Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/roland/infiniband: (62 commits)
  mlx4_core: Deprecate log_num_vlan module param
  IB/mlx4: Don't set VLAN in IBoE WQEs' control segment
  IB/mlx4: Enable 4K mtu for IBoE
  RDMA/cxgb4: Mark QP in error before disabling the queue in firmware
  RDMA/cxgb4: Serialize calls to CQ's comp_handler
  RDMA/cxgb3: Serialize calls to CQ's comp_handler
  IB/qib: Fix issue with link states and QSFP cables
  IB/mlx4: Configure extended active speeds
  mlx4_core: Add extended port capabilities support
  IB/qib: Hold links until tuning data is available
  IB/qib: Clean up checkpatch issue
  IB/qib: Remove s_lock around header validation
  IB/qib: Precompute timeout jiffies to optimize latency
  IB/qib: Use RCU for qpn lookup
  IB/qib: Eliminate divide/mod in converting idx to egr buf pointer
  IB/qib: Decode path MTU optimization
  IB/qib: Optimize RC/UC code by IB operation
  IPoIB: Use the right function to do DMA unmap pages
  RDMA/cxgb4: Use correct QID in insert_recv_cqe()
  RDMA/cxgb4: Make sure flush CQ entries are collected on connection close
  ...

drivers/infiniband/core/cm.c

@@ -889,6 +889,8 @@ retest:
 		break;
 	case IB_CM_ESTABLISHED:
 		spin_unlock_irq(&cm_id_priv->lock);
+		if (cm_id_priv->qp_type == IB_QPT_XRC_TGT)
+			break;
 		ib_send_cm_dreq(cm_id, NULL, 0);
 		goto retest;
 	case IB_CM_DREQ_SENT:
@@ -1008,7 +1010,6 @@ static void cm_format_req(struct cm_req_msg *req_msg,
 	req_msg->service_id = param->service_id;
 	req_msg->local_ca_guid = cm_id_priv->id.device->node_guid;
 	cm_req_set_local_qpn(req_msg, cpu_to_be32(param->qp_num));
-	cm_req_set_resp_res(req_msg, param->responder_resources);
 	cm_req_set_init_depth(req_msg, param->initiator_depth);
 	cm_req_set_remote_resp_timeout(req_msg,
 				       param->remote_cm_response_timeout);
@@ -1017,12 +1018,16 @@ static void cm_format_req(struct cm_req_msg *req_msg,
 	cm_req_set_starting_psn(req_msg, cpu_to_be32(param->starting_psn));
 	cm_req_set_local_resp_timeout(req_msg,
 				      param->local_cm_response_timeout);
-	cm_req_set_retry_count(req_msg, param->retry_count);
 	req_msg->pkey = param->primary_path->pkey;
 	cm_req_set_path_mtu(req_msg, param->primary_path->mtu);
-	cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count);
 	cm_req_set_max_cm_retries(req_msg, param->max_cm_retries);
-	cm_req_set_srq(req_msg, param->srq);
+
+	if (param->qp_type != IB_QPT_XRC_INI) {
+		cm_req_set_resp_res(req_msg, param->responder_resources);
+		cm_req_set_retry_count(req_msg, param->retry_count);
+		cm_req_set_rnr_retry_count(req_msg, param->rnr_retry_count);
+		cm_req_set_srq(req_msg, param->srq);
+	}
 
 	if (pri_path->hop_limit <= 1) {
 		req_msg->primary_local_lid = pri_path->slid;
@@ -1080,7 +1085,8 @@ static int cm_validate_req_param(struct ib_cm_req_param *param)
 	if (!param->primary_path)
 		return -EINVAL;
 
-	if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC)
+	if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC &&
+	    param->qp_type != IB_QPT_XRC_INI)
 		return -EINVAL;
 
 	if (param->private_data &&
@@ -1601,18 +1607,24 @@ static void cm_format_rep(struct cm_rep_msg *rep_msg,
 	cm_format_mad_hdr(&rep_msg->hdr, CM_REP_ATTR_ID, cm_id_priv->tid);
 	rep_msg->local_comm_id = cm_id_priv->id.local_id;
 	rep_msg->remote_comm_id = cm_id_priv->id.remote_id;
-	cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num));
 	cm_rep_set_starting_psn(rep_msg, cpu_to_be32(param->starting_psn));
 	rep_msg->resp_resources = param->responder_resources;
-	rep_msg->initiator_depth = param->initiator_depth;
 	cm_rep_set_target_ack_delay(rep_msg,
 				    cm_id_priv->av.port->cm_dev->ack_delay);
 	cm_rep_set_failover(rep_msg, param->failover_accepted);
-	cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
 	cm_rep_set_rnr_retry_count(rep_msg, param->rnr_retry_count);
-	cm_rep_set_srq(rep_msg, param->srq);
 	rep_msg->local_ca_guid = cm_id_priv->id.device->node_guid;
 
+	if (cm_id_priv->qp_type != IB_QPT_XRC_TGT) {
+		rep_msg->initiator_depth = param->initiator_depth;
+		cm_rep_set_flow_ctrl(rep_msg, param->flow_control);
+		cm_rep_set_srq(rep_msg, param->srq);
+		cm_rep_set_local_qpn(rep_msg, cpu_to_be32(param->qp_num));
+	} else {
+		cm_rep_set_srq(rep_msg, 1);
+		cm_rep_set_local_eecn(rep_msg, cpu_to_be32(param->qp_num));
+	}
+
 	if (param->private_data && param->private_data_len)
 		memcpy(rep_msg->private_data, param->private_data,
 		       param->private_data_len);
@@ -1660,7 +1672,7 @@ int ib_send_cm_rep(struct ib_cm_id *cm_id,
 	cm_id_priv->initiator_depth = param->initiator_depth;
 	cm_id_priv->responder_resources = param->responder_resources;
 	cm_id_priv->rq_psn = cm_rep_get_starting_psn(rep_msg);
-	cm_id_priv->local_qpn = cm_rep_get_local_qpn(rep_msg);
+	cm_id_priv->local_qpn = cpu_to_be32(param->qp_num & 0xFFFFFF);
 
 out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
 	return ret;
@@ -1731,7 +1743,7 @@ error:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
 }
 EXPORT_SYMBOL(ib_send_cm_rtu);
 
-static void cm_format_rep_event(struct cm_work *work)
+static void cm_format_rep_event(struct cm_work *work, enum ib_qp_type qp_type)
 {
 	struct cm_rep_msg *rep_msg;
 	struct ib_cm_rep_event_param *param;
@@ -1740,7 +1752,7 @@ static void cm_format_rep_event(struct cm_work *work)
 	param = &work->cm_event.param.rep_rcvd;
 	param->remote_ca_guid = rep_msg->local_ca_guid;
 	param->remote_qkey = be32_to_cpu(rep_msg->local_qkey);
-	param->remote_qpn = be32_to_cpu(cm_rep_get_local_qpn(rep_msg));
+	param->remote_qpn = be32_to_cpu(cm_rep_get_qpn(rep_msg, qp_type));
 	param->starting_psn = be32_to_cpu(cm_rep_get_starting_psn(rep_msg));
 	param->responder_resources = rep_msg->initiator_depth;
 	param->initiator_depth = rep_msg->resp_resources;
@@ -1808,7 +1820,7 @@ static int cm_rep_handler(struct cm_work *work)
 		return -EINVAL;
 	}
 
-	cm_format_rep_event(work);
+	cm_format_rep_event(work, cm_id_priv->qp_type);
 
 	spin_lock_irq(&cm_id_priv->lock);
 	switch (cm_id_priv->id.state) {
@@ -1823,7 +1835,7 @@ static int cm_rep_handler(struct cm_work *work)
 
 	cm_id_priv->timewait_info->work.remote_id = rep_msg->local_comm_id;
 	cm_id_priv->timewait_info->remote_ca_guid = rep_msg->local_ca_guid;
-	cm_id_priv->timewait_info->remote_qpn = cm_rep_get_local_qpn(rep_msg);
+	cm_id_priv->timewait_info->remote_qpn = cm_rep_get_qpn(rep_msg, cm_id_priv->qp_type);
 
 	spin_lock(&cm.lock);
 	/* Check for duplicate REP. */
@@ -1850,7 +1862,7 @@ static int cm_rep_handler(struct cm_work *work)
 
 	cm_id_priv->id.state = IB_CM_REP_RCVD;
 	cm_id_priv->id.remote_id = rep_msg->local_comm_id;
-	cm_id_priv->remote_qpn = cm_rep_get_local_qpn(rep_msg);
+	cm_id_priv->remote_qpn = cm_rep_get_qpn(rep_msg, cm_id_priv->qp_type);
 	cm_id_priv->initiator_depth = rep_msg->resp_resources;
 	cm_id_priv->responder_resources = rep_msg->initiator_depth;
 	cm_id_priv->sq_psn = cm_rep_get_starting_psn(rep_msg);
@@ -3492,7 +3504,8 @@ static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv,
 		qp_attr->path_mtu = cm_id_priv->path_mtu;
 		qp_attr->dest_qp_num = be32_to_cpu(cm_id_priv->remote_qpn);
 		qp_attr->rq_psn = be32_to_cpu(cm_id_priv->rq_psn);
-		if (cm_id_priv->qp_type == IB_QPT_RC) {
+		if (cm_id_priv->qp_type == IB_QPT_RC ||
+		    cm_id_priv->qp_type == IB_QPT_XRC_TGT) {
 			*qp_attr_mask |= IB_QP_MAX_DEST_RD_ATOMIC |
 					 IB_QP_MIN_RNR_TIMER;
 			qp_attr->max_dest_rd_atomic =
@@ -3537,15 +3550,21 @@ static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
 		if (cm_id_priv->id.lap_state == IB_CM_LAP_UNINIT) {
 			*qp_attr_mask = IB_QP_STATE | IB_QP_SQ_PSN;
 			qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn);
-			if (cm_id_priv->qp_type == IB_QPT_RC) {
-				*qp_attr_mask |= IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
-						 IB_QP_RNR_RETRY |
+			switch (cm_id_priv->qp_type) {
+			case IB_QPT_RC:
+			case IB_QPT_XRC_INI:
+				*qp_attr_mask |= IB_QP_RETRY_CNT | IB_QP_RNR_RETRY |
 						 IB_QP_MAX_QP_RD_ATOMIC;
-				qp_attr->timeout = cm_id_priv->av.timeout;
 				qp_attr->retry_cnt = cm_id_priv->retry_count;
 				qp_attr->rnr_retry = cm_id_priv->rnr_retry_count;
-				qp_attr->max_rd_atomic =
-					cm_id_priv->initiator_depth;
+				qp_attr->max_rd_atomic = cm_id_priv->initiator_depth;
+				/* fall through */
+			case IB_QPT_XRC_TGT:
+				*qp_attr_mask |= IB_QP_TIMEOUT;
+				qp_attr->timeout = cm_id_priv->av.timeout;
+				break;
+			default:
+				break;
 			}
 			if (cm_id_priv->alt_av.ah_attr.dlid) {
 				*qp_attr_mask |= IB_QP_PATH_MIG_STATE;

drivers/infiniband/core/cm_msgs.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004, 2011 Intel Corporation.  All rights reserved.
  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
  *
@@ -86,7 +86,7 @@ struct cm_req_msg {
 	__be16 pkey;
 	/* path MTU:4, RDC exists:1, RNR retry count:3. */
 	u8 offset50;
-	/* max CM Retries:4, SRQ:1, rsvd:3 */
+	/* max CM Retries:4, SRQ:1, extended transport type:3 */
 	u8 offset51;
 
 	__be16 primary_local_lid;
@@ -175,6 +175,11 @@ static inline enum ib_qp_type cm_req_get_qp_type(struct cm_req_msg *req_msg)
 	switch(transport_type) {
 	case 0: return IB_QPT_RC;
 	case 1: return IB_QPT_UC;
+	case 3:
+		switch (req_msg->offset51 & 0x7) {
+		case 1: return IB_QPT_XRC_TGT;
+		default: return 0;
+		}
 	default: return 0;
 	}
 }
@@ -188,6 +193,12 @@ static inline void cm_req_set_qp_type(struct cm_req_msg *req_msg,
 						  req_msg->offset40) &
 						   0xFFFFFFF9) | 0x2);
 		break;
+	case IB_QPT_XRC_INI:
+		req_msg->offset40 = cpu_to_be32((be32_to_cpu(
+						 req_msg->offset40) &
+						   0xFFFFFFF9) | 0x6);
+		req_msg->offset51 = (req_msg->offset51 & 0xF8) | 1;
+		break;
 	default:
 		req_msg->offset40 = cpu_to_be32(be32_to_cpu(
 						 req_msg->offset40) &
@@ -527,6 +538,23 @@ static inline void cm_rep_set_local_qpn(struct cm_rep_msg *rep_msg, __be32 qpn)
 			    (be32_to_cpu(rep_msg->offset12) & 0x000000FF));
 }
 
+static inline __be32 cm_rep_get_local_eecn(struct cm_rep_msg *rep_msg)
+{
+	return cpu_to_be32(be32_to_cpu(rep_msg->offset16) >> 8);
+}
+
+static inline void cm_rep_set_local_eecn(struct cm_rep_msg *rep_msg, __be32 eecn)
+{
+	rep_msg->offset16 = cpu_to_be32((be32_to_cpu(eecn) << 8) |
+			    (be32_to_cpu(rep_msg->offset16) & 0x000000FF));
+}
+
+static inline __be32 cm_rep_get_qpn(struct cm_rep_msg *rep_msg, enum ib_qp_type qp_type)
+{
+	return (qp_type == IB_QPT_XRC_INI) ?
+		cm_rep_get_local_eecn(rep_msg) : cm_rep_get_local_qpn(rep_msg);
+}
+
 static inline __be32 cm_rep_get_starting_psn(struct cm_rep_msg *rep_msg)
 {
 	return cpu_to_be32(be32_to_cpu(rep_msg->offset20) >> 8);

drivers/infiniband/core/cma.c

@@ -81,6 +81,7 @@ static DEFINE_IDR(sdp_ps);
 static DEFINE_IDR(tcp_ps);
 static DEFINE_IDR(udp_ps);
 static DEFINE_IDR(ipoib_ps);
+static DEFINE_IDR(ib_ps);
 
 struct cma_device {
 	struct list_head	list;
@@ -1179,6 +1180,15 @@ static void cma_set_req_event_data(struct rdma_cm_event *event,
 	event->param.conn.qp_num = req_data->remote_qpn;
 }
 
+static int cma_check_req_qp_type(struct rdma_cm_id *id, struct ib_cm_event *ib_event)
+{
+	return (((ib_event->event == IB_CM_REQ_RECEIVED) ||
+		 (ib_event->param.req_rcvd.qp_type == id->qp_type)) ||
+		((ib_event->event == IB_CM_SIDR_REQ_RECEIVED) &&
+		 (id->qp_type == IB_QPT_UD)) ||
+		(!id->qp_type));
+}
+
 static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
 {
 	struct rdma_id_private *listen_id, *conn_id;
@@ -1186,13 +1196,16 @@ static int cma_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *ib_event)
 	int offset, ret;
 
 	listen_id = cm_id->context;
+	if (!cma_check_req_qp_type(&listen_id->id, ib_event))
+		return -EINVAL;
+
 	if (cma_disable_callback(listen_id, RDMA_CM_LISTEN))
 		return -ECONNABORTED;
 
 	memset(&event, 0, sizeof event);
 	offset = cma_user_data_offset(listen_id->id.ps);
 	event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
-	if (listen_id->id.qp_type == IB_QPT_UD) {
+	if (ib_event->event == IB_CM_SIDR_REQ_RECEIVED) {
 		conn_id = cma_new_udp_id(&listen_id->id, ib_event);
 		event.param.ud.private_data = ib_event->private_data + offset;
 		event.param.ud.private_data_len =
@@ -1328,6 +1341,8 @@ static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
 		switch (iw_event->status) {
 		case 0:
 			event.event = RDMA_CM_EVENT_ESTABLISHED;
+			event.param.conn.initiator_depth = iw_event->ird;
+			event.param.conn.responder_resources = iw_event->ord;
 			break;
 		case -ECONNRESET:
 		case -ECONNREFUSED:
@@ -1343,6 +1358,8 @@ static int cma_iw_handler(struct iw_cm_id *iw_id, struct iw_cm_event *iw_event)
 		break;
 	case IW_CM_EVENT_ESTABLISHED:
 		event.event = RDMA_CM_EVENT_ESTABLISHED;
+		event.param.conn.initiator_depth = iw_event->ird;
+		event.param.conn.responder_resources = iw_event->ord;
 		break;
 	default:
 		BUG_ON(1);
@@ -1433,8 +1450,8 @@ static int iw_conn_req_handler(struct iw_cm_id *cm_id,
 	event.event = RDMA_CM_EVENT_CONNECT_REQUEST;
 	event.param.conn.private_data = iw_event->private_data;
 	event.param.conn.private_data_len = iw_event->private_data_len;
-	event.param.conn.initiator_depth = attr.max_qp_init_rd_atom;
-	event.param.conn.responder_resources = attr.max_qp_rd_atom;
+	event.param.conn.initiator_depth = iw_event->ird;
+	event.param.conn.responder_resources = iw_event->ord;
 
 	/*
 	 * Protect against the user destroying conn_id from another thread
@@ -2234,6 +2251,9 @@ static int cma_get_port(struct rdma_id_private *id_priv)
 	case RDMA_PS_IPOIB:
 		ps = &ipoib_ps;
 		break;
+	case RDMA_PS_IB:
+		ps = &ib_ps;
+		break;
 	default:
 		return -EPROTONOSUPPORT;
 	}
@@ -2569,7 +2589,7 @@ static int cma_connect_ib(struct rdma_id_private *id_priv,
 	req.service_id = cma_get_service_id(id_priv->id.ps,
 					    (struct sockaddr *) &route->addr.dst_addr);
 	req.qp_num = id_priv->qp_num;
-	req.qp_type = IB_QPT_RC;
+	req.qp_type = id_priv->id.qp_type;
 	req.starting_psn = id_priv->seq_num;
 	req.responder_resources = conn_param->responder_resources;
 	req.initiator_depth = conn_param->initiator_depth;
@@ -2616,14 +2636,16 @@ static int cma_connect_iw(struct rdma_id_private *id_priv,
 	if (ret)
 		goto out;
 
-	iw_param.ord = conn_param->initiator_depth;
-	iw_param.ird = conn_param->responder_resources;
-	iw_param.private_data = conn_param->private_data;
-	iw_param.private_data_len = conn_param->private_data_len;
-	if (id_priv->id.qp)
+	if (conn_param) {
+		iw_param.ord = conn_param->initiator_depth;
+		iw_param.ird = conn_param->responder_resources;
+		iw_param.private_data = conn_param->private_data;
+		iw_param.private_data_len = conn_param->private_data_len;
+		iw_param.qpn = id_priv->id.qp ? id_priv->qp_num : conn_param->qp_num;
+	} else {
+		memset(&iw_param, 0, sizeof iw_param);
 		iw_param.qpn = id_priv->qp_num;
-	else
-		iw_param.qpn = conn_param->qp_num;
+	}
 	ret = iw_cm_connect(cm_id, &iw_param);
 out:
 	if (ret) {
@@ -2765,14 +2787,20 @@ int rdma_accept(struct rdma_cm_id *id, struct rdma_conn_param *conn_param)
 
 	switch (rdma_node_get_transport(id->device->node_type)) {
 	case RDMA_TRANSPORT_IB:
-		if (id->qp_type == IB_QPT_UD)
-			ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
-						conn_param->private_data,
-						conn_param->private_data_len);
-		else if (conn_param)
-			ret = cma_accept_ib(id_priv, conn_param);
-		else
-			ret = cma_rep_recv(id_priv);
+		if (id->qp_type == IB_QPT_UD) {
+			if (conn_param)
+				ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
+							conn_param->private_data,
+							conn_param->private_data_len);
+			else
+				ret = cma_send_sidr_rep(id_priv, IB_SIDR_SUCCESS,
+							NULL, 0);
+		} else {
+			if (conn_param)
+				ret = cma_accept_ib(id_priv, conn_param);
+			else
+				ret = cma_rep_recv(id_priv);
+		}
 		break;
 	case RDMA_TRANSPORT_IWARP:
 		ret = cma_accept_iw(id_priv, conn_param);
@@ -3460,6 +3488,7 @@ static void __exit cma_cleanup(void)
 	idr_destroy(&tcp_ps);
 	idr_destroy(&udp_ps);
 	idr_destroy(&ipoib_ps);
+	idr_destroy(&ib_ps);
 }
 
 module_init(cma_init);

drivers/infiniband/core/mad.c

@@ -1596,6 +1596,9 @@ find_mad_agent(struct ib_mad_port_private *port_priv,
 					mad->mad_hdr.class_version].class;
 			if (!class)
 				goto out;
+			if (convert_mgmt_class(mad->mad_hdr.mgmt_class) >=
+			    IB_MGMT_MAX_METHODS)
+				goto out;
 			method = class->method_table[convert_mgmt_class(
 							mad->mad_hdr.mgmt_class)];
 			if (method)

drivers/infiniband/core/sysfs.c

@@ -185,17 +185,35 @@ static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
 	if (ret)
 		return ret;
 
+	rate = (25 * attr.active_speed) / 10;
+
 	switch (attr.active_speed) {
-	case 2: speed = " DDR"; break;
-	case 4: speed = " QDR"; break;
+	case 2:
+		speed = " DDR";
+		break;
+	case 4:
+		speed = " QDR";
+		break;
+	case 8:
+		speed = " FDR10";
+		rate = 10;
+		break;
+	case 16:
+		speed = " FDR";
+		rate = 14;
+		break;
+	case 32:
+		speed = " EDR";
+		rate = 25;
+		break;
 	}
 
-	rate = 25 * ib_width_enum_to_int(attr.active_width) * attr.active_speed;
+	rate *= ib_width_enum_to_int(attr.active_width);
 	if (rate < 0)
 		return -EINVAL;
 
 	return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
-		       rate / 10, rate % 10 ? ".5" : "",
+		       rate, (attr.active_speed == 1) ? ".5" : "",
 		       ib_width_enum_to_int(attr.active_width), speed);
 }
 

drivers/infiniband/core/ucm.c

@@ -1122,7 +1122,7 @@ static ssize_t ib_ucm_write(struct file *filp, const char __user *buf,
 	if (copy_from_user(&hdr, buf, sizeof(hdr)))
 		return -EFAULT;
 
-	if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucm_cmd_table))
+	if (hdr.cmd >= ARRAY_SIZE(ucm_cmd_table))
 		return -EINVAL;
 
 	if (hdr.in + sizeof(hdr) > len)

drivers/infiniband/core/ucma.c

@@ -276,7 +276,7 @@ static int ucma_event_handler(struct rdma_cm_id *cm_id,
 	ucma_set_event_context(ctx, event, uevent);
 	uevent->resp.event = event->event;
 	uevent->resp.status = event->status;
-	if (cm_id->ps == RDMA_PS_UDP || cm_id->ps == RDMA_PS_IPOIB)
+	if (cm_id->qp_type == IB_QPT_UD)
 		ucma_copy_ud_event(&uevent->resp.param.ud, &event->param.ud);
 	else
 		ucma_copy_conn_event(&uevent->resp.param.conn,
@@ -377,6 +377,9 @@ static int ucma_get_qp_type(struct rdma_ucm_create_id *cmd, enum ib_qp_type *qp_
 	case RDMA_PS_IPOIB:
 		*qp_type = IB_QPT_UD;
 		return 0;
+	case RDMA_PS_IB:
+		*qp_type = cmd->qp_type;
+		return 0;
 	default:
 		return -EINVAL;
 	}
@@ -1270,7 +1273,7 @@ static ssize_t ucma_write(struct file *filp, const char __user *buf,
 	if (copy_from_user(&hdr, buf, sizeof(hdr)))
 		return -EFAULT;
 
-	if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
+	if (hdr.cmd >= ARRAY_SIZE(ucma_cmd_table))
 		return -EINVAL;
 
 	if (hdr.in + sizeof(hdr) > len)

drivers/infiniband/core/user_mad.c

@@ -458,8 +458,7 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
 		goto err;
 	}
 
-	if (packet->mad.hdr.id < 0 ||
-	    packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
+	if (packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
 		ret = -EINVAL;
 		goto err;
 	}
@@ -703,7 +702,7 @@ static int ib_umad_unreg_agent(struct ib_umad_file *file, u32 __user *arg)
 	mutex_lock(&file->port->file_mutex);
 	mutex_lock(&file->mutex);
 
-	if (id < 0 || id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
+	if (id >= IB_UMAD_MAX_AGENTS || !__get_agent(file, id)) {
 		ret = -EINVAL;
 		goto out;
 	}

drivers/infiniband/core/uverbs.h

@@ -76,6 +76,8 @@ struct ib_uverbs_device {
 	struct ib_device		       *ib_dev;
 	int					devnum;
 	struct cdev			        cdev;
+	struct rb_root				xrcd_tree;
+	struct mutex				xrcd_tree_mutex;
 };
 
 struct ib_uverbs_event_file {
@@ -120,6 +122,16 @@ struct ib_uevent_object {
 	u32			events_reported;
 };
 
+struct ib_uxrcd_object {
+	struct ib_uobject	uobject;
+	atomic_t		refcnt;
+};
+
+struct ib_usrq_object {
+	struct ib_uevent_object	uevent;
+	struct ib_uxrcd_object *uxrcd;
+};
+
 struct ib_uqp_object {
 	struct ib_uevent_object	uevent;
 	struct list_head 	mcast_list;
@@ -142,6 +154,7 @@ extern struct idr ib_uverbs_ah_idr;
 extern struct idr ib_uverbs_cq_idr;
 extern struct idr ib_uverbs_qp_idr;
 extern struct idr ib_uverbs_srq_idr;
+extern struct idr ib_uverbs_xrcd_idr;
 
 void idr_remove_uobj(struct idr *idp, struct ib_uobject *uobj);
 
@@ -161,6 +174,7 @@ void ib_uverbs_qp_event_handler(struct ib_event *event, void *context_ptr);
 void ib_uverbs_srq_event_handler(struct ib_event *event, void *context_ptr);
 void ib_uverbs_event_handler(struct ib_event_handler *handler,
 			     struct ib_event *event);
+void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev, struct ib_xrcd *xrcd);
 
 #define IB_UVERBS_DECLARE_CMD(name)					\
 	ssize_t ib_uverbs_##name(struct ib_uverbs_file *file,		\
@@ -181,6 +195,7 @@ IB_UVERBS_DECLARE_CMD(poll_cq);
 IB_UVERBS_DECLARE_CMD(req_notify_cq);
 IB_UVERBS_DECLARE_CMD(destroy_cq);
 IB_UVERBS_DECLARE_CMD(create_qp);
+IB_UVERBS_DECLARE_CMD(open_qp);
 IB_UVERBS_DECLARE_CMD(query_qp);
 IB_UVERBS_DECLARE_CMD(modify_qp);
 IB_UVERBS_DECLARE_CMD(destroy_qp);
@@ -195,5 +210,8 @@ IB_UVERBS_DECLARE_CMD(create_srq);
 IB_UVERBS_DECLARE_CMD(modify_srq);
 IB_UVERBS_DECLARE_CMD(query_srq);
 IB_UVERBS_DECLARE_CMD(destroy_srq);
+IB_UVERBS_DECLARE_CMD(create_xsrq);
+IB_UVERBS_DECLARE_CMD(open_xrcd);
+IB_UVERBS_DECLARE_CMD(close_xrcd);
 
 #endif /* UVERBS_H */

drivers/infiniband/core/uverbs_cmd.c

@@ -47,6 +47,7 @@ static struct lock_class_key cq_lock_key;
 static struct lock_class_key qp_lock_key;
 static struct lock_class_key ah_lock_key;
 static struct lock_class_key srq_lock_key;
+static struct lock_class_key xrcd_lock_key;
 
 #define INIT_UDATA(udata, ibuf, obuf, ilen, olen)			\
 	do {								\
@@ -255,6 +256,18 @@ static void put_srq_read(struct ib_srq *srq)
 	put_uobj_read(srq->uobject);
 }
 
+static struct ib_xrcd *idr_read_xrcd(int xrcd_handle, struct ib_ucontext *context,
+				     struct ib_uobject **uobj)
+{
+	*uobj = idr_read_uobj(&ib_uverbs_xrcd_idr, xrcd_handle, context, 0);
+	return *uobj ? (*uobj)->object : NULL;
+}
+
+static void put_xrcd_read(struct ib_uobject *uobj)
+{
+	put_uobj_read(uobj);
+}
+
 ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
 			      const char __user *buf,
 			      int in_len, int out_len)
@@ -298,6 +311,7 @@ ssize_t ib_uverbs_get_context(struct ib_uverbs_file *file,
 	INIT_LIST_HEAD(&ucontext->qp_list);
 	INIT_LIST_HEAD(&ucontext->srq_list);
 	INIT_LIST_HEAD(&ucontext->ah_list);
+	INIT_LIST_HEAD(&ucontext->xrcd_list);
 	ucontext->closing = 0;
 
 	resp.num_comp_vectors = file->device->num_comp_vectors;
@@ -579,6 +593,310 @@ ssize_t ib_uverbs_dealloc_pd(struct ib_uverbs_file *file,
 	return in_len;
 }
 
+struct xrcd_table_entry {
+	struct rb_node  node;
+	struct ib_xrcd *xrcd;
+	struct inode   *inode;
+};
+
+static int xrcd_table_insert(struct ib_uverbs_device *dev,
+			    struct inode *inode,
+			    struct ib_xrcd *xrcd)
+{
+	struct xrcd_table_entry *entry, *scan;
+	struct rb_node **p = &dev->xrcd_tree.rb_node;
+	struct rb_node *parent = NULL;
+
+	entry = kmalloc(sizeof *entry, GFP_KERNEL);
+	if (!entry)
+		return -ENOMEM;
+
+	entry->xrcd  = xrcd;
+	entry->inode = inode;
+
+	while (*p) {
+		parent = *p;
+		scan = rb_entry(parent, struct xrcd_table_entry, node);
+
+		if (inode < scan->inode) {
+			p = &(*p)->rb_left;
+		} else if (inode > scan->inode) {
+			p = &(*p)->rb_right;
+		} else {
+			kfree(entry);
+			return -EEXIST;
+		}
+	}
+
+	rb_link_node(&entry->node, parent, p);
+	rb_insert_color(&entry->node, &dev->xrcd_tree);
+	igrab(inode);
+	return 0;
+}
+
+static struct xrcd_table_entry *xrcd_table_search(struct ib_uverbs_device *dev,
+						  struct inode *inode)
+{
+	struct xrcd_table_entry *entry;
+	struct rb_node *p = dev->xrcd_tree.rb_node;
+
+	while (p) {
+		entry = rb_entry(p, struct xrcd_table_entry, node);
+
+		if (inode < entry->inode)
+			p = p->rb_left;
+		else if (inode > entry->inode)
+			p = p->rb_right;
+		else
+			return entry;
+	}
+
+	return NULL;
+}
+
+static struct ib_xrcd *find_xrcd(struct ib_uverbs_device *dev, struct inode *inode)
+{
+	struct xrcd_table_entry *entry;
+
+	entry = xrcd_table_search(dev, inode);
+	if (!entry)
+		return NULL;
+
+	return entry->xrcd;
+}
+
+static void xrcd_table_delete(struct ib_uverbs_device *dev,
+			      struct inode *inode)
+{
+	struct xrcd_table_entry *entry;
+
+	entry = xrcd_table_search(dev, inode);
+	if (entry) {
+		iput(inode);
+		rb_erase(&entry->node, &dev->xrcd_tree);
+		kfree(entry);
+	}
+}
+
+ssize_t ib_uverbs_open_xrcd(struct ib_uverbs_file *file,
+			    const char __user *buf, int in_len,
+			    int out_len)
+{
+	struct ib_uverbs_open_xrcd	cmd;
+	struct ib_uverbs_open_xrcd_resp	resp;
+	struct ib_udata			udata;
+	struct ib_uxrcd_object         *obj;
+	struct ib_xrcd                 *xrcd = NULL;
+	struct file                    *f = NULL;
+	struct inode                   *inode = NULL;
+	int				ret = 0;
+	int				new_xrcd = 0;
+
+	if (out_len < sizeof resp)
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	INIT_UDATA(&udata, buf + sizeof cmd,
+		   (unsigned long) cmd.response + sizeof resp,
+		   in_len - sizeof cmd, out_len - sizeof  resp);
+
+	mutex_lock(&file->device->xrcd_tree_mutex);
+
+	if (cmd.fd != -1) {
+		/* search for file descriptor */
+		f = fget(cmd.fd);
+		if (!f) {
+			ret = -EBADF;
+			goto err_tree_mutex_unlock;
+		}
+
+		inode = f->f_dentry->d_inode;
+		if (!inode) {
+			ret = -EBADF;
+			goto err_tree_mutex_unlock;
+		}
+
+		xrcd = find_xrcd(file->device, inode);
+		if (!xrcd && !(cmd.oflags & O_CREAT)) {
+			/* no file descriptor. Need CREATE flag */
+			ret = -EAGAIN;
+			goto err_tree_mutex_unlock;
+		}
+
+		if (xrcd && cmd.oflags & O_EXCL) {
+			ret = -EINVAL;
+			goto err_tree_mutex_unlock;
+		}
+	}
+
+	obj = kmalloc(sizeof *obj, GFP_KERNEL);
+	if (!obj) {
+		ret = -ENOMEM;
+		goto err_tree_mutex_unlock;
+	}
+
+	init_uobj(&obj->uobject, 0, file->ucontext, &xrcd_lock_key);
+
+	down_write(&obj->uobject.mutex);
+
+	if (!xrcd) {
+		xrcd = file->device->ib_dev->alloc_xrcd(file->device->ib_dev,
+							file->ucontext, &udata);
+		if (IS_ERR(xrcd)) {
+			ret = PTR_ERR(xrcd);
+			goto err;
+		}
+
+		xrcd->inode   = inode;
+		xrcd->device  = file->device->ib_dev;
+		atomic_set(&xrcd->usecnt, 0);
+		mutex_init(&xrcd->tgt_qp_mutex);
+		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
+		new_xrcd = 1;
+	}
+
+	atomic_set(&obj->refcnt, 0);
+	obj->uobject.object = xrcd;
+	ret = idr_add_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
+	if (ret)
+		goto err_idr;
+
+	memset(&resp, 0, sizeof resp);
+	resp.xrcd_handle = obj->uobject.id;
+
+	if (inode) {
+		if (new_xrcd) {
+			/* create new inode/xrcd table entry */
+			ret = xrcd_table_insert(file->device, inode, xrcd);
+			if (ret)
+				goto err_insert_xrcd;
+		}
+		atomic_inc(&xrcd->usecnt);
+	}
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp)) {
+		ret = -EFAULT;
+		goto err_copy;
+	}
+
+	if (f)
+		fput(f);
+
+	mutex_lock(&file->mutex);
+	list_add_tail(&obj->uobject.list, &file->ucontext->xrcd_list);
+	mutex_unlock(&file->mutex);
+
+	obj->uobject.live = 1;
+	up_write(&obj->uobject.mutex);
+
+	mutex_unlock(&file->device->xrcd_tree_mutex);
+	return in_len;
+
+err_copy:
+	if (inode) {
+		if (new_xrcd)
+			xrcd_table_delete(file->device, inode);
+		atomic_dec(&xrcd->usecnt);
+	}
+
+err_insert_xrcd:
+	idr_remove_uobj(&ib_uverbs_xrcd_idr, &obj->uobject);
+
+err_idr:
+	ib_dealloc_xrcd(xrcd);
+
+err:
+	put_uobj_write(&obj->uobject);
+
+err_tree_mutex_unlock:
+	if (f)
+		fput(f);
+
+	mutex_unlock(&file->device->xrcd_tree_mutex);
+
+	return ret;
+}
+
+ssize_t ib_uverbs_close_xrcd(struct ib_uverbs_file *file,
+			     const char __user *buf, int in_len,
+			     int out_len)
+{
+	struct ib_uverbs_close_xrcd cmd;
+	struct ib_uobject           *uobj;
+	struct ib_xrcd              *xrcd = NULL;
+	struct inode                *inode = NULL;
+	struct ib_uxrcd_object      *obj;
+	int                         live;
+	int                         ret = 0;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	mutex_lock(&file->device->xrcd_tree_mutex);
+	uobj = idr_write_uobj(&ib_uverbs_xrcd_idr, cmd.xrcd_handle, file->ucontext);
+	if (!uobj) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	xrcd  = uobj->object;
+	inode = xrcd->inode;
+	obj   = container_of(uobj, struct ib_uxrcd_object, uobject);
+	if (atomic_read(&obj->refcnt)) {
+		put_uobj_write(uobj);
+		ret = -EBUSY;
+		goto out;
+	}
+
+	if (!inode || atomic_dec_and_test(&xrcd->usecnt)) {
+		ret = ib_dealloc_xrcd(uobj->object);
+		if (!ret)
+			uobj->live = 0;
+	}
+
+	live = uobj->live;
+	if (inode && ret)
+		atomic_inc(&xrcd->usecnt);
+
+	put_uobj_write(uobj);
+
+	if (ret)
+		goto out;
+
+	if (inode && !live)
+		xrcd_table_delete(file->device, inode);
+
+	idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
+	mutex_lock(&file->mutex);
+	list_del(&uobj->list);
+	mutex_unlock(&file->mutex);
+
+	put_uobj(uobj);
+	ret = in_len;
+
+out:
+	mutex_unlock(&file->device->xrcd_tree_mutex);
+	return ret;
+}
+
+void ib_uverbs_dealloc_xrcd(struct ib_uverbs_device *dev,
+			    struct ib_xrcd *xrcd)
+{
+	struct inode *inode;
+
+	inode = xrcd->inode;
+	if (inode && !atomic_dec_and_test(&xrcd->usecnt))
+		return;
+
+	ib_dealloc_xrcd(xrcd);
+
+	if (inode)
+		xrcd_table_delete(dev, inode);
+}
+
 ssize_t ib_uverbs_reg_mr(struct ib_uverbs_file *file,
 			 const char __user *buf, int in_len,
 			 int out_len)
@@ -1052,9 +1370,12 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
 	struct ib_uverbs_create_qp_resp resp;
 	struct ib_udata                 udata;
 	struct ib_uqp_object           *obj;
-	struct ib_pd                   *pd;
-	struct ib_cq                   *scq, *rcq;
-	struct ib_srq                  *srq;
+	struct ib_device	       *device;
+	struct ib_pd                   *pd = NULL;
+	struct ib_xrcd		       *xrcd = NULL;
+	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
+	struct ib_cq                   *scq = NULL, *rcq = NULL;
+	struct ib_srq                  *srq = NULL;
 	struct ib_qp                   *qp;
 	struct ib_qp_init_attr          attr;
 	int ret;
@@ -1076,15 +1397,39 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
 	init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key);
 	down_write(&obj->uevent.uobject.mutex);
 
-	srq = cmd.is_srq ? idr_read_srq(cmd.srq_handle, file->ucontext) : NULL;
-	pd  = idr_read_pd(cmd.pd_handle, file->ucontext);
-	scq = idr_read_cq(cmd.send_cq_handle, file->ucontext, 0);
-	rcq = cmd.recv_cq_handle == cmd.send_cq_handle ?
-		scq : idr_read_cq(cmd.recv_cq_handle, file->ucontext, 1);
+	if (cmd.qp_type == IB_QPT_XRC_TGT) {
+		xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
+		if (!xrcd) {
+			ret = -EINVAL;
+			goto err_put;
+		}
+		device = xrcd->device;
+	} else {
+		pd  = idr_read_pd(cmd.pd_handle, file->ucontext);
+		scq = idr_read_cq(cmd.send_cq_handle, file->ucontext, 0);
+		if (!pd || !scq) {
+			ret = -EINVAL;
+			goto err_put;
+		}
 
-	if (!pd || !scq || !rcq || (cmd.is_srq && !srq)) {
-		ret = -EINVAL;
-		goto err_put;
+		if (cmd.qp_type == IB_QPT_XRC_INI) {
+			cmd.max_recv_wr = cmd.max_recv_sge = 0;
+		} else {
+			if (cmd.is_srq) {
+				srq = idr_read_srq(cmd.srq_handle, file->ucontext);
+				if (!srq || srq->srq_type != IB_SRQT_BASIC) {
+					ret = -EINVAL;
+					goto err_put;
+				}
+			}
+			rcq = (cmd.recv_cq_handle == cmd.send_cq_handle) ?
+			       scq : idr_read_cq(cmd.recv_cq_handle, file->ucontext, 1);
+			if (!rcq) {
+				ret = -EINVAL;
+				goto err_put;
+			}
+		}
+		device = pd->device;
 	}
 
 	attr.event_handler = ib_uverbs_qp_event_handler;
@@ -1092,6 +1437,7 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
 	attr.send_cq       = scq;
 	attr.recv_cq       = rcq;
 	attr.srq           = srq;
+	attr.xrcd	   = xrcd;
 	attr.sq_sig_type   = cmd.sq_sig_all ? IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
 	attr.qp_type       = cmd.qp_type;
 	attr.create_flags  = 0;
@@ -1106,26 +1452,34 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
 	INIT_LIST_HEAD(&obj->uevent.event_list);
 	INIT_LIST_HEAD(&obj->mcast_list);
 
-	qp = pd->device->create_qp(pd, &attr, &udata);
+	if (cmd.qp_type == IB_QPT_XRC_TGT)
+		qp = ib_create_qp(pd, &attr);
+	else
+		qp = device->create_qp(pd, &attr, &udata);
+
 	if (IS_ERR(qp)) {
 		ret = PTR_ERR(qp);
 		goto err_put;
 	}
 
-	qp->device     	  = pd->device;
-	qp->pd         	  = pd;
-	qp->send_cq    	  = attr.send_cq;
-	qp->recv_cq    	  = attr.recv_cq;
-	qp->srq	       	  = attr.srq;
-	qp->uobject       = &obj->uevent.uobject;
-	qp->event_handler = attr.event_handler;
-	qp->qp_context    = attr.qp_context;
-	qp->qp_type	  = attr.qp_type;
-	atomic_inc(&pd->usecnt);
-	atomic_inc(&attr.send_cq->usecnt);
-	atomic_inc(&attr.recv_cq->usecnt);
-	if (attr.srq)
-		atomic_inc(&attr.srq->usecnt);
+	if (cmd.qp_type != IB_QPT_XRC_TGT) {
+		qp->real_qp	  = qp;
+		qp->device	  = device;
+		qp->pd		  = pd;
+		qp->send_cq	  = attr.send_cq;
+		qp->recv_cq	  = attr.recv_cq;
+		qp->srq		  = attr.srq;
+		qp->event_handler = attr.event_handler;
+		qp->qp_context	  = attr.qp_context;
+		qp->qp_type	  = attr.qp_type;
+		atomic_inc(&pd->usecnt);
+		atomic_inc(&attr.send_cq->usecnt);
+		if (attr.recv_cq)
+			atomic_inc(&attr.recv_cq->usecnt);
+		if (attr.srq)
+			atomic_inc(&attr.srq->usecnt);
+	}
+	qp->uobject = &obj->uevent.uobject;
 
 	obj->uevent.uobject.object = qp;
 	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
@@ -1147,9 +1501,13 @@ ssize_t ib_uverbs_create_qp(struct ib_uverbs_file *file,
 		goto err_copy;
 	}
 
-	put_pd_read(pd);
-	put_cq_read(scq);
-	if (rcq != scq)
+	if (xrcd)
+		put_xrcd_read(xrcd_uobj);
+	if (pd)
+		put_pd_read(pd);
+	if (scq)
+		put_cq_read(scq);
+	if (rcq && rcq != scq)
 		put_cq_read(rcq);
 	if (srq)
 		put_srq_read(srq);
@@ -1171,6 +1529,8 @@ err_destroy:
 	ib_destroy_qp(qp);
 
 err_put:
+	if (xrcd)
+		put_xrcd_read(xrcd_uobj);
 	if (pd)
 		put_pd_read(pd);
 	if (scq)
@@ -1184,6 +1544,98 @@ err_put:
 	return ret;
 }
 
+ssize_t ib_uverbs_open_qp(struct ib_uverbs_file *file,
+			  const char __user *buf, int in_len, int out_len)
+{
+	struct ib_uverbs_open_qp        cmd;
+	struct ib_uverbs_create_qp_resp resp;
+	struct ib_udata                 udata;
+	struct ib_uqp_object           *obj;
+	struct ib_xrcd		       *xrcd;
+	struct ib_uobject	       *uninitialized_var(xrcd_uobj);
+	struct ib_qp                   *qp;
+	struct ib_qp_open_attr          attr;
+	int ret;
+
+	if (out_len < sizeof resp)
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	INIT_UDATA(&udata, buf + sizeof cmd,
+		   (unsigned long) cmd.response + sizeof resp,
+		   in_len - sizeof cmd, out_len - sizeof resp);
+
+	obj = kmalloc(sizeof *obj, GFP_KERNEL);
+	if (!obj)
+		return -ENOMEM;
+
+	init_uobj(&obj->uevent.uobject, cmd.user_handle, file->ucontext, &qp_lock_key);
+	down_write(&obj->uevent.uobject.mutex);
+
+	xrcd = idr_read_xrcd(cmd.pd_handle, file->ucontext, &xrcd_uobj);
+	if (!xrcd) {
+		ret = -EINVAL;
+		goto err_put;
+	}
+
+	attr.event_handler = ib_uverbs_qp_event_handler;
+	attr.qp_context    = file;
+	attr.qp_num        = cmd.qpn;
+	attr.qp_type       = cmd.qp_type;
+
+	obj->uevent.events_reported = 0;
+	INIT_LIST_HEAD(&obj->uevent.event_list);
+	INIT_LIST_HEAD(&obj->mcast_list);
+
+	qp = ib_open_qp(xrcd, &attr);
+	if (IS_ERR(qp)) {
+		ret = PTR_ERR(qp);
+		goto err_put;
+	}
+
+	qp->uobject = &obj->uevent.uobject;
+
+	obj->uevent.uobject.object = qp;
+	ret = idr_add_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
+	if (ret)
+		goto err_destroy;
+
+	memset(&resp, 0, sizeof resp);
+	resp.qpn       = qp->qp_num;
+	resp.qp_handle = obj->uevent.uobject.id;
+
+	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+			 &resp, sizeof resp)) {
+		ret = -EFAULT;
+		goto err_remove;
+	}
+
+	put_xrcd_read(xrcd_uobj);
+
+	mutex_lock(&file->mutex);
+	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->qp_list);
+	mutex_unlock(&file->mutex);
+
+	obj->uevent.uobject.live = 1;
+
+	up_write(&obj->uevent.uobject.mutex);
+
+	return in_len;
+
+err_remove:
+	idr_remove_uobj(&ib_uverbs_qp_idr, &obj->uevent.uobject);
+
+err_destroy:
+	ib_destroy_qp(qp);
+
+err_put:
+	put_xrcd_read(xrcd_uobj);
+	put_uobj_write(&obj->uevent.uobject);
+	return ret;
+}
+
 ssize_t ib_uverbs_query_qp(struct ib_uverbs_file *file,
 			   const char __user *buf, int in_len,
 			   int out_len)
@@ -1284,6 +1736,20 @@ out:
 	return ret ? ret : in_len;
 }
 
+/* Remove ignored fields set in the attribute mask */
+static int modify_qp_mask(enum ib_qp_type qp_type, int mask)
+{
+	switch (qp_type) {
+	case IB_QPT_XRC_INI:
+		return mask & ~(IB_QP_MAX_DEST_RD_ATOMIC | IB_QP_MIN_RNR_TIMER);
+	case IB_QPT_XRC_TGT:
+		return mask & ~(IB_QP_MAX_QP_RD_ATOMIC | IB_QP_RETRY_CNT |
+				IB_QP_RNR_RETRY);
+	default:
+		return mask;
+	}
+}
+
 ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
 			    const char __user *buf, int in_len,
 			    int out_len)
@@ -1356,7 +1822,12 @@ ssize_t ib_uverbs_modify_qp(struct ib_uverbs_file *file,
 	attr->alt_ah_attr.ah_flags 	    = cmd.alt_dest.is_global ? IB_AH_GRH : 0;
 	attr->alt_ah_attr.port_num 	    = cmd.alt_dest.port_num;
 
-	ret = qp->device->modify_qp(qp, attr, cmd.attr_mask, &udata);
+	if (qp->real_qp == qp) {
+		ret = qp->device->modify_qp(qp, attr,
+			modify_qp_mask(qp->qp_type, cmd.attr_mask), &udata);
+	} else {
+		ret = ib_modify_qp(qp, attr, modify_qp_mask(qp->qp_type, cmd.attr_mask));
+	}
 
 	put_qp_read(qp);
 
@@ -1553,7 +2024,7 @@ ssize_t ib_uverbs_post_send(struct ib_uverbs_file *file,
 	}
 
 	resp.bad_wr = 0;
-	ret = qp->device->post_send(qp, wr, &bad_wr);
+	ret = qp->device->post_send(qp->real_qp, wr, &bad_wr);
 	if (ret)
 		for (next = wr; next; next = next->next) {
 			++resp.bad_wr;
@@ -1691,7 +2162,7 @@ ssize_t ib_uverbs_post_recv(struct ib_uverbs_file *file,
 		goto out;
 
 	resp.bad_wr = 0;
-	ret = qp->device->post_recv(qp, wr, &bad_wr);
+	ret = qp->device->post_recv(qp->real_qp, wr, &bad_wr);
 
 	put_qp_read(qp);
 
@@ -1975,107 +2446,199 @@ out_put:
 	return ret ? ret : in_len;
 }
 
-ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
-			     const char __user *buf, int in_len,
-			     int out_len)
+int __uverbs_create_xsrq(struct ib_uverbs_file *file,
+			 struct ib_uverbs_create_xsrq *cmd,
+			 struct ib_udata *udata)
 {
-	struct ib_uverbs_create_srq      cmd;
 	struct ib_uverbs_create_srq_resp resp;
-	struct ib_udata                  udata;
-	struct ib_uevent_object         *obj;
+	struct ib_usrq_object           *obj;
 	struct ib_pd                    *pd;
 	struct ib_srq                   *srq;
+	struct ib_uobject               *uninitialized_var(xrcd_uobj);
 	struct ib_srq_init_attr          attr;
 	int ret;
 
-	if (out_len < sizeof resp)
-		return -ENOSPC;
-
-	if (copy_from_user(&cmd, buf, sizeof cmd))
-		return -EFAULT;
-
-	INIT_UDATA(&udata, buf + sizeof cmd,
-		   (unsigned long) cmd.response + sizeof resp,
-		   in_len - sizeof cmd, out_len - sizeof resp);
-
 	obj = kmalloc(sizeof *obj, GFP_KERNEL);
 	if (!obj)
 		return -ENOMEM;
 
-	init_uobj(&obj->uobject, cmd.user_handle, file->ucontext, &srq_lock_key);
-	down_write(&obj->uobject.mutex);
+	init_uobj(&obj->uevent.uobject, cmd->user_handle, file->ucontext, &srq_lock_key);
+	down_write(&obj->uevent.uobject.mutex);
 
-	pd  = idr_read_pd(cmd.pd_handle, file->ucontext);
+	pd  = idr_read_pd(cmd->pd_handle, file->ucontext);
 	if (!pd) {
 		ret = -EINVAL;
 		goto err;
 	}
 
+	if (cmd->srq_type == IB_SRQT_XRC) {
+		attr.ext.xrc.cq  = idr_read_cq(cmd->cq_handle, file->ucontext, 0);
+		if (!attr.ext.xrc.cq) {
+			ret = -EINVAL;
+			goto err_put_pd;
+		}
+
+		attr.ext.xrc.xrcd  = idr_read_xrcd(cmd->xrcd_handle, file->ucontext, &xrcd_uobj);
+		if (!attr.ext.xrc.xrcd) {
+			ret = -EINVAL;
+			goto err_put_cq;
+		}
+
+		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object, uobject);
+		atomic_inc(&obj->uxrcd->refcnt);
+	}
+
 	attr.event_handler  = ib_uverbs_srq_event_handler;
 	attr.srq_context    = file;
-	attr.attr.max_wr    = cmd.max_wr;
-	attr.attr.max_sge   = cmd.max_sge;
-	attr.attr.srq_limit = cmd.srq_limit;
+	attr.srq_type       = cmd->srq_type;
+	attr.attr.max_wr    = cmd->max_wr;
+	attr.attr.max_sge   = cmd->max_sge;
+	attr.attr.srq_limit = cmd->srq_limit;
 
-	obj->events_reported     = 0;
-	INIT_LIST_HEAD(&obj->event_list);
+	obj->uevent.events_reported = 0;
+	INIT_LIST_HEAD(&obj->uevent.event_list);
 
-	srq = pd->device->create_srq(pd, &attr, &udata);
+	srq = pd->device->create_srq(pd, &attr, udata);
 	if (IS_ERR(srq)) {
 		ret = PTR_ERR(srq);
 		goto err_put;
 	}
 
-	srq->device    	   = pd->device;
-	srq->pd        	   = pd;
-	srq->uobject       = &obj->uobject;
+	srq->device        = pd->device;
+	srq->pd            = pd;
+	srq->srq_type	   = cmd->srq_type;
+	srq->uobject       = &obj->uevent.uobject;
 	srq->event_handler = attr.event_handler;
 	srq->srq_context   = attr.srq_context;
+
+	if (cmd->srq_type == IB_SRQT_XRC) {
+		srq->ext.xrc.cq   = attr.ext.xrc.cq;
+		srq->ext.xrc.xrcd = attr.ext.xrc.xrcd;
+		atomic_inc(&attr.ext.xrc.cq->usecnt);
+		atomic_inc(&attr.ext.xrc.xrcd->usecnt);
+	}
+
 	atomic_inc(&pd->usecnt);
 	atomic_set(&srq->usecnt, 0);
 
-	obj->uobject.object = srq;
-	ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uobject);
+	obj->uevent.uobject.object = srq;
+	ret = idr_add_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
 	if (ret)
 		goto err_destroy;
 
 	memset(&resp, 0, sizeof resp);
-	resp.srq_handle = obj->uobject.id;
+	resp.srq_handle = obj->uevent.uobject.id;
 	resp.max_wr     = attr.attr.max_wr;
 	resp.max_sge    = attr.attr.max_sge;
+	if (cmd->srq_type == IB_SRQT_XRC)
+		resp.srqn = srq->ext.xrc.srq_num;
 
-	if (copy_to_user((void __user *) (unsigned long) cmd.response,
+	if (copy_to_user((void __user *) (unsigned long) cmd->response,
 			 &resp, sizeof resp)) {
 		ret = -EFAULT;
 		goto err_copy;
 	}
 
+	if (cmd->srq_type == IB_SRQT_XRC) {
+		put_uobj_read(xrcd_uobj);
+		put_cq_read(attr.ext.xrc.cq);
+	}
 	put_pd_read(pd);
 
 	mutex_lock(&file->mutex);
-	list_add_tail(&obj->uobject.list, &file->ucontext->srq_list);
+	list_add_tail(&obj->uevent.uobject.list, &file->ucontext->srq_list);
 	mutex_unlock(&file->mutex);
 
-	obj->uobject.live = 1;
+	obj->uevent.uobject.live = 1;
 
-	up_write(&obj->uobject.mutex);
+	up_write(&obj->uevent.uobject.mutex);
 
-	return in_len;
+	return 0;
 
 err_copy:
-	idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uobject);
+	idr_remove_uobj(&ib_uverbs_srq_idr, &obj->uevent.uobject);
 
 err_destroy:
 	ib_destroy_srq(srq);
 
 err_put:
+	if (cmd->srq_type == IB_SRQT_XRC) {
+		atomic_dec(&obj->uxrcd->refcnt);
+		put_uobj_read(xrcd_uobj);
+	}
+
+err_put_cq:
+	if (cmd->srq_type == IB_SRQT_XRC)
+		put_cq_read(attr.ext.xrc.cq);
+
+err_put_pd:
 	put_pd_read(pd);
 
 err:
-	put_uobj_write(&obj->uobject);
+	put_uobj_write(&obj->uevent.uobject);
 	return ret;
 }
 
+ssize_t ib_uverbs_create_srq(struct ib_uverbs_file *file,
+			     const char __user *buf, int in_len,
+			     int out_len)
+{
+	struct ib_uverbs_create_srq      cmd;
+	struct ib_uverbs_create_xsrq     xcmd;
+	struct ib_uverbs_create_srq_resp resp;
+	struct ib_udata                  udata;
+	int ret;
+
+	if (out_len < sizeof resp)
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	xcmd.response	 = cmd.response;
+	xcmd.user_handle = cmd.user_handle;
+	xcmd.srq_type	 = IB_SRQT_BASIC;
+	xcmd.pd_handle	 = cmd.pd_handle;
+	xcmd.max_wr	 = cmd.max_wr;
+	xcmd.max_sge	 = cmd.max_sge;
+	xcmd.srq_limit	 = cmd.srq_limit;
+
+	INIT_UDATA(&udata, buf + sizeof cmd,
+		   (unsigned long) cmd.response + sizeof resp,
+		   in_len - sizeof cmd, out_len - sizeof resp);
+
+	ret = __uverbs_create_xsrq(file, &xcmd, &udata);
+	if (ret)
+		return ret;
+
+	return in_len;
+}
+
+ssize_t ib_uverbs_create_xsrq(struct ib_uverbs_file *file,
+			      const char __user *buf, int in_len, int out_len)
+{
+	struct ib_uverbs_create_xsrq     cmd;
+	struct ib_uverbs_create_srq_resp resp;
+	struct ib_udata                  udata;
+	int ret;
+
+	if (out_len < sizeof resp)
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, buf, sizeof cmd))
+		return -EFAULT;
+
+	INIT_UDATA(&udata, buf + sizeof cmd,
+		   (unsigned long) cmd.response + sizeof resp,
+		   in_len - sizeof cmd, out_len - sizeof resp);
+
+	ret = __uverbs_create_xsrq(file, &cmd, &udata);
+	if (ret)
+		return ret;
+
+	return in_len;
+}
+
 ssize_t ib_uverbs_modify_srq(struct ib_uverbs_file *file,
 			     const char __user *buf, int in_len,
 			     int out_len)

drivers/infiniband/core/uverbs_main.c

@@ -72,6 +72,7 @@ DEFINE_IDR(ib_uverbs_ah_idr);
 DEFINE_IDR(ib_uverbs_cq_idr);
 DEFINE_IDR(ib_uverbs_qp_idr);
 DEFINE_IDR(ib_uverbs_srq_idr);
+DEFINE_IDR(ib_uverbs_xrcd_idr);
 
 static DEFINE_SPINLOCK(map_lock);
 static DECLARE_BITMAP(dev_map, IB_UVERBS_MAX_DEVICES);
@@ -107,6 +108,10 @@ static ssize_t (*uverbs_cmd_table[])(struct ib_uverbs_file *file,
 	[IB_USER_VERBS_CMD_MODIFY_SRQ]		= ib_uverbs_modify_srq,
 	[IB_USER_VERBS_CMD_QUERY_SRQ]		= ib_uverbs_query_srq,
 	[IB_USER_VERBS_CMD_DESTROY_SRQ]		= ib_uverbs_destroy_srq,
+	[IB_USER_VERBS_CMD_OPEN_XRCD]		= ib_uverbs_open_xrcd,
+	[IB_USER_VERBS_CMD_CLOSE_XRCD]		= ib_uverbs_close_xrcd,
+	[IB_USER_VERBS_CMD_CREATE_XSRQ]		= ib_uverbs_create_xsrq,
+	[IB_USER_VERBS_CMD_OPEN_QP]		= ib_uverbs_open_qp
 };
 
 static void ib_uverbs_add_one(struct ib_device *device);
@@ -202,8 +207,12 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
 			container_of(uobj, struct ib_uqp_object, uevent.uobject);
 
 		idr_remove_uobj(&ib_uverbs_qp_idr, uobj);
-		ib_uverbs_detach_umcast(qp, uqp);
-		ib_destroy_qp(qp);
+		if (qp != qp->real_qp) {
+			ib_close_qp(qp);
+		} else {
+			ib_uverbs_detach_umcast(qp, uqp);
+			ib_destroy_qp(qp);
+		}
 		ib_uverbs_release_uevent(file, &uqp->uevent);
 		kfree(uqp);
 	}
@@ -241,6 +250,18 @@ static int ib_uverbs_cleanup_ucontext(struct ib_uverbs_file *file,
 		kfree(uobj);
 	}
 
+	mutex_lock(&file->device->xrcd_tree_mutex);
+	list_for_each_entry_safe(uobj, tmp, &context->xrcd_list, list) {
+		struct ib_xrcd *xrcd = uobj->object;
+		struct ib_uxrcd_object *uxrcd =
+			container_of(uobj, struct ib_uxrcd_object, uobject);
+
+		idr_remove_uobj(&ib_uverbs_xrcd_idr, uobj);
+		ib_uverbs_dealloc_xrcd(file->device, xrcd);
+		kfree(uxrcd);
+	}
+	mutex_unlock(&file->device->xrcd_tree_mutex);
+
 	list_for_each_entry_safe(uobj, tmp, &context->pd_list, list) {
 		struct ib_pd *pd = uobj->object;
 
@@ -557,8 +578,7 @@ static ssize_t ib_uverbs_write(struct file *filp, const char __user *buf,
 	if (hdr.in_words * 4 != count)
 		return -EINVAL;
 
-	if (hdr.command < 0				||
-	    hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
+	if (hdr.command >= ARRAY_SIZE(uverbs_cmd_table) ||
 	    !uverbs_cmd_table[hdr.command])
 		return -EINVAL;
 
@@ -741,6 +761,8 @@ static void ib_uverbs_add_one(struct ib_device *device)
 
 	kref_init(&uverbs_dev->ref);
 	init_completion(&uverbs_dev->comp);
+	uverbs_dev->xrcd_tree = RB_ROOT;
+	mutex_init(&uverbs_dev->xrcd_tree_mutex);
 
 	spin_lock(&map_lock);
 	devnum = find_first_zero_bit(dev_map, IB_UVERBS_MAX_DEVICES);

drivers/infiniband/core/verbs.c

@@ -39,6 +39,7 @@
 #include <linux/errno.h>
 #include <linux/err.h>
 #include <linux/string.h>
+#include <linux/slab.h>
 
 #include <rdma/ib_verbs.h>
 #include <rdma/ib_cache.h>
@@ -77,6 +78,31 @@ enum ib_rate mult_to_ib_rate(int mult)
 }
 EXPORT_SYMBOL(mult_to_ib_rate);
 
+int ib_rate_to_mbps(enum ib_rate rate)
+{
+	switch (rate) {
+	case IB_RATE_2_5_GBPS: return 2500;
+	case IB_RATE_5_GBPS:   return 5000;
+	case IB_RATE_10_GBPS:  return 10000;
+	case IB_RATE_20_GBPS:  return 20000;
+	case IB_RATE_30_GBPS:  return 30000;
+	case IB_RATE_40_GBPS:  return 40000;
+	case IB_RATE_60_GBPS:  return 60000;
+	case IB_RATE_80_GBPS:  return 80000;
+	case IB_RATE_120_GBPS: return 120000;
+	case IB_RATE_14_GBPS:  return 14062;
+	case IB_RATE_56_GBPS:  return 56250;
+	case IB_RATE_112_GBPS: return 112500;
+	case IB_RATE_168_GBPS: return 168750;
+	case IB_RATE_25_GBPS:  return 25781;
+	case IB_RATE_100_GBPS: return 103125;
+	case IB_RATE_200_GBPS: return 206250;
+	case IB_RATE_300_GBPS: return 309375;
+	default:	       return -1;
+	}
+}
+EXPORT_SYMBOL(ib_rate_to_mbps);
+
 enum rdma_transport_type
 rdma_node_get_transport(enum rdma_node_type node_type)
 {
@@ -250,6 +276,13 @@ struct ib_srq *ib_create_srq(struct ib_pd *pd,
 		srq->uobject       = NULL;
 		srq->event_handler = srq_init_attr->event_handler;
 		srq->srq_context   = srq_init_attr->srq_context;
+		srq->srq_type      = srq_init_attr->srq_type;
+		if (srq->srq_type == IB_SRQT_XRC) {
+			srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
+			srq->ext.xrc.cq   = srq_init_attr->ext.xrc.cq;
+			atomic_inc(&srq->ext.xrc.xrcd->usecnt);
+			atomic_inc(&srq->ext.xrc.cq->usecnt);
+		}
 		atomic_inc(&pd->usecnt);
 		atomic_set(&srq->usecnt, 0);
 	}
@@ -279,16 +312,29 @@ EXPORT_SYMBOL(ib_query_srq);
 int ib_destroy_srq(struct ib_srq *srq)
 {
 	struct ib_pd *pd;
+	enum ib_srq_type srq_type;
+	struct ib_xrcd *uninitialized_var(xrcd);
+	struct ib_cq *uninitialized_var(cq);
 	int ret;
 
 	if (atomic_read(&srq->usecnt))
 		return -EBUSY;
 
 	pd = srq->pd;
+	srq_type = srq->srq_type;
+	if (srq_type == IB_SRQT_XRC) {
+		xrcd = srq->ext.xrc.xrcd;
+		cq = srq->ext.xrc.cq;
+	}
 
 	ret = srq->device->destroy_srq(srq);
-	if (!ret)
+	if (!ret) {
 		atomic_dec(&pd->usecnt);
+		if (srq_type == IB_SRQT_XRC) {
+			atomic_dec(&xrcd->usecnt);
+			atomic_dec(&cq->usecnt);
+		}
+	}
 
 	return ret;
 }
@@ -296,28 +342,123 @@ EXPORT_SYMBOL(ib_destroy_srq);
 
 /* Queue pairs */
 
+static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
+{
+	struct ib_qp *qp = context;
+
+	list_for_each_entry(event->element.qp, &qp->open_list, open_list)
+		event->element.qp->event_handler(event, event->element.qp->qp_context);
+}
+
+static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
+{
+	mutex_lock(&xrcd->tgt_qp_mutex);
+	list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
+	mutex_unlock(&xrcd->tgt_qp_mutex);
+}
+
+static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
+				  void (*event_handler)(struct ib_event *, void *),
+				  void *qp_context)
+{
+	struct ib_qp *qp;
+	unsigned long flags;
+
+	qp = kzalloc(sizeof *qp, GFP_KERNEL);
+	if (!qp)
+		return ERR_PTR(-ENOMEM);
+
+	qp->real_qp = real_qp;
+	atomic_inc(&real_qp->usecnt);
+	qp->device = real_qp->device;
+	qp->event_handler = event_handler;
+	qp->qp_context = qp_context;
+	qp->qp_num = real_qp->qp_num;
+	qp->qp_type = real_qp->qp_type;
+
+	spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
+	list_add(&qp->open_list, &real_qp->open_list);
+	spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
+
+	return qp;
+}
+
+struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
+			 struct ib_qp_open_attr *qp_open_attr)
+{
+	struct ib_qp *qp, *real_qp;
+
+	if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
+		return ERR_PTR(-EINVAL);
+
+	qp = ERR_PTR(-EINVAL);
+	mutex_lock(&xrcd->tgt_qp_mutex);
+	list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
+		if (real_qp->qp_num == qp_open_attr->qp_num) {
+			qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
+					  qp_open_attr->qp_context);
+			break;
+		}
+	}
+	mutex_unlock(&xrcd->tgt_qp_mutex);
+	return qp;
+}
+EXPORT_SYMBOL(ib_open_qp);
+
 struct ib_qp *ib_create_qp(struct ib_pd *pd,
 			   struct ib_qp_init_attr *qp_init_attr)
 {
-	struct ib_qp *qp;
+	struct ib_qp *qp, *real_qp;
+	struct ib_device *device;
 
-	qp = pd->device->create_qp(pd, qp_init_attr, NULL);
+	device = pd ? pd->device : qp_init_attr->xrcd->device;
+	qp = device->create_qp(pd, qp_init_attr, NULL);
 
 	if (!IS_ERR(qp)) {
-		qp->device     	  = pd->device;
-		qp->pd         	  = pd;
-		qp->send_cq    	  = qp_init_attr->send_cq;
-		qp->recv_cq    	  = qp_init_attr->recv_cq;
-		qp->srq	       	  = qp_init_attr->srq;
-		qp->uobject       = NULL;
-		qp->event_handler = qp_init_attr->event_handler;
-		qp->qp_context    = qp_init_attr->qp_context;
-		qp->qp_type	  = qp_init_attr->qp_type;
-		atomic_inc(&pd->usecnt);
-		atomic_inc(&qp_init_attr->send_cq->usecnt);
-		atomic_inc(&qp_init_attr->recv_cq->usecnt);
-		if (qp_init_attr->srq)
-			atomic_inc(&qp_init_attr->srq->usecnt);
+		qp->device     = device;
+		qp->real_qp    = qp;
+		qp->uobject    = NULL;
+		qp->qp_type    = qp_init_attr->qp_type;
+
+		if (qp_init_attr->qp_type == IB_QPT_XRC_TGT) {
+			qp->event_handler = __ib_shared_qp_event_handler;
+			qp->qp_context = qp;
+			qp->pd = NULL;
+			qp->send_cq = qp->recv_cq = NULL;
+			qp->srq = NULL;
+			qp->xrcd = qp_init_attr->xrcd;
+			atomic_inc(&qp_init_attr->xrcd->usecnt);
+			INIT_LIST_HEAD(&qp->open_list);
+			atomic_set(&qp->usecnt, 0);
+
+			real_qp = qp;
+			qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
+					  qp_init_attr->qp_context);
+			if (!IS_ERR(qp))
+				__ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
+			else
+				real_qp->device->destroy_qp(real_qp);
+		} else {
+			qp->event_handler = qp_init_attr->event_handler;
+			qp->qp_context = qp_init_attr->qp_context;
+			if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
+				qp->recv_cq = NULL;
+				qp->srq = NULL;
+			} else {
+				qp->recv_cq = qp_init_attr->recv_cq;
+				atomic_inc(&qp_init_attr->recv_cq->usecnt);
+				qp->srq = qp_init_attr->srq;
+				if (qp->srq)
+					atomic_inc(&qp_init_attr->srq->usecnt);
+			}
+
+			qp->pd	    = pd;
+			qp->send_cq = qp_init_attr->send_cq;
+			qp->xrcd    = NULL;
+
+			atomic_inc(&pd->usecnt);
+			atomic_inc(&qp_init_attr->send_cq->usecnt);
+		}
 	}
 
 	return qp;
@@ -326,8 +467,8 @@ EXPORT_SYMBOL(ib_create_qp);
 
 static const struct {
 	int			valid;
-	enum ib_qp_attr_mask	req_param[IB_QPT_RAW_ETHERTYPE + 1];
-	enum ib_qp_attr_mask	opt_param[IB_QPT_RAW_ETHERTYPE + 1];
+	enum ib_qp_attr_mask	req_param[IB_QPT_MAX];
+	enum ib_qp_attr_mask	opt_param[IB_QPT_MAX];
 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
 	[IB_QPS_RESET] = {
 		[IB_QPS_RESET] = { .valid = 1 },
@@ -343,6 +484,12 @@ static const struct {
 				[IB_QPT_RC]  = (IB_QP_PKEY_INDEX		|
 						IB_QP_PORT			|
 						IB_QP_ACCESS_FLAGS),
+				[IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX		|
+						IB_QP_PORT			|
+						IB_QP_ACCESS_FLAGS),
+				[IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX		|
+						IB_QP_PORT			|
+						IB_QP_ACCESS_FLAGS),
 				[IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
 						IB_QP_QKEY),
 				[IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
@@ -365,6 +512,12 @@ static const struct {
 				[IB_QPT_RC]  = (IB_QP_PKEY_INDEX		|
 						IB_QP_PORT			|
 						IB_QP_ACCESS_FLAGS),
+				[IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX		|
+						IB_QP_PORT			|
+						IB_QP_ACCESS_FLAGS),
+				[IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX		|
+						IB_QP_PORT			|
+						IB_QP_ACCESS_FLAGS),
 				[IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
 						IB_QP_QKEY),
 				[IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
@@ -384,6 +537,16 @@ static const struct {
 						IB_QP_RQ_PSN			|
 						IB_QP_MAX_DEST_RD_ATOMIC	|
 						IB_QP_MIN_RNR_TIMER),
+				[IB_QPT_XRC_INI] = (IB_QP_AV			|
+						IB_QP_PATH_MTU			|
+						IB_QP_DEST_QPN			|
+						IB_QP_RQ_PSN),
+				[IB_QPT_XRC_TGT] = (IB_QP_AV			|
+						IB_QP_PATH_MTU			|
+						IB_QP_DEST_QPN			|
+						IB_QP_RQ_PSN			|
+						IB_QP_MAX_DEST_RD_ATOMIC	|
+						IB_QP_MIN_RNR_TIMER),
 			},
 			.opt_param = {
 				 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX		|
@@ -394,6 +557,12 @@ static const struct {
 				 [IB_QPT_RC]  = (IB_QP_ALT_PATH			|
 						 IB_QP_ACCESS_FLAGS		|
 						 IB_QP_PKEY_INDEX),
+				 [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH		|
+						 IB_QP_ACCESS_FLAGS		|
+						 IB_QP_PKEY_INDEX),
+				 [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH		|
+						 IB_QP_ACCESS_FLAGS		|
+						 IB_QP_PKEY_INDEX),
 				 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
 						 IB_QP_QKEY),
 				 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
@@ -414,6 +583,13 @@ static const struct {
 						IB_QP_RNR_RETRY			|
 						IB_QP_SQ_PSN			|
 						IB_QP_MAX_QP_RD_ATOMIC),
+				[IB_QPT_XRC_INI] = (IB_QP_TIMEOUT		|
+						IB_QP_RETRY_CNT			|
+						IB_QP_RNR_RETRY			|
+						IB_QP_SQ_PSN			|
+						IB_QP_MAX_QP_RD_ATOMIC),
+				[IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT		|
+						IB_QP_SQ_PSN),
 				[IB_QPT_SMI] = IB_QP_SQ_PSN,
 				[IB_QPT_GSI] = IB_QP_SQ_PSN,
 			},
@@ -429,6 +605,15 @@ static const struct {
 						 IB_QP_ACCESS_FLAGS		|
 						 IB_QP_MIN_RNR_TIMER		|
 						 IB_QP_PATH_MIG_STATE),
+				 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE		|
+						 IB_QP_ALT_PATH			|
+						 IB_QP_ACCESS_FLAGS		|
+						 IB_QP_PATH_MIG_STATE),
+				 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE		|
+						 IB_QP_ALT_PATH			|
+						 IB_QP_ACCESS_FLAGS		|
+						 IB_QP_MIN_RNR_TIMER		|
+						 IB_QP_PATH_MIG_STATE),
 				 [IB_QPT_SMI] = (IB_QP_CUR_STATE		|
 						 IB_QP_QKEY),
 				 [IB_QPT_GSI] = (IB_QP_CUR_STATE		|
@@ -453,6 +638,15 @@ static const struct {
 						IB_QP_ALT_PATH			|
 						IB_QP_PATH_MIG_STATE		|
 						IB_QP_MIN_RNR_TIMER),
+				[IB_QPT_XRC_INI] = (IB_QP_CUR_STATE		|
+						IB_QP_ACCESS_FLAGS		|
+						IB_QP_ALT_PATH			|
+						IB_QP_PATH_MIG_STATE),
+				[IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE		|
+						IB_QP_ACCESS_FLAGS		|
+						IB_QP_ALT_PATH			|
+						IB_QP_PATH_MIG_STATE		|
+						IB_QP_MIN_RNR_TIMER),
 				[IB_QPT_SMI] = (IB_QP_CUR_STATE			|
 						IB_QP_QKEY),
 				[IB_QPT_GSI] = (IB_QP_CUR_STATE			|
@@ -465,6 +659,8 @@ static const struct {
 				[IB_QPT_UD]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
 				[IB_QPT_UC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
 				[IB_QPT_RC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
+				[IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
+				[IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
 				[IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
 				[IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
 			}
@@ -487,6 +683,15 @@ static const struct {
 						IB_QP_ACCESS_FLAGS		|
 						IB_QP_MIN_RNR_TIMER		|
 						IB_QP_PATH_MIG_STATE),
+				[IB_QPT_XRC_INI] = (IB_QP_CUR_STATE		|
+						IB_QP_ALT_PATH			|
+						IB_QP_ACCESS_FLAGS		|
+						IB_QP_PATH_MIG_STATE),
+				[IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE		|
+						IB_QP_ALT_PATH			|
+						IB_QP_ACCESS_FLAGS		|
+						IB_QP_MIN_RNR_TIMER		|
+						IB_QP_PATH_MIG_STATE),
 				[IB_QPT_SMI] = (IB_QP_CUR_STATE			|
 						IB_QP_QKEY),
 				[IB_QPT_GSI] = (IB_QP_CUR_STATE			|
@@ -515,6 +720,25 @@ static const struct {
 						IB_QP_PKEY_INDEX		|
 						IB_QP_MIN_RNR_TIMER		|
 						IB_QP_PATH_MIG_STATE),
+				[IB_QPT_XRC_INI] = (IB_QP_PORT			|
+						IB_QP_AV			|
+						IB_QP_TIMEOUT			|
+						IB_QP_RETRY_CNT			|
+						IB_QP_RNR_RETRY			|
+						IB_QP_MAX_QP_RD_ATOMIC		|
+						IB_QP_ALT_PATH			|
+						IB_QP_ACCESS_FLAGS		|
+						IB_QP_PKEY_INDEX		|
+						IB_QP_PATH_MIG_STATE),
+				[IB_QPT_XRC_TGT] = (IB_QP_PORT			|
+						IB_QP_AV			|
+						IB_QP_TIMEOUT			|
+						IB_QP_MAX_DEST_RD_ATOMIC	|
+						IB_QP_ALT_PATH			|
+						IB_QP_ACCESS_FLAGS		|
+						IB_QP_PKEY_INDEX		|
+						IB_QP_MIN_RNR_TIMER		|
+						IB_QP_PATH_MIG_STATE),
 				[IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
 						IB_QP_QKEY),
 				[IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
@@ -579,7 +803,7 @@ int ib_modify_qp(struct ib_qp *qp,
 		 struct ib_qp_attr *qp_attr,
 		 int qp_attr_mask)
 {
-	return qp->device->modify_qp(qp, qp_attr, qp_attr_mask, NULL);
+	return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
 }
 EXPORT_SYMBOL(ib_modify_qp);
 
@@ -589,11 +813,59 @@ int ib_query_qp(struct ib_qp *qp,
 		struct ib_qp_init_attr *qp_init_attr)
 {
 	return qp->device->query_qp ?
-		qp->device->query_qp(qp, qp_attr, qp_attr_mask, qp_init_attr) :
+		qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
 		-ENOSYS;
 }
 EXPORT_SYMBOL(ib_query_qp);
 
+int ib_close_qp(struct ib_qp *qp)
+{
+	struct ib_qp *real_qp;
+	unsigned long flags;
+
+	real_qp = qp->real_qp;
+	if (real_qp == qp)
+		return -EINVAL;
+
+	spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
+	list_del(&qp->open_list);
+	spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
+
+	atomic_dec(&real_qp->usecnt);
+	kfree(qp);
+
+	return 0;
+}
+EXPORT_SYMBOL(ib_close_qp);
+
+static int __ib_destroy_shared_qp(struct ib_qp *qp)
+{
+	struct ib_xrcd *xrcd;
+	struct ib_qp *real_qp;
+	int ret;
+
+	real_qp = qp->real_qp;
+	xrcd = real_qp->xrcd;
+
+	mutex_lock(&xrcd->tgt_qp_mutex);
+	ib_close_qp(qp);
+	if (atomic_read(&real_qp->usecnt) == 0)
+		list_del(&real_qp->xrcd_list);
+	else
+		real_qp = NULL;
+	mutex_unlock(&xrcd->tgt_qp_mutex);
+
+	if (real_qp) {
+		ret = ib_destroy_qp(real_qp);
+		if (!ret)
+			atomic_dec(&xrcd->usecnt);
+		else
+			__ib_insert_xrcd_qp(xrcd, real_qp);
+	}
+
+	return 0;
+}
+
 int ib_destroy_qp(struct ib_qp *qp)
 {
 	struct ib_pd *pd;
@@ -601,16 +873,25 @@ int ib_destroy_qp(struct ib_qp *qp)
 	struct ib_srq *srq;
 	int ret;
 
-	pd  = qp->pd;
-	scq = qp->send_cq;
-	rcq = qp->recv_cq;
-	srq = qp->srq;
+	if (atomic_read(&qp->usecnt))
+		return -EBUSY;
+
+	if (qp->real_qp != qp)
+		return __ib_destroy_shared_qp(qp);
+
+	pd   = qp->pd;
+	scq  = qp->send_cq;
+	rcq  = qp->recv_cq;
+	srq  = qp->srq;
 
 	ret = qp->device->destroy_qp(qp);
 	if (!ret) {
-		atomic_dec(&pd->usecnt);
-		atomic_dec(&scq->usecnt);
-		atomic_dec(&rcq->usecnt);
+		if (pd)
+			atomic_dec(&pd->usecnt);
+		if (scq)
+			atomic_dec(&scq->usecnt);
+		if (rcq)
+			atomic_dec(&rcq->usecnt);
 		if (srq)
 			atomic_dec(&srq->usecnt);
 	}
@@ -920,3 +1201,42 @@ int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
 	return qp->device->detach_mcast(qp, gid, lid);
 }
 EXPORT_SYMBOL(ib_detach_mcast);
+
+struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
+{
+	struct ib_xrcd *xrcd;
+
+	if (!device->alloc_xrcd)
+		return ERR_PTR(-ENOSYS);
+
+	xrcd = device->alloc_xrcd(device, NULL, NULL);
+	if (!IS_ERR(xrcd)) {
+		xrcd->device = device;
+		xrcd->inode = NULL;
+		atomic_set(&xrcd->usecnt, 0);
+		mutex_init(&xrcd->tgt_qp_mutex);
+		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
+	}
+
+	return xrcd;
+}
+EXPORT_SYMBOL(ib_alloc_xrcd);
+
+int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
+{
+	struct ib_qp *qp;
+	int ret;
+
+	if (atomic_read(&xrcd->usecnt))
+		return -EBUSY;
+
+	while (!list_empty(&xrcd->tgt_qp_list)) {
+		qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
+		ret = ib_destroy_qp(qp);
+		if (ret)
+			return ret;
+	}
+
+	return xrcd->device->dealloc_xrcd(xrcd);
+}
+EXPORT_SYMBOL(ib_dealloc_xrcd);

drivers/infiniband/hw/amso1100/c2_ae.c

@@ -288,6 +288,11 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
 		cm_event.private_data_len =
 			be32_to_cpu(req->private_data_length);
 		cm_event.private_data = req->private_data;
+		/*
+		 * Until ird/ord negotiation via MPAv2 support is added, send
+		 * max supported values
+		 */
+		cm_event.ird = cm_event.ord = 128;
 
 		if (cm_id->event_handler)
 			cm_id->event_handler(cm_id, &cm_event);

drivers/infiniband/hw/amso1100/c2_intr.c

@@ -183,6 +183,11 @@ static void handle_vq(struct c2_dev *c2dev, u32 mq_index)
 	case IW_CM_EVENT_ESTABLISHED:
 		c2_set_qp_state(req->qp,
 				C2_QP_STATE_RTS);
+		/*
+		 * Until ird/ord negotiation via MPAv2 support is added, send
+		 * max supported values
+		 */
+		cm_event.ird = cm_event.ord = 128;
 	case IW_CM_EVENT_CLOSE:
 
 		/*

drivers/infiniband/hw/amso1100/c2_provider.c

@@ -753,10 +753,7 @@ static struct net_device *c2_pseudo_netdev_init(struct c2_dev *c2dev)
 	memcpy_fromio(netdev->dev_addr, c2dev->kva + C2_REGS_RDMA_ENADDR, 6);
 
 	/* Print out the MAC address */
-	pr_debug("%s: MAC %02X:%02X:%02X:%02X:%02X:%02X\n",
-		netdev->name,
-		netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
-		netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
+	pr_debug("%s: MAC %pM\n", netdev->name, netdev->dev_addr);
 
 #if 0
 	/* Disable network packets */

drivers/infiniband/hw/cxgb3/iwch_cm.c

@@ -753,6 +753,11 @@ static void connect_request_upcall(struct iwch_ep *ep)
 	event.private_data_len = ep->plen;
 	event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 	event.provider_data = ep;
+	/*
+	 * Until ird/ord negotiation via MPAv2 support is added, send max
+	 * supported values
+	 */
+	event.ird = event.ord = 8;
 	if (state_read(&ep->parent_ep->com) != DEAD) {
 		get_ep(&ep->com);
 		ep->parent_ep->com.cm_id->event_handler(
@@ -770,6 +775,11 @@ static void established_upcall(struct iwch_ep *ep)
 	PDBG("%s ep %p\n", __func__, ep);
 	memset(&event, 0, sizeof(event));
 	event.event = IW_CM_EVENT_ESTABLISHED;
+	/*
+	 * Until ird/ord negotiation via MPAv2 support is added, send max
+	 * supported values
+	 */
+	event.ird = event.ord = 8;
 	if (ep->com.cm_id) {
 		PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);

drivers/infiniband/hw/cxgb3/iwch_ev.c

@@ -46,6 +46,7 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
 	struct ib_event event;
 	struct iwch_qp_attributes attrs;
 	struct iwch_qp *qhp;
+	unsigned long flag;
 
 	spin_lock(&rnicp->lock);
 	qhp = get_qhp(rnicp, CQE_QPID(rsp_msg->cqe));
@@ -94,7 +95,9 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
 	if (qhp->ibqp.event_handler)
 		(*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
 
+	spin_lock_irqsave(&chp->comp_handler_lock, flag);
 	(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+	spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 
 	if (atomic_dec_and_test(&qhp->refcnt))
 		wake_up(&qhp->wait);
@@ -107,6 +110,7 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
 	struct iwch_cq *chp;
 	struct iwch_qp *qhp;
 	u32 cqid = RSPQ_CQID(rsp_msg);
+	unsigned long flag;
 
 	rnicp = (struct iwch_dev *) rdev_p->ulp;
 	spin_lock(&rnicp->lock);
@@ -170,7 +174,9 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
 		 */
 		if (qhp->ep && SQ_TYPE(rsp_msg->cqe))
 			dst_confirm(qhp->ep->dst);
+		spin_lock_irqsave(&chp->comp_handler_lock, flag);
 		(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+		spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 		break;
 
 	case TPT_ERR_STAG:

drivers/infiniband/hw/cxgb3/iwch_provider.c

@@ -190,6 +190,7 @@ static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int ve
 	chp->rhp = rhp;
 	chp->ibcq.cqe = 1 << chp->cq.size_log2;
 	spin_lock_init(&chp->lock);
+	spin_lock_init(&chp->comp_handler_lock);
 	atomic_set(&chp->refcnt, 1);
 	init_waitqueue_head(&chp->wait);
 	if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {

drivers/infiniband/hw/cxgb3/iwch_provider.h

@@ -103,6 +103,7 @@ struct iwch_cq {
 	struct iwch_dev *rhp;
 	struct t3_cq cq;
 	spinlock_t lock;
+	spinlock_t comp_handler_lock;
 	atomic_t refcnt;
 	wait_queue_head_t wait;
 	u32 __user *user_rptr_addr;

drivers/infiniband/hw/cxgb3/iwch_qp.c

@@ -822,8 +822,11 @@ static void __flush_qp(struct iwch_qp *qhp, struct iwch_cq *rchp,
 	flushed = cxio_flush_rq(&qhp->wq, &rchp->cq, count);
 	spin_unlock(&qhp->lock);
 	spin_unlock_irqrestore(&rchp->lock, *flag);
-	if (flushed)
+	if (flushed) {
+		spin_lock_irqsave(&rchp->comp_handler_lock, *flag);
 		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock_irqrestore(&rchp->comp_handler_lock, *flag);
+	}
 
 	/* locking hierarchy: cq lock first, then qp lock. */
 	spin_lock_irqsave(&schp->lock, *flag);
@@ -833,8 +836,11 @@ static void __flush_qp(struct iwch_qp *qhp, struct iwch_cq *rchp,
 	flushed = cxio_flush_sq(&qhp->wq, &schp->cq, count);
 	spin_unlock(&qhp->lock);
 	spin_unlock_irqrestore(&schp->lock, *flag);
-	if (flushed)
+	if (flushed) {
+		spin_lock_irqsave(&schp->comp_handler_lock, *flag);
 		(*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
+		spin_unlock_irqrestore(&schp->comp_handler_lock, *flag);
+	}
 
 	/* deref */
 	if (atomic_dec_and_test(&qhp->refcnt))
@@ -853,11 +859,15 @@ static void flush_qp(struct iwch_qp *qhp, unsigned long *flag)
 	if (qhp->ibqp.uobject) {
 		cxio_set_wq_in_error(&qhp->wq);
 		cxio_set_cq_in_error(&rchp->cq);
+		spin_lock_irqsave(&rchp->comp_handler_lock, *flag);
 		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock_irqrestore(&rchp->comp_handler_lock, *flag);
 		if (schp != rchp) {
 			cxio_set_cq_in_error(&schp->cq);
+			spin_lock_irqsave(&schp->comp_handler_lock, *flag);
 			(*schp->ibcq.comp_handler)(&schp->ibcq,
 						   schp->ibcq.cq_context);
+			spin_unlock_irqrestore(&schp->comp_handler_lock, *flag);
 		}
 		return;
 	}

drivers/infiniband/hw/cxgb4/cm.c

@@ -103,7 +103,8 @@ MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
 static int mpa_rev = 1;
 module_param(mpa_rev, int, 0644);
 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
-		 "1 is spec compliant. (default=1)");
+		"1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
+		" compliant (default=1)");
 
 static int markers_enabled;
 module_param(markers_enabled, int, 0644);
@@ -497,17 +498,21 @@ static int send_connect(struct c4iw_ep *ep)
 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
 }
 
-static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb)
+static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
+		u8 mpa_rev_to_use)
 {
 	int mpalen, wrlen;
 	struct fw_ofld_tx_data_wr *req;
 	struct mpa_message *mpa;
+	struct mpa_v2_conn_params mpa_v2_params;
 
 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
 
 	BUG_ON(skb_cloned(skb));
 
 	mpalen = sizeof(*mpa) + ep->plen;
+	if (mpa_rev_to_use == 2)
+		mpalen += sizeof(struct mpa_v2_conn_params);
 	wrlen = roundup(mpalen + sizeof *req, 16);
 	skb = get_skb(skb, wrlen, GFP_KERNEL);
 	if (!skb) {
@@ -533,12 +538,39 @@ static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb)
 	mpa = (struct mpa_message *)(req + 1);
 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
-		     (markers_enabled ? MPA_MARKERS : 0);
+		     (markers_enabled ? MPA_MARKERS : 0) |
+		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
 	mpa->private_data_size = htons(ep->plen);
-	mpa->revision = mpa_rev;
+	mpa->revision = mpa_rev_to_use;
+	if (mpa_rev_to_use == 1)
+		ep->tried_with_mpa_v1 = 1;
+
+	if (mpa_rev_to_use == 2) {
+		mpa->private_data_size +=
+			htons(sizeof(struct mpa_v2_conn_params));
+		mpa_v2_params.ird = htons((u16)ep->ird);
+		mpa_v2_params.ord = htons((u16)ep->ord);
+
+		if (peer2peer) {
+			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
+			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_WRITE_RTR);
+			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_READ_RTR);
+		}
+		memcpy(mpa->private_data, &mpa_v2_params,
+		       sizeof(struct mpa_v2_conn_params));
 
-	if (ep->plen)
-		memcpy(mpa->private_data, ep->mpa_pkt + sizeof(*mpa), ep->plen);
+		if (ep->plen)
+			memcpy(mpa->private_data +
+			       sizeof(struct mpa_v2_conn_params),
+			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
+	} else
+		if (ep->plen)
+			memcpy(mpa->private_data,
+					ep->mpa_pkt + sizeof(*mpa), ep->plen);
 
 	/*
 	 * Reference the mpa skb.  This ensures the data area
@@ -562,10 +594,13 @@ static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
 	struct fw_ofld_tx_data_wr *req;
 	struct mpa_message *mpa;
 	struct sk_buff *skb;
+	struct mpa_v2_conn_params mpa_v2_params;
 
 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
 
 	mpalen = sizeof(*mpa) + plen;
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
+		mpalen += sizeof(struct mpa_v2_conn_params);
 	wrlen = roundup(mpalen + sizeof *req, 16);
 
 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
@@ -595,8 +630,29 @@ static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
 	mpa->flags = MPA_REJECT;
 	mpa->revision = mpa_rev;
 	mpa->private_data_size = htons(plen);
-	if (plen)
-		memcpy(mpa->private_data, pdata, plen);
+
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
+		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
+		mpa->private_data_size +=
+			htons(sizeof(struct mpa_v2_conn_params));
+		mpa_v2_params.ird = htons(((u16)ep->ird) |
+					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
+					   0));
+		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
+					  (p2p_type ==
+					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
+					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
+					   FW_RI_INIT_P2PTYPE_READ_REQ ?
+					   MPA_V2_RDMA_READ_RTR : 0) : 0));
+		memcpy(mpa->private_data, &mpa_v2_params,
+		       sizeof(struct mpa_v2_conn_params));
+
+		if (ep->plen)
+			memcpy(mpa->private_data +
+			       sizeof(struct mpa_v2_conn_params), pdata, plen);
+	} else
+		if (plen)
+			memcpy(mpa->private_data, pdata, plen);
 
 	/*
 	 * Reference the mpa skb again.  This ensures the data area
@@ -617,10 +673,13 @@ static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
 	struct fw_ofld_tx_data_wr *req;
 	struct mpa_message *mpa;
 	struct sk_buff *skb;
+	struct mpa_v2_conn_params mpa_v2_params;
 
 	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
 
 	mpalen = sizeof(*mpa) + plen;
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
+		mpalen += sizeof(struct mpa_v2_conn_params);
 	wrlen = roundup(mpalen + sizeof *req, 16);
 
 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
@@ -649,10 +708,36 @@ static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
 		     (markers_enabled ? MPA_MARKERS : 0);
-	mpa->revision = mpa_rev;
+	mpa->revision = ep->mpa_attr.version;
 	mpa->private_data_size = htons(plen);
-	if (plen)
-		memcpy(mpa->private_data, pdata, plen);
+
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
+		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
+		mpa->private_data_size +=
+			htons(sizeof(struct mpa_v2_conn_params));
+		mpa_v2_params.ird = htons((u16)ep->ird);
+		mpa_v2_params.ord = htons((u16)ep->ord);
+		if (peer2peer && (ep->mpa_attr.p2p_type !=
+					FW_RI_INIT_P2PTYPE_DISABLED)) {
+			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
+
+			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_WRITE_RTR);
+			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_READ_RTR);
+		}
+
+		memcpy(mpa->private_data, &mpa_v2_params,
+		       sizeof(struct mpa_v2_conn_params));
+
+		if (ep->plen)
+			memcpy(mpa->private_data +
+			       sizeof(struct mpa_v2_conn_params), pdata, plen);
+	} else
+		if (plen)
+			memcpy(mpa->private_data, pdata, plen);
 
 	/*
 	 * Reference the mpa skb.  This ensures the data area
@@ -695,7 +780,10 @@ static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
 
 	/* start MPA negotiation */
 	send_flowc(ep, NULL);
-	send_mpa_req(ep, skb);
+	if (ep->retry_with_mpa_v1)
+		send_mpa_req(ep, skb, 1);
+	else
+		send_mpa_req(ep, skb, mpa_rev);
 
 	return 0;
 }
@@ -769,8 +857,19 @@ static void connect_reply_upcall(struct c4iw_ep *ep, int status)
 	event.remote_addr = ep->com.remote_addr;
 
 	if ((status == 0) || (status == -ECONNREFUSED)) {
-		event.private_data_len = ep->plen;
-		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
+		if (!ep->tried_with_mpa_v1) {
+			/* this means MPA_v2 is used */
+			event.private_data_len = ep->plen -
+				sizeof(struct mpa_v2_conn_params);
+			event.private_data = ep->mpa_pkt +
+				sizeof(struct mpa_message) +
+				sizeof(struct mpa_v2_conn_params);
+		} else {
+			/* this means MPA_v1 is used */
+			event.private_data_len = ep->plen;
+			event.private_data = ep->mpa_pkt +
+				sizeof(struct mpa_message);
+		}
 	}
 
 	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
@@ -793,9 +892,22 @@ static void connect_request_upcall(struct c4iw_ep *ep)
 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
 	event.local_addr = ep->com.local_addr;
 	event.remote_addr = ep->com.remote_addr;
-	event.private_data_len = ep->plen;
-	event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 	event.provider_data = ep;
+	if (!ep->tried_with_mpa_v1) {
+		/* this means MPA_v2 is used */
+		event.ord = ep->ord;
+		event.ird = ep->ird;
+		event.private_data_len = ep->plen -
+			sizeof(struct mpa_v2_conn_params);
+		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
+			sizeof(struct mpa_v2_conn_params);
+	} else {
+		/* this means MPA_v1 is used. Send max supported */
+		event.ord = c4iw_max_read_depth;
+		event.ird = c4iw_max_read_depth;
+		event.private_data_len = ep->plen;
+		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
+	}
 	if (state_read(&ep->parent_ep->com) != DEAD) {
 		c4iw_get_ep(&ep->com);
 		ep->parent_ep->com.cm_id->event_handler(
@@ -813,6 +925,8 @@ static void established_upcall(struct c4iw_ep *ep)
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
 	memset(&event, 0, sizeof(event));
 	event.event = IW_CM_EVENT_ESTABLISHED;
+	event.ird = ep->ird;
+	event.ord = ep->ord;
 	if (ep->com.cm_id) {
 		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
@@ -848,7 +962,10 @@ static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
 static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
 {
 	struct mpa_message *mpa;
+	struct mpa_v2_conn_params *mpa_v2_params;
 	u16 plen;
+	u16 resp_ird, resp_ord;
+	u8 rtr_mismatch = 0, insuff_ird = 0;
 	struct c4iw_qp_attributes attrs;
 	enum c4iw_qp_attr_mask mask;
 	int err;
@@ -888,7 +1005,9 @@ static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
 	mpa = (struct mpa_message *) ep->mpa_pkt;
 
 	/* Validate MPA header. */
-	if (mpa->revision != mpa_rev) {
+	if (mpa->revision > mpa_rev) {
+		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
+		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
 		err = -EPROTO;
 		goto err;
 	}
@@ -938,13 +1057,66 @@ static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
-	ep->mpa_attr.version = mpa_rev;
-	ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
-					    FW_RI_INIT_P2PTYPE_DISABLED;
+	ep->mpa_attr.version = mpa->revision;
+	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
+
+	if (mpa->revision == 2) {
+		ep->mpa_attr.enhanced_rdma_conn =
+			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
+		if (ep->mpa_attr.enhanced_rdma_conn) {
+			mpa_v2_params = (struct mpa_v2_conn_params *)
+				(ep->mpa_pkt + sizeof(*mpa));
+			resp_ird = ntohs(mpa_v2_params->ird) &
+				MPA_V2_IRD_ORD_MASK;
+			resp_ord = ntohs(mpa_v2_params->ord) &
+				MPA_V2_IRD_ORD_MASK;
+
+			/*
+			 * This is a double-check. Ideally, below checks are
+			 * not required since ird/ord stuff has been taken
+			 * care of in c4iw_accept_cr
+			 */
+			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
+				err = -ENOMEM;
+				ep->ird = resp_ord;
+				ep->ord = resp_ird;
+				insuff_ird = 1;
+			}
+
+			if (ntohs(mpa_v2_params->ird) &
+					MPA_V2_PEER2PEER_MODEL) {
+				if (ntohs(mpa_v2_params->ord) &
+						MPA_V2_RDMA_WRITE_RTR)
+					ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
+				else if (ntohs(mpa_v2_params->ord) &
+						MPA_V2_RDMA_READ_RTR)
+					ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_READ_REQ;
+			}
+		}
+	} else if (mpa->revision == 1)
+		if (peer2peer)
+			ep->mpa_attr.p2p_type = p2p_type;
+
 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
-	     "xmit_marker_enabled=%d, version=%d\n", __func__,
-	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
-	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
+	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
+	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
+	     ep->mpa_attr.recv_marker_enabled,
+	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
+	     ep->mpa_attr.p2p_type, p2p_type);
+
+	/*
+	 * If responder's RTR does not match with that of initiator, assign
+	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
+	 * generated when moving QP to RTS state.
+	 * A TERM message will be sent after QP has moved to RTS state
+	 */
+	if ((ep->mpa_attr.version == 2) &&
+			(ep->mpa_attr.p2p_type != p2p_type)) {
+		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
+		rtr_mismatch = 1;
+	}
 
 	attrs.mpa_attr = ep->mpa_attr;
 	attrs.max_ird = ep->ird;
@@ -961,6 +1133,39 @@ static void process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
 			     ep->com.qp, mask, &attrs, 1);
 	if (err)
 		goto err;
+
+	/*
+	 * If responder's RTR requirement did not match with what initiator
+	 * supports, generate TERM message
+	 */
+	if (rtr_mismatch) {
+		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
+		attrs.layer_etype = LAYER_MPA | DDP_LLP;
+		attrs.ecode = MPA_NOMATCH_RTR;
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
+		err = -ENOMEM;
+		goto out;
+	}
+
+	/*
+	 * Generate TERM if initiator IRD is not sufficient for responder
+	 * provided ORD. Currently, we do the same behaviour even when
+	 * responder provided IRD is also not sufficient as regards to
+	 * initiator ORD.
+	 */
+	if (insuff_ird) {
+		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
+				__func__);
+		attrs.layer_etype = LAYER_MPA | DDP_LLP;
+		attrs.ecode = MPA_INSUFF_IRD;
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
+		err = -ENOMEM;
+		goto out;
+	}
 	goto out;
 err:
 	state_set(&ep->com, ABORTING);
@@ -973,6 +1178,7 @@ out:
 static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
 {
 	struct mpa_message *mpa;
+	struct mpa_v2_conn_params *mpa_v2_params;
 	u16 plen;
 
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
@@ -1013,7 +1219,9 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
 	/*
 	 * Validate MPA Header.
 	 */
-	if (mpa->revision != mpa_rev) {
+	if (mpa->revision > mpa_rev) {
+		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
+		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
 		abort_connection(ep, skb, GFP_KERNEL);
 		return;
 	}
@@ -1056,9 +1264,37 @@ static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
-	ep->mpa_attr.version = mpa_rev;
-	ep->mpa_attr.p2p_type = peer2peer ? p2p_type :
-					    FW_RI_INIT_P2PTYPE_DISABLED;
+	ep->mpa_attr.version = mpa->revision;
+	if (mpa->revision == 1)
+		ep->tried_with_mpa_v1 = 1;
+	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
+
+	if (mpa->revision == 2) {
+		ep->mpa_attr.enhanced_rdma_conn =
+			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
+		if (ep->mpa_attr.enhanced_rdma_conn) {
+			mpa_v2_params = (struct mpa_v2_conn_params *)
+				(ep->mpa_pkt + sizeof(*mpa));
+			ep->ird = ntohs(mpa_v2_params->ird) &
+				MPA_V2_IRD_ORD_MASK;
+			ep->ord = ntohs(mpa_v2_params->ord) &
+				MPA_V2_IRD_ORD_MASK;
+			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
+				if (peer2peer) {
+					if (ntohs(mpa_v2_params->ord) &
+							MPA_V2_RDMA_WRITE_RTR)
+						ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
+					else if (ntohs(mpa_v2_params->ord) &
+							MPA_V2_RDMA_READ_RTR)
+						ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_READ_REQ;
+				}
+		}
+	} else if (mpa->revision == 1)
+		if (peer2peer)
+			ep->mpa_attr.p2p_type = p2p_type;
+
 	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
 	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
 	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
@@ -1550,6 +1786,112 @@ static int is_neg_adv_abort(unsigned int status)
 	       status == CPL_ERR_PERSIST_NEG_ADVICE;
 }
 
+static int c4iw_reconnect(struct c4iw_ep *ep)
+{
+	int err = 0;
+	struct rtable *rt;
+	struct net_device *pdev;
+	struct neighbour *neigh;
+	int step;
+
+	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
+	init_timer(&ep->timer);
+
+	/*
+	 * Allocate an active TID to initiate a TCP connection.
+	 */
+	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
+	if (ep->atid == -1) {
+		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
+		err = -ENOMEM;
+		goto fail2;
+	}
+
+	/* find a route */
+	rt = find_route(ep->com.dev,
+			ep->com.cm_id->local_addr.sin_addr.s_addr,
+			ep->com.cm_id->remote_addr.sin_addr.s_addr,
+			ep->com.cm_id->local_addr.sin_port,
+			ep->com.cm_id->remote_addr.sin_port, 0);
+	if (!rt) {
+		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
+		err = -EHOSTUNREACH;
+		goto fail3;
+	}
+	ep->dst = &rt->dst;
+
+	neigh = dst_get_neighbour(ep->dst);
+
+	/* get a l2t entry */
+	if (neigh->dev->flags & IFF_LOOPBACK) {
+		PDBG("%s LOOPBACK\n", __func__);
+		pdev = ip_dev_find(&init_net,
+				   ep->com.cm_id->remote_addr.sin_addr.s_addr);
+		ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
+					neigh, pdev, 0);
+		ep->mtu = pdev->mtu;
+		ep->tx_chan = cxgb4_port_chan(pdev);
+		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
+		step = ep->com.dev->rdev.lldi.ntxq /
+			ep->com.dev->rdev.lldi.nchan;
+		ep->txq_idx = cxgb4_port_idx(pdev) * step;
+		step = ep->com.dev->rdev.lldi.nrxq /
+			ep->com.dev->rdev.lldi.nchan;
+		ep->ctrlq_idx = cxgb4_port_idx(pdev);
+		ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
+			cxgb4_port_idx(pdev) * step];
+		dev_put(pdev);
+	} else {
+		ep->l2t = cxgb4_l2t_get(ep->com.dev->rdev.lldi.l2t,
+					neigh, neigh->dev, 0);
+		ep->mtu = dst_mtu(ep->dst);
+		ep->tx_chan = cxgb4_port_chan(neigh->dev);
+		ep->smac_idx = (cxgb4_port_viid(neigh->dev) & 0x7F) << 1;
+		step = ep->com.dev->rdev.lldi.ntxq /
+			ep->com.dev->rdev.lldi.nchan;
+		ep->txq_idx = cxgb4_port_idx(neigh->dev) * step;
+		ep->ctrlq_idx = cxgb4_port_idx(neigh->dev);
+		step = ep->com.dev->rdev.lldi.nrxq /
+			ep->com.dev->rdev.lldi.nchan;
+		ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
+			cxgb4_port_idx(neigh->dev) * step];
+	}
+	if (!ep->l2t) {
+		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
+		err = -ENOMEM;
+		goto fail4;
+	}
+
+	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
+	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
+	     ep->l2t->idx);
+
+	state_set(&ep->com, CONNECTING);
+	ep->tos = 0;
+
+	/* send connect request to rnic */
+	err = send_connect(ep);
+	if (!err)
+		goto out;
+
+	cxgb4_l2t_release(ep->l2t);
+fail4:
+	dst_release(ep->dst);
+fail3:
+	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
+fail2:
+	/*
+	 * remember to send notification to upper layer.
+	 * We are in here so the upper layer is not aware that this is
+	 * re-connect attempt and so, upper layer is still waiting for
+	 * response of 1st connect request.
+	 */
+	connect_reply_upcall(ep, -ECONNRESET);
+	c4iw_put_ep(&ep->com);
+out:
+	return err;
+}
+
 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 {
 	struct cpl_abort_req_rss *req = cplhdr(skb);
@@ -1573,8 +1915,11 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 
 	/*
 	 * Wake up any threads in rdma_init() or rdma_fini().
+	 * However, this is not needed if com state is just
+	 * MPA_REQ_SENT
 	 */
-	c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+	if (ep->com.state != MPA_REQ_SENT)
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
 
 	mutex_lock(&ep->com.mutex);
 	switch (ep->com.state) {
@@ -1585,7 +1930,21 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 		break;
 	case MPA_REQ_SENT:
 		stop_ep_timer(ep);
-		connect_reply_upcall(ep, -ECONNRESET);
+		if (mpa_rev == 2 && ep->tried_with_mpa_v1)
+			connect_reply_upcall(ep, -ECONNRESET);
+		else {
+			/*
+			 * we just don't send notification upwards because we
+			 * want to retry with mpa_v1 without upper layers even
+			 * knowing it.
+			 *
+			 * do some housekeeping so as to re-initiate the
+			 * connection
+			 */
+			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
+			     mpa_rev);
+			ep->retry_with_mpa_v1 = 1;
+		}
 		break;
 	case MPA_REP_SENT:
 		break;
@@ -1621,7 +1980,9 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 	dst_confirm(ep->dst);
 	if (ep->com.state != ABORTING) {
 		__state_set(&ep->com, DEAD);
-		release = 1;
+		/* we don't release if we want to retry with mpa_v1 */
+		if (!ep->retry_with_mpa_v1)
+			release = 1;
 	}
 	mutex_unlock(&ep->com.mutex);
 
@@ -1641,6 +2002,15 @@ static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
 out:
 	if (release)
 		release_ep_resources(ep);
+
+	/* retry with mpa-v1 */
+	if (ep && ep->retry_with_mpa_v1) {
+		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
+		dst_release(ep->dst);
+		cxgb4_l2t_release(ep->l2t);
+		c4iw_reconnect(ep);
+	}
+
 	return 0;
 }
 
@@ -1792,18 +2162,40 @@ int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		goto err;
 	}
 
-	cm_id->add_ref(cm_id);
-	ep->com.cm_id = cm_id;
-	ep->com.qp = qp;
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
+		if (conn_param->ord > ep->ird) {
+			ep->ird = conn_param->ird;
+			ep->ord = conn_param->ord;
+			send_mpa_reject(ep, conn_param->private_data,
+					conn_param->private_data_len);
+			abort_connection(ep, NULL, GFP_KERNEL);
+			err = -ENOMEM;
+			goto err;
+		}
+		if (conn_param->ird > ep->ord) {
+			if (!ep->ord)
+				conn_param->ird = 1;
+			else {
+				abort_connection(ep, NULL, GFP_KERNEL);
+				err = -ENOMEM;
+				goto err;
+			}
+		}
 
+	}
 	ep->ird = conn_param->ird;
 	ep->ord = conn_param->ord;
 
-	if (peer2peer && ep->ird == 0)
-		ep->ird = 1;
+	if (ep->mpa_attr.version != 2)
+		if (peer2peer && ep->ird == 0)
+			ep->ird = 1;
 
 	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
 
+	cm_id->add_ref(cm_id);
+	ep->com.cm_id = cm_id;
+	ep->com.qp = qp;
+
 	/* bind QP to EP and move to RTS */
 	attrs.mpa_attr = ep->mpa_attr;
 	attrs.max_ird = ep->ird;
@@ -1944,6 +2336,8 @@ int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		       ep->com.dev->rdev.lldi.nchan;
 		ep->rss_qid = ep->com.dev->rdev.lldi.rxq_ids[
 			      cxgb4_port_idx(neigh->dev) * step];
+		ep->retry_with_mpa_v1 = 0;
+		ep->tried_with_mpa_v1 = 0;
 	}
 	if (!ep->l2t) {
 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
@@ -2323,8 +2717,11 @@ static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
 
 	/*
 	 * Wake up any threads in rdma_init() or rdma_fini().
+	 * However, this is not needed if com state is just
+	 * MPA_REQ_SENT
 	 */
-	c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+	if (ep->com.state != MPA_REQ_SENT)
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
 	sched(dev, skb);
 	return 0;
 }

drivers/infiniband/hw/cxgb4/cq.c

@@ -185,7 +185,7 @@ static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
 				 V_CQE_OPCODE(FW_RI_SEND) |
 				 V_CQE_TYPE(0) |
 				 V_CQE_SWCQE(1) |
-				 V_CQE_QPID(wq->rq.qid));
+				 V_CQE_QPID(wq->sq.qid));
 	cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
 	cq->sw_queue[cq->sw_pidx] = cqe;
 	t4_swcq_produce(cq);
@@ -818,6 +818,7 @@ struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
 	chp->cq.size--;				/* status page */
 	chp->ibcq.cqe = entries - 2;
 	spin_lock_init(&chp->lock);
+	spin_lock_init(&chp->comp_handler_lock);
 	atomic_set(&chp->refcnt, 1);
 	init_waitqueue_head(&chp->wait);
 	ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);

drivers/infiniband/hw/cxgb4/device.c

@@ -376,10 +376,8 @@ struct uld_ctx {
 	struct c4iw_dev *dev;
 };
 
-static void c4iw_remove(struct uld_ctx *ctx)
+static void c4iw_dealloc(struct uld_ctx *ctx)
 {
-	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
-	c4iw_unregister_device(ctx->dev);
 	c4iw_rdev_close(&ctx->dev->rdev);
 	idr_destroy(&ctx->dev->cqidr);
 	idr_destroy(&ctx->dev->qpidr);
@@ -389,11 +387,30 @@ static void c4iw_remove(struct uld_ctx *ctx)
 	ctx->dev = NULL;
 }
 
+static void c4iw_remove(struct uld_ctx *ctx)
+{
+	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
+	c4iw_unregister_device(ctx->dev);
+	c4iw_dealloc(ctx);
+}
+
+static int rdma_supported(const struct cxgb4_lld_info *infop)
+{
+	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
+	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
+	       infop->vr->cq.size > 0 && infop->vr->ocq.size > 0;
+}
+
 static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
 {
 	struct c4iw_dev *devp;
 	int ret;
 
+	if (!rdma_supported(infop)) {
+		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
+		       pci_name(infop->pdev));
+		return ERR_PTR(-ENOSYS);
+	}
 	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
 	if (!devp) {
 		printk(KERN_ERR MOD "Cannot allocate ib device\n");
@@ -414,7 +431,6 @@ static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
 
 	ret = c4iw_rdev_open(&devp->rdev);
 	if (ret) {
-		mutex_unlock(&dev_mutex);
 		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
 		ib_dealloc_device(&devp->ibdev);
 		return ERR_PTR(ret);
@@ -519,15 +535,24 @@ static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
 	case CXGB4_STATE_UP:
 		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
 		if (!ctx->dev) {
-			int ret = 0;
+			int ret;
 
 			ctx->dev = c4iw_alloc(&ctx->lldi);
-			if (!IS_ERR(ctx->dev))
-				ret = c4iw_register_device(ctx->dev);
-			if (IS_ERR(ctx->dev) || ret)
+			if (IS_ERR(ctx->dev)) {
+				printk(KERN_ERR MOD
+				       "%s: initialization failed: %ld\n",
+				       pci_name(ctx->lldi.pdev),
+				       PTR_ERR(ctx->dev));
+				ctx->dev = NULL;
+				break;
+			}
+			ret = c4iw_register_device(ctx->dev);
+			if (ret) {
 				printk(KERN_ERR MOD
 				       "%s: RDMA registration failed: %d\n",
 				       pci_name(ctx->lldi.pdev), ret);
+				c4iw_dealloc(ctx);
+			}
 		}
 		break;
 	case CXGB4_STATE_DOWN:

drivers/infiniband/hw/cxgb4/ev.c

@@ -42,6 +42,7 @@ static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
 {
 	struct ib_event event;
 	struct c4iw_qp_attributes attrs;
+	unsigned long flag;
 
 	if ((qhp->attr.state == C4IW_QP_STATE_ERROR) ||
 	    (qhp->attr.state == C4IW_QP_STATE_TERMINATE)) {
@@ -72,7 +73,9 @@ static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
 	if (qhp->ibqp.event_handler)
 		(*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
 
+	spin_lock_irqsave(&chp->comp_handler_lock, flag);
 	(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+	spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 }
 
 void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe)
@@ -183,11 +186,14 @@ out:
 int c4iw_ev_handler(struct c4iw_dev *dev, u32 qid)
 {
 	struct c4iw_cq *chp;
+	unsigned long flag;
 
 	chp = get_chp(dev, qid);
-	if (chp)
+	if (chp) {
+		spin_lock_irqsave(&chp->comp_handler_lock, flag);
 		(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
-	else
+		spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
+	} else
 		PDBG("%s unknown cqid 0x%x\n", __func__, qid);
 	return 0;
 }

drivers/infiniband/hw/cxgb4/iw_cxgb4.h

@@ -309,6 +309,7 @@ struct c4iw_cq {
 	struct c4iw_dev *rhp;
 	struct t4_cq cq;
 	spinlock_t lock;
+	spinlock_t comp_handler_lock;
 	atomic_t refcnt;
 	wait_queue_head_t wait;
 };
@@ -323,6 +324,7 @@ struct c4iw_mpa_attributes {
 	u8 recv_marker_enabled;
 	u8 xmit_marker_enabled;
 	u8 crc_enabled;
+	u8 enhanced_rdma_conn;
 	u8 version;
 	u8 p2p_type;
 };
@@ -349,6 +351,8 @@ struct c4iw_qp_attributes {
 	u8 is_terminate_local;
 	struct c4iw_mpa_attributes mpa_attr;
 	struct c4iw_ep *llp_stream_handle;
+	u8 layer_etype;
+	u8 ecode;
 };
 
 struct c4iw_qp {
@@ -501,11 +505,18 @@ enum c4iw_mmid_state {
 #define MPA_KEY_REP "MPA ID Rep Frame"
 
 #define MPA_MAX_PRIVATE_DATA	256
+#define MPA_ENHANCED_RDMA_CONN	0x10
 #define MPA_REJECT		0x20
 #define MPA_CRC			0x40
 #define MPA_MARKERS		0x80
 #define MPA_FLAGS_MASK		0xE0
 
+#define MPA_V2_PEER2PEER_MODEL          0x8000
+#define MPA_V2_ZERO_LEN_FPDU_RTR        0x4000
+#define MPA_V2_RDMA_WRITE_RTR           0x8000
+#define MPA_V2_RDMA_READ_RTR            0x4000
+#define MPA_V2_IRD_ORD_MASK             0x3FFF
+
 #define c4iw_put_ep(ep) { \
 	PDBG("put_ep (via %s:%u) ep %p refcnt %d\n", __func__, __LINE__,  \
 	     ep, atomic_read(&((ep)->kref.refcount))); \
@@ -528,6 +539,11 @@ struct mpa_message {
 	u8 private_data[0];
 };
 
+struct mpa_v2_conn_params {
+	__be16 ird;
+	__be16 ord;
+};
+
 struct terminate_message {
 	u8 layer_etype;
 	u8 ecode;
@@ -580,7 +596,10 @@ enum c4iw_ddp_ecodes {
 
 enum c4iw_mpa_ecodes {
 	MPA_CRC_ERR		= 0x02,
-	MPA_MARKER_ERR		= 0x03
+	MPA_MARKER_ERR          = 0x03,
+	MPA_LOCAL_CATA          = 0x05,
+	MPA_INSUFF_IRD          = 0x06,
+	MPA_NOMATCH_RTR         = 0x07,
 };
 
 enum c4iw_ep_state {
@@ -651,6 +670,8 @@ struct c4iw_ep {
 	u16 txq_idx;
 	u16 ctrlq_idx;
 	u8 tos;
+	u8 retry_with_mpa_v1;
+	u8 tried_with_mpa_v1;
 };
 
 static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id)

drivers/infiniband/hw/cxgb4/qp.c

@@ -917,7 +917,11 @@ static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe,
 	wqe->u.terminate.type = FW_RI_TYPE_TERMINATE;
 	wqe->u.terminate.immdlen = cpu_to_be32(sizeof *term);
 	term = (struct terminate_message *)wqe->u.terminate.termmsg;
-	build_term_codes(err_cqe, &term->layer_etype, &term->ecode);
+	if (qhp->attr.layer_etype == (LAYER_MPA|DDP_LLP)) {
+		term->layer_etype = qhp->attr.layer_etype;
+		term->ecode = qhp->attr.ecode;
+	} else
+		build_term_codes(err_cqe, &term->layer_etype, &term->ecode);
 	c4iw_ofld_send(&qhp->rhp->rdev, skb);
 }
 
@@ -941,8 +945,11 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
 	flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
 	spin_unlock(&qhp->lock);
 	spin_unlock_irqrestore(&rchp->lock, flag);
-	if (flushed)
+	if (flushed) {
+		spin_lock_irqsave(&rchp->comp_handler_lock, flag);
 		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
+	}
 
 	/* locking hierarchy: cq lock first, then qp lock. */
 	spin_lock_irqsave(&schp->lock, flag);
@@ -952,13 +959,17 @@ static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
 	flushed = c4iw_flush_sq(&qhp->wq, &schp->cq, count);
 	spin_unlock(&qhp->lock);
 	spin_unlock_irqrestore(&schp->lock, flag);
-	if (flushed)
+	if (flushed) {
+		spin_lock_irqsave(&schp->comp_handler_lock, flag);
 		(*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
+		spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
+	}
 }
 
 static void flush_qp(struct c4iw_qp *qhp)
 {
 	struct c4iw_cq *rchp, *schp;
+	unsigned long flag;
 
 	rchp = get_chp(qhp->rhp, qhp->attr.rcq);
 	schp = get_chp(qhp->rhp, qhp->attr.scq);
@@ -966,8 +977,16 @@ static void flush_qp(struct c4iw_qp *qhp)
 	if (qhp->ibqp.uobject) {
 		t4_set_wq_in_error(&qhp->wq);
 		t4_set_cq_in_error(&rchp->cq);
-		if (schp != rchp)
+		spin_lock_irqsave(&rchp->comp_handler_lock, flag);
+		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
+		if (schp != rchp) {
 			t4_set_cq_in_error(&schp->cq);
+			spin_lock_irqsave(&schp->comp_handler_lock, flag);
+			(*schp->ibcq.comp_handler)(&schp->ibcq,
+					schp->ibcq.cq_context);
+			spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
+		}
 		return;
 	}
 	__flush_qp(qhp, rchp, schp);
@@ -1012,6 +1031,7 @@ out:
 
 static void build_rtr_msg(u8 p2p_type, struct fw_ri_init *init)
 {
+	PDBG("%s p2p_type = %d\n", __func__, p2p_type);
 	memset(&init->u, 0, sizeof init->u);
 	switch (p2p_type) {
 	case FW_RI_INIT_P2PTYPE_RDMA_WRITE:
@@ -1206,12 +1226,16 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 				disconnect = 1;
 				c4iw_get_ep(&qhp->ep->com);
 			}
+			if (qhp->ibqp.uobject)
+				t4_set_wq_in_error(&qhp->wq);
 			ret = rdma_fini(rhp, qhp, ep);
 			if (ret)
 				goto err;
 			break;
 		case C4IW_QP_STATE_TERMINATE:
 			set_state(qhp, C4IW_QP_STATE_TERMINATE);
+			qhp->attr.layer_etype = attrs->layer_etype;
+			qhp->attr.ecode = attrs->ecode;
 			if (qhp->ibqp.uobject)
 				t4_set_wq_in_error(&qhp->wq);
 			ep = qhp->ep;
@@ -1222,6 +1246,8 @@ int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
 			break;
 		case C4IW_QP_STATE_ERROR:
 			set_state(qhp, C4IW_QP_STATE_ERROR);
+			if (qhp->ibqp.uobject)
+				t4_set_wq_in_error(&qhp->wq);
 			if (!internal) {
 				abort = 1;
 				disconnect = 1;
@@ -1334,7 +1360,10 @@ int c4iw_destroy_qp(struct ib_qp *ib_qp)
 	rhp = qhp->rhp;
 
 	attrs.next_state = C4IW_QP_STATE_ERROR;
-	c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
+	if (qhp->attr.state == C4IW_QP_STATE_TERMINATE)
+		c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+	else
+		c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
 	wait_event(qhp->wait, !qhp->ep);
 
 	remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid);

drivers/infiniband/hw/ehca/ehca_eq.c

@@ -125,7 +125,7 @@ int ehca_create_eq(struct ehca_shca *shca,
 		tasklet_init(&eq->interrupt_task, ehca_tasklet_eq, (long)shca);
 
 		ret = ibmebus_request_irq(eq->ist, ehca_interrupt_eq,
-					  IRQF_DISABLED, "ehca_eq",
+					  0, "ehca_eq",
 					  (void *)shca);
 		if (ret < 0)
 			ehca_err(ib_dev, "Can't map interrupt handler.");
@@ -133,7 +133,7 @@ int ehca_create_eq(struct ehca_shca *shca,
 		tasklet_init(&eq->interrupt_task, ehca_tasklet_neq, (long)shca);
 
 		ret = ibmebus_request_irq(eq->ist, ehca_interrupt_neq,
-					  IRQF_DISABLED, "ehca_neq",
+					  0, "ehca_neq",
 					  (void *)shca);
 		if (ret < 0)
 			ehca_err(ib_dev, "Can't map interrupt handler.");

drivers/infiniband/hw/ehca/ehca_qp.c

@@ -977,6 +977,9 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd,
 	struct hcp_modify_qp_control_block *mqpcb;
 	u64 hret, update_mask;
 
+	if (srq_init_attr->srq_type != IB_SRQT_BASIC)
+		return ERR_PTR(-ENOSYS);
+
 	/* For common attributes, internal_create_qp() takes its info
 	 * out of qp_init_attr, so copy all common attrs there.
 	 */

drivers/infiniband/hw/ipath/ipath_init_chip.c

@@ -34,6 +34,7 @@
 #include <linux/pci.h>
 #include <linux/netdevice.h>
 #include <linux/slab.h>
+#include <linux/stat.h>
 #include <linux/vmalloc.h>
 
 #include "ipath_kernel.h"

drivers/infiniband/hw/ipath/ipath_srq.c

@@ -107,6 +107,11 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
 	u32 sz;
 	struct ib_srq *ret;
 
+	if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
+		ret = ERR_PTR(-ENOSYS);
+		goto done;
+	}
+
 	if (srq_init_attr->attr.max_wr == 0) {
 		ret = ERR_PTR(-EINVAL);
 		goto done;

drivers/infiniband/hw/mlx4/main.c

@@ -128,6 +128,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
+		props->device_cap_flags |= IB_DEVICE_XRC;
 
 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 		0xffffff;
@@ -181,8 +183,12 @@ mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 
 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 			      struct ib_port_attr *props,
+			      struct ib_smp *in_mad,
 			      struct ib_smp *out_mad)
 {
+	int ext_active_speed;
+	int err;
+
 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
 	props->lmc		= out_mad->data[34] & 0x7;
 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
@@ -203,6 +209,39 @@ static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 	props->max_vl_num	= out_mad->data[37] >> 4;
 	props->init_type_reply	= out_mad->data[41] >> 4;
 
+	/* Check if extended speeds (EDR/FDR/...) are supported */
+	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
+		ext_active_speed = out_mad->data[62] >> 4;
+
+		switch (ext_active_speed) {
+		case 1:
+			props->active_speed = 16; /* FDR */
+			break;
+		case 2:
+			props->active_speed = 32; /* EDR */
+			break;
+		}
+	}
+
+	/* If reported active speed is QDR, check if is FDR-10 */
+	if (props->active_speed == 4) {
+		if (to_mdev(ibdev)->dev->caps.ext_port_cap[port] &
+		    MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
+			init_query_mad(in_mad);
+			in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
+			in_mad->attr_mod = cpu_to_be32(port);
+
+			err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port,
+					   NULL, NULL, in_mad, out_mad);
+			if (err)
+				return err;
+
+			/* Checking LinkSpeedActive for FDR-10 */
+			if (out_mad->data[15] & 0x1)
+				props->active_speed = 8;
+		}
+	}
+
 	return 0;
 }
 
@@ -227,7 +266,7 @@ static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 	props->pkey_tbl_len	= 1;
 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
 	props->qkey_viol_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 48));
-	props->max_mtu		= IB_MTU_2048;
+	props->max_mtu		= IB_MTU_4096;
 	props->subnet_timeout	= 0;
 	props->max_vl_num	= out_mad->data[37] >> 4;
 	props->init_type_reply	= 0;
@@ -274,7 +313,7 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 		goto out;
 
 	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
-		ib_link_query_port(ibdev, port, props, out_mad) :
+		ib_link_query_port(ibdev, port, props, in_mad, out_mad) :
 		eth_link_query_port(ibdev, port, props, out_mad);
 
 out:
@@ -566,6 +605,57 @@ static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 	return 0;
 }
 
+static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata)
+{
+	struct mlx4_ib_xrcd *xrcd;
+	int err;
+
+	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+		return ERR_PTR(-ENOSYS);
+
+	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
+	if (!xrcd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
+	if (err)
+		goto err1;
+
+	xrcd->pd = ib_alloc_pd(ibdev);
+	if (IS_ERR(xrcd->pd)) {
+		err = PTR_ERR(xrcd->pd);
+		goto err2;
+	}
+
+	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
+	if (IS_ERR(xrcd->cq)) {
+		err = PTR_ERR(xrcd->cq);
+		goto err3;
+	}
+
+	return &xrcd->ibxrcd;
+
+err3:
+	ib_dealloc_pd(xrcd->pd);
+err2:
+	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
+err1:
+	kfree(xrcd);
+	return ERR_PTR(err);
+}
+
+static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
+{
+	ib_destroy_cq(to_mxrcd(xrcd)->cq);
+	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
+	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
+	kfree(xrcd);
+
+	return 0;
+}
+
 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 {
 	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
@@ -1044,7 +1134,9 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
-		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
+		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
 
 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
@@ -1093,6 +1185,14 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
 	ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
 
+	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
+		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
+		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
+		ibdev->ib_dev.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
+			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
+	}
+
 	spin_lock_init(&iboe->lock);
 
 	if (init_node_data(ibdev))

drivers/infiniband/hw/mlx4/mlx4_ib.h

@@ -56,6 +56,13 @@ struct mlx4_ib_pd {
 	u32			pdn;
 };
 
+struct mlx4_ib_xrcd {
+	struct ib_xrcd		ibxrcd;
+	u32			xrcdn;
+	struct ib_pd	       *pd;
+	struct ib_cq	       *cq;
+};
+
 struct mlx4_ib_cq_buf {
 	struct mlx4_buf		buf;
 	struct mlx4_mtt		mtt;
@@ -138,6 +145,7 @@ struct mlx4_ib_qp {
 	struct mlx4_mtt		mtt;
 	int			buf_size;
 	struct mutex		mutex;
+	u16			xrcdn;
 	u32			flags;
 	u8			port;
 	u8			alt_port;
@@ -211,6 +219,11 @@ static inline struct mlx4_ib_pd *to_mpd(struct ib_pd *ibpd)
 	return container_of(ibpd, struct mlx4_ib_pd, ibpd);
 }
 
+static inline struct mlx4_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
+{
+	return container_of(ibxrcd, struct mlx4_ib_xrcd, ibxrcd);
+}
+
 static inline struct mlx4_ib_cq *to_mcq(struct ib_cq *ibcq)
 {
 	return container_of(ibcq, struct mlx4_ib_cq, ibcq);

drivers/infiniband/hw/mlx4/qp.c

@@ -302,15 +302,14 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
 }
 
 static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
-		       int is_user, int has_srq, struct mlx4_ib_qp *qp)
+		       int is_user, int has_rq, struct mlx4_ib_qp *qp)
 {
 	/* Sanity check RQ size before proceeding */
 	if (cap->max_recv_wr  > dev->dev->caps.max_wqes  ||
 	    cap->max_recv_sge > dev->dev->caps.max_rq_sg)
 		return -EINVAL;
 
-	if (has_srq) {
-		/* QPs attached to an SRQ should have no RQ */
+	if (!has_rq) {
 		if (cap->max_recv_wr)
 			return -EINVAL;
 
@@ -463,6 +462,14 @@ static int set_user_sq_size(struct mlx4_ib_dev *dev,
 	return 0;
 }
 
+static int qp_has_rq(struct ib_qp_init_attr *attr)
+{
+	if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT)
+		return 0;
+
+	return !attr->srq;
+}
+
 static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
 			    struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp)
@@ -479,7 +486,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
 		qp->sq_signal_bits = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
 
-	err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, !!init_attr->srq, qp);
+	err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, qp_has_rq(init_attr), qp);
 	if (err)
 		goto err;
 
@@ -513,7 +520,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			goto err_mtt;
 
-		if (!init_attr->srq) {
+		if (qp_has_rq(init_attr)) {
 			err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
 						  ucmd.db_addr, &qp->db);
 			if (err)
@@ -532,7 +539,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			goto err;
 
-		if (!init_attr->srq) {
+		if (qp_has_rq(init_attr)) {
 			err = mlx4_db_alloc(dev->dev, &qp->db, 0);
 			if (err)
 				goto err;
@@ -575,6 +582,9 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	if (err)
 		goto err_qpn;
 
+	if (init_attr->qp_type == IB_QPT_XRC_TGT)
+		qp->mqp.qpn |= (1 << 23);
+
 	/*
 	 * Hardware wants QPN written in big-endian order (after
 	 * shifting) for send doorbell.  Precompute this value to save
@@ -592,9 +602,8 @@ err_qpn:
 
 err_wrid:
 	if (pd->uobject) {
-		if (!init_attr->srq)
-			mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context),
-					      &qp->db);
+		if (qp_has_rq(init_attr))
+			mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &qp->db);
 	} else {
 		kfree(qp->sq.wrid);
 		kfree(qp->rq.wrid);
@@ -610,7 +619,7 @@ err_buf:
 		mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
 
 err_db:
-	if (!pd->uobject && !init_attr->srq)
+	if (!pd->uobject && qp_has_rq(init_attr))
 		mlx4_db_free(dev->dev, &qp->db);
 
 err:
@@ -671,6 +680,33 @@ static void del_gid_entries(struct mlx4_ib_qp *qp)
 	}
 }
 
+static struct mlx4_ib_pd *get_pd(struct mlx4_ib_qp *qp)
+{
+	if (qp->ibqp.qp_type == IB_QPT_XRC_TGT)
+		return to_mpd(to_mxrcd(qp->ibqp.xrcd)->pd);
+	else
+		return to_mpd(qp->ibqp.pd);
+}
+
+static void get_cqs(struct mlx4_ib_qp *qp,
+		    struct mlx4_ib_cq **send_cq, struct mlx4_ib_cq **recv_cq)
+{
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_XRC_TGT:
+		*send_cq = to_mcq(to_mxrcd(qp->ibqp.xrcd)->cq);
+		*recv_cq = *send_cq;
+		break;
+	case IB_QPT_XRC_INI:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = *send_cq;
+		break;
+	default:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = to_mcq(qp->ibqp.recv_cq);
+		break;
+	}
+}
+
 static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 			      int is_user)
 {
@@ -682,8 +718,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 			printk(KERN_WARNING "mlx4_ib: modify QP %06x to RESET failed.\n",
 			       qp->mqp.qpn);
 
-	send_cq = to_mcq(qp->ibqp.send_cq);
-	recv_cq = to_mcq(qp->ibqp.recv_cq);
+	get_cqs(qp, &send_cq, &recv_cq);
 
 	mlx4_ib_lock_cqs(send_cq, recv_cq);
 
@@ -706,7 +741,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 	mlx4_mtt_cleanup(dev->dev, &qp->mtt);
 
 	if (is_user) {
-		if (!qp->ibqp.srq)
+		if (qp->rq.wqe_cnt)
 			mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context),
 					      &qp->db);
 		ib_umem_release(qp->umem);
@@ -714,7 +749,7 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 		kfree(qp->sq.wrid);
 		kfree(qp->rq.wrid);
 		mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
-		if (!qp->ibqp.srq)
+		if (qp->rq.wqe_cnt)
 			mlx4_db_free(dev->dev, &qp->db);
 	}
 
@@ -725,10 +760,10 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
 {
-	struct mlx4_ib_dev *dev = to_mdev(pd->device);
 	struct mlx4_ib_sqp *sqp;
 	struct mlx4_ib_qp *qp;
 	int err;
+	u16 xrcdn = 0;
 
 	/*
 	 * We only support LSO and multicast loopback blocking, and
@@ -739,10 +774,20 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 		return ERR_PTR(-EINVAL);
 
 	if (init_attr->create_flags &&
-	    (pd->uobject || init_attr->qp_type != IB_QPT_UD))
+	    (udata || init_attr->qp_type != IB_QPT_UD))
 		return ERR_PTR(-EINVAL);
 
 	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+		pd = to_mxrcd(init_attr->xrcd)->pd;
+		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+		init_attr->send_cq = to_mxrcd(init_attr->xrcd)->cq;
+		/* fall through */
+	case IB_QPT_XRC_INI:
+		if (!(to_mdev(pd->device)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+			return ERR_PTR(-ENOSYS);
+		init_attr->recv_cq = init_attr->send_cq;
+		/* fall through */
 	case IB_QPT_RC:
 	case IB_QPT_UC:
 	case IB_QPT_UD:
@@ -751,13 +796,14 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 		if (!qp)
 			return ERR_PTR(-ENOMEM);
 
-		err = create_qp_common(dev, pd, init_attr, udata, 0, qp);
+		err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata, 0, qp);
 		if (err) {
 			kfree(qp);
 			return ERR_PTR(err);
 		}
 
 		qp->ibqp.qp_num = qp->mqp.qpn;
+		qp->xrcdn = xrcdn;
 
 		break;
 	}
@@ -765,7 +811,7 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 	case IB_QPT_GSI:
 	{
 		/* Userspace is not allowed to create special QPs: */
-		if (pd->uobject)
+		if (udata)
 			return ERR_PTR(-EINVAL);
 
 		sqp = kzalloc(sizeof *sqp, GFP_KERNEL);
@@ -774,8 +820,8 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 
 		qp = &sqp->qp;
 
-		err = create_qp_common(dev, pd, init_attr, udata,
-				       dev->dev->caps.sqp_start +
+		err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
+				       to_mdev(pd->device)->dev->caps.sqp_start +
 				       (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
 				       init_attr->port_num - 1,
 				       qp);
@@ -801,11 +847,13 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp)
 {
 	struct mlx4_ib_dev *dev = to_mdev(qp->device);
 	struct mlx4_ib_qp *mqp = to_mqp(qp);
+	struct mlx4_ib_pd *pd;
 
 	if (is_qp0(dev, mqp))
 		mlx4_CLOSE_PORT(dev->dev, mqp->port);
 
-	destroy_qp_common(dev, mqp, !!qp->pd->uobject);
+	pd = get_pd(mqp);
+	destroy_qp_common(dev, mqp, !!pd->ibpd.uobject);
 
 	if (is_sqp(dev, mqp))
 		kfree(to_msqp(mqp));
@@ -821,6 +869,8 @@ static int to_mlx4_st(enum ib_qp_type type)
 	case IB_QPT_RC:		return MLX4_QP_ST_RC;
 	case IB_QPT_UC:		return MLX4_QP_ST_UC;
 	case IB_QPT_UD:		return MLX4_QP_ST_UD;
+	case IB_QPT_XRC_INI:
+	case IB_QPT_XRC_TGT:	return MLX4_QP_ST_XRC;
 	case IB_QPT_SMI:
 	case IB_QPT_GSI:	return MLX4_QP_ST_MLX;
 	default:		return -1;
@@ -959,6 +1009,8 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 {
 	struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
+	struct mlx4_ib_pd *pd;
+	struct mlx4_ib_cq *send_cq, *recv_cq;
 	struct mlx4_qp_context *context;
 	enum mlx4_qp_optpar optpar = 0;
 	int sqd_event;
@@ -1014,8 +1066,10 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		context->sq_size_stride = ilog2(qp->sq.wqe_cnt) << 3;
 	context->sq_size_stride |= qp->sq.wqe_shift - 4;
 
-	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
 		context->sq_size_stride |= !!qp->sq_no_prefetch << 7;
+		context->xrcd = cpu_to_be32((u32) qp->xrcdn);
+	}
 
 	if (qp->ibqp.uobject)
 		context->usr_page = cpu_to_be32(to_mucontext(ibqp->uobject->context)->uar.index);
@@ -1079,8 +1133,12 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		optpar |= MLX4_QP_OPTPAR_ALT_ADDR_PATH;
 	}
 
-	context->pd	    = cpu_to_be32(to_mpd(ibqp->pd)->pdn);
-	context->params1    = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
+	pd = get_pd(qp);
+	get_cqs(qp, &send_cq, &recv_cq);
+	context->pd       = cpu_to_be32(pd->pdn);
+	context->cqn_send = cpu_to_be32(send_cq->mcq.cqn);
+	context->cqn_recv = cpu_to_be32(recv_cq->mcq.cqn);
+	context->params1  = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
 
 	/* Set "fast registration enabled" for all kernel QPs */
 	if (!qp->ibqp.uobject)
@@ -1106,8 +1164,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (attr_mask & IB_QP_SQ_PSN)
 		context->next_send_psn = cpu_to_be32(attr->sq_psn);
 
-	context->cqn_send = cpu_to_be32(to_mcq(ibqp->send_cq)->mcq.cqn);
-
 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
 		if (attr->max_dest_rd_atomic)
 			context->params2 |=
@@ -1130,8 +1186,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (attr_mask & IB_QP_RQ_PSN)
 		context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
 
-	context->cqn_recv = cpu_to_be32(to_mcq(ibqp->recv_cq)->mcq.cqn);
-
 	if (attr_mask & IB_QP_QKEY) {
 		context->qkey = cpu_to_be32(attr->qkey);
 		optpar |= MLX4_QP_OPTPAR_Q_KEY;
@@ -1140,7 +1194,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (ibqp->srq)
 		context->srqn = cpu_to_be32(1 << 24 | to_msrq(ibqp->srq)->msrq.srqn);
 
-	if (!ibqp->srq && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+	if (qp->rq.wqe_cnt && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
 		context->db_rec_addr = cpu_to_be64(qp->db.dma);
 
 	if (cur_state == IB_QPS_INIT &&
@@ -1225,17 +1279,17 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	 * entries and reinitialize the QP.
 	 */
 	if (new_state == IB_QPS_RESET && !ibqp->uobject) {
-		mlx4_ib_cq_clean(to_mcq(ibqp->recv_cq), qp->mqp.qpn,
+		mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
 				 ibqp->srq ? to_msrq(ibqp->srq): NULL);
-		if (ibqp->send_cq != ibqp->recv_cq)
-			mlx4_ib_cq_clean(to_mcq(ibqp->send_cq), qp->mqp.qpn, NULL);
+		if (send_cq != recv_cq)
+			mlx4_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
 
 		qp->rq.head = 0;
 		qp->rq.tail = 0;
 		qp->sq.head = 0;
 		qp->sq.tail = 0;
 		qp->sq_next_wqe = 0;
-		if (!ibqp->srq)
+		if (qp->rq.wqe_cnt)
 			*qp->db.db  = 0;
 	}
 
@@ -1547,14 +1601,13 @@ static void set_masked_atomic_seg(struct mlx4_wqe_masked_atomic_seg *aseg,
 }
 
 static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg,
-			     struct ib_send_wr *wr, __be16 *vlan)
+			     struct ib_send_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);
-	*vlan = dseg->vlan;
 }
 
 static void set_mlx_icrc_seg(void *dseg)
@@ -1657,7 +1710,6 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	__be32 uninitialized_var(lso_hdr_sz);
 	__be32 blh;
 	int i;
-	__be16 vlan = cpu_to_be16(0xffff);
 
 	spin_lock_irqsave(&qp->sq.lock, flags);
 
@@ -1761,7 +1813,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			break;
 
 		case IB_QPT_UD:
-			set_datagram_seg(wqe, wr, &vlan);
+			set_datagram_seg(wqe, wr);
 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
 			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
 
@@ -1824,11 +1876,6 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		ctrl->fence_size = (wr->send_flags & IB_SEND_FENCE ?
 				    MLX4_WQE_CTRL_FENCE : 0) | size;
 
-		if (be16_to_cpu(vlan) < 0x1000) {
-			ctrl->ins_vlan = 1 << 6;
-			ctrl->vlan_tag = vlan;
-		}
-
 		/*
 		 * Make sure descriptor is fully written before
 		 * setting ownership bit (because HW can start

drivers/infiniband/hw/mlx4/srq.c

@@ -76,6 +76,8 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 	struct mlx4_ib_srq *srq;
 	struct mlx4_wqe_srq_next_seg *next;
 	struct mlx4_wqe_data_seg *scatter;
+	u32 cqn;
+	u16 xrcdn;
 	int desc_size;
 	int buf_size;
 	int err;
@@ -174,12 +176,18 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 		}
 	}
 
-	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt,
+	cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
+		to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
+	xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
+		to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
+		(u16) dev->dev->caps.reserved_xrcds;
+	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
 			     srq->db.dma, &srq->msrq);
 	if (err)
 		goto err_wrid;
 
 	srq->msrq.event = mlx4_ib_srq_event;
+	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
 
 	if (pd->uobject)
 		if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {

drivers/infiniband/hw/mthca/mthca_provider.c

@@ -438,6 +438,9 @@ static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
 	struct mthca_srq *srq;
 	int err;
 
+	if (init_attr->srq_type != IB_SRQT_BASIC)
+		return ERR_PTR(-ENOSYS);
+
 	srq = kmalloc(sizeof *srq, GFP_KERNEL);
 	if (!srq)
 		return ERR_PTR(-ENOMEM);

drivers/infiniband/hw/nes/Makefile

@@ -1,3 +1,3 @@
 obj-$(CONFIG_INFINIBAND_NES) += iw_nes.o
 
-iw_nes-objs := nes.o nes_hw.o nes_nic.o nes_utils.o nes_verbs.o nes_cm.o
+iw_nes-objs := nes.o nes_hw.o nes_nic.o nes_utils.o nes_verbs.o nes_cm.o nes_mgt.o

drivers/infiniband/hw/nes/nes.c

@@ -84,7 +84,7 @@ module_param(send_first, int, 0644);
 MODULE_PARM_DESC(send_first, "Send RDMA Message First on Active Connection");
 
 
-unsigned int nes_drv_opt = 0;
+unsigned int nes_drv_opt = NES_DRV_OPT_DISABLE_INT_MOD | NES_DRV_OPT_ENABLE_PAU;
 module_param(nes_drv_opt, int, 0644);
 MODULE_PARM_DESC(nes_drv_opt, "Driver option parameters");
 
@@ -130,9 +130,6 @@ static struct notifier_block nes_net_notifier = {
 	.notifier_call = nes_net_event
 };
 
-
-
-
 /**
  * nes_inetaddr_event
  */
@@ -321,6 +318,9 @@ void nes_rem_ref(struct ib_qp *ibqp)
 	}
 
 	if (atomic_dec_and_test(&nesqp->refcount)) {
+		if (nesqp->pau_mode)
+			nes_destroy_pau_qp(nesdev, nesqp);
+
 		/* Destroy the QP */
 		cqp_request = nes_get_cqp_request(nesdev);
 		if (cqp_request == NULL) {

drivers/infiniband/hw/nes/nes.h

@@ -102,6 +102,7 @@
 #define NES_DRV_OPT_NO_INLINE_DATA       0x00000080
 #define NES_DRV_OPT_DISABLE_INT_MOD      0x00000100
 #define NES_DRV_OPT_DISABLE_VIRT_WQ      0x00000200
+#define NES_DRV_OPT_ENABLE_PAU           0x00000400
 
 #define NES_AEQ_EVENT_TIMEOUT         2500
 #define NES_DISCONNECT_EVENT_TIMEOUT  2000
@@ -128,6 +129,7 @@
 #define NES_DBG_IW_RX       0x00020000
 #define NES_DBG_IW_TX       0x00040000
 #define NES_DBG_SHUTDOWN    0x00080000
+#define NES_DBG_PAU         0x00100000
 #define NES_DBG_RSVD1       0x10000000
 #define NES_DBG_RSVD2       0x20000000
 #define NES_DBG_RSVD3       0x40000000
@@ -162,6 +164,7 @@ do { \
 #include "nes_context.h"
 #include "nes_user.h"
 #include "nes_cm.h"
+#include "nes_mgt.h"
 
 extern int max_mtu;
 #define max_frame_len (max_mtu+ETH_HLEN)
@@ -202,6 +205,8 @@ extern atomic_t cm_nodes_created;
 extern atomic_t cm_nodes_destroyed;
 extern atomic_t cm_accel_dropped_pkts;
 extern atomic_t cm_resets_recvd;
+extern atomic_t pau_qps_created;
+extern atomic_t pau_qps_destroyed;
 
 extern u32 int_mod_timer_init;
 extern u32 int_mod_cq_depth_256;
@@ -273,6 +278,14 @@ struct nes_device {
 	u8                     link_recheck;
 };
 
+/* Receive skb private area - must fit in skb->cb area */
+struct nes_rskb_cb {
+	u64                    busaddr;
+	u32                    maplen;
+	u32                    seqnum;
+	u8                     *data_start;
+	struct nes_qp          *nesqp;
+};
 
 static inline __le32 get_crc_value(struct nes_v4_quad *nes_quad)
 {
@@ -305,8 +318,8 @@ set_wqe_32bit_value(__le32 *wqe_words, u32 index, u32 value)
 static inline void
 nes_fill_init_cqp_wqe(struct nes_hw_cqp_wqe *cqp_wqe, struct nes_device *nesdev)
 {
-	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_COMP_CTX_LOW_IDX,
-			(u64)((unsigned long) &nesdev->cqp));
+	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_CTX_LOW_IDX]       = 0;
+	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_CTX_HIGH_IDX]      = 0;
 	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_LOW_IDX]   = 0;
 	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX]  = 0;
 	cqp_wqe->wqe_words[NES_CQP_STAG_WQE_PBL_BLK_COUNT_IDX] = 0;

drivers/infiniband/hw/nes/nes_cm.h

@@ -48,7 +48,16 @@
 #define IETF_MPA_KEY_SIZE 16
 #define IETF_MPA_VERSION  1
 #define IETF_MAX_PRIV_DATA_LEN 512
-#define IETF_MPA_FRAME_SIZE     20
+#define IETF_MPA_FRAME_SIZE    20
+#define IETF_RTR_MSG_SIZE      4
+#define IETF_MPA_V2_FLAG       0x10
+
+/* IETF RTR MSG Fields               */
+#define IETF_PEER_TO_PEER       0x8000
+#define IETF_FLPDU_ZERO_LEN     0x4000
+#define IETF_RDMA0_WRITE        0x8000
+#define IETF_RDMA0_READ         0x4000
+#define IETF_NO_IRD_ORD         0x3FFF
 
 enum ietf_mpa_flags {
 	IETF_MPA_FLAGS_MARKERS = 0x80,	/* receive Markers */
@@ -56,7 +65,7 @@ enum ietf_mpa_flags {
 	IETF_MPA_FLAGS_REJECT  = 0x20,	/* Reject */
 };
 
-struct ietf_mpa_frame {
+struct ietf_mpa_v1 {
 	u8 key[IETF_MPA_KEY_SIZE];
 	u8 flags;
 	u8 rev;
@@ -66,6 +75,20 @@ struct ietf_mpa_frame {
 
 #define ietf_mpa_req_resp_frame ietf_mpa_frame
 
+struct ietf_rtr_msg {
+	__be16 ctrl_ird;
+	__be16 ctrl_ord;
+};
+
+struct ietf_mpa_v2 {
+	u8 key[IETF_MPA_KEY_SIZE];
+	u8 flags;
+	u8 rev;
+	 __be16 priv_data_len;
+	struct ietf_rtr_msg rtr_msg;
+	u8 priv_data[0];
+};
+
 struct nes_v4_quad {
 	u32 rsvd0;
 	__le32 DstIpAdrIndex;	/* Only most significant 5 bits are valid */
@@ -171,8 +194,7 @@ struct nes_timer_entry {
 
 #define NES_CM_DEF_SEQ2      0x18ed5740
 #define NES_CM_DEF_LOCAL_ID2 0xb807
-#define	MAX_CM_BUFFER	(IETF_MPA_FRAME_SIZE + IETF_MAX_PRIV_DATA_LEN)
-
+#define	MAX_CM_BUFFER	(IETF_MPA_FRAME_SIZE + IETF_RTR_MSG_SIZE + IETF_MAX_PRIV_DATA_LEN)
 
 typedef u32 nes_addr_t;
 
@@ -204,6 +226,21 @@ enum nes_cm_node_state {
 	NES_CM_STATE_CLOSED
 };
 
+enum mpa_frame_version {
+	IETF_MPA_V1 = 1,
+	IETF_MPA_V2 = 2
+};
+
+enum mpa_frame_key {
+	MPA_KEY_REQUEST,
+	MPA_KEY_REPLY
+};
+
+enum send_rdma0 {
+	SEND_RDMA_READ_ZERO = 1,
+	SEND_RDMA_WRITE_ZERO = 2
+};
+
 enum nes_tcpip_pkt_type {
 	NES_PKT_TYPE_UNKNOWN,
 	NES_PKT_TYPE_SYN,
@@ -245,9 +282,9 @@ struct nes_cm_tcp_context {
 
 
 enum nes_cm_listener_state {
-	NES_CM_LISTENER_PASSIVE_STATE=1,
-	NES_CM_LISTENER_ACTIVE_STATE=2,
-	NES_CM_LISTENER_EITHER_STATE=3
+	NES_CM_LISTENER_PASSIVE_STATE = 1,
+	NES_CM_LISTENER_ACTIVE_STATE = 2,
+	NES_CM_LISTENER_EITHER_STATE = 3
 };
 
 struct nes_cm_listener {
@@ -283,16 +320,20 @@ struct nes_cm_node {
 
 	struct nes_cm_node        *loopbackpartner;
 
-	struct nes_timer_entry	*send_entry;
-
+	struct nes_timer_entry	  *send_entry;
+	struct nes_timer_entry    *recv_entry;
 	spinlock_t                retrans_list_lock;
-	struct nes_timer_entry  *recv_entry;
+	enum send_rdma0           send_rdma0_op;
 
-	int                       send_write0;
 	union {
-		struct ietf_mpa_frame mpa_frame;
-		u8                    mpa_frame_buf[MAX_CM_BUFFER];
+		struct ietf_mpa_v1 mpa_frame;
+		struct ietf_mpa_v2 mpa_v2_frame;
+		u8                 mpa_frame_buf[MAX_CM_BUFFER];
 	};
+	enum mpa_frame_version    mpa_frame_rev;
+	u16			  ird_size;
+	u16                       ord_size;
+
 	u16                       mpa_frame_size;
 	struct iw_cm_id           *cm_id;
 	struct list_head          list;
@@ -399,10 +440,8 @@ struct nes_cm_ops {
 			struct nes_vnic *, u16, void *,
 			struct nes_cm_info *);
 	int (*close)(struct nes_cm_core *, struct nes_cm_node *);
-	int (*accept)(struct nes_cm_core *, struct ietf_mpa_frame *,
-			struct nes_cm_node *);
-	int (*reject)(struct nes_cm_core *, struct ietf_mpa_frame *,
-			struct nes_cm_node *);
+	int (*accept)(struct nes_cm_core *, struct nes_cm_node *);
+	int (*reject)(struct nes_cm_core *, struct nes_cm_node *);
 	int (*recv_pkt)(struct nes_cm_core *, struct nes_vnic *,
 			struct sk_buff *);
 	int (*destroy_cm_core)(struct nes_cm_core *);
@@ -422,5 +461,7 @@ int nes_destroy_listen(struct iw_cm_id *);
 int nes_cm_recv(struct sk_buff *, struct net_device *);
 int nes_cm_start(void);
 int nes_cm_stop(void);
+int nes_add_ref_cm_node(struct nes_cm_node *cm_node);
+int nes_rem_ref_cm_node(struct nes_cm_node *cm_node);
 
 #endif			/* NES_CM_H */

drivers/infiniband/hw/nes/nes_hw.c

@@ -110,6 +110,14 @@ static unsigned char *nes_tcp_state_str[] = {
 };
 #endif
 
+static inline void print_ip(struct nes_cm_node *cm_node)
+{
+	unsigned char *rem_addr;
+	if (cm_node) {
+		rem_addr = (unsigned char *)&cm_node->rem_addr;
+		printk(KERN_ERR PFX "Remote IP addr: %pI4\n", rem_addr);
+	}
+}
 
 /**
  * nes_nic_init_timer_defaults
@@ -1555,6 +1563,7 @@ static void nes_replenish_nic_rq(struct nes_vnic *nesvnic)
 	struct nes_hw_nic_rq_wqe *nic_rqe;
 	struct nes_hw_nic *nesnic;
 	struct nes_device *nesdev;
+	struct nes_rskb_cb *cb;
 	u32 rx_wqes_posted = 0;
 
 	nesnic = &nesvnic->nic;
@@ -1580,6 +1589,9 @@ static void nes_replenish_nic_rq(struct nes_vnic *nesvnic)
 
 			bus_address = pci_map_single(nesdev->pcidev,
 					skb->data, nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->busaddr = bus_address;
+			cb->maplen = nesvnic->max_frame_size;
 
 			nic_rqe = &nesnic->rq_vbase[nesvnic->nic.rq_head];
 			nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] =
@@ -1669,6 +1681,7 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 	u32 cqp_head;
 	u32 counter;
 	u32 wqe_count;
+	struct nes_rskb_cb *cb;
 	u8 jumbomode=0;
 
 	/* Allocate fragment, SQ, RQ, and CQ; Reuse CEQ based on the PCI function */
@@ -1845,6 +1858,9 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 
 		pmem = pci_map_single(nesdev->pcidev, skb->data,
 				nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+		cb = (struct nes_rskb_cb *)&skb->cb[0];
+		cb->busaddr = pmem;
+		cb->maplen = nesvnic->max_frame_size;
 
 		nic_rqe = &nesvnic->nic.rq_vbase[counter];
 		nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] = cpu_to_le32(nesvnic->max_frame_size);
@@ -1873,6 +1889,13 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 			jumbomode = 1;
 		nes_nic_init_timer_defaults(nesdev, jumbomode);
 	}
+	if ((nesdev->nesadapter->allow_unaligned_fpdus) &&
+		(nes_init_mgt_qp(nesdev, netdev, nesvnic))) {
+			nes_debug(NES_DBG_INIT, "%s: Out of memory for pau nic\n", netdev->name);
+			nes_destroy_nic_qp(nesvnic);
+		return -ENOMEM;
+	}
+
 	nesvnic->lro_mgr.max_aggr       = nes_lro_max_aggr;
 	nesvnic->lro_mgr.max_desc       = NES_MAX_LRO_DESCRIPTORS;
 	nesvnic->lro_mgr.lro_arr        = nesvnic->lro_desc;
@@ -1895,28 +1918,29 @@ void nes_destroy_nic_qp(struct nes_vnic *nesvnic)
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_hw_cqp_wqe *cqp_wqe;
 	struct nes_hw_nic_sq_wqe *nic_sqe;
-	struct nes_hw_nic_rq_wqe *nic_rqe;
 	__le16 *wqe_fragment_length;
 	u16  wqe_fragment_index;
-	u64 wqe_frag;
 	u32 cqp_head;
 	u32 wqm_cfg0;
 	unsigned long flags;
+	struct sk_buff *rx_skb;
+	struct nes_rskb_cb *cb;
 	int ret;
 
+	if (nesdev->nesadapter->allow_unaligned_fpdus)
+		nes_destroy_mgt(nesvnic);
+
 	/* clear wqe stall before destroying NIC QP */
 	wqm_cfg0 = nes_read_indexed(nesdev, NES_IDX_WQM_CONFIG0);
 	nes_write_indexed(nesdev, NES_IDX_WQM_CONFIG0, wqm_cfg0 & 0xFFFF7FFF);
 
 	/* Free remaining NIC receive buffers */
 	while (nesvnic->nic.rq_head != nesvnic->nic.rq_tail) {
-		nic_rqe   = &nesvnic->nic.rq_vbase[nesvnic->nic.rq_tail];
-		wqe_frag  = (u64)le32_to_cpu(
-			nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_LOW_IDX]);
-		wqe_frag |= ((u64)le32_to_cpu(
-			nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX]))<<32;
-		pci_unmap_single(nesdev->pcidev, (dma_addr_t)wqe_frag,
-				nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+		rx_skb = nesvnic->nic.rx_skb[nesvnic->nic.rq_tail];
+		cb = (struct nes_rskb_cb *)&rx_skb->cb[0];
+		pci_unmap_single(nesdev->pcidev, cb->busaddr, cb->maplen,
+			PCI_DMA_FROMDEVICE);
+
 		dev_kfree_skb(nesvnic->nic.rx_skb[nesvnic->nic.rq_tail++]);
 		nesvnic->nic.rq_tail &= (nesvnic->nic.rq_size - 1);
 	}
@@ -2775,6 +2799,7 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 	struct nes_hw_nic_sq_wqe *nic_sqe;
 	struct sk_buff *skb;
 	struct sk_buff *rx_skb;
+	struct nes_rskb_cb *cb;
 	__le16 *wqe_fragment_length;
 	u32 head;
 	u32 cq_size;
@@ -2859,6 +2884,8 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 				bus_address += ((u64)le32_to_cpu(nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX])) << 32;
 				pci_unmap_single(nesdev->pcidev, bus_address,
 						nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+				cb = (struct nes_rskb_cb *)&rx_skb->cb[0];
+				cb->busaddr = 0;
 				/* rx_skb->tail = rx_skb->data + rx_pkt_size; */
 				/* rx_skb->len = rx_pkt_size; */
 				rx_skb->len = 0;  /* TODO: see if this is necessary */
@@ -2983,6 +3010,7 @@ skip_rx_indicate0:
 }
 
 
+
 /**
  * nes_cqp_ce_handler
  */
@@ -2997,6 +3025,8 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 	u32 cq_size;
 	u32 cqe_count=0;
 	u32 error_code;
+	u32 opcode;
+	u32 ctx_index;
 	/* u32 counter; */
 
 	head = cq->cq_head;
@@ -3007,12 +3037,9 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 		/* nes_debug(NES_DBG_CQP, "head=%u cqe_words=%08X\n", head,
 			  le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX])); */
 
-		if (le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX]) & NES_CQE_VALID) {
-			u64temp = (((u64)(le32_to_cpu(cq->cq_vbase[head].
-					cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX]))) << 32) |
-					((u64)(le32_to_cpu(cq->cq_vbase[head].
-					cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX])));
-			cqp = *((struct nes_hw_cqp **)&u64temp);
+		opcode = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX]);
+		if (opcode & NES_CQE_VALID) {
+			cqp = &nesdev->cqp;
 
 			error_code = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_ERROR_CODE_IDX]);
 			if (error_code) {
@@ -3021,15 +3048,14 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 						le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX])&0x3f,
 						(u16)(error_code >> 16),
 						(u16)error_code);
-				nes_debug(NES_DBG_CQP, "cqp: qp_id=%u, sq_head=%u, sq_tail=%u\n",
-						cqp->qp_id, cqp->sq_head, cqp->sq_tail);
 			}
 
-			u64temp = (((u64)(le32_to_cpu(nesdev->cqp.sq_vbase[cqp->sq_tail].
-					wqe_words[NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX]))) << 32) |
-					((u64)(le32_to_cpu(nesdev->cqp.sq_vbase[cqp->sq_tail].
-					wqe_words[NES_CQP_WQE_COMP_SCRATCH_LOW_IDX])));
-			cqp_request = *((struct nes_cqp_request **)&u64temp);
+			u64temp = (((u64)(le32_to_cpu(cq->cq_vbase[head].
+					cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX]))) << 32) |
+					((u64)(le32_to_cpu(cq->cq_vbase[head].
+					cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX])));
+
+			cqp_request = (struct nes_cqp_request *)(unsigned long)u64temp;
 			if (cqp_request) {
 				if (cqp_request->waiting) {
 					/* nes_debug(NES_DBG_CQP, "%s: Waking up requestor\n"); */
@@ -3075,9 +3101,15 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 		cqp_wqe = &nesdev->cqp.sq_vbase[head];
 		memcpy(cqp_wqe, &cqp_request->cqp_wqe, sizeof(*cqp_wqe));
 		barrier();
-		cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_LOW_IDX] =
+
+		opcode = cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX];
+		if ((opcode & NES_CQP_OPCODE_MASK) == NES_CQP_DOWNLOAD_SEGMENT)
+			ctx_index = NES_CQP_WQE_DL_COMP_CTX_LOW_IDX;
+		else
+			ctx_index = NES_CQP_WQE_COMP_CTX_LOW_IDX;
+		cqp_wqe->wqe_words[ctx_index] =
 			cpu_to_le32((u32)((unsigned long)cqp_request));
-		cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX] =
+		cqp_wqe->wqe_words[ctx_index + 1] =
 			cpu_to_le32((u32)(upper_32_bits((unsigned long)cqp_request)));
 		nes_debug(NES_DBG_CQP, "CQP request %p (opcode 0x%02X) put on CQPs SQ wqe%u.\n",
 				cqp_request, le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX])&0x3f, head);
@@ -3093,7 +3125,6 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 	nes_read32(nesdev->regs+NES_CQE_ALLOC);
 }
 
-
 static u8 *locate_mpa(u8 *pkt, u32 aeq_info)
 {
 	if (aeq_info & NES_AEQE_Q2_DATA_ETHERNET) {
@@ -3553,9 +3584,9 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 
 	aeqe_cq_id = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]);
 	if (aeq_info & NES_AEQE_QP) {
-		if ((!nes_is_resource_allocated(nesadapter, nesadapter->allocated_qps,
-				aeqe_cq_id)) ||
-				(atomic_read(&nesqp->close_timer_started)))
+		if (!nes_is_resource_allocated(nesadapter,
+				nesadapter->allocated_qps,
+				aeqe_cq_id))
 			return;
 	}
 
@@ -3566,8 +3597,7 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 
 			if (atomic_inc_return(&nesqp->close_timer_started) == 1) {
 				if ((tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) &&
-					(nesqp->ibqp_state == IB_QPS_RTS) &&
-					((nesadapter->eeprom_version >> 16) != NES_A0)) {
+					(nesqp->ibqp_state == IB_QPS_RTS)) {
 					spin_lock_irqsave(&nesqp->lock, flags);
 					nesqp->hw_iwarp_state = iwarp_state;
 					nesqp->hw_tcp_state = tcp_state;
@@ -3594,9 +3624,10 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				return;
 			}
 			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_CLOSING;
+			nesqp->hw_iwarp_state = iwarp_state;
+			nesqp->hw_tcp_state = tcp_state;
+			nesqp->last_aeq = async_event_id;
 			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_hw_modify_qp(nesdev, nesqp, NES_CQP_QP_IWARP_STATE_CLOSING, 0, 0);
 			nes_cm_disconn(nesqp);
 			break;
 
@@ -3694,7 +3725,9 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 		case NES_AEQE_AEID_ROE_INVALID_RDMA_WRITE_OR_READ_RESP:
 			printk(KERN_ERR PFX "QP[%u] async_event_id=0x%04X IB_EVENT_QP_FATAL\n",
 					nesqp->hwqp.qp_id, async_event_id);
-			nes_terminate_connection(nesdev, nesqp, aeqe, IB_EVENT_QP_FATAL);
+			print_ip(nesqp->cm_node);
+			if (!atomic_read(&nesqp->close_timer_started))
+				nes_terminate_connection(nesdev, nesqp, aeqe, IB_EVENT_QP_FATAL);
 			break;
 
 		case NES_AEQE_AEID_CQ_OPERATION_ERROR:

drivers/infiniband/hw/nes/nes_hw.h

@@ -47,6 +47,11 @@
 #define NES_MULTICAST_PF_MAX 8
 #define NES_A0 3
 
+#define NES_ENABLE_PAU 0x07000001
+#define NES_DISABLE_PAU 0x07000000
+#define NES_PAU_COUNTER 10
+#define NES_CQP_OPCODE_MASK 0x3f
+
 enum pci_regs {
 	NES_INT_STAT = 0x0000,
 	NES_INT_MASK = 0x0004,
@@ -73,8 +78,10 @@ enum indexed_regs {
 	NES_IDX_QP_CONTROL = 0x0040,
 	NES_IDX_FLM_CONTROL = 0x0080,
 	NES_IDX_INT_CPU_STATUS = 0x00a0,
+	NES_IDX_GPR_TRIGGER = 0x00bc,
 	NES_IDX_GPIO_CONTROL = 0x00f0,
 	NES_IDX_GPIO_DATA = 0x00f4,
+	NES_IDX_GPR2 = 0x010c,
 	NES_IDX_TCP_CONFIG0 = 0x01e4,
 	NES_IDX_TCP_TIMER_CONFIG = 0x01ec,
 	NES_IDX_TCP_NOW = 0x01f0,
@@ -202,6 +209,7 @@ enum nes_cqp_opcodes {
 	NES_CQP_REGISTER_SHARED_STAG = 0x0c,
 	NES_CQP_DEALLOCATE_STAG = 0x0d,
 	NES_CQP_MANAGE_ARP_CACHE = 0x0f,
+	NES_CQP_DOWNLOAD_SEGMENT = 0x10,
 	NES_CQP_SUSPEND_QPS = 0x11,
 	NES_CQP_UPLOAD_CONTEXT = 0x13,
 	NES_CQP_CREATE_CEQ = 0x16,
@@ -210,7 +218,8 @@ enum nes_cqp_opcodes {
 	NES_CQP_DESTROY_AEQ = 0x1b,
 	NES_CQP_LMI_ACCESS = 0x20,
 	NES_CQP_FLUSH_WQES = 0x22,
-	NES_CQP_MANAGE_APBVT = 0x23
+	NES_CQP_MANAGE_APBVT = 0x23,
+	NES_CQP_MANAGE_QUAD_HASH = 0x25
 };
 
 enum nes_cqp_wqe_word_idx {
@@ -222,6 +231,14 @@ enum nes_cqp_wqe_word_idx {
 	NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX = 5,
 };
 
+enum nes_cqp_wqe_word_download_idx { /* format differs from other cqp ops */
+	NES_CQP_WQE_DL_OPCODE_IDX = 0,
+	NES_CQP_WQE_DL_COMP_CTX_LOW_IDX = 1,
+	NES_CQP_WQE_DL_COMP_CTX_HIGH_IDX = 2,
+	NES_CQP_WQE_DL_LENGTH_0_TOTAL_IDX = 3
+	/* For index values 4-15 use NES_NIC_SQ_WQE_ values */
+};
+
 enum nes_cqp_cq_wqeword_idx {
 	NES_CQP_CQ_WQE_PBL_LOW_IDX = 6,
 	NES_CQP_CQ_WQE_PBL_HIGH_IDX = 7,
@@ -242,6 +259,7 @@ enum nes_cqp_stag_wqeword_idx {
 	NES_CQP_STAG_WQE_PBL_LEN_IDX = 14
 };
 
+#define NES_CQP_OP_LOGICAL_PORT_SHIFT 26
 #define NES_CQP_OP_IWARP_STATE_SHIFT 28
 #define NES_CQP_OP_TERMLEN_SHIFT     28
 
@@ -599,6 +617,7 @@ enum nes_nic_sq_wqe_bits {
 
 enum nes_nic_cqe_word_idx {
 	NES_NIC_CQE_ACCQP_ID_IDX = 0,
+	NES_NIC_CQE_HASH_RCVNXT = 1,
 	NES_NIC_CQE_TAG_PKT_TYPE_IDX = 2,
 	NES_NIC_CQE_MISC_IDX = 3,
 };
@@ -1005,6 +1024,11 @@ struct nes_arp_entry {
 #define NES_NIC_CQ_DOWNWARD_TREND   16
 #define NES_PFT_SIZE		    48
 
+#define NES_MGT_WQ_COUNT 32
+#define NES_MGT_CTX_SIZE ((NES_NIC_CTX_RQ_SIZE_32) | (NES_NIC_CTX_SQ_SIZE_32))
+#define NES_MGT_QP_OFFSET 36
+#define NES_MGT_QP_COUNT 4
+
 struct nes_hw_tune_timer {
     /* u16 cq_count; */
     u16 threshold_low;
@@ -1118,6 +1142,7 @@ struct nes_adapter {
 	u32 et_rate_sample_interval;
 	u32 timer_int_limit;
 	u32 wqm_quanta;
+	u8 allow_unaligned_fpdus;
 
 	/* Adapter base MAC address */
 	u32 mac_addr_low;
@@ -1251,6 +1276,14 @@ struct nes_vnic {
 	enum ib_event_type delayed_event;
 	enum ib_event_type last_dispatched_event;
 	spinlock_t port_ibevent_lock;
+	u32 mgt_mem_size;
+	void *mgt_vbase;
+	dma_addr_t mgt_pbase;
+	struct nes_vnic_mgt *mgtvnic[NES_MGT_QP_COUNT];
+	struct task_struct *mgt_thread;
+	wait_queue_head_t mgt_wait_queue;
+	struct sk_buff_head mgt_skb_list;
+
 };
 
 struct nes_ib_device {

drivers/infiniband/hw/nes/nes_mgt.c

@@ -0,0 +1,1162 @@
+/*
+ * Copyright (c) 2006 - 2009 Intel-NE, 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 <linux/etherdevice.h>
+#include <linux/kthread.h>
+#include <linux/ip.h>
+#include <linux/tcp.h>
+#include <net/tcp.h>
+#include "nes.h"
+#include "nes_mgt.h"
+
+atomic_t pau_qps_created;
+atomic_t pau_qps_destroyed;
+
+static void nes_replenish_mgt_rq(struct nes_vnic_mgt *mgtvnic)
+{
+	unsigned long flags;
+	dma_addr_t bus_address;
+	struct sk_buff *skb;
+	struct nes_hw_nic_rq_wqe *nic_rqe;
+	struct nes_hw_mgt *nesmgt;
+	struct nes_device *nesdev;
+	struct nes_rskb_cb *cb;
+	u32 rx_wqes_posted = 0;
+
+	nesmgt = &mgtvnic->mgt;
+	nesdev = mgtvnic->nesvnic->nesdev;
+	spin_lock_irqsave(&nesmgt->rq_lock, flags);
+	if (nesmgt->replenishing_rq != 0) {
+		if (((nesmgt->rq_size - 1) == atomic_read(&mgtvnic->rx_skbs_needed)) &&
+		    (atomic_read(&mgtvnic->rx_skb_timer_running) == 0)) {
+			atomic_set(&mgtvnic->rx_skb_timer_running, 1);
+			spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+			mgtvnic->rq_wqes_timer.expires = jiffies + (HZ / 2);      /* 1/2 second */
+			add_timer(&mgtvnic->rq_wqes_timer);
+		} else {
+			spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+		}
+		return;
+	}
+	nesmgt->replenishing_rq = 1;
+	spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+	do {
+		skb = dev_alloc_skb(mgtvnic->nesvnic->max_frame_size);
+		if (skb) {
+			skb->dev = mgtvnic->nesvnic->netdev;
+
+			bus_address = pci_map_single(nesdev->pcidev,
+						     skb->data, mgtvnic->nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->busaddr = bus_address;
+			cb->maplen = mgtvnic->nesvnic->max_frame_size;
+
+			nic_rqe = &nesmgt->rq_vbase[mgtvnic->mgt.rq_head];
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] =
+				cpu_to_le32(mgtvnic->nesvnic->max_frame_size);
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_3_2_IDX] = 0;
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_LOW_IDX] =
+				cpu_to_le32((u32)bus_address);
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX] =
+				cpu_to_le32((u32)((u64)bus_address >> 32));
+			nesmgt->rx_skb[nesmgt->rq_head] = skb;
+			nesmgt->rq_head++;
+			nesmgt->rq_head &= nesmgt->rq_size - 1;
+			atomic_dec(&mgtvnic->rx_skbs_needed);
+			barrier();
+			if (++rx_wqes_posted == 255) {
+				nes_write32(nesdev->regs + NES_WQE_ALLOC, (rx_wqes_posted << 24) | nesmgt->qp_id);
+				rx_wqes_posted = 0;
+			}
+		} else {
+			spin_lock_irqsave(&nesmgt->rq_lock, flags);
+			if (((nesmgt->rq_size - 1) == atomic_read(&mgtvnic->rx_skbs_needed)) &&
+			    (atomic_read(&mgtvnic->rx_skb_timer_running) == 0)) {
+				atomic_set(&mgtvnic->rx_skb_timer_running, 1);
+				spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+				mgtvnic->rq_wqes_timer.expires = jiffies + (HZ / 2);      /* 1/2 second */
+				add_timer(&mgtvnic->rq_wqes_timer);
+			} else {
+				spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+			}
+			break;
+		}
+	} while (atomic_read(&mgtvnic->rx_skbs_needed));
+	barrier();
+	if (rx_wqes_posted)
+		nes_write32(nesdev->regs + NES_WQE_ALLOC, (rx_wqes_posted << 24) | nesmgt->qp_id);
+	nesmgt->replenishing_rq = 0;
+}
+
+/**
+ * nes_mgt_rq_wqes_timeout
+ */
+static void nes_mgt_rq_wqes_timeout(unsigned long parm)
+{
+	struct nes_vnic_mgt *mgtvnic = (struct nes_vnic_mgt *)parm;
+
+	atomic_set(&mgtvnic->rx_skb_timer_running, 0);
+	if (atomic_read(&mgtvnic->rx_skbs_needed))
+		nes_replenish_mgt_rq(mgtvnic);
+}
+
+/**
+ * nes_mgt_free_skb - unmap and free skb
+ */
+static void nes_mgt_free_skb(struct nes_device *nesdev, struct sk_buff *skb, u32 dir)
+{
+	struct nes_rskb_cb *cb;
+
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	pci_unmap_single(nesdev->pcidev, cb->busaddr, cb->maplen, dir);
+	cb->busaddr = 0;
+	dev_kfree_skb_any(skb);
+}
+
+/**
+ * nes_download_callback - handle download completions
+ */
+static void nes_download_callback(struct nes_device *nesdev, struct nes_cqp_request *cqp_request)
+{
+	struct pau_fpdu_info *fpdu_info = cqp_request->cqp_callback_pointer;
+	struct nes_qp *nesqp = fpdu_info->nesqp;
+	struct sk_buff *skb;
+	int i;
+
+	for (i = 0; i < fpdu_info->frag_cnt; i++) {
+		skb = fpdu_info->frags[i].skb;
+		if (fpdu_info->frags[i].cmplt) {
+			nes_mgt_free_skb(nesdev, skb, PCI_DMA_TODEVICE);
+			nes_rem_ref_cm_node(nesqp->cm_node);
+		}
+	}
+
+	if (fpdu_info->hdr_vbase)
+		pci_free_consistent(nesdev->pcidev, fpdu_info->hdr_len,
+				    fpdu_info->hdr_vbase, fpdu_info->hdr_pbase);
+	kfree(fpdu_info);
+}
+
+/**
+ * nes_get_seq - Get the seq, ack_seq and window from the packet
+ */
+static u32 nes_get_seq(struct sk_buff *skb, u32 *ack, u16 *wnd, u32 *fin_rcvd, u32 *rst_rcvd)
+{
+	struct nes_rskb_cb *cb = (struct nes_rskb_cb *)&skb->cb[0];
+	struct iphdr *iph = (struct iphdr *)(cb->data_start + ETH_HLEN);
+	struct tcphdr *tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+
+	*ack = be32_to_cpu(tcph->ack_seq);
+	*wnd = be16_to_cpu(tcph->window);
+	*fin_rcvd = tcph->fin;
+	*rst_rcvd = tcph->rst;
+	return be32_to_cpu(tcph->seq);
+}
+
+/**
+ * nes_get_next_skb - Get the next skb based on where current skb is in the queue
+ */
+static struct sk_buff *nes_get_next_skb(struct nes_device *nesdev, struct nes_qp *nesqp,
+					struct sk_buff *skb, u32 nextseq, u32 *ack,
+					u16 *wnd, u32 *fin_rcvd, u32 *rst_rcvd)
+{
+	u32 seq;
+	bool processacks;
+	struct sk_buff *old_skb;
+
+	if (skb) {
+		/* Continue processing fpdu */
+		if (skb->next == (struct sk_buff *)&nesqp->pau_list)
+			goto out;
+		skb = skb->next;
+		processacks = false;
+	} else {
+		/* Starting a new one */
+		if (skb_queue_empty(&nesqp->pau_list))
+			goto out;
+		skb = skb_peek(&nesqp->pau_list);
+		processacks = true;
+	}
+
+	while (1) {
+		seq = nes_get_seq(skb, ack, wnd, fin_rcvd, rst_rcvd);
+		if (seq == nextseq) {
+			if (skb->len || processacks)
+				break;
+		} else if (after(seq, nextseq)) {
+			goto out;
+		}
+
+		if (skb->next == (struct sk_buff *)&nesqp->pau_list)
+			goto out;
+
+		old_skb = skb;
+		skb = skb->next;
+		skb_unlink(old_skb, &nesqp->pau_list);
+		nes_mgt_free_skb(nesdev, old_skb, PCI_DMA_TODEVICE);
+		nes_rem_ref_cm_node(nesqp->cm_node);
+	}
+	return skb;
+
+out:
+	return NULL;
+}
+
+/**
+ * get_fpdu_info - Find the next complete fpdu and return its fragments.
+ */
+static int get_fpdu_info(struct nes_device *nesdev, struct nes_qp *nesqp,
+			 struct pau_fpdu_info **pau_fpdu_info)
+{
+	struct sk_buff *skb;
+	struct iphdr *iph;
+	struct tcphdr *tcph;
+	struct nes_rskb_cb *cb;
+	struct pau_fpdu_info *fpdu_info = NULL;
+	struct pau_fpdu_frag frags[MAX_FPDU_FRAGS];
+	unsigned long flags;
+	u32 fpdu_len = 0;
+	u32 tmp_len;
+	int frag_cnt = 0;
+	u32 tot_len;
+	u32 frag_tot;
+	u32 ack;
+	u32 fin_rcvd;
+	u32 rst_rcvd;
+	u16 wnd;
+	int i;
+	int rc = 0;
+
+	*pau_fpdu_info = NULL;
+
+	spin_lock_irqsave(&nesqp->pau_lock, flags);
+	skb = nes_get_next_skb(nesdev, nesqp, NULL, nesqp->pau_rcv_nxt, &ack, &wnd, &fin_rcvd, &rst_rcvd);
+	if (!skb) {
+		spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+		goto out;
+	}
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	if (skb->len) {
+		fpdu_len = be16_to_cpu(*(__be16 *) skb->data) + MPA_FRAMING;
+		fpdu_len = (fpdu_len + 3) & 0xfffffffc;
+		tmp_len = fpdu_len;
+
+		/* See if we have all of the fpdu */
+		frag_tot = 0;
+		memset(&frags, 0, sizeof frags);
+		for (i = 0; i < MAX_FPDU_FRAGS; i++) {
+			frags[i].physaddr = cb->busaddr;
+			frags[i].physaddr += skb->data - cb->data_start;
+			frags[i].frag_len = min(tmp_len, skb->len);
+			frags[i].skb = skb;
+			frags[i].cmplt = (skb->len == frags[i].frag_len);
+			frag_tot += frags[i].frag_len;
+			frag_cnt++;
+
+			tmp_len -= frags[i].frag_len;
+			if (tmp_len == 0)
+				break;
+
+			skb = nes_get_next_skb(nesdev, nesqp, skb,
+					       nesqp->pau_rcv_nxt + frag_tot, &ack, &wnd, &fin_rcvd, &rst_rcvd);
+			if (!skb) {
+				spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+				goto out;
+			} else if (rst_rcvd) {
+				/* rst received in the middle of fpdu */
+				for (; i >= 0; i--) {
+					skb_unlink(frags[i].skb, &nesqp->pau_list);
+					nes_mgt_free_skb(nesdev, frags[i].skb, PCI_DMA_TODEVICE);
+				}
+				cb = (struct nes_rskb_cb *)&skb->cb[0];
+				frags[0].physaddr = cb->busaddr;
+				frags[0].physaddr += skb->data - cb->data_start;
+				frags[0].frag_len = skb->len;
+				frags[0].skb = skb;
+				frags[0].cmplt = true;
+				frag_cnt = 1;
+				break;
+			}
+
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+		}
+	} else {
+		/* no data */
+		frags[0].physaddr = cb->busaddr;
+		frags[0].frag_len = 0;
+		frags[0].skb = skb;
+		frags[0].cmplt = true;
+		frag_cnt = 1;
+	}
+
+	spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+
+	/* Found one */
+	fpdu_info = kzalloc(sizeof(*fpdu_info), GFP_ATOMIC);
+	if (fpdu_info == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to alloc a fpdu_info.\n");
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	fpdu_info->cqp_request = nes_get_cqp_request(nesdev);
+	if (fpdu_info->cqp_request == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to get a cqp_request.\n");
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	cb = (struct nes_rskb_cb *)&frags[0].skb->cb[0];
+	iph = (struct iphdr *)(cb->data_start + ETH_HLEN);
+	tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+	fpdu_info->hdr_len = (((unsigned char *)tcph) + 4 * (tcph->doff)) - cb->data_start;
+	fpdu_info->data_len = fpdu_len;
+	tot_len = fpdu_info->hdr_len + fpdu_len - ETH_HLEN;
+
+	if (frags[0].cmplt) {
+		fpdu_info->hdr_pbase = cb->busaddr;
+		fpdu_info->hdr_vbase = NULL;
+	} else {
+		fpdu_info->hdr_vbase = pci_alloc_consistent(nesdev->pcidev,
+							    fpdu_info->hdr_len, &fpdu_info->hdr_pbase);
+		if (!fpdu_info->hdr_vbase) {
+			nes_debug(NES_DBG_PAU, "Unable to allocate memory for pau first frag\n");
+			rc = -ENOMEM;
+			goto out;
+		}
+
+		/* Copy hdrs, adjusting len and seqnum */
+		memcpy(fpdu_info->hdr_vbase, cb->data_start, fpdu_info->hdr_len);
+		iph = (struct iphdr *)(fpdu_info->hdr_vbase + ETH_HLEN);
+		tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+	}
+
+	iph->tot_len = cpu_to_be16(tot_len);
+	iph->saddr = cpu_to_be32(0x7f000001);
+
+	tcph->seq = cpu_to_be32(nesqp->pau_rcv_nxt);
+	tcph->ack_seq = cpu_to_be32(ack);
+	tcph->window = cpu_to_be16(wnd);
+
+	nesqp->pau_rcv_nxt += fpdu_len + fin_rcvd;
+
+	memcpy(fpdu_info->frags, frags, sizeof(fpdu_info->frags));
+	fpdu_info->frag_cnt = frag_cnt;
+	fpdu_info->nesqp = nesqp;
+	*pau_fpdu_info = fpdu_info;
+
+	/* Update skb's for next pass */
+	for (i = 0; i < frag_cnt; i++) {
+		cb = (struct nes_rskb_cb *)&frags[i].skb->cb[0];
+		skb_pull(frags[i].skb, frags[i].frag_len);
+
+		if (frags[i].skb->len == 0) {
+			/* Pull skb off the list - it will be freed in the callback */
+			spin_lock_irqsave(&nesqp->pau_lock, flags);
+			skb_unlink(frags[i].skb, &nesqp->pau_list);
+			spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+		} else {
+			/* Last skb still has data so update the seq */
+			iph = (struct iphdr *)(cb->data_start + ETH_HLEN);
+			tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+			tcph->seq = cpu_to_be32(nesqp->pau_rcv_nxt);
+		}
+	}
+
+out:
+	if (rc) {
+		if (fpdu_info) {
+			if (fpdu_info->cqp_request)
+				nes_put_cqp_request(nesdev, fpdu_info->cqp_request);
+			kfree(fpdu_info);
+		}
+	}
+	return rc;
+}
+
+/**
+ * forward_fpdu - send complete fpdus, one at a time
+ */
+static int forward_fpdus(struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct pau_fpdu_info *fpdu_info;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	struct nes_cqp_request *cqp_request;
+	u64 u64tmp;
+	u32 u32tmp;
+	int rc;
+
+	while (1) {
+		rc = get_fpdu_info(nesdev, nesqp, &fpdu_info);
+		if (fpdu_info == NULL)
+			return rc;
+
+		cqp_request = fpdu_info->cqp_request;
+		cqp_wqe = &cqp_request->cqp_wqe;
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_DL_OPCODE_IDX,
+				    NES_CQP_DOWNLOAD_SEGMENT |
+				    (((u32)nesvnic->logical_port) << NES_CQP_OP_LOGICAL_PORT_SHIFT));
+
+		u32tmp = fpdu_info->hdr_len << 16;
+		u32tmp |= fpdu_info->hdr_len + (u32)fpdu_info->data_len;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_DL_LENGTH_0_TOTAL_IDX,
+				    u32tmp);
+
+		u32tmp = (fpdu_info->frags[1].frag_len << 16) | fpdu_info->frags[0].frag_len;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_LENGTH_2_1_IDX,
+				    u32tmp);
+
+		u32tmp = (fpdu_info->frags[3].frag_len << 16) | fpdu_info->frags[2].frag_len;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_LENGTH_4_3_IDX,
+				    u32tmp);
+
+		u64tmp = (u64)fpdu_info->hdr_pbase;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG0_LOW_IDX,
+				    lower_32_bits(u64tmp));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG0_HIGH_IDX,
+				    upper_32_bits(u64tmp >> 32));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG1_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[0].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG1_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[0].physaddr));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG2_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[1].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG2_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[1].physaddr));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG3_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[2].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG3_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[2].physaddr));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG4_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[3].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG4_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[3].physaddr));
+
+		cqp_request->cqp_callback_pointer = fpdu_info;
+		cqp_request->callback = 1;
+		cqp_request->cqp_callback = nes_download_callback;
+
+		atomic_set(&cqp_request->refcount, 1);
+		nes_post_cqp_request(nesdev, cqp_request);
+	}
+
+	return 0;
+}
+
+static void process_fpdus(struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	int again = 1;
+	unsigned long flags;
+
+	do {
+		/* Ignore rc - if it failed, tcp retries will cause it to try again */
+		forward_fpdus(nesvnic, nesqp);
+
+		spin_lock_irqsave(&nesqp->pau_lock, flags);
+		if (nesqp->pau_pending) {
+			nesqp->pau_pending = 0;
+		} else {
+			nesqp->pau_busy = 0;
+			again = 0;
+		}
+
+		spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+	} while (again);
+}
+
+/**
+ * queue_fpdus - Handle fpdu's that hw passed up to sw
+ */
+static void queue_fpdus(struct sk_buff *skb, struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct sk_buff *tmpskb;
+	struct nes_rskb_cb *cb;
+	struct iphdr *iph;
+	struct tcphdr *tcph;
+	unsigned char *tcph_end;
+	u32 rcv_nxt;
+	u32 rcv_wnd;
+	u32 seqnum;
+	u32 len;
+	bool process_it = false;
+	unsigned long flags;
+
+	/* Move data ptr to after tcp header */
+	iph = (struct iphdr *)skb->data;
+	tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+	seqnum = be32_to_cpu(tcph->seq);
+	tcph_end = (((char *)tcph) + (4 * tcph->doff));
+
+	len = be16_to_cpu(iph->tot_len);
+	if (skb->len > len)
+		skb_trim(skb, len);
+	skb_pull(skb, tcph_end - skb->data);
+
+	/* Initialize tracking values */
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	cb->seqnum = seqnum;
+
+	/* Make sure data is in the receive window */
+	rcv_nxt = nesqp->pau_rcv_nxt;
+	rcv_wnd = le32_to_cpu(nesqp->nesqp_context->rcv_wnd);
+	if (!between(seqnum, rcv_nxt, (rcv_nxt + rcv_wnd))) {
+		nes_mgt_free_skb(nesvnic->nesdev, skb, PCI_DMA_TODEVICE);
+		nes_rem_ref_cm_node(nesqp->cm_node);
+		return;
+	}
+
+	spin_lock_irqsave(&nesqp->pau_lock, flags);
+
+	if (nesqp->pau_busy)
+		nesqp->pau_pending = 1;
+	else
+		nesqp->pau_busy = 1;
+
+	/* Queue skb by sequence number */
+	if (skb_queue_len(&nesqp->pau_list) == 0) {
+		skb_queue_head(&nesqp->pau_list, skb);
+	} else {
+		tmpskb = nesqp->pau_list.next;
+		while (tmpskb != (struct sk_buff *)&nesqp->pau_list) {
+			cb = (struct nes_rskb_cb *)&tmpskb->cb[0];
+			if (before(seqnum, cb->seqnum))
+				break;
+			tmpskb = tmpskb->next;
+		}
+		skb_insert(tmpskb, skb, &nesqp->pau_list);
+	}
+	if (nesqp->pau_state == PAU_READY)
+		process_it = true;
+	spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+
+	if (process_it)
+		process_fpdus(nesvnic, nesqp);
+
+	return;
+}
+
+/**
+ * mgt_thread - Handle mgt skbs in a safe context
+ */
+static int mgt_thread(void *context)
+{
+	struct nes_vnic *nesvnic = context;
+	struct sk_buff *skb;
+	struct nes_rskb_cb *cb;
+
+	while (!kthread_should_stop()) {
+		wait_event_interruptible(nesvnic->mgt_wait_queue,
+					 skb_queue_len(&nesvnic->mgt_skb_list) || kthread_should_stop());
+		while ((skb_queue_len(&nesvnic->mgt_skb_list)) && !kthread_should_stop()) {
+			skb = skb_dequeue(&nesvnic->mgt_skb_list);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->data_start = skb->data - ETH_HLEN;
+			cb->busaddr = pci_map_single(nesvnic->nesdev->pcidev, cb->data_start,
+						     nesvnic->max_frame_size, PCI_DMA_TODEVICE);
+			queue_fpdus(skb, nesvnic, cb->nesqp);
+		}
+	}
+
+	/* Closing down so delete any entries on the queue */
+	while (skb_queue_len(&nesvnic->mgt_skb_list)) {
+		skb = skb_dequeue(&nesvnic->mgt_skb_list);
+		cb = (struct nes_rskb_cb *)&skb->cb[0];
+		nes_rem_ref_cm_node(cb->nesqp->cm_node);
+		dev_kfree_skb_any(skb);
+	}
+	return 0;
+}
+
+/**
+ * nes_queue_skbs - Queue skb so it can be handled in a thread context
+ */
+void nes_queue_mgt_skbs(struct sk_buff *skb, struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct nes_rskb_cb *cb;
+
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	cb->nesqp = nesqp;
+	skb_queue_tail(&nesvnic->mgt_skb_list, skb);
+	wake_up_interruptible(&nesvnic->mgt_wait_queue);
+}
+
+void nes_destroy_pau_qp(struct nes_device *nesdev, struct nes_qp *nesqp)
+{
+	struct sk_buff *skb;
+	unsigned long flags;
+	atomic_inc(&pau_qps_destroyed);
+
+	/* Free packets that have not yet been forwarded */
+	/* Lock is acquired by skb_dequeue when removing the skb */
+	spin_lock_irqsave(&nesqp->pau_lock, flags);
+	while (skb_queue_len(&nesqp->pau_list)) {
+		skb = skb_dequeue(&nesqp->pau_list);
+		nes_mgt_free_skb(nesdev, skb, PCI_DMA_TODEVICE);
+		nes_rem_ref_cm_node(nesqp->cm_node);
+	}
+	spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+}
+
+static void nes_chg_qh_handler(struct nes_device *nesdev, struct nes_cqp_request *cqp_request)
+{
+	struct pau_qh_chg *qh_chg = cqp_request->cqp_callback_pointer;
+	struct nes_cqp_request *new_request;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	struct nes_adapter *nesadapter;
+	struct nes_qp *nesqp;
+	struct nes_v4_quad nes_quad;
+	u32 crc_value;
+	u64 u64temp;
+
+	nesadapter = nesdev->nesadapter;
+	nesqp = qh_chg->nesqp;
+
+	/* Should we handle the bad completion */
+	if (cqp_request->major_code) {
+		printk(KERN_ERR PFX "Invalid cqp_request major_code=0x%x\n",
+		       cqp_request->major_code);
+		WARN_ON(1);
+	}
+
+	switch (nesqp->pau_state) {
+	case PAU_DEL_QH:
+		/* Old hash code deleted, now set the new one */
+		nesqp->pau_state = PAU_ADD_LB_QH;
+		new_request = nes_get_cqp_request(nesdev);
+		if (new_request == NULL) {
+			nes_debug(NES_DBG_PAU, "Failed to get a new_request.\n");
+			WARN_ON(1);
+			return;
+		}
+
+		memset(&nes_quad, 0, sizeof(nes_quad));
+		nes_quad.DstIpAdrIndex =
+			cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24);
+		nes_quad.SrcIpadr = cpu_to_be32(0x7f000001);
+		nes_quad.TcpPorts[0] = swab16(nesqp->nesqp_context->tcpPorts[1]);
+		nes_quad.TcpPorts[1] = swab16(nesqp->nesqp_context->tcpPorts[0]);
+
+		/* Produce hash key */
+		crc_value = get_crc_value(&nes_quad);
+		nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff);
+		nes_debug(NES_DBG_PAU, "new HTE Index = 0x%08X, CRC = 0x%08X\n",
+			  nesqp->hte_index, nesqp->hte_index & nesadapter->hte_index_mask);
+
+		nesqp->hte_index &= nesadapter->hte_index_mask;
+		nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index);
+		nesqp->nesqp_context->ip0 = cpu_to_le32(0x7f000001);
+		nesqp->nesqp_context->rcv_nxt = cpu_to_le32(nesqp->pau_rcv_nxt);
+
+		cqp_wqe = &new_request->cqp_wqe;
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+		set_wqe_32bit_value(cqp_wqe->wqe_words,
+				    NES_CQP_WQE_OPCODE_IDX, NES_CQP_MANAGE_QUAD_HASH |
+				    NES_CQP_QP_TYPE_IWARP | NES_CQP_QP_CONTEXT_VALID | NES_CQP_QP_IWARP_STATE_RTS);
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX, nesqp->hwqp.qp_id);
+		u64temp = (u64)nesqp->nesqp_context_pbase;
+		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, u64temp);
+
+		nes_debug(NES_DBG_PAU, "Waiting for CQP completion for adding the quad hash.\n");
+
+		new_request->cqp_callback_pointer = qh_chg;
+		new_request->callback = 1;
+		new_request->cqp_callback = nes_chg_qh_handler;
+		atomic_set(&new_request->refcount, 1);
+		nes_post_cqp_request(nesdev, new_request);
+		break;
+
+	case PAU_ADD_LB_QH:
+		/* Start processing the queued fpdu's */
+		nesqp->pau_state = PAU_READY;
+		process_fpdus(qh_chg->nesvnic, qh_chg->nesqp);
+		kfree(qh_chg);
+		break;
+	}
+}
+
+/**
+ * nes_change_quad_hash
+ */
+static int nes_change_quad_hash(struct nes_device *nesdev,
+				struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct nes_cqp_request *cqp_request = NULL;
+	struct pau_qh_chg *qh_chg = NULL;
+	u64 u64temp;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	int ret = 0;
+
+	cqp_request = nes_get_cqp_request(nesdev);
+	if (cqp_request == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to get a cqp_request.\n");
+		ret = -ENOMEM;
+		goto chg_qh_err;
+	}
+
+	qh_chg = kmalloc(sizeof *qh_chg, GFP_ATOMIC);
+	if (qh_chg == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to get a cqp_request.\n");
+		ret = -ENOMEM;
+		goto chg_qh_err;
+	}
+	qh_chg->nesdev = nesdev;
+	qh_chg->nesvnic = nesvnic;
+	qh_chg->nesqp = nesqp;
+	nesqp->pau_state = PAU_DEL_QH;
+
+	cqp_wqe = &cqp_request->cqp_wqe;
+	nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+	set_wqe_32bit_value(cqp_wqe->wqe_words,
+			    NES_CQP_WQE_OPCODE_IDX, NES_CQP_MANAGE_QUAD_HASH | NES_CQP_QP_DEL_HTE |
+			    NES_CQP_QP_TYPE_IWARP | NES_CQP_QP_CONTEXT_VALID | NES_CQP_QP_IWARP_STATE_RTS);
+	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX, nesqp->hwqp.qp_id);
+	u64temp = (u64)nesqp->nesqp_context_pbase;
+	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, u64temp);
+
+	nes_debug(NES_DBG_PAU, "Waiting for CQP completion for deleting the quad hash.\n");
+
+	cqp_request->cqp_callback_pointer = qh_chg;
+	cqp_request->callback = 1;
+	cqp_request->cqp_callback = nes_chg_qh_handler;
+	atomic_set(&cqp_request->refcount, 1);
+	nes_post_cqp_request(nesdev, cqp_request);
+
+	return ret;
+
+chg_qh_err:
+	kfree(qh_chg);
+	if (cqp_request)
+		nes_put_cqp_request(nesdev, cqp_request);
+	return ret;
+}
+
+/**
+ * nes_mgt_ce_handler
+ * This management code deals with any packed and unaligned (pau) fpdu's
+ * that the hardware cannot handle.
+ */
+static void nes_mgt_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
+{
+	struct nes_vnic_mgt *mgtvnic = container_of(cq, struct nes_vnic_mgt, mgt_cq);
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
+	u32 head;
+	u32 cq_size;
+	u32 cqe_count = 0;
+	u32 cqe_misc;
+	u32 qp_id = 0;
+	u32 skbs_needed;
+	unsigned long context;
+	struct nes_qp *nesqp;
+	struct sk_buff *rx_skb;
+	struct nes_rskb_cb *cb;
+
+	head = cq->cq_head;
+	cq_size = cq->cq_size;
+
+	while (1) {
+		cqe_misc = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_NIC_CQE_MISC_IDX]);
+		if (!(cqe_misc & NES_NIC_CQE_VALID))
+			break;
+
+		nesqp = NULL;
+		if (cqe_misc & NES_NIC_CQE_ACCQP_VALID) {
+			qp_id = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_NIC_CQE_ACCQP_ID_IDX]);
+			qp_id &= 0x001fffff;
+			if (qp_id < nesadapter->max_qp) {
+				context = (unsigned long)nesadapter->qp_table[qp_id - NES_FIRST_QPN];
+				nesqp = (struct nes_qp *)context;
+			}
+		}
+
+		if (nesqp) {
+			if (nesqp->pau_mode == false) {
+				nesqp->pau_mode = true; /* First time for this qp */
+				nesqp->pau_rcv_nxt = le32_to_cpu(
+					cq->cq_vbase[head].cqe_words[NES_NIC_CQE_HASH_RCVNXT]);
+				skb_queue_head_init(&nesqp->pau_list);
+				spin_lock_init(&nesqp->pau_lock);
+				atomic_inc(&pau_qps_created);
+				nes_change_quad_hash(nesdev, mgtvnic->nesvnic, nesqp);
+			}
+
+			rx_skb = mgtvnic->mgt.rx_skb[mgtvnic->mgt.rq_tail];
+			rx_skb->len = 0;
+			skb_put(rx_skb, cqe_misc & 0x0000ffff);
+			rx_skb->protocol = eth_type_trans(rx_skb, mgtvnic->nesvnic->netdev);
+			cb = (struct nes_rskb_cb *)&rx_skb->cb[0];
+			pci_unmap_single(nesdev->pcidev, cb->busaddr, cb->maplen, PCI_DMA_FROMDEVICE);
+			cb->busaddr = 0;
+			mgtvnic->mgt.rq_tail++;
+			mgtvnic->mgt.rq_tail &= mgtvnic->mgt.rq_size - 1;
+
+			nes_add_ref_cm_node(nesqp->cm_node);
+			nes_queue_mgt_skbs(rx_skb, mgtvnic->nesvnic, nesqp);
+		} else {
+			printk(KERN_ERR PFX "Invalid QP %d for packed/unaligned handling\n", qp_id);
+		}
+
+		cq->cq_vbase[head].cqe_words[NES_NIC_CQE_MISC_IDX] = 0;
+		cqe_count++;
+		if (++head >= cq_size)
+			head = 0;
+
+		if (cqe_count == 255) {
+			/* Replenish mgt CQ */
+			nes_write32(nesdev->regs + NES_CQE_ALLOC, cq->cq_number | (cqe_count << 16));
+			nesdev->currcq_count += cqe_count;
+			cqe_count = 0;
+		}
+
+		skbs_needed = atomic_inc_return(&mgtvnic->rx_skbs_needed);
+		if (skbs_needed > (mgtvnic->mgt.rq_size >> 1))
+			nes_replenish_mgt_rq(mgtvnic);
+	}
+
+	cq->cq_head = head;
+	nes_write32(nesdev->regs + NES_CQE_ALLOC, NES_CQE_ALLOC_NOTIFY_NEXT |
+		    cq->cq_number | (cqe_count << 16));
+	nes_read32(nesdev->regs + NES_CQE_ALLOC);
+	nesdev->currcq_count += cqe_count;
+}
+
+/**
+ * nes_init_mgt_qp
+ */
+int nes_init_mgt_qp(struct nes_device *nesdev, struct net_device *netdev, struct nes_vnic *nesvnic)
+{
+	struct nes_vnic_mgt *mgtvnic;
+	u32 counter;
+	void *vmem;
+	dma_addr_t pmem;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	u32 cqp_head;
+	unsigned long flags;
+	struct nes_hw_nic_qp_context *mgt_context;
+	u64 u64temp;
+	struct nes_hw_nic_rq_wqe *mgt_rqe;
+	struct sk_buff *skb;
+	u32 wqe_count;
+	struct nes_rskb_cb *cb;
+	u32 mgt_mem_size;
+	void *mgt_vbase;
+	dma_addr_t mgt_pbase;
+	int i;
+	int ret;
+
+	/* Allocate space the all mgt QPs once */
+	mgtvnic = kzalloc(NES_MGT_QP_COUNT * sizeof(struct nes_vnic_mgt), GFP_KERNEL);
+	if (mgtvnic == NULL) {
+		nes_debug(NES_DBG_INIT, "Unable to allocate memory for mgt structure\n");
+		return -ENOMEM;
+	}
+
+	/* Allocate fragment, RQ, and CQ; Reuse CEQ based on the PCI function */
+	/* We are not sending from this NIC so sq is not allocated */
+	mgt_mem_size = 256 +
+		       (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_rq_wqe)) +
+		       (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_cqe)) +
+		       sizeof(struct nes_hw_nic_qp_context);
+	mgt_mem_size = (mgt_mem_size + PAGE_SIZE - 1) & ~(PAGE_SIZE - 1);
+	mgt_vbase = pci_alloc_consistent(nesdev->pcidev, NES_MGT_QP_COUNT * mgt_mem_size, &mgt_pbase);
+	if (!mgt_vbase) {
+		kfree(mgtvnic);
+		nes_debug(NES_DBG_INIT, "Unable to allocate memory for mgt host descriptor rings\n");
+		return -ENOMEM;
+	}
+
+	nesvnic->mgt_mem_size = NES_MGT_QP_COUNT * mgt_mem_size;
+	nesvnic->mgt_vbase = mgt_vbase;
+	nesvnic->mgt_pbase = mgt_pbase;
+
+	skb_queue_head_init(&nesvnic->mgt_skb_list);
+	init_waitqueue_head(&nesvnic->mgt_wait_queue);
+	nesvnic->mgt_thread = kthread_run(mgt_thread, nesvnic, "nes_mgt_thread");
+
+	for (i = 0; i < NES_MGT_QP_COUNT; i++) {
+		mgtvnic->nesvnic = nesvnic;
+		mgtvnic->mgt.qp_id = nesdev->mac_index + NES_MGT_QP_OFFSET + i;
+		memset(mgt_vbase, 0, mgt_mem_size);
+		nes_debug(NES_DBG_INIT, "Allocated mgt QP structures at %p (phys = %016lX), size = %u.\n",
+			  mgt_vbase, (unsigned long)mgt_pbase, mgt_mem_size);
+
+		vmem = (void *)(((unsigned long)mgt_vbase + (256 - 1)) &
+				~(unsigned long)(256 - 1));
+		pmem = (dma_addr_t)(((unsigned long long)mgt_pbase + (256 - 1)) &
+				    ~(unsigned long long)(256 - 1));
+
+		spin_lock_init(&mgtvnic->mgt.rq_lock);
+
+		/* setup the RQ */
+		mgtvnic->mgt.rq_vbase = vmem;
+		mgtvnic->mgt.rq_pbase = pmem;
+		mgtvnic->mgt.rq_head = 0;
+		mgtvnic->mgt.rq_tail = 0;
+		mgtvnic->mgt.rq_size = NES_MGT_WQ_COUNT;
+
+		/* setup the CQ */
+		vmem += (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_rq_wqe));
+		pmem += (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_rq_wqe));
+
+		mgtvnic->mgt_cq.cq_number = mgtvnic->mgt.qp_id;
+		mgtvnic->mgt_cq.cq_vbase = vmem;
+		mgtvnic->mgt_cq.cq_pbase = pmem;
+		mgtvnic->mgt_cq.cq_head = 0;
+		mgtvnic->mgt_cq.cq_size = NES_MGT_WQ_COUNT;
+
+		mgtvnic->mgt_cq.ce_handler = nes_mgt_ce_handler;
+
+		/* Send CreateCQ request to CQP */
+		spin_lock_irqsave(&nesdev->cqp.lock, flags);
+		cqp_head = nesdev->cqp.sq_head;
+
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+
+		cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] = cpu_to_le32(
+			NES_CQP_CREATE_CQ | NES_CQP_CQ_CEQ_VALID |
+			((u32)mgtvnic->mgt_cq.cq_size << 16));
+		cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX] = cpu_to_le32(
+			mgtvnic->mgt_cq.cq_number | ((u32)nesdev->ceq_index << 16));
+		u64temp = (u64)mgtvnic->mgt_cq.cq_pbase;
+		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_CQ_WQE_PBL_LOW_IDX, u64temp);
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_CQ_CONTEXT_HIGH_IDX] = 0;
+		u64temp = (unsigned long)&mgtvnic->mgt_cq;
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_CQ_CONTEXT_LOW_IDX] = cpu_to_le32((u32)(u64temp >> 1));
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_CQ_CONTEXT_HIGH_IDX] =
+			cpu_to_le32(((u32)((u64temp) >> 33)) & 0x7FFFFFFF);
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_DOORBELL_INDEX_HIGH_IDX] = 0;
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+
+		/* Send CreateQP request to CQP */
+		mgt_context = (void *)(&mgtvnic->mgt_cq.cq_vbase[mgtvnic->mgt_cq.cq_size]);
+		mgt_context->context_words[NES_NIC_CTX_MISC_IDX] =
+			cpu_to_le32((u32)NES_MGT_CTX_SIZE |
+				    ((u32)PCI_FUNC(nesdev->pcidev->devfn) << 12));
+		nes_debug(NES_DBG_INIT, "RX_WINDOW_BUFFER_PAGE_TABLE_SIZE = 0x%08X, RX_WINDOW_BUFFER_SIZE = 0x%08X\n",
+			  nes_read_indexed(nesdev, NES_IDX_RX_WINDOW_BUFFER_PAGE_TABLE_SIZE),
+			  nes_read_indexed(nesdev, NES_IDX_RX_WINDOW_BUFFER_SIZE));
+		if (nes_read_indexed(nesdev, NES_IDX_RX_WINDOW_BUFFER_SIZE) != 0)
+			mgt_context->context_words[NES_NIC_CTX_MISC_IDX] |= cpu_to_le32(NES_NIC_BACK_STORE);
+
+		u64temp = (u64)mgtvnic->mgt.rq_pbase;
+		mgt_context->context_words[NES_NIC_CTX_SQ_LOW_IDX] = cpu_to_le32((u32)u64temp);
+		mgt_context->context_words[NES_NIC_CTX_SQ_HIGH_IDX] = cpu_to_le32((u32)(u64temp >> 32));
+		u64temp = (u64)mgtvnic->mgt.rq_pbase;
+		mgt_context->context_words[NES_NIC_CTX_RQ_LOW_IDX] = cpu_to_le32((u32)u64temp);
+		mgt_context->context_words[NES_NIC_CTX_RQ_HIGH_IDX] = cpu_to_le32((u32)(u64temp >> 32));
+
+		cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] = cpu_to_le32(NES_CQP_CREATE_QP |
+									 NES_CQP_QP_TYPE_NIC);
+		cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX] = cpu_to_le32(mgtvnic->mgt.qp_id);
+		u64temp = (u64)mgtvnic->mgt_cq.cq_pbase +
+			  (mgtvnic->mgt_cq.cq_size * sizeof(struct nes_hw_nic_cqe));
+		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, u64temp);
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+		nesdev->cqp.sq_head = cqp_head;
+
+		barrier();
+
+		/* Ring doorbell (2 WQEs) */
+		nes_write32(nesdev->regs + NES_WQE_ALLOC, 0x02800000 | nesdev->cqp.qp_id);
+
+		spin_unlock_irqrestore(&nesdev->cqp.lock, flags);
+		nes_debug(NES_DBG_INIT, "Waiting for create MGT QP%u to complete.\n",
+			  mgtvnic->mgt.qp_id);
+
+		ret = wait_event_timeout(nesdev->cqp.waitq, (nesdev->cqp.sq_tail == cqp_head),
+					 NES_EVENT_TIMEOUT);
+		nes_debug(NES_DBG_INIT, "Create MGT QP%u completed, wait_event_timeout ret = %u.\n",
+			  mgtvnic->mgt.qp_id, ret);
+		if (!ret) {
+			nes_debug(NES_DBG_INIT, "MGT QP%u create timeout expired\n", mgtvnic->mgt.qp_id);
+			if (i == 0) {
+				pci_free_consistent(nesdev->pcidev, nesvnic->mgt_mem_size, nesvnic->mgt_vbase,
+						    nesvnic->mgt_pbase);
+				kfree(mgtvnic);
+			} else {
+				nes_destroy_mgt(nesvnic);
+			}
+			return -EIO;
+		}
+
+		/* Populate the RQ */
+		for (counter = 0; counter < (NES_MGT_WQ_COUNT - 1); counter++) {
+			skb = dev_alloc_skb(nesvnic->max_frame_size);
+			if (!skb) {
+				nes_debug(NES_DBG_INIT, "%s: out of memory for receive skb\n", netdev->name);
+				return -ENOMEM;
+			}
+
+			skb->dev = netdev;
+
+			pmem = pci_map_single(nesdev->pcidev, skb->data,
+					      nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->busaddr = pmem;
+			cb->maplen = nesvnic->max_frame_size;
+
+			mgt_rqe = &mgtvnic->mgt.rq_vbase[counter];
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] = cpu_to_le32((u32)nesvnic->max_frame_size);
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_3_2_IDX] = 0;
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_LOW_IDX] = cpu_to_le32((u32)pmem);
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX] = cpu_to_le32((u32)((u64)pmem >> 32));
+			mgtvnic->mgt.rx_skb[counter] = skb;
+		}
+
+		init_timer(&mgtvnic->rq_wqes_timer);
+		mgtvnic->rq_wqes_timer.function = nes_mgt_rq_wqes_timeout;
+		mgtvnic->rq_wqes_timer.data = (unsigned long)mgtvnic;
+
+		wqe_count = NES_MGT_WQ_COUNT - 1;
+		mgtvnic->mgt.rq_head = wqe_count;
+		barrier();
+		do {
+			counter = min(wqe_count, ((u32)255));
+			wqe_count -= counter;
+			nes_write32(nesdev->regs + NES_WQE_ALLOC, (counter << 24) | mgtvnic->mgt.qp_id);
+		} while (wqe_count);
+
+		nes_write32(nesdev->regs + NES_CQE_ALLOC, NES_CQE_ALLOC_NOTIFY_NEXT |
+			    mgtvnic->mgt_cq.cq_number);
+		nes_read32(nesdev->regs + NES_CQE_ALLOC);
+
+		mgt_vbase += mgt_mem_size;
+		mgt_pbase += mgt_mem_size;
+		nesvnic->mgtvnic[i] = mgtvnic++;
+	}
+	return 0;
+}
+
+
+void nes_destroy_mgt(struct nes_vnic *nesvnic)
+{
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct nes_vnic_mgt *mgtvnic;
+	struct nes_vnic_mgt *first_mgtvnic;
+	unsigned long flags;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	u32 cqp_head;
+	struct sk_buff *rx_skb;
+	int i;
+	int ret;
+
+	kthread_stop(nesvnic->mgt_thread);
+
+	/* Free remaining NIC receive buffers */
+	first_mgtvnic = nesvnic->mgtvnic[0];
+	for (i = 0; i < NES_MGT_QP_COUNT; i++) {
+		mgtvnic = nesvnic->mgtvnic[i];
+		if (mgtvnic == NULL)
+			continue;
+
+		while (mgtvnic->mgt.rq_head != mgtvnic->mgt.rq_tail) {
+			rx_skb = mgtvnic->mgt.rx_skb[mgtvnic->mgt.rq_tail];
+			nes_mgt_free_skb(nesdev, rx_skb, PCI_DMA_FROMDEVICE);
+			mgtvnic->mgt.rq_tail++;
+			mgtvnic->mgt.rq_tail &= (mgtvnic->mgt.rq_size - 1);
+		}
+
+		spin_lock_irqsave(&nesdev->cqp.lock, flags);
+
+		/* Destroy NIC QP */
+		cqp_head = nesdev->cqp.sq_head;
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX,
+				    (NES_CQP_DESTROY_QP | NES_CQP_QP_TYPE_NIC));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX,
+				    mgtvnic->mgt.qp_id);
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+
+		/* Destroy NIC CQ */
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX,
+				    (NES_CQP_DESTROY_CQ | ((u32)mgtvnic->mgt_cq.cq_size << 16)));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX,
+				    (mgtvnic->mgt_cq.cq_number | ((u32)nesdev->ceq_index << 16)));
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+
+		nesdev->cqp.sq_head = cqp_head;
+		barrier();
+
+		/* Ring doorbell (2 WQEs) */
+		nes_write32(nesdev->regs + NES_WQE_ALLOC, 0x02800000 | nesdev->cqp.qp_id);
+
+		spin_unlock_irqrestore(&nesdev->cqp.lock, flags);
+		nes_debug(NES_DBG_SHUTDOWN, "Waiting for CQP, cqp_head=%u, cqp.sq_head=%u,"
+			  " cqp.sq_tail=%u, cqp.sq_size=%u\n",
+			  cqp_head, nesdev->cqp.sq_head,
+			  nesdev->cqp.sq_tail, nesdev->cqp.sq_size);
+
+		ret = wait_event_timeout(nesdev->cqp.waitq, (nesdev->cqp.sq_tail == cqp_head),
+					 NES_EVENT_TIMEOUT);
+
+		nes_debug(NES_DBG_SHUTDOWN, "Destroy MGT QP returned, wait_event_timeout ret = %u, cqp_head=%u,"
+			  " cqp.sq_head=%u, cqp.sq_tail=%u\n",
+			  ret, cqp_head, nesdev->cqp.sq_head, nesdev->cqp.sq_tail);
+		if (!ret)
+			nes_debug(NES_DBG_SHUTDOWN, "MGT QP%u destroy timeout expired\n",
+				  mgtvnic->mgt.qp_id);
+
+		nesvnic->mgtvnic[i] = NULL;
+	}
+
+	if (nesvnic->mgt_vbase) {
+		pci_free_consistent(nesdev->pcidev, nesvnic->mgt_mem_size, nesvnic->mgt_vbase,
+				    nesvnic->mgt_pbase);
+		nesvnic->mgt_vbase = NULL;
+		nesvnic->mgt_pbase = 0;
+	}
+
+	kfree(first_mgtvnic);
+}

drivers/infiniband/hw/nes/nes_mgt.h

@@ -0,0 +1,97 @@
+/*
+* Copyright (c) 2010 Intel-NE, 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.
+*/
+
+#ifndef __NES_MGT_H
+#define __NES_MGT_H
+
+#define MPA_FRAMING 6	/* length is 2 bytes, crc is 4 bytes */
+
+int nes_init_mgt_qp(struct nes_device *nesdev, struct net_device *netdev, struct nes_vnic *nesvnic);
+void nes_queue_mgt_skbs(struct sk_buff *skb, struct nes_vnic *nesvnic, struct nes_qp *nesqp);
+void nes_destroy_mgt(struct nes_vnic *nesvnic);
+void nes_destroy_pau_qp(struct nes_device *nesdev, struct nes_qp *nesqp);
+
+struct nes_hw_mgt {
+	struct nes_hw_nic_rq_wqe *rq_vbase;	/* virtual address of rq */
+	dma_addr_t rq_pbase;			/* PCI memory for host rings */
+	struct sk_buff *rx_skb[NES_NIC_WQ_SIZE];
+	u16 qp_id;
+	u16 sq_head;
+	u16 rq_head;
+	u16 rq_tail;
+	u16 rq_size;
+	u8 replenishing_rq;
+	u8 reserved;
+	spinlock_t rq_lock;
+};
+
+struct nes_vnic_mgt {
+	struct nes_vnic        *nesvnic;
+	struct nes_hw_mgt      mgt;
+	struct nes_hw_nic_cq   mgt_cq;
+	atomic_t               rx_skbs_needed;
+	struct timer_list      rq_wqes_timer;
+	atomic_t               rx_skb_timer_running;
+};
+
+#define MAX_FPDU_FRAGS 4
+struct pau_fpdu_frag {
+	struct sk_buff         *skb;
+	u64                    physaddr;
+	u32                    frag_len;
+	bool                   cmplt;
+};
+
+struct pau_fpdu_info {
+	struct nes_qp          *nesqp;
+	struct nes_cqp_request *cqp_request;
+	void                   *hdr_vbase;
+	dma_addr_t             hdr_pbase;
+	int                    hdr_len;
+	u16                    data_len;
+	u16                    frag_cnt;
+	struct pau_fpdu_frag   frags[MAX_FPDU_FRAGS];
+};
+
+enum pau_qh_state {
+	PAU_DEL_QH,
+	PAU_ADD_LB_QH,
+	PAU_READY
+};
+
+struct pau_qh_chg {
+	struct nes_device *nesdev;
+	struct nes_vnic *nesvnic;
+	struct nes_qp *nesqp;
+};
+
+#endif          /* __NES_MGT_H */

drivers/infiniband/hw/nes/nes_nic.c

@@ -1091,6 +1091,8 @@ static const char nes_ethtool_stringset[][ETH_GSTRING_LEN] = {
 	"LRO aggregated",
 	"LRO flushed",
 	"LRO no_desc",
+	"PAU CreateQPs",
+	"PAU DestroyQPs",
 };
 #define NES_ETHTOOL_STAT_COUNT  ARRAY_SIZE(nes_ethtool_stringset)
 
@@ -1306,6 +1308,8 @@ static void nes_netdev_get_ethtool_stats(struct net_device *netdev,
 	target_stat_values[++index] = nesvnic->lro_mgr.stats.aggregated;
 	target_stat_values[++index] = nesvnic->lro_mgr.stats.flushed;
 	target_stat_values[++index] = nesvnic->lro_mgr.stats.no_desc;
+	target_stat_values[++index] = atomic_read(&pau_qps_created);
+	target_stat_values[++index] = atomic_read(&pau_qps_destroyed);
 }
 
 /**

drivers/infiniband/hw/nes/nes_utils.c

@@ -51,13 +51,34 @@
 
 #include "nes.h"
 
-
-
 static u16 nes_read16_eeprom(void __iomem *addr, u16 offset);
 
 u32 mh_detected;
 u32 mh_pauses_sent;
 
+u32 nes_set_pau(struct nes_device *nesdev)
+{
+	u32 ret = 0;
+	u32 counter;
+
+	nes_write_indexed(nesdev, NES_IDX_GPR2, NES_ENABLE_PAU);
+	nes_write_indexed(nesdev, NES_IDX_GPR_TRIGGER, 1);
+
+	for (counter = 0; counter < NES_PAU_COUNTER; counter++) {
+		udelay(30);
+		if (!nes_read_indexed(nesdev, NES_IDX_GPR2)) {
+			printk(KERN_INFO PFX "PAU is supported.\n");
+			break;
+		}
+		nes_write_indexed(nesdev, NES_IDX_GPR_TRIGGER, 1);
+	}
+	if (counter == NES_PAU_COUNTER) {
+		printk(KERN_INFO PFX "PAU is not supported.\n");
+		return -EPERM;
+	}
+	return ret;
+}
+
 /**
  * nes_read_eeprom_values -
  */
@@ -187,6 +208,11 @@ int nes_read_eeprom_values(struct nes_device *nesdev, struct nes_adapter *nesada
 		if (((major_ver == 3) && (minor_ver >= 16)) || (major_ver > 3))
 			nesadapter->send_term_ok = 1;
 
+		if (nes_drv_opt & NES_DRV_OPT_ENABLE_PAU) {
+			if (!nes_set_pau(nesdev))
+				nesadapter->allow_unaligned_fpdus = 1;
+		}
+
 		nesadapter->firmware_version = (((u32)(u8)(eeprom_data>>8))  <<  16) +
 				(u32)((u8)eeprom_data);
 
@@ -594,6 +620,7 @@ void nes_put_cqp_request(struct nes_device *nesdev,
 		nes_free_cqp_request(nesdev, cqp_request);
 }
 
+
 /**
  * nes_post_cqp_request
  */
@@ -604,6 +631,8 @@ void nes_post_cqp_request(struct nes_device *nesdev,
 	unsigned long flags;
 	u32 cqp_head;
 	u64 u64temp;
+	u32 opcode;
+	int ctx_index = NES_CQP_WQE_COMP_CTX_LOW_IDX;
 
 	spin_lock_irqsave(&nesdev->cqp.lock, flags);
 
@@ -614,17 +643,20 @@ void nes_post_cqp_request(struct nes_device *nesdev,
 		nesdev->cqp.sq_head &= nesdev->cqp.sq_size-1;
 		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
 		memcpy(cqp_wqe, &cqp_request->cqp_wqe, sizeof(*cqp_wqe));
+		opcode = le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX]);
+		if ((opcode & NES_CQP_OPCODE_MASK) == NES_CQP_DOWNLOAD_SEGMENT)
+			ctx_index = NES_CQP_WQE_DL_COMP_CTX_LOW_IDX;
 		barrier();
 		u64temp = (unsigned long)cqp_request;
-		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_COMP_SCRATCH_LOW_IDX,
-				    u64temp);
+		set_wqe_64bit_value(cqp_wqe->wqe_words, ctx_index, u64temp);
 		nes_debug(NES_DBG_CQP, "CQP request (opcode 0x%02X), line 1 = 0x%08X put on CQPs SQ,"
-				" request = %p, cqp_head = %u, cqp_tail = %u, cqp_size = %u,"
-				" waiting = %d, refcount = %d.\n",
-				le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX])&0x3f,
-				le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX]), cqp_request,
-				nesdev->cqp.sq_head, nesdev->cqp.sq_tail, nesdev->cqp.sq_size,
-				cqp_request->waiting, atomic_read(&cqp_request->refcount));
+			" request = %p, cqp_head = %u, cqp_tail = %u, cqp_size = %u,"
+			" waiting = %d, refcount = %d.\n",
+			opcode & NES_CQP_OPCODE_MASK,
+			le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX]), cqp_request,
+			nesdev->cqp.sq_head, nesdev->cqp.sq_tail, nesdev->cqp.sq_size,
+			cqp_request->waiting, atomic_read(&cqp_request->refcount));
+
 		barrier();
 
 		/* Ring doorbell (1 WQEs) */
@@ -645,7 +677,6 @@ void nes_post_cqp_request(struct nes_device *nesdev,
 	return;
 }
 
-
 /**
  * nes_arp_table
  */

drivers/infiniband/hw/nes/nes_verbs.c

@@ -1458,7 +1458,7 @@ static int nes_destroy_qp(struct ib_qp *ibqp)
 	struct ib_qp_attr attr;
 	struct iw_cm_id *cm_id;
 	struct iw_cm_event cm_event;
-	int ret;
+	int ret = 0;
 
 	atomic_inc(&sw_qps_destroyed);
 	nesqp->destroyed = 1;
@@ -1511,7 +1511,6 @@ static int nes_destroy_qp(struct ib_qp *ibqp)
 		if ((nesqp->nesrcq) && (nesqp->nesrcq != nesqp->nesscq))
 			nes_clean_cq(nesqp, nesqp->nesrcq);
 	}
-
 	nes_rem_ref(&nesqp->ibqp);
 	return 0;
 }
@@ -2338,8 +2337,10 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	skip_pages = ((u32)region->offset) >> 12;
 
-	if (ib_copy_from_udata(&req, udata, sizeof(req)))
+	if (ib_copy_from_udata(&req, udata, sizeof(req))) {
+		ib_umem_release(region);
 		return ERR_PTR(-EFAULT);
+	}
 	nes_debug(NES_DBG_MR, "Memory Registration type = %08X.\n", req.reg_type);
 
 	switch (req.reg_type) {
@@ -2631,6 +2632,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			return &nesmr->ibmr;
 	}
 
+	ib_umem_release(region);
 	return ERR_PTR(-ENOSYS);
 }
 

drivers/infiniband/hw/nes/nes_verbs.h

@@ -139,7 +139,8 @@ struct nes_qp {
 	struct nes_cq         *nesrcq;
 	struct nes_pd         *nespd;
 	void *cm_node; /* handle of the node this QP is associated with */
-	struct ietf_mpa_frame *ietf_frame;
+	void                  *ietf_frame;
+	u8                    ietf_frame_size;
 	dma_addr_t            ietf_frame_pbase;
 	struct ib_mr          *lsmm_mr;
 	struct nes_hw_qp      hwqp;
@@ -154,6 +155,7 @@ struct nes_qp {
 	u32                   mmap_sq_db_index;
 	u32                   mmap_rq_db_index;
 	spinlock_t            lock;
+	spinlock_t            pau_lock;
 	struct nes_qp_context *nesqp_context;
 	dma_addr_t            nesqp_context_pbase;
 	void	              *pbl_vbase;
@@ -161,6 +163,8 @@ struct nes_qp {
 	struct page           *page;
 	struct timer_list     terminate_timer;
 	enum ib_event_type    terminate_eventtype;
+	struct sk_buff_head   pau_list;
+	u32                   pau_rcv_nxt;
 	u16                   active_conn:1;
 	u16                   skip_lsmm:1;
 	u16                   user_mode:1;
@@ -168,7 +172,8 @@ struct nes_qp {
 	u16                   flush_issued:1;
 	u16                   destroyed:1;
 	u16                   sig_all:1;
-	u16                   rsvd:9;
+	u16                   pau_mode:1;
+	u16                   rsvd:8;
 	u16                   private_data_len;
 	u16                   term_sq_flush_code;
 	u16                   term_rq_flush_code;
@@ -176,5 +181,8 @@ struct nes_qp {
 	u8                    hw_tcp_state;
 	u8                    term_flags;
 	u8                    sq_kmapped;
+	u8                    pau_busy;
+	u8                    pau_pending;
+	u8                    pau_state;
 };
 #endif			/* NES_VERBS_H */

drivers/infiniband/hw/qib/qib.h

@@ -171,7 +171,9 @@ struct qib_ctxtdata {
 	/* how many alloc_pages() chunks in rcvegrbuf_pages */
 	u32 rcvegrbuf_chunks;
 	/* how many egrbufs per chunk */
-	u32 rcvegrbufs_perchunk;
+	u16 rcvegrbufs_perchunk;
+	/* ilog2 of above */
+	u16 rcvegrbufs_perchunk_shift;
 	/* order for rcvegrbuf_pages */
 	size_t rcvegrbuf_size;
 	/* rcvhdrq size (for freeing) */
@@ -221,6 +223,9 @@ struct qib_ctxtdata {
 	/* ctxt rcvhdrq head offset */
 	u32 head;
 	u32 pkt_count;
+	/* lookaside fields */
+	struct qib_qp *lookaside_qp;
+	u32 lookaside_qpn;
 	/* QPs waiting for context processing */
 	struct list_head qp_wait_list;
 };
@@ -807,6 +812,10 @@ struct qib_devdata {
 	 * supports, less gives more pio bufs/ctxt, etc.
 	 */
 	u32 cfgctxts;
+	/*
+	 * number of ctxts available for PSM open
+	 */
+	u32 freectxts;
 
 	/*
 	 * hint that we should update pioavailshadow before
@@ -936,7 +945,9 @@ struct qib_devdata {
 	/* chip address space used by 4k pio buffers */
 	u32 align4k;
 	/* size of each rcvegrbuffer */
-	u32 rcvegrbufsize;
+	u16 rcvegrbufsize;
+	/* log2 of above */
+	u16 rcvegrbufsize_shift;
 	/* localbus width (1, 2,4,8,16,32) from config space  */
 	u32 lbus_width;
 	/* localbus speed in MHz */

drivers/infiniband/hw/qib/qib_driver.c

@@ -279,10 +279,10 @@ bail:
  */
 static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail)
 {
-	const u32 chunk = etail / rcd->rcvegrbufs_perchunk;
-	const u32 idx =  etail % rcd->rcvegrbufs_perchunk;
+	const u32 chunk = etail >> rcd->rcvegrbufs_perchunk_shift;
+	const u32 idx =  etail & ((u32)rcd->rcvegrbufs_perchunk - 1);
 
-	return rcd->rcvegrbuf[chunk] + idx * rcd->dd->rcvegrbufsize;
+	return rcd->rcvegrbuf[chunk] + (idx << rcd->dd->rcvegrbufsize_shift);
 }
 
 /*
@@ -310,7 +310,6 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
 		u32 opcode;
 		u32 psn;
 		int diff;
-		unsigned long flags;
 
 		/* Sanity check packet */
 		if (tlen < 24)
@@ -365,7 +364,6 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
 
 			switch (qp->ibqp.qp_type) {
 			case IB_QPT_RC:
-				spin_lock_irqsave(&qp->s_lock, flags);
 				ruc_res =
 					qib_ruc_check_hdr(
 						ibp, hdr,
@@ -373,11 +371,8 @@ static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
 						qp,
 						be32_to_cpu(ohdr->bth[0]));
 				if (ruc_res) {
-					spin_unlock_irqrestore(&qp->s_lock,
-							       flags);
 					goto unlock;
 				}
-				spin_unlock_irqrestore(&qp->s_lock, flags);
 
 				/* Only deal with RDMA Writes for now */
 				if (opcode <
@@ -547,6 +542,15 @@ move_along:
 			updegr = 0;
 		}
 	}
+	/*
+	 * Notify qib_destroy_qp() if it is waiting
+	 * for lookaside_qp to finish.
+	 */
+	if (rcd->lookaside_qp) {
+		if (atomic_dec_and_test(&rcd->lookaside_qp->refcount))
+			wake_up(&rcd->lookaside_qp->wait);
+		rcd->lookaside_qp = NULL;
+	}
 
 	rcd->head = l;
 	rcd->pkt_count += i;

drivers/infiniband/hw/qib/qib_file_ops.c

@@ -1284,6 +1284,7 @@ static int setup_ctxt(struct qib_pportdata *ppd, int ctxt,
 	strlcpy(rcd->comm, current->comm, sizeof(rcd->comm));
 	ctxt_fp(fp) = rcd;
 	qib_stats.sps_ctxts++;
+	dd->freectxts++;
 	ret = 0;
 	goto bail;
 
@@ -1792,6 +1793,7 @@ static int qib_close(struct inode *in, struct file *fp)
 		if (dd->pageshadow)
 			unlock_expected_tids(rcd);
 		qib_stats.sps_ctxts--;
+		dd->freectxts--;
 	}
 
 	mutex_unlock(&qib_mutex);

drivers/infiniband/hw/qib/qib_iba6120.c

@@ -3273,6 +3273,8 @@ static int init_6120_variables(struct qib_devdata *dd)
 	/* we always allocate at least 2048 bytes for eager buffers */
 	ret = ib_mtu_enum_to_int(qib_ibmtu);
 	dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU;
+	BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
+	dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
 
 	qib_6120_tidtemplate(dd);
 

drivers/infiniband/hw/qib/qib_iba7220.c

@@ -4085,6 +4085,8 @@ static int qib_init_7220_variables(struct qib_devdata *dd)
 	/* we always allocate at least 2048 bytes for eager buffers */
 	ret = ib_mtu_enum_to_int(qib_ibmtu);
 	dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU;
+	BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
+	dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
 
 	qib_7220_tidtemplate(dd);
 

drivers/infiniband/hw/qib/qib_iba7322.c

@@ -2310,12 +2310,15 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd)
 	val = ppd->cpspec->ibcctrl_a | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE <<
 		QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
 
+	ppd->cpspec->ibcctrl_a = val;
 	/*
 	 * Reset the PCS interface to the serdes (and also ibc, which is still
 	 * in reset from above).  Writes new value of ibcctrl_a as last step.
 	 */
 	qib_7322_mini_pcs_reset(ppd);
 	qib_write_kreg(dd, kr_scratch, 0ULL);
+	/* clear the linkinit cmds */
+	ppd->cpspec->ibcctrl_a &= ~SYM_MASK(IBCCtrlA_0, LinkInitCmd);
 
 	if (!ppd->cpspec->ibcctrl_b) {
 		unsigned lse = ppd->link_speed_enabled;
@@ -2387,11 +2390,6 @@ static int qib_7322_bringup_serdes(struct qib_pportdata *ppd)
 	qib_write_kreg_port(ppd, krp_rcvctrl, ppd->p_rcvctrl);
 	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
 
-	/* Hold the link state machine for mezz boards */
-	if (IS_QMH(dd) || IS_QME(dd))
-		qib_set_ib_7322_lstate(ppd, 0,
-				       QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
-
 	/* Also enable IBSTATUSCHG interrupt.  */
 	val = qib_read_kreg_port(ppd, krp_errmask);
 	qib_write_kreg_port(ppd, krp_errmask,
@@ -2853,9 +2851,8 @@ static irqreturn_t qib_7322intr(int irq, void *data)
 		for (i = 0; i < dd->first_user_ctxt; i++) {
 			if (ctxtrbits & rmask) {
 				ctxtrbits &= ~rmask;
-				if (dd->rcd[i]) {
+				if (dd->rcd[i])
 					qib_kreceive(dd->rcd[i], NULL, &npkts);
-				}
 			}
 			rmask <<= 1;
 		}
@@ -5230,6 +5227,8 @@ static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
 				     QIBL_IB_AUTONEG_INPROG)))
 			set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled);
 		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
+			struct qib_qsfp_data *qd =
+				&ppd->cpspec->qsfp_data;
 			/* unlock the Tx settings, speed may change */
 			qib_write_kreg_port(ppd, krp_tx_deemph_override,
 				SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
@@ -5237,6 +5236,12 @@ static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
 			qib_cancel_sends(ppd);
 			/* on link down, ensure sane pcs state */
 			qib_7322_mini_pcs_reset(ppd);
+			/* schedule the qsfp refresh which should turn the link
+			   off */
+			if (ppd->dd->flags & QIB_HAS_QSFP) {
+				qd->t_insert = get_jiffies_64();
+				schedule_work(&qd->work);
+			}
 			spin_lock_irqsave(&ppd->sdma_lock, flags);
 			if (__qib_sdma_running(ppd))
 				__qib_sdma_process_event(ppd,
@@ -5587,43 +5592,79 @@ static void qsfp_7322_event(struct work_struct *work)
 	struct qib_qsfp_data *qd;
 	struct qib_pportdata *ppd;
 	u64 pwrup;
+	unsigned long flags;
 	int ret;
 	u32 le2;
 
 	qd = container_of(work, struct qib_qsfp_data, work);
 	ppd = qd->ppd;
-	pwrup = qd->t_insert + msecs_to_jiffies(QSFP_PWR_LAG_MSEC);
+	pwrup = qd->t_insert +
+		msecs_to_jiffies(QSFP_PWR_LAG_MSEC - QSFP_MODPRS_LAG_MSEC);
 
-	/*
-	 * Some QSFP's not only do not respond until the full power-up
-	 * time, but may behave badly if we try. So hold off responding
-	 * to insertion.
-	 */
-	while (1) {
-		u64 now = get_jiffies_64();
-		if (time_after64(now, pwrup))
-			break;
-		msleep(20);
-	}
-	ret = qib_refresh_qsfp_cache(ppd, &qd->cache);
-	/*
-	 * Need to change LE2 back to defaults if we couldn't
-	 * read the cable type (to handle cable swaps), so do this
-	 * even on failure to read cable information.  We don't
-	 * get here for QME, so IS_QME check not needed here.
-	 */
-	if (!ret && !ppd->dd->cspec->r1) {
-		if (QSFP_IS_ACTIVE_FAR(qd->cache.tech))
-			le2 = LE2_QME;
-		else if (qd->cache.atten[1] >= qib_long_atten &&
-			 QSFP_IS_CU(qd->cache.tech))
-			le2 = LE2_5m;
-		else
+	/* Delay for 20 msecs to allow ModPrs resistor to setup */
+	mdelay(QSFP_MODPRS_LAG_MSEC);
+
+	if (!qib_qsfp_mod_present(ppd)) {
+		ppd->cpspec->qsfp_data.modpresent = 0;
+		/* Set the physical link to disabled */
+		qib_set_ib_7322_lstate(ppd, 0,
+				       QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	} else {
+		/*
+		 * Some QSFP's not only do not respond until the full power-up
+		 * time, but may behave badly if we try. So hold off responding
+		 * to insertion.
+		 */
+		while (1) {
+			u64 now = get_jiffies_64();
+			if (time_after64(now, pwrup))
+				break;
+			msleep(20);
+		}
+
+		ret = qib_refresh_qsfp_cache(ppd, &qd->cache);
+
+		/*
+		 * Need to change LE2 back to defaults if we couldn't
+		 * read the cable type (to handle cable swaps), so do this
+		 * even on failure to read cable information.  We don't
+		 * get here for QME, so IS_QME check not needed here.
+		 */
+		if (!ret && !ppd->dd->cspec->r1) {
+			if (QSFP_IS_ACTIVE_FAR(qd->cache.tech))
+				le2 = LE2_QME;
+			else if (qd->cache.atten[1] >= qib_long_atten &&
+				 QSFP_IS_CU(qd->cache.tech))
+				le2 = LE2_5m;
+			else
+				le2 = LE2_DEFAULT;
+		} else
 			le2 = LE2_DEFAULT;
-	} else
-		le2 = LE2_DEFAULT;
-	ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7));
-	init_txdds_table(ppd, 0);
+		ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7));
+		/*
+		 * We always change parameteters, since we can choose
+		 * values for cables without eeproms, and the cable may have
+		 * changed from a cable with full or partial eeprom content
+		 * to one with partial or no content.
+		 */
+		init_txdds_table(ppd, 0);
+		/* The physical link is being re-enabled only when the
+		 * previous state was DISABLED and the VALID bit is not
+		 * set. This should only happen when  the cable has been
+		 * physically pulled. */
+		if (!ppd->cpspec->qsfp_data.modpresent &&
+		    (ppd->lflags & (QIBL_LINKV | QIBL_IB_LINK_DISABLED))) {
+			ppd->cpspec->qsfp_data.modpresent = 1;
+			qib_set_ib_7322_lstate(ppd, 0,
+				QLOGIC_IB_IBCC_LINKINITCMD_SLEEP);
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags |= QIBL_LINKV;
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		}
+	}
 }
 
 /*
@@ -5727,7 +5768,8 @@ static void set_no_qsfp_atten(struct qib_devdata *dd, int change)
 			/* now change the IBC and serdes, overriding generic */
 			init_txdds_table(ppd, 1);
 			/* Re-enable the physical state machine on mezz boards
-			 * now that the correct settings have been set. */
+			 * now that the correct settings have been set.
+			 * QSFP boards are handles by the QSFP event handler */
 			if (IS_QMH(dd) || IS_QME(dd))
 				qib_set_ib_7322_lstate(ppd, 0,
 					    QLOGIC_IB_IBCC_LINKINITCMD_SLEEP);
@@ -6205,6 +6247,8 @@ static int qib_init_7322_variables(struct qib_devdata *dd)
 
 	/* we always allocate at least 2048 bytes for eager buffers */
 	dd->rcvegrbufsize = max(mtu, 2048);
+	BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
+	dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
 
 	qib_7322_tidtemplate(dd);
 
@@ -7147,7 +7191,8 @@ static void find_best_ent(struct qib_pportdata *ppd,
 		}
 	}
 
-	/* Lookup serdes setting by cable type and attenuation */
+	/* Active cables don't have attenuation so we only set SERDES
+	 * settings to account for the attenuation of the board traces. */
 	if (!override && QSFP_IS_ACTIVE(qd->tech)) {
 		*sdr_dds = txdds_sdr + ppd->dd->board_atten;
 		*ddr_dds = txdds_ddr + ppd->dd->board_atten;
@@ -7464,12 +7509,6 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
 	u32 le_val, rxcaldone;
 	int chan, chan_done = (1 << SERDES_CHANS) - 1;
 
-	/*
-	 * Initialize the Tx DDS tables.  Also done every QSFP event,
-	 * for adapters with QSFP
-	 */
-	init_txdds_table(ppd, 0);
-
 	/* Clear cmode-override, may be set from older driver */
 	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14);
 
@@ -7655,6 +7694,12 @@ static int serdes_7322_init_new(struct qib_pportdata *ppd)
 	/* VGA output common mode */
 	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
 
+	/*
+	 * Initialize the Tx DDS tables.  Also done every QSFP event,
+	 * for adapters with QSFP
+	 */
+	init_txdds_table(ppd, 0);
+
 	return 0;
 }
 

drivers/infiniband/hw/qib/qib_init.c

@@ -183,6 +183,9 @@ struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt)
 		rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt +
 			rcd->rcvegrbufs_perchunk - 1) /
 			rcd->rcvegrbufs_perchunk;
+		BUG_ON(!is_power_of_2(rcd->rcvegrbufs_perchunk));
+		rcd->rcvegrbufs_perchunk_shift =
+			ilog2(rcd->rcvegrbufs_perchunk);
 	}
 	return rcd;
 }
@@ -398,6 +401,7 @@ static void enable_chip(struct qib_devdata *dd)
 		if (rcd)
 			dd->f_rcvctrl(rcd->ppd, rcvmask, i);
 	}
+	dd->freectxts = dd->cfgctxts - dd->first_user_ctxt;
 }
 
 static void verify_interrupt(unsigned long opaque)
@@ -581,10 +585,6 @@ int qib_init(struct qib_devdata *dd, int reinit)
 			continue;
 		}
 
-		/* let link come up, and enable IBC */
-		spin_lock_irqsave(&ppd->lflags_lock, flags);
-		ppd->lflags &= ~QIBL_IB_LINK_DISABLED;
-		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
 		portok++;
 	}
 

drivers/infiniband/hw/qib/qib_qp.c

@@ -34,6 +34,7 @@
 
 #include <linux/err.h>
 #include <linux/vmalloc.h>
+#include <linux/jhash.h>
 
 #include "qib.h"
 
@@ -204,6 +205,13 @@ static void free_qpn(struct qib_qpn_table *qpt, u32 qpn)
 		clear_bit(qpn & BITS_PER_PAGE_MASK, map->page);
 }
 
+static inline unsigned qpn_hash(struct qib_ibdev *dev, u32 qpn)
+{
+	return jhash_1word(qpn, dev->qp_rnd) &
+		(dev->qp_table_size - 1);
+}
+
+
 /*
  * Put the QP into the hash table.
  * The hash table holds a reference to the QP.
@@ -211,22 +219,23 @@ static void free_qpn(struct qib_qpn_table *qpt, u32 qpn)
 static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp)
 {
 	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
-	unsigned n = qp->ibqp.qp_num % dev->qp_table_size;
 	unsigned long flags;
+	unsigned n = qpn_hash(dev, qp->ibqp.qp_num);
 
 	spin_lock_irqsave(&dev->qpt_lock, flags);
+	atomic_inc(&qp->refcount);
 
 	if (qp->ibqp.qp_num == 0)
-		ibp->qp0 = qp;
+		rcu_assign_pointer(ibp->qp0, qp);
 	else if (qp->ibqp.qp_num == 1)
-		ibp->qp1 = qp;
+		rcu_assign_pointer(ibp->qp1, qp);
 	else {
 		qp->next = dev->qp_table[n];
-		dev->qp_table[n] = qp;
+		rcu_assign_pointer(dev->qp_table[n], qp);
 	}
-	atomic_inc(&qp->refcount);
 
 	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	synchronize_rcu();
 }
 
 /*
@@ -236,29 +245,32 @@ static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp)
 static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp)
 {
 	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
-	struct qib_qp *q, **qpp;
+	unsigned n = qpn_hash(dev, qp->ibqp.qp_num);
 	unsigned long flags;
 
-	qpp = &dev->qp_table[qp->ibqp.qp_num % dev->qp_table_size];
-
 	spin_lock_irqsave(&dev->qpt_lock, flags);
 
 	if (ibp->qp0 == qp) {
-		ibp->qp0 = NULL;
 		atomic_dec(&qp->refcount);
+		rcu_assign_pointer(ibp->qp0, NULL);
 	} else if (ibp->qp1 == qp) {
-		ibp->qp1 = NULL;
 		atomic_dec(&qp->refcount);
-	} else
+		rcu_assign_pointer(ibp->qp1, NULL);
+	} else {
+		struct qib_qp *q, **qpp;
+
+		qpp = &dev->qp_table[n];
 		for (; (q = *qpp) != NULL; qpp = &q->next)
 			if (q == qp) {
-				*qpp = qp->next;
-				qp->next = NULL;
 				atomic_dec(&qp->refcount);
+				rcu_assign_pointer(*qpp, qp->next);
+				qp->next = NULL;
 				break;
 			}
+	}
 
 	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	synchronize_rcu();
 }
 
 /**
@@ -280,21 +292,24 @@ unsigned qib_free_all_qps(struct qib_devdata *dd)
 
 		if (!qib_mcast_tree_empty(ibp))
 			qp_inuse++;
-		if (ibp->qp0)
+		rcu_read_lock();
+		if (rcu_dereference(ibp->qp0))
 			qp_inuse++;
-		if (ibp->qp1)
+		if (rcu_dereference(ibp->qp1))
 			qp_inuse++;
+		rcu_read_unlock();
 	}
 
 	spin_lock_irqsave(&dev->qpt_lock, flags);
 	for (n = 0; n < dev->qp_table_size; n++) {
 		qp = dev->qp_table[n];
-		dev->qp_table[n] = NULL;
+		rcu_assign_pointer(dev->qp_table[n], NULL);
 
 		for (; qp; qp = qp->next)
 			qp_inuse++;
 	}
 	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	synchronize_rcu();
 
 	return qp_inuse;
 }
@@ -309,25 +324,28 @@ unsigned qib_free_all_qps(struct qib_devdata *dd)
  */
 struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn)
 {
-	struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev;
-	unsigned long flags;
-	struct qib_qp *qp;
+	struct qib_qp *qp = NULL;
 
-	spin_lock_irqsave(&dev->qpt_lock, flags);
+	if (unlikely(qpn <= 1)) {
+		rcu_read_lock();
+		if (qpn == 0)
+			qp = rcu_dereference(ibp->qp0);
+		else
+			qp = rcu_dereference(ibp->qp1);
+	} else {
+		struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev;
+		unsigned n = qpn_hash(dev, qpn);
 
-	if (qpn == 0)
-		qp = ibp->qp0;
-	else if (qpn == 1)
-		qp = ibp->qp1;
-	else
-		for (qp = dev->qp_table[qpn % dev->qp_table_size]; qp;
-		     qp = qp->next)
+		rcu_read_lock();
+		for (qp = dev->qp_table[n]; rcu_dereference(qp); qp = qp->next)
 			if (qp->ibqp.qp_num == qpn)
 				break;
+	}
 	if (qp)
-		atomic_inc(&qp->refcount);
+		if (unlikely(!atomic_inc_not_zero(&qp->refcount)))
+			qp = NULL;
 
-	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	rcu_read_unlock();
 	return qp;
 }
 
@@ -765,8 +783,10 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 		}
 	}
 
-	if (attr_mask & IB_QP_PATH_MTU)
+	if (attr_mask & IB_QP_PATH_MTU) {
 		qp->path_mtu = pmtu;
+		qp->pmtu = ib_mtu_enum_to_int(pmtu);
+	}
 
 	if (attr_mask & IB_QP_RETRY_CNT) {
 		qp->s_retry_cnt = attr->retry_cnt;
@@ -781,8 +801,12 @@ int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	if (attr_mask & IB_QP_MIN_RNR_TIMER)
 		qp->r_min_rnr_timer = attr->min_rnr_timer;
 
-	if (attr_mask & IB_QP_TIMEOUT)
+	if (attr_mask & IB_QP_TIMEOUT) {
 		qp->timeout = attr->timeout;
+		qp->timeout_jiffies =
+			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
+				1000UL);
+	}
 
 	if (attr_mask & IB_QP_QKEY)
 		qp->qkey = attr->qkey;
@@ -1013,6 +1037,10 @@ struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
 			ret = ERR_PTR(-ENOMEM);
 			goto bail_swq;
 		}
+		RCU_INIT_POINTER(qp->next, NULL);
+		qp->timeout_jiffies =
+			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
+				1000UL);
 		if (init_attr->srq)
 			sz = 0;
 		else {

drivers/infiniband/hw/qib/qib_qsfp.c

@@ -273,18 +273,12 @@ int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp)
 	int ret;
 	int idx;
 	u16 cks;
-	u32 mask;
 	u8 peek[4];
 
 	/* ensure sane contents on invalid reads, for cable swaps */
 	memset(cp, 0, sizeof(*cp));
 
-	mask = QSFP_GPIO_MOD_PRS_N;
-	if (ppd->hw_pidx)
-		mask <<= QSFP_GPIO_PORT2_SHIFT;
-
-	ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
-	if (ret & mask) {
+	if (!qib_qsfp_mod_present(ppd)) {
 		ret = -ENODEV;
 		goto bail;
 	}
@@ -444,6 +438,19 @@ const char * const qib_qsfp_devtech[16] = {
 
 static const char *pwr_codes = "1.5W2.0W2.5W3.5W";
 
+int qib_qsfp_mod_present(struct qib_pportdata *ppd)
+{
+	u32 mask;
+	int ret;
+
+	mask = QSFP_GPIO_MOD_PRS_N <<
+		(ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT);
+	ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
+
+	return !((ret & mask) >>
+		 ((ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT) + 3));
+}
+
 /*
  * Initialize structures that control access to QSFP. Called once per port
  * on cards that support QSFP.
@@ -452,7 +459,6 @@ void qib_qsfp_init(struct qib_qsfp_data *qd,
 		   void (*fevent)(struct work_struct *))
 {
 	u32 mask, highs;
-	int pins;
 
 	struct qib_devdata *dd = qd->ppd->dd;
 
@@ -480,8 +486,7 @@ void qib_qsfp_init(struct qib_qsfp_data *qd,
 		mask <<= QSFP_GPIO_PORT2_SHIFT;
 
 	/* Do not try to wait here. Better to let event handle it */
-	pins = dd->f_gpio_mod(dd, 0, 0, 0);
-	if (pins & mask)
+	if (!qib_qsfp_mod_present(qd->ppd))
 		goto bail;
 	/* We see a module, but it may be unwise to look yet. Just schedule */
 	qd->t_insert = get_jiffies_64();

drivers/infiniband/hw/qib/qib_qsfp.h

@@ -34,6 +34,7 @@
 
 #define QSFP_DEV 0xA0
 #define QSFP_PWR_LAG_MSEC 2000
+#define QSFP_MODPRS_LAG_MSEC 20
 
 /*
  * Below are masks for various QSFP signals, for Port 1.
@@ -177,10 +178,12 @@ struct qib_qsfp_data {
 	struct work_struct work;
 	struct qib_qsfp_cache cache;
 	u64 t_insert;
+	u8 modpresent;
 };
 
 extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd,
 				  struct qib_qsfp_cache *cp);
+extern int qib_qsfp_mod_present(struct qib_pportdata *ppd);
 extern void qib_qsfp_init(struct qib_qsfp_data *qd,
 			  void (*fevent)(struct work_struct *));
 extern void qib_qsfp_deinit(struct qib_qsfp_data *qd);

drivers/infiniband/hw/qib/qib_rc.c

@@ -59,8 +59,7 @@ static void start_timer(struct qib_qp *qp)
 	qp->s_flags |= QIB_S_TIMER;
 	qp->s_timer.function = rc_timeout;
 	/* 4.096 usec. * (1 << qp->timeout) */
-	qp->s_timer.expires = jiffies +
-		usecs_to_jiffies((4096UL * (1UL << qp->timeout)) / 1000UL);
+	qp->s_timer.expires = jiffies + qp->timeout_jiffies;
 	add_timer(&qp->s_timer);
 }
 
@@ -239,7 +238,7 @@ int qib_make_rc_req(struct qib_qp *qp)
 	u32 len;
 	u32 bth0;
 	u32 bth2;
-	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	u32 pmtu = qp->pmtu;
 	char newreq;
 	unsigned long flags;
 	int ret = 0;
@@ -1519,9 +1518,7 @@ read_middle:
 		 * 4.096 usec. * (1 << qp->timeout)
 		 */
 		qp->s_flags |= QIB_S_TIMER;
-		mod_timer(&qp->s_timer, jiffies +
-			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
-					 1000UL));
+		mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies);
 		if (qp->s_flags & QIB_S_WAIT_ACK) {
 			qp->s_flags &= ~QIB_S_WAIT_ACK;
 			qib_schedule_send(qp);
@@ -1732,7 +1729,7 @@ static int qib_rc_rcv_error(struct qib_other_headers *ohdr,
 		 * same request.
 		 */
 		offset = ((psn - e->psn) & QIB_PSN_MASK) *
-			ib_mtu_enum_to_int(qp->path_mtu);
+			qp->pmtu;
 		len = be32_to_cpu(reth->length);
 		if (unlikely(offset + len != e->rdma_sge.sge_length))
 			goto unlock_done;
@@ -1876,7 +1873,7 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
 	u32 psn;
 	u32 pad;
 	struct ib_wc wc;
-	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	u32 pmtu = qp->pmtu;
 	int diff;
 	struct ib_reth *reth;
 	unsigned long flags;
@@ -1892,10 +1889,8 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
 	}
 
 	opcode = be32_to_cpu(ohdr->bth[0]);
-	spin_lock_irqsave(&qp->s_lock, flags);
 	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
-		goto sunlock;
-	spin_unlock_irqrestore(&qp->s_lock, flags);
+		return;
 
 	psn = be32_to_cpu(ohdr->bth[2]);
 	opcode >>= 24;
@@ -1955,8 +1950,6 @@ void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
 		break;
 	}
 
-	memset(&wc, 0, sizeof wc);
-
 	if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
 		qp->r_flags |= QIB_R_COMM_EST;
 		if (qp->ibqp.event_handler) {
@@ -2009,16 +2002,19 @@ send_middle:
 			goto rnr_nak;
 		qp->r_rcv_len = 0;
 		if (opcode == OP(SEND_ONLY))
-			goto send_last;
-		/* FALLTHROUGH */
+			goto no_immediate_data;
+		/* FALLTHROUGH for SEND_ONLY_WITH_IMMEDIATE */
 	case OP(SEND_LAST_WITH_IMMEDIATE):
 send_last_imm:
 		wc.ex.imm_data = ohdr->u.imm_data;
 		hdrsize += 4;
 		wc.wc_flags = IB_WC_WITH_IMM;
-		/* FALLTHROUGH */
+		goto send_last;
 	case OP(SEND_LAST):
 	case OP(RDMA_WRITE_LAST):
+no_immediate_data:
+		wc.wc_flags = 0;
+		wc.ex.imm_data = 0;
 send_last:
 		/* Get the number of bytes the message was padded by. */
 		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
@@ -2051,6 +2047,12 @@ send_last:
 		wc.src_qp = qp->remote_qpn;
 		wc.slid = qp->remote_ah_attr.dlid;
 		wc.sl = qp->remote_ah_attr.sl;
+		/* zero fields that are N/A */
+		wc.vendor_err = 0;
+		wc.pkey_index = 0;
+		wc.dlid_path_bits = 0;
+		wc.port_num = 0;
+		wc.csum_ok = 0;
 		/* Signal completion event if the solicited bit is set. */
 		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 			     (ohdr->bth[0] &
@@ -2089,7 +2091,7 @@ send_last:
 		if (opcode == OP(RDMA_WRITE_FIRST))
 			goto send_middle;
 		else if (opcode == OP(RDMA_WRITE_ONLY))
-			goto send_last;
+			goto no_immediate_data;
 		ret = qib_get_rwqe(qp, 1);
 		if (ret < 0)
 			goto nack_op_err;

drivers/infiniband/hw/qib/qib_ruc.c

@@ -260,12 +260,15 @@ static int gid_ok(union ib_gid *gid, __be64 gid_prefix, __be64 id)
 
 /*
  *
- * This should be called with the QP s_lock held.
+ * This should be called with the QP r_lock held.
+ *
+ * The s_lock will be acquired around the qib_migrate_qp() call.
  */
 int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 		      int has_grh, struct qib_qp *qp, u32 bth0)
 {
 	__be64 guid;
+	unsigned long flags;
 
 	if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) {
 		if (!has_grh) {
@@ -295,7 +298,9 @@ int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 		if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid ||
 		    ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num)
 			goto err;
+		spin_lock_irqsave(&qp->s_lock, flags);
 		qib_migrate_qp(qp);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
 	} else {
 		if (!has_grh) {
 			if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)

drivers/infiniband/hw/qib/qib_srq.c

@@ -107,6 +107,11 @@ struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
 	u32 sz;
 	struct ib_srq *ret;
 
+	if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
+		ret = ERR_PTR(-ENOSYS);
+		goto done;
+	}
+
 	if (srq_init_attr->attr.max_sge == 0 ||
 	    srq_init_attr->attr.max_sge > ib_qib_max_srq_sges ||
 	    srq_init_attr->attr.max_wr == 0 ||

drivers/infiniband/hw/qib/qib_sysfs.c

@@ -515,8 +515,7 @@ static ssize_t show_nfreectxts(struct device *device,
 	struct qib_devdata *dd = dd_from_dev(dev);
 
 	/* Return the number of free user ports (contexts) available. */
-	return scnprintf(buf, PAGE_SIZE, "%u\n", dd->cfgctxts -
-		dd->first_user_ctxt - (u32)qib_stats.sps_ctxts);
+	return scnprintf(buf, PAGE_SIZE, "%u\n", dd->freectxts);
 }
 
 static ssize_t show_serial(struct device *device,

drivers/infiniband/hw/qib/qib_uc.c

@@ -51,7 +51,7 @@ int qib_make_uc_req(struct qib_qp *qp)
 	u32 hwords;
 	u32 bth0;
 	u32 len;
-	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	u32 pmtu = qp->pmtu;
 	int ret = 0;
 
 	spin_lock_irqsave(&qp->s_lock, flags);
@@ -243,13 +243,12 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
 {
 	struct qib_other_headers *ohdr;
-	unsigned long flags;
 	u32 opcode;
 	u32 hdrsize;
 	u32 psn;
 	u32 pad;
 	struct ib_wc wc;
-	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	u32 pmtu = qp->pmtu;
 	struct ib_reth *reth;
 	int ret;
 
@@ -263,14 +262,11 @@ void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
 	}
 
 	opcode = be32_to_cpu(ohdr->bth[0]);
-	spin_lock_irqsave(&qp->s_lock, flags);
 	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
-		goto sunlock;
-	spin_unlock_irqrestore(&qp->s_lock, flags);
+		return;
 
 	psn = be32_to_cpu(ohdr->bth[2]);
 	opcode >>= 24;
-	memset(&wc, 0, sizeof wc);
 
 	/* Compare the PSN verses the expected PSN. */
 	if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
@@ -370,7 +366,7 @@ send_first:
 		}
 		qp->r_rcv_len = 0;
 		if (opcode == OP(SEND_ONLY))
-			goto send_last;
+			goto no_immediate_data;
 		else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
 			goto send_last_imm;
 		/* FALLTHROUGH */
@@ -389,8 +385,11 @@ send_last_imm:
 		wc.ex.imm_data = ohdr->u.imm_data;
 		hdrsize += 4;
 		wc.wc_flags = IB_WC_WITH_IMM;
-		/* FALLTHROUGH */
+		goto send_last;
 	case OP(SEND_LAST):
+no_immediate_data:
+		wc.ex.imm_data = 0;
+		wc.wc_flags = 0;
 send_last:
 		/* Get the number of bytes the message was padded by. */
 		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
@@ -418,6 +417,12 @@ last_imm:
 		wc.src_qp = qp->remote_qpn;
 		wc.slid = qp->remote_ah_attr.dlid;
 		wc.sl = qp->remote_ah_attr.sl;
+		/* zero fields that are N/A */
+		wc.vendor_err = 0;
+		wc.pkey_index = 0;
+		wc.dlid_path_bits = 0;
+		wc.port_num = 0;
+		wc.csum_ok = 0;
 		/* Signal completion event if the solicited bit is set. */
 		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 			     (ohdr->bth[0] &
@@ -546,6 +551,4 @@ op_err:
 	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
 	return;
 
-sunlock:
-	spin_unlock_irqrestore(&qp->s_lock, flags);
 }

drivers/infiniband/hw/qib/qib_verbs.c

@@ -38,11 +38,12 @@
 #include <linux/utsname.h>
 #include <linux/rculist.h>
 #include <linux/mm.h>
+#include <linux/random.h>
 
 #include "qib.h"
 #include "qib_common.h"
 
-static unsigned int ib_qib_qp_table_size = 251;
+static unsigned int ib_qib_qp_table_size = 256;
 module_param_named(qp_table_size, ib_qib_qp_table_size, uint, S_IRUGO);
 MODULE_PARM_DESC(qp_table_size, "QP table size");
 
@@ -659,17 +660,25 @@ void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
 		if (atomic_dec_return(&mcast->refcount) <= 1)
 			wake_up(&mcast->wait);
 	} else {
-		qp = qib_lookup_qpn(ibp, qp_num);
-		if (!qp)
-			goto drop;
+		if (rcd->lookaside_qp) {
+			if (rcd->lookaside_qpn != qp_num) {
+				if (atomic_dec_and_test(
+					&rcd->lookaside_qp->refcount))
+					wake_up(
+					 &rcd->lookaside_qp->wait);
+					rcd->lookaside_qp = NULL;
+				}
+		}
+		if (!rcd->lookaside_qp) {
+			qp = qib_lookup_qpn(ibp, qp_num);
+			if (!qp)
+				goto drop;
+			rcd->lookaside_qp = qp;
+			rcd->lookaside_qpn = qp_num;
+		} else
+			qp = rcd->lookaside_qp;
 		ibp->n_unicast_rcv++;
 		qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp);
-		/*
-		 * Notify qib_destroy_qp() if it is waiting
-		 * for us to finish.
-		 */
-		if (atomic_dec_and_test(&qp->refcount))
-			wake_up(&qp->wait);
 	}
 	return;
 
@@ -1974,6 +1983,8 @@ static void init_ibport(struct qib_pportdata *ppd)
 	ibp->z_excessive_buffer_overrun_errors =
 		cntrs.excessive_buffer_overrun_errors;
 	ibp->z_vl15_dropped = cntrs.vl15_dropped;
+	RCU_INIT_POINTER(ibp->qp0, NULL);
+	RCU_INIT_POINTER(ibp->qp1, NULL);
 }
 
 /**
@@ -1990,12 +2001,15 @@ int qib_register_ib_device(struct qib_devdata *dd)
 	int ret;
 
 	dev->qp_table_size = ib_qib_qp_table_size;
-	dev->qp_table = kzalloc(dev->qp_table_size * sizeof *dev->qp_table,
+	get_random_bytes(&dev->qp_rnd, sizeof(dev->qp_rnd));
+	dev->qp_table = kmalloc(dev->qp_table_size * sizeof *dev->qp_table,
 				GFP_KERNEL);
 	if (!dev->qp_table) {
 		ret = -ENOMEM;
 		goto err_qpt;
 	}
+	for (i = 0; i < dev->qp_table_size; i++)
+		RCU_INIT_POINTER(dev->qp_table[i], NULL);
 
 	for (i = 0; i < dd->num_pports; i++)
 		init_ibport(ppd + i);

drivers/infiniband/hw/qib/qib_verbs.h

@@ -485,6 +485,7 @@ struct qib_qp {
 	u8 alt_timeout;         /* Alternate path timeout for this QP */
 	u8 port_num;
 	enum ib_mtu path_mtu;
+	u32 pmtu;		/* decoded from path_mtu */
 	u32 remote_qpn;
 	u32 qkey;               /* QKEY for this QP (for UD or RD) */
 	u32 s_size;             /* send work queue size */
@@ -495,6 +496,7 @@ struct qib_qp {
 	u32 s_last;             /* last completed entry */
 	u32 s_ssn;              /* SSN of tail entry */
 	u32 s_lsn;              /* limit sequence number (credit) */
+	unsigned long timeout_jiffies;  /* computed from timeout */
 	struct qib_swqe *s_wq;  /* send work queue */
 	struct qib_swqe *s_wqe;
 	struct qib_rq r_rq;             /* receive work queue */
@@ -723,7 +725,8 @@ struct qib_ibdev {
 	dma_addr_t pio_hdrs_phys;
 	/* list of QPs waiting for RNR timer */
 	spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */
-	unsigned qp_table_size; /* size of the hash table */
+	u32 qp_table_size; /* size of the hash table */
+	u32 qp_rnd; /* random bytes for hash */
 	spinlock_t qpt_lock;
 
 	u32 n_piowait;

drivers/infiniband/ulp/ipoib/ipoib_cm.c

@@ -84,7 +84,7 @@ static void ipoib_cm_dma_unmap_rx(struct ipoib_dev_priv *priv, int frags,
 	ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE, DMA_FROM_DEVICE);
 
 	for (i = 0; i < frags; ++i)
-		ib_dma_unmap_single(priv->ca, mapping[i + 1], PAGE_SIZE, DMA_FROM_DEVICE);
+		ib_dma_unmap_page(priv->ca, mapping[i + 1], PAGE_SIZE, DMA_FROM_DEVICE);
 }
 
 static int ipoib_cm_post_receive_srq(struct net_device *dev, int id)
@@ -183,7 +183,7 @@ partial_error:
 	ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_CM_HEAD_SIZE, DMA_FROM_DEVICE);
 
 	for (; i > 0; --i)
-		ib_dma_unmap_single(priv->ca, mapping[i], PAGE_SIZE, DMA_FROM_DEVICE);
+		ib_dma_unmap_page(priv->ca, mapping[i], PAGE_SIZE, DMA_FROM_DEVICE);
 
 	dev_kfree_skb_any(skb);
 	return NULL;
@@ -1497,6 +1497,7 @@ static void ipoib_cm_create_srq(struct net_device *dev, int max_sge)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ib_srq_init_attr srq_init_attr = {
+		.srq_type = IB_SRQT_BASIC,
 		.attr = {
 			.max_wr  = ipoib_recvq_size,
 			.max_sge = max_sge

drivers/infiniband/ulp/ipoib/ipoib_fs.c

@@ -212,16 +212,15 @@ static int ipoib_path_seq_show(struct seq_file *file, void *iter_ptr)
 		   gid_buf, path.pathrec.dlid ? "yes" : "no");
 
 	if (path.pathrec.dlid) {
-		rate = ib_rate_to_mult(path.pathrec.rate) * 25;
+		rate = ib_rate_to_mbps(path.pathrec.rate);
 
 		seq_printf(file,
 			   "  DLID:     0x%04x\n"
 			   "  SL: %12d\n"
-			   "  rate: %*d%s Gb/sec\n",
+			   "  rate: %8d.%d Gb/sec\n",
 			   be16_to_cpu(path.pathrec.dlid),
 			   path.pathrec.sl,
-			   10 - ((rate % 10) ? 2 : 0),
-			   rate / 10, rate % 10 ? ".5" : "");
+			   rate / 1000, rate % 1000);
 	}
 
 	seq_putc(file, '\n');

drivers/net/ethernet/mellanox/mlx4/eq.c

@@ -484,7 +484,7 @@ static void mlx4_free_eq(struct mlx4_dev *dev,
 
 	mlx4_mtt_cleanup(dev, &eq->mtt);
 	for (i = 0; i < npages; ++i)
-		pci_free_consistent(dev->pdev, PAGE_SIZE,
+		dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
 				    eq->page_list[i].buf,
 				    eq->page_list[i].map);
 

drivers/net/ethernet/mellanox/mlx4/fw.c

@@ -205,6 +205,8 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 #define QUERY_DEV_CAP_MAX_MCG_OFFSET		0x63
 #define QUERY_DEV_CAP_RSVD_PD_OFFSET		0x64
 #define QUERY_DEV_CAP_MAX_PD_OFFSET		0x65
+#define QUERY_DEV_CAP_RSVD_XRC_OFFSET		0x66
+#define QUERY_DEV_CAP_MAX_XRC_OFFSET		0x67
 #define QUERY_DEV_CAP_MAX_COUNTERS_OFFSET	0x68
 #define QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET	0x80
 #define QUERY_DEV_CAP_QPC_ENTRY_SZ_OFFSET	0x82
@@ -319,6 +321,10 @@ int mlx4_QUERY_DEV_CAP(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	dev_cap->reserved_pds = field >> 4;
 	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET);
 	dev_cap->max_pds = 1 << (field & 0x3f);
+	MLX4_GET(field, outbox, QUERY_DEV_CAP_RSVD_XRC_OFFSET);
+	dev_cap->reserved_xrcds = field >> 4;
+	MLX4_GET(field, outbox, QUERY_DEV_CAP_MAX_PD_OFFSET);
+	dev_cap->max_xrcds = 1 << (field & 0x1f);
 
 	MLX4_GET(size, outbox, QUERY_DEV_CAP_RDMARC_ENTRY_SZ_OFFSET);
 	dev_cap->rdmarc_entry_sz = size;

drivers/net/ethernet/mellanox/mlx4/fw.h

@@ -93,6 +93,8 @@ struct mlx4_dev_cap {
 	int max_mcgs;
 	int reserved_pds;
 	int max_pds;
+	int reserved_xrcds;
+	int max_xrcds;
 	int qpc_entry_sz;
 	int rdmarc_entry_sz;
 	int altc_entry_sz;

drivers/net/ethernet/mellanox/mlx4/main.c

@@ -96,6 +96,8 @@ MODULE_PARM_DESC(log_num_mac, "Log2 max number of MACs per ETH port (1-7)");
 static int log_num_vlan;
 module_param_named(log_num_vlan, log_num_vlan, int, 0444);
 MODULE_PARM_DESC(log_num_vlan, "Log2 max number of VLANs per ETH port (0-7)");
+/* Log2 max number of VLANs per ETH port (0-7) */
+#define MLX4_LOG_NUM_VLANS 7
 
 static int use_prio;
 module_param_named(use_prio, use_prio, bool, 0444);
@@ -220,6 +222,10 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	dev->caps.reserved_mrws	     = dev_cap->reserved_mrws;
 	dev->caps.reserved_uars	     = dev_cap->reserved_uars;
 	dev->caps.reserved_pds	     = dev_cap->reserved_pds;
+	dev->caps.reserved_xrcds     = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
+					dev_cap->reserved_xrcds : 0;
+	dev->caps.max_xrcds          = (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) ?
+					dev_cap->max_xrcds : 0;
 	dev->caps.mtt_entry_sz	     = dev->caps.mtts_per_seg * dev_cap->mtt_entry_sz;
 	dev->caps.max_msg_sz         = dev_cap->max_msg_sz;
 	dev->caps.page_size_cap	     = ~(u32) (dev_cap->min_page_sz - 1);
@@ -230,7 +236,7 @@ static int mlx4_dev_cap(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap)
 	dev->caps.max_gso_sz	     = dev_cap->max_gso_sz;
 
 	dev->caps.log_num_macs  = log_num_mac;
-	dev->caps.log_num_vlans = log_num_vlan;
+	dev->caps.log_num_vlans = MLX4_LOG_NUM_VLANS;
 	dev->caps.log_num_prios = use_prio ? 3 : 0;
 
 	for (i = 1; i <= dev->caps.num_ports; ++i) {
@@ -912,11 +918,18 @@ static int mlx4_setup_hca(struct mlx4_dev *dev)
 		goto err_kar_unmap;
 	}
 
+	err = mlx4_init_xrcd_table(dev);
+	if (err) {
+		mlx4_err(dev, "Failed to initialize "
+			 "reliable connection domain table, aborting.\n");
+		goto err_pd_table_free;
+	}
+
 	err = mlx4_init_mr_table(dev);
 	if (err) {
 		mlx4_err(dev, "Failed to initialize "
 			 "memory region table, aborting.\n");
-		goto err_pd_table_free;
+		goto err_xrcd_table_free;
 	}
 
 	err = mlx4_init_eq_table(dev);
@@ -998,6 +1011,13 @@ static int mlx4_setup_hca(struct mlx4_dev *dev)
 				  "ib capabilities (%d). Continuing with "
 				  "caps = 0\n", port, err);
 		dev->caps.ib_port_def_cap[port] = ib_port_default_caps;
+
+		err = mlx4_check_ext_port_caps(dev, port);
+		if (err)
+			mlx4_warn(dev, "failed to get port %d extended "
+				  "port capabilities support info (%d)."
+				  " Assuming not supported\n", port, err);
+
 		err = mlx4_SET_PORT(dev, port);
 		if (err) {
 			mlx4_err(dev, "Failed to set port %d, aborting\n",
@@ -1033,6 +1053,9 @@ err_eq_table_free:
 err_mr_table_free:
 	mlx4_cleanup_mr_table(dev);
 
+err_xrcd_table_free:
+	mlx4_cleanup_xrcd_table(dev);
+
 err_pd_table_free:
 	mlx4_cleanup_pd_table(dev);
 
@@ -1355,6 +1378,7 @@ err_port:
 	mlx4_cmd_use_polling(dev);
 	mlx4_cleanup_eq_table(dev);
 	mlx4_cleanup_mr_table(dev);
+	mlx4_cleanup_xrcd_table(dev);
 	mlx4_cleanup_pd_table(dev);
 	mlx4_cleanup_uar_table(dev);
 
@@ -1416,6 +1440,7 @@ static void mlx4_remove_one(struct pci_dev *pdev)
 		mlx4_cmd_use_polling(dev);
 		mlx4_cleanup_eq_table(dev);
 		mlx4_cleanup_mr_table(dev);
+		mlx4_cleanup_xrcd_table(dev);
 		mlx4_cleanup_pd_table(dev);
 
 		iounmap(priv->kar);
@@ -1489,10 +1514,9 @@ static int __init mlx4_verify_params(void)
 		return -1;
 	}
 
-	if ((log_num_vlan < 0) || (log_num_vlan > 7)) {
-		pr_warning("mlx4_core: bad num_vlan: %d\n", log_num_vlan);
-		return -1;
-	}
+	if (log_num_vlan != 0)
+		pr_warning("mlx4_core: log_num_vlan - obsolete module param, using %d\n",
+			   MLX4_LOG_NUM_VLANS);
 
 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 7)) {
 		pr_warning("mlx4_core: bad log_mtts_per_seg: %d\n", log_mtts_per_seg);

drivers/net/ethernet/mellanox/mlx4/mlx4.h

@@ -335,6 +335,7 @@ struct mlx4_priv {
 	struct mlx4_cmd		cmd;
 
 	struct mlx4_bitmap	pd_bitmap;
+	struct mlx4_bitmap	xrcd_bitmap;
 	struct mlx4_uar_table	uar_table;
 	struct mlx4_mr_table	mr_table;
 	struct mlx4_cq_table	cq_table;
@@ -384,6 +385,7 @@ int mlx4_alloc_eq_table(struct mlx4_dev *dev);
 void mlx4_free_eq_table(struct mlx4_dev *dev);
 
 int mlx4_init_pd_table(struct mlx4_dev *dev);
+int mlx4_init_xrcd_table(struct mlx4_dev *dev);
 int mlx4_init_uar_table(struct mlx4_dev *dev);
 int mlx4_init_mr_table(struct mlx4_dev *dev);
 int mlx4_init_eq_table(struct mlx4_dev *dev);
@@ -393,6 +395,7 @@ int mlx4_init_srq_table(struct mlx4_dev *dev);
 int mlx4_init_mcg_table(struct mlx4_dev *dev);
 
 void mlx4_cleanup_pd_table(struct mlx4_dev *dev);
+void mlx4_cleanup_xrcd_table(struct mlx4_dev *dev);
 void mlx4_cleanup_uar_table(struct mlx4_dev *dev);
 void mlx4_cleanup_mr_table(struct mlx4_dev *dev);
 void mlx4_cleanup_eq_table(struct mlx4_dev *dev);
@@ -450,6 +453,7 @@ void mlx4_init_vlan_table(struct mlx4_dev *dev, struct mlx4_vlan_table *table);
 
 int mlx4_SET_PORT(struct mlx4_dev *dev, u8 port);
 int mlx4_get_port_ib_caps(struct mlx4_dev *dev, u8 port, __be32 *caps);
+int mlx4_check_ext_port_caps(struct mlx4_dev *dev, u8 port);
 
 int mlx4_qp_detach_common(struct mlx4_dev *dev, struct mlx4_qp *qp, u8 gid[16],
 			  enum mlx4_protocol prot, enum mlx4_steer_type steer);

drivers/net/ethernet/mellanox/mlx4/mr.c

@@ -139,7 +139,7 @@ static int mlx4_buddy_init(struct mlx4_buddy *buddy, int max_order)
 
 	buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *),
 			      GFP_KERNEL);
-	buddy->num_free = kzalloc((buddy->max_order + 1) * sizeof (int *),
+	buddy->num_free = kcalloc((buddy->max_order + 1), sizeof *buddy->num_free,
 				  GFP_KERNEL);
 	if (!buddy->bits || !buddy->num_free)
 		goto err_out;

drivers/net/ethernet/mellanox/mlx4/pd.c

@@ -61,6 +61,24 @@ void mlx4_pd_free(struct mlx4_dev *dev, u32 pdn)
 }
 EXPORT_SYMBOL_GPL(mlx4_pd_free);
 
+int mlx4_xrcd_alloc(struct mlx4_dev *dev, u32 *xrcdn)
+{
+	struct mlx4_priv *priv = mlx4_priv(dev);
+
+	*xrcdn = mlx4_bitmap_alloc(&priv->xrcd_bitmap);
+	if (*xrcdn == -1)
+		return -ENOMEM;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(mlx4_xrcd_alloc);
+
+void mlx4_xrcd_free(struct mlx4_dev *dev, u32 xrcdn)
+{
+	mlx4_bitmap_free(&mlx4_priv(dev)->xrcd_bitmap, xrcdn);
+}
+EXPORT_SYMBOL_GPL(mlx4_xrcd_free);
+
 int mlx4_init_pd_table(struct mlx4_dev *dev)
 {
 	struct mlx4_priv *priv = mlx4_priv(dev);
@@ -74,6 +92,18 @@ void mlx4_cleanup_pd_table(struct mlx4_dev *dev)
 	mlx4_bitmap_cleanup(&mlx4_priv(dev)->pd_bitmap);
 }
 
+int mlx4_init_xrcd_table(struct mlx4_dev *dev)
+{
+	struct mlx4_priv *priv = mlx4_priv(dev);
+
+	return mlx4_bitmap_init(&priv->xrcd_bitmap, (1 << 16),
+				(1 << 16) - 1, dev->caps.reserved_xrcds + 1, 0);
+}
+
+void mlx4_cleanup_xrcd_table(struct mlx4_dev *dev)
+{
+	mlx4_bitmap_cleanup(&mlx4_priv(dev)->xrcd_bitmap);
+}
 
 int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar)
 {

drivers/net/ethernet/mellanox/mlx4/port.c

@@ -148,22 +148,26 @@ int mlx4_register_mac(struct mlx4_dev *dev, u8 port, u64 mac, int *qpn, u8 wrap)
 
 	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_VEP_UC_STEER) {
 		err = mlx4_uc_steer_add(dev, port, mac, qpn, 1);
-		if (!err) {
-			entry = kmalloc(sizeof *entry, GFP_KERNEL);
-			if (!entry) {
-				mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
-				return -ENOMEM;
-			}
-			entry->mac = mac;
-			err = radix_tree_insert(&info->mac_tree, *qpn, entry);
-			if (err) {
-				mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
-				return err;
-			}
-		} else
+		if (err)
 			return err;
+
+		entry = kmalloc(sizeof *entry, GFP_KERNEL);
+		if (!entry) {
+			mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
+			return -ENOMEM;
+		}
+
+		entry->mac = mac;
+		err = radix_tree_insert(&info->mac_tree, *qpn, entry);
+		if (err) {
+			kfree(entry);
+			mlx4_uc_steer_release(dev, port, mac, *qpn, 1);
+			return err;
+		}
 	}
+
 	mlx4_dbg(dev, "Registering MAC: 0x%llx\n", (unsigned long long) mac);
+
 	mutex_lock(&table->mutex);
 	for (i = 0; i < MLX4_MAX_MAC_NUM - 1; i++) {
 		if (free < 0 && !table->refs[i]) {
@@ -465,6 +469,48 @@ int mlx4_get_port_ib_caps(struct mlx4_dev *dev, u8 port, __be32 *caps)
 	return err;
 }
 
+int mlx4_check_ext_port_caps(struct mlx4_dev *dev, u8 port)
+{
+	struct mlx4_cmd_mailbox *inmailbox, *outmailbox;
+	u8 *inbuf, *outbuf;
+	int err, packet_error;
+
+	inmailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(inmailbox))
+		return PTR_ERR(inmailbox);
+
+	outmailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(outmailbox)) {
+		mlx4_free_cmd_mailbox(dev, inmailbox);
+		return PTR_ERR(outmailbox);
+	}
+
+	inbuf = inmailbox->buf;
+	outbuf = outmailbox->buf;
+	memset(inbuf, 0, 256);
+	memset(outbuf, 0, 256);
+	inbuf[0] = 1;
+	inbuf[1] = 1;
+	inbuf[2] = 1;
+	inbuf[3] = 1;
+
+	*(__be16 *) (&inbuf[16]) = MLX4_ATTR_EXTENDED_PORT_INFO;
+	*(__be32 *) (&inbuf[20]) = cpu_to_be32(port);
+
+	err = mlx4_cmd_box(dev, inmailbox->dma, outmailbox->dma, port, 3,
+			   MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C);
+
+	packet_error = be16_to_cpu(*(__be16 *) (outbuf + 4));
+
+	dev->caps.ext_port_cap[port] = (!err && !packet_error) ?
+				       MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO
+				       : 0;
+
+	mlx4_free_cmd_mailbox(dev, inmailbox);
+	mlx4_free_cmd_mailbox(dev, outmailbox);
+	return err;
+}
+
 int mlx4_SET_PORT(struct mlx4_dev *dev, u8 port)
 {
 	struct mlx4_cmd_mailbox *mailbox;

drivers/net/ethernet/mellanox/mlx4/qp.c

@@ -280,6 +280,9 @@ int mlx4_init_qp_table(struct mlx4_dev *dev)
 	 * We reserve 2 extra QPs per port for the special QPs.  The
 	 * block of special QPs must be aligned to a multiple of 8, so
 	 * round up.
+	 *
+	 * We also reserve the MSB of the 24-bit QP number to indicate
+	 * that a QP is an XRC QP.
 	 */
 	dev->caps.sqp_start =
 		ALIGN(dev->caps.reserved_qps_cnt[MLX4_QP_REGION_FW], 8);

drivers/net/ethernet/mellanox/mlx4/srq.c

@@ -40,20 +40,20 @@
 struct mlx4_srq_context {
 	__be32			state_logsize_srqn;
 	u8			logstride;
-	u8			reserved1[3];
-	u8			pg_offset;
-	u8			reserved2[3];
-	u32			reserved3;
+	u8			reserved1;
+	__be16			xrcd;
+	__be32			pg_offset_cqn;
+	u32			reserved2;
 	u8			log_page_size;
-	u8			reserved4[2];
+	u8			reserved3[2];
 	u8			mtt_base_addr_h;
 	__be32			mtt_base_addr_l;
 	__be32			pd;
 	__be16			limit_watermark;
 	__be16			wqe_cnt;
-	u16			reserved5;
+	u16			reserved4;
 	__be16			wqe_counter;
-	u32			reserved6;
+	u32			reserved5;
 	__be64			db_rec_addr;
 };
 
@@ -109,8 +109,8 @@ static int mlx4_QUERY_SRQ(struct mlx4_dev *dev, struct mlx4_cmd_mailbox *mailbox
 			    MLX4_CMD_TIME_CLASS_A);
 }
 
-int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
-		   u64 db_rec, struct mlx4_srq *srq)
+int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, u32 cqn, u16 xrcd,
+		   struct mlx4_mtt *mtt, u64 db_rec, struct mlx4_srq *srq)
 {
 	struct mlx4_srq_table *srq_table = &mlx4_priv(dev)->srq_table;
 	struct mlx4_cmd_mailbox *mailbox;
@@ -148,6 +148,8 @@ int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
 	srq_context->state_logsize_srqn = cpu_to_be32((ilog2(srq->max) << 24) |
 						      srq->srqn);
 	srq_context->logstride          = srq->wqe_shift - 4;
+	srq_context->xrcd		= cpu_to_be16(xrcd);
+	srq_context->pg_offset_cqn	= cpu_to_be32(cqn & 0xffffff);
 	srq_context->log_page_size      = mtt->page_shift - MLX4_ICM_PAGE_SHIFT;
 
 	mtt_addr = mlx4_mtt_addr(dev, mtt);

include/linux/mlx4/device.h

@@ -61,6 +61,7 @@ enum {
 	MLX4_DEV_CAP_FLAG_RC		= 1LL <<  0,
 	MLX4_DEV_CAP_FLAG_UC		= 1LL <<  1,
 	MLX4_DEV_CAP_FLAG_UD		= 1LL <<  2,
+	MLX4_DEV_CAP_FLAG_XRC		= 1LL <<  3,
 	MLX4_DEV_CAP_FLAG_SRQ		= 1LL <<  6,
 	MLX4_DEV_CAP_FLAG_IPOIB_CSUM	= 1LL <<  7,
 	MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR	= 1LL <<  8,
@@ -83,6 +84,12 @@ enum {
 	MLX4_DEV_CAP_FLAG_COUNTERS	= 1LL << 48
 };
 
+#define MLX4_ATTR_EXTENDED_PORT_INFO	cpu_to_be16(0xff90)
+
+enum {
+	MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO	= 1 <<  0
+};
+
 enum {
 	MLX4_BMME_FLAG_LOCAL_INV	= 1 <<  6,
 	MLX4_BMME_FLAG_REMOTE_INV	= 1 <<  7,
@@ -257,6 +264,8 @@ struct mlx4_caps {
 	int			num_qp_per_mgm;
 	int			num_pds;
 	int			reserved_pds;
+	int			max_xrcds;
+	int			reserved_xrcds;
 	int			mtt_entry_sz;
 	u32			max_msg_sz;
 	u32			page_size_cap;
@@ -277,6 +286,7 @@ struct mlx4_caps {
 	u32			port_mask;
 	enum mlx4_port_type	possible_type[MLX4_MAX_PORTS + 1];
 	u32			max_counters;
+	u8			ext_port_cap[MLX4_MAX_PORTS + 1];
 };
 
 struct mlx4_buf_list {
@@ -500,6 +510,8 @@ static inline void *mlx4_buf_offset(struct mlx4_buf *buf, int offset)
 
 int mlx4_pd_alloc(struct mlx4_dev *dev, u32 *pdn);
 void mlx4_pd_free(struct mlx4_dev *dev, u32 pdn);
+int mlx4_xrcd_alloc(struct mlx4_dev *dev, u32 *xrcdn);
+void mlx4_xrcd_free(struct mlx4_dev *dev, u32 xrcdn);
 
 int mlx4_uar_alloc(struct mlx4_dev *dev, struct mlx4_uar *uar);
 void mlx4_uar_free(struct mlx4_dev *dev, struct mlx4_uar *uar);
@@ -539,8 +551,8 @@ void mlx4_qp_release_range(struct mlx4_dev *dev, int base_qpn, int cnt);
 int mlx4_qp_alloc(struct mlx4_dev *dev, int qpn, struct mlx4_qp *qp);
 void mlx4_qp_free(struct mlx4_dev *dev, struct mlx4_qp *qp);
 
-int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, struct mlx4_mtt *mtt,
-		   u64 db_rec, struct mlx4_srq *srq);
+int mlx4_srq_alloc(struct mlx4_dev *dev, u32 pdn, u32 cqn, u16 xrcdn,
+		   struct mlx4_mtt *mtt, u64 db_rec, struct mlx4_srq *srq);
 void mlx4_srq_free(struct mlx4_dev *dev, struct mlx4_srq *srq);
 int mlx4_srq_arm(struct mlx4_dev *dev, struct mlx4_srq *srq, int limit_watermark);
 int mlx4_srq_query(struct mlx4_dev *dev, struct mlx4_srq *srq, int *limit_watermark);

include/linux/mlx4/qp.h

@@ -75,6 +75,7 @@ enum {
 	MLX4_QP_ST_UC				= 0x1,
 	MLX4_QP_ST_RD				= 0x2,
 	MLX4_QP_ST_UD				= 0x3,
+	MLX4_QP_ST_XRC				= 0x6,
 	MLX4_QP_ST_MLX				= 0x7
 };
 
@@ -137,7 +138,7 @@ struct mlx4_qp_context {
 	__be32			ssn;
 	__be32			params2;
 	__be32			rnr_nextrecvpsn;
-	__be32			srcd;
+	__be32			xrcd;
 	__be32			cqn_recv;
 	__be64			db_rec_addr;
 	__be32			qkey;

include/rdma/ib_user_verbs.h

@@ -81,7 +81,11 @@ enum {
 	IB_USER_VERBS_CMD_MODIFY_SRQ,
 	IB_USER_VERBS_CMD_QUERY_SRQ,
 	IB_USER_VERBS_CMD_DESTROY_SRQ,
-	IB_USER_VERBS_CMD_POST_SRQ_RECV
+	IB_USER_VERBS_CMD_POST_SRQ_RECV,
+	IB_USER_VERBS_CMD_OPEN_XRCD,
+	IB_USER_VERBS_CMD_CLOSE_XRCD,
+	IB_USER_VERBS_CMD_CREATE_XSRQ,
+	IB_USER_VERBS_CMD_OPEN_QP
 };
 
 /*
@@ -222,6 +226,21 @@ struct ib_uverbs_dealloc_pd {
 	__u32 pd_handle;
 };
 
+struct ib_uverbs_open_xrcd {
+	__u64 response;
+	__u32 fd;
+	__u32 oflags;
+	__u64 driver_data[0];
+};
+
+struct ib_uverbs_open_xrcd_resp {
+	__u32 xrcd_handle;
+};
+
+struct ib_uverbs_close_xrcd {
+	__u32 xrcd_handle;
+};
+
 struct ib_uverbs_reg_mr {
 	__u64 response;
 	__u64 start;
@@ -404,6 +423,17 @@ struct ib_uverbs_create_qp {
 	__u64 driver_data[0];
 };
 
+struct ib_uverbs_open_qp {
+	__u64 response;
+	__u64 user_handle;
+	__u32 pd_handle;
+	__u32 qpn;
+	__u8  qp_type;
+	__u8  reserved[7];
+	__u64 driver_data[0];
+};
+
+/* also used for open response */
 struct ib_uverbs_create_qp_resp {
 	__u32 qp_handle;
 	__u32 qpn;
@@ -648,11 +678,25 @@ struct ib_uverbs_create_srq {
 	__u64 driver_data[0];
 };
 
+struct ib_uverbs_create_xsrq {
+	__u64 response;
+	__u64 user_handle;
+	__u32 srq_type;
+	__u32 pd_handle;
+	__u32 max_wr;
+	__u32 max_sge;
+	__u32 srq_limit;
+	__u32 reserved;
+	__u32 xrcd_handle;
+	__u32 cq_handle;
+	__u64 driver_data[0];
+};
+
 struct ib_uverbs_create_srq_resp {
 	__u32 srq_handle;
 	__u32 max_wr;
 	__u32 max_sge;
-	__u32 reserved;
+	__u32 srqn;
 };
 
 struct ib_uverbs_modify_srq {

include/rdma/ib_verbs.h

@@ -112,6 +112,7 @@ enum ib_device_cap_flags {
 	 */
 	IB_DEVICE_UD_IP_CSUM		= (1<<18),
 	IB_DEVICE_UD_TSO		= (1<<19),
+	IB_DEVICE_XRC			= (1<<20),
 	IB_DEVICE_MEM_MGT_EXTENSIONS	= (1<<21),
 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1<<22),
 };
@@ -207,6 +208,7 @@ enum ib_port_cap_flags {
 	IB_PORT_SM_DISABLED			= 1 << 10,
 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
+	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
 	IB_PORT_CM_SUP				= 1 << 16,
 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
 	IB_PORT_REINIT_SUP			= 1 << 18,
@@ -415,7 +417,15 @@ enum ib_rate {
 	IB_RATE_40_GBPS  = 7,
 	IB_RATE_60_GBPS  = 8,
 	IB_RATE_80_GBPS  = 9,
-	IB_RATE_120_GBPS = 10
+	IB_RATE_120_GBPS = 10,
+	IB_RATE_14_GBPS  = 11,
+	IB_RATE_56_GBPS  = 12,
+	IB_RATE_112_GBPS = 13,
+	IB_RATE_168_GBPS = 14,
+	IB_RATE_25_GBPS  = 15,
+	IB_RATE_100_GBPS = 16,
+	IB_RATE_200_GBPS = 17,
+	IB_RATE_300_GBPS = 18
 };
 
 /**
@@ -426,6 +436,13 @@ enum ib_rate {
  */
 int ib_rate_to_mult(enum ib_rate rate) __attribute_const__;
 
+/**
+ * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
+ * For example, IB_RATE_2_5_GBPS will be converted to 2500.
+ * @rate: rate to convert.
+ */
+int ib_rate_to_mbps(enum ib_rate rate) __attribute_const__;
+
 /**
  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
  * enum.
@@ -522,6 +539,11 @@ enum ib_cq_notify_flags {
 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
 };
 
+enum ib_srq_type {
+	IB_SRQT_BASIC,
+	IB_SRQT_XRC
+};
+
 enum ib_srq_attr_mask {
 	IB_SRQ_MAX_WR	= 1 << 0,
 	IB_SRQ_LIMIT	= 1 << 1,
@@ -537,6 +559,14 @@ struct ib_srq_init_attr {
 	void		      (*event_handler)(struct ib_event *, void *);
 	void		       *srq_context;
 	struct ib_srq_attr	attr;
+	enum ib_srq_type	srq_type;
+
+	union {
+		struct {
+			struct ib_xrcd *xrcd;
+			struct ib_cq   *cq;
+		} xrc;
+	} ext;
 };
 
 struct ib_qp_cap {
@@ -565,7 +595,11 @@ enum ib_qp_type {
 	IB_QPT_UC,
 	IB_QPT_UD,
 	IB_QPT_RAW_IPV6,
-	IB_QPT_RAW_ETHERTYPE
+	IB_QPT_RAW_ETHERTYPE,
+	/* Save 8 for RAW_PACKET */
+	IB_QPT_XRC_INI = 9,
+	IB_QPT_XRC_TGT,
+	IB_QPT_MAX
 };
 
 enum ib_qp_create_flags {
@@ -579,6 +613,7 @@ struct ib_qp_init_attr {
 	struct ib_cq	       *send_cq;
 	struct ib_cq	       *recv_cq;
 	struct ib_srq	       *srq;
+	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
 	struct ib_qp_cap	cap;
 	enum ib_sig_type	sq_sig_type;
 	enum ib_qp_type		qp_type;
@@ -586,6 +621,13 @@ struct ib_qp_init_attr {
 	u8			port_num; /* special QP types only */
 };
 
+struct ib_qp_open_attr {
+	void                  (*event_handler)(struct ib_event *, void *);
+	void		       *qp_context;
+	u32			qp_num;
+	enum ib_qp_type		qp_type;
+};
+
 enum ib_rnr_timeout {
 	IB_RNR_TIMER_655_36 =  0,
 	IB_RNR_TIMER_000_01 =  1,
@@ -770,6 +812,7 @@ struct ib_send_wr {
 			u32				rkey;
 		} fast_reg;
 	} wr;
+	u32			xrc_remote_srq_num;	/* XRC TGT QPs only */
 };
 
 struct ib_recv_wr {
@@ -831,6 +874,7 @@ struct ib_ucontext {
 	struct list_head	qp_list;
 	struct list_head	srq_list;
 	struct list_head	ah_list;
+	struct list_head	xrcd_list;
 	int			closing;
 };
 
@@ -858,6 +902,15 @@ struct ib_pd {
 	atomic_t          	usecnt; /* count all resources */
 };
 
+struct ib_xrcd {
+	struct ib_device       *device;
+	atomic_t		usecnt; /* count all exposed resources */
+	struct inode	       *inode;
+
+	struct mutex		tgt_qp_mutex;
+	struct list_head	tgt_qp_list;
+};
+
 struct ib_ah {
 	struct ib_device	*device;
 	struct ib_pd		*pd;
@@ -882,7 +935,16 @@ struct ib_srq {
 	struct ib_uobject      *uobject;
 	void		      (*event_handler)(struct ib_event *, void *);
 	void		       *srq_context;
+	enum ib_srq_type	srq_type;
 	atomic_t		usecnt;
+
+	union {
+		struct {
+			struct ib_xrcd *xrcd;
+			struct ib_cq   *cq;
+			u32		srq_num;
+		} xrc;
+	} ext;
 };
 
 struct ib_qp {
@@ -891,6 +953,11 @@ struct ib_qp {
 	struct ib_cq	       *send_cq;
 	struct ib_cq	       *recv_cq;
 	struct ib_srq	       *srq;
+	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
+	struct list_head	xrcd_list;
+	atomic_t		usecnt; /* count times opened */
+	struct list_head	open_list;
+	struct ib_qp           *real_qp;
 	struct ib_uobject      *uobject;
 	void                  (*event_handler)(struct ib_event *, void *);
 	void		       *qp_context;
@@ -1149,6 +1216,10 @@ struct ib_device {
 						  struct ib_grh *in_grh,
 						  struct ib_mad *in_mad,
 						  struct ib_mad *out_mad);
+	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
+						 struct ib_ucontext *ucontext,
+						 struct ib_udata *udata);
+	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
 
 	struct ib_dma_mapping_ops   *dma_ops;
 
@@ -1442,6 +1513,25 @@ int ib_query_qp(struct ib_qp *qp,
  */
 int ib_destroy_qp(struct ib_qp *qp);
 
+/**
+ * ib_open_qp - Obtain a reference to an existing sharable QP.
+ * @xrcd - XRC domain
+ * @qp_open_attr: Attributes identifying the QP to open.
+ *
+ * Returns a reference to a sharable QP.
+ */
+struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
+			 struct ib_qp_open_attr *qp_open_attr);
+
+/**
+ * ib_close_qp - Release an external reference to a QP.
+ * @qp: The QP handle to release
+ *
+ * The opened QP handle is released by the caller.  The underlying
+ * shared QP is not destroyed until all internal references are released.
+ */
+int ib_close_qp(struct ib_qp *qp);
+
 /**
  * ib_post_send - Posts a list of work requests to the send queue of
  *   the specified QP.
@@ -2060,4 +2150,16 @@ int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
  */
 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
 
+/**
+ * ib_alloc_xrcd - Allocates an XRC domain.
+ * @device: The device on which to allocate the XRC domain.
+ */
+struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
+
+/**
+ * ib_dealloc_xrcd - Deallocates an XRC domain.
+ * @xrcd: The XRC domain to deallocate.
+ */
+int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
+
 #endif /* IB_VERBS_H */

include/rdma/iw_cm.h

@@ -52,8 +52,10 @@ struct iw_cm_event {
 	struct sockaddr_in local_addr;
 	struct sockaddr_in remote_addr;
 	void *private_data;
-	u8 private_data_len;
 	void *provider_data;
+	u8 private_data_len;
+	u8 ord;
+	u8 ird;
 };
 
 /**

include/rdma/rdma_cm.h

@@ -65,6 +65,7 @@ enum rdma_cm_event_type {
 enum rdma_port_space {
 	RDMA_PS_SDP   = 0x0001,
 	RDMA_PS_IPOIB = 0x0002,
+	RDMA_PS_IB    = 0x013F,
 	RDMA_PS_TCP   = 0x0106,
 	RDMA_PS_UDP   = 0x0111,
 };

include/rdma/rdma_user_cm.h

@@ -77,7 +77,8 @@ struct rdma_ucm_create_id {
 	__u64 uid;
 	__u64 response;
 	__u16 ps;
-	__u8  reserved[6];
+	__u8  qp_type;
+	__u8  reserved[5];
 };
 
 struct rdma_ucm_create_id_resp {