qed: iWARP CM add passive side connect

This patch implements the passive side connect.
It addresses pre-allocating resources, creating a connection
element upon valid SYN packet received. Calling upper layer and
implementation of the accept/reject calls.

Error handling is not part of this patch.

Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

9 files changed, 1039 insertions, 20 deletions

diff --git a/drivers/net/ethernet/qlogic/qed/qed.h b/drivers/net/ethernet/qlogic/qed/qed.h
index fd8cd5e17121..91003bc6f00b 100644
--- a/drivers/net/ethernet/qlogic/qed/qed.h
+++ b/drivers/net/ethernet/qlogic/qed/qed.h
@@ -789,6 +789,8 @@ void qed_configure_vp_wfq_on_link_change(struct qed_dev *cdev,
 void qed_clean_wfq_db(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt);
 int qed_device_num_engines(struct qed_dev *cdev);
 int qed_device_get_port_id(struct qed_dev *cdev);
+void qed_set_fw_mac_addr(__le16 *fw_msb,
+			 __le16 *fw_mid, __le16 *fw_lsb, u8 *mac);
 
 #define QED_LEADING_HWFN(dev)   (&dev->hwfns[0])
 
diff --git a/drivers/net/ethernet/qlogic/qed/qed_dev.c b/drivers/net/ethernet/qlogic/qed/qed_dev.c
index 6c8505dc5c31..4060a6ad9be3 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_dev.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_dev.c
@@ -4127,3 +4127,14 @@ int qed_device_get_port_id(struct qed_dev *cdev)
 {
 	return (QED_LEADING_HWFN(cdev)->abs_pf_id) % qed_device_num_ports(cdev);
 }
+
+void qed_set_fw_mac_addr(__le16 *fw_msb,
+			 __le16 *fw_mid, __le16 *fw_lsb, u8 *mac)
+{
+	((u8 *)fw_msb)[0] = mac[1];
+	((u8 *)fw_msb)[1] = mac[0];
+	((u8 *)fw_mid)[0] = mac[3];
+	((u8 *)fw_mid)[1] = mac[2];
+	((u8 *)fw_lsb)[0] = mac[5];
+	((u8 *)fw_lsb)[1] = mac[4];
+}
diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c
index 2bab57c6bae8..a6dadae1f998 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.c
@@ -44,9 +44,30 @@
 
 #define QED_IWARP_ORD_DEFAULT		32
 #define QED_IWARP_IRD_DEFAULT		32
+#define QED_IWARP_MAX_FW_MSS		4120
+
+#define QED_EP_SIG 0xecabcdef
+
+struct mpa_v2_hdr {
+	__be16 ird;
+	__be16 ord;
+};
+
+#define MPA_V2_PEER2PEER_MODEL  0x8000
+#define MPA_V2_SEND_RTR         0x4000	/* on ird */
+#define MPA_V2_READ_RTR         0x4000	/* on ord */
+#define MPA_V2_WRITE_RTR        0x8000
+#define MPA_V2_IRD_ORD_MASK     0x3FFF
+
+#define MPA_REV2(_mpa_rev) ((_mpa_rev) == MPA_NEGOTIATION_TYPE_ENHANCED)
+
+#define QED_IWARP_INVALID_TCP_CID	0xffffffff
 #define QED_IWARP_RCV_WND_SIZE_DEF	(256 * 1024)
 #define QED_IWARP_RCV_WND_SIZE_MIN	(64 * 1024)
+#define TIMESTAMP_HEADER_SIZE		(12)
+
 #define QED_IWARP_TS_EN			BIT(0)
+#define QED_IWARP_DA_EN			BIT(1)
 #define QED_IWARP_PARAM_CRC_NEEDED	(1)
 #define QED_IWARP_PARAM_P2P		(1)
 
@@ -63,7 +84,8 @@ void qed_iwarp_init_devinfo(struct qed_hwfn *p_hwfn)
 	dev->max_inline = IWARP_REQ_MAX_INLINE_DATA_SIZE;
 	dev->max_qp = min_t(u32,
 			    IWARP_MAX_QPS,
-			    p_hwfn->p_rdma_info->num_qps);
+			    p_hwfn->p_rdma_info->num_qps) -
+		      QED_IWARP_PREALLOC_CNT;
 
 	dev->max_cq = dev->max_qp;
 
@@ -78,12 +100,22 @@ void qed_iwarp_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
 	p_hwfn->b_rdma_enabled_in_prs = true;
 }
 
+/* We have two cid maps, one for tcp which should be used only from passive
+ * syn processing and replacing a pre-allocated ep in the list. The second
+ * for active tcp and for QPs.
+ */
 static void qed_iwarp_cid_cleaned(struct qed_hwfn *p_hwfn, u32 cid)
 {
 	cid -= qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
 
 	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
-	qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, cid);
+
+	if (cid < QED_IWARP_PREALLOC_CNT)
+		qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map,
+				    cid);
+	else
+		qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, cid);
+
 	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
 }
 
@@ -107,6 +139,45 @@ static int qed_iwarp_alloc_cid(struct qed_hwfn *p_hwfn, u32 *cid)
 	return rc;
 }
 
+static void qed_iwarp_set_tcp_cid(struct qed_hwfn *p_hwfn, u32 cid)
+{
+	cid -= qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
+
+	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+	qed_bmap_set_id(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, cid);
+	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+}
+
+/* This function allocates a cid for passive tcp (called from syn receive)
+ * the reason it's separate from the regular cid allocation is because it
+ * is assured that these cids already have ilt allocated. They are preallocated
+ * to ensure that we won't need to allocate memory during syn processing
+ */
+static int qed_iwarp_alloc_tcp_cid(struct qed_hwfn *p_hwfn, u32 *cid)
+{
+	int rc;
+
+	spin_lock_bh(&p_hwfn->p_rdma_info->lock);
+
+	rc = qed_rdma_bmap_alloc_id(p_hwfn,
+				    &p_hwfn->p_rdma_info->tcp_cid_map, cid);
+
+	spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
+
+	if (rc) {
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "can't allocate iwarp tcp cid max-count=%d\n",
+			   p_hwfn->p_rdma_info->tcp_cid_map.max_count);
+
+		*cid = QED_IWARP_INVALID_TCP_CID;
+		return rc;
+	}
+
+	*cid += qed_cxt_get_proto_cid_start(p_hwfn,
+					    p_hwfn->p_rdma_info->proto);
+	return 0;
+}
+
 int qed_iwarp_create_qp(struct qed_hwfn *p_hwfn,
 			struct qed_rdma_qp *qp,
 			struct qed_rdma_create_qp_out_params *out_params)
@@ -403,6 +474,26 @@ int qed_iwarp_fw_destroy(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
 	return rc;
 }
 
+static void qed_iwarp_destroy_ep(struct qed_hwfn *p_hwfn,
+				 struct qed_iwarp_ep *ep,
+				 bool remove_from_active_list)
+{
+	dma_free_coherent(&p_hwfn->cdev->pdev->dev,
+			  sizeof(*ep->ep_buffer_virt),
+			  ep->ep_buffer_virt, ep->ep_buffer_phys);
+
+	if (remove_from_active_list) {
+		spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+		list_del(&ep->list_entry);
+		spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+	}
+
+	if (ep->qp)
+		ep->qp->ep = NULL;
+
+	kfree(ep);
+}
+
 int qed_iwarp_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
 {
 	int rc = 0;
@@ -424,6 +515,507 @@ int qed_iwarp_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
 	return rc;
 }
 
+static int
+qed_iwarp_create_ep(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep **ep_out)
+{
+	struct qed_iwarp_ep *ep;
+	int rc;
+
+	ep = kzalloc(sizeof(*ep), GFP_KERNEL);
+	if (!ep)
+		return -ENOMEM;
+
+	ep->state = QED_IWARP_EP_INIT;
+
+	ep->ep_buffer_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
+						sizeof(*ep->ep_buffer_virt),
+						&ep->ep_buffer_phys,
+						GFP_KERNEL);
+	if (!ep->ep_buffer_virt) {
+		rc = -ENOMEM;
+		goto err;
+	}
+
+	ep->sig = QED_EP_SIG;
+
+	*ep_out = ep;
+
+	return 0;
+
+err:
+	kfree(ep);
+	return rc;
+}
+
+static void
+qed_iwarp_print_tcp_ramrod(struct qed_hwfn *p_hwfn,
+			   struct iwarp_tcp_offload_ramrod_data *p_tcp_ramrod)
+{
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "local_mac=%x %x %x, remote_mac=%x %x %x\n",
+		   p_tcp_ramrod->tcp.local_mac_addr_lo,
+		   p_tcp_ramrod->tcp.local_mac_addr_mid,
+		   p_tcp_ramrod->tcp.local_mac_addr_hi,
+		   p_tcp_ramrod->tcp.remote_mac_addr_lo,
+		   p_tcp_ramrod->tcp.remote_mac_addr_mid,
+		   p_tcp_ramrod->tcp.remote_mac_addr_hi);
+
+	if (p_tcp_ramrod->tcp.ip_version == TCP_IPV4) {
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "local_ip=%pI4h:%x, remote_ip=%pI4h%x, vlan=%x\n",
+			   p_tcp_ramrod->tcp.local_ip,
+			   p_tcp_ramrod->tcp.local_port,
+			   p_tcp_ramrod->tcp.remote_ip,
+			   p_tcp_ramrod->tcp.remote_port,
+			   p_tcp_ramrod->tcp.vlan_id);
+	} else {
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "local_ip=%pI6h:%x, remote_ip=%pI6h:%x, vlan=%x\n",
+			   p_tcp_ramrod->tcp.local_ip,
+			   p_tcp_ramrod->tcp.local_port,
+			   p_tcp_ramrod->tcp.remote_ip,
+			   p_tcp_ramrod->tcp.remote_port,
+			   p_tcp_ramrod->tcp.vlan_id);
+	}
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+		   "flow_label=%x, ttl=%x, tos_or_tc=%x, mss=%x, rcv_wnd_scale=%x, connect_mode=%x, flags=%x\n",
+		   p_tcp_ramrod->tcp.flow_label,
+		   p_tcp_ramrod->tcp.ttl,
+		   p_tcp_ramrod->tcp.tos_or_tc,
+		   p_tcp_ramrod->tcp.mss,
+		   p_tcp_ramrod->tcp.rcv_wnd_scale,
+		   p_tcp_ramrod->tcp.connect_mode,
+		   p_tcp_ramrod->tcp.flags);
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "syn_ip_payload_length=%x, lo=%x, hi=%x\n",
+		   p_tcp_ramrod->tcp.syn_ip_payload_length,
+		   p_tcp_ramrod->tcp.syn_phy_addr_lo,
+		   p_tcp_ramrod->tcp.syn_phy_addr_hi);
+}
+
+static int
+qed_iwarp_tcp_offload(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
+{
+	struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
+	struct iwarp_tcp_offload_ramrod_data *p_tcp_ramrod;
+	struct tcp_offload_params_opt2 *tcp;
+	struct qed_sp_init_data init_data;
+	struct qed_spq_entry *p_ent;
+	dma_addr_t async_output_phys;
+	dma_addr_t in_pdata_phys;
+	u16 physical_q;
+	u8 tcp_flags;
+	int rc;
+	int i;
+
+	memset(&init_data, 0, sizeof(init_data));
+	init_data.cid = ep->tcp_cid;
+	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+	init_data.comp_mode = QED_SPQ_MODE_CB;
+
+	rc = qed_sp_init_request(p_hwfn, &p_ent,
+				 IWARP_RAMROD_CMD_ID_TCP_OFFLOAD,
+				 PROTOCOLID_IWARP, &init_data);
+	if (rc)
+		return rc;
+
+	p_tcp_ramrod = &p_ent->ramrod.iwarp_tcp_offload;
+
+	in_pdata_phys = ep->ep_buffer_phys +
+			offsetof(struct qed_iwarp_ep_memory, in_pdata);
+	DMA_REGPAIR_LE(p_tcp_ramrod->iwarp.incoming_ulp_buffer.addr,
+		       in_pdata_phys);
+
+	p_tcp_ramrod->iwarp.incoming_ulp_buffer.len =
+	    cpu_to_le16(sizeof(ep->ep_buffer_virt->in_pdata));
+
+	async_output_phys = ep->ep_buffer_phys +
+			    offsetof(struct qed_iwarp_ep_memory, async_output);
+	DMA_REGPAIR_LE(p_tcp_ramrod->iwarp.async_eqe_output_buf,
+		       async_output_phys);
+
+	p_tcp_ramrod->iwarp.handle_for_async.hi = cpu_to_le32(PTR_HI(ep));
+	p_tcp_ramrod->iwarp.handle_for_async.lo = cpu_to_le32(PTR_LO(ep));
+
+	physical_q = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_OFLD);
+	p_tcp_ramrod->iwarp.physical_q0 = cpu_to_le16(physical_q);
+	physical_q = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_ACK);
+	p_tcp_ramrod->iwarp.physical_q1 = cpu_to_le16(physical_q);
+	p_tcp_ramrod->iwarp.mpa_mode = iwarp_info->mpa_rev;
+
+	tcp = &p_tcp_ramrod->tcp;
+	qed_set_fw_mac_addr(&tcp->remote_mac_addr_hi,
+			    &tcp->remote_mac_addr_mid,
+			    &tcp->remote_mac_addr_lo, ep->remote_mac_addr);
+	qed_set_fw_mac_addr(&tcp->local_mac_addr_hi, &tcp->local_mac_addr_mid,
+			    &tcp->local_mac_addr_lo, ep->local_mac_addr);
+
+	tcp->vlan_id = cpu_to_le16(ep->cm_info.vlan);
+
+	tcp_flags = p_hwfn->p_rdma_info->iwarp.tcp_flags;
+	tcp->flags = 0;
+	SET_FIELD(tcp->flags, TCP_OFFLOAD_PARAMS_OPT2_TS_EN,
+		  !!(tcp_flags & QED_IWARP_TS_EN));
+
+	SET_FIELD(tcp->flags, TCP_OFFLOAD_PARAMS_OPT2_DA_EN,
+		  !!(tcp_flags & QED_IWARP_DA_EN));
+
+	tcp->ip_version = ep->cm_info.ip_version;
+
+	for (i = 0; i < 4; i++) {
+		tcp->remote_ip[i] = cpu_to_le32(ep->cm_info.remote_ip[i]);
+		tcp->local_ip[i] = cpu_to_le32(ep->cm_info.local_ip[i]);
+	}
+
+	tcp->remote_port = cpu_to_le16(ep->cm_info.remote_port);
+	tcp->local_port = cpu_to_le16(ep->cm_info.local_port);
+	tcp->mss = cpu_to_le16(ep->mss);
+	tcp->flow_label = 0;
+	tcp->ttl = 0x40;
+	tcp->tos_or_tc = 0;
+
+	tcp->rcv_wnd_scale = (u8)p_hwfn->p_rdma_info->iwarp.rcv_wnd_scale;
+	tcp->connect_mode = ep->connect_mode;
+
+	if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
+		tcp->syn_ip_payload_length =
+			cpu_to_le16(ep->syn_ip_payload_length);
+		tcp->syn_phy_addr_hi = DMA_HI_LE(ep->syn_phy_addr);
+		tcp->syn_phy_addr_lo = DMA_LO_LE(ep->syn_phy_addr);
+	}
+
+	qed_iwarp_print_tcp_ramrod(p_hwfn, p_tcp_ramrod);
+
+	rc = qed_spq_post(p_hwfn, p_ent, NULL);
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+		   "EP(0x%x) Offload completed rc=%d\n", ep->tcp_cid, rc);
+
+	return rc;
+}
+
+static void
+qed_iwarp_mpa_received(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
+{
+	struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
+	struct qed_iwarp_cm_event_params params;
+	struct mpa_v2_hdr *mpa_v2;
+	union async_output *async_data;
+	u16 mpa_ord, mpa_ird;
+	u8 mpa_hdr_size = 0;
+	u8 mpa_rev;
+
+	async_data = &ep->ep_buffer_virt->async_output;
+
+	mpa_rev = async_data->mpa_request.mpa_handshake_mode;
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+		   "private_data_len=%x handshake_mode=%x private_data=(%x)\n",
+		   async_data->mpa_request.ulp_data_len,
+		   mpa_rev, *((u32 *)((u8 *)ep->ep_buffer_virt->in_pdata)));
+
+	if (mpa_rev == MPA_NEGOTIATION_TYPE_ENHANCED) {
+		/* Read ord/ird values from private data buffer */
+		mpa_v2 = (struct mpa_v2_hdr *)ep->ep_buffer_virt->in_pdata;
+		mpa_hdr_size = sizeof(*mpa_v2);
+
+		mpa_ord = ntohs(mpa_v2->ord);
+		mpa_ird = ntohs(mpa_v2->ird);
+
+		/* Temprary store in cm_info incoming ord/ird requested, later
+		 * replace with negotiated value during accept
+		 */
+		ep->cm_info.ord = (u8)min_t(u16,
+					    (mpa_ord & MPA_V2_IRD_ORD_MASK),
+					    QED_IWARP_ORD_DEFAULT);
+
+		ep->cm_info.ird = (u8)min_t(u16,
+					    (mpa_ird & MPA_V2_IRD_ORD_MASK),
+					    QED_IWARP_IRD_DEFAULT);
+
+		/* Peer2Peer negotiation */
+		ep->rtr_type = MPA_RTR_TYPE_NONE;
+		if (mpa_ird & MPA_V2_PEER2PEER_MODEL) {
+			if (mpa_ord & MPA_V2_WRITE_RTR)
+				ep->rtr_type |= MPA_RTR_TYPE_ZERO_WRITE;
+
+			if (mpa_ord & MPA_V2_READ_RTR)
+				ep->rtr_type |= MPA_RTR_TYPE_ZERO_READ;
+
+			if (mpa_ird & MPA_V2_SEND_RTR)
+				ep->rtr_type |= MPA_RTR_TYPE_ZERO_SEND;
+
+			ep->rtr_type &= iwarp_info->rtr_type;
+
+			/* if we're left with no match send our capabilities */
+			if (ep->rtr_type == MPA_RTR_TYPE_NONE)
+				ep->rtr_type = iwarp_info->rtr_type;
+		}
+
+		ep->mpa_rev = MPA_NEGOTIATION_TYPE_ENHANCED;
+	} else {
+		ep->cm_info.ord = QED_IWARP_ORD_DEFAULT;
+		ep->cm_info.ird = QED_IWARP_IRD_DEFAULT;
+		ep->mpa_rev = MPA_NEGOTIATION_TYPE_BASIC;
+	}
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+		   "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x rtr:0x%x ulp_data_len = %x mpa_hdr_size = %x\n",
+		   mpa_rev, ep->cm_info.ord, ep->cm_info.ird, ep->rtr_type,
+		   async_data->mpa_request.ulp_data_len, mpa_hdr_size);
+
+	/* Strip mpa v2 hdr from private data before sending to upper layer */
+	ep->cm_info.private_data = ep->ep_buffer_virt->in_pdata + mpa_hdr_size;
+
+	ep->cm_info.private_data_len = async_data->mpa_request.ulp_data_len -
+				       mpa_hdr_size;
+
+	params.event = QED_IWARP_EVENT_MPA_REQUEST;
+	params.cm_info = &ep->cm_info;
+	params.ep_context = ep;
+	params.status = 0;
+
+	ep->state = QED_IWARP_EP_MPA_REQ_RCVD;
+	ep->event_cb(ep->cb_context, &params);
+}
+
+static int
+qed_iwarp_mpa_offload(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
+{
+	struct iwarp_mpa_offload_ramrod_data *p_mpa_ramrod;
+	struct qed_sp_init_data init_data;
+	dma_addr_t async_output_phys;
+	struct qed_spq_entry *p_ent;
+	dma_addr_t out_pdata_phys;
+	dma_addr_t in_pdata_phys;
+	struct qed_rdma_qp *qp;
+	bool reject;
+	int rc;
+
+	if (!ep)
+		return -EINVAL;
+
+	qp = ep->qp;
+	reject = !qp;
+
+	memset(&init_data, 0, sizeof(init_data));
+	init_data.cid = reject ? ep->tcp_cid : qp->icid;
+	init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
+
+	init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
+
+	rc = qed_sp_init_request(p_hwfn, &p_ent,
+				 IWARP_RAMROD_CMD_ID_MPA_OFFLOAD,
+				 PROTOCOLID_IWARP, &init_data);
+	if (rc)
+		return rc;
+
+	p_mpa_ramrod = &p_ent->ramrod.iwarp_mpa_offload;
+	out_pdata_phys = ep->ep_buffer_phys +
+			 offsetof(struct qed_iwarp_ep_memory, out_pdata);
+	DMA_REGPAIR_LE(p_mpa_ramrod->common.outgoing_ulp_buffer.addr,
+		       out_pdata_phys);
+	p_mpa_ramrod->common.outgoing_ulp_buffer.len =
+	    ep->cm_info.private_data_len;
+	p_mpa_ramrod->common.crc_needed = p_hwfn->p_rdma_info->iwarp.crc_needed;
+
+	p_mpa_ramrod->common.out_rq.ord = ep->cm_info.ord;
+	p_mpa_ramrod->common.out_rq.ird = ep->cm_info.ird;
+
+	p_mpa_ramrod->tcp_cid = p_hwfn->hw_info.opaque_fid << 16 | ep->tcp_cid;
+
+	in_pdata_phys = ep->ep_buffer_phys +
+			offsetof(struct qed_iwarp_ep_memory, in_pdata);
+	p_mpa_ramrod->tcp_connect_side = ep->connect_mode;
+	DMA_REGPAIR_LE(p_mpa_ramrod->incoming_ulp_buffer.addr,
+		       in_pdata_phys);
+	p_mpa_ramrod->incoming_ulp_buffer.len =
+	    cpu_to_le16(sizeof(ep->ep_buffer_virt->in_pdata));
+	async_output_phys = ep->ep_buffer_phys +
+			    offsetof(struct qed_iwarp_ep_memory, async_output);
+	DMA_REGPAIR_LE(p_mpa_ramrod->async_eqe_output_buf,
+		       async_output_phys);
+	p_mpa_ramrod->handle_for_async.hi = cpu_to_le32(PTR_HI(ep));
+	p_mpa_ramrod->handle_for_async.lo = cpu_to_le32(PTR_LO(ep));
+
+	if (!reject) {
+		DMA_REGPAIR_LE(p_mpa_ramrod->shared_queue_addr,
+			       qp->shared_queue_phys_addr);
+		p_mpa_ramrod->stats_counter_id =
+		    RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) + qp->stats_queue;
+	} else {
+		p_mpa_ramrod->common.reject = 1;
+	}
+
+	p_mpa_ramrod->mode = ep->mpa_rev;
+	SET_FIELD(p_mpa_ramrod->rtr_pref,
+		  IWARP_MPA_OFFLOAD_RAMROD_DATA_RTR_SUPPORTED, ep->rtr_type);
+
+	ep->state = QED_IWARP_EP_MPA_OFFLOADED;
+	rc = qed_spq_post(p_hwfn, p_ent, NULL);
+	if (!reject)
+		ep->cid = qp->icid;	/* Now they're migrated. */
+
+	DP_VERBOSE(p_hwfn,
+		   QED_MSG_RDMA,
+		   "QP(0x%x) EP(0x%x) MPA Offload rc = %d IRD=0x%x ORD=0x%x rtr_type=%d mpa_rev=%d reject=%d\n",
+		   reject ? 0xffff : qp->icid,
+		   ep->tcp_cid,
+		   rc,
+		   ep->cm_info.ird,
+		   ep->cm_info.ord, ep->rtr_type, ep->mpa_rev, reject);
+	return rc;
+}
+
+static void
+qed_iwarp_return_ep(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
+{
+	ep->state = QED_IWARP_EP_INIT;
+	if (ep->qp)
+		ep->qp->ep = NULL;
+	ep->qp = NULL;
+	memset(&ep->cm_info, 0, sizeof(ep->cm_info));
+
+	if (ep->tcp_cid == QED_IWARP_INVALID_TCP_CID) {
+		/* We don't care about the return code, it's ok if tcp_cid
+		 * remains invalid...in this case we'll defer allocation
+		 */
+		qed_iwarp_alloc_tcp_cid(p_hwfn, &ep->tcp_cid);
+	}
+	spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+
+	list_del(&ep->list_entry);
+	list_add_tail(&ep->list_entry,
+		      &p_hwfn->p_rdma_info->iwarp.ep_free_list);
+
+	spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+}
+
+#define QED_IWARP_CONNECT_MODE_STRING(ep) \
+	((ep)->connect_mode == TCP_CONNECT_PASSIVE) ? "Passive" : "Active"
+
+/* Called as a result of the event:
+ * IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE
+ */
+static void
+qed_iwarp_mpa_complete(struct qed_hwfn *p_hwfn,
+		       struct qed_iwarp_ep *ep, u8 fw_return_code)
+{
+	struct qed_iwarp_cm_event_params params;
+
+	params.event = QED_IWARP_EVENT_PASSIVE_COMPLETE;
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+		   "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x\n",
+		   ep->mpa_rev, ep->cm_info.ord, ep->cm_info.ird);
+
+	params.cm_info = &ep->cm_info;
+
+	params.ep_context = ep;
+
+	ep->state = QED_IWARP_EP_CLOSED;
+
+	switch (fw_return_code) {
+	case RDMA_RETURN_OK:
+		ep->qp->max_rd_atomic_req = ep->cm_info.ord;
+		ep->qp->max_rd_atomic_resp = ep->cm_info.ird;
+		qed_iwarp_modify_qp(p_hwfn, ep->qp, QED_IWARP_QP_STATE_RTS, 1);
+		ep->state = QED_IWARP_EP_ESTABLISHED;
+		params.status = 0;
+		break;
+	default:
+		params.status = -ECONNRESET;
+		break;
+	}
+
+	ep->event_cb(ep->cb_context, &params);
+}
+
+static void
+qed_iwarp_mpa_v2_set_private(struct qed_hwfn *p_hwfn,
+			     struct qed_iwarp_ep *ep, u8 *mpa_data_size)
+{
+	struct mpa_v2_hdr *mpa_v2_params;
+	u16 mpa_ird, mpa_ord;
+
+	*mpa_data_size = 0;
+	if (MPA_REV2(ep->mpa_rev)) {
+		mpa_v2_params =
+		    (struct mpa_v2_hdr *)ep->ep_buffer_virt->out_pdata;
+		*mpa_data_size = sizeof(*mpa_v2_params);
+
+		mpa_ird = (u16)ep->cm_info.ird;
+		mpa_ord = (u16)ep->cm_info.ord;
+
+		if (ep->rtr_type != MPA_RTR_TYPE_NONE) {
+			mpa_ird |= MPA_V2_PEER2PEER_MODEL;
+
+			if (ep->rtr_type & MPA_RTR_TYPE_ZERO_SEND)
+				mpa_ird |= MPA_V2_SEND_RTR;
+
+			if (ep->rtr_type & MPA_RTR_TYPE_ZERO_WRITE)
+				mpa_ord |= MPA_V2_WRITE_RTR;
+
+			if (ep->rtr_type & MPA_RTR_TYPE_ZERO_READ)
+				mpa_ord |= MPA_V2_READ_RTR;
+		}
+
+		mpa_v2_params->ird = htons(mpa_ird);
+		mpa_v2_params->ord = htons(mpa_ord);
+
+		DP_VERBOSE(p_hwfn,
+			   QED_MSG_RDMA,
+			   "MPA_NEGOTIATE Header: [%x ord:%x ird] %x ord:%x ird:%x peer2peer:%x rtr_send:%x rtr_write:%x rtr_read:%x\n",
+			   mpa_v2_params->ird,
+			   mpa_v2_params->ord,
+			   *((u32 *)mpa_v2_params),
+			   mpa_ord & MPA_V2_IRD_ORD_MASK,
+			   mpa_ird & MPA_V2_IRD_ORD_MASK,
+			   !!(mpa_ird & MPA_V2_PEER2PEER_MODEL),
+			   !!(mpa_ird & MPA_V2_SEND_RTR),
+			   !!(mpa_ord & MPA_V2_WRITE_RTR),
+			   !!(mpa_ord & MPA_V2_READ_RTR));
+	}
+}
+
+static struct qed_iwarp_ep *qed_iwarp_get_free_ep(struct qed_hwfn *p_hwfn)
+{
+	struct qed_iwarp_ep *ep = NULL;
+	int rc;
+
+	spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+
+	if (list_empty(&p_hwfn->p_rdma_info->iwarp.ep_free_list)) {
+		DP_ERR(p_hwfn, "Ep list is empty\n");
+		goto out;
+	}
+
+	ep = list_first_entry(&p_hwfn->p_rdma_info->iwarp.ep_free_list,
+			      struct qed_iwarp_ep, list_entry);
+
+	/* in some cases we could have failed allocating a tcp cid when added
+	 * from accept / failure... retry now..this is not the common case.
+	 */
+	if (ep->tcp_cid == QED_IWARP_INVALID_TCP_CID) {
+		rc = qed_iwarp_alloc_tcp_cid(p_hwfn, &ep->tcp_cid);
+
+		/* if we fail we could look for another entry with a valid
+		 * tcp_cid, but since we don't expect to reach this anyway
+		 * it's not worth the handling
+		 */
+		if (rc) {
+			ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
+			ep = NULL;
+			goto out;
+		}
+	}
+
+	list_del(&ep->list_entry);
+
+out:
+	spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+	return ep;
+}
+
 #define QED_IWARP_MAX_CID_CLEAN_TIME  100
 #define QED_IWARP_MAX_NO_PROGRESS_CNT 5
 
@@ -465,20 +1057,213 @@ qed_iwarp_wait_cid_map_cleared(struct qed_hwfn *p_hwfn, struct qed_bmap *bmap)
 
 static int qed_iwarp_wait_for_all_cids(struct qed_hwfn *p_hwfn)
 {
+	int rc;
+	int i;
+
+	rc = qed_iwarp_wait_cid_map_cleared(p_hwfn,
+					    &p_hwfn->p_rdma_info->tcp_cid_map);
+	if (rc)
+		return rc;
+
+	/* Now free the tcp cids from the main cid map */
+	for (i = 0; i < QED_IWARP_PREALLOC_CNT; i++)
+		qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, i);
+
 	/* Now wait for all cids to be completed */
 	return qed_iwarp_wait_cid_map_cleared(p_hwfn,
 					      &p_hwfn->p_rdma_info->cid_map);
 }
 
+static void qed_iwarp_free_prealloc_ep(struct qed_hwfn *p_hwfn)
+{
+	struct qed_iwarp_ep *ep;
+
+	while (!list_empty(&p_hwfn->p_rdma_info->iwarp.ep_free_list)) {
+		spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+
+		ep = list_first_entry(&p_hwfn->p_rdma_info->iwarp.ep_free_list,
+				      struct qed_iwarp_ep, list_entry);
+
+		if (!ep) {
+			spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+			break;
+		}
+		list_del(&ep->list_entry);
+
+		spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+
+		if (ep->tcp_cid != QED_IWARP_INVALID_TCP_CID)
+			qed_iwarp_cid_cleaned(p_hwfn, ep->tcp_cid);
+
+		qed_iwarp_destroy_ep(p_hwfn, ep, false);
+	}
+}
+
+static int qed_iwarp_prealloc_ep(struct qed_hwfn *p_hwfn, bool init)
+{
+	struct qed_iwarp_ep *ep;
+	int rc = 0;
+	int count;
+	u32 cid;
+	int i;
+
+	count = init ? QED_IWARP_PREALLOC_CNT : 1;
+	for (i = 0; i < count; i++) {
+		rc = qed_iwarp_create_ep(p_hwfn, &ep);
+		if (rc)
+			return rc;
+
+		/* During initialization we allocate from the main pool,
+		 * afterwards we allocate only from the tcp_cid.
+		 */
+		if (init) {
+			rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
+			if (rc)
+				goto err;
+			qed_iwarp_set_tcp_cid(p_hwfn, cid);
+		} else {
+			/* We don't care about the return code, it's ok if
+			 * tcp_cid remains invalid...in this case we'll
+			 * defer allocation
+			 */
+			qed_iwarp_alloc_tcp_cid(p_hwfn, &cid);
+		}
+
+		ep->tcp_cid = cid;
+
+		spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+		list_add_tail(&ep->list_entry,
+			      &p_hwfn->p_rdma_info->iwarp.ep_free_list);
+		spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+	}
+
+	return rc;
+
+err:
+	qed_iwarp_destroy_ep(p_hwfn, ep, false);
+
+	return rc;
+}
+
 int qed_iwarp_alloc(struct qed_hwfn *p_hwfn)
 {
+	int rc;
+
+	/* Allocate bitmap for tcp cid. These are used by passive side
+	 * to ensure it can allocate a tcp cid during dpc that was
+	 * pre-acquired and doesn't require dynamic allocation of ilt
+	 */
+	rc = qed_rdma_bmap_alloc(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map,
+				 QED_IWARP_PREALLOC_CNT, "TCP_CID");
+	if (rc) {
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "Failed to allocate tcp cid, rc = %d\n", rc);
+		return rc;
+	}
+
+	INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.ep_free_list);
 	spin_lock_init(&p_hwfn->p_rdma_info->iwarp.iw_lock);
 
-	return 0;
+	return qed_iwarp_prealloc_ep(p_hwfn, true);
 }
 
 void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn)
 {
+	qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1);
+}
+
+int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams)
+{
+	struct qed_hwfn *p_hwfn = (struct qed_hwfn *)rdma_cxt;
+	struct qed_iwarp_ep *ep;
+	u8 mpa_data_size = 0;
+	int rc;
+
+	ep = (struct qed_iwarp_ep *)iparams->ep_context;
+	if (!ep) {
+		DP_ERR(p_hwfn, "Ep Context receive in accept is NULL\n");
+		return -EINVAL;
+	}
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x)\n",
+		   iparams->qp->icid, ep->tcp_cid);
+
+	if ((iparams->ord > QED_IWARP_ORD_DEFAULT) ||
+	    (iparams->ird > QED_IWARP_IRD_DEFAULT)) {
+		DP_VERBOSE(p_hwfn,
+			   QED_MSG_RDMA,
+			   "QP(0x%x) EP(0x%x) ERROR: Invalid ord(0x%x)/ird(0x%x)\n",
+			   iparams->qp->icid,
+			   ep->tcp_cid, iparams->ord, iparams->ord);
+		return -EINVAL;
+	}
+
+	qed_iwarp_prealloc_ep(p_hwfn, false);
+
+	ep->cb_context = iparams->cb_context;
+	ep->qp = iparams->qp;
+	ep->qp->ep = ep;
+
+	if (ep->mpa_rev == MPA_NEGOTIATION_TYPE_ENHANCED) {
+		/* Negotiate ord/ird: if upperlayer requested ord larger than
+		 * ird advertised by remote, we need to decrease our ord
+		 */
+		if (iparams->ord > ep->cm_info.ird)
+			iparams->ord = ep->cm_info.ird;
+
+		if ((ep->rtr_type & MPA_RTR_TYPE_ZERO_READ) &&
+		    (iparams->ird == 0))
+			iparams->ird = 1;
+	}
+
+	/* Update cm_info ord/ird to be negotiated values */
+	ep->cm_info.ord = iparams->ord;
+	ep->cm_info.ird = iparams->ird;
+
+	qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
+
+	ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
+	ep->cm_info.private_data_len = iparams->private_data_len +
+				       mpa_data_size;
+
+	memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
+	       iparams->private_data, iparams->private_data_len);
+
+	rc = qed_iwarp_mpa_offload(p_hwfn, ep);
+	if (rc)
+		qed_iwarp_modify_qp(p_hwfn,
+				    iparams->qp, QED_IWARP_QP_STATE_ERROR, 1);
+
+	return rc;
+}
+
+int qed_iwarp_reject(void *rdma_cxt, struct qed_iwarp_reject_in *iparams)
+{
+	struct qed_hwfn *p_hwfn = rdma_cxt;
+	struct qed_iwarp_ep *ep;
+	u8 mpa_data_size = 0;
+
+	ep = iparams->ep_context;
+	if (!ep) {
+		DP_ERR(p_hwfn, "Ep Context receive in reject is NULL\n");
+		return -EINVAL;
+	}
+
+	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "EP(0x%x)\n", ep->tcp_cid);
+
+	ep->cb_context = iparams->cb_context;
+	ep->qp = NULL;
+
+	qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
+
+	ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
+	ep->cm_info.private_data_len = iparams->private_data_len +
+				       mpa_data_size;
+
+	memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
+	       iparams->private_data, iparams->private_data_len);
+
+	return qed_iwarp_mpa_offload(p_hwfn, ep);
 }
 
 static void
@@ -526,6 +1311,38 @@ qed_iwarp_ll2_post_rx(struct qed_hwfn *p_hwfn,
 	return rc;
 }
 
+static bool
+qed_iwarp_ep_exists(struct qed_hwfn *p_hwfn, struct qed_iwarp_cm_info *cm_info)
+{
+	struct qed_iwarp_ep *ep = NULL;
+	bool found = false;
+
+	list_for_each_entry(ep,
+			    &p_hwfn->p_rdma_info->iwarp.ep_list,
+			    list_entry) {
+		if ((ep->cm_info.local_port == cm_info->local_port) &&
+		    (ep->cm_info.remote_port == cm_info->remote_port) &&
+		    (ep->cm_info.vlan == cm_info->vlan) &&
+		    !memcmp(&ep->cm_info.local_ip, cm_info->local_ip,
+			    sizeof(cm_info->local_ip)) &&
+		    !memcmp(&ep->cm_info.remote_ip, cm_info->remote_ip,
+			    sizeof(cm_info->remote_ip))) {
+			found = true;
+			break;
+		}
+	}
+
+	if (found) {
+		DP_NOTICE(p_hwfn,
+			  "SYN received on active connection - dropping\n");
+		qed_iwarp_print_cm_info(p_hwfn, cm_info);
+
+		return true;
+	}
+
+	return false;
+}
+
 static struct qed_iwarp_listener *
 qed_iwarp_get_listener(struct qed_hwfn *p_hwfn,
 		       struct qed_iwarp_cm_info *cm_info)
@@ -596,9 +1413,8 @@ qed_iwarp_parse_rx_pkt(struct qed_hwfn *p_hwfn,
 
 	eth_hlen = ETH_HLEN + (vlan_valid ? sizeof(u32) : 0);
 
-	memcpy(remote_mac_addr, ethh->h_source, ETH_ALEN);
-
-	memcpy(local_mac_addr, ethh->h_dest, ETH_ALEN);
+	ether_addr_copy(remote_mac_addr, ethh->h_source);
+	ether_addr_copy(local_mac_addr, ethh->h_dest);
 
 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "eth_type =%d source mac: %pM\n",
 		   eth_type, ethh->h_source);
@@ -661,9 +1477,12 @@ qed_iwarp_ll2_comp_syn_pkt(void *cxt, struct qed_ll2_comp_rx_data *data)
 	struct qed_hwfn *p_hwfn = cxt;
 	u8 remote_mac_addr[ETH_ALEN];
 	u8 local_mac_addr[ETH_ALEN];
+	struct qed_iwarp_ep *ep;
 	int tcp_start_offset;
+	u8 ts_hdr_size = 0;
 	u8 ll2_syn_handle;
 	int payload_len;
+	u32 hdr_size;
 	int rc;
 
 	memset(&cm_info, 0, sizeof(cm_info));
@@ -719,6 +1538,49 @@ qed_iwarp_ll2_comp_syn_pkt(void *cxt, struct qed_ll2_comp_rx_data *data)
 	}
 
 	DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Received syn on listening port\n");
+	/* There may be an open ep on this connection if this is a syn
+	 * retrasnmit... need to make sure there isn't...
+	 */
+	if (qed_iwarp_ep_exists(p_hwfn, &cm_info))
+		goto err;
+
+	ep = qed_iwarp_get_free_ep(p_hwfn);
+	if (!ep)
+		goto err;
+
+	spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+	list_add_tail(&ep->list_entry, &p_hwfn->p_rdma_info->iwarp.ep_list);
+	spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
+
+	ether_addr_copy(ep->remote_mac_addr, remote_mac_addr);
+	ether_addr_copy(ep->local_mac_addr, local_mac_addr);
+
+	memcpy(&ep->cm_info, &cm_info, sizeof(ep->cm_info));
+
+	if (p_hwfn->p_rdma_info->iwarp.tcp_flags & QED_IWARP_TS_EN)
+		ts_hdr_size = TIMESTAMP_HEADER_SIZE;
+
+	hdr_size = ((cm_info.ip_version == QED_TCP_IPV4) ? 40 : 60) +
+		   ts_hdr_size;
+	ep->mss = p_hwfn->p_rdma_info->iwarp.max_mtu - hdr_size;
+	ep->mss = min_t(u16, QED_IWARP_MAX_FW_MSS, ep->mss);
+
+	ep->event_cb = listener->event_cb;
+	ep->cb_context = listener->cb_context;
+	ep->connect_mode = TCP_CONNECT_PASSIVE;
+
+	ep->syn = buf;
+	ep->syn_ip_payload_length = (u16)payload_len;
+	ep->syn_phy_addr = buf->data_phys_addr + data->u.placement_offset +
+			   tcp_start_offset;
+
+	rc = qed_iwarp_tcp_offload(p_hwfn, ep);
+	if (rc) {
+		qed_iwarp_return_ep(p_hwfn, ep);
+		goto err;
+	}
+
+	return;
 err:
 	qed_iwarp_ll2_post_rx(p_hwfn, buf, ll2_syn_handle);
 }
@@ -905,7 +1767,12 @@ int qed_iwarp_setup(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
 
 	iwarp_info->peer2peer = QED_IWARP_PARAM_P2P;
 
+	iwarp_info->rtr_type =  MPA_RTR_TYPE_ZERO_SEND |
+				MPA_RTR_TYPE_ZERO_WRITE |
+				MPA_RTR_TYPE_ZERO_READ;
+
 	spin_lock_init(&p_hwfn->p_rdma_info->iwarp.qp_lock);
+	INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.ep_list);
 	INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.listen_list);
 
 	qed_spq_register_async_cb(p_hwfn, PROTOCOLID_IWARP,
@@ -918,6 +1785,7 @@ int qed_iwarp_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
 {
 	int rc;
 
+	qed_iwarp_free_prealloc_ep(p_hwfn);
 	rc = qed_iwarp_wait_for_all_cids(p_hwfn);
 	if (rc)
 		return rc;
@@ -927,11 +1795,70 @@ int qed_iwarp_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
 	return qed_iwarp_ll2_stop(p_hwfn, p_ptt);
 }
 
+void
+qed_iwarp_connect_complete(struct qed_hwfn *p_hwfn,
+			   struct qed_iwarp_ep *ep, u8 fw_return_code)
+{
+	/* Done with the SYN packet, post back to ll2 rx */
+	qed_iwarp_ll2_post_rx(p_hwfn, ep->syn,
+			      p_hwfn->p_rdma_info->iwarp.ll2_syn_handle);
+	ep->syn = NULL;
+
+	/* If connect failed - upper layer doesn't know about it */
+	qed_iwarp_mpa_received(p_hwfn, ep);
+}
+
+static inline bool
+qed_iwarp_check_ep_ok(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
+{
+	if (!ep || (ep->sig != QED_EP_SIG)) {
+		DP_ERR(p_hwfn, "ERROR ON ASYNC ep=%p\n", ep);
+		return false;
+	}
+
+	return true;
+}
+
 static int qed_iwarp_async_event(struct qed_hwfn *p_hwfn,
 				 u8 fw_event_code, u16 echo,
 				 union event_ring_data *data,
 				 u8 fw_return_code)
 {
+	struct regpair *fw_handle = &data->rdma_data.async_handle;
+	struct qed_iwarp_ep *ep = NULL;
+	u16 cid;
+
+	ep = (struct qed_iwarp_ep *)(uintptr_t)HILO_64(fw_handle->hi,
+						       fw_handle->lo);
+
+	switch (fw_event_code) {
+	case IWARP_EVENT_TYPE_ASYNC_CONNECT_COMPLETE:
+		/* Async completion after TCP 3-way handshake */
+		if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
+			return -EINVAL;
+		DP_VERBOSE(p_hwfn,
+			   QED_MSG_RDMA,
+			   "EP(0x%x) IWARP_EVENT_TYPE_ASYNC_CONNECT_COMPLETE fw_ret_code=%d\n",
+			   ep->tcp_cid, fw_return_code);
+		qed_iwarp_connect_complete(p_hwfn, ep, fw_return_code);
+		break;
+	case IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE:
+		if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
+			return -EINVAL;
+		DP_VERBOSE(p_hwfn,
+			   QED_MSG_RDMA,
+			   "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE fw_ret_code=%d\n",
+			   ep->cid, fw_return_code);
+		qed_iwarp_mpa_complete(p_hwfn, ep, fw_return_code);
+		break;
+	case IWARP_EVENT_TYPE_ASYNC_CID_CLEANED:
+		cid = (u16)le32_to_cpu(fw_handle->lo);
+		DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
+			   "(0x%x)IWARP_EVENT_TYPE_ASYNC_CID_CLEANED\n", cid);
+		qed_iwarp_cid_cleaned(p_hwfn, cid);
+
+		break;
+	}
 	return 0;
 }
 
diff --git a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h
index 29005ac83a6a..bedac98c834c 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_iwarp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_iwarp.h
@@ -42,6 +42,8 @@ enum qed_iwarp_qp_state {
 
 enum qed_iwarp_qp_state qed_roce2iwarp_state(enum qed_roce_qp_state state);
 
+#define QED_IWARP_PREALLOC_CNT  (256)
+
 #define QED_IWARP_LL2_SYN_TX_SIZE       (128)
 #define QED_IWARP_LL2_SYN_RX_SIZE       (256)
 #define QED_IWARP_MAX_SYN_PKT_SIZE      (128)
@@ -55,6 +57,8 @@ struct qed_iwarp_ll2_buff {
 
 struct qed_iwarp_info {
 	struct list_head listen_list;	/* qed_iwarp_listener */
+	struct list_head ep_list;	/* qed_iwarp_ep */
+	struct list_head ep_free_list;	/* pre-allocated ep's */
 	spinlock_t iw_lock;	/* for iwarp resources */
 	spinlock_t qp_lock;	/* for teardown races */
 	u32 rcv_wnd_scale;
@@ -68,6 +72,61 @@ struct qed_iwarp_info {
 	enum mpa_rtr_type rtr_type;
 };
 
+enum qed_iwarp_ep_state {
+	QED_IWARP_EP_INIT,
+	QED_IWARP_EP_MPA_REQ_RCVD,
+	QED_IWARP_EP_MPA_OFFLOADED,
+	QED_IWARP_EP_ESTABLISHED,
+	QED_IWARP_EP_CLOSED
+};
+
+union async_output {
+	struct iwarp_eqe_data_mpa_async_completion mpa_response;
+	struct iwarp_eqe_data_tcp_async_completion mpa_request;
+};
+
+#define QED_MAX_PRIV_DATA_LEN (512)
+struct qed_iwarp_ep_memory {
+	u8 in_pdata[QED_MAX_PRIV_DATA_LEN];
+	u8 out_pdata[QED_MAX_PRIV_DATA_LEN];
+	union async_output async_output;
+};
+
+/* Endpoint structure represents a TCP connection. This connection can be
+ * associated with a QP or not (in which case QP==NULL)
+ */
+struct qed_iwarp_ep {
+	struct list_head list_entry;
+	struct qed_rdma_qp *qp;
+	struct qed_iwarp_ep_memory *ep_buffer_virt;
+	dma_addr_t ep_buffer_phys;
+	enum qed_iwarp_ep_state state;
+	int sig;
+	struct qed_iwarp_cm_info cm_info;
+	enum tcp_connect_mode connect_mode;
+	enum mpa_rtr_type rtr_type;
+	enum mpa_negotiation_mode mpa_rev;
+	u32 tcp_cid;
+	u32 cid;
+	u16 mss;
+	u8 remote_mac_addr[6];
+	u8 local_mac_addr[6];
+	bool mpa_reply_processed;
+
+	/* For Passive side - syn packet related data */
+	u16 syn_ip_payload_length;
+	struct qed_iwarp_ll2_buff *syn;
+	dma_addr_t syn_phy_addr;
+
+	/* The event_cb function is called for asynchrounous events associated
+	 * with the ep. It is initialized at different entry points depending
+	 * on whether the ep is the tcp connection active side or passive side
+	 * The cb_context is passed to the event_cb function.
+	 */
+	iwarp_event_handler event_cb;
+	void *cb_context;
+};
+
 struct qed_iwarp_listener {
 	struct list_head list_entry;
 
@@ -115,6 +174,9 @@ qed_iwarp_create_listen(void *rdma_cxt,
 			struct qed_iwarp_listen_in *iparams,
 			struct qed_iwarp_listen_out *oparams);
 
+int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams);
+
+int qed_iwarp_reject(void *rdma_cxt, struct qed_iwarp_reject_in *iparams);
 int qed_iwarp_destroy_listen(void *rdma_cxt, void *handle);
 
 #endif
diff --git a/drivers/net/ethernet/qlogic/qed/qed_l2.c b/drivers/net/ethernet/qlogic/qed/qed_l2.c
index 27ea54ba7e1b..0ba5ec8a9814 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_l2.c
+++ b/drivers/net/ethernet/qlogic/qed/qed_l2.c
@@ -1227,19 +1227,6 @@ static enum eth_filter_action qed_filter_action(enum qed_filter_opcode opcode)
 	return action;
 }
 
-static void qed_set_fw_mac_addr(__le16 *fw_msb,
-				__le16 *fw_mid,
-				__le16 *fw_lsb,
-				u8 *mac)
-{
-	((u8 *)fw_msb)[0] = mac[1];
-	((u8 *)fw_msb)[1] = mac[0];
-	((u8 *)fw_mid)[0] = mac[3];
-	((u8 *)fw_mid)[1] = mac[2];
-	((u8 *)fw_lsb)[0] = mac[5];
-	((u8 *)fw_lsb)[1] = mac[4];
-}
-
 static int
 qed_filter_ucast_common(struct qed_hwfn *p_hwfn,
 			u16 opaque_fid,
diff --git a/drivers/net/ethernet/qlogic/qed/qed_rdma.h b/drivers/net/ethernet/qlogic/qed/qed_rdma.h
index 90e4e0f13727..18ec9cbd84f5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_rdma.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_rdma.h
@@ -85,6 +85,7 @@ struct qed_rdma_info {
 	struct qed_bmap qp_map;
 	struct qed_bmap srq_map;
 	struct qed_bmap cid_map;
+	struct qed_bmap tcp_cid_map;
 	struct qed_bmap real_cid_map;
 	struct qed_bmap dpi_map;
 	struct qed_bmap toggle_bits;
@@ -167,6 +168,7 @@ struct qed_rdma_qp {
 
 	void *shared_queue;
 	dma_addr_t shared_queue_phys_addr;
+	struct qed_iwarp_ep *ep;
 };
 
 #if IS_ENABLED(CONFIG_QED_RDMA)
diff --git a/drivers/net/ethernet/qlogic/qed/qed_sp.h b/drivers/net/ethernet/qlogic/qed/qed_sp.h
index c3752c56e54d..ab4ad8a1e2a5 100644
--- a/drivers/net/ethernet/qlogic/qed/qed_sp.h
+++ b/drivers/net/ethernet/qlogic/qed/qed_sp.h
@@ -111,6 +111,8 @@ union ramrod_data {
 	struct rdma_srq_destroy_ramrod_data rdma_destroy_srq;
 	struct rdma_srq_modify_ramrod_data rdma_modify_srq;
 	struct iwarp_create_qp_ramrod_data iwarp_create_qp;
+	struct iwarp_tcp_offload_ramrod_data iwarp_tcp_offload;
+	struct iwarp_mpa_offload_ramrod_data iwarp_mpa_offload;
 	struct iwarp_modify_qp_ramrod_data iwarp_modify_qp;
 	struct iwarp_init_func_ramrod_data iwarp_init_func;
 	struct fcoe_init_ramrod_params fcoe_init;
diff --git a/include/linux/qed/common_hsi.h b/include/linux/qed/common_hsi.h
index 885ae1379b5a..39e2a2ac2471 100644
--- a/include/linux/qed/common_hsi.h
+++ b/include/linux/qed/common_hsi.h
@@ -38,6 +38,8 @@
 #include <linux/slab.h>
 
 /* dma_addr_t manip */
+#define PTR_LO(x)               ((u32)(((uintptr_t)(x)) & 0xffffffff))
+#define PTR_HI(x)               ((u32)((((uintptr_t)(x)) >> 16) >> 16))
 #define DMA_LO_LE(x)		cpu_to_le32(lower_32_bits(x))
 #define DMA_HI_LE(x)		cpu_to_le32(upper_32_bits(x))
 #define DMA_REGPAIR_LE(x, val)	do { \
diff --git a/include/linux/qed/qed_rdma_if.h b/include/linux/qed/qed_rdma_if.h
index 28df5688ad0c..c4c241fe2579 100644
--- a/include/linux/qed/qed_rdma_if.h
+++ b/include/linux/qed/qed_rdma_if.h
@@ -471,7 +471,8 @@ struct qed_rdma_counters_out_params {
 #define QED_ROCE_TX_FRAG_FAILURE        (2)
 
 enum qed_iwarp_event_type {
-	QED_IWARP_EVENT_MPA_REQUEST,	/* Passive side request received */
+	QED_IWARP_EVENT_MPA_REQUEST,	  /* Passive side request received */
+	QED_IWARP_EVENT_PASSIVE_COMPLETE, /* ack on mpa response */
 };
 
 enum qed_tcp_ip_version {
@@ -516,6 +517,23 @@ struct qed_iwarp_listen_out {
 	void *handle;
 };
 
+struct qed_iwarp_accept_in {
+	void *ep_context;
+	void *cb_context;
+	struct qed_rdma_qp *qp;
+	const void *private_data;
+	u16 private_data_len;
+	u8 ord;
+	u8 ird;
+};
+
+struct qed_iwarp_reject_in {
+	void *ep_context;
+	void *cb_context;
+	const void *private_data;
+	u16 private_data_len;
+};
+
 struct qed_roce_ll2_header {
 	void *vaddr;
 	dma_addr_t baddr;
@@ -626,6 +644,12 @@ struct qed_rdma_ops {
 				   struct qed_iwarp_listen_in *iparams,
 				   struct qed_iwarp_listen_out *oparams);
 
+	int (*iwarp_accept)(void *rdma_cxt,
+			    struct qed_iwarp_accept_in *iparams);
+
+	int (*iwarp_reject)(void *rdma_cxt,
+			    struct qed_iwarp_reject_in *iparams);
+
 	int (*iwarp_destroy_listen)(void *rdma_cxt, void *handle);
 
 };