Soft RoCE driver

Soft RoCE (RXE) - The software RoCE driver

ib_rxe implements the RDMA transport and registers to the RDMA core
device as a kernel verbs provider. It also implements the packet IO
layer. On the other hand ib_rxe registers to the Linux netdev stack
as a udp encapsulating protocol, in that case RDMA, for sending and
receiving packets over any Ethernet device.  This yields a RDMA
transport over the UDP/Ethernet network layer forming a RoCEv2
compatible device.

The configuration procedure of the Soft RoCE drivers requires
binding to any existing Ethernet network device. This is done with
/sys interface.

A userspace Soft RoCE library (librxe) provides user applications
the ability to run with Soft RoCE devices.  The use of rxe verbs ins
user space requires the inclusion of librxe as a device specifics
plug-in to libibverbs. librxe is packaged separately.

Architecture:

     +-----------------------------------------------------------+
     |                          Application                      |
     +-----------------------------------------------------------+
                            +-----------------------------------+
                            |             libibverbs            |
User                        +-----------------------------------+
                            +----------------+ +----------------+
                            | librxe         | | HW RoCE lib    |
                            +----------------+ +----------------+
+---------------------------------------------------------------+
     +--------------+                           +------------+
     | Sockets      |                           | RDMA ULP   |
     +--------------+                           +------------+
     +--------------+                  +---------------------+
     | TCP/IP       |                  | ib_core             |
     +--------------+                  +---------------------+
                             +------------+ +----------------+
Kernel                       | ib_rxe     | | HW RoCE driver |
                             +------------+ +----------------+
     +------------------------------------+
     | NIC driver                         |
     +------------------------------------+

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
     +-----------------------------------------------------------+
     |                          Application                      |
     +-----------------------------------------------------------+
                            +-----------------------------------+
                            |             libibverbs            |
User                        +-----------------------------------+
                            +----------------+ +----------------+
                            | librxe         | | HW RoCE lib    |
                            +----------------+ +----------------+
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
     +--------------+                           +------------+
     | Sockets      |                           | RDMA ULP   |
     +--------------+                           +------------+
     +--------------+                  +---------------------+
     | TCP/IP       |                  | ib_core             |
     +--------------+                  +---------------------+
                             +------------+ +----------------+
Kernel                       | ib_rxe     | | HW RoCE driver |
                             +------------+ +----------------+
     +------------------------------------+
     | NIC driver                         |
     +------------------------------------+
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Soft RoCE resources:

[1[ https://github.com/SoftRoCE/librxe-dev librxe - source code in
Github
[2] https://github.com/SoftRoCE/rxe-dev/wiki/rxe-dev:-Home - Soft RoCE
Wiki page
[3] https://github.com/SoftRoCE/librxe-dev - Soft RoCE userspace library

Signed-off-by: Kamal Heib <kamalh@mellanox.com>
Signed-off-by: Amir Vadai <amirv@mellanox.com>
Signed-off-by: Moni Shoua <monis@mellanox.com>
Reviewed-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>

drivers/infiniband/sw/rxe/rxe_req.c

@@ -0,0 +1,726 @@
+/*
+ * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
+ * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *	- Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *
+ *	- Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/skbuff.h>
+
+#include "rxe.h"
+#include "rxe_loc.h"
+#include "rxe_queue.h"
+
+static int next_opcode(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
+		       unsigned opcode);
+
+static inline void retry_first_write_send(struct rxe_qp *qp,
+					  struct rxe_send_wqe *wqe,
+					  unsigned mask, int npsn)
+{
+	int i;
+
+	for (i = 0; i < npsn; i++) {
+		int to_send = (wqe->dma.resid > qp->mtu) ?
+				qp->mtu : wqe->dma.resid;
+
+		qp->req.opcode = next_opcode(qp, wqe,
+					     wqe->wr.opcode);
+
+		if (wqe->wr.send_flags & IB_SEND_INLINE) {
+			wqe->dma.resid -= to_send;
+			wqe->dma.sge_offset += to_send;
+		} else {
+			advance_dma_data(&wqe->dma, to_send);
+		}
+		if (mask & WR_WRITE_MASK)
+			wqe->iova += qp->mtu;
+	}
+}
+
+static void req_retry(struct rxe_qp *qp)
+{
+	struct rxe_send_wqe *wqe;
+	unsigned int wqe_index;
+	unsigned int mask;
+	int npsn;
+	int first = 1;
+
+	wqe = queue_head(qp->sq.queue);
+	npsn = (qp->comp.psn - wqe->first_psn) & BTH_PSN_MASK;
+
+	qp->req.wqe_index	= consumer_index(qp->sq.queue);
+	qp->req.psn		= qp->comp.psn;
+	qp->req.opcode		= -1;
+
+	for (wqe_index = consumer_index(qp->sq.queue);
+		wqe_index != producer_index(qp->sq.queue);
+		wqe_index = next_index(qp->sq.queue, wqe_index)) {
+		wqe = addr_from_index(qp->sq.queue, wqe_index);
+		mask = wr_opcode_mask(wqe->wr.opcode, qp);
+
+		if (wqe->state == wqe_state_posted)
+			break;
+
+		if (wqe->state == wqe_state_done)
+			continue;
+
+		wqe->iova = (mask & WR_ATOMIC_MASK) ?
+			     wqe->wr.wr.atomic.remote_addr :
+			     (mask & WR_READ_OR_WRITE_MASK) ?
+			     wqe->wr.wr.rdma.remote_addr :
+			     0;
+
+		if (!first || (mask & WR_READ_MASK) == 0) {
+			wqe->dma.resid = wqe->dma.length;
+			wqe->dma.cur_sge = 0;
+			wqe->dma.sge_offset = 0;
+		}
+
+		if (first) {
+			first = 0;
+
+			if (mask & WR_WRITE_OR_SEND_MASK)
+				retry_first_write_send(qp, wqe, mask, npsn);
+
+			if (mask & WR_READ_MASK)
+				wqe->iova += npsn * qp->mtu;
+		}
+
+		wqe->state = wqe_state_posted;
+	}
+}
+
+void rnr_nak_timer(unsigned long data)
+{
+	struct rxe_qp *qp = (struct rxe_qp *)data;
+
+	pr_debug("rnr nak timer fired\n");
+	rxe_run_task(&qp->req.task, 1);
+}
+
+static struct rxe_send_wqe *req_next_wqe(struct rxe_qp *qp)
+{
+	struct rxe_send_wqe *wqe = queue_head(qp->sq.queue);
+	unsigned long flags;
+
+	if (unlikely(qp->req.state == QP_STATE_DRAIN)) {
+		/* check to see if we are drained;
+		 * state_lock used by requester and completer
+		 */
+		spin_lock_irqsave(&qp->state_lock, flags);
+		do {
+			if (qp->req.state != QP_STATE_DRAIN) {
+				/* comp just finished */
+				spin_unlock_irqrestore(&qp->state_lock,
+						       flags);
+				break;
+			}
+
+			if (wqe && ((qp->req.wqe_index !=
+				consumer_index(qp->sq.queue)) ||
+				(wqe->state != wqe_state_posted))) {
+				/* comp not done yet */
+				spin_unlock_irqrestore(&qp->state_lock,
+						       flags);
+				break;
+			}
+
+			qp->req.state = QP_STATE_DRAINED;
+			spin_unlock_irqrestore(&qp->state_lock, flags);
+
+			if (qp->ibqp.event_handler) {
+				struct ib_event ev;
+
+				ev.device = qp->ibqp.device;
+				ev.element.qp = &qp->ibqp;
+				ev.event = IB_EVENT_SQ_DRAINED;
+				qp->ibqp.event_handler(&ev,
+					qp->ibqp.qp_context);
+			}
+		} while (0);
+	}
+
+	if (qp->req.wqe_index == producer_index(qp->sq.queue))
+		return NULL;
+
+	wqe = addr_from_index(qp->sq.queue, qp->req.wqe_index);
+
+	if (unlikely((qp->req.state == QP_STATE_DRAIN ||
+		      qp->req.state == QP_STATE_DRAINED) &&
+		     (wqe->state != wqe_state_processing)))
+		return NULL;
+
+	if (unlikely((wqe->wr.send_flags & IB_SEND_FENCE) &&
+		     (qp->req.wqe_index != consumer_index(qp->sq.queue)))) {
+		qp->req.wait_fence = 1;
+		return NULL;
+	}
+
+	wqe->mask = wr_opcode_mask(wqe->wr.opcode, qp);
+	return wqe;
+}
+
+static int next_opcode_rc(struct rxe_qp *qp, unsigned opcode, int fits)
+{
+	switch (opcode) {
+	case IB_WR_RDMA_WRITE:
+		if (qp->req.opcode == IB_OPCODE_RC_RDMA_WRITE_FIRST ||
+		    qp->req.opcode == IB_OPCODE_RC_RDMA_WRITE_MIDDLE)
+			return fits ?
+				IB_OPCODE_RC_RDMA_WRITE_LAST :
+				IB_OPCODE_RC_RDMA_WRITE_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_RC_RDMA_WRITE_ONLY :
+				IB_OPCODE_RC_RDMA_WRITE_FIRST;
+
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		if (qp->req.opcode == IB_OPCODE_RC_RDMA_WRITE_FIRST ||
+		    qp->req.opcode == IB_OPCODE_RC_RDMA_WRITE_MIDDLE)
+			return fits ?
+				IB_OPCODE_RC_RDMA_WRITE_LAST_WITH_IMMEDIATE :
+				IB_OPCODE_RC_RDMA_WRITE_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_RC_RDMA_WRITE_ONLY_WITH_IMMEDIATE :
+				IB_OPCODE_RC_RDMA_WRITE_FIRST;
+
+	case IB_WR_SEND:
+		if (qp->req.opcode == IB_OPCODE_RC_SEND_FIRST ||
+		    qp->req.opcode == IB_OPCODE_RC_SEND_MIDDLE)
+			return fits ?
+				IB_OPCODE_RC_SEND_LAST :
+				IB_OPCODE_RC_SEND_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_RC_SEND_ONLY :
+				IB_OPCODE_RC_SEND_FIRST;
+
+	case IB_WR_SEND_WITH_IMM:
+		if (qp->req.opcode == IB_OPCODE_RC_SEND_FIRST ||
+		    qp->req.opcode == IB_OPCODE_RC_SEND_MIDDLE)
+			return fits ?
+				IB_OPCODE_RC_SEND_LAST_WITH_IMMEDIATE :
+				IB_OPCODE_RC_SEND_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_RC_SEND_ONLY_WITH_IMMEDIATE :
+				IB_OPCODE_RC_SEND_FIRST;
+
+	case IB_WR_RDMA_READ:
+		return IB_OPCODE_RC_RDMA_READ_REQUEST;
+
+	case IB_WR_ATOMIC_CMP_AND_SWP:
+		return IB_OPCODE_RC_COMPARE_SWAP;
+
+	case IB_WR_ATOMIC_FETCH_AND_ADD:
+		return IB_OPCODE_RC_FETCH_ADD;
+
+	case IB_WR_SEND_WITH_INV:
+		if (qp->req.opcode == IB_OPCODE_RC_SEND_FIRST ||
+		    qp->req.opcode == IB_OPCODE_RC_SEND_MIDDLE)
+			return fits ? IB_OPCODE_RC_SEND_LAST_WITH_INVALIDATE :
+				IB_OPCODE_RC_SEND_MIDDLE;
+		else
+			return fits ? IB_OPCODE_RC_SEND_ONLY_WITH_INVALIDATE :
+				IB_OPCODE_RC_SEND_FIRST;
+	case IB_WR_REG_MR:
+	case IB_WR_LOCAL_INV:
+		return opcode;
+	}
+
+	return -EINVAL;
+}
+
+static int next_opcode_uc(struct rxe_qp *qp, unsigned opcode, int fits)
+{
+	switch (opcode) {
+	case IB_WR_RDMA_WRITE:
+		if (qp->req.opcode == IB_OPCODE_UC_RDMA_WRITE_FIRST ||
+		    qp->req.opcode == IB_OPCODE_UC_RDMA_WRITE_MIDDLE)
+			return fits ?
+				IB_OPCODE_UC_RDMA_WRITE_LAST :
+				IB_OPCODE_UC_RDMA_WRITE_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_UC_RDMA_WRITE_ONLY :
+				IB_OPCODE_UC_RDMA_WRITE_FIRST;
+
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		if (qp->req.opcode == IB_OPCODE_UC_RDMA_WRITE_FIRST ||
+		    qp->req.opcode == IB_OPCODE_UC_RDMA_WRITE_MIDDLE)
+			return fits ?
+				IB_OPCODE_UC_RDMA_WRITE_LAST_WITH_IMMEDIATE :
+				IB_OPCODE_UC_RDMA_WRITE_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_UC_RDMA_WRITE_ONLY_WITH_IMMEDIATE :
+				IB_OPCODE_UC_RDMA_WRITE_FIRST;
+
+	case IB_WR_SEND:
+		if (qp->req.opcode == IB_OPCODE_UC_SEND_FIRST ||
+		    qp->req.opcode == IB_OPCODE_UC_SEND_MIDDLE)
+			return fits ?
+				IB_OPCODE_UC_SEND_LAST :
+				IB_OPCODE_UC_SEND_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_UC_SEND_ONLY :
+				IB_OPCODE_UC_SEND_FIRST;
+
+	case IB_WR_SEND_WITH_IMM:
+		if (qp->req.opcode == IB_OPCODE_UC_SEND_FIRST ||
+		    qp->req.opcode == IB_OPCODE_UC_SEND_MIDDLE)
+			return fits ?
+				IB_OPCODE_UC_SEND_LAST_WITH_IMMEDIATE :
+				IB_OPCODE_UC_SEND_MIDDLE;
+		else
+			return fits ?
+				IB_OPCODE_UC_SEND_ONLY_WITH_IMMEDIATE :
+				IB_OPCODE_UC_SEND_FIRST;
+	}
+
+	return -EINVAL;
+}
+
+static int next_opcode(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
+		       unsigned opcode)
+{
+	int fits = (wqe->dma.resid <= qp->mtu);
+
+	switch (qp_type(qp)) {
+	case IB_QPT_RC:
+		return next_opcode_rc(qp, opcode, fits);
+
+	case IB_QPT_UC:
+		return next_opcode_uc(qp, opcode, fits);
+
+	case IB_QPT_SMI:
+	case IB_QPT_UD:
+	case IB_QPT_GSI:
+		switch (opcode) {
+		case IB_WR_SEND:
+			return IB_OPCODE_UD_SEND_ONLY;
+
+		case IB_WR_SEND_WITH_IMM:
+			return IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE;
+		}
+		break;
+
+	default:
+		break;
+	}
+
+	return -EINVAL;
+}
+
+static inline int check_init_depth(struct rxe_qp *qp, struct rxe_send_wqe *wqe)
+{
+	int depth;
+
+	if (wqe->has_rd_atomic)
+		return 0;
+
+	qp->req.need_rd_atomic = 1;
+	depth = atomic_dec_return(&qp->req.rd_atomic);
+
+	if (depth >= 0) {
+		qp->req.need_rd_atomic = 0;
+		wqe->has_rd_atomic = 1;
+		return 0;
+	}
+
+	atomic_inc(&qp->req.rd_atomic);
+	return -EAGAIN;
+}
+
+static inline int get_mtu(struct rxe_qp *qp, struct rxe_send_wqe *wqe)
+{
+	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
+	struct rxe_port *port;
+	struct rxe_av *av;
+
+	if ((qp_type(qp) == IB_QPT_RC) || (qp_type(qp) == IB_QPT_UC))
+		return qp->mtu;
+
+	av = &wqe->av;
+	port = &rxe->port;
+
+	return port->mtu_cap;
+}
+
+static struct sk_buff *init_req_packet(struct rxe_qp *qp,
+				       struct rxe_send_wqe *wqe,
+				       int opcode, int payload,
+				       struct rxe_pkt_info *pkt)
+{
+	struct rxe_dev		*rxe = to_rdev(qp->ibqp.device);
+	struct rxe_port		*port = &rxe->port;
+	struct sk_buff		*skb;
+	struct rxe_send_wr	*ibwr = &wqe->wr;
+	struct rxe_av		*av;
+	int			pad = (-payload) & 0x3;
+	int			paylen;
+	int			solicited;
+	u16			pkey;
+	u32			qp_num;
+	int			ack_req;
+
+	/* length from start of bth to end of icrc */
+	paylen = rxe_opcode[opcode].length + payload + pad + RXE_ICRC_SIZE;
+
+	/* pkt->hdr, rxe, port_num and mask are initialized in ifc
+	 * layer
+	 */
+	pkt->opcode	= opcode;
+	pkt->qp		= qp;
+	pkt->psn	= qp->req.psn;
+	pkt->mask	= rxe_opcode[opcode].mask;
+	pkt->paylen	= paylen;
+	pkt->offset	= 0;
+	pkt->wqe	= wqe;
+
+	/* init skb */
+	av = rxe_get_av(pkt);
+	skb = rxe->ifc_ops->init_packet(rxe, av, paylen, pkt);
+	if (unlikely(!skb))
+		return NULL;
+
+	/* init bth */
+	solicited = (ibwr->send_flags & IB_SEND_SOLICITED) &&
+			(pkt->mask & RXE_END_MASK) &&
+			((pkt->mask & (RXE_SEND_MASK)) ||
+			(pkt->mask & (RXE_WRITE_MASK | RXE_IMMDT_MASK)) ==
+			(RXE_WRITE_MASK | RXE_IMMDT_MASK));
+
+	pkey = (qp_type(qp) == IB_QPT_GSI) ?
+		 port->pkey_tbl[ibwr->wr.ud.pkey_index] :
+		 port->pkey_tbl[qp->attr.pkey_index];
+
+	qp_num = (pkt->mask & RXE_DETH_MASK) ? ibwr->wr.ud.remote_qpn :
+					 qp->attr.dest_qp_num;
+
+	ack_req = ((pkt->mask & RXE_END_MASK) ||
+		(qp->req.noack_pkts++ > RXE_MAX_PKT_PER_ACK));
+	if (ack_req)
+		qp->req.noack_pkts = 0;
+
+	bth_init(pkt, pkt->opcode, solicited, 0, pad, pkey, qp_num,
+		 ack_req, pkt->psn);
+
+	/* init optional headers */
+	if (pkt->mask & RXE_RETH_MASK) {
+		reth_set_rkey(pkt, ibwr->wr.rdma.rkey);
+		reth_set_va(pkt, wqe->iova);
+		reth_set_len(pkt, wqe->dma.length);
+	}
+
+	if (pkt->mask & RXE_IMMDT_MASK)
+		immdt_set_imm(pkt, ibwr->ex.imm_data);
+
+	if (pkt->mask & RXE_IETH_MASK)
+		ieth_set_rkey(pkt, ibwr->ex.invalidate_rkey);
+
+	if (pkt->mask & RXE_ATMETH_MASK) {
+		atmeth_set_va(pkt, wqe->iova);
+		if (opcode == IB_OPCODE_RC_COMPARE_SWAP ||
+		    opcode == IB_OPCODE_RD_COMPARE_SWAP) {
+			atmeth_set_swap_add(pkt, ibwr->wr.atomic.swap);
+			atmeth_set_comp(pkt, ibwr->wr.atomic.compare_add);
+		} else {
+			atmeth_set_swap_add(pkt, ibwr->wr.atomic.compare_add);
+		}
+		atmeth_set_rkey(pkt, ibwr->wr.atomic.rkey);
+	}
+
+	if (pkt->mask & RXE_DETH_MASK) {
+		if (qp->ibqp.qp_num == 1)
+			deth_set_qkey(pkt, GSI_QKEY);
+		else
+			deth_set_qkey(pkt, ibwr->wr.ud.remote_qkey);
+		deth_set_sqp(pkt, qp->ibqp.qp_num);
+	}
+
+	return skb;
+}
+
+static int fill_packet(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
+		       struct rxe_pkt_info *pkt, struct sk_buff *skb,
+		       int paylen)
+{
+	struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
+	u32 crc = 0;
+	u32 *p;
+	int err;
+
+	err = rxe->ifc_ops->prepare(rxe, pkt, skb, &crc);
+	if (err)
+		return err;
+
+	if (pkt->mask & RXE_WRITE_OR_SEND) {
+		if (wqe->wr.send_flags & IB_SEND_INLINE) {
+			u8 *tmp = &wqe->dma.inline_data[wqe->dma.sge_offset];
+
+			crc = crc32_le(crc, tmp, paylen);
+
+			memcpy(payload_addr(pkt), tmp, paylen);
+
+			wqe->dma.resid -= paylen;
+			wqe->dma.sge_offset += paylen;
+		} else {
+			err = copy_data(rxe, qp->pd, 0, &wqe->dma,
+					payload_addr(pkt), paylen,
+					from_mem_obj,
+					&crc);
+			if (err)
+				return err;
+		}
+	}
+	p = payload_addr(pkt) + paylen + bth_pad(pkt);
+
+	*p = ~crc;
+
+	return 0;
+}
+
+static void update_wqe_state(struct rxe_qp *qp,
+			     struct rxe_send_wqe *wqe,
+			     struct rxe_pkt_info *pkt,
+			     enum wqe_state *prev_state)
+{
+	enum wqe_state prev_state_ = wqe->state;
+
+	if (pkt->mask & RXE_END_MASK) {
+		if (qp_type(qp) == IB_QPT_RC)
+			wqe->state = wqe_state_pending;
+	} else {
+		wqe->state = wqe_state_processing;
+	}
+
+	*prev_state = prev_state_;
+}
+
+static void update_state(struct rxe_qp *qp, struct rxe_send_wqe *wqe,
+			 struct rxe_pkt_info *pkt, int payload)
+{
+	/* number of packets left to send including current one */
+	int num_pkt = (wqe->dma.resid + payload + qp->mtu - 1) / qp->mtu;
+
+	/* handle zero length packet case */
+	if (num_pkt == 0)
+		num_pkt = 1;
+
+	if (pkt->mask & RXE_START_MASK) {
+		wqe->first_psn = qp->req.psn;
+		wqe->last_psn = (qp->req.psn + num_pkt - 1) & BTH_PSN_MASK;
+	}
+
+	if (pkt->mask & RXE_READ_MASK)
+		qp->req.psn = (wqe->first_psn + num_pkt) & BTH_PSN_MASK;
+	else
+		qp->req.psn = (qp->req.psn + 1) & BTH_PSN_MASK;
+
+	qp->req.opcode = pkt->opcode;
+
+
+	if (pkt->mask & RXE_END_MASK)
+		qp->req.wqe_index = next_index(qp->sq.queue, qp->req.wqe_index);
+
+	qp->need_req_skb = 0;
+
+	if (qp->qp_timeout_jiffies && !timer_pending(&qp->retrans_timer))
+		mod_timer(&qp->retrans_timer,
+			  jiffies + qp->qp_timeout_jiffies);
+}
+
+int rxe_requester(void *arg)
+{
+	struct rxe_qp *qp = (struct rxe_qp *)arg;
+	struct rxe_pkt_info pkt;
+	struct sk_buff *skb;
+	struct rxe_send_wqe *wqe;
+	unsigned mask;
+	int payload;
+	int mtu;
+	int opcode;
+	int ret;
+	enum wqe_state prev_state;
+
+next_wqe:
+	if (unlikely(!qp->valid || qp->req.state == QP_STATE_ERROR))
+		goto exit;
+
+	if (unlikely(qp->req.state == QP_STATE_RESET)) {
+		qp->req.wqe_index = consumer_index(qp->sq.queue);
+		qp->req.opcode = -1;
+		qp->req.need_rd_atomic = 0;
+		qp->req.wait_psn = 0;
+		qp->req.need_retry = 0;
+		goto exit;
+	}
+
+	if (unlikely(qp->req.need_retry)) {
+		req_retry(qp);
+		qp->req.need_retry = 0;
+	}
+
+	wqe = req_next_wqe(qp);
+	if (unlikely(!wqe))
+		goto exit;
+
+	if (wqe->mask & WR_REG_MASK) {
+		if (wqe->wr.opcode == IB_WR_LOCAL_INV) {
+			struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
+			struct rxe_mem *rmr;
+
+			rmr = rxe_pool_get_index(&rxe->mr_pool,
+						 wqe->wr.ex.invalidate_rkey >> 8);
+			if (!rmr) {
+				pr_err("No mr for key %#x\n", wqe->wr.ex.invalidate_rkey);
+				wqe->state = wqe_state_error;
+				wqe->status = IB_WC_MW_BIND_ERR;
+				goto exit;
+			}
+			rmr->state = RXE_MEM_STATE_FREE;
+			wqe->state = wqe_state_done;
+			wqe->status = IB_WC_SUCCESS;
+		} else if (wqe->wr.opcode == IB_WR_REG_MR) {
+			struct rxe_mem *rmr = to_rmr(wqe->wr.wr.reg.mr);
+
+			rmr->state = RXE_MEM_STATE_VALID;
+			rmr->access = wqe->wr.wr.reg.access;
+			rmr->lkey = wqe->wr.wr.reg.key;
+			rmr->rkey = wqe->wr.wr.reg.key;
+			wqe->state = wqe_state_done;
+			wqe->status = IB_WC_SUCCESS;
+		} else {
+			goto exit;
+		}
+		qp->req.wqe_index = next_index(qp->sq.queue,
+						qp->req.wqe_index);
+		goto next_wqe;
+	}
+
+	if (unlikely(qp_type(qp) == IB_QPT_RC &&
+		     qp->req.psn > (qp->comp.psn + RXE_MAX_UNACKED_PSNS))) {
+		qp->req.wait_psn = 1;
+		goto exit;
+	}
+
+	/* Limit the number of inflight SKBs per QP */
+	if (unlikely(atomic_read(&qp->skb_out) >
+		     RXE_INFLIGHT_SKBS_PER_QP_HIGH)) {
+		qp->need_req_skb = 1;
+		goto exit;
+	}
+
+	opcode = next_opcode(qp, wqe, wqe->wr.opcode);
+	if (unlikely(opcode < 0)) {
+		wqe->status = IB_WC_LOC_QP_OP_ERR;
+		goto exit;
+	}
+
+	mask = rxe_opcode[opcode].mask;
+	if (unlikely(mask & RXE_READ_OR_ATOMIC)) {
+		if (check_init_depth(qp, wqe))
+			goto exit;
+	}
+
+	mtu = get_mtu(qp, wqe);
+	payload = (mask & RXE_WRITE_OR_SEND) ? wqe->dma.resid : 0;
+	if (payload > mtu) {
+		if (qp_type(qp) == IB_QPT_UD) {
+			/* C10-93.1.1: If the total sum of all the buffer lengths specified for a
+			 * UD message exceeds the MTU of the port as returned by QueryHCA, the CI
+			 * shall not emit any packets for this message. Further, the CI shall not
+			 * generate an error due to this condition.
+			 */
+
+			/* fake a successful UD send */
+			wqe->first_psn = qp->req.psn;
+			wqe->last_psn = qp->req.psn;
+			qp->req.psn = (qp->req.psn + 1) & BTH_PSN_MASK;
+			qp->req.opcode = IB_OPCODE_UD_SEND_ONLY;
+			qp->req.wqe_index = next_index(qp->sq.queue,
+						       qp->req.wqe_index);
+			wqe->state = wqe_state_done;
+			wqe->status = IB_WC_SUCCESS;
+			goto complete;
+		}
+		payload = mtu;
+	}
+
+	skb = init_req_packet(qp, wqe, opcode, payload, &pkt);
+	if (unlikely(!skb)) {
+		pr_err("Failed allocating skb\n");
+		goto err;
+	}
+
+	if (fill_packet(qp, wqe, &pkt, skb, payload)) {
+		pr_debug("Error during fill packet\n");
+		goto err;
+	}
+
+	update_wqe_state(qp, wqe, &pkt, &prev_state);
+	ret = rxe_xmit_packet(to_rdev(qp->ibqp.device), qp, &pkt, skb);
+	if (ret) {
+		qp->need_req_skb = 1;
+		kfree_skb(skb);
+
+		wqe->state = prev_state;
+
+		if (ret == -EAGAIN) {
+			rxe_run_task(&qp->req.task, 1);
+			goto exit;
+		}
+
+		goto err;
+	}
+
+	update_state(qp, wqe, &pkt, payload);
+
+	goto next_wqe;
+
+err:
+	kfree_skb(skb);
+	wqe->status = IB_WC_LOC_PROT_ERR;
+	wqe->state = wqe_state_error;
+
+complete:
+	if (qp_type(qp) != IB_QPT_RC) {
+		while (rxe_completer(qp) == 0)
+			;
+	}
+
+	return 0;
+
+exit:
+	return -EAGAIN;
+}