdr->lkey = wr->rkey;
+	hdr->total_len = wr->length;
 
-	fbo = wr->wr.fast_reg.iova_start -
-	    (wr->wr.fast_reg.page_list->page_list[0] & PAGE_MASK);
+	fbo = wr->iova_start - (wr->page_list->page_list[0] & PAGE_MASK);
 
-	fast_reg->va_hi = upper_32_bits(wr->wr.fast_reg.iova_start);
-	fast_reg->va_lo = (u32) (wr->wr.fast_reg.iova_start & 0xffffffff);
+	fast_reg->va_hi = upper_32_bits(wr->iova_start);
+	fast_reg->va_lo = (u32) (wr->iova_start & 0xffffffff);
 	fast_reg->fbo_hi = upper_32_bits(fbo);
 	fast_reg->fbo_lo = (u32) fbo & 0xffffffff;
-	fast_reg->num_sges = wr->wr.fast_reg.page_list_len;
+	fast_reg->num_sges = wr->page_list_len;
 	fast_reg->size_sge =
-		get_encoded_page_size(1 << wr->wr.fast_reg.page_shift);
+		get_encoded_page_size(1 << wr->page_shift);
 	mr = (struct ocrdma_mr *) (unsigned long)
 		dev->stag_arr[(hdr->lkey >> 8) & (OCRDMA_MAX_STAG - 1)];
 	build_frmr_pbes(wr, mr->hwmr.pbl_table, &mr->hwmr);
diff --git a/drivers/infiniband/hw/qib/qib_keys.c b/drivers/infiniband/hw/qib/qib_keys.c
index 5afaa218508d..eaf139a33b2e 100644
--- a/drivers/infiniband/hw/qib/qib_keys.c
+++ b/drivers/infiniband/hw/qib/qib_keys.c
@@ -338,12 +338,13 @@ bail:
 /*
  * Initialize the memory region specified by the work reqeust.
  */
-int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr)
+int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *send_wr)
 {
+	struct ib_fast_reg_wr *wr = fast_reg_wr(send_wr);
 	struct qib_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
 	struct qib_pd *pd = to_ipd(qp->ibqp.pd);
 	struct qib_mregion *mr;
-	u32 rkey = wr->wr.fast_reg.rkey;
+	u32 rkey = wr->rkey;
 	unsigned i, n, m;
 	int ret = -EINVAL;
 	unsigned long flags;
@@ -360,22 +361,22 @@ int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr)
 	if (unlikely(mr == NULL || qp->ibqp.pd != mr->pd))
 		goto bail;
 
-	if (wr->wr.fast_reg.page_list_len > mr->max_segs)
+	if (wr->page_list_len > mr->max_segs)
 		goto bail;
 
-	ps = 1UL << wr->wr.fast_reg.page_shift;
-	if (wr->wr.fast_reg.length > ps * wr->wr.fast_reg.page_list_len)
+	ps = 1UL << wr->page_shift;
+	if (wr->length > ps * wr->page_list_len)
 		goto bail;
 
-	mr->user_base = wr->wr.fast_reg.iova_start;
-	mr->iova = wr->wr.fast_reg.iova_start;
+	mr->user_base = wr->iova_start;
+	mr->iova = wr->iova_start;
 	mr->lkey = rkey;
-	mr->length = wr->wr.fast_reg.length;
-	mr->access_flags = wr->wr.fast_reg.access_flags;
-	page_list = wr->wr.fast_reg.page_list->page_list;
+	mr->length = wr->length;
+	mr->access_flags = wr->access_flags;
+	page_list = wr->page_list->page_list;
 	m = 0;
 	n = 0;
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+	for (i = 0; i < wr->page_list_len; i++) {
 		mr->map[m]->segs[n].vaddr = (void *) page_list[i];
 		mr->map[m]->segs[n].length = ps;
 		if (++n == QIB_SEGSZ) {
diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
index 4fa88ba2963e..40f85bb3e0d3 100644
--- a/drivers/infiniband/hw/qib/qib_qp.c
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -436,7 +436,7 @@ static void clear_mr_refs(struct qib_qp *qp, int clr_sends)
 			if (qp->ibqp.qp_type == IB_QPT_UD ||
 			    qp->ibqp.qp_type == IB_QPT_SMI ||
 			    qp->ibqp.qp_type == IB_QPT_GSI)
-				atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+				atomic_dec(&to_iah(wqe->ud_wr.ah)->refcount);
 			if (++qp->s_last >= qp->s_size)
 				qp->s_last = 0;
 		}
diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
index 4544d6f88ad7..e6b7556d5221 100644
--- a/drivers/infiniband/hw/qib/qib_rc.c
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -373,10 +373,11 @@ int qib_make_rc_req(struct qib_qp *qp)
 				qp->s_flags |= QIB_S_WAIT_SSN_CREDIT;
 				goto bail;
 			}
+
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			hwords += sizeof(struct ib_reth) / sizeof(u32);
 			wqe->lpsn = wqe->psn;
@@ -386,15 +387,15 @@ int qib_make_rc_req(struct qib_qp *qp)
 				len = pmtu;
 				break;
 			}
-			if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+			if (wqe->rdma_wr.wr.opcode == IB_WR_RDMA_WRITE)
 				qp->s_state = OP(RDMA_WRITE_ONLY);
 			else {
-				qp->s_state =
-					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
+				qp->s_state = OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
 				/* Immediate data comes after RETH */
-				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
+				ohdr->u.rc.imm_data =
+					wqe->rdma_wr.wr.ex.imm_data;
 				hwords += 1;
-				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				if (wqe->rdma_wr.wr.send_flags & IB_SEND_SOLICITED)
 					bth0 |= IB_BTH_SOLICITED;
 			}
 			bth2 |= IB_BTH_REQ_ACK;
@@ -424,10 +425,11 @@ int qib_make_rc_req(struct qib_qp *qp)
 					qp->s_next_psn += (len - 1) / pmtu;
 				wqe->lpsn = qp->s_next_psn++;
 			}
+
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			qp->s_state = OP(RDMA_READ_REQUEST);
 			hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
@@ -455,24 +457,24 @@ int qib_make_rc_req(struct qib_qp *qp)
 					qp->s_lsn++;
 				wqe->lpsn = wqe->psn;
 			}
-			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+			if (wqe->atomic_wr.wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
 				qp->s_state = OP(COMPARE_SWAP);
 				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
-					wqe->wr.wr.atomic.swap);
+					wqe->atomic_wr.swap);
 				ohdr->u.atomic_eth.compare_data = cpu_to_be64(
-					wqe->wr.wr.atomic.compare_add);
+					wqe->atomic_wr.compare_add);
 			} else {
 				qp->s_state = OP(FETCH_ADD);
 				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
-					wqe->wr.wr.atomic.compare_add);
+					wqe->atomic_wr.compare_add);
 				ohdr->u.atomic_eth.compare_data = 0;
 			}
 			ohdr->u.atomic_eth.vaddr[0] = cpu_to_be32(
-				wqe->wr.wr.atomic.remote_addr >> 32);
+				wqe->atomic_wr.remote_addr >> 32);
 			ohdr->u.atomic_eth.vaddr[1] = cpu_to_be32(
-				wqe->wr.wr.atomic.remote_addr);
+				wqe->atomic_wr.remote_addr);
 			ohdr->u.atomic_eth.rkey = cpu_to_be32(
-				wqe->wr.wr.atomic.rkey);
+				wqe->atomic_wr.rkey);
 			hwords += sizeof(struct ib_atomic_eth) / sizeof(u32);
 			ss = NULL;
 			len = 0;
@@ -597,9 +599,9 @@ int qib_make_rc_req(struct qib_qp *qp)
 		 */
 		len = ((qp->s_psn - wqe->psn) & QIB_PSN_MASK) * pmtu;
 		ohdr->u.rc.reth.vaddr =
-			cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
+			cpu_to_be64(wqe->rdma_wr.remote_addr + len);
 		ohdr->u.rc.reth.rkey =
-			cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			cpu_to_be32(wqe->rdma_wr.rkey);
 		ohdr->u.rc.reth.length = cpu_to_be32(wqe->length - len);
 		qp->s_state = OP(RDMA_READ_REQUEST);
 		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
diff --git a/drivers/infiniband/hw/qib/qib_ruc.c b/drivers/infiniband/hw/qib/qib_ruc.c
index 22e356ca8058..b1aa21bdd484 100644
--- a/drivers/infiniband/hw/qib/qib_ruc.c
+++ b/drivers/infiniband/hw/qib/qib_ruc.c
@@ -459,8 +459,8 @@ again:
 		if (wqe->length == 0)
 			break;
 		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, wqe->length,
-					  wqe->wr.wr.rdma.remote_addr,
-					  wqe->wr.wr.rdma.rkey,
+					  wqe->rdma_wr.remote_addr,
+					  wqe->rdma_wr.rkey,
 					  IB_ACCESS_REMOTE_WRITE)))
 			goto acc_err;
 		qp->r_sge.sg_list = NULL;
@@ -472,8 +472,8 @@ again:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
 			goto inv_err;
 		if (unlikely(!qib_rkey_ok(qp, &sqp->s_sge.sge, wqe->length,
-					  wqe->wr.wr.rdma.remote_addr,
-					  wqe->wr.wr.rdma.rkey,
+					  wqe->rdma_wr.remote_addr,
+					  wqe->rdma_wr.rkey,
 					  IB_ACCESS_REMOTE_READ)))
 			goto acc_err;
 		release = 0;
@@ -490,18 +490,18 @@ again:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
 			goto inv_err;
 		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
-					  wqe->wr.wr.atomic.remote_addr,
-					  wqe->wr.wr.atomic.rkey,
+					  wqe->atomic_wr.remote_addr,
+					  wqe->atomic_wr.rkey,
 					  IB_ACCESS_REMOTE_ATOMIC)))
 			goto acc_err;
 		/* Perform atomic OP and save result. */
 		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
-		sdata = wqe->wr.wr.atomic.compare_add;
+		sdata = wqe->atomic_wr.compare_add;
 		*(u64 *) sqp->s_sge.sge.vaddr =
-			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
+			(wqe->atomic_wr.wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
 			(u64) atomic64_add_return(sdata, maddr) - sdata :
 			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
-				      sdata, wqe->wr.wr.atomic.swap);
+				      sdata, wqe->atomic_wr.swap);
 		qib_put_mr(qp->r_sge.sge.mr);
 		qp->r_sge.num_sge = 0;
 		goto send_comp;
@@ -785,7 +785,7 @@ void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
 	if (qp->ibqp.qp_type == IB_QPT_UD ||
 	    qp->ibqp.qp_type == IB_QPT_SMI ||
 	    qp->ibqp.qp_type == IB_QPT_GSI)
-		atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+		atomic_dec(&to_iah(wqe->ud_wr.ah)->refcount);
 
 	/* See ch. 11.2.4.1 and 10.7.3.1 */
 	if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
diff --git a/drivers/infiniband/hw/qib/qib_uc.c b/drivers/infiniband/hw/qib/qib_uc.c
index aa3a8035bb68..06a564589c35 100644
--- a/drivers/infiniband/hw/qib/qib_uc.c
+++ b/drivers/infiniband/hw/qib/qib_uc.c
@@ -129,9 +129,9 @@ int qib_make_uc_req(struct qib_qp *qp)
 		case IB_WR_RDMA_WRITE:
 		case IB_WR_RDMA_WRITE_WITH_IMM:
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			hwords += sizeof(struct ib_reth) / 4;
 			if (len > pmtu) {
diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
index 26243b722b5e..59193f67ea78 100644
--- a/drivers/infiniband/hw/qib/qib_ud.c
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -59,7 +59,7 @@ static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
 	u32 length;
 	enum ib_qp_type sqptype, dqptype;
 
-	qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
+	qp = qib_lookup_qpn(ibp, swqe->ud_wr.remote_qpn);
 	if (!qp) {
 		ibp->n_pkt_drops++;
 		return;
@@ -76,7 +76,7 @@ static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
 		goto drop;
 	}
 
-	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
+	ah_attr = &to_iah(swqe->ud_wr.ah)->attr;
 	ppd = ppd_from_ibp(ibp);
 
 	if (qp->ibqp.qp_num > 1) {
@@ -106,8 +106,8 @@ static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
 	if (qp->ibqp.qp_num) {
 		u32 qkey;
 
-		qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
-			sqp->qkey : swqe->wr.wr.ud.remote_qkey;
+		qkey = (int)swqe->ud_wr.remote_qkey < 0 ?
+			sqp->qkey : swqe->ud_wr.remote_qkey;
 		if (unlikely(qkey != qp->qkey)) {
 			u16 lid;
 
@@ -210,7 +210,7 @@ static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
 	wc.qp = &qp->ibqp;
 	wc.src_qp = sqp->ibqp.qp_num;
 	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
-		swqe->wr.wr.ud.pkey_index : 0;
+		swqe->ud_wr.pkey_index : 0;
 	wc.slid = ppd->lid | (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
 	wc.sl = ah_attr->sl;
 	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
@@ -277,7 +277,7 @@ int qib_make_ud_req(struct qib_qp *qp)
 	/* Construct the header. */
 	ibp = to_iport(qp->ibqp.device, qp->port_num);
 	ppd = ppd_from_ibp(ibp);
-	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
+	ah_attr = &to_iah(wqe->ud_wr.ah)->attr;
 	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
 		if (ah_attr->dlid != QIB_PERMISSIVE_LID)
 			this_cpu_inc(ibp->pmastats->n_multicast_xmit);
@@ -363,7 +363,7 @@ int qib_make_ud_req(struct qib_qp *qp)
 	bth0 |= extra_bytes << 20;
 	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
 		qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
-			     wqe->wr.wr.ud.pkey_index : qp->s_pkey_index);
+			     wqe->ud_wr.pkey_index : qp->s_pkey_index);
 	ohdr->bth[0] = cpu_to_be32(bth0);
 	/*
 	 * Use the multicast QP if the destination LID is a multicast LID.
@@ -371,14 +371,14 @@ int qib_make_ud_req(struct qib_qp *qp)
 	ohdr->bth[1] = ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
 		ah_attr->dlid != QIB_PERMISSIVE_LID ?
 		cpu_to_be32(QIB_MULTICAST_QPN) :
-		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
+		cpu_to_be32(wqe->ud_wr.remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
 	/*
 	 * Qkeys with the high order bit set mean use the
 	 * qkey from the QP context instead of the WR (see 10.2.5).
 	 */
-	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
-					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
+	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->ud_wr.remote_qkey < 0 ?
+					 qp->qkey : wqe->ud_wr.remote_qkey);
 	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
 
 done:
diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c
index 3dcc4985b60f..a6b0b098ff30 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.c
+++ b/drivers/infiniband/hw/qib/qib_verbs.c
@@ -374,7 +374,7 @@ static int qib_post_one_send(struct qib_qp *qp, struct ib_send_wr *wr,
 		    wr->opcode != IB_WR_SEND_WITH_IMM)
 			goto bail_inval;
 		/* Check UD destination address PD */
-		if (qp->ibqp.pd != wr->wr.ud.ah->pd)
+		if (qp->ibqp.pd != ud_wr(wr)->ah->pd)
 			goto bail_inval;
 	} else if ((unsigned) wr->opcode > IB_WR_ATOMIC_FETCH_AND_ADD)
 		goto bail_inval;
@@ -397,7 +397,23 @@ static int qib_post_one_send(struct qib_qp *qp, struct ib_send_wr *wr,
 	rkt = &to_idev(qp->ibqp.device)->lk_table;
 	pd = to_ipd(qp->ibqp.pd);
 	wqe = get_swqe_ptr(qp, qp->s_head);
-	wqe->wr = *wr;
+
+	if (qp->ibqp.qp_type != IB_QPT_UC &&
+	    qp->ibqp.qp_type != IB_QPT_RC)
+		memcpy(&wqe->ud_wr, ud_wr(wr), sizeof(wqe->ud_wr));
+	else if (wr->opcode == IB_WR_FAST_REG_MR)
+		memcpy(&wqe->fast_reg_wr, fast_reg_wr(wr),
+			sizeof(wqe->fast_reg_wr));
+	else if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
+		 wr->opcode == IB_WR_RDMA_WRITE ||
+		 wr->opcode == IB_WR_RDMA_READ)
+		memcpy(&wqe->rdma_wr, rdma_wr(wr), sizeof(wqe->rdma_wr));
+	else if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		 wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+		memcpy(&wqe->atomic_wr, atomic_wr(wr), sizeof(wqe->atomic_wr));
+	else
+		memcpy(&wqe->wr, wr, sizeof(wqe->wr));
+
 	wqe->length = 0;
 	j = 0;
 	if (wr->num_sge) {
@@ -426,7 +442,7 @@ static int qib_post_one_send(struct qib_qp *qp, struct ib_send_wr *wr,
 				  qp->port_num - 1)->ibmtu)
 		goto bail_inval_free;
 	else
-		atomic_inc(&to_iah(wr->wr.ud.ah)->refcount);
+		atomic_inc(&to_iah(ud_wr(wr)->ah)->refcount);
 	wqe->ssn = qp->s_ssn++;
 	qp->s_head = next;
 
diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
index a08df70e8503..8aa16851a5e6 100644
--- a/drivers/infiniband/hw/qib/qib_verbs.h
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -338,7 +338,13 @@ struct qib_mr {
  * in qp->s_max_sge.
  */
 struct qib_swqe {
-	struct ib_send_wr wr;   /* don't use wr.sg_list */
+	union {
+		struct ib_send_wr wr;   /* don't use wr.sg_list */
+		struct ib_ud_wr ud_wr;
+		struct ib_fast_reg_wr fast_reg_wr;
+		struct ib_rdma_wr rdma_wr;
+		struct ib_atomic_wr atomic_wr;
+	};
 	u32 psn;                /* first packet sequence number */
 	u32 lpsn;               /* last packet sequence number */
 	u32 ssn;                /* send sequence number */
diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index edc5b8565d6d..3ede10309754 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -360,7 +360,7 @@ struct ipoib_dev_priv {
 	unsigned	     tx_head;
 	unsigned	     tx_tail;
 	struct ib_sge	     tx_sge[MAX_SKB_FRAGS + 1];
-	struct ib_send_wr    tx_wr;
+	struct ib_ud_wr      tx_wr;
 	unsigned	     tx_outstanding;
 	struct ib_wc	     send_wc[MAX_SEND_CQE];
 
@@ -528,7 +528,7 @@ static inline void ipoib_build_sge(struct ipoib_dev_priv *priv,
 		priv->tx_sge[i + off].addr = mapping[i + off];
 		priv->tx_sge[i + off].length = skb_frag_size(&frags[i]);
 	}
-	priv->tx_wr.num_sge	     = nr_frags + off;
+	priv->tx_wr.wr.num_sge	     = nr_frags + off;
 }
 
 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
index c78dc1638030..3ae9726efb98 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
@@ -700,9 +700,9 @@ static inline int post_send(struct ipoib_dev_priv *priv,
 
 	ipoib_build_sge(priv, tx_req);
 
-	priv->tx_wr.wr_id	= wr_id | IPOIB_OP_CM;
+	priv->tx_wr.wr.wr_id	= wr_id | IPOIB_OP_CM;
 
-	return ib_post_send(tx->qp, &priv->tx_wr, &bad_wr);
+	return ib_post_send(tx->qp, &priv->tx_wr.wr, &bad_wr);
 }
 
 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index d266667ca9b8..5ea0c14070d1 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -518,19 +518,19 @@ static inline int post_send(struct ipoib_dev_priv *priv,
 
 	ipoib_build_sge(priv, tx_req);
 
-	priv->tx_wr.wr_id 	     = wr_id;
-	priv->tx_wr.wr.ud.remote_qpn = qpn;
-	priv->tx_wr.wr.ud.ah 	     = address;
+	priv->tx_wr.wr.wr_id	= wr_id;
+	priv->tx_wr.remote_qpn	= qpn;
+	priv->tx_wr.ah		= address;
 
 	if (head) {
-		priv->tx_wr.wr.ud.mss	 = skb_shinfo(skb)->gso_size;
-		priv->tx_wr.wr.ud.header = head;
-		priv->tx_wr.wr.ud.hlen	 = hlen;
-		priv->tx_wr.opcode	 = IB_WR_LSO;
+		priv->tx_wr.mss		= skb_shinfo(skb)->gso_size;
+		priv->tx_wr.header	= head;
+		priv->tx_wr.hlen	= hlen;
+		priv->tx_wr.wr.opcode	= IB_WR_LSO;
 	} else
-		priv->tx_wr.opcode	 = IB_WR_SEND;
+		priv->tx_wr.wr.opcode	= IB_WR_SEND;
 
-	return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr);
+	return ib_post_send(priv->qp, &priv->tx_wr.wr, &bad_wr);
 }
 
 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
@@ -583,9 +583,9 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 	}
 
 	if (skb->ip_summed == CHECKSUM_PARTIAL)
-		priv->tx_wr.send_flags |= IB_SEND_IP_CSUM;
+		priv->tx_wr.wr.send_flags |= IB_SEND_IP_CSUM;
 	else
-		priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM;
+		priv->tx_wr.wr.send_flags &= ~IB_SEND_IP_CSUM;
 
 	if (++priv->tx_outstanding == ipoib_sendq_size) {
 		ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index cbb6721d0a65..7d3281866ffc 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -461,7 +461,7 @@ int ipoib_set_mode(struct net_device *dev, const char *buf)
 		netdev_update_features(dev);
 		dev_set_mtu(dev, ipoib_cm_max_mtu(dev));
 		rtnl_unlock();
-		priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM;
+		priv->tx_wr.wr.send_flags &= ~IB_SEND_IP_CSUM;
 
 		ipoib_flush_paths(dev);
 		rtnl_lock();
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
index d4b97614196c..f357ca67a41c 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
@@ -245,7 +245,7 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
 
 		priv->qkey = be32_to_cpu(priv->broadcast->mcmember.qkey);
 		spin_unlock_irq(&priv->lock);
-		priv->tx_wr.wr.ud.remote_qkey = priv->qkey;
+		priv->tx_wr.remote_qkey = priv->qkey;
 		set_qkey = 1;
 	}
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
index 78845b6e8b81..d48c5bae7877 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
@@ -221,9 +221,9 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 	for (i = 0; i < MAX_SKB_FRAGS + 1; ++i)
 		priv->tx_sge[i].lkey = priv->pd->local_dma_lkey;
 
-	priv->tx_wr.opcode	= IB_WR_SEND;
-	priv->tx_wr.sg_list	= priv->tx_sge;
-	priv->tx_wr.send_flags	= IB_SEND_SIGNALED;
+	priv->tx_wr.wr.opcode		= IB_WR_SEND;
+	priv->tx_wr.wr.sg_list		= priv->tx_sge;
+	priv->tx_wr.wr.send_flags	= IB_SEND_SIGNALED;
 
 	priv->rx_sge[0].lkey = priv->pd->local_dma_lkey;
 
diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.h b/drivers/infiniband/ulp/iser/iscsi_iser.h
index d8bbad9eb59b..2484bee993ec 100644
--- a/drivers/infiniband/ulp/iser/iscsi_iser.h
+++ b/drivers/infiniband/ulp/iser/iscsi_iser.h
@@ -295,7 +295,11 @@ struct iser_tx_desc {
 	int                          num_sge;
 	bool			     mapped;
 	u8                           wr_idx;
-	struct ib_send_wr            wrs[ISER_MAX_WRS];
+	union iser_wr {
+		struct ib_send_wr		send;
+		struct ib_fast_reg_wr		fast_reg;
+		struct ib_sig_handover_wr	sig;
+	} wrs[ISER_MAX_WRS];
 	struct iser_mem_reg          data_reg;
 	struct iser_mem_reg          prot_reg;
 	struct ib_sig_attrs          sig_attrs;
@@ -707,11 +711,11 @@ iser_reg_desc_put_fmr(struct ib_conn *ib_conn,
 static inline struct ib_send_wr *
 iser_tx_next_wr(struct iser_tx_desc *tx_desc)
 {
-	struct ib_send_wr *cur_wr = &tx_desc->wrs[tx_desc->wr_idx];
+	struct ib_send_wr *cur_wr = &tx_desc->wrs[tx_desc->wr_idx].send;
 	struct ib_send_wr *last_wr;
 
 	if (tx_desc->wr_idx) {
-		last_wr = &tx_desc->wrs[tx_desc->wr_idx - 1];
+		last_wr = &tx_desc->wrs[tx_desc->wr_idx - 1].send;
 		last_wr->next = cur_wr;
 	}
 	tx_desc->wr_idx++;
diff --git a/drivers/infiniband/ulp/iser/iser_memory.c b/drivers/infiniband/ulp/iser/iser_memory.c
index 3e0452c4248f..b29fda3e8e74 100644
--- a/drivers/infiniband/ulp/iser/iser_memory.c
+++ b/drivers/infiniband/ulp/iser/iser_memory.c
@@ -437,7 +437,7 @@ iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
 {
 	struct iser_tx_desc *tx_desc = &iser_task->desc;
 	struct ib_sig_attrs *sig_attrs = &tx_desc->sig_attrs;
-	struct ib_send_wr *wr;
+	struct ib_sig_handover_wr *wr;
 	int ret;
 
 	memset(sig_attrs, 0, sizeof(*sig_attrs));
@@ -447,26 +447,24 @@ iser_reg_sig_mr(struct iscsi_iser_task *iser_task,
 
 	iser_set_prot_checks(iser_task->sc, &sig_attrs->check_mask);
 
-	if (!pi_ctx->sig_mr_valid) {
-		wr = iser_tx_next_wr(tx_desc);
-		iser_inv_rkey(wr, pi_ctx->sig_mr);
-	}
-
-	wr = iser_tx_next_wr(tx_desc);
-	wr->opcode = IB_WR_REG_SIG_MR;
-	wr->wr_id = ISER_FASTREG_LI_WRID;
-	wr->sg_list = &data_reg->sge;
-	wr->num_sge = 1;
-	wr->send_flags = 0;
-	wr->wr.sig_handover.sig_attrs = sig_attrs;
-	wr->wr.sig_handover.sig_mr = pi_ctx->sig_mr;
+	if (!pi_ctx->sig_mr_valid)
+		iser_inv_rkey(iser_tx_next_wr(tx_desc), pi_ctx->sig_mr);
+
+	wr = sig_handover_wr(iser_tx_next_wr(tx_desc));
+	wr->wr.opcode = IB_WR_REG_SIG_MR;
+	wr->wr.wr_id = ISER_FASTREG_LI_WRID;
+	wr->wr.sg_list = &data_reg->sge;
+	wr->wr.num_sge = 1;
+	wr->wr.send_flags = 0;
+	wr->sig_attrs = sig_attrs;
+	wr->sig_mr = pi_ctx->sig_mr;
 	if (scsi_prot_sg_count(iser_task->sc))
-		wr->wr.sig_handover.prot = &prot_reg->sge;
+		wr->prot = &prot_reg->sge;
 	else
-		wr->wr.sig_handover.prot = NULL;
-	wr->wr.sig_handover.access_flags = IB_ACCESS_LOCAL_WRITE |
-					   IB_ACCESS_REMOTE_READ |
-					   IB_ACCESS_REMOTE_WRITE;
+		wr->prot = NULL;
+	wr->access_flags = IB_ACCESS_LOCAL_WRITE |
+			   IB_ACCESS_REMOTE_READ |
+			   IB_ACCESS_REMOTE_WRITE;
 	pi_ctx->sig_mr_valid = 0;
 
 	sig_reg->sge.lkey = pi_ctx->sig_mr->lkey;
@@ -491,7 +489,7 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
 	struct ib_mr *mr = rsc->mr;
 	struct ib_fast_reg_page_list *frpl = rsc->frpl;
 	struct iser_tx_desc *tx_desc = &iser_task->desc;
-	struct ib_send_wr *wr;
+	struct ib_fast_reg_wr *wr;
 	int offset, size, plen;
 
 	plen = iser_sg_to_page_vec(mem, device->ib_device, frpl->page_list,
@@ -501,24 +499,22 @@ static int iser_fast_reg_mr(struct iscsi_iser_task *iser_task,
 		return -EINVAL;
 	}
 
-	if (!rsc->mr_valid) {
-		wr = iser_tx_next_wr(tx_desc);
-		iser_inv_rkey(wr, mr);
-	}
-
-	wr = iser_tx_next_wr(tx_desc);
-	wr->opcode = IB_WR_FAST_REG_MR;
-	wr->wr_id = ISER_FASTREG_LI_WRID;
-	wr->send_flags = 0;
-	wr->wr.fast_reg.iova_start = frpl->page_list[0] + offset;
-	wr->wr.fast_reg.page_list = frpl;
-	wr->wr.fast_reg.page_list_len = plen;
-	wr->wr.fast_reg.page_shift = SHIFT_4K;
-	wr->wr.fast_reg.length = size;
-	wr->wr.fast_reg.rkey = mr->rkey;
-	wr->wr.fast_reg.access_flags = (IB_ACCESS_LOCAL_WRITE  |
-					IB_ACCESS_REMOTE_WRITE |
-					IB_ACCESS_REMOTE_READ);
+	if (!rsc->mr_valid)
+		iser_inv_rkey(iser_tx_next_wr(tx_desc), mr);
+
+	wr = fast_reg_wr(iser_tx_next_wr(tx_desc));
+	wr->wr.opcode = IB_WR_FAST_REG_MR;
+	wr->wr.wr_id = ISER_FASTREG_LI_WRID;
+	wr->wr.send_flags = 0;
+	wr->iova_start = frpl->page_list[0] + offset;
+	wr->page_list = frpl;
+	wr->page_list_len = plen;
+	wr->page_shift = SHIFT_4K;
+	wr->length = size;
+	wr->rkey = mr->rkey;
+	wr->access_flags = (IB_ACCESS_LOCAL_WRITE  |
+			    IB_ACCESS_REMOTE_WRITE |
+			    IB_ACCESS_REMOTE_READ);
 	rsc->mr_valid = 0;
 
 	reg->sge.lkey = mr->lkey;
diff --git a/drivers/infiniband/ulp/iser/iser_verbs.c b/drivers/infiniband/ulp/iser/iser_verbs.c
index f3122372d49f..e7f3b204239b 100644
--- a/drivers/infiniband/ulp/iser/iser_verbs.c
+++ b/drivers/infiniband/ulp/iser/iser_verbs.c
@@ -1135,7 +1135,7 @@ int iser_post_send(struct ib_conn *ib_conn, struct iser_tx_desc *tx_desc,
 	wr->opcode = IB_WR_SEND;
 	wr->send_flags = signal ? IB_SEND_SIGNALED : 0;
 
-	ib_ret = ib_post_send(ib_conn->qp, &tx_desc->wrs[0], &bad_wr);
+	ib_ret = ib_post_send(ib_conn->qp, &tx_desc->wrs[0].send, &bad_wr);
 	if (ib_ret)
 		iser_err("ib_post_send failed, ret:%d opcode:%d\n",
 			 ib_ret, bad_wr->opcode);
diff --git a/drivers/infiniband/ulp/isert/ib_isert.c b/drivers/infiniband/ulp/isert/ib_isert.c
index 7d41c9d6d6bf..1b4d13d323b6 100644
--- a/drivers/infiniband/ulp/isert/ib_isert.c
+++ b/drivers/infiniband/ulp/isert/ib_isert.c
@@ -1715,10 +1715,10 @@ isert_unmap_cmd(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn)
 		isert_unmap_data_buf(isert_conn, &wr->data);
 	}
 
-	if (wr->send_wr) {
+	if (wr->rdma_wr) {
 		isert_dbg("Cmd %p free send_wr\n", isert_cmd);
-		kfree(wr->send_wr);
-		wr->send_wr = NULL;
+		kfree(wr->rdma_wr);
+		wr->rdma_wr = NULL;
 	}
 
 	if (wr->ib_sge) {
@@ -1753,7 +1753,7 @@ isert_unreg_rdma(struct isert_cmd *isert_cmd, struct isert_conn *isert_conn)
 	}
 
 	wr->ib_sge = NULL;
-	wr->send_wr = NULL;
+	wr->rdma_wr = NULL;
 }
 
 static void
@@ -1922,7 +1922,7 @@ isert_completion_rdma_write(struct iser_tx_desc *tx_desc,
 	}
 
 	device->unreg_rdma_mem(isert_cmd, isert_conn);
-	wr->send_wr_num = 0;
+	wr->rdma_wr_num = 0;
 	if (ret)
 		transport_send_check_condition_and_sense(se_cmd,
 							 se_cmd->pi_err, 0);
@@ -1950,7 +1950,7 @@ isert_completion_rdma_read(struct iser_tx_desc *tx_desc,
 	iscsit_stop_dataout_timer(cmd);
 	device->unreg_rdma_mem(isert_cmd, isert_conn);
 	cmd->write_data_done = wr->data.len;
-	wr->send_wr_num = 0;
+	wr->rdma_wr_num = 0;
 
 	isert_dbg("Cmd: %p RDMA_READ comp calling execute_cmd\n", isert_cmd);
 	spin_lock_bh(&cmd->istate_lock);
@@ -2402,7 +2402,7 @@ isert_put_text_rsp(struct iscsi_cmd *cmd, struct iscsi_conn *conn)
 
 static int
 isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
-		    struct ib_sge *ib_sge, struct ib_send_wr *send_wr,
+		    struct ib_sge *ib_sge, struct ib_rdma_wr *rdma_wr,
 		    u32 data_left, u32 offset)
 {
 	struct iscsi_cmd *cmd = isert_cmd->iscsi_cmd;
@@ -2417,8 +2417,8 @@ isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
 	sg_nents = min(cmd->se_cmd.t_data_nents - sg_off, isert_conn->max_sge);
 	page_off = offset % PAGE_SIZE;
 
-	send_wr->sg_list = ib_sge;
-	send_wr->wr_id = (uintptr_t)&isert_cmd->tx_desc;
+	rdma_wr->wr.sg_list = ib_sge;
+	rdma_wr->wr.wr_id = (uintptr_t)&isert_cmd->tx_desc;
 	/*
 	 * Perform mapping of TCM scatterlist memory ib_sge dma_addr.
 	 */
@@ -2443,11 +2443,11 @@ isert_build_rdma_wr(struct isert_conn *isert_conn, struct isert_cmd *isert_cmd,
 		isert_dbg("Incrementing ib_sge pointer to %p\n", ib_sge);
 	}
 
-	send_wr->num_sge = ++i;
+	rdma_wr->wr.num_sge = ++i;
 	isert_dbg("Set outgoing sg_list: %p num_sg: %u from TCM SGLs\n",
-		  send_wr->sg_list, send_wr->num_sge);
+		  rdma_wr->wr.sg_list, rdma_wr->wr.num_sge);
 
-	return send_wr->num_sge;
+	return rdma_wr->wr.num_sge;
 }
 
 static int
@@ -2458,7 +2458,7 @@ isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
 	struct isert_conn *isert_conn = conn->context;
 	struct isert_data_buf *data = &wr->data;
-	struct ib_send_wr *send_wr;
+	struct ib_rdma_wr *rdma_wr;
 	struct ib_sge *ib_sge;
 	u32 offset, data_len, data_left, rdma_write_max, va_offset = 0;
 	int ret = 0, i, ib_sge_cnt;
@@ -2483,11 +2483,11 @@ isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
 	}
 	wr->ib_sge = ib_sge;
 
-	wr->send_wr_num = DIV_ROUND_UP(data->nents, isert_conn->max_sge);
-	wr->send_wr = kzalloc(sizeof(struct ib_send_wr) * wr->send_wr_num,
+	wr->rdma_wr_num = DIV_ROUND_UP(data->nents, isert_conn->max_sge);
+	wr->rdma_wr = kzalloc(sizeof(struct ib_rdma_wr) * wr->rdma_wr_num,
 				GFP_KERNEL);
-	if (!wr->send_wr) {
-		isert_dbg("Unable to allocate wr->send_wr\n");
+	if (!wr->rdma_wr) {
+		isert_dbg("Unable to allocate wr->rdma_wr\n");
 		ret = -ENOMEM;
 		goto unmap_cmd;
 	}
@@ -2495,31 +2495,31 @@ isert_map_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
 	wr->isert_cmd = isert_cmd;
 	rdma_write_max = isert_conn->max_sge * PAGE_SIZE;
 
-	for (i = 0; i < wr->send_wr_num; i++) {
-		send_wr = &isert_cmd->rdma_wr.send_wr[i];
+	for (i = 0; i < wr->rdma_wr_num; i++) {
+		rdma_wr = &isert_cmd->rdma_wr.rdma_wr[i];
 		data_len = min(data_left, rdma_write_max);
 
-		send_wr->send_flags = 0;
+		rdma_wr->wr.send_flags = 0;
 		if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) {
-			send_wr->opcode = IB_WR_RDMA_WRITE;
-			send_wr->wr.rdma.remote_addr = isert_cmd->read_va + offset;
-			send_wr->wr.rdma.rkey = isert_cmd->read_stag;
-			if (i + 1 == wr->send_wr_num)
-				send_wr->next = &isert_cmd->tx_desc.send_wr;
+			rdma_wr->wr.opcode = IB_WR_RDMA_WRITE;
+			rdma_wr->remote_addr = isert_cmd->read_va + offset;
+			rdma_wr->rkey = isert_cmd->read_stag;
+			if (i + 1 == wr->rdma_wr_num)
+				rdma_wr->wr.next = &isert_cmd->tx_desc.send_wr;
 			else
-				send_wr->next = &wr->send_wr[i + 1];
+				rdma_wr->wr.next = &wr->rdma_wr[i + 1].wr;
 		} else {
-			send_wr->opcode = IB_WR_RDMA_READ;
-			send_wr->wr.rdma.remote_addr = isert_cmd->write_va + va_offset;
-			send_wr->wr.rdma.rkey = isert_cmd->write_stag;
-			if (i + 1 == wr->send_wr_num)
-				send_wr->send_flags = IB_SEND_SIGNALED;
+			rdma_wr->wr.opcode = IB_WR_RDMA_READ;
+			rdma_wr->remote_addr = isert_cmd->write_va + va_offset;
+			rdma_wr->rkey = isert_cmd->write_stag;
+			if (i + 1 == wr->rdma_wr_num)
+				rdma_wr->wr.send_flags = IB_SEND_SIGNALED;
 			else
-				send_wr->next = &wr->send_wr[i + 1];
+				rdma_wr->wr.next = &wr->rdma_wr[i + 1].wr;
 		}
 
 		ib_sge_cnt = isert_build_rdma_wr(isert_conn, isert_cmd, ib_sge,
-					send_wr, data_len, offset);
+					rdma_wr, data_len, offset);
 		ib_sge += ib_sge_cnt;
 
 		offset += data_len;
@@ -2599,8 +2599,8 @@ isert_fast_reg_mr(struct isert_conn *isert_conn,
 	struct ib_device *ib_dev = device->ib_device;
 	struct ib_mr *mr;
 	struct ib_fast_reg_page_list *frpl;
-	struct ib_send_wr fr_wr, inv_wr;
-	struct ib_send_wr *bad_wr, *wr = NULL;
+	struct ib_fast_reg_wr fr_wr;
+	struct ib_send_wr inv_wr, *bad_wr, *wr = NULL;
 	int ret, pagelist_len;
 	u32 page_off;
 
@@ -2638,20 +2638,20 @@ isert_fast_reg_mr(struct isert_conn *isert_conn,
 
 	/* Prepare FASTREG WR */
 	memset(&fr_wr, 0, sizeof(fr_wr));
-	fr_wr.wr_id = ISER_FASTREG_LI_WRID;
-	fr_wr.opcode = IB_WR_FAST_REG_MR;
-	fr_wr.wr.fast_reg.iova_start = frpl->page_list[0] + page_off;
-	fr_wr.wr.fast_reg.page_list = frpl;
-	fr_wr.wr.fast_reg.page_list_len = pagelist_len;
-	fr_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
-	fr_wr.wr.fast_reg.length = mem->len;
-	fr_wr.wr.fast_reg.rkey = mr->rkey;
-	fr_wr.wr.fast_reg.access_flags = IB_ACCESS_LOCAL_WRITE;
+	fr_wr.wr.wr_id = ISER_FASTREG_LI_WRID;
+	fr_wr.wr.opcode = IB_WR_FAST_REG_MR;
+	fr_wr.iova_start = frpl->page_list[0] + page_off;
+	fr_wr.page_list = frpl;
+	fr_wr.page_list_len = pagelist_len;
+	fr_wr.page_shift = PAGE_SHIFT;
+	fr_wr.length = mem->len;
+	fr_wr.rkey = mr->rkey;
+	fr_wr.access_flags = IB_ACCESS_LOCAL_WRITE;
 
 	if (!wr)
-		wr = &fr_wr;
+		wr = &fr_wr.wr;
 	else
-		wr->next = &fr_wr;
+		wr->next = &fr_wr.wr;
 
 	ret = ib_post_send(isert_conn->qp, wr, &bad_wr);
 	if (ret) {
@@ -2732,8 +2732,8 @@ isert_reg_sig_mr(struct isert_conn *isert_conn,
 		 struct isert_rdma_wr *rdma_wr,
 		 struct fast_reg_descriptor *fr_desc)
 {
-	struct ib_send_wr sig_wr, inv_wr;
-	struct ib_send_wr *bad_wr, *wr = NULL;
+	struct ib_sig_handover_wr sig_wr;
+	struct ib_send_wr inv_wr, *bad_wr, *wr = NULL;
 	struct pi_context *pi_ctx = fr_desc->pi_ctx;
 	struct ib_sig_attrs sig_attrs;
 	int ret;
@@ -2751,20 +2751,20 @@ isert_reg_sig_mr(struct isert_conn *isert_conn,
 	}
 
 	memset(&sig_wr, 0, sizeof(sig_wr));
-	sig_wr.opcode = IB_WR_REG_SIG_MR;
-	sig_wr.wr_id = ISER_FASTREG_LI_WRID;
-	sig_wr.sg_list = &rdma_wr->ib_sg[DATA];
-	sig_wr.num_sge = 1;
-	sig_wr.wr.sig_handover.access_flags = IB_ACCESS_LOCAL_WRITE;
-	sig_wr.wr.sig_handover.sig_attrs = &sig_attrs;
-	sig_wr.wr.sig_handover.sig_mr = pi_ctx->sig_mr;
+	sig_wr.wr.opcode = IB_WR_REG_SIG_MR;
+	sig_wr.wr.wr_id = ISER_FASTREG_LI_WRID;
+	sig_wr.wr.sg_list = &rdma_wr->ib_sg[DATA];
+	sig_wr.wr.num_sge = 1;
+	sig_wr.access_flags = IB_ACCESS_LOCAL_WRITE;
+	sig_wr.sig_attrs = &sig_attrs;
+	sig_wr.sig_mr = pi_ctx->sig_mr;
 	if (se_cmd->t_prot_sg)
-		sig_wr.wr.sig_handover.prot = &rdma_wr->ib_sg[PROT];
+		sig_wr.prot = &rdma_wr->ib_sg[PROT];
 
 	if (!wr)
-		wr = &sig_wr;
+		wr = &sig_wr.wr;
 	else
-		wr->next = &sig_wr;
+		wr->next = &sig_wr.wr;
 
 	ret = ib_post_send(isert_conn->qp, wr, &bad_wr);
 	if (ret) {
@@ -2858,7 +2858,7 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
 	struct isert_cmd *isert_cmd = iscsit_priv_cmd(cmd);
 	struct isert_conn *isert_conn = conn->context;
 	struct fast_reg_descriptor *fr_desc = NULL;
-	struct ib_send_wr *send_wr;
+	struct ib_rdma_wr *rdma_wr;
 	struct ib_sge *ib_sg;
 	u32 offset;
 	int ret = 0;
@@ -2899,26 +2899,26 @@ isert_reg_rdma(struct iscsi_conn *conn, struct iscsi_cmd *cmd,
 
 	memcpy(&wr->s_ib_sge, ib_sg, sizeof(*ib_sg));
 	wr->ib_sge = &wr->s_ib_sge;
-	wr->send_wr_num = 1;
-	memset(&wr->s_send_wr, 0, sizeof(*send_wr));
-	wr->send_wr = &wr->s_send_wr;
+	wr->rdma_wr_num = 1;
+	memset(&wr->s_rdma_wr, 0, sizeof(wr->s_rdma_wr));
+	wr->rdma_wr = &wr->s_rdma_wr;
 	wr->isert_cmd = isert_cmd;
 
-	send_wr = &isert_cmd->rdma_wr.s_send_wr;
-	send_wr->sg_list = &wr->s_ib_sge;
-	send_wr->num_sge = 1;
-	send_wr->wr_id = (uintptr_t)&isert_cmd->tx_desc;
+	rdma_wr = &isert_cmd->rdma_wr.s_rdma_wr;
+	rdma_wr->wr.sg_list = &wr->s_ib_sge;
+	rdma_wr->wr.num_sge = 1;
+	rdma_wr->wr.wr_id = (uintptr_t)&isert_cmd->tx_desc;
 	if (wr->iser_ib_op == ISER_IB_RDMA_WRITE) {
-		send_wr->opcode = IB_WR_RDMA_WRITE;
-		send_wr->wr.rdma.remote_addr = isert_cmd->read_va;
-		send_wr->wr.rdma.rkey = isert_cmd->read_stag;
-		send_wr->send_flags = !isert_prot_cmd(isert_conn, se_cmd) ?
+		rdma_wr->wr.opcode = IB_WR_RDMA_WRITE;
+		rdma_wr->remote_addr = isert_cmd->read_va;
+		rdma_wr->rkey = isert_cmd->read_stag;
+		rdma_wr->wr.send_flags = !isert_prot_cmd(isert_conn, se_cmd) ?
 				      0 : IB_SEND_SIGNALED;
 	} else {
-		send_wr->opcode = IB_WR_RDMA_READ;
-		send_wr->wr.rdma.remote_addr = isert_cmd->write_va;
-		send_wr->wr.rdma.rkey = isert_cmd->write_stag;
-		send_wr->send_flags = IB_SEND_SIGNALED;
+		rdma_wr->wr.opcode = IB_WR_RDMA_READ;
+		rdma_wr->remote_addr = isert_cmd->write_va;
+		rdma_wr->rkey = isert_cmd->write_stag;
+		rdma_wr->wr.send_flags = IB_SEND_SIGNALED;
 	}
 
 	return 0;
@@ -2966,8 +2966,8 @@ isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
 		isert_init_tx_hdrs(isert_conn, &isert_cmd->tx_desc);
 		isert_init_send_wr(isert_conn, isert_cmd,
 				   &isert_cmd->tx_desc.send_wr);
-		isert_cmd->rdma_wr.s_send_wr.next = &isert_cmd->tx_desc.send_wr;
-		wr->send_wr_num += 1;
+		isert_cmd->rdma_wr.s_rdma_wr.wr.next = &isert_cmd->tx_desc.send_wr;
+		wr->rdma_wr_num += 1;
 
 		rc = isert_post_recv(isert_conn, isert_cmd->rx_desc);
 		if (rc) {
@@ -2976,7 +2976,7 @@ isert_put_datain(struct iscsi_conn *conn, struct iscsi_cmd *cmd)
 		}
 	}
 
-	rc = ib_post_send(isert_conn->qp, wr->send_wr, &wr_failed);
+	rc = ib_post_send(isert_conn->qp, &wr->rdma_wr->wr, &wr_failed);
 	if (rc)
 		isert_warn("ib_post_send() failed for IB_WR_RDMA_WRITE\n");
 
@@ -3010,7 +3010,7 @@ isert_get_dataout(struct iscsi_conn *conn, struct iscsi_cmd *cmd, bool recovery)
 		return rc;
 	}
 
-	rc = ib_post_send(isert_conn->qp, wr->send_wr, &wr_failed);
+	rc = ib_post_send(isert_conn->qp, &wr->rdma_wr->wr, &wr_failed);
 	if (rc)
 		isert_warn("ib_post_send() failed for IB_WR_RDMA_READ\n");
 
diff --git a/drivers/infiniband/ulp/isert/ib_isert.h b/drivers/infiniband/ulp/isert/ib_isert.h
index c5b99bcecbcf..360461819452 100644
--- a/drivers/infiniband/ulp/isert/ib_isert.h
+++ b/drivers/infiniband/ulp/isert/ib_isert.h
@@ -117,9 +117,9 @@ struct isert_rdma_wr {
 	enum iser_ib_op_code	iser_ib_op;
 	struct ib_sge		*ib_sge;
 	struct ib_sge		s_ib_sge;
-	int			send_wr_num;
-	struct ib_send_wr	*send_wr;
-	struct ib_send_wr	s_send_wr;
+	int			rdma_wr_num;
+	struct ib_rdma_wr	*rdma_wr;
+	struct ib_rdma_wr	s_rdma_wr;
 	struct ib_sge		ib_sg[3];
 	struct isert_data_buf	data;
 	struct isert_data_buf	prot;
diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 8ba887824d05..235f9b8b27f1 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -1306,7 +1306,7 @@ static int srp_map_finish_fr(struct srp_map_state *state,
 	struct srp_target_port *target = ch->target;
 	struct srp_device *dev = target->srp_host->srp_dev;
 	struct ib_send_wr *bad_wr;
-	struct ib_send_wr wr;
+	struct ib_fast_reg_wr wr;
 	struct srp_fr_desc *desc;
 	u32 rkey;
 
@@ -1324,17 +1324,17 @@ static int srp_map_finish_fr(struct srp_map_state *state,
 	       sizeof(state->pages[0]) * state->npages);
 
 	memset(&wr, 0, sizeof(wr));
-	wr.opcode = IB_WR_FAST_REG_MR;
-	wr.wr_id = FAST_REG_WR_ID_MASK;
-	wr.wr.fast_reg.iova_start = state->base_dma_addr;
-	wr.wr.fast_reg.page_list = desc->frpl;
-	wr.wr.fast_reg.page_list_len = state->npages;
-	wr.wr.fast_reg.page_shift = ilog2(dev->mr_page_size);
-	wr.wr.fast_reg.length = state->dma_len;
-	wr.wr.fast_reg.access_flags = (IB_ACCESS_LOCAL_WRITE |
-				       IB_ACCESS_REMOTE_READ |
-				       IB_ACCESS_REMOTE_WRITE);
-	wr.wr.fast_reg.rkey = desc->mr->lkey;
+	wr.wr.opcode = IB_WR_FAST_REG_MR;
+	wr.wr.wr_id = FAST_REG_WR_ID_MASK;
+	wr.iova_start = state->base_dma_addr;
+	wr.page_list = desc->frpl;
+	wr.page_list_len = state->npages;
+	wr.page_shift = ilog2(dev->mr_page_size);
+	wr.length = state->dma_len;
+	wr.access_flags = (IB_ACCESS_LOCAL_WRITE |
+			   IB_ACCESS_REMOTE_READ |
+			   IB_ACCESS_REMOTE_WRITE);
+	wr.rkey = desc->mr->lkey;
 
 	*state->fr.next++ = desc;
 	state->nmdesc++;
@@ -1342,7 +1342,7 @@ static int srp_map_finish_fr(struct srp_map_state *state,
 	srp_map_desc(state, state->base_dma_addr, state->dma_len,
 		     desc->mr->rkey);
 
-	return ib_post_send(ch->qp, &wr, &bad_wr);
+	return ib_post_send(ch->qp, &wr.wr, &bad_wr);
 }
 
 static int srp_finish_mapping(struct srp_map_state *state,
diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
index a7ac77a02593..47c4022fda76 100644
--- a/drivers/infiniband/ulp/srpt/ib_srpt.c
+++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
@@ -2823,7 +2823,7 @@ static int srpt_cm_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
 static int srpt_perform_rdmas(struct srpt_rdma_ch *ch,
 			      struct srpt_send_ioctx *ioctx)
 {
-	struct ib_send_wr wr;
+	struct ib_rdma_wr wr;
 	struct ib_send_wr *bad_wr;
 	struct rdma_iu *riu;
 	int i;
@@ -2851,29 +2851,29 @@ static int srpt_perform_rdmas(struct srpt_rdma_ch *ch,
 
 	for (i = 0; i < n_rdma; ++i, ++riu) {
 		if (dir == DMA_FROM_DEVICE) {
-			wr.opcode = IB_WR_RDMA_WRITE;
-			wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
+			wr.wr.opcode = IB_WR_RDMA_WRITE;
+			wr.wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
 						SRPT_RDMA_WRITE_LAST :
 						SRPT_RDMA_MID,
 						ioctx->ioctx.index);
 		} else {
-			wr.opcode = IB_WR_RDMA_READ;
-			wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
+			wr.wr.opcode = IB_WR_RDMA_READ;
+			wr.wr.wr_id = encode_wr_id(i == n_rdma - 1 ?
 						SRPT_RDMA_READ_LAST :
 						SRPT_RDMA_MID,
 						ioctx->ioctx.index);
 		}
-		wr.next = NULL;
-		wr.wr.rdma.remote_addr = riu->raddr;
-		wr.wr.rdma.rkey = riu->rkey;
-		wr.num_sge = riu->sge_cnt;
-		wr.sg_list = riu->sge;
+		wr.wr.next = NULL;
+		wr.remote_addr = riu->raddr;
+		wr.rkey = riu->rkey;
+		wr.wr.num_sge = riu->sge_cnt;
+		wr.wr.sg_list = riu->sge;
 
 		/* only get completion event for the last rdma write */
 		if (i == (n_rdma - 1) && dir == DMA_TO_DEVICE)
-			wr.send_flags = IB_SEND_SIGNALED;
+			wr.wr.send_flags = IB_SEND_SIGNALED;
 
-		ret = ib_post_send(ch->qp, &wr, &bad_wr);
+		ret = ib_post_send(ch->qp, &wr.wr, &bad_wr);
 		if (ret)
 			break;
 	}
@@ -2882,11 +2882,11 @@ static int srpt_perform_rdmas(struct srpt_rdma_ch *ch,
 		pr_err("%s[%d]: ib_post_send() returned %d for %d/%d\n",
 				 __func__, __LINE__, ret, i, n_rdma);
 	if (ret && i > 0) {
-		wr.num_sge = 0;
-		wr.wr_id = encode_wr_id(SRPT_RDMA_ABORT, ioctx->ioctx.index);
-		wr.send_flags = IB_SEND_SIGNALED;
+		wr.wr.num_sge = 0;
+		wr.wr.wr_id = encode_wr_id(SRPT_RDMA_ABORT, ioctx->ioctx.index);
+		wr.wr.send_flags = IB_SEND_SIGNALED;
 		while (ch->state == CH_LIVE &&
-			ib_post_send(ch->qp, &wr, &bad_wr) != 0) {
+			ib_post_send(ch->qp, &wr.wr, &bad_wr) != 0) {
 			pr_info("Trying to abort failed RDMA transfer [%d]\n",
 				ioctx->ioctx.index);
 			msleep(1000);
diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h
index 5015a9e830bd..993d1ff9ba21 100644
--- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h
+++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd.h
@@ -527,7 +527,7 @@ typedef struct kib_tx                         /* transmit message */
 	__u64                  tx_msgaddr;    /* message buffer (I/O addr) */
 	DECLARE_PCI_UNMAP_ADDR(tx_msgunmap);  /* for dma_unmap_single() */
 	int                    tx_nwrq;       /* # send work items */
-	struct ib_send_wr      *tx_wrq;       /* send work items... */
+	struct ib_rdma_wr      *tx_wrq;       /* send work items... */
 	struct ib_sge          *tx_sge;       /* ...and their memory */
 	kib_rdma_desc_t        *tx_rd;        /* rdma descriptor */
 	int                    tx_nfrags;     /* # entries in... */
diff --git a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
index a23a6d956a4d..a34f1707c167 100644
--- a/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
+++ b/drivers/staging/lustre/lnet/klnds/o2iblnd/o2iblnd_cb.c
@@ -834,7 +834,7 @@ kiblnd_post_tx_locked(kib_conn_t *conn, kib_tx_t *tx, int credit)
 		/* close_conn will launch failover */
 		rc = -ENETDOWN;
 	} else {
-		rc = ib_post_send(conn->ibc_cmid->qp, tx->tx_wrq, &bad_wrq);
+		rc = ib_post_send(conn->ibc_cmid->qp, &tx->tx_wrq->wr, &bad_wrq);
 	}
 
 	conn->ibc_last_send = jiffies;
@@ -1008,7 +1008,7 @@ kiblnd_init_tx_msg(lnet_ni_t *ni, kib_tx_t *tx, int type, int body_nob)
 {
 	kib_hca_dev_t *hdev = tx->tx_pool->tpo_hdev;
 	struct ib_sge *sge = &tx->tx_sge[tx->tx_nwrq];
-	struct ib_send_wr *wrq = &tx->tx_wrq[tx->tx_nwrq];
+	struct ib_rdma_wr *wrq = &tx->tx_wrq[tx->tx_nwrq];
 	int nob = offsetof(kib_msg_t, ibm_u) + body_nob;
 	struct ib_mr *mr;
 
@@ -1027,12 +1027,12 @@ kiblnd_init_tx_msg(lnet_ni_t *ni, kib_tx_t *tx, int type, int body_nob)
 
 	memset(wrq, 0, sizeof(*wrq));
 
-	wrq->next       = NULL;
-	wrq->wr_id      = kiblnd_ptr2wreqid(tx, IBLND_WID_TX);
-	wrq->sg_list    = sge;
-	wrq->num_sge    = 1;
-	wrq->opcode     = IB_WR_SEND;
-	wrq->send_flags = IB_SEND_SIGNALED;
+	wrq->wr.next       = NULL;
+	wrq->wr.wr_id      = kiblnd_ptr2wreqid(tx, IBLND_WID_TX);
+	wrq->wr.sg_list    = sge;
+	wrq->wr.num_sge    = 1;
+	wrq->wr.opcode     = IB_WR_SEND;
+	wrq->wr.send_flags = IB_SEND_SIGNALED;
 
 	tx->tx_nwrq++;
 }
@@ -1044,7 +1044,7 @@ kiblnd_init_rdma(kib_conn_t *conn, kib_tx_t *tx, int type,
 	kib_msg_t *ibmsg = tx->tx_msg;
 	kib_rdma_desc_t *srcrd = tx->tx_rd;
 	struct ib_sge *sge = &tx->tx_sge[0];
-	struct ib_send_wr *wrq = &tx->tx_wrq[0];
+	struct ib_rdma_wr *wrq = &tx->tx_wrq[0], *next;
 	int rc  = resid;
 	int srcidx;
 	int dstidx;
@@ -1090,16 +1090,17 @@ kiblnd_init_rdma(kib_conn_t *conn, kib_tx_t *tx, int type,
 		sge->length = wrknob;
 
 		wrq = &tx->tx_wrq[tx->tx_nwrq];
+		next = wrq + 1;
 
-		wrq->next       = wrq + 1;
-		wrq->wr_id      = kiblnd_ptr2wreqid(tx, IBLND_WID_RDMA);
-		wrq->sg_list    = sge;
-		wrq->num_sge    = 1;
-		wrq->opcode     = IB_WR_RDMA_WRITE;
-		wrq->send_flags = 0;
+		wrq->wr.next       = &next->wr;
+		wrq->wr.wr_id      = kiblnd_ptr2wreqid(tx, IBLND_WID_RDMA);
+		wrq->wr.sg_list    = sge;
+		wrq->wr.num_sge    = 1;
+		wrq->wr.opcode     = IB_WR_RDMA_WRITE;
+		wrq->wr.send_flags = 0;
 
-		wrq->wr.rdma.remote_addr = kiblnd_rd_frag_addr(dstrd, dstidx);
-		wrq->wr.rdma.rkey        = kiblnd_rd_frag_key(dstrd, dstidx);
+		wrq->remote_addr = kiblnd_rd_frag_addr(dstrd, dstidx);
+		wrq->rkey        = kiblnd_rd_frag_key(dstrd, dstidx);
 
 		srcidx = kiblnd_rd_consume_frag(srcrd, srcidx, wrknob);
 		dstidx = kiblnd_rd_consume_frag(dstrd, dstidx, wrknob);
diff --git a/drivers/staging/rdma/amso1100/c2_qp.c b/drivers/staging/rdma/amso1100/c2_qp.c
index 86708dee58b1..4c43ca935cc7 100644
--- a/drivers/staging/rdma/amso1100/c2_qp.c
+++ b/drivers/staging/rdma/amso1100/c2_qp.c
@@ -860,9 +860,9 @@ int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 				flags |= SQ_READ_FENCE;
 			}
 			wr.sqwr.rdma_write.remote_stag =
-			    cpu_to_be32(ib_wr->wr.rdma.rkey);
+			    cpu_to_be32(rdma_wr(ib_wr)->rkey);
 			wr.sqwr.rdma_write.remote_to =
-			    cpu_to_be64(ib_wr->wr.rdma.remote_addr);
+			    cpu_to_be64(rdma_wr(ib_wr)->remote_addr);
 			err = move_sgl((struct c2_data_addr *)
 				       & (wr.sqwr.rdma_write.data),
 				       ib_wr->sg_list,
@@ -889,9 +889,9 @@ int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 			wr.sqwr.rdma_read.local_to =
 			    cpu_to_be64(ib_wr->sg_list->addr);
 			wr.sqwr.rdma_read.remote_stag =
-			    cpu_to_be32(ib_wr->wr.rdma.rkey);
+			    cpu_to_be32(rdma_wr(ib_wr)->rkey);
 			wr.sqwr.rdma_read.remote_to =
-			    cpu_to_be64(ib_wr->wr.rdma.remote_addr);
+			    cpu_to_be64(rdma_wr(ib_wr)->remote_addr);
 			wr.sqwr.rdma_read.length =
 			    cpu_to_be32(ib_wr->sg_list->length);
 			break;
diff --git a/drivers/staging/rdma/ehca/ehca_reqs.c b/drivers/staging/rdma/ehca/ehca_reqs.c
index 47f94984353d..10e2074384f5 100644
--- a/drivers/staging/rdma/ehca/ehca_reqs.c
+++ b/drivers/staging/rdma/ehca/ehca_reqs.c
@@ -110,19 +110,19 @@ static inline int ehca_write_rwqe(struct ipz_queue *ipz_rqueue,
 /* need ib_mad struct */
 #include <rdma/ib_mad.h>
 
-static void trace_send_wr_ud(const struct ib_send_wr *send_wr)
+static void trace_ud_wr(const struct ib_ud_wr *ud_wr)
 {
 	int idx;
 	int j;
-	while (send_wr) {
-		struct ib_mad_hdr *mad_hdr = send_wr->wr.ud.mad_hdr;
-		struct ib_sge *sge = send_wr->sg_list;
-		ehca_gen_dbg("send_wr#%x wr_id=%lx num_sge=%x "
-			     "send_flags=%x opcode=%x", idx, send_wr->wr_id,
-			     send_wr->num_sge, send_wr->send_flags,
-			     send_wr->opcode);
+	while (ud_wr) {
+		struct ib_mad_hdr *mad_hdr = ud_wrmad_hdr;
+		struct ib_sge *sge = ud_wr->wr.sg_list;
+		ehca_gen_dbg("ud_wr#%x wr_id=%lx num_sge=%x "
+			     "send_flags=%x opcode=%x", idx, ud_wr->wr.wr_id,
+			     ud_wr->wr.num_sge, ud_wr->wr.send_flags,
+			     ud_wr->.wr.opcode);
 		if (mad_hdr) {
-			ehca_gen_dbg("send_wr#%x mad_hdr base_version=%x "
+			ehca_gen_dbg("ud_wr#%x mad_hdr base_version=%x "
 				     "mgmt_class=%x class_version=%x method=%x "
 				     "status=%x class_specific=%x tid=%lx "
 				     "attr_id=%x resv=%x attr_mod=%x",
@@ -134,33 +134,33 @@ static void trace_send_wr_ud(const struct ib_send_wr *send_wr)
 				     mad_hdr->resv,
 				     mad_hdr->attr_mod);
 		}
-		for (j = 0; j < send_wr->num_sge; j++) {
+		for (j = 0; j < ud_wr->wr.num_sge; j++) {
 			u8 *data = __va(sge->addr);
-			ehca_gen_dbg("send_wr#%x sge#%x addr=%p length=%x "
+			ehca_gen_dbg("ud_wr#%x sge#%x addr=%p length=%x "
 				     "lkey=%x",
 				     idx, j, data, sge->length, sge->lkey);
 			/* assume length is n*16 */
-			ehca_dmp(data, sge->length, "send_wr#%x sge#%x",
+			ehca_dmp(data, sge->length, "ud_wr#%x sge#%x",
 				 idx, j);
 			sge++;
 		} /* eof for j */
 		idx++;
-		send_wr = send_wr->next;
-	} /* eof while send_wr */
+		ud_wr = ud_wr(ud_wr->wr.next);
+	} /* eof while ud_wr */
 }
 
 #endif /* DEBUG_GSI_SEND_WR */
 
 static inline int ehca_write_swqe(struct ehca_qp *qp,
 				  struct ehca_wqe *wqe_p,
-				  const struct ib_send_wr *send_wr,
+				  struct ib_send_wr *send_wr,
 				  u32 sq_map_idx,
 				  int hidden)
 {
 	u32 idx;
 	u64 dma_length;
 	struct ehca_av *my_av;
-	u32 remote_qkey = send_wr->wr.ud.remote_qkey;
+	u32 remote_qkey;
 	struct ehca_qmap_entry *qmap_entry = &qp->sq_map.map[sq_map_idx];
 
 	if (unlikely((send_wr->num_sge < 0) ||
@@ -223,20 +223,21 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 		/* no break is intential here */
 	case IB_QPT_UD:
 		/* IB 1.2 spec C10-15 compliance */
-		if (send_wr->wr.ud.remote_qkey & 0x80000000)
+		remote_qkey = ud_wr(send_wr)->remote_qkey;
+		if (remote_qkey & 0x80000000)
 			remote_qkey = qp->qkey;
 
-		wqe_p->destination_qp_number = send_wr->wr.ud.remote_qpn << 8;
+		wqe_p->destination_qp_number = ud_wr(send_wr)->remote_qpn << 8;
 		wqe_p->local_ee_context_qkey = remote_qkey;
-		if (unlikely(!send_wr->wr.ud.ah)) {
-			ehca_gen_err("wr.ud.ah is NULL. qp=%p", qp);
+		if (unlikely(!ud_wr(send_wr)->ah)) {
+			ehca_gen_err("ud_wr(send_wr) is NULL. qp=%p", qp);
 			return -EINVAL;
 		}
-		if (unlikely(send_wr->wr.ud.remote_qpn == 0)) {
+		if (unlikely(ud_wr(send_wr)->remote_qpn == 0)) {
 			ehca_gen_err("dest QP# is 0. qp=%x", qp->real_qp_num);
 			return -EINVAL;
 		}
-		my_av = container_of(send_wr->wr.ud.ah, struct ehca_av, ib_ah);
+		my_av = container_of(ud_wr(send_wr)->ah, struct ehca_av, ib_ah);
 		wqe_p->u.ud_av.ud_av = my_av->av;
 
 		/*
@@ -255,9 +256,9 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 		    qp->qp_type == IB_QPT_GSI)
 			wqe_p->u.ud_av.ud_av.pmtu = 1;
 		if (qp->qp_type == IB_QPT_GSI) {
-			wqe_p->pkeyi = send_wr->wr.ud.pkey_index;
+			wqe_p->pkeyi = ud_wr(send_wr)->pkey_index;
 #ifdef DEBUG_GSI_SEND_WR
-			trace_send_wr_ud(send_wr);
+			trace_ud_wr(ud_wr(send_wr));
 #endif /* DEBUG_GSI_SEND_WR */
 		}
 		break;
@@ -269,8 +270,8 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 	case IB_QPT_RC:
 		/* TODO: atomic not implemented */
 		wqe_p->u.nud.remote_virtual_address =
-			send_wr->wr.rdma.remote_addr;
-		wqe_p->u.nud.rkey = send_wr->wr.rdma.rkey;
+			rdma_wr(send_wr)->remote_addr;
+		wqe_p->u.nud.rkey = rdma_wr(send_wr)->rkey;
 
 		/*
 		 * omitted checking of IB_SEND_INLINE
diff --git a/drivers/staging/rdma/hfi1/keys.c b/drivers/staging/rdma/hfi1/keys.c
index f6eff177ace1..82c21b1c0263 100644
--- a/drivers/staging/rdma/hfi1/keys.c
+++ b/drivers/staging/rdma/hfi1/keys.c
@@ -358,12 +358,12 @@ bail:
 /*
  * Initialize the memory region specified by the work request.
  */
-int hfi1_fast_reg_mr(struct hfi1_qp *qp, struct ib_send_wr *wr)
+int hfi1_fast_reg_mr(struct hfi1_qp *qp, struct ib_fast_reg_wr *wr)
 {
 	struct hfi1_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
 	struct hfi1_pd *pd = to_ipd(qp->ibqp.pd);
 	struct hfi1_mregion *mr;
-	u32 rkey = wr->wr.fast_reg.rkey;
+	u32 rkey = wr->rkey;
 	unsigned i, n, m;
 	int ret = -EINVAL;
 	unsigned long flags;
@@ -380,22 +380,22 @@ int hfi1_fast_reg_mr(struct hfi1_qp *qp, struct ib_send_wr *wr)
 	if (unlikely(mr == NULL || qp->ibqp.pd != mr->pd))
 		goto bail;
 
-	if (wr->wr.fast_reg.page_list_len > mr->max_segs)
+	if (wr->page_list_len > mr->max_segs)
 		goto bail;
 
-	ps = 1UL << wr->wr.fast_reg.page_shift;
-	if (wr->wr.fast_reg.length > ps * wr->wr.fast_reg.page_list_len)
+	ps = 1UL << wr->page_shift;
+	if (wr->length > ps * wr->page_list_len)
 		goto bail;
 
-	mr->user_base = wr->wr.fast_reg.iova_start;
-	mr->iova = wr->wr.fast_reg.iova_start;
+	mr->user_base = wr->iova_start;
+	mr->iova = wr->iova_start;
 	mr->lkey = rkey;
-	mr->length = wr->wr.fast_reg.length;
-	mr->access_flags = wr->wr.fast_reg.access_flags;
-	page_list = wr->wr.fast_reg.page_list->page_list;
+	mr->length = wr->length;
+	mr->access_flags = wr->access_flags;
+	page_list = wr->page_list->page_list;
 	m = 0;
 	n = 0;
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+	for (i = 0; i < wr->page_list_len; i++) {
 		mr->map[m]->segs[n].vaddr = (void *) page_list[i];
 		mr->map[m]->segs[n].length = ps;
 		if (++n == HFI1_SEGSZ) {
diff --git a/drivers/staging/rdma/hfi1/qp.c b/drivers/staging/rdma/hfi1/qp.c
index df1fa56eaf85..f8c36166962f 100644
--- a/drivers/staging/rdma/hfi1/qp.c
+++ b/drivers/staging/rdma/hfi1/qp.c
@@ -422,7 +422,7 @@ static void clear_mr_refs(struct hfi1_qp *qp, int clr_sends)
 			if (qp->ibqp.qp_type == IB_QPT_UD ||
 			    qp->ibqp.qp_type == IB_QPT_SMI ||
 			    qp->ibqp.qp_type == IB_QPT_GSI)
-				atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+				atomic_dec(&to_iah(wqe->ud_wr.ah)->refcount);
 			if (++qp->s_last >= qp->s_size)
 				qp->s_last = 0;
 		}
diff --git a/drivers/staging/rdma/hfi1/rc.c b/drivers/staging/rdma/hfi1/rc.c
index 632dd5ba7dfd..fd0ac608c62d 100644
--- a/drivers/staging/rdma/hfi1/rc.c
+++ b/drivers/staging/rdma/hfi1/rc.c
@@ -404,9 +404,9 @@ int hfi1_make_rc_req(struct hfi1_qp *qp)
 				goto bail;
 			}
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			hwords += sizeof(struct ib_reth) / sizeof(u32);
 			wqe->lpsn = wqe->psn;
@@ -455,9 +455,9 @@ int hfi1_make_rc_req(struct hfi1_qp *qp)
 				wqe->lpsn = qp->s_next_psn++;
 			}
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			qp->s_state = OP(RDMA_READ_REQUEST);
 			hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
@@ -488,21 +488,21 @@ int hfi1_make_rc_req(struct hfi1_qp *qp)
 			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
 				qp->s_state = OP(COMPARE_SWAP);
 				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
-					wqe->wr.wr.atomic.swap);
+					wqe->atomic_wr.swap);
 				ohdr->u.atomic_eth.compare_data = cpu_to_be64(
-					wqe->wr.wr.atomic.compare_add);
+					wqe->atomic_wr.compare_add);
 			} else {
 				qp->s_state = OP(FETCH_ADD);
 				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
-					wqe->wr.wr.atomic.compare_add);
+					wqe->atomic_wr.compare_add);
 				ohdr->u.atomic_eth.compare_data = 0;
 			}
 			ohdr->u.atomic_eth.vaddr[0] = cpu_to_be32(
-				wqe->wr.wr.atomic.remote_addr >> 32);
+				wqe->atomic_wr.remote_addr >> 32);
 			ohdr->u.atomic_eth.vaddr[1] = cpu_to_be32(
-				wqe->wr.wr.atomic.remote_addr);
+				wqe->atomic_wr.remote_addr);
 			ohdr->u.atomic_eth.rkey = cpu_to_be32(
-				wqe->wr.wr.atomic.rkey);
+				wqe->atomic_wr.rkey);
 			hwords += sizeof(struct ib_atomic_eth) / sizeof(u32);
 			ss = NULL;
 			len = 0;
@@ -629,9 +629,9 @@ int hfi1_make_rc_req(struct hfi1_qp *qp)
 		 */
 		len = (delta_psn(qp->s_psn, wqe->psn)) * pmtu;
 		ohdr->u.rc.reth.vaddr =
-			cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
+			cpu_to_be64(wqe->rdma_wr.remote_addr + len);
 		ohdr->u.rc.reth.rkey =
-			cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			cpu_to_be32(wqe->rdma_wr.rkey);
 		ohdr->u.rc.reth.length = cpu_to_be32(wqe->length - len);
 		qp->s_state = OP(RDMA_READ_REQUEST);
 		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
diff --git a/drivers/staging/rdma/hfi1/ruc.c b/drivers/staging/rdma/hfi1/ruc.c
index a4115288db66..d614474770b3 100644
--- a/drivers/staging/rdma/hfi1/ruc.c
+++ b/drivers/staging/rdma/hfi1/ruc.c
@@ -481,8 +481,8 @@ again:
 		if (wqe->length == 0)
 			break;
 		if (unlikely(!hfi1_rkey_ok(qp, &qp->r_sge.sge, wqe->length,
-					   wqe->wr.wr.rdma.remote_addr,
-					   wqe->wr.wr.rdma.rkey,
+					   wqe->rdma_wr.remote_addr,
+					   wqe->rdma_wr.rkey,
 					   IB_ACCESS_REMOTE_WRITE)))
 			goto acc_err;
 		qp->r_sge.sg_list = NULL;
@@ -494,8 +494,8 @@ again:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
 			goto inv_err;
 		if (unlikely(!hfi1_rkey_ok(qp, &sqp->s_sge.sge, wqe->length,
-					   wqe->wr.wr.rdma.remote_addr,
-					   wqe->wr.wr.rdma.rkey,
+					   wqe->rdma_wr.remote_addr,
+					   wqe->rdma_wr.rkey,
 					   IB_ACCESS_REMOTE_READ)))
 			goto acc_err;
 		release = 0;
@@ -512,18 +512,18 @@ again:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
 			goto inv_err;
 		if (unlikely(!hfi1_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
-					   wqe->wr.wr.atomic.remote_addr,
-					   wqe->wr.wr.atomic.rkey,
+					   wqe->atomic_wr.remote_addr,
+					   wqe->atomic_wr.rkey,
 					   IB_ACCESS_REMOTE_ATOMIC)))
 			goto acc_err;
 		/* Perform atomic OP and save result. */
 		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
-		sdata = wqe->wr.wr.atomic.compare_add;
+		sdata = wqe->atomic_wr.compare_add;
 		*(u64 *) sqp->s_sge.sge.vaddr =
 			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
 			(u64) atomic64_add_return(sdata, maddr) - sdata :
 			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
-				      sdata, wqe->wr.wr.atomic.swap);
+				      sdata, wqe->atomic_wr.swap);
 		hfi1_put_mr(qp->r_sge.sge.mr);
 		qp->r_sge.num_sge = 0;
 		goto send_comp;
@@ -913,7 +913,7 @@ void hfi1_send_complete(struct hfi1_qp *qp, struct hfi1_swqe *wqe,
 	if (qp->ibqp.qp_type == IB_QPT_UD ||
 	    qp->ibqp.qp_type == IB_QPT_SMI ||
 	    qp->ibqp.qp_type == IB_QPT_GSI)
-		atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+		atomic_dec(&to_iah(wqe->ud_wr.ah)->refcount);
 
 	/* See ch. 11.2.4.1 and 10.7.3.1 */
 	if (!(qp->s_flags & HFI1_S_SIGNAL_REQ_WR) ||
diff --git a/drivers/staging/rdma/hfi1/uc.c b/drivers/staging/rdma/hfi1/uc.c
index b536f397737c..6095039c4485 100644
--- a/drivers/staging/rdma/hfi1/uc.c
+++ b/drivers/staging/rdma/hfi1/uc.c
@@ -147,9 +147,9 @@ int hfi1_make_uc_req(struct hfi1_qp *qp)
 		case IB_WR_RDMA_WRITE:
 		case IB_WR_RDMA_WRITE_WITH_IMM:
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			hwords += sizeof(struct ib_reth) / 4;
 			if (len > pmtu) {
diff --git a/drivers/staging/rdma/hfi1/ud.c b/drivers/staging/rdma/hfi1/ud.c
index d40d1a1e10aa..5a9c784bec04 100644
--- a/drivers/staging/rdma/hfi1/ud.c
+++ b/drivers/staging/rdma/hfi1/ud.c
@@ -80,7 +80,7 @@ static void ud_loopback(struct hfi1_qp *sqp, struct hfi1_swqe *swqe)
 
 	rcu_read_lock();
 
-	qp = hfi1_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
+	qp = hfi1_lookup_qpn(ibp, swqe->ud_wr.remote_qpn);
 	if (!qp) {
 		ibp->n_pkt_drops++;
 		rcu_read_unlock();
@@ -98,7 +98,7 @@ static void ud_loopback(struct hfi1_qp *sqp, struct hfi1_swqe *swqe)
 		goto drop;
 	}
 
-	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
+	ah_attr = &to_iah(swqe->ud_wr.ah)->attr;
 	ppd = ppd_from_ibp(ibp);
 
 	if (qp->ibqp.qp_num > 1) {
@@ -128,8 +128,8 @@ static void ud_loopback(struct hfi1_qp *sqp, struct hfi1_swqe *swqe)
 	if (qp->ibqp.qp_num) {
 		u32 qkey;
 
-		qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
-			sqp->qkey : swqe->wr.wr.ud.remote_qkey;
+		qkey = (int)swqe->ud_wr.remote_qkey < 0 ?
+			sqp->qkey : swqe->ud_wr.remote_qkey;
 		if (unlikely(qkey != qp->qkey)) {
 			u16 lid;
 
@@ -234,7 +234,7 @@ static void ud_loopback(struct hfi1_qp *sqp, struct hfi1_swqe *swqe)
 	if (qp->ibqp.qp_type == IB_QPT_GSI || qp->ibqp.qp_type == IB_QPT_SMI) {
 		if (sqp->ibqp.qp_type == IB_QPT_GSI ||
 		    sqp->ibqp.qp_type == IB_QPT_SMI)
-			wc.pkey_index = swqe->wr.wr.ud.pkey_index;
+			wc.pkey_index = swqe->ud_wr.pkey_index;
 		else
 			wc.pkey_index = sqp->s_pkey_index;
 	} else {
@@ -309,7 +309,7 @@ int hfi1_make_ud_req(struct hfi1_qp *qp)
 	/* Construct the header. */
 	ibp = to_iport(qp->ibqp.device, qp->port_num);
 	ppd = ppd_from_ibp(ibp);
-	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
+	ah_attr = &to_iah(wqe->ud_wr.ah)->attr;
 	if (ah_attr->dlid < HFI1_MULTICAST_LID_BASE ||
 	    ah_attr->dlid == HFI1_PERMISSIVE_LID) {
 		lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
@@ -401,18 +401,18 @@ int hfi1_make_ud_req(struct hfi1_qp *qp)
 		bth0 |= IB_BTH_SOLICITED;
 	bth0 |= extra_bytes << 20;
 	if (qp->ibqp.qp_type == IB_QPT_GSI || qp->ibqp.qp_type == IB_QPT_SMI)
-		bth0 |= hfi1_get_pkey(ibp, wqe->wr.wr.ud.pkey_index);
+		bth0 |= hfi1_get_pkey(ibp, wqe->ud_wr.pkey_index);
 	else
 		bth0 |= hfi1_get_pkey(ibp, qp->s_pkey_index);
 	ohdr->bth[0] = cpu_to_be32(bth0);
-	ohdr->bth[1] = cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
+	ohdr->bth[1] = cpu_to_be32(wqe->ud_wr.remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(mask_psn(qp->s_next_psn++));
 	/*
 	 * Qkeys with the high order bit set mean use the
 	 * qkey from the QP context instead of the WR (see 10.2.5).
 	 */
-	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
-					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
+	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->ud_wr.remote_qkey < 0 ?
+					 qp->qkey : wqe->ud_wr.remote_qkey);
 	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
 	/* disarm any ahg */
 	qp->s_hdr->ahgcount = 0;
diff --git a/drivers/staging/rdma/hfi1/verbs.c b/drivers/staging/rdma/hfi1/verbs.c
index 41bb59eb001c..981e6c1b79a3 100644
--- a/drivers/staging/rdma/hfi1/verbs.c
+++ b/drivers/staging/rdma/hfi1/verbs.c
@@ -391,7 +391,7 @@ static int post_one_send(struct hfi1_qp *qp, struct ib_send_wr *wr)
 		    wr->opcode != IB_WR_SEND_WITH_IMM)
 			return -EINVAL;
 		/* Check UD destination address PD */
-		if (qp->ibqp.pd != wr->wr.ud.ah->pd)
+		if (qp->ibqp.pd != ud_wr(wr)->ah->pd)
 			return -EINVAL;
 	} else if ((unsigned) wr->opcode > IB_WR_ATOMIC_FETCH_AND_ADD)
 		return -EINVAL;
@@ -412,7 +412,24 @@ static int post_one_send(struct hfi1_qp *qp, struct ib_send_wr *wr)
 	rkt = &to_idev(qp->ibqp.device)->lk_table;
 	pd = to_ipd(qp->ibqp.pd);
 	wqe = get_swqe_ptr(qp, qp->s_head);
-	wqe->wr = *wr;
+
+
+	if (qp->ibqp.qp_type != IB_QPT_UC &&
+	    qp->ibqp.qp_type != IB_QPT_RC)
+		memcpy(&wqe->ud_wr, ud_wr(wr), sizeof(wqe->ud_wr));
+	else if (wr->opcode == IB_WR_FAST_REG_MR)
+		memcpy(&wqe->fast_reg_wr, fast_reg_wr(wr),
+			sizeof(wqe->fast_reg_wr));
+	else if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
+		 wr->opcode == IB_WR_RDMA_WRITE ||
+		 wr->opcode == IB_WR_RDMA_READ)
+		memcpy(&wqe->rdma_wr, rdma_wr(wr), sizeof(wqe->rdma_wr));
+	else if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		 wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+		memcpy(&wqe->atomic_wr, atomic_wr(wr), sizeof(wqe->atomic_wr));
+	else
+		memcpy(&wqe->wr, wr, sizeof(wqe->wr));
+
 	wqe->length = 0;
 	j = 0;
 	if (wr->num_sge) {
@@ -438,7 +455,7 @@ static int post_one_send(struct hfi1_qp *qp, struct ib_send_wr *wr)
 		if (wqe->length > 0x80000000U)
 			goto bail_inval_free;
 	} else {
-		struct hfi1_ah *ah = to_iah(wr->wr.ud.ah);
+		struct hfi1_ah *ah = to_iah(ud_wr(wr)->ah);
 
 		atomic_inc(&ah->refcount);
 	}
diff --git a/drivers/staging/rdma/hfi1/verbs.h b/drivers/staging/rdma/hfi1/verbs.h
index ed903a93baf7..cf5a3c956284 100644
--- a/drivers/staging/rdma/hfi1/verbs.h
+++ b/drivers/staging/rdma/hfi1/verbs.h
@@ -348,7 +348,13 @@ struct hfi1_mr {
  * in qp->s_max_sge.
  */
 struct hfi1_swqe {
-	struct ib_send_wr wr;   /* don't use wr.sg_list */
+	union {
+		struct ib_send_wr wr;   /* don't use wr.sg_list */
+		struct ib_rdma_wr rdma_wr;
+		struct ib_atomic_wr atomic_wr;
+		struct ib_ud_wr ud_wr;
+		struct ib_fast_reg_wr fast_reg_wr;
+	};
 	u32 psn;                /* first packet sequence number */
 	u32 lpsn;               /* last packet sequence number */
 	u32 ssn;                /* send sequence number */
@@ -1025,7 +1031,7 @@ struct ib_fast_reg_page_list *hfi1_alloc_fast_reg_page_list(
 
 void hfi1_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl);
 
-int hfi1_fast_reg_mr(struct hfi1_qp *qp, struct ib_send_wr *wr);
+int hfi1_fast_reg_mr(struct hfi1_qp *qp, struct ib_fast_reg_wr *wr);
 
 struct ib_fmr *hfi1_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
 			      struct ib_fmr_attr *fmr_attr);
diff --git a/drivers/staging/rdma/ipath/ipath_rc.c b/drivers/staging/rdma/ipath/ipath_rc.c
index 79b3dbc97179..d4aa53574e57 100644
--- a/drivers/staging/rdma/ipath/ipath_rc.c
+++ b/drivers/staging/rdma/ipath/ipath_rc.c
@@ -350,9 +350,9 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 				goto bail;
 			}
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			hwords += sizeof(struct ib_reth) / sizeof(u32);
 			wqe->lpsn = wqe->psn;
@@ -401,9 +401,9 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 				wqe->lpsn = qp->s_next_psn++;
 			}
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			qp->s_state = OP(RDMA_READ_REQUEST);
 			hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
@@ -433,21 +433,21 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
 				qp->s_state = OP(COMPARE_SWAP);
 				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
-					wqe->wr.wr.atomic.swap);
+					wqe->atomic_wr.swap);
 				ohdr->u.atomic_eth.compare_data = cpu_to_be64(
-					wqe->wr.wr.atomic.compare_add);
+					wqe->atomic_wr.compare_add);
 			} else {
 				qp->s_state = OP(FETCH_ADD);
 				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
-					wqe->wr.wr.atomic.compare_add);
+					wqe->atomic_wr.compare_add);
 				ohdr->u.atomic_eth.compare_data = 0;
 			}
 			ohdr->u.atomic_eth.vaddr[0] = cpu_to_be32(
-				wqe->wr.wr.atomic.remote_addr >> 32);
+				wqe->atomic_wr.remote_addr >> 32);
 			ohdr->u.atomic_eth.vaddr[1] = cpu_to_be32(
-				wqe->wr.wr.atomic.remote_addr);
+				wqe->atomic_wr.remote_addr);
 			ohdr->u.atomic_eth.rkey = cpu_to_be32(
-				wqe->wr.wr.atomic.rkey);
+				wqe->atomic_wr.rkey);
 			hwords += sizeof(struct ib_atomic_eth) / sizeof(u32);
 			ss = NULL;
 			len = 0;
@@ -567,9 +567,9 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 		ipath_init_restart(qp, wqe);
 		len = ((qp->s_psn - wqe->psn) & IPATH_PSN_MASK) * pmtu;
 		ohdr->u.rc.reth.vaddr =
-			cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
+			cpu_to_be64(wqe->rdma_wr.remote_addr + len);
 		ohdr->u.rc.reth.rkey =
-			cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			cpu_to_be32(wqe->rdma_wr.rkey);
 		ohdr->u.rc.reth.length = cpu_to_be32(qp->s_len);
 		qp->s_state = OP(RDMA_READ_REQUEST);
 		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
diff --git a/drivers/staging/rdma/ipath/ipath_ruc.c b/drivers/staging/rdma/ipath/ipath_ruc.c
index 1f95bbaf7602..46af8b03d3d4 100644
--- a/drivers/staging/rdma/ipath/ipath_ruc.c
+++ b/drivers/staging/rdma/ipath/ipath_ruc.c
@@ -353,8 +353,8 @@ again:
 		if (wqe->length == 0)
 			break;
 		if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, wqe->length,
-					    wqe->wr.wr.rdma.remote_addr,
-					    wqe->wr.wr.rdma.rkey,
+					    wqe->rdma_wr.remote_addr,
+					    wqe->rdma_wr.rkey,
 					    IB_ACCESS_REMOTE_WRITE)))
 			goto acc_err;
 		break;
@@ -363,8 +363,8 @@ again:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
 			goto inv_err;
 		if (unlikely(!ipath_rkey_ok(qp, &sqp->s_sge, wqe->length,
-					    wqe->wr.wr.rdma.remote_addr,
-					    wqe->wr.wr.rdma.rkey,
+					    wqe->rdma_wr.remote_addr,
+					    wqe->rdma_wr.rkey,
 					    IB_ACCESS_REMOTE_READ)))
 			goto acc_err;
 		qp->r_sge.sge = wqe->sg_list[0];
@@ -377,18 +377,18 @@ again:
 		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
 			goto inv_err;
 		if (unlikely(!ipath_rkey_ok(qp, &qp->r_sge, sizeof(u64),
-					    wqe->wr.wr.atomic.remote_addr,
-					    wqe->wr.wr.atomic.rkey,
+					    wqe->atomic_wr.remote_addr,
+					    wqe->atomic_wr.rkey,
 					    IB_ACCESS_REMOTE_ATOMIC)))
 			goto acc_err;
 		/* Perform atomic OP and save result. */
 		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
-		sdata = wqe->wr.wr.atomic.compare_add;
+		sdata = wqe->atomic_wr.compare_add;
 		*(u64 *) sqp->s_sge.sge.vaddr =
 			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
 			(u64) atomic64_add_return(sdata, maddr) - sdata :
 			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
-				      sdata, wqe->wr.wr.atomic.swap);
+				      sdata, wqe->atomic_wr.swap);
 		goto send_comp;
 
 	default:
diff --git a/drivers/staging/rdma/ipath/ipath_uc.c b/drivers/staging/rdma/ipath/ipath_uc.c
index 22e60998f1a7..0246b30280b9 100644
--- a/drivers/staging/rdma/ipath/ipath_uc.c
+++ b/drivers/staging/rdma/ipath/ipath_uc.c
@@ -126,9 +126,9 @@ int ipath_make_uc_req(struct ipath_qp *qp)
 		case IB_WR_RDMA_WRITE:
 		case IB_WR_RDMA_WRITE_WITH_IMM:
 			ohdr->u.rc.reth.vaddr =
-				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+				cpu_to_be64(wqe->rdma_wr.remote_addr);
 			ohdr->u.rc.reth.rkey =
-				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+				cpu_to_be32(wqe->rdma_wr.rkey);
 			ohdr->u.rc.reth.length = cpu_to_be32(len);
 			hwords += sizeof(struct ib_reth) / 4;
 			if (len > pmtu) {
diff --git a/drivers/staging/rdma/ipath/ipath_ud.c b/drivers/staging/rdma/ipath/ipath_ud.c
index e8a2a915251e..3ffc1565d03d 100644
--- a/drivers/staging/rdma/ipath/ipath_ud.c
+++ b/drivers/staging/rdma/ipath/ipath_ud.c
@@ -65,7 +65,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 	u32 rlen;
 	u32 length;
 
-	qp = ipath_lookup_qpn(&dev->qp_table, swqe->wr.wr.ud.remote_qpn);
+	qp = ipath_lookup_qpn(&dev->qp_table, swqe->ud_wr.remote_qpn);
 	if (!qp || !(ib_ipath_state_ops[qp->state] & IPATH_PROCESS_RECV_OK)) {
 		dev->n_pkt_drops++;
 		goto done;
@@ -77,8 +77,8 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 	 * qkey from the QP context instead of the WR (see 10.2.5).
 	 */
 	if (unlikely(qp->ibqp.qp_num &&
-		     ((int) swqe->wr.wr.ud.remote_qkey < 0 ?
-		      sqp->qkey : swqe->wr.wr.ud.remote_qkey) != qp->qkey)) {
+		     ((int) swqe->ud_wr.remote_qkey < 0 ?
+		      sqp->qkey : swqe->ud_wr.remote_qkey) != qp->qkey)) {
 		/* XXX OK to lose a count once in a while. */
 		dev->qkey_violations++;
 		dev->n_pkt_drops++;
@@ -175,7 +175,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 	} else
 		spin_unlock_irqrestore(&rq->lock, flags);
 
-	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
+	ah_attr = &to_iah(swqe->ud_wr.ah)->attr;
 	if (ah_attr->ah_flags & IB_AH_GRH) {
 		ipath_copy_sge(&rsge, &ah_attr->grh, sizeof(struct ib_grh));
 		wc.wc_flags |= IB_WC_GRH;
@@ -225,7 +225,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 	wc.port_num = 1;
 	/* Signal completion event if the solicited bit is set. */
 	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
-		       swqe->wr.send_flags & IB_SEND_SOLICITED);
+		       swqe->ud_wr.wr.send_flags & IB_SEND_SOLICITED);
 drop:
 	if (atomic_dec_and_test(&qp->refcount))
 		wake_up(&qp->wait);
@@ -280,7 +280,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 		next_cur = 0;
 
 	/* Construct the header. */
-	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
+	ah_attr = &to_iah(wqe->ud_wr.ah)->attr;
 	if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE) {
 		if (ah_attr->dlid != IPATH_PERMISSIVE_LID)
 			dev->n_multicast_xmit++;
@@ -322,7 +322,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 	qp->s_wqe = wqe;
 	qp->s_sge.sge = wqe->sg_list[0];
 	qp->s_sge.sg_list = wqe->sg_list + 1;
-	qp->s_sge.num_sge = wqe->wr.num_sge;
+	qp->s_sge.num_sge = wqe->ud_wr.wr.num_sge;
 
 	if (ah_attr->ah_flags & IB_AH_GRH) {
 		/* Header size in 32-bit words. */
@@ -340,9 +340,9 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 		lrh0 = IPATH_LRH_BTH;
 		ohdr = &qp->s_hdr.u.oth;
 	}
-	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+	if (wqe->ud_wr.wr.opcode == IB_WR_SEND_WITH_IMM) {
 		qp->s_hdrwords++;
-		ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
+		ohdr->u.ud.imm_data = wqe->ud_wr.wr.ex.imm_data;
 		bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
 	} else
 		bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
@@ -360,7 +360,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 		qp->s_hdr.lrh[3] = cpu_to_be16(lid);
 	} else
 		qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
-	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+	if (wqe->ud_wr.wr.send_flags & IB_SEND_SOLICITED)
 		bth0 |= 1 << 23;
 	bth0 |= extra_bytes << 20;
 	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? IPATH_DEFAULT_P_KEY :
@@ -372,14 +372,14 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 	ohdr->bth[1] = ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
 		ah_attr->dlid != IPATH_PERMISSIVE_LID ?
 		cpu_to_be32(IPATH_MULTICAST_QPN) :
-		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
+		cpu_to_be32(wqe->ud_wr.remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & IPATH_PSN_MASK);
 	/*
 	 * Qkeys with the high order bit set mean use the
 	 * qkey from the QP context instead of the WR (see 10.2.5).
 	 */
-	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
-					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
+	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->ud_wr.remote_qkey < 0 ?
+					 qp->qkey : wqe->ud_wr.remote_qkey);
 	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
 
 done:
diff --git a/drivers/staging/rdma/ipath/ipath_verbs.c b/drivers/staging/rdma/ipath/ipath_verbs.c
index ed2bbc2f7eae..15633ec1843f 100644
--- a/drivers/staging/rdma/ipath/ipath_verbs.c
+++ b/drivers/staging/rdma/ipath/ipath_verbs.c
@@ -374,7 +374,7 @@ static int ipath_post_one_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 		    wr->opcode != IB_WR_SEND_WITH_IMM)
 			goto bail_inval;
 		/* Check UD destination address PD */
-		if (qp->ibqp.pd != wr->wr.ud.ah->pd)
+		if (qp->ibqp.pd != ud_wr(wr)->ah->pd)
 			goto bail_inval;
 	} else if ((unsigned) wr->opcode > IB_WR_ATOMIC_FETCH_AND_ADD)
 		goto bail_inval;
@@ -395,7 +395,23 @@ static int ipath_post_one_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 	}
 
 	wqe = get_swqe_ptr(qp, qp->s_head);
-	wqe->wr = *wr;
+
+	if (qp->ibqp.qp_type != IB_QPT_UC &&
+	    qp->ibqp.qp_type != IB_QPT_RC)
+		memcpy(&wqe->ud_wr, ud_wr(wr), sizeof(wqe->ud_wr));
+	else if (wr->opcode == IB_WR_FAST_REG_MR)
+		memcpy(&wqe->fast_reg_wr, fast_reg_wr(wr),
+			sizeof(wqe->fast_reg_wr));
+	else if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM ||
+		 wr->opcode == IB_WR_RDMA_WRITE ||
+		 wr->opcode == IB_WR_RDMA_READ)
+		memcpy(&wqe->rdma_wr, rdma_wr(wr), sizeof(wqe->rdma_wr));
+	else if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		 wr->opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+		memcpy(&wqe->atomic_wr, atomic_wr(wr), sizeof(wqe->atomic_wr));
+	else
+		memcpy(&wqe->wr, wr, sizeof(wqe->wr));
+
 	wqe->length = 0;
 	if (wr->num_sge) {
 		acc = wr->opcode >= IB_WR_RDMA_READ ?
diff --git a/drivers/staging/rdma/ipath/ipath_verbs.h b/drivers/staging/rdma/ipath/ipath_verbs.h
index ec167e545e15..ed102a26ec08 100644
--- a/drivers/staging/rdma/ipath/ipath_verbs.h
+++ b/drivers/staging/rdma/ipath/ipath_verbs.h
@@ -277,7 +277,14 @@ struct ipath_mr {
  * in qp->s_max_sge.
  */
 struct ipath_swqe {
-	struct ib_send_wr wr;	/* don't use wr.sg_list */
+	union {
+		struct ib_send_wr wr;   /* don't use wr.sg_list */
+		struct ib_ud_wr ud_wr;
+		struct ib_fast_reg_wr fast_reg_wr;
+		struct ib_rdma_wr rdma_wr;
+		struct ib_atomic_wr atomic_wr;
+	};
+
 	u32 psn;		/* first packet sequence number */
 	u32 lpsn;		/* last packet sequence number */
 	u32 ssn;		/* send sequence number */
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index e4cc389c43cb..85103aff909b 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -1096,54 +1096,93 @@ struct ib_send_wr {
 		__be32		imm_data;
 		u32		invalidate_rkey;
 	} ex;
-	union {
-		struct {
-			u64	remote_addr;
-			u32	rkey;
-		} rdma;
-		struct {
-			u64	remote_addr;
-			u64	compare_add;
-			u64	swap;
-			u64	compare_add_mask;
-			u64	swap_mask;
-			u32	rkey;
-		} atomic;
-		struct {
-			struct ib_ah *ah;
-			void   *header;
-			int     hlen;
-			int     mss;
-			u32	remote_qpn;
-			u32	remote_qkey;
-			u16	pkey_index; /* valid for GSI only */
-			u8	port_num;   /* valid for DR SMPs on switch only */
-		} ud;
-		struct {
-			u64				iova_start;
-			struct ib_fast_reg_page_list   *page_list;
-			unsigned int			page_shift;
-			unsigned int			page_list_len;
-			u32				length;
-			int				access_flags;
-			u32				rkey;
-		} fast_reg;
-		struct {
-			struct ib_mw            *mw;
-			/* The new rkey for the memory window. */
-			u32                      rkey;
-			struct ib_mw_bind_info   bind_info;
-		} bind_mw;
-		struct {
-			struct ib_sig_attrs    *sig_attrs;
-			struct ib_mr	       *sig_mr;
-			int			access_flags;
-			struct ib_sge	       *prot;
-		} sig_handover;
-	} wr;
-	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
 };
 
+struct ib_rdma_wr {
+	struct ib_send_wr	wr;
+	u64			remote_addr;
+	u32			rkey;
+};
+
+static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct ib_rdma_wr, wr);
+}
+
+struct ib_atomic_wr {
+	struct ib_send_wr	wr;
+	u64			remote_addr;
+	u64			compare_add;
+	u64			swap;
+	u64			compare_add_mask;
+	u64			swap_mask;
+	u32			rkey;
+};
+
+static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct ib_atomic_wr, wr);
+}
+
+struct ib_ud_wr {
+	struct ib_send_wr	wr;
+	struct ib_ah		*ah;
+	void			*header;
+	int			hlen;
+	int			mss;
+	u32			remote_qpn;
+	u32			remote_qkey;
+	u16			pkey_index; /* valid for GSI only */
+	u8			port_num;   /* valid for DR SMPs on switch only */
+};
+
+static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct ib_ud_wr, wr);
+}
+
+struct ib_fast_reg_wr {
+	struct ib_send_wr	wr;
+	u64			iova_start;
+	struct ib_fast_reg_page_list *page_list;
+	unsigned int		page_shift;
+	unsigned int		page_list_len;
+	u32			length;
+	int			access_flags;
+	u32			rkey;
+};
+
+static inline struct ib_fast_reg_wr *fast_reg_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct ib_fast_reg_wr, wr);
+}
+
+struct ib_bind_mw_wr {
+	struct ib_send_wr	wr;
+	struct ib_mw		*mw;
+	/* The new rkey for the memory window. */
+	u32			rkey;
+	struct ib_mw_bind_info	bind_info;
+};
+
+static inline struct ib_bind_mw_wr *bind_mw_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct ib_bind_mw_wr, wr);
+}
+
+struct ib_sig_handover_wr {
+	struct ib_send_wr	wr;
+	struct ib_sig_attrs    *sig_attrs;
+	struct ib_mr	       *sig_mr;
+	int			access_flags;
+	struct ib_sge	       *prot;
+};
+
+static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
+{
+	return container_of(wr, struct ib_sig_handover_wr, wr);
+}
+
 struct ib_recv_wr {
 	struct ib_recv_wr      *next;
 	u64			wr_id;
diff --git a/net/rds/ib.h b/net/rds/ib.h
index aae60fda77f6..301c48385166 100644
--- a/net/rds/ib.h
+++ b/net/rds/ib.h
@@ -69,7 +69,11 @@ struct rds_ib_connect_private {
 
 struct rds_ib_send_work {
 	void			*s_op;
-	struct ib_send_wr	s_wr;
+	union {
+		struct ib_send_wr	s_wr;
+		struct ib_rdma_wr	s_rdma_wr;
+		struct ib_atomic_wr	s_atomic_wr;
+	};
 	struct ib_sge		s_sge[RDS_IB_MAX_SGE];
 	unsigned long		s_queued;
 };
diff --git a/net/rds/ib_send.c b/net/rds/ib_send.c
index 4e88047086b6..987386e9931b 100644
--- a/net/rds/ib_send.c
+++ b/net/rds/ib_send.c
@@ -785,23 +785,23 @@ int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op)
 	send->s_queued = jiffies;
 
 	if (op->op_type == RDS_ATOMIC_TYPE_CSWP) {
-		send->s_wr.opcode = IB_WR_MASKED_ATOMIC_CMP_AND_SWP;
-		send->s_wr.wr.atomic.compare_add = op->op_m_cswp.compare;
-		send->s_wr.wr.atomic.swap = op->op_m_cswp.swap;
-		send->s_wr.wr.atomic.compare_add_mask = op->op_m_cswp.compare_mask;
-		send->s_wr.wr.atomic.swap_mask = op->op_m_cswp.swap_mask;
+		send->s_atomic_wr.wr.opcode = IB_WR_MASKED_ATOMIC_CMP_AND_SWP;
+		send->s_atomic_wr.compare_add = op->op_m_cswp.compare;
+		send->s_atomic_wr.swap = op->op_m_cswp.swap;
+		send->s_atomic_wr.compare_add_mask = op->op_m_cswp.compare_mask;
+		send->s_atomic_wr.swap_mask = op->op_m_cswp.swap_mask;
 	} else { /* FADD */
-		send->s_wr.opcode = IB_WR_MASKED_ATOMIC_FETCH_AND_ADD;
-		send->s_wr.wr.atomic.compare_add = op->op_m_fadd.add;
-		send->s_wr.wr.atomic.swap = 0;
-		send->s_wr.wr.atomic.compare_add_mask = op->op_m_fadd.nocarry_mask;
-		send->s_wr.wr.atomic.swap_mask = 0;
+		send->s_atomic_wr.wr.opcode = IB_WR_MASKED_ATOMIC_FETCH_AND_ADD;
+		send->s_atomic_wr.compare_add = op->op_m_fadd.add;
+		send->s_atomic_wr.swap = 0;
+		send->s_atomic_wr.compare_add_mask = op->op_m_fadd.nocarry_mask;
+		send->s_atomic_wr.swap_mask = 0;
 	}
 	nr_sig = rds_ib_set_wr_signal_state(ic, send, op->op_notify);
-	send->s_wr.num_sge = 1;
-	send->s_wr.next = NULL;
-	send->s_wr.wr.atomic.remote_addr = op->op_remote_addr;
-	send->s_wr.wr.atomic.rkey = op->op_rkey;
+	send->s_atomic_wr.wr.num_sge = 1;
+	send->s_atomic_wr.wr.next = NULL;
+	send->s_atomic_wr.remote_addr = op->op_remote_addr;
+	send->s_atomic_wr.rkey = op->op_rkey;
 	send->s_op = op;
 	rds_message_addref(container_of(send->s_op, struct rds_message, atomic));
 
@@ -826,11 +826,11 @@ int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op)
 	if (nr_sig)
 		atomic_add(nr_sig, &ic->i_signaled_sends);
 
-	failed_wr = &send->s_wr;
-	ret = ib_post_send(ic->i_cm_id->qp, &send->s_wr, &failed_wr);
+	failed_wr = &send->s_atomic_wr.wr;
+	ret = ib_post_send(ic->i_cm_id->qp, &send->s_atomic_wr.wr, &failed_wr);
 	rdsdebug("ic %p send %p (wr %p) ret %d wr %p\n", ic,
-		 send, &send->s_wr, ret, failed_wr);
-	BUG_ON(failed_wr != &send->s_wr);
+		 send, &send->s_atomic_wr, ret, failed_wr);
+	BUG_ON(failed_wr != &send->s_atomic_wr.wr);
 	if (ret) {
 		printk(KERN_WARNING "RDS/IB: atomic ib_post_send to %pI4 "
 		       "returned %d\n", &conn->c_faddr, ret);
@@ -839,9 +839,9 @@ int rds_ib_xmit_atomic(struct rds_connection *conn, struct rm_atomic_op *op)
 		goto out;
 	}
 
-	if (unlikely(failed_wr != &send->s_wr)) {
+	if (unlikely(failed_wr != &send->s_atomic_wr.wr)) {
 		printk(KERN_WARNING "RDS/IB: atomic ib_post_send() rc=%d, but failed_wqe updated!\n", ret);
-		BUG_ON(failed_wr != &send->s_wr);
+		BUG_ON(failed_wr != &send->s_atomic_wr.wr);
 	}
 
 out:
@@ -912,22 +912,23 @@ int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 		nr_sig += rds_ib_set_wr_signal_state(ic, send, op->op_notify);
 
 		send->s_wr.opcode = op->op_write ? IB_WR_RDMA_WRITE : IB_WR_RDMA_READ;
-		send->s_wr.wr.rdma.remote_addr = remote_addr;
-		send->s_wr.wr.rdma.rkey = op->op_rkey;
+		send->s_rdma_wr.remote_addr = remote_addr;
+		send->s_rdma_wr.rkey = op->op_rkey;
 
 		if (num_sge > max_sge) {
-			send->s_wr.num_sge = max_sge;
+			send->s_rdma_wr.wr.num_sge = max_sge;
 			num_sge -= max_sge;
 		} else {
-			send->s_wr.num_sge = num_sge;
+			send->s_rdma_wr.wr.num_sge = num_sge;
 		}
 
-		send->s_wr.next = NULL;
+		send->s_rdma_wr.wr.next = NULL;
 
 		if (prev)
-			prev->s_wr.next = &send->s_wr;
+			prev->s_rdma_wr.wr.next = &send->s_rdma_wr.wr;
 
-		for (j = 0; j < send->s_wr.num_sge && scat != &op->op_sg[op->op_count]; j++) {
+		for (j = 0; j < send->s_rdma_wr.wr.num_sge &&
+		     scat != &op->op_sg[op->op_count]; j++) {
 			len = ib_sg_dma_len(ic->i_cm_id->device, scat);
 			send->s_sge[j].addr =
 				 ib_sg_dma_address(ic->i_cm_id->device, scat);
@@ -942,7 +943,9 @@ int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 		}
 
 		rdsdebug("send %p wr %p num_sge %u next %p\n", send,
-			&send->s_wr, send->s_wr.num_sge, send->s_wr.next);
+			&send->s_rdma_wr.wr,
+			send->s_rdma_wr.wr.num_sge,
+			send->s_rdma_wr.wr.next);
 
 		prev = send;
 		if (++send == &ic->i_sends[ic->i_send_ring.w_nr])
@@ -963,11 +966,11 @@ int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 	if (nr_sig)
 		atomic_add(nr_sig, &ic->i_signaled_sends);
 
-	failed_wr = &first->s_wr;
-	ret = ib_post_send(ic->i_cm_id->qp, &first->s_wr, &failed_wr);
+	failed_wr = &first->s_rdma_wr.wr;
+	ret = ib_post_send(ic->i_cm_id->qp, &first->s_rdma_wr.wr, &failed_wr);
 	rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
-		 first, &first->s_wr, ret, failed_wr);
-	BUG_ON(failed_wr != &first->s_wr);
+		 first, &first->s_rdma_wr.wr, ret, failed_wr);
+	BUG_ON(failed_wr != &first->s_rdma_wr.wr);
 	if (ret) {
 		printk(KERN_WARNING "RDS/IB: rdma ib_post_send to %pI4 "
 		       "returned %d\n", &conn->c_faddr, ret);
@@ -976,9 +979,9 @@ int rds_ib_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 		goto out;
 	}
 
-	if (unlikely(failed_wr != &first->s_wr)) {
+	if (unlikely(failed_wr != &first->s_rdma_wr.wr)) {
 		printk(KERN_WARNING "RDS/IB: ib_post_send() rc=%d, but failed_wqe updated!\n", ret);
-		BUG_ON(failed_wr != &first->s_wr);
+		BUG_ON(failed_wr != &first->s_rdma_wr.wr);
 	}
 
 
diff --git a/net/rds/iw.h b/net/rds/iw.h
index cbe6674e31ee..fe858e5dd8d1 100644
--- a/net/rds/iw.h
+++ b/net/rds/iw.h
@@ -77,7 +77,11 @@ struct rds_iw_send_work {
 	struct ib_fast_reg_page_list *s_page_list;
 	unsigned char		s_remap_count;
 
-	struct ib_send_wr	s_wr;
+	union {
+		struct ib_send_wr	s_send_wr;
+		struct ib_rdma_wr	s_rdma_wr;
+		struct ib_fast_reg_wr	s_fast_reg_wr;
+	};
 	struct ib_sge		s_sge[RDS_IW_MAX_SGE];
 	unsigned long		s_queued;
 };
diff --git a/net/rds/iw_rdma.c b/net/rds/iw_rdma.c
index 6a8fbd6e69e7..f8a612cc69e6 100644
--- a/net/rds/iw_rdma.c
+++ b/net/rds/iw_rdma.c
@@ -696,7 +696,8 @@ static int rds_iw_init_fastreg(struct rds_iw_mr_pool *pool,
 static int rds_iw_rdma_build_fastreg(struct rds_iw_mapping *mapping)
 {
 	struct rds_iw_mr *ibmr = mapping->m_mr;
-	struct ib_send_wr f_wr, *failed_wr;
+	struct ib_fast_reg_wr f_wr;
+	struct ib_send_wr *failed_wr;
 	int ret;
 
 	/*
@@ -709,22 +710,22 @@ static int rds_iw_rdma_build_fastreg(struct rds_iw_mapping *mapping)
 	mapping->m_rkey = ibmr->mr->rkey;
 
 	memset(&f_wr, 0, sizeof(f_wr));
-	f_wr.wr_id = RDS_IW_FAST_REG_WR_ID;
-	f_wr.opcode = IB_WR_FAST_REG_MR;
-	f_wr.wr.fast_reg.length = mapping->m_sg.bytes;
-	f_wr.wr.fast_reg.rkey = mapping->m_rkey;
-	f_wr.wr.fast_reg.page_list = ibmr->page_list;
-	f_wr.wr.fast_reg.page_list_len = mapping->m_sg.dma_len;
-	f_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
-	f_wr.wr.fast_reg.access_flags = IB_ACCESS_LOCAL_WRITE |
+	f_wr.wr.wr_id = RDS_IW_FAST_REG_WR_ID;
+	f_wr.wr.opcode = IB_WR_FAST_REG_MR;
+	f_wr.length = mapping->m_sg.bytes;
+	f_wr.rkey = mapping->m_rkey;
+	f_wr.page_list = ibmr->page_list;
+	f_wr.page_list_len = mapping->m_sg.dma_len;
+	f_wr.page_shift = PAGE_SHIFT;
+	f_wr.access_flags = IB_ACCESS_LOCAL_WRITE |
 				IB_ACCESS_REMOTE_READ |
 				IB_ACCESS_REMOTE_WRITE;
-	f_wr.wr.fast_reg.iova_start = 0;
-	f_wr.send_flags = IB_SEND_SIGNALED;
+	f_wr.iova_start = 0;
+	f_wr.wr.send_flags = IB_SEND_SIGNALED;
 
-	failed_wr = &f_wr;
-	ret = ib_post_send(ibmr->cm_id->qp, &f_wr, &failed_wr);
-	BUG_ON(failed_wr != &f_wr);
+	failed_wr = &f_wr.wr;
+	ret = ib_post_send(ibmr->cm_id->qp, &f_wr.wr, &failed_wr);
+	BUG_ON(failed_wr != &f_wr.wr);
 	if (ret)
 		printk_ratelimited(KERN_WARNING "RDS/IW: %s:%d ib_post_send returned %d\n",
 			__func__, __LINE__, ret);
diff --git a/net/rds/iw_send.c b/net/rds/iw_send.c
index 86152ec3b887..f6e23c515b44 100644
--- a/net/rds/iw_send.c
+++ b/net/rds/iw_send.c
@@ -137,13 +137,13 @@ void rds_iw_send_init_ring(struct rds_iw_connection *ic)
 		send->s_op = NULL;
 		send->s_mapping = NULL;
 
-		send->s_wr.next = NULL;
-		send->s_wr.wr_id = i;
-		send->s_wr.sg_list = send->s_sge;
-		send->s_wr.num_sge = 1;
-		send->s_wr.opcode = IB_WR_SEND;
-		send->s_wr.send_flags = 0;
-		send->s_wr.ex.imm_data = 0;
+		send->s_send_wr.next = NULL;
+		send->s_send_wr.wr_id = i;
+		send->s_send_wr.sg_list = send->s_sge;
+		send->s_send_wr.num_sge = 1;
+		send->s_send_wr.opcode = IB_WR_SEND;
+		send->s_send_wr.send_flags = 0;
+		send->s_send_wr.ex.imm_data = 0;
 
 		sge = rds_iw_data_sge(ic, send->s_sge);
 		sge->lkey = 0;
@@ -179,7 +179,7 @@ void rds_iw_send_clear_ring(struct rds_iw_connection *ic)
 		ib_dereg_mr(send->s_mr);
 		BUG_ON(!send->s_page_list);
 		ib_free_fast_reg_page_list(send->s_page_list);
-		if (send->s_wr.opcode == 0xdead)
+		if (send->s_send_wr.opcode == 0xdead)
 			continue;
 		if (send->s_rm)
 			rds_iw_send_unmap_rm(ic, send, IB_WC_WR_FLUSH_ERR);
@@ -247,7 +247,7 @@ void rds_iw_send_cq_comp_handler(struct ib_cq *cq, void *context)
 			send = &ic->i_sends[oldest];
 
 			/* In the error case, wc.opcode sometimes contains garbage */
-			switch (send->s_wr.opcode) {
+			switch (send->s_send_wr.opcode) {
 			case IB_WR_SEND:
 				if (send->s_rm)
 					rds_iw_send_unmap_rm(ic, send, wc.status);
@@ -262,12 +262,12 @@ void rds_iw_send_cq_comp_handler(struct ib_cq *cq, void *context)
 			default:
 				printk_ratelimited(KERN_NOTICE
 						"RDS/IW: %s: unexpected opcode 0x%x in WR!\n",
-						__func__, send->s_wr.opcode);
+						__func__, send->s_send_wr.opcode);
 				break;
 			}
 
-			send->s_wr.opcode = 0xdead;
-			send->s_wr.num_sge = 1;
+			send->s_send_wr.opcode = 0xdead;
+			send->s_send_wr.num_sge = 1;
 			if (time_after(jiffies, send->s_queued + HZ/2))
 				rds_iw_stats_inc(s_iw_tx_stalled);
 
@@ -455,10 +455,10 @@ rds_iw_xmit_populate_wr(struct rds_iw_connection *ic,
 
 	WARN_ON(pos != send - ic->i_sends);
 
-	send->s_wr.send_flags = send_flags;
-	send->s_wr.opcode = IB_WR_SEND;
-	send->s_wr.num_sge = 2;
-	send->s_wr.next = NULL;
+	send->s_send_wr.send_flags = send_flags;
+	send->s_send_wr.opcode = IB_WR_SEND;
+	send->s_send_wr.num_sge = 2;
+	send->s_send_wr.next = NULL;
 	send->s_queued = jiffies;
 	send->s_op = NULL;
 
@@ -472,7 +472,7 @@ rds_iw_xmit_populate_wr(struct rds_iw_connection *ic,
 	} else {
 		/* We're sending a packet with no payload. There is only
 		 * one SGE */
-		send->s_wr.num_sge = 1;
+		send->s_send_wr.num_sge = 1;
 		sge = &send->s_sge[0];
 	}
 
@@ -672,23 +672,23 @@ int rds_iw_xmit(struct rds_connection *conn, struct rds_message *rm,
 		 */
 		if (ic->i_unsignaled_wrs-- == 0) {
 			ic->i_unsignaled_wrs = rds_iw_sysctl_max_unsig_wrs;
-			send->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
+			send->s_send_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
 		}
 
 		ic->i_unsignaled_bytes -= len;
 		if (ic->i_unsignaled_bytes <= 0) {
 			ic->i_unsignaled_bytes = rds_iw_sysctl_max_unsig_bytes;
-			send->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
+			send->s_send_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
 		}
 
 		/*
 		 * Always signal the last one if we're stopping due to flow control.
 		 */
 		if (flow_controlled && i == (work_alloc-1))
-			send->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
+			send->s_send_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
 
 		rdsdebug("send %p wr %p num_sge %u next %p\n", send,
-			 &send->s_wr, send->s_wr.num_sge, send->s_wr.next);
+			 &send->s_send_wr, send->s_send_wr.num_sge, send->s_send_wr.next);
 
 		sent += len;
 		rm->data.op_dmaoff += len;
@@ -722,7 +722,7 @@ add_header:
 		}
 
 		if (prev)
-			prev->s_wr.next = &send->s_wr;
+			prev->s_send_wr.next = &send->s_send_wr;
 		prev = send;
 
 		pos = (pos + 1) % ic->i_send_ring.w_nr;
@@ -736,7 +736,7 @@ add_header:
 	/* if we finished the message then send completion owns it */
 	if (scat == &rm->data.op_sg[rm->data.op_count]) {
 		prev->s_rm = ic->i_rm;
-		prev->s_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
+		prev->s_send_wr.send_flags |= IB_SEND_SIGNALED | IB_SEND_SOLICITED;
 		ic->i_rm = NULL;
 	}
 
@@ -748,11 +748,11 @@ add_header:
 		rds_iw_send_add_credits(conn, credit_alloc - i);
 
 	/* XXX need to worry about failed_wr and partial sends. */
-	failed_wr = &first->s_wr;
-	ret = ib_post_send(ic->i_cm_id->qp, &first->s_wr, &failed_wr);
+	failed_wr = &first->s_send_wr;
+	ret = ib_post_send(ic->i_cm_id->qp, &first->s_send_wr, &failed_wr);
 	rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
-		 first, &first->s_wr, ret, failed_wr);
-	BUG_ON(failed_wr != &first->s_wr);
+		 first, &first->s_send_wr, ret, failed_wr);
+	BUG_ON(failed_wr != &first->s_send_wr);
 	if (ret) {
 		printk(KERN_WARNING "RDS/IW: ib_post_send to %pI4 "
 		       "returned %d\n", &conn->c_faddr, ret);
@@ -778,14 +778,14 @@ static void rds_iw_build_send_fastreg(struct rds_iw_device *rds_iwdev, struct rd
 	 * in the sg list is added to the fast reg page list and placed
 	 * inside the fast_reg_mr WR.
 	 */
-	send->s_wr.opcode = IB_WR_FAST_REG_MR;
-	send->s_wr.wr.fast_reg.length = len;
-	send->s_wr.wr.fast_reg.rkey = send->s_mr->rkey;
-	send->s_wr.wr.fast_reg.page_list = send->s_page_list;
-	send->s_wr.wr.fast_reg.page_list_len = nent;
-	send->s_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
-	send->s_wr.wr.fast_reg.access_flags = IB_ACCESS_REMOTE_WRITE;
-	send->s_wr.wr.fast_reg.iova_start = sg_addr;
+	send->s_fast_reg_wr.wr.opcode = IB_WR_FAST_REG_MR;
+	send->s_fast_reg_wr.length = len;
+	send->s_fast_reg_wr.rkey = send->s_mr->rkey;
+	send->s_fast_reg_wr.page_list = send->s_page_list;
+	send->s_fast_reg_wr.page_list_len = nent;
+	send->s_fast_reg_wr.page_shift = PAGE_SHIFT;
+	send->s_fast_reg_wr.access_flags = IB_ACCESS_REMOTE_WRITE;
+	send->s_fast_reg_wr.iova_start = sg_addr;
 
 	ib_update_fast_reg_key(send->s_mr, send->s_remap_count++);
 }
@@ -863,7 +863,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 	num_sge = op->op_count;
 
 	for (i = 0; i < work_alloc && scat != &op->op_sg[op->op_count]; i++) {
-		send->s_wr.send_flags = 0;
+		send->s_rdma_wr.wr.send_flags = 0;
 		send->s_queued = jiffies;
 
 		/*
@@ -872,7 +872,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 		 */
 		if (ic->i_unsignaled_wrs-- == 0) {
 			ic->i_unsignaled_wrs = rds_iw_sysctl_max_unsig_wrs;
-			send->s_wr.send_flags = IB_SEND_SIGNALED;
+			send->s_rdma_wr.wr.send_flags = IB_SEND_SIGNALED;
 		}
 
 		/* To avoid the need to have the plumbing to invalidate the fastreg_mr used
@@ -880,29 +880,30 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 		 * IB_WR_RDMA_READ_WITH_INV will invalidate it after the read has completed.
 		 */
 		if (op->op_write)
-			send->s_wr.opcode = IB_WR_RDMA_WRITE;
+			send->s_rdma_wr.wr.opcode = IB_WR_RDMA_WRITE;
 		else
-			send->s_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
+			send->s_rdma_wr.wr.opcode = IB_WR_RDMA_READ_WITH_INV;
 
-		send->s_wr.wr.rdma.remote_addr = remote_addr;
-		send->s_wr.wr.rdma.rkey = op->op_rkey;
+		send->s_rdma_wr.remote_addr = remote_addr;
+		send->s_rdma_wr.rkey = op->op_rkey;
 		send->s_op = op;
 
 		if (num_sge > rds_iwdev->max_sge) {
-			send->s_wr.num_sge = rds_iwdev->max_sge;
+			send->s_rdma_wr.wr.num_sge = rds_iwdev->max_sge;
 			num_sge -= rds_iwdev->max_sge;
 		} else
-			send->s_wr.num_sge = num_sge;
+			send->s_rdma_wr.wr.num_sge = num_sge;
 
-		send->s_wr.next = NULL;
+		send->s_rdma_wr.wr.next = NULL;
 
 		if (prev)
-			prev->s_wr.next = &send->s_wr;
+			prev->s_send_wr.next = &send->s_rdma_wr.wr;
 
-		for (j = 0; j < send->s_wr.num_sge && scat != &op->op_sg[op->op_count]; j++) {
+		for (j = 0; j < send->s_rdma_wr.wr.num_sge &&
+		     scat != &op->op_sg[op->op_count]; j++) {
 			len = ib_sg_dma_len(ic->i_cm_id->device, scat);
 
-			if (send->s_wr.opcode == IB_WR_RDMA_READ_WITH_INV)
+			if (send->s_rdma_wr.wr.opcode == IB_WR_RDMA_READ_WITH_INV)
 				send->s_page_list->page_list[j] = ib_sg_dma_address(ic->i_cm_id->device, scat);
 			else {
 				send->s_sge[j].addr = ib_sg_dma_address(ic->i_cm_id->device, scat);
@@ -917,15 +918,17 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 			scat++;
 		}
 
-		if (send->s_wr.opcode == IB_WR_RDMA_READ_WITH_INV) {
-			send->s_wr.num_sge = 1;
+		if (send->s_rdma_wr.wr.opcode == IB_WR_RDMA_READ_WITH_INV) {
+			send->s_rdma_wr.wr.num_sge = 1;
 			send->s_sge[0].addr = conn->c_xmit_rm->m_rs->rs_user_addr;
 			send->s_sge[0].length = conn->c_xmit_rm->m_rs->rs_user_bytes;
 			send->s_sge[0].lkey = ic->i_sends[fr_pos].s_mr->lkey;
 		}
 
 		rdsdebug("send %p wr %p num_sge %u next %p\n", send,
-			&send->s_wr, send->s_wr.num_sge, send->s_wr.next);
+			&send->s_rdma_wr,
+			send->s_rdma_wr.wr.num_sge,
+			send->s_rdma_wr.wr.next);
 
 		prev = send;
 		if (++send == &ic->i_sends[ic->i_send_ring.w_nr])
@@ -934,7 +937,7 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 
 	/* if we finished the message then send completion owns it */
 	if (scat == &op->op_sg[op->op_count])
-		first->s_wr.send_flags = IB_SEND_SIGNALED;
+		first->s_rdma_wr.wr.send_flags = IB_SEND_SIGNALED;
 
 	if (i < work_alloc) {
 		rds_iw_ring_unalloc(&ic->i_send_ring, work_alloc - i);
@@ -953,11 +956,11 @@ int rds_iw_xmit_rdma(struct rds_connection *conn, struct rm_rdma_op *op)
 		work_alloc++;
 	}
 
-	failed_wr = &first->s_wr;
-	ret = ib_post_send(ic->i_cm_id->qp, &first->s_wr, &failed_wr);
+	failed_wr = &first->s_rdma_wr.wr;
+	ret = ib_post_send(ic->i_cm_id->qp, &first->s_rdma_wr.wr, &failed_wr);
 	rdsdebug("ic %p first %p (wr %p) ret %d wr %p\n", ic,
-		 first, &first->s_wr, ret, failed_wr);
-	BUG_ON(failed_wr != &first->s_wr);
+		 first, &first->s_rdma_wr, ret, failed_wr);
+	BUG_ON(failed_wr != &first->s_rdma_wr.wr);
 	if (ret) {
 		printk(KERN_WARNING "RDS/IW: rdma ib_post_send to %pI4 "
 		       "returned %d\n", &conn->c_faddr, ret);
diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c
index 5318951b3b53..0d2f46f600b6 100644
--- a/net/sunrpc/xprtrdma/frwr_ops.c
+++ b/net/sunrpc/xprtrdma/frwr_ops.c
@@ -312,7 +312,8 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
 	struct rpcrdma_mw *mw;
 	struct rpcrdma_frmr *frmr;
 	struct ib_mr *mr;
-	struct ib_send_wr fastreg_wr, *bad_wr;
+	struct ib_fast_reg_wr fastreg_wr;
+	struct ib_send_wr *bad_wr;
 	u8 key;
 	int len, pageoff;
 	int i, rc;
@@ -358,23 +359,23 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
 		__func__, mw, i, len);
 
 	memset(&fastreg_wr, 0, sizeof(fastreg_wr));
-	fastreg_wr.wr_id = (unsigned long)(void *)mw;
-	fastreg_wr.opcode = IB_WR_FAST_REG_MR;
-	fastreg_wr.wr.fast_reg.iova_start = seg1->mr_dma + pageoff;
-	fastreg_wr.wr.fast_reg.page_list = frmr->fr_pgl;
-	fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
-	fastreg_wr.wr.fast_reg.page_list_len = page_no;
-	fastreg_wr.wr.fast_reg.length = len;
-	fastreg_wr.wr.fast_reg.access_flags = writing ?
+	fastreg_wr.wr.wr_id = (unsigned long)(void *)mw;
+	fastreg_wr.wr.opcode = IB_WR_FAST_REG_MR;
+	fastreg_wr.iova_start = seg1->mr_dma + pageoff;
+	fastreg_wr.page_list = frmr->fr_pgl;
+	fastreg_wr.page_shift = PAGE_SHIFT;
+	fastreg_wr.page_list_len = page_no;
+	fastreg_wr.length = len;
+	fastreg_wr.access_flags = writing ?
 				IB_ACCESS_REMOTE_WRITE | IB_ACCESS_LOCAL_WRITE :
 				IB_ACCESS_REMOTE_READ;
 	mr = frmr->fr_mr;
 	key = (u8)(mr->rkey & 0x000000FF);
 	ib_update_fast_reg_key(mr, ++key);
-	fastreg_wr.wr.fast_reg.rkey = mr->rkey;
+	fastreg_wr.rkey = mr->rkey;
 
 	DECR_CQCOUNT(&r_xprt->rx_ep);
-	rc = ib_post_send(ia->ri_id->qp, &fastreg_wr, &bad_wr);
+	rc = ib_post_send(ia->ri_id->qp, &fastreg_wr.wr, &bad_wr);
 	if (rc)
 		goto out_senderr;
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index cb5174284074..7be42d0da19e 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -126,7 +126,7 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
 			u64 rs_offset,
 			bool last)
 {
-	struct ib_send_wr read_wr;
+	struct ib_rdma_wr read_wr;
 	int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
 	struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
 	int ret, read, pno;
@@ -179,16 +179,16 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
 		clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
 
 	memset(&read_wr, 0, sizeof(read_wr));
-	read_wr.wr_id = (unsigned long)ctxt;
-	read_wr.opcode = IB_WR_RDMA_READ;
-	ctxt->wr_op = read_wr.opcode;
-	read_wr.send_flags = IB_SEND_SIGNALED;
-	read_wr.wr.rdma.rkey = rs_handle;
-	read_wr.wr.rdma.remote_addr = rs_offset;
-	read_wr.sg_list = ctxt->sge;
-	read_wr.num_sge = pages_needed;
-
-	ret = svc_rdma_send(xprt, &read_wr);
+	read_wr.wr.wr_id = (unsigned long)ctxt;
+	read_wr.wr.opcode = IB_WR_RDMA_READ;
+	ctxt->wr_op = read_wr.wr.opcode;
+	read_wr.wr.send_flags = IB_SEND_SIGNALED;
+	read_wr.rkey = rs_handle;
+	read_wr.remote_addr = rs_offset;
+	read_wr.wr.sg_list = ctxt->sge;
+	read_wr.wr.num_sge = pages_needed;
+
+	ret = svc_rdma_send(xprt, &read_wr.wr);
 	if (ret) {
 		pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
@@ -218,9 +218,9 @@ int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
 			 u64 rs_offset,
 			 bool last)
 {
-	struct ib_send_wr read_wr;
+	struct ib_rdma_wr read_wr;
 	struct ib_send_wr inv_wr;
-	struct ib_send_wr fastreg_wr;
+	struct ib_fast_reg_wr fastreg_wr;
 	u8 key;
 	int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
 	struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
@@ -289,31 +289,31 @@ int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
 
 	/* Prepare FASTREG WR */
 	memset(&fastreg_wr, 0, sizeof(fastreg_wr));
-	fastreg_wr.opcode = IB_WR_FAST_REG_MR;
-	fastreg_wr.send_flags = IB_SEND_SIGNALED;
-	fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
-	fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
-	fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
-	fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
-	fastreg_wr.wr.fast_reg.length = frmr->map_len;
-	fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
-	fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
-	fastreg_wr.next = &read_wr;
+	fastreg_wr.wr.opcode = IB_WR_FAST_REG_MR;
+	fastreg_wr.wr.send_flags = IB_SEND_SIGNALED;
+	fastreg_wr.iova_start = (unsigned long)frmr->kva;
+	fastreg_wr.page_list = frmr->page_list;
+	fastreg_wr.page_list_len = frmr->page_list_len;
+	fastreg_wr.page_shift = PAGE_SHIFT;
+	fastreg_wr.length = frmr->map_len;
+	fastreg_wr.access_flags = frmr->access_flags;
+	fastreg_wr.rkey = frmr->mr->lkey;
+	fastreg_wr.wr.next = &read_wr.wr;
 
 	/* Prepare RDMA_READ */
 	memset(&read_wr, 0, sizeof(read_wr));
-	read_wr.send_flags = IB_SEND_SIGNALED;
-	read_wr.wr.rdma.rkey = rs_handle;
-	read_wr.wr.rdma.remote_addr = rs_offset;
-	read_wr.sg_list = ctxt->sge;
-	read_wr.num_sge = 1;
+	read_wr.wr.send_flags = IB_SEND_SIGNALED;
+	read_wr.rkey = rs_handle;
+	read_wr.remote_addr = rs_offset;
+	read_wr.wr.sg_list = ctxt->sge;
+	read_wr.wr.num_sge = 1;
 	if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) {
-		read_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
-		read_wr.wr_id = (unsigned long)ctxt;
-		read_wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey;
+		read_wr.wr.opcode = IB_WR_RDMA_READ_WITH_INV;
+		read_wr.wr.wr_id = (unsigned long)ctxt;
+		read_wr.wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey;
 	} else {
-		read_wr.opcode = IB_WR_RDMA_READ;
-		read_wr.next = &inv_wr;
+		read_wr.wr.opcode = IB_WR_RDMA_READ;
+		read_wr.wr.next = &inv_wr;
 		/* Prepare invalidate */
 		memset(&inv_wr, 0, sizeof(inv_wr));
 		inv_wr.wr_id = (unsigned long)ctxt;
@@ -321,10 +321,10 @@ int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
 		inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE;
 		inv_wr.ex.invalidate_rkey = frmr->mr->lkey;
 	}
-	ctxt->wr_op = read_wr.opcode;
+	ctxt->wr_op = read_wr.wr.opcode;
 
 	/* Post the chain */
-	ret = svc_rdma_send(xprt, &fastreg_wr);
+	ret = svc_rdma_send(xprt, &fastreg_wr.wr);
 	if (ret) {
 		pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 1dfae8317065..969a1ab75fc3 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -217,7 +217,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
 		      u32 xdr_off, int write_len,
 		      struct svc_rdma_req_map *vec)
 {
-	struct ib_send_wr write_wr;
+	struct ib_rdma_wr write_wr;
 	struct ib_sge *sge;
 	int xdr_sge_no;
 	int sge_no;
@@ -282,17 +282,17 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
 	/* Prepare WRITE WR */
 	memset(&write_wr, 0, sizeof write_wr);
 	ctxt->wr_op = IB_WR_RDMA_WRITE;
-	write_wr.wr_id = (unsigned long)ctxt;
-	write_wr.sg_list = &sge[0];
-	write_wr.num_sge = sge_no;
-	write_wr.opcode = IB_WR_RDMA_WRITE;
-	write_wr.send_flags = IB_SEND_SIGNALED;
-	write_wr.wr.rdma.rkey = rmr;
-	write_wr.wr.rdma.remote_addr = to;
+	write_wr.wr.wr_id = (unsigned long)ctxt;
+	write_wr.wr.sg_list = &sge[0];
+	write_wr.wr.num_sge = sge_no;
+	write_wr.wr.opcode = IB_WR_RDMA_WRITE;
+	write_wr.wr.send_flags = IB_SEND_SIGNALED;
+	write_wr.rkey = rmr;
+	write_wr.remote_addr = to;
 
 	/* Post It */
 	atomic_inc(&rdma_stat_write);
-	if (svc_rdma_send(xprt, &write_wr))
+	if (svc_rdma_send(xprt, &write_wr.wr))
 		goto err;
 	return write_len - bc;
  err: