Merge /scratch/Ksrc/linux-git/

22 files changed, 8448 insertions, 468 deletions

diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile
index e1a7cf3e8636..10be36731ed7 100644
--- a/drivers/infiniband/core/Makefile
+++ b/drivers/infiniband/core/Makefile
@@ -1,15 +1,20 @@
 EXTRA_CFLAGS += -Idrivers/infiniband/include
 
-obj-$(CONFIG_INFINIBAND) +=		ib_core.o ib_mad.o ib_sa.o ib_umad.o
+obj-$(CONFIG_INFINIBAND) +=		ib_core.o ib_mad.o ib_sa.o \
+					ib_cm.o ib_umad.o ib_ucm.o
 obj-$(CONFIG_INFINIBAND_USER_VERBS) +=	ib_uverbs.o
 
 ib_core-y :=			packer.o ud_header.o verbs.o sysfs.o \
 				device.o fmr_pool.o cache.o
 
-ib_mad-y :=			mad.o smi.o agent.o
+ib_mad-y :=			mad.o smi.o agent.o mad_rmpp.o
 
 ib_sa-y :=			sa_query.o
 
+ib_cm-y :=			cm.o
+
 ib_umad-y :=			user_mad.o
 
+ib_ucm-y :=			ucm.o
+
 ib_uverbs-y :=			uverbs_main.o uverbs_cmd.o uverbs_mem.o
diff --git a/drivers/infiniband/core/agent.c b/drivers/infiniband/core/agent.c
index 23d1957c4b29..729f0b0d983a 100644
--- a/drivers/infiniband/core/agent.c
+++ b/drivers/infiniband/core/agent.c
@@ -134,7 +134,7 @@ static int agent_mad_send(struct ib_mad_agent *mad_agent,
 					  sizeof(mad_priv->mad),
 					  DMA_TO_DEVICE);
 	gather_list.length = sizeof(mad_priv->mad);
-	gather_list.lkey = (*port_priv->mr).lkey;
+	gather_list.lkey = mad_agent->mr->lkey;
 
 	send_wr.next = NULL;
 	send_wr.opcode = IB_WR_SEND;
@@ -156,10 +156,10 @@ static int agent_mad_send(struct ib_mad_agent *mad_agent,
 			/* Should sgid be looked up ? */
 			ah_attr.grh.sgid_index = 0;
 			ah_attr.grh.hop_limit = grh->hop_limit;
-			ah_attr.grh.flow_label = be32_to_cpup(
-				&grh->version_tclass_flow)  & 0xfffff;
-			ah_attr.grh.traffic_class = (be32_to_cpup(
-				&grh->version_tclass_flow) >> 20) & 0xff;
+			ah_attr.grh.flow_label = be32_to_cpu(
+				grh->version_tclass_flow)  & 0xfffff;
+			ah_attr.grh.traffic_class = (be32_to_cpu(
+				grh->version_tclass_flow) >> 20) & 0xff;
 			memcpy(ah_attr.grh.dgid.raw,
 			       grh->sgid.raw,
 			       sizeof(ah_attr.grh.dgid));
@@ -322,22 +322,12 @@ int ib_agent_port_open(struct ib_device *device, int port_num)
 		goto error3;
 	}
 
-	port_priv->mr = ib_get_dma_mr(port_priv->smp_agent->qp->pd,
-				      IB_ACCESS_LOCAL_WRITE);
-	if (IS_ERR(port_priv->mr)) {
-		printk(KERN_ERR SPFX "Couldn't get DMA MR\n");
-		ret = PTR_ERR(port_priv->mr);
-		goto error4;
-	}
-
 	spin_lock_irqsave(&ib_agent_port_list_lock, flags);
 	list_add_tail(&port_priv->port_list, &ib_agent_port_list);
 	spin_unlock_irqrestore(&ib_agent_port_list_lock, flags);
 
 	return 0;
 
-error4:
-	ib_unregister_mad_agent(port_priv->perf_mgmt_agent);
 error3:
 	ib_unregister_mad_agent(port_priv->smp_agent);
 error2:
@@ -361,8 +351,6 @@ int ib_agent_port_close(struct ib_device *device, int port_num)
 	list_del(&port_priv->port_list);
 	spin_unlock_irqrestore(&ib_agent_port_list_lock, flags);
 
-	ib_dereg_mr(port_priv->mr);
-
 	ib_unregister_mad_agent(port_priv->perf_mgmt_agent);
 	ib_unregister_mad_agent(port_priv->smp_agent);
 	kfree(port_priv);
diff --git a/drivers/infiniband/core/agent_priv.h b/drivers/infiniband/core/agent_priv.h
index 17a0cce5813c..17435af1e914 100644
--- a/drivers/infiniband/core/agent_priv.h
+++ b/drivers/infiniband/core/agent_priv.h
@@ -33,7 +33,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: agent_priv.h 1389 2004-12-27 22:56:47Z roland $
+ * $Id: agent_priv.h 1640 2005-01-24 22:39:02Z halr $
  */
 
 #ifndef __IB_AGENT_PRIV_H__
@@ -57,7 +57,6 @@ struct ib_agent_port_private {
 	int port_num;
 	struct ib_mad_agent *smp_agent;	      /* SM class */
 	struct ib_mad_agent *perf_mgmt_agent; /* PerfMgmt class */
-	struct ib_mr *mr;
 };
 
 #endif	/* __IB_AGENT_PRIV_H__ */
diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c
new file mode 100644
index 000000000000..403ed125d8f4
--- /dev/null
+++ b/drivers/infiniband/core/cm.c
@@ -0,0 +1,3324 @@
+/*
+ * Copyright (c) 2004, 2005 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004, 2005 Voltaire Corporation.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, 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.
+ *
+ * $Id: cm.c 2821 2005-07-08 17:07:28Z sean.hefty $
+ */
+#include <linux/dma-mapping.h>
+#include <linux/err.h>
+#include <linux/idr.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/rbtree.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+
+#include <ib_cache.h>
+#include <ib_cm.h>
+#include "cm_msgs.h"
+
+MODULE_AUTHOR("Sean Hefty");
+MODULE_DESCRIPTION("InfiniBand CM");
+MODULE_LICENSE("Dual BSD/GPL");
+
+static void cm_add_one(struct ib_device *device);
+static void cm_remove_one(struct ib_device *device);
+
+static struct ib_client cm_client = {
+	.name   = "cm",
+	.add    = cm_add_one,
+	.remove = cm_remove_one
+};
+
+static struct ib_cm {
+	spinlock_t lock;
+	struct list_head device_list;
+	rwlock_t device_lock;
+	struct rb_root listen_service_table;
+	u64 listen_service_id;
+	/* struct rb_root peer_service_table; todo: fix peer to peer */
+	struct rb_root remote_qp_table;
+	struct rb_root remote_id_table;
+	struct rb_root remote_sidr_table;
+	struct idr local_id_table;
+	struct workqueue_struct *wq;
+} cm;
+
+struct cm_port {
+	struct cm_device *cm_dev;
+	struct ib_mad_agent *mad_agent;
+	u8 port_num;
+};
+
+struct cm_device {
+	struct list_head list;
+	struct ib_device *device;
+	u64 ca_guid;
+	struct cm_port port[0];
+};
+
+struct cm_av {
+	struct cm_port *port;
+	union ib_gid dgid;
+	struct ib_ah_attr ah_attr;
+	u16 pkey_index;
+	u8 packet_life_time;
+};
+
+struct cm_work {
+	struct work_struct work;
+	struct list_head list;
+	struct cm_port *port;
+	struct ib_mad_recv_wc *mad_recv_wc;	/* Received MADs */
+	u32 local_id;				/* Established / timewait */
+	u32 remote_id;
+	struct ib_cm_event cm_event;
+	struct ib_sa_path_rec path[0];
+};
+
+struct cm_timewait_info {
+	struct cm_work work;			/* Must be first. */
+	struct rb_node remote_qp_node;
+	struct rb_node remote_id_node;
+	u64 remote_ca_guid;
+	u32 remote_qpn;
+	u8 inserted_remote_qp;
+	u8 inserted_remote_id;
+};
+
+struct cm_id_private {
+	struct ib_cm_id	id;
+
+	struct rb_node service_node;
+	struct rb_node sidr_id_node;
+	spinlock_t lock;
+	wait_queue_head_t wait;
+	atomic_t refcount;
+
+	struct ib_mad_send_buf *msg;
+	struct cm_timewait_info *timewait_info;
+	/* todo: use alternate port on send failure */
+	struct cm_av av;
+	struct cm_av alt_av;
+
+	void *private_data;
+	u64 tid;
+	u32 local_qpn;
+	u32 remote_qpn;
+	u32 sq_psn;
+	u32 rq_psn;
+	int timeout_ms;
+	enum ib_mtu path_mtu;
+	u8 private_data_len;
+	u8 max_cm_retries;
+	u8 peer_to_peer;
+	u8 responder_resources;
+	u8 initiator_depth;
+	u8 local_ack_timeout;
+	u8 retry_count;
+	u8 rnr_retry_count;
+	u8 service_timeout;
+
+	struct list_head work_list;
+	atomic_t work_count;
+};
+
+static void cm_work_handler(void *data);
+
+static inline void cm_deref_id(struct cm_id_private *cm_id_priv)
+{
+	if (atomic_dec_and_test(&cm_id_priv->refcount))
+		wake_up(&cm_id_priv->wait);
+}
+
+static int cm_alloc_msg(struct cm_id_private *cm_id_priv,
+			struct ib_mad_send_buf **msg)
+{
+	struct ib_mad_agent *mad_agent;
+	struct ib_mad_send_buf *m;
+	struct ib_ah *ah;
+
+	mad_agent = cm_id_priv->av.port->mad_agent;
+	ah = ib_create_ah(mad_agent->qp->pd, &cm_id_priv->av.ah_attr);
+	if (IS_ERR(ah))
+		return PTR_ERR(ah);
+
+	m = ib_create_send_mad(mad_agent, 1, cm_id_priv->av.pkey_index,
+			       ah, 0, sizeof(struct ib_mad_hdr),
+			       sizeof(struct ib_mad)-sizeof(struct ib_mad_hdr),
+			       GFP_ATOMIC);
+	if (IS_ERR(m)) {
+		ib_destroy_ah(ah);
+		return PTR_ERR(m);
+	}
+
+	/* Timeout set by caller if response is expected. */
+	m->send_wr.wr.ud.retries = cm_id_priv->max_cm_retries;
+
+	atomic_inc(&cm_id_priv->refcount);
+	m->context[0] = cm_id_priv;
+	*msg = m;
+	return 0;
+}
+
+static int cm_alloc_response_msg(struct cm_port *port,
+				 struct ib_mad_recv_wc *mad_recv_wc,
+				 struct ib_mad_send_buf **msg)
+{
+	struct ib_mad_send_buf *m;
+	struct ib_ah *ah;
+
+	ah = ib_create_ah_from_wc(port->mad_agent->qp->pd, mad_recv_wc->wc,
+				  mad_recv_wc->recv_buf.grh, port->port_num);
+	if (IS_ERR(ah))
+		return PTR_ERR(ah);
+
+	m = ib_create_send_mad(port->mad_agent, 1, mad_recv_wc->wc->pkey_index,
+			       ah, 0, sizeof(struct ib_mad_hdr),
+			       sizeof(struct ib_mad)-sizeof(struct ib_mad_hdr),
+			       GFP_ATOMIC);
+	if (IS_ERR(m)) {
+		ib_destroy_ah(ah);
+		return PTR_ERR(m);
+	}
+	*msg = m;
+	return 0;
+}
+
+static void cm_free_msg(struct ib_mad_send_buf *msg)
+{
+	ib_destroy_ah(msg->send_wr.wr.ud.ah);
+	if (msg->context[0])
+		cm_deref_id(msg->context[0]);
+	ib_free_send_mad(msg);
+}
+
+static void * cm_copy_private_data(const void *private_data,
+				   u8 private_data_len)
+{
+	void *data;
+
+	if (!private_data || !private_data_len)
+		return NULL;
+
+	data = kmalloc(private_data_len, GFP_KERNEL);
+	if (!data)
+		return ERR_PTR(-ENOMEM);
+
+	memcpy(data, private_data, private_data_len);
+	return data;
+}
+
+static void cm_set_private_data(struct cm_id_private *cm_id_priv,
+				 void *private_data, u8 private_data_len)
+{
+	if (cm_id_priv->private_data && cm_id_priv->private_data_len)
+		kfree(cm_id_priv->private_data);
+
+	cm_id_priv->private_data = private_data;
+	cm_id_priv->private_data_len = private_data_len;
+}
+
+static void cm_set_ah_attr(struct ib_ah_attr *ah_attr, u8 port_num,
+			   u16 dlid, u8 sl, u16 src_path_bits)
+{
+	memset(ah_attr, 0, sizeof ah_attr);
+	ah_attr->dlid = be16_to_cpu(dlid);
+	ah_attr->sl = sl;
+	ah_attr->src_path_bits = src_path_bits;
+	ah_attr->port_num = port_num;
+}
+
+static void cm_init_av_for_response(struct cm_port *port,
+				    struct ib_wc *wc, struct cm_av *av)
+{
+	av->port = port;
+	av->pkey_index = wc->pkey_index;
+	cm_set_ah_attr(&av->ah_attr, port->port_num, cpu_to_be16(wc->slid),
+		       wc->sl, wc->dlid_path_bits);
+}
+
+static int cm_init_av_by_path(struct ib_sa_path_rec *path, struct cm_av *av)
+{
+	struct cm_device *cm_dev;
+	struct cm_port *port = NULL;
+	unsigned long flags;
+	int ret;
+	u8 p;
+
+	read_lock_irqsave(&cm.device_lock, flags);
+	list_for_each_entry(cm_dev, &cm.device_list, list) {
+		if (!ib_find_cached_gid(cm_dev->device, &path->sgid,
+					&p, NULL)) {
+			port = &cm_dev->port[p-1];
+			break;
+		}
+	}
+	read_unlock_irqrestore(&cm.device_lock, flags);
+
+	if (!port)
+		return -EINVAL;
+
+	ret = ib_find_cached_pkey(cm_dev->device, port->port_num,
+				  be16_to_cpu(path->pkey), &av->pkey_index);
+	if (ret)
+		return ret;
+
+	av->port = port;
+	cm_set_ah_attr(&av->ah_attr, av->port->port_num, path->dlid,
+		       path->sl, path->slid & 0x7F);
+	av->packet_life_time = path->packet_life_time;
+	return 0;
+}
+
+static int cm_alloc_id(struct cm_id_private *cm_id_priv)
+{
+	unsigned long flags;
+	int ret;
+
+	do {
+		spin_lock_irqsave(&cm.lock, flags);
+		ret = idr_get_new_above(&cm.local_id_table, cm_id_priv, 1,
+					(int *) &cm_id_priv->id.local_id);
+		spin_unlock_irqrestore(&cm.lock, flags);
+	} while( (ret == -EAGAIN) && idr_pre_get(&cm.local_id_table, GFP_KERNEL) );
+	return ret;
+}
+
+static void cm_free_id(u32 local_id)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&cm.lock, flags);
+	idr_remove(&cm.local_id_table, (int) local_id);
+	spin_unlock_irqrestore(&cm.lock, flags);
+}
+
+static struct cm_id_private * cm_get_id(u32 local_id, u32 remote_id)
+{
+	struct cm_id_private *cm_id_priv;
+
+	cm_id_priv = idr_find(&cm.local_id_table, (int) local_id);
+	if (cm_id_priv) {
+		if (cm_id_priv->id.remote_id == remote_id)
+			atomic_inc(&cm_id_priv->refcount);
+		else
+			cm_id_priv = NULL;
+	}
+
+	return cm_id_priv;
+}
+
+static struct cm_id_private * cm_acquire_id(u32 local_id, u32 remote_id)
+{
+	struct cm_id_private *cm_id_priv;
+	unsigned long flags;
+
+	spin_lock_irqsave(&cm.lock, flags);
+	cm_id_priv = cm_get_id(local_id, remote_id);
+	spin_unlock_irqrestore(&cm.lock, flags);
+
+	return cm_id_priv;
+}
+
+static struct cm_id_private * cm_insert_listen(struct cm_id_private *cm_id_priv)
+{
+	struct rb_node **link = &cm.listen_service_table.rb_node;
+	struct rb_node *parent = NULL;
+	struct cm_id_private *cur_cm_id_priv;
+	u64 service_id = cm_id_priv->id.service_id;
+	u64 service_mask = cm_id_priv->id.service_mask;
+
+	while (*link) {
+		parent = *link;
+		cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
+					  service_node);
+		if ((cur_cm_id_priv->id.service_mask & service_id) ==
+		    (service_mask & cur_cm_id_priv->id.service_id))
+			return cm_id_priv;
+		if (service_id < cur_cm_id_priv->id.service_id)
+			link = &(*link)->rb_left;
+		else
+			link = &(*link)->rb_right;
+	}
+	rb_link_node(&cm_id_priv->service_node, parent, link);
+	rb_insert_color(&cm_id_priv->service_node, &cm.listen_service_table);
+	return NULL;
+}
+
+static struct cm_id_private * cm_find_listen(u64 service_id)
+{
+	struct rb_node *node = cm.listen_service_table.rb_node;
+	struct cm_id_private *cm_id_priv;
+
+	while (node) {
+		cm_id_priv = rb_entry(node, struct cm_id_private, service_node);
+		if ((cm_id_priv->id.service_mask & service_id) ==
+		    (cm_id_priv->id.service_mask & cm_id_priv->id.service_id))
+			return cm_id_priv;
+		if (service_id < cm_id_priv->id.service_id)
+			node = node->rb_left;
+		else
+			node = node->rb_right;
+	}
+	return NULL;
+}
+
+static struct cm_timewait_info * cm_insert_remote_id(struct cm_timewait_info
+						     *timewait_info)
+{
+	struct rb_node **link = &cm.remote_id_table.rb_node;
+	struct rb_node *parent = NULL;
+	struct cm_timewait_info *cur_timewait_info;
+	u64 remote_ca_guid = timewait_info->remote_ca_guid;
+	u32 remote_id = timewait_info->work.remote_id;
+
+	while (*link) {
+		parent = *link;
+		cur_timewait_info = rb_entry(parent, struct cm_timewait_info,
+					     remote_id_node);
+		if (remote_id < cur_timewait_info->work.remote_id)
+			link = &(*link)->rb_left;
+		else if (remote_id > cur_timewait_info->work.remote_id)
+			link = &(*link)->rb_right;
+		else if (remote_ca_guid < cur_timewait_info->remote_ca_guid)
+			link = &(*link)->rb_left;
+		else if (remote_ca_guid > cur_timewait_info->remote_ca_guid)
+			link = &(*link)->rb_right;
+		else
+			return cur_timewait_info;
+	}
+	timewait_info->inserted_remote_id = 1;
+	rb_link_node(&timewait_info->remote_id_node, parent, link);
+	rb_insert_color(&timewait_info->remote_id_node, &cm.remote_id_table);
+	return NULL;
+}
+
+static struct cm_timewait_info * cm_find_remote_id(u64 remote_ca_guid,
+						   u32 remote_id)
+{
+	struct rb_node *node = cm.remote_id_table.rb_node;
+	struct cm_timewait_info *timewait_info;
+
+	while (node) {
+		timewait_info = rb_entry(node, struct cm_timewait_info,
+					 remote_id_node);
+		if (remote_id < timewait_info->work.remote_id)
+			node = node->rb_left;
+		else if (remote_id > timewait_info->work.remote_id)
+			node = node->rb_right;
+		else if (remote_ca_guid < timewait_info->remote_ca_guid)
+			node = node->rb_left;
+		else if (remote_ca_guid > timewait_info->remote_ca_guid)
+			node = node->rb_right;
+		else
+			return timewait_info;
+	}
+	return NULL;
+}
+
+static struct cm_timewait_info * cm_insert_remote_qpn(struct cm_timewait_info
+						      *timewait_info)
+{
+	struct rb_node **link = &cm.remote_qp_table.rb_node;
+	struct rb_node *parent = NULL;
+	struct cm_timewait_info *cur_timewait_info;
+	u64 remote_ca_guid = timewait_info->remote_ca_guid;
+	u32 remote_qpn = timewait_info->remote_qpn;
+
+	while (*link) {
+		parent = *link;
+		cur_timewait_info = rb_entry(parent, struct cm_timewait_info,
+					     remote_qp_node);
+		if (remote_qpn < cur_timewait_info->remote_qpn)
+			link = &(*link)->rb_left;
+		else if (remote_qpn > cur_timewait_info->remote_qpn)
+			link = &(*link)->rb_right;
+		else if (remote_ca_guid < cur_timewait_info->remote_ca_guid)
+			link = &(*link)->rb_left;
+		else if (remote_ca_guid > cur_timewait_info->remote_ca_guid)
+			link = &(*link)->rb_right;
+		else
+			return cur_timewait_info;
+	}
+	timewait_info->inserted_remote_qp = 1;
+	rb_link_node(&timewait_info->remote_qp_node, parent, link);
+	rb_insert_color(&timewait_info->remote_qp_node, &cm.remote_qp_table);
+	return NULL;
+}
+
+static struct cm_id_private * cm_insert_remote_sidr(struct cm_id_private
+						    *cm_id_priv)
+{
+	struct rb_node **link = &cm.remote_sidr_table.rb_node;
+	struct rb_node *parent = NULL;
+	struct cm_id_private *cur_cm_id_priv;
+	union ib_gid *port_gid = &cm_id_priv->av.dgid;
+	u32 remote_id = cm_id_priv->id.remote_id;
+
+	while (*link) {
+		parent = *link;
+		cur_cm_id_priv = rb_entry(parent, struct cm_id_private,
+					  sidr_id_node);
+		if (remote_id < cur_cm_id_priv->id.remote_id)
+			link = &(*link)->rb_left;
+		else if (remote_id > cur_cm_id_priv->id.remote_id)
+			link = &(*link)->rb_right;
+		else {
+			int cmp;
+			cmp = memcmp(port_gid, &cur_cm_id_priv->av.dgid,
+				     sizeof *port_gid);
+			if (cmp < 0)
+				link = &(*link)->rb_left;
+			else if (cmp > 0)
+				link = &(*link)->rb_right;
+			else
+				return cur_cm_id_priv;
+		}
+	}
+	rb_link_node(&cm_id_priv->sidr_id_node, parent, link);
+	rb_insert_color(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
+	return NULL;
+}
+
+static void cm_reject_sidr_req(struct cm_id_private *cm_id_priv,
+			       enum ib_cm_sidr_status status)
+{
+	struct ib_cm_sidr_rep_param param;
+
+	memset(&param, 0, sizeof param);
+	param.status = status;
+	ib_send_cm_sidr_rep(&cm_id_priv->id, &param);
+}
+
+struct ib_cm_id *ib_create_cm_id(ib_cm_handler cm_handler,
+				 void *context)
+{
+	struct cm_id_private *cm_id_priv;
+	int ret;
+
+	cm_id_priv = kmalloc(sizeof *cm_id_priv, GFP_KERNEL);
+	if (!cm_id_priv)
+		return ERR_PTR(-ENOMEM);
+
+	memset(cm_id_priv, 0, sizeof *cm_id_priv);
+	cm_id_priv->id.state = IB_CM_IDLE;
+	cm_id_priv->id.cm_handler = cm_handler;
+	cm_id_priv->id.context = context;
+	ret = cm_alloc_id(cm_id_priv);
+	if (ret)
+		goto error;
+
+	spin_lock_init(&cm_id_priv->lock);
+	init_waitqueue_head(&cm_id_priv->wait);
+	INIT_LIST_HEAD(&cm_id_priv->work_list);
+	atomic_set(&cm_id_priv->work_count, -1);
+	atomic_set(&cm_id_priv->refcount, 1);
+	return &cm_id_priv->id;
+
+error:
+	kfree(cm_id_priv);
+	return ERR_PTR(-ENOMEM);
+}
+EXPORT_SYMBOL(ib_create_cm_id);
+
+static struct cm_work * cm_dequeue_work(struct cm_id_private *cm_id_priv)
+{
+	struct cm_work *work;
+
+	if (list_empty(&cm_id_priv->work_list))
+		return NULL;
+
+	work = list_entry(cm_id_priv->work_list.next, struct cm_work, list);
+	list_del(&work->list);
+	return work;
+}
+
+static void cm_free_work(struct cm_work *work)
+{
+	if (work->mad_recv_wc)
+		ib_free_recv_mad(work->mad_recv_wc);
+	kfree(work);
+}
+
+static inline int cm_convert_to_ms(int iba_time)
+{
+	/* approximate conversion to ms from 4.096us x 2^iba_time */
+	return 1 << max(iba_time - 8, 0);
+}
+
+static void cm_cleanup_timewait(struct cm_timewait_info *timewait_info)
+{
+	unsigned long flags;
+
+	if (!timewait_info->inserted_remote_id &&
+	    !timewait_info->inserted_remote_qp)
+	    return;
+
+	spin_lock_irqsave(&cm.lock, flags);
+	if (timewait_info->inserted_remote_id) {
+		rb_erase(&timewait_info->remote_id_node, &cm.remote_id_table);
+		timewait_info->inserted_remote_id = 0;
+	}
+
+	if (timewait_info->inserted_remote_qp) {
+		rb_erase(&timewait_info->remote_qp_node, &cm.remote_qp_table);
+		timewait_info->inserted_remote_qp = 0;
+	}
+	spin_unlock_irqrestore(&cm.lock, flags);
+}
+
+static struct cm_timewait_info * cm_create_timewait_info(u32 local_id)
+{
+	struct cm_timewait_info *timewait_info;
+
+	timewait_info = kmalloc(sizeof *timewait_info, GFP_KERNEL);
+	if (!timewait_info)
+		return ERR_PTR(-ENOMEM);
+	memset(timewait_info, 0, sizeof *timewait_info);
+
+	timewait_info->work.local_id = local_id;
+	INIT_WORK(&timewait_info->work.work, cm_work_handler,
+		  &timewait_info->work);
+	timewait_info->work.cm_event.event = IB_CM_TIMEWAIT_EXIT;
+	return timewait_info;
+}
+
+static void cm_enter_timewait(struct cm_id_private *cm_id_priv)
+{
+	int wait_time;
+
+	/*
+	 * The cm_id could be destroyed by the user before we exit timewait.
+	 * To protect against this, we search for the cm_id after exiting
+	 * timewait before notifying the user that we've exited timewait.
+	 */
+	cm_id_priv->id.state = IB_CM_TIMEWAIT;
+	wait_time = cm_convert_to_ms(cm_id_priv->local_ack_timeout);
+	queue_delayed_work(cm.wq, &cm_id_priv->timewait_info->work.work,
+			   msecs_to_jiffies(wait_time));
+	cm_id_priv->timewait_info = NULL;
+}
+
+static void cm_reset_to_idle(struct cm_id_private *cm_id_priv)
+{
+	cm_id_priv->id.state = IB_CM_IDLE;
+	if (cm_id_priv->timewait_info) {
+		cm_cleanup_timewait(cm_id_priv->timewait_info);
+		kfree(cm_id_priv->timewait_info);
+		cm_id_priv->timewait_info = NULL;
+	}
+}
+
+void ib_destroy_cm_id(struct ib_cm_id *cm_id)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_work *work;
+	unsigned long flags;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+retest:
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IB_CM_LISTEN:
+		cm_id->state = IB_CM_IDLE;
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		spin_lock_irqsave(&cm.lock, flags);
+		rb_erase(&cm_id_priv->service_node, &cm.listen_service_table);
+		spin_unlock_irqrestore(&cm.lock, flags);
+		break;
+	case IB_CM_SIDR_REQ_SENT:
+		cm_id->state = IB_CM_IDLE;
+		ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+			      (unsigned long) cm_id_priv->msg);
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		break;
+	case IB_CM_SIDR_REQ_RCVD:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_reject_sidr_req(cm_id_priv, IB_SIDR_REJECT);
+		break;
+	case IB_CM_REQ_SENT:
+	case IB_CM_MRA_REQ_RCVD:
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+		ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+			      (unsigned long) cm_id_priv->msg);
+		/* Fall through */
+	case IB_CM_REQ_RCVD:
+	case IB_CM_MRA_REQ_SENT:
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		ib_send_cm_rej(cm_id, IB_CM_REJ_TIMEOUT,
+			       &cm_id_priv->av.port->cm_dev->ca_guid,
+			       sizeof cm_id_priv->av.port->cm_dev->ca_guid,
+			       NULL, 0);
+		break;
+	case IB_CM_ESTABLISHED:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		ib_send_cm_dreq(cm_id, NULL, 0);
+		goto retest;
+	case IB_CM_DREQ_SENT:
+		ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+			      (unsigned long) cm_id_priv->msg);
+		cm_enter_timewait(cm_id_priv);
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		break;
+	case IB_CM_DREQ_RCVD:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		ib_send_cm_drep(cm_id, NULL, 0);
+		break;
+	default:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		break;
+	}
+
+	cm_free_id(cm_id->local_id);
+	atomic_dec(&cm_id_priv->refcount);
+	wait_event(cm_id_priv->wait, !atomic_read(&cm_id_priv->refcount));
+	while ((work = cm_dequeue_work(cm_id_priv)) != NULL)
+		cm_free_work(work);
+	if (cm_id_priv->private_data && cm_id_priv->private_data_len)
+		kfree(cm_id_priv->private_data);
+	kfree(cm_id_priv);
+}
+EXPORT_SYMBOL(ib_destroy_cm_id);
+
+int ib_cm_listen(struct ib_cm_id *cm_id,
+		 u64 service_id,
+		 u64 service_mask)
+{
+	struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
+	unsigned long flags;
+	int ret = 0;
+
+	service_mask = service_mask ? service_mask : ~0ULL;
+	service_id &= service_mask;
+	if ((service_id & IB_SERVICE_ID_AGN_MASK) == IB_CM_ASSIGN_SERVICE_ID &&
+	    (service_id != IB_CM_ASSIGN_SERVICE_ID))
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	BUG_ON(cm_id->state != IB_CM_IDLE);
+
+	cm_id->state = IB_CM_LISTEN;
+
+	spin_lock_irqsave(&cm.lock, flags);
+	if (service_id == IB_CM_ASSIGN_SERVICE_ID) {
+		cm_id->service_id = __cpu_to_be64(cm.listen_service_id++);
+		cm_id->service_mask = ~0ULL;
+	} else {
+		cm_id->service_id = service_id;
+		cm_id->service_mask = service_mask;
+	}
+	cur_cm_id_priv = cm_insert_listen(cm_id_priv);
+	spin_unlock_irqrestore(&cm.lock, flags);
+
+	if (cur_cm_id_priv) {
+		cm_id->state = IB_CM_IDLE;
+		ret = -EBUSY;
+	}
+	return ret;
+}
+EXPORT_SYMBOL(ib_cm_listen);
+
+static u64 cm_form_tid(struct cm_id_private *cm_id_priv,
+		       enum cm_msg_sequence msg_seq)
+{
+	u64 hi_tid, low_tid;
+
+	hi_tid   = ((u64) cm_id_priv->av.port->mad_agent->hi_tid) << 32;
+	low_tid  = (u64) (cm_id_priv->id.local_id | (msg_seq << 30));
+	return cpu_to_be64(hi_tid | low_tid);
+}
+
+static void cm_format_mad_hdr(struct ib_mad_hdr *hdr,
+			      enum cm_msg_attr_id attr_id, u64 tid)
+{
+	hdr->base_version  = IB_MGMT_BASE_VERSION;
+	hdr->mgmt_class	   = IB_MGMT_CLASS_CM;
+	hdr->class_version = IB_CM_CLASS_VERSION;
+	hdr->method	   = IB_MGMT_METHOD_SEND;
+	hdr->attr_id	   = attr_id;
+	hdr->tid	   = tid;
+}
+
+static void cm_format_req(struct cm_req_msg *req_msg,
+			  struct cm_id_private *cm_id_priv,
+			  struct ib_cm_req_param *param)
+{
+	cm_format_mad_hdr(&req_msg->hdr, CM_REQ_ATTR_ID,
+			  cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_REQ));
+
+	req_msg->local_comm_id = cm_id_priv->id.local_id;
+	req_msg->service_id = param->service_id;
+	req_msg->local_ca_guid = cm_id_priv->av.port->cm_dev->ca_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);
+	cm_req_set_qp_type(req_msg, param->qp_type);
+	cm_req_set_flow_ctrl(req_msg, param->flow_control);
+	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);
+
+	req_msg->primary_local_lid = param->primary_path->slid;
+	req_msg->primary_remote_lid = param->primary_path->dlid;
+	req_msg->primary_local_gid = param->primary_path->sgid;
+	req_msg->primary_remote_gid = param->primary_path->dgid;
+	cm_req_set_primary_flow_label(req_msg, param->primary_path->flow_label);
+	cm_req_set_primary_packet_rate(req_msg, param->primary_path->rate);
+	req_msg->primary_traffic_class = param->primary_path->traffic_class;
+	req_msg->primary_hop_limit = param->primary_path->hop_limit;
+	cm_req_set_primary_sl(req_msg, param->primary_path->sl);
+	cm_req_set_primary_subnet_local(req_msg, 1); /* local only... */
+	cm_req_set_primary_local_ack_timeout(req_msg,
+		min(31, param->primary_path->packet_life_time + 1));
+
+	if (param->alternate_path) {
+		req_msg->alt_local_lid = param->alternate_path->slid;
+		req_msg->alt_remote_lid = param->alternate_path->dlid;
+		req_msg->alt_local_gid = param->alternate_path->sgid;
+		req_msg->alt_remote_gid = param->alternate_path->dgid;
+		cm_req_set_alt_flow_label(req_msg,
+					  param->alternate_path->flow_label);
+		cm_req_set_alt_packet_rate(req_msg, param->alternate_path->rate);
+		req_msg->alt_traffic_class = param->alternate_path->traffic_class;
+		req_msg->alt_hop_limit = param->alternate_path->hop_limit;
+		cm_req_set_alt_sl(req_msg, param->alternate_path->sl);
+		cm_req_set_alt_subnet_local(req_msg, 1); /* local only... */
+		cm_req_set_alt_local_ack_timeout(req_msg,
+			min(31, param->alternate_path->packet_life_time + 1));
+	}
+
+	if (param->private_data && param->private_data_len)
+		memcpy(req_msg->private_data, param->private_data,
+		       param->private_data_len);
+}
+
+static inline int cm_validate_req_param(struct ib_cm_req_param *param)
+{
+	/* peer-to-peer not supported */
+	if (param->peer_to_peer)
+		return -EINVAL;
+
+	if (!param->primary_path)
+		return -EINVAL;
+
+	if (param->qp_type != IB_QPT_RC && param->qp_type != IB_QPT_UC)
+		return -EINVAL;
+
+	if (param->private_data &&
+	    param->private_data_len > IB_CM_REQ_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	if (param->alternate_path &&
+	    (param->alternate_path->pkey != param->primary_path->pkey ||
+	     param->alternate_path->mtu != param->primary_path->mtu))
+		return -EINVAL;
+
+	return 0;
+}
+
+int ib_send_cm_req(struct ib_cm_id *cm_id,
+		   struct ib_cm_req_param *param)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_send_wr *bad_send_wr;
+	struct cm_req_msg *req_msg;
+	unsigned long flags;
+	int ret;
+
+	ret = cm_validate_req_param(param);
+	if (ret)
+		return ret;
+
+	/* Verify that we're not in timewait. */
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_IDLE) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		ret = -EINVAL;
+		goto out;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
+							    id.local_id);
+	if (IS_ERR(cm_id_priv->timewait_info))
+		goto out;
+
+	ret = cm_init_av_by_path(param->primary_path, &cm_id_priv->av);
+	if (ret)
+		goto error1;
+	if (param->alternate_path) {
+		ret = cm_init_av_by_path(param->alternate_path,
+					 &cm_id_priv->alt_av);
+		if (ret)
+			goto error1;
+	}
+	cm_id->service_id = param->service_id;
+	cm_id->service_mask = ~0ULL;
+	cm_id_priv->timeout_ms = cm_convert_to_ms(
+				    param->primary_path->packet_life_time) * 2 +
+				 cm_convert_to_ms(
+				    param->remote_cm_response_timeout);
+	cm_id_priv->max_cm_retries = param->max_cm_retries;
+	cm_id_priv->initiator_depth = param->initiator_depth;
+	cm_id_priv->responder_resources = param->responder_resources;
+	cm_id_priv->retry_count = param->retry_count;
+	cm_id_priv->path_mtu = param->primary_path->mtu;
+
+	ret = cm_alloc_msg(cm_id_priv, &cm_id_priv->msg);
+	if (ret)
+		goto error1;
+
+	req_msg = (struct cm_req_msg *) cm_id_priv->msg->mad;
+	cm_format_req(req_msg, cm_id_priv, param);
+	cm_id_priv->tid = req_msg->hdr.tid;
+	cm_id_priv->msg->send_wr.wr.ud.timeout_ms = cm_id_priv->timeout_ms;
+	cm_id_priv->msg->context[1] = (void *) (unsigned long) IB_CM_REQ_SENT;
+
+	cm_id_priv->local_qpn = cm_req_get_local_qpn(req_msg);
+	cm_id_priv->rq_psn = cm_req_get_starting_psn(req_msg);
+	cm_id_priv->local_ack_timeout =
+				cm_req_get_primary_local_ack_timeout(req_msg);
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				&cm_id_priv->msg->send_wr, &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto error2;
+	}
+	BUG_ON(cm_id->state != IB_CM_IDLE);
+	cm_id->state = IB_CM_REQ_SENT;
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return 0;
+
+error2:	cm_free_msg(cm_id_priv->msg);
+error1:	kfree(cm_id_priv->timewait_info);
+out:	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_req);
+
+static int cm_issue_rej(struct cm_port *port,
+			struct ib_mad_recv_wc *mad_recv_wc,
+			enum ib_cm_rej_reason reason,
+			enum cm_msg_response msg_rejected,
+			void *ari, u8 ari_length)
+{
+	struct ib_mad_send_buf *msg = NULL;
+	struct ib_send_wr *bad_send_wr;
+	struct cm_rej_msg *rej_msg, *rcv_msg;
+	int ret;
+
+	ret = cm_alloc_response_msg(port, mad_recv_wc, &msg);
+	if (ret)
+		return ret;
+
+	/* We just need common CM header information.  Cast to any message. */
+	rcv_msg = (struct cm_rej_msg *) mad_recv_wc->recv_buf.mad;
+	rej_msg = (struct cm_rej_msg *) msg->mad;
+
+	cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, rcv_msg->hdr.tid);
+	rej_msg->remote_comm_id = rcv_msg->local_comm_id;
+	rej_msg->local_comm_id = rcv_msg->remote_comm_id;
+	cm_rej_set_msg_rejected(rej_msg, msg_rejected);
+	rej_msg->reason = reason;
+
+	if (ari && ari_length) {
+		cm_rej_set_reject_info_len(rej_msg, ari_length);
+		memcpy(rej_msg->ari, ari, ari_length);
+	}
+
+	ret = ib_post_send_mad(port->mad_agent, &msg->send_wr, &bad_send_wr);
+	if (ret)
+		cm_free_msg(msg);
+
+	return ret;
+}
+
+static inline int cm_is_active_peer(u64 local_ca_guid, u64 remote_ca_guid,
+				    u32 local_qpn, u32 remote_qpn)
+{
+	return (be64_to_cpu(local_ca_guid) > be64_to_cpu(remote_ca_guid) ||
+		((local_ca_guid == remote_ca_guid) &&
+		 (be32_to_cpu(local_qpn) > be32_to_cpu(remote_qpn))));
+}
+
+static inline void cm_format_paths_from_req(struct cm_req_msg *req_msg,
+					    struct ib_sa_path_rec *primary_path,
+					    struct ib_sa_path_rec *alt_path)
+{
+	memset(primary_path, 0, sizeof *primary_path);
+	primary_path->dgid = req_msg->primary_local_gid;
+	primary_path->sgid = req_msg->primary_remote_gid;
+	primary_path->dlid = req_msg->primary_local_lid;
+	primary_path->slid = req_msg->primary_remote_lid;
+	primary_path->flow_label = cm_req_get_primary_flow_label(req_msg);
+	primary_path->hop_limit = req_msg->primary_hop_limit;
+	primary_path->traffic_class = req_msg->primary_traffic_class;
+	primary_path->reversible = 1;
+	primary_path->pkey = req_msg->pkey;
+	primary_path->sl = cm_req_get_primary_sl(req_msg);
+	primary_path->mtu_selector = IB_SA_EQ;
+	primary_path->mtu = cm_req_get_path_mtu(req_msg);
+	primary_path->rate_selector = IB_SA_EQ;
+	primary_path->rate = cm_req_get_primary_packet_rate(req_msg);
+	primary_path->packet_life_time_selector = IB_SA_EQ;
+	primary_path->packet_life_time =
+		cm_req_get_primary_local_ack_timeout(req_msg);
+	primary_path->packet_life_time -= (primary_path->packet_life_time > 0);
+
+	if (req_msg->alt_local_lid) {
+		memset(alt_path, 0, sizeof *alt_path);
+		alt_path->dgid = req_msg->alt_local_gid;
+		alt_path->sgid = req_msg->alt_remote_gid;
+		alt_path->dlid = req_msg->alt_local_lid;
+		alt_path->slid = req_msg->alt_remote_lid;
+		alt_path->flow_label = cm_req_get_alt_flow_label(req_msg);
+		alt_path->hop_limit = req_msg->alt_hop_limit;
+		alt_path->traffic_class = req_msg->alt_traffic_class;
+		alt_path->reversible = 1;
+		alt_path->pkey = req_msg->pkey;
+		alt_path->sl = cm_req_get_alt_sl(req_msg);
+		alt_path->mtu_selector = IB_SA_EQ;
+		alt_path->mtu = cm_req_get_path_mtu(req_msg);
+		alt_path->rate_selector = IB_SA_EQ;
+		alt_path->rate = cm_req_get_alt_packet_rate(req_msg);
+		alt_path->packet_life_time_selector = IB_SA_EQ;
+		alt_path->packet_life_time =
+			cm_req_get_alt_local_ack_timeout(req_msg);
+		alt_path->packet_life_time -= (alt_path->packet_life_time > 0);
+	}
+}
+
+static void cm_format_req_event(struct cm_work *work,
+				struct cm_id_private *cm_id_priv,
+				struct ib_cm_id *listen_id)
+{
+	struct cm_req_msg *req_msg;
+	struct ib_cm_req_event_param *param;
+
+	req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
+	param = &work->cm_event.param.req_rcvd;
+	param->listen_id = listen_id;
+	param->device = cm_id_priv->av.port->mad_agent->device;
+	param->port = cm_id_priv->av.port->port_num;
+	param->primary_path = &work->path[0];
+	if (req_msg->alt_local_lid)
+		param->alternate_path = &work->path[1];
+	else
+		param->alternate_path = NULL;
+	param->remote_ca_guid = req_msg->local_ca_guid;
+	param->remote_qkey = be32_to_cpu(req_msg->local_qkey);
+	param->remote_qpn = be32_to_cpu(cm_req_get_local_qpn(req_msg));
+	param->qp_type = cm_req_get_qp_type(req_msg);
+	param->starting_psn = be32_to_cpu(cm_req_get_starting_psn(req_msg));
+	param->responder_resources = cm_req_get_init_depth(req_msg);
+	param->initiator_depth = cm_req_get_resp_res(req_msg);
+	param->local_cm_response_timeout =
+					cm_req_get_remote_resp_timeout(req_msg);
+	param->flow_control = cm_req_get_flow_ctrl(req_msg);
+	param->remote_cm_response_timeout =
+					cm_req_get_local_resp_timeout(req_msg);
+	param->retry_count = cm_req_get_retry_count(req_msg);
+	param->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg);
+	param->srq = cm_req_get_srq(req_msg);
+	work->cm_event.private_data = &req_msg->private_data;
+}
+
+static void cm_process_work(struct cm_id_private *cm_id_priv,
+			    struct cm_work *work)
+{
+	unsigned long flags;
+	int ret;
+
+	/* We will typically only have the current event to report. */
+	ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &work->cm_event);
+	cm_free_work(work);
+
+	while (!ret && !atomic_add_negative(-1, &cm_id_priv->work_count)) {
+		spin_lock_irqsave(&cm_id_priv->lock, flags);
+		work = cm_dequeue_work(cm_id_priv);
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		BUG_ON(!work);
+		ret = cm_id_priv->id.cm_handler(&cm_id_priv->id,
+						&work->cm_event);
+		cm_free_work(work);
+	}
+	cm_deref_id(cm_id_priv);
+	if (ret)
+		ib_destroy_cm_id(&cm_id_priv->id);
+}
+
+static void cm_format_mra(struct cm_mra_msg *mra_msg,
+			  struct cm_id_private *cm_id_priv,
+			  enum cm_msg_response msg_mraed, u8 service_timeout,
+			  const void *private_data, u8 private_data_len)
+{
+	cm_format_mad_hdr(&mra_msg->hdr, CM_MRA_ATTR_ID, cm_id_priv->tid);
+	cm_mra_set_msg_mraed(mra_msg, msg_mraed);
+	mra_msg->local_comm_id = cm_id_priv->id.local_id;
+	mra_msg->remote_comm_id = cm_id_priv->id.remote_id;
+	cm_mra_set_service_timeout(mra_msg, service_timeout);
+
+	if (private_data && private_data_len)
+		memcpy(mra_msg->private_data, private_data, private_data_len);
+}
+
+static void cm_format_rej(struct cm_rej_msg *rej_msg,
+			  struct cm_id_private *cm_id_priv,
+			  enum ib_cm_rej_reason reason,
+			  void *ari,
+			  u8 ari_length,
+			  const void *private_data,
+			  u8 private_data_len)
+{
+	cm_format_mad_hdr(&rej_msg->hdr, CM_REJ_ATTR_ID, cm_id_priv->tid);
+	rej_msg->remote_comm_id = cm_id_priv->id.remote_id;
+
+	switch(cm_id_priv->id.state) {
+	case IB_CM_REQ_RCVD:
+		rej_msg->local_comm_id = 0;
+		cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ);
+		break;
+	case IB_CM_MRA_REQ_SENT:
+		rej_msg->local_comm_id = cm_id_priv->id.local_id;
+		cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REQ);
+		break;
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+		rej_msg->local_comm_id = cm_id_priv->id.local_id;
+		cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_REP);
+		break;
+	default:
+		rej_msg->local_comm_id = cm_id_priv->id.local_id;
+		cm_rej_set_msg_rejected(rej_msg, CM_MSG_RESPONSE_OTHER);
+		break;
+	}
+
+	rej_msg->reason = reason;
+	if (ari && ari_length) {
+		cm_rej_set_reject_info_len(rej_msg, ari_length);
+		memcpy(rej_msg->ari, ari, ari_length);
+	}
+
+	if (private_data && private_data_len)
+		memcpy(rej_msg->private_data, private_data, private_data_len);
+}
+
+static void cm_dup_req_handler(struct cm_work *work,
+			       struct cm_id_private *cm_id_priv)
+{
+	struct ib_mad_send_buf *msg = NULL;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	/* Quick state check to discard duplicate REQs. */
+	if (cm_id_priv->id.state == IB_CM_REQ_RCVD)
+		return;
+
+	ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
+	if (ret)
+		return;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_MRA_REQ_SENT:
+		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
+			      CM_MSG_RESPONSE_REQ, cm_id_priv->service_timeout,
+			      cm_id_priv->private_data,
+			      cm_id_priv->private_data_len);
+		break;
+	case IB_CM_TIMEWAIT:
+		cm_format_rej((struct cm_rej_msg *) msg->mad, cm_id_priv,
+			      IB_CM_REJ_STALE_CONN, NULL, 0, NULL, 0);
+		break;
+	default:
+		goto unlock;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent, &msg->send_wr,
+			       &bad_send_wr);
+	if (ret)
+		goto free;
+	return;
+
+unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+free:	cm_free_msg(msg);
+}
+
+static struct cm_id_private * cm_match_req(struct cm_work *work,
+					   struct cm_id_private *cm_id_priv)
+{
+	struct cm_id_private *listen_cm_id_priv, *cur_cm_id_priv;
+	struct cm_timewait_info *timewait_info;
+	struct cm_req_msg *req_msg;
+	unsigned long flags;
+
+	req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
+
+	/* Check for duplicate REQ and stale connections. */
+	spin_lock_irqsave(&cm.lock, flags);
+	timewait_info = cm_insert_remote_id(cm_id_priv->timewait_info);
+	if (!timewait_info)
+		timewait_info = cm_insert_remote_qpn(cm_id_priv->timewait_info);
+
+	if (timewait_info) {
+		cur_cm_id_priv = cm_get_id(timewait_info->work.local_id,
+					   timewait_info->work.remote_id);
+		spin_unlock_irqrestore(&cm.lock, flags);
+		if (cur_cm_id_priv) {
+			cm_dup_req_handler(work, cur_cm_id_priv);
+			cm_deref_id(cur_cm_id_priv);
+		} else
+			cm_issue_rej(work->port, work->mad_recv_wc,
+				     IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REQ,
+				     NULL, 0);
+		goto error;
+	}
+
+	/* Find matching listen request. */
+	listen_cm_id_priv = cm_find_listen(req_msg->service_id);
+	if (!listen_cm_id_priv) {
+		spin_unlock_irqrestore(&cm.lock, flags);
+		cm_issue_rej(work->port, work->mad_recv_wc,
+			     IB_CM_REJ_INVALID_SERVICE_ID, CM_MSG_RESPONSE_REQ,
+			     NULL, 0);
+		goto error;
+	}
+	atomic_inc(&listen_cm_id_priv->refcount);
+	atomic_inc(&cm_id_priv->refcount);
+	cm_id_priv->id.state = IB_CM_REQ_RCVD;
+	atomic_inc(&cm_id_priv->work_count);
+	spin_unlock_irqrestore(&cm.lock, flags);
+	return listen_cm_id_priv;
+
+error:	cm_cleanup_timewait(cm_id_priv->timewait_info);
+	return NULL;
+}
+
+static int cm_req_handler(struct cm_work *work)
+{
+	struct ib_cm_id *cm_id;
+	struct cm_id_private *cm_id_priv, *listen_cm_id_priv;
+	struct cm_req_msg *req_msg;
+	int ret;
+
+	req_msg = (struct cm_req_msg *)work->mad_recv_wc->recv_buf.mad;
+
+	cm_id = ib_create_cm_id(NULL, NULL);
+	if (IS_ERR(cm_id))
+		return PTR_ERR(cm_id);
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	cm_id_priv->id.remote_id = req_msg->local_comm_id;
+	cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
+				&cm_id_priv->av);
+	cm_id_priv->timewait_info = cm_create_timewait_info(cm_id_priv->
+							    id.local_id);
+	if (IS_ERR(cm_id_priv->timewait_info)) {
+		ret = PTR_ERR(cm_id_priv->timewait_info);
+		goto error1;
+	}
+	cm_id_priv->timewait_info->work.remote_id = req_msg->local_comm_id;
+	cm_id_priv->timewait_info->remote_ca_guid = req_msg->local_ca_guid;
+	cm_id_priv->timewait_info->remote_qpn = cm_req_get_local_qpn(req_msg);
+
+	listen_cm_id_priv = cm_match_req(work, cm_id_priv);
+	if (!listen_cm_id_priv) {
+		ret = -EINVAL;
+		goto error2;
+	}
+
+	cm_id_priv->id.cm_handler = listen_cm_id_priv->id.cm_handler;
+	cm_id_priv->id.context = listen_cm_id_priv->id.context;
+	cm_id_priv->id.service_id = req_msg->service_id;
+	cm_id_priv->id.service_mask = ~0ULL;
+
+	cm_format_paths_from_req(req_msg, &work->path[0], &work->path[1]);
+	ret = cm_init_av_by_path(&work->path[0], &cm_id_priv->av);
+	if (ret)
+		goto error3;
+	if (req_msg->alt_local_lid) {
+		ret = cm_init_av_by_path(&work->path[1], &cm_id_priv->alt_av);
+		if (ret)
+			goto error3;
+	}
+	cm_id_priv->tid = req_msg->hdr.tid;
+	cm_id_priv->timeout_ms = cm_convert_to_ms(
+					cm_req_get_local_resp_timeout(req_msg));
+	cm_id_priv->max_cm_retries = cm_req_get_max_cm_retries(req_msg);
+	cm_id_priv->remote_qpn = cm_req_get_local_qpn(req_msg);
+	cm_id_priv->initiator_depth = cm_req_get_resp_res(req_msg);
+	cm_id_priv->responder_resources = cm_req_get_init_depth(req_msg);
+	cm_id_priv->path_mtu = cm_req_get_path_mtu(req_msg);
+	cm_id_priv->sq_psn = cm_req_get_starting_psn(req_msg);
+	cm_id_priv->local_ack_timeout =
+				cm_req_get_primary_local_ack_timeout(req_msg);
+	cm_id_priv->retry_count = cm_req_get_retry_count(req_msg);
+	cm_id_priv->rnr_retry_count = cm_req_get_rnr_retry_count(req_msg);
+
+	cm_format_req_event(work, cm_id_priv, &listen_cm_id_priv->id);
+	cm_process_work(cm_id_priv, work);
+	cm_deref_id(listen_cm_id_priv);
+	return 0;
+
+error3:	atomic_dec(&cm_id_priv->refcount);
+	cm_deref_id(listen_cm_id_priv);
+	cm_cleanup_timewait(cm_id_priv->timewait_info);
+error2:	kfree(cm_id_priv->timewait_info);
+error1:	ib_destroy_cm_id(&cm_id_priv->id);
+	return ret;
+}
+
+static void cm_format_rep(struct cm_rep_msg *rep_msg,
+			  struct cm_id_private *cm_id_priv,
+			  struct ib_cm_rep_param *param)
+{
+	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, param->target_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->av.port->cm_dev->ca_guid;
+
+	if (param->private_data && param->private_data_len)
+		memcpy(rep_msg->private_data, param->private_data,
+		       param->private_data_len);
+}
+
+int ib_send_cm_rep(struct ib_cm_id *cm_id,
+		   struct ib_cm_rep_param *param)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct cm_rep_msg *rep_msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if (param->private_data &&
+	    param->private_data_len > IB_CM_REP_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_REQ_RCVD &&
+	    cm_id->state != IB_CM_MRA_REQ_SENT) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto out;
+
+	rep_msg = (struct cm_rep_msg *) msg->mad;
+	cm_format_rep(rep_msg, cm_id_priv, param);
+	msg->send_wr.wr.ud.timeout_ms = cm_id_priv->timeout_ms;
+	msg->context[1] = (void *) (unsigned long) IB_CM_REP_SENT;
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		return ret;
+	}
+
+	cm_id->state = IB_CM_REP_SENT;
+	cm_id_priv->msg = msg;
+	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);
+
+out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_rep);
+
+static void cm_format_rtu(struct cm_rtu_msg *rtu_msg,
+			  struct cm_id_private *cm_id_priv,
+			  const void *private_data,
+			  u8 private_data_len)
+{
+	cm_format_mad_hdr(&rtu_msg->hdr, CM_RTU_ATTR_ID, cm_id_priv->tid);
+	rtu_msg->local_comm_id = cm_id_priv->id.local_id;
+	rtu_msg->remote_comm_id = cm_id_priv->id.remote_id;
+
+	if (private_data && private_data_len)
+		memcpy(rtu_msg->private_data, private_data, private_data_len);
+}
+
+int ib_send_cm_rtu(struct ib_cm_id *cm_id,
+		   const void *private_data,
+		   u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	void *data;
+	int ret;
+
+	if (private_data && private_data_len > IB_CM_RTU_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	data = cm_copy_private_data(private_data, private_data_len);
+	if (IS_ERR(data))
+		return PTR_ERR(data);
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_REP_RCVD &&
+	    cm_id->state != IB_CM_MRA_REP_SENT) {
+		ret = -EINVAL;
+		goto error;
+	}
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto error;
+
+	cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
+		      private_data, private_data_len);
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		kfree(data);
+		return ret;
+	}
+
+	cm_id->state = IB_CM_ESTABLISHED;
+	cm_set_private_data(cm_id_priv, data, private_data_len);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return 0;
+
+error:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	kfree(data);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_rtu);
+
+static void cm_format_rep_event(struct cm_work *work)
+{
+	struct cm_rep_msg *rep_msg;
+	struct ib_cm_rep_event_param *param;
+
+	rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
+	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->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;
+	param->target_ack_delay = cm_rep_get_target_ack_delay(rep_msg);
+	param->failover_accepted = cm_rep_get_failover(rep_msg);
+	param->flow_control = cm_rep_get_flow_ctrl(rep_msg);
+	param->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
+	param->srq = cm_rep_get_srq(rep_msg);
+	work->cm_event.private_data = &rep_msg->private_data;
+}
+
+static void cm_dup_rep_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_rep_msg *rep_msg;
+	struct ib_mad_send_buf *msg = NULL;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	rep_msg = (struct cm_rep_msg *) work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id,
+				   rep_msg->local_comm_id);
+	if (!cm_id_priv)
+		return;
+
+	ret = cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg);
+	if (ret)
+		goto deref;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state == IB_CM_ESTABLISHED)
+		cm_format_rtu((struct cm_rtu_msg *) msg->mad, cm_id_priv,
+			      cm_id_priv->private_data,
+			      cm_id_priv->private_data_len);
+	else if (cm_id_priv->id.state == IB_CM_MRA_REP_SENT)
+		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
+			      CM_MSG_RESPONSE_REP, cm_id_priv->service_timeout,
+			      cm_id_priv->private_data,
+			      cm_id_priv->private_data_len);
+	else
+		goto unlock;
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent, &msg->send_wr,
+			       &bad_send_wr);
+	if (ret)
+		goto free;
+	goto deref;
+
+unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+free:	cm_free_msg(msg);
+deref:	cm_deref_id(cm_id_priv);
+}
+
+static int cm_rep_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_rep_msg *rep_msg;
+	unsigned long flags;
+	int ret;
+
+	rep_msg = (struct cm_rep_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(rep_msg->remote_comm_id, 0);
+	if (!cm_id_priv) {
+		cm_dup_rep_handler(work);
+		return -EINVAL;
+	}
+
+	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);
+
+	spin_lock_irqsave(&cm.lock, flags);
+	/* Check for duplicate REP. */
+	if (cm_insert_remote_id(cm_id_priv->timewait_info)) {
+		spin_unlock_irqrestore(&cm.lock, flags);
+		ret = -EINVAL;
+		goto error;
+	}
+	/* Check for a stale connection. */
+	if (cm_insert_remote_qpn(cm_id_priv->timewait_info)) {
+		spin_unlock_irqrestore(&cm.lock, flags);
+		cm_issue_rej(work->port, work->mad_recv_wc,
+			     IB_CM_REJ_STALE_CONN, CM_MSG_RESPONSE_REP,
+			     NULL, 0);
+		ret = -EINVAL;
+		goto error;
+	}
+	spin_unlock_irqrestore(&cm.lock, flags);
+
+	cm_format_rep_event(work);
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REQ_SENT:
+	case IB_CM_MRA_REQ_RCVD:
+		break;
+	default:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		ret = -EINVAL;
+		goto error;
+	}
+	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->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);
+	cm_id_priv->rnr_retry_count = cm_rep_get_rnr_retry_count(rep_msg);
+
+	/* todo: handle peer_to_peer */
+
+	ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+		      (unsigned long) cm_id_priv->msg);
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+
+error:	cm_cleanup_timewait(cm_id_priv->timewait_info);
+	cm_deref_id(cm_id_priv);
+	return ret;
+}
+
+static int cm_establish_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	unsigned long flags;
+	int ret;
+
+	/* See comment in ib_cm_establish about lookup. */
+	cm_id_priv = cm_acquire_id(work->local_id, work->remote_id);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_ESTABLISHED) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto out;
+	}
+
+	ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+		      (unsigned long) cm_id_priv->msg);
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static int cm_rtu_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_rtu_msg *rtu_msg;
+	unsigned long flags;
+	int ret;
+
+	rtu_msg = (struct cm_rtu_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(rtu_msg->remote_comm_id,
+				   rtu_msg->local_comm_id);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	work->cm_event.private_data = &rtu_msg->private_data;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_REP_SENT &&
+	    cm_id_priv->id.state != IB_CM_MRA_REP_RCVD) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto out;
+	}
+	cm_id_priv->id.state = IB_CM_ESTABLISHED;
+
+	ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+		      (unsigned long) cm_id_priv->msg);
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static void cm_format_dreq(struct cm_dreq_msg *dreq_msg,
+			  struct cm_id_private *cm_id_priv,
+			  const void *private_data,
+			  u8 private_data_len)
+{
+	cm_format_mad_hdr(&dreq_msg->hdr, CM_DREQ_ATTR_ID,
+			  cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_DREQ));
+	dreq_msg->local_comm_id = cm_id_priv->id.local_id;
+	dreq_msg->remote_comm_id = cm_id_priv->id.remote_id;
+	cm_dreq_set_remote_qpn(dreq_msg, cm_id_priv->remote_qpn);
+
+	if (private_data && private_data_len)
+		memcpy(dreq_msg->private_data, private_data, private_data_len);
+}
+
+int ib_send_cm_dreq(struct ib_cm_id *cm_id,
+		    const void *private_data,
+		    u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if (private_data && private_data_len > IB_CM_DREQ_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_ESTABLISHED) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret) {
+		cm_enter_timewait(cm_id_priv);
+		goto out;
+	}
+
+	cm_format_dreq((struct cm_dreq_msg *) msg->mad, cm_id_priv,
+		       private_data, private_data_len);
+	msg->send_wr.wr.ud.timeout_ms = cm_id_priv->timeout_ms;
+	msg->context[1] = (void *) (unsigned long) IB_CM_DREQ_SENT;
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret) {
+		cm_enter_timewait(cm_id_priv);
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		return ret;
+	}
+
+	cm_id->state = IB_CM_DREQ_SENT;
+	cm_id_priv->msg = msg;
+out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_dreq);
+
+static void cm_format_drep(struct cm_drep_msg *drep_msg,
+			  struct cm_id_private *cm_id_priv,
+			  const void *private_data,
+			  u8 private_data_len)
+{
+	cm_format_mad_hdr(&drep_msg->hdr, CM_DREP_ATTR_ID, cm_id_priv->tid);
+	drep_msg->local_comm_id = cm_id_priv->id.local_id;
+	drep_msg->remote_comm_id = cm_id_priv->id.remote_id;
+
+	if (private_data && private_data_len)
+		memcpy(drep_msg->private_data, private_data, private_data_len);
+}
+
+int ib_send_cm_drep(struct ib_cm_id *cm_id,
+		    const void *private_data,
+		    u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	void *data;
+	int ret;
+
+	if (private_data && private_data_len > IB_CM_DREP_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	data = cm_copy_private_data(private_data, private_data_len);
+	if (IS_ERR(data))
+		return PTR_ERR(data);
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_DREQ_RCVD) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		kfree(data);
+		return -EINVAL;
+	}
+
+	cm_set_private_data(cm_id_priv, data, private_data_len);
+	cm_enter_timewait(cm_id_priv);
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto out;
+
+	cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
+		       private_data, private_data_len);
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent, &msg->send_wr,
+			       &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		return ret;
+	}
+
+out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_drep);
+
+static int cm_dreq_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_dreq_msg *dreq_msg;
+	struct ib_mad_send_buf *msg = NULL;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	dreq_msg = (struct cm_dreq_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(dreq_msg->remote_comm_id,
+				   dreq_msg->local_comm_id);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	work->cm_event.private_data = &dreq_msg->private_data;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->local_qpn != cm_dreq_get_remote_qpn(dreq_msg))
+		goto unlock;
+
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REP_SENT:
+	case IB_CM_DREQ_SENT:
+		ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+			      (unsigned long) cm_id_priv->msg);
+		break;
+	case IB_CM_ESTABLISHED:
+	case IB_CM_MRA_REP_RCVD:
+		break;
+	case IB_CM_TIMEWAIT:
+		if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
+			goto unlock;
+
+		cm_format_drep((struct cm_drep_msg *) msg->mad, cm_id_priv,
+			       cm_id_priv->private_data,
+			       cm_id_priv->private_data_len);
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+		if (ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				     &msg->send_wr, &bad_send_wr))
+			cm_free_msg(msg);
+		goto deref;
+	default:
+		goto unlock;
+	}
+	cm_id_priv->id.state = IB_CM_DREQ_RCVD;
+	cm_id_priv->tid = dreq_msg->hdr.tid;
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+
+unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+deref:	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static int cm_drep_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_drep_msg *drep_msg;
+	unsigned long flags;
+	int ret;
+
+	drep_msg = (struct cm_drep_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(drep_msg->remote_comm_id,
+				   drep_msg->local_comm_id);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	work->cm_event.private_data = &drep_msg->private_data;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_DREQ_SENT &&
+	    cm_id_priv->id.state != IB_CM_DREQ_RCVD) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto out;
+	}
+	cm_enter_timewait(cm_id_priv);
+
+	ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+		      (unsigned long) cm_id_priv->msg);
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+int ib_send_cm_rej(struct ib_cm_id *cm_id,
+		   enum ib_cm_rej_reason reason,
+		   void *ari,
+		   u8 ari_length,
+		   const void *private_data,
+		   u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if ((private_data && private_data_len > IB_CM_REJ_PRIVATE_DATA_SIZE) ||
+	    (ari && ari_length > IB_CM_REJ_ARI_LENGTH))
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id->state) {
+	case IB_CM_REQ_SENT:
+	case IB_CM_MRA_REQ_RCVD:
+	case IB_CM_REQ_RCVD:
+	case IB_CM_MRA_REQ_SENT:
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+		ret = cm_alloc_msg(cm_id_priv, &msg);
+		if (!ret)
+			cm_format_rej((struct cm_rej_msg *) msg->mad,
+				      cm_id_priv, reason, ari, ari_length,
+				      private_data, private_data_len);
+
+		cm_reset_to_idle(cm_id_priv);
+		break;
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+		ret = cm_alloc_msg(cm_id_priv, &msg);
+		if (!ret)
+			cm_format_rej((struct cm_rej_msg *) msg->mad,
+				      cm_id_priv, reason, ari, ari_length,
+				      private_data, private_data_len);
+
+		cm_enter_timewait(cm_id_priv);
+		break;
+	default:
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (ret)
+		goto out;
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret)
+		cm_free_msg(msg);
+
+out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_rej);
+
+static void cm_format_rej_event(struct cm_work *work)
+{
+	struct cm_rej_msg *rej_msg;
+	struct ib_cm_rej_event_param *param;
+
+	rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
+	param = &work->cm_event.param.rej_rcvd;
+	param->ari = rej_msg->ari;
+	param->ari_length = cm_rej_get_reject_info_len(rej_msg);
+	param->reason = rej_msg->reason;
+	work->cm_event.private_data = &rej_msg->private_data;
+}
+
+static struct cm_id_private * cm_acquire_rejected_id(struct cm_rej_msg *rej_msg)
+{
+	struct cm_timewait_info *timewait_info;
+	struct cm_id_private *cm_id_priv;
+	unsigned long flags;
+	u32 remote_id;
+
+	remote_id = rej_msg->local_comm_id;
+
+	if (rej_msg->reason == IB_CM_REJ_TIMEOUT) {
+		spin_lock_irqsave(&cm.lock, flags);
+		timewait_info = cm_find_remote_id( *((u64 *) rej_msg->ari),
+						  remote_id);
+		if (!timewait_info) {
+			spin_unlock_irqrestore(&cm.lock, flags);
+			return NULL;
+		}
+		cm_id_priv = idr_find(&cm.local_id_table,
+				      (int) timewait_info->work.local_id);
+		if (cm_id_priv) {
+			if (cm_id_priv->id.remote_id == remote_id)
+				atomic_inc(&cm_id_priv->refcount);
+			else
+				cm_id_priv = NULL;
+		}
+		spin_unlock_irqrestore(&cm.lock, flags);
+	} else if (cm_rej_get_msg_rejected(rej_msg) == CM_MSG_RESPONSE_REQ)
+		cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, 0);
+	else
+		cm_id_priv = cm_acquire_id(rej_msg->remote_comm_id, remote_id);
+
+	return cm_id_priv;
+}
+
+static int cm_rej_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_rej_msg *rej_msg;
+	unsigned long flags;
+	int ret;
+
+	rej_msg = (struct cm_rej_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_rejected_id(rej_msg);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	cm_format_rej_event(work);
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REQ_SENT:
+	case IB_CM_MRA_REQ_RCVD:
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+		ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+			      (unsigned long) cm_id_priv->msg);
+		/* fall through */
+	case IB_CM_REQ_RCVD:
+	case IB_CM_MRA_REQ_SENT:
+		if (rej_msg->reason == IB_CM_REJ_STALE_CONN)
+			cm_enter_timewait(cm_id_priv);
+		else
+			cm_reset_to_idle(cm_id_priv);
+		break;
+	case IB_CM_DREQ_SENT:
+		ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+			      (unsigned long) cm_id_priv->msg);
+		/* fall through */
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+	case IB_CM_ESTABLISHED:
+		cm_enter_timewait(cm_id_priv);
+		break;
+	default:
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+int ib_send_cm_mra(struct ib_cm_id *cm_id,
+		   u8 service_timeout,
+		   const void *private_data,
+		   u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	void *data;
+	unsigned long flags;
+	int ret;
+
+	if (private_data && private_data_len > IB_CM_MRA_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	data = cm_copy_private_data(private_data, private_data_len);
+	if (IS_ERR(data))
+		return PTR_ERR(data);
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch(cm_id_priv->id.state) {
+	case IB_CM_REQ_RCVD:
+		ret = cm_alloc_msg(cm_id_priv, &msg);
+		if (ret)
+			goto error1;
+
+		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
+			      CM_MSG_RESPONSE_REQ, service_timeout,
+			      private_data, private_data_len);
+		ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				       &msg->send_wr, &bad_send_wr);
+		if (ret)
+			goto error2;
+		cm_id->state = IB_CM_MRA_REQ_SENT;
+		break;
+	case IB_CM_REP_RCVD:
+		ret = cm_alloc_msg(cm_id_priv, &msg);
+		if (ret)
+			goto error1;
+
+		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
+			      CM_MSG_RESPONSE_REP, service_timeout,
+			      private_data, private_data_len);
+		ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				       &msg->send_wr, &bad_send_wr);
+		if (ret)
+			goto error2;
+		cm_id->state = IB_CM_MRA_REP_SENT;
+		break;
+	case IB_CM_ESTABLISHED:
+		ret = cm_alloc_msg(cm_id_priv, &msg);
+		if (ret)
+			goto error1;
+
+		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
+			      CM_MSG_RESPONSE_OTHER, service_timeout,
+			      private_data, private_data_len);
+		ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				       &msg->send_wr, &bad_send_wr);
+		if (ret)
+			goto error2;
+		cm_id->lap_state = IB_CM_MRA_LAP_SENT;
+		break;
+	default:
+		ret = -EINVAL;
+		goto error1;
+	}
+	cm_id_priv->service_timeout = service_timeout;
+	cm_set_private_data(cm_id_priv, data, private_data_len);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return 0;
+
+error1:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	kfree(data);
+	return ret;
+
+error2:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	kfree(data);
+	cm_free_msg(msg);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_mra);
+
+static struct cm_id_private * cm_acquire_mraed_id(struct cm_mra_msg *mra_msg)
+{
+	switch (cm_mra_get_msg_mraed(mra_msg)) {
+	case CM_MSG_RESPONSE_REQ:
+		return cm_acquire_id(mra_msg->remote_comm_id, 0);
+	case CM_MSG_RESPONSE_REP:
+	case CM_MSG_RESPONSE_OTHER:
+		return cm_acquire_id(mra_msg->remote_comm_id,
+				     mra_msg->local_comm_id);
+	default:
+		return NULL;
+	}
+}
+
+static int cm_mra_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_mra_msg *mra_msg;
+	unsigned long flags;
+	int timeout, ret;
+
+	mra_msg = (struct cm_mra_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_mraed_id(mra_msg);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	work->cm_event.private_data = &mra_msg->private_data;
+	work->cm_event.param.mra_rcvd.service_timeout =
+					cm_mra_get_service_timeout(mra_msg);
+	timeout = cm_convert_to_ms(cm_mra_get_service_timeout(mra_msg)) +
+		  cm_convert_to_ms(cm_id_priv->av.packet_life_time);
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REQ_SENT:
+		if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REQ ||
+		    ib_modify_mad(cm_id_priv->av.port->mad_agent,
+				  (unsigned long) cm_id_priv->msg, timeout))
+			goto out;
+		cm_id_priv->id.state = IB_CM_MRA_REQ_RCVD;
+		break;
+	case IB_CM_REP_SENT:
+		if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_REP ||
+		    ib_modify_mad(cm_id_priv->av.port->mad_agent,
+				  (unsigned long) cm_id_priv->msg, timeout))
+			goto out;
+		cm_id_priv->id.state = IB_CM_MRA_REP_RCVD;
+		break;
+	case IB_CM_ESTABLISHED:
+		if (cm_mra_get_msg_mraed(mra_msg) != CM_MSG_RESPONSE_OTHER ||
+		    cm_id_priv->id.lap_state != IB_CM_LAP_SENT ||
+		    ib_modify_mad(cm_id_priv->av.port->mad_agent,
+				  (unsigned long) cm_id_priv->msg, timeout))
+			goto out;
+		cm_id_priv->id.lap_state = IB_CM_MRA_LAP_RCVD;
+		break;
+	default:
+		goto out;
+	}
+
+	cm_id_priv->msg->context[1] = (void *) (unsigned long)
+				      cm_id_priv->id.state;
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static void cm_format_lap(struct cm_lap_msg *lap_msg,
+			  struct cm_id_private *cm_id_priv,
+			  struct ib_sa_path_rec *alternate_path,
+			  const void *private_data,
+			  u8 private_data_len)
+{
+	cm_format_mad_hdr(&lap_msg->hdr, CM_LAP_ATTR_ID,
+			  cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_LAP));
+	lap_msg->local_comm_id = cm_id_priv->id.local_id;
+	lap_msg->remote_comm_id = cm_id_priv->id.remote_id;
+	cm_lap_set_remote_qpn(lap_msg, cm_id_priv->remote_qpn);
+	/* todo: need remote CM response timeout */
+	cm_lap_set_remote_resp_timeout(lap_msg, 0x1F);
+	lap_msg->alt_local_lid = alternate_path->slid;
+	lap_msg->alt_remote_lid = alternate_path->dlid;
+	lap_msg->alt_local_gid = alternate_path->sgid;
+	lap_msg->alt_remote_gid = alternate_path->dgid;
+	cm_lap_set_flow_label(lap_msg, alternate_path->flow_label);
+	cm_lap_set_traffic_class(lap_msg, alternate_path->traffic_class);
+	lap_msg->alt_hop_limit = alternate_path->hop_limit;
+	cm_lap_set_packet_rate(lap_msg, alternate_path->rate);
+	cm_lap_set_sl(lap_msg, alternate_path->sl);
+	cm_lap_set_subnet_local(lap_msg, 1); /* local only... */
+	cm_lap_set_local_ack_timeout(lap_msg,
+		min(31, alternate_path->packet_life_time + 1));
+
+	if (private_data && private_data_len)
+		memcpy(lap_msg->private_data, private_data, private_data_len);
+}
+
+int ib_send_cm_lap(struct ib_cm_id *cm_id,
+		   struct ib_sa_path_rec *alternate_path,
+		   const void *private_data,
+		   u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if (private_data && private_data_len > IB_CM_LAP_PRIVATE_DATA_SIZE)
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_ESTABLISHED ||
+	    cm_id->lap_state != IB_CM_LAP_IDLE) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto out;
+
+	cm_format_lap((struct cm_lap_msg *) msg->mad, cm_id_priv,
+		      alternate_path, private_data, private_data_len);
+	msg->send_wr.wr.ud.timeout_ms = cm_id_priv->timeout_ms;
+	msg->context[1] = (void *) (unsigned long) IB_CM_ESTABLISHED;
+
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		return ret;
+	}
+
+	cm_id->lap_state = IB_CM_LAP_SENT;
+	cm_id_priv->msg = msg;
+
+out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_lap);
+
+static void cm_format_path_from_lap(struct ib_sa_path_rec *path,
+				    struct cm_lap_msg *lap_msg)
+{
+	memset(path, 0, sizeof *path);
+	path->dgid = lap_msg->alt_local_gid;
+	path->sgid = lap_msg->alt_remote_gid;
+	path->dlid = lap_msg->alt_local_lid;
+	path->slid = lap_msg->alt_remote_lid;
+	path->flow_label = cm_lap_get_flow_label(lap_msg);
+	path->hop_limit = lap_msg->alt_hop_limit;
+	path->traffic_class = cm_lap_get_traffic_class(lap_msg);
+	path->reversible = 1;
+	/* pkey is same as in REQ */
+	path->sl = cm_lap_get_sl(lap_msg);
+	path->mtu_selector = IB_SA_EQ;
+	/* mtu is same as in REQ */
+	path->rate_selector = IB_SA_EQ;
+	path->rate = cm_lap_get_packet_rate(lap_msg);
+	path->packet_life_time_selector = IB_SA_EQ;
+	path->packet_life_time = cm_lap_get_local_ack_timeout(lap_msg);
+	path->packet_life_time -= (path->packet_life_time > 0);
+}
+
+static int cm_lap_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_lap_msg *lap_msg;
+	struct ib_cm_lap_event_param *param;
+	struct ib_mad_send_buf *msg = NULL;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	/* todo: verify LAP request and send reject APR if invalid. */
+	lap_msg = (struct cm_lap_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(lap_msg->remote_comm_id,
+				   lap_msg->local_comm_id);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	param = &work->cm_event.param.lap_rcvd;
+	param->alternate_path = &work->path[0];
+	cm_format_path_from_lap(param->alternate_path, lap_msg);
+	work->cm_event.private_data = &lap_msg->private_data;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_ESTABLISHED)
+		goto unlock;
+
+	switch (cm_id_priv->id.lap_state) {
+	case IB_CM_LAP_IDLE:
+		break;
+	case IB_CM_MRA_LAP_SENT:
+		if (cm_alloc_response_msg(work->port, work->mad_recv_wc, &msg))
+			goto unlock;
+
+		cm_format_mra((struct cm_mra_msg *) msg->mad, cm_id_priv,
+			      CM_MSG_RESPONSE_OTHER,
+			      cm_id_priv->service_timeout,
+			      cm_id_priv->private_data,
+			      cm_id_priv->private_data_len);
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+		if (ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				     &msg->send_wr, &bad_send_wr))
+			cm_free_msg(msg);
+		goto deref;
+	default:
+		goto unlock;
+	}
+
+	cm_id_priv->id.lap_state = IB_CM_LAP_RCVD;
+	cm_id_priv->tid = lap_msg->hdr.tid;
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+
+unlock:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+deref:	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static void cm_format_apr(struct cm_apr_msg *apr_msg,
+			  struct cm_id_private *cm_id_priv,
+			  enum ib_cm_apr_status status,
+			  void *info,
+			  u8 info_length,
+			  const void *private_data,
+			  u8 private_data_len)
+{
+	cm_format_mad_hdr(&apr_msg->hdr, CM_APR_ATTR_ID, cm_id_priv->tid);
+	apr_msg->local_comm_id = cm_id_priv->id.local_id;
+	apr_msg->remote_comm_id = cm_id_priv->id.remote_id;
+	apr_msg->ap_status = (u8) status;
+
+	if (info && info_length) {
+		apr_msg->info_length = info_length;
+		memcpy(apr_msg->info, info, info_length);
+	}
+
+	if (private_data && private_data_len)
+		memcpy(apr_msg->private_data, private_data, private_data_len);
+}
+
+int ib_send_cm_apr(struct ib_cm_id *cm_id,
+		   enum ib_cm_apr_status status,
+		   void *info,
+		   u8 info_length,
+		   const void *private_data,
+		   u8 private_data_len)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if ((private_data && private_data_len > IB_CM_APR_PRIVATE_DATA_SIZE) ||
+	    (info && info_length > IB_CM_APR_INFO_LENGTH))
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_ESTABLISHED ||
+	    (cm_id->lap_state != IB_CM_LAP_RCVD &&
+	     cm_id->lap_state != IB_CM_MRA_LAP_SENT)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto out;
+
+	cm_format_apr((struct cm_apr_msg *) msg->mad, cm_id_priv, status,
+		      info, info_length, private_data, private_data_len);
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		return ret;
+	}
+
+	cm_id->lap_state = IB_CM_LAP_IDLE;
+out:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_apr);
+
+static int cm_apr_handler(struct cm_work *work)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_apr_msg *apr_msg;
+	unsigned long flags;
+	int ret;
+
+	apr_msg = (struct cm_apr_msg *)work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(apr_msg->remote_comm_id,
+				   apr_msg->local_comm_id);
+	if (!cm_id_priv)
+		return -EINVAL; /* Unmatched reply. */
+
+	work->cm_event.param.apr_rcvd.ap_status = apr_msg->ap_status;
+	work->cm_event.param.apr_rcvd.apr_info = &apr_msg->info;
+	work->cm_event.param.apr_rcvd.info_len = apr_msg->info_length;
+	work->cm_event.private_data = &apr_msg->private_data;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_ESTABLISHED ||
+	    (cm_id_priv->id.lap_state != IB_CM_LAP_SENT &&
+	     cm_id_priv->id.lap_state != IB_CM_MRA_LAP_RCVD)) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto out;
+	}
+	cm_id_priv->id.lap_state = IB_CM_LAP_IDLE;
+	ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+		      (unsigned long) cm_id_priv->msg);
+	cm_id_priv->msg = NULL;
+
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static int cm_timewait_handler(struct cm_work *work)
+{
+	struct cm_timewait_info *timewait_info;
+	struct cm_id_private *cm_id_priv;
+	unsigned long flags;
+	int ret;
+
+	timewait_info = (struct cm_timewait_info *)work;
+	cm_cleanup_timewait(timewait_info);
+
+	cm_id_priv = cm_acquire_id(timewait_info->work.local_id,
+				   timewait_info->work.remote_id);
+	if (!cm_id_priv)
+		return -EINVAL;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_TIMEWAIT ||
+	    cm_id_priv->remote_qpn != timewait_info->remote_qpn) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto out;
+	}
+	cm_id_priv->id.state = IB_CM_IDLE;
+	ret = atomic_inc_and_test(&cm_id_priv->work_count);
+	if (!ret)
+		list_add_tail(&work->list, &cm_id_priv->work_list);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret)
+		cm_process_work(cm_id_priv, work);
+	else
+		cm_deref_id(cm_id_priv);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static void cm_format_sidr_req(struct cm_sidr_req_msg *sidr_req_msg,
+			       struct cm_id_private *cm_id_priv,
+			       struct ib_cm_sidr_req_param *param)
+{
+	cm_format_mad_hdr(&sidr_req_msg->hdr, CM_SIDR_REQ_ATTR_ID,
+			  cm_form_tid(cm_id_priv, CM_MSG_SEQUENCE_SIDR));
+	sidr_req_msg->request_id = cm_id_priv->id.local_id;
+	sidr_req_msg->pkey = param->pkey;
+	sidr_req_msg->service_id = param->service_id;
+
+	if (param->private_data && param->private_data_len)
+		memcpy(sidr_req_msg->private_data, param->private_data,
+		       param->private_data_len);
+}
+
+int ib_send_cm_sidr_req(struct ib_cm_id *cm_id,
+			struct ib_cm_sidr_req_param *param)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if (!param->path || (param->private_data &&
+	     param->private_data_len > IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE))
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	ret = cm_init_av_by_path(param->path, &cm_id_priv->av);
+	if (ret)
+		goto out;
+
+	cm_id->service_id = param->service_id;
+	cm_id->service_mask = ~0ULL;
+	cm_id_priv->timeout_ms = param->timeout_ms;
+	cm_id_priv->max_cm_retries = param->max_cm_retries;
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto out;
+
+	cm_format_sidr_req((struct cm_sidr_req_msg *) msg->mad, cm_id_priv,
+			   param);
+	msg->send_wr.wr.ud.timeout_ms = cm_id_priv->timeout_ms;
+	msg->context[1] = (void *) (unsigned long) IB_CM_SIDR_REQ_SENT;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state == IB_CM_IDLE)
+		ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+				       &msg->send_wr, &bad_send_wr);
+	else
+		ret = -EINVAL;
+
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		goto out;
+	}
+	cm_id->state = IB_CM_SIDR_REQ_SENT;
+	cm_id_priv->msg = msg;
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+out:
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_sidr_req);
+
+static void cm_format_sidr_req_event(struct cm_work *work,
+				     struct ib_cm_id *listen_id)
+{
+	struct cm_sidr_req_msg *sidr_req_msg;
+	struct ib_cm_sidr_req_event_param *param;
+
+	sidr_req_msg = (struct cm_sidr_req_msg *)
+				work->mad_recv_wc->recv_buf.mad;
+	param = &work->cm_event.param.sidr_req_rcvd;
+	param->pkey = sidr_req_msg->pkey;
+	param->listen_id = listen_id;
+	param->device = work->port->mad_agent->device;
+	param->port = work->port->port_num;
+	work->cm_event.private_data = &sidr_req_msg->private_data;
+}
+
+static int cm_sidr_req_handler(struct cm_work *work)
+{
+	struct ib_cm_id *cm_id;
+	struct cm_id_private *cm_id_priv, *cur_cm_id_priv;
+	struct cm_sidr_req_msg *sidr_req_msg;
+	struct ib_wc *wc;
+	unsigned long flags;
+
+	cm_id = ib_create_cm_id(NULL, NULL);
+	if (IS_ERR(cm_id))
+		return PTR_ERR(cm_id);
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+
+	/* Record SGID/SLID and request ID for lookup. */
+	sidr_req_msg = (struct cm_sidr_req_msg *)
+				work->mad_recv_wc->recv_buf.mad;
+	wc = work->mad_recv_wc->wc;
+	cm_id_priv->av.dgid.global.subnet_prefix = wc->slid;
+	cm_id_priv->av.dgid.global.interface_id = 0;
+	cm_init_av_for_response(work->port, work->mad_recv_wc->wc,
+				&cm_id_priv->av);
+	cm_id_priv->id.remote_id = sidr_req_msg->request_id;
+	cm_id_priv->id.state = IB_CM_SIDR_REQ_RCVD;
+	cm_id_priv->tid = sidr_req_msg->hdr.tid;
+	atomic_inc(&cm_id_priv->work_count);
+
+	spin_lock_irqsave(&cm.lock, flags);
+	cur_cm_id_priv = cm_insert_remote_sidr(cm_id_priv);
+	if (cur_cm_id_priv) {
+		spin_unlock_irqrestore(&cm.lock, flags);
+		goto out; /* Duplicate message. */
+	}
+	cur_cm_id_priv = cm_find_listen(sidr_req_msg->service_id);
+	if (!cur_cm_id_priv) {
+		rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
+		spin_unlock_irqrestore(&cm.lock, flags);
+		/* todo: reply with no match */
+		goto out; /* No match. */
+	}
+	atomic_inc(&cur_cm_id_priv->refcount);
+	spin_unlock_irqrestore(&cm.lock, flags);
+
+	cm_id_priv->id.cm_handler = cur_cm_id_priv->id.cm_handler;
+	cm_id_priv->id.context = cur_cm_id_priv->id.context;
+	cm_id_priv->id.service_id = sidr_req_msg->service_id;
+	cm_id_priv->id.service_mask = ~0ULL;
+
+	cm_format_sidr_req_event(work, &cur_cm_id_priv->id);
+	cm_process_work(cm_id_priv, work);
+	cm_deref_id(cur_cm_id_priv);
+	return 0;
+out:
+	ib_destroy_cm_id(&cm_id_priv->id);
+	return -EINVAL;
+}
+
+static void cm_format_sidr_rep(struct cm_sidr_rep_msg *sidr_rep_msg,
+			       struct cm_id_private *cm_id_priv,
+			       struct ib_cm_sidr_rep_param *param)
+{
+	cm_format_mad_hdr(&sidr_rep_msg->hdr, CM_SIDR_REP_ATTR_ID,
+			  cm_id_priv->tid);
+	sidr_rep_msg->request_id = cm_id_priv->id.remote_id;
+	sidr_rep_msg->status = param->status;
+	cm_sidr_rep_set_qpn(sidr_rep_msg, cpu_to_be32(param->qp_num));
+	sidr_rep_msg->service_id = cm_id_priv->id.service_id;
+	sidr_rep_msg->qkey = cpu_to_be32(param->qkey);
+
+	if (param->info && param->info_length)
+		memcpy(sidr_rep_msg->info, param->info, param->info_length);
+
+	if (param->private_data && param->private_data_len)
+		memcpy(sidr_rep_msg->private_data, param->private_data,
+		       param->private_data_len);
+}
+
+int ib_send_cm_sidr_rep(struct ib_cm_id *cm_id,
+			struct ib_cm_sidr_rep_param *param)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	unsigned long flags;
+	int ret;
+
+	if ((param->info && param->info_length > IB_CM_SIDR_REP_INFO_LENGTH) ||
+	    (param->private_data &&
+	     param->private_data_len > IB_CM_SIDR_REP_PRIVATE_DATA_SIZE))
+		return -EINVAL;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id->state != IB_CM_SIDR_REQ_RCVD) {
+		ret = -EINVAL;
+		goto error;
+	}
+
+	ret = cm_alloc_msg(cm_id_priv, &msg);
+	if (ret)
+		goto error;
+
+	cm_format_sidr_rep((struct cm_sidr_rep_msg *) msg->mad, cm_id_priv,
+			   param);
+	ret = ib_post_send_mad(cm_id_priv->av.port->mad_agent,
+			       &msg->send_wr, &bad_send_wr);
+	if (ret) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		cm_free_msg(msg);
+		return ret;
+	}
+	cm_id->state = IB_CM_IDLE;
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	spin_lock_irqsave(&cm.lock, flags);
+	rb_erase(&cm_id_priv->sidr_id_node, &cm.remote_sidr_table);
+	spin_unlock_irqrestore(&cm.lock, flags);
+	return 0;
+
+error:	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+EXPORT_SYMBOL(ib_send_cm_sidr_rep);
+
+static void cm_format_sidr_rep_event(struct cm_work *work)
+{
+	struct cm_sidr_rep_msg *sidr_rep_msg;
+	struct ib_cm_sidr_rep_event_param *param;
+
+	sidr_rep_msg = (struct cm_sidr_rep_msg *)
+				work->mad_recv_wc->recv_buf.mad;
+	param = &work->cm_event.param.sidr_rep_rcvd;
+	param->status = sidr_rep_msg->status;
+	param->qkey = be32_to_cpu(sidr_rep_msg->qkey);
+	param->qpn = be32_to_cpu(cm_sidr_rep_get_qpn(sidr_rep_msg));
+	param->info = &sidr_rep_msg->info;
+	param->info_len = sidr_rep_msg->info_length;
+	work->cm_event.private_data = &sidr_rep_msg->private_data;
+}
+
+static int cm_sidr_rep_handler(struct cm_work *work)
+{
+	struct cm_sidr_rep_msg *sidr_rep_msg;
+	struct cm_id_private *cm_id_priv;
+	unsigned long flags;
+
+	sidr_rep_msg = (struct cm_sidr_rep_msg *)
+				work->mad_recv_wc->recv_buf.mad;
+	cm_id_priv = cm_acquire_id(sidr_rep_msg->request_id, 0);
+	if (!cm_id_priv)
+		return -EINVAL; /* Unmatched reply. */
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	if (cm_id_priv->id.state != IB_CM_SIDR_REQ_SENT) {
+		spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+		goto out;
+	}
+	cm_id_priv->id.state = IB_CM_IDLE;
+	ib_cancel_mad(cm_id_priv->av.port->mad_agent,
+		      (unsigned long) cm_id_priv->msg);
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	cm_format_sidr_rep_event(work);
+	cm_process_work(cm_id_priv, work);
+	return 0;
+out:
+	cm_deref_id(cm_id_priv);
+	return -EINVAL;
+}
+
+static void cm_process_send_error(struct ib_mad_send_buf *msg,
+				  enum ib_wc_status wc_status)
+{
+	struct cm_id_private *cm_id_priv;
+	struct ib_cm_event cm_event;
+	enum ib_cm_state state;
+	unsigned long flags;
+	int ret;
+
+	memset(&cm_event, 0, sizeof cm_event);
+	cm_id_priv = msg->context[0];
+
+	/* Discard old sends or ones without a response. */
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	state = (enum ib_cm_state) (unsigned long) msg->context[1];
+	if (msg != cm_id_priv->msg || state != cm_id_priv->id.state)
+		goto discard;
+
+	switch (state) {
+	case IB_CM_REQ_SENT:
+	case IB_CM_MRA_REQ_RCVD:
+		cm_reset_to_idle(cm_id_priv);
+		cm_event.event = IB_CM_REQ_ERROR;
+		break;
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+		cm_reset_to_idle(cm_id_priv);
+		cm_event.event = IB_CM_REP_ERROR;
+		break;
+	case IB_CM_DREQ_SENT:
+		cm_enter_timewait(cm_id_priv);
+		cm_event.event = IB_CM_DREQ_ERROR;
+		break;
+	case IB_CM_SIDR_REQ_SENT:
+		cm_id_priv->id.state = IB_CM_IDLE;
+		cm_event.event = IB_CM_SIDR_REQ_ERROR;
+		break;
+	default:
+		goto discard;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	cm_event.param.send_status = wc_status;
+
+	/* No other events can occur on the cm_id at this point. */
+	ret = cm_id_priv->id.cm_handler(&cm_id_priv->id, &cm_event);
+	cm_free_msg(msg);
+	if (ret)
+		ib_destroy_cm_id(&cm_id_priv->id);
+	return;
+discard:
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	cm_free_msg(msg);
+}
+
+static void cm_send_handler(struct ib_mad_agent *mad_agent,
+			    struct ib_mad_send_wc *mad_send_wc)
+{
+	struct ib_mad_send_buf *msg;
+
+	msg = (struct ib_mad_send_buf *)(unsigned long)mad_send_wc->wr_id;
+
+	switch (mad_send_wc->status) {
+	case IB_WC_SUCCESS:
+	case IB_WC_WR_FLUSH_ERR:
+		cm_free_msg(msg);
+		break;
+	default:
+		if (msg->context[0] && msg->context[1])
+			cm_process_send_error(msg, mad_send_wc->status);
+		else
+			cm_free_msg(msg);
+		break;
+	}
+}
+
+static void cm_work_handler(void *data)
+{
+	struct cm_work *work = data;
+	int ret;
+
+	switch (work->cm_event.event) {
+	case IB_CM_REQ_RECEIVED:
+		ret = cm_req_handler(work);
+		break;
+	case IB_CM_MRA_RECEIVED:
+		ret = cm_mra_handler(work);
+		break;
+	case IB_CM_REJ_RECEIVED:
+		ret = cm_rej_handler(work);
+		break;
+	case IB_CM_REP_RECEIVED:
+		ret = cm_rep_handler(work);
+		break;
+	case IB_CM_RTU_RECEIVED:
+		ret = cm_rtu_handler(work);
+		break;
+	case IB_CM_USER_ESTABLISHED:
+		ret = cm_establish_handler(work);
+		break;
+	case IB_CM_DREQ_RECEIVED:
+		ret = cm_dreq_handler(work);
+		break;
+	case IB_CM_DREP_RECEIVED:
+		ret = cm_drep_handler(work);
+		break;
+	case IB_CM_SIDR_REQ_RECEIVED:
+		ret = cm_sidr_req_handler(work);
+		break;
+	case IB_CM_SIDR_REP_RECEIVED:
+		ret = cm_sidr_rep_handler(work);
+		break;
+	case IB_CM_LAP_RECEIVED:
+		ret = cm_lap_handler(work);
+		break;
+	case IB_CM_APR_RECEIVED:
+		ret = cm_apr_handler(work);
+		break;
+	case IB_CM_TIMEWAIT_EXIT:
+		ret = cm_timewait_handler(work);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	if (ret)
+		cm_free_work(work);
+}
+
+int ib_cm_establish(struct ib_cm_id *cm_id)
+{
+	struct cm_id_private *cm_id_priv;
+	struct cm_work *work;
+	unsigned long flags;
+	int ret = 0;
+
+	work = kmalloc(sizeof *work, GFP_ATOMIC);
+	if (!work)
+		return -ENOMEM;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id->state)
+	{
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+		cm_id->state = IB_CM_ESTABLISHED;
+		break;
+	case IB_CM_ESTABLISHED:
+		ret = -EISCONN;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+
+	if (ret) {
+		kfree(work);
+		goto out;
+	}
+
+	/*
+	 * The CM worker thread may try to destroy the cm_id before it
+	 * can execute this work item.  To prevent potential deadlock,
+	 * we need to find the cm_id once we're in the context of the
+	 * worker thread, rather than holding a reference on it.
+	 */
+	INIT_WORK(&work->work, cm_work_handler, work);
+	work->local_id = cm_id->local_id;
+	work->remote_id = cm_id->remote_id;
+	work->mad_recv_wc = NULL;
+	work->cm_event.event = IB_CM_USER_ESTABLISHED;
+	queue_work(cm.wq, &work->work);
+out:
+	return ret;
+}
+EXPORT_SYMBOL(ib_cm_establish);
+
+static void cm_recv_handler(struct ib_mad_agent *mad_agent,
+			    struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct cm_work *work;
+	enum ib_cm_event_type event;
+	int paths = 0;
+
+	switch (mad_recv_wc->recv_buf.mad->mad_hdr.attr_id) {
+	case CM_REQ_ATTR_ID:
+		paths = 1 + (((struct cm_req_msg *) mad_recv_wc->recv_buf.mad)->
+						    alt_local_lid != 0);
+		event = IB_CM_REQ_RECEIVED;
+		break;
+	case CM_MRA_ATTR_ID:
+		event = IB_CM_MRA_RECEIVED;
+		break;
+	case CM_REJ_ATTR_ID:
+		event = IB_CM_REJ_RECEIVED;
+		break;
+	case CM_REP_ATTR_ID:
+		event = IB_CM_REP_RECEIVED;
+		break;
+	case CM_RTU_ATTR_ID:
+		event = IB_CM_RTU_RECEIVED;
+		break;
+	case CM_DREQ_ATTR_ID:
+		event = IB_CM_DREQ_RECEIVED;
+		break;
+	case CM_DREP_ATTR_ID:
+		event = IB_CM_DREP_RECEIVED;
+		break;
+	case CM_SIDR_REQ_ATTR_ID:
+		event = IB_CM_SIDR_REQ_RECEIVED;
+		break;
+	case CM_SIDR_REP_ATTR_ID:
+		event = IB_CM_SIDR_REP_RECEIVED;
+		break;
+	case CM_LAP_ATTR_ID:
+		paths = 1;
+		event = IB_CM_LAP_RECEIVED;
+		break;
+	case CM_APR_ATTR_ID:
+		event = IB_CM_APR_RECEIVED;
+		break;
+	default:
+		ib_free_recv_mad(mad_recv_wc);
+		return;
+	}
+
+	work = kmalloc(sizeof *work + sizeof(struct ib_sa_path_rec) * paths,
+		       GFP_KERNEL);
+	if (!work) {
+		ib_free_recv_mad(mad_recv_wc);
+		return;
+	}
+
+	INIT_WORK(&work->work, cm_work_handler, work);
+	work->cm_event.event = event;
+	work->mad_recv_wc = mad_recv_wc;
+	work->port = (struct cm_port *)mad_agent->context;
+	queue_work(cm.wq, &work->work);
+}
+
+static int cm_init_qp_init_attr(struct cm_id_private *cm_id_priv,
+				struct ib_qp_attr *qp_attr,
+				int *qp_attr_mask)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REQ_SENT:
+	case IB_CM_MRA_REQ_RCVD:
+	case IB_CM_REQ_RCVD:
+	case IB_CM_MRA_REQ_SENT:
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+	case IB_CM_ESTABLISHED:
+		*qp_attr_mask = IB_QP_STATE | IB_QP_ACCESS_FLAGS |
+				IB_QP_PKEY_INDEX | IB_QP_PORT;
+		qp_attr->qp_access_flags = IB_ACCESS_LOCAL_WRITE;
+		if (cm_id_priv->responder_resources)
+			qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_WRITE |
+						    IB_ACCESS_REMOTE_READ;
+		qp_attr->pkey_index = cm_id_priv->av.pkey_index;
+		qp_attr->port_num = cm_id_priv->av.port->port_num;
+		ret = 0;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+
+static int cm_init_qp_rtr_attr(struct cm_id_private *cm_id_priv,
+			       struct ib_qp_attr *qp_attr,
+			       int *qp_attr_mask)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REQ_RCVD:
+	case IB_CM_MRA_REQ_SENT:
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+	case IB_CM_ESTABLISHED:
+		*qp_attr_mask = IB_QP_STATE | 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;
+		qp_attr->ah_attr = cm_id_priv->av.ah_attr;
+		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);
+		qp_attr->max_dest_rd_atomic = cm_id_priv->responder_resources;
+		qp_attr->min_rnr_timer = 0;
+		if (cm_id_priv->alt_av.ah_attr.dlid) {
+			*qp_attr_mask |= IB_QP_ALT_PATH;
+			qp_attr->alt_ah_attr = cm_id_priv->alt_av.ah_attr;
+		}
+		ret = 0;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+
+static int cm_init_qp_rts_attr(struct cm_id_private *cm_id_priv,
+			       struct ib_qp_attr *qp_attr,
+			       int *qp_attr_mask)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&cm_id_priv->lock, flags);
+	switch (cm_id_priv->id.state) {
+	case IB_CM_REP_RCVD:
+	case IB_CM_MRA_REP_SENT:
+	case IB_CM_REP_SENT:
+	case IB_CM_MRA_REP_RCVD:
+	case IB_CM_ESTABLISHED:
+		*qp_attr_mask = IB_QP_STATE | IB_QP_TIMEOUT | IB_QP_RETRY_CNT |
+				IB_QP_RNR_RETRY | IB_QP_SQ_PSN |
+				IB_QP_MAX_QP_RD_ATOMIC;
+		qp_attr->timeout = cm_id_priv->local_ack_timeout;
+		qp_attr->retry_cnt = cm_id_priv->retry_count;
+		qp_attr->rnr_retry = cm_id_priv->rnr_retry_count;
+		qp_attr->sq_psn = be32_to_cpu(cm_id_priv->sq_psn);
+		qp_attr->max_rd_atomic = cm_id_priv->initiator_depth;
+		if (cm_id_priv->alt_av.ah_attr.dlid) {
+			*qp_attr_mask |= IB_QP_PATH_MIG_STATE;
+			qp_attr->path_mig_state = IB_MIG_REARM;
+		}
+		ret = 0;
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	spin_unlock_irqrestore(&cm_id_priv->lock, flags);
+	return ret;
+}
+
+int ib_cm_init_qp_attr(struct ib_cm_id *cm_id,
+		       struct ib_qp_attr *qp_attr,
+		       int *qp_attr_mask)
+{
+	struct cm_id_private *cm_id_priv;
+	int ret;
+
+	cm_id_priv = container_of(cm_id, struct cm_id_private, id);
+	switch (qp_attr->qp_state) {
+	case IB_QPS_INIT:
+		ret = cm_init_qp_init_attr(cm_id_priv, qp_attr, qp_attr_mask);
+		break;
+	case IB_QPS_RTR:
+		ret = cm_init_qp_rtr_attr(cm_id_priv, qp_attr, qp_attr_mask);
+		break;
+	case IB_QPS_RTS:
+		ret = cm_init_qp_rts_attr(cm_id_priv, qp_attr, qp_attr_mask);
+		break;
+	default:
+		ret = -EINVAL;
+		break;
+	}
+	return ret;
+}
+EXPORT_SYMBOL(ib_cm_init_qp_attr);
+
+static u64 cm_get_ca_guid(struct ib_device *device)
+{
+	struct ib_device_attr *device_attr;
+	u64 guid;
+	int ret;
+
+	device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
+	if (!device_attr)
+		return 0;
+
+	ret = ib_query_device(device, device_attr);
+	guid = ret ? 0 : device_attr->node_guid;
+	kfree(device_attr);
+	return guid;
+}
+
+static void cm_add_one(struct ib_device *device)
+{
+	struct cm_device *cm_dev;
+	struct cm_port *port;
+	struct ib_mad_reg_req reg_req = {
+		.mgmt_class = IB_MGMT_CLASS_CM,
+		.mgmt_class_version = IB_CM_CLASS_VERSION
+	};
+	struct ib_port_modify port_modify = {
+		.set_port_cap_mask = IB_PORT_CM_SUP
+	};
+	unsigned long flags;
+	int ret;
+	u8 i;
+
+	cm_dev = kmalloc(sizeof(*cm_dev) + sizeof(*port) *
+			 device->phys_port_cnt, GFP_KERNEL);
+	if (!cm_dev)
+		return;
+
+	cm_dev->device = device;
+	cm_dev->ca_guid = cm_get_ca_guid(device);
+	if (!cm_dev->ca_guid)
+		goto error1;
+
+	set_bit(IB_MGMT_METHOD_SEND, reg_req.method_mask);
+	for (i = 1; i <= device->phys_port_cnt; i++) {
+		port = &cm_dev->port[i-1];
+		port->cm_dev = cm_dev;
+		port->port_num = i;
+		port->mad_agent = ib_register_mad_agent(device, i,
+							IB_QPT_GSI,
+							&reg_req,
+							0,
+							cm_send_handler,
+							cm_recv_handler,
+							port);
+		if (IS_ERR(port->mad_agent))
+			goto error2;
+
+		ret = ib_modify_port(device, i, 0, &port_modify);
+		if (ret)
+			goto error3;
+	}
+	ib_set_client_data(device, &cm_client, cm_dev);
+
+	write_lock_irqsave(&cm.device_lock, flags);
+	list_add_tail(&cm_dev->list, &cm.device_list);
+	write_unlock_irqrestore(&cm.device_lock, flags);
+	return;
+
+error3:
+	ib_unregister_mad_agent(port->mad_agent);
+error2:
+	port_modify.set_port_cap_mask = 0;
+	port_modify.clr_port_cap_mask = IB_PORT_CM_SUP;
+	while (--i) {
+		port = &cm_dev->port[i-1];
+		ib_modify_port(device, port->port_num, 0, &port_modify);
+		ib_unregister_mad_agent(port->mad_agent);
+	}
+error1:
+	kfree(cm_dev);
+}
+
+static void cm_remove_one(struct ib_device *device)
+{
+	struct cm_device *cm_dev;
+	struct cm_port *port;
+	struct ib_port_modify port_modify = {
+		.clr_port_cap_mask = IB_PORT_CM_SUP
+	};
+	unsigned long flags;
+	int i;
+
+	cm_dev = ib_get_client_data(device, &cm_client);
+	if (!cm_dev)
+		return;
+
+	write_lock_irqsave(&cm.device_lock, flags);
+	list_del(&cm_dev->list);
+	write_unlock_irqrestore(&cm.device_lock, flags);
+
+	for (i = 1; i <= device->phys_port_cnt; i++) {
+		port = &cm_dev->port[i-1];
+		ib_modify_port(device, port->port_num, 0, &port_modify);
+		ib_unregister_mad_agent(port->mad_agent);
+	}
+	kfree(cm_dev);
+}
+
+static int __init ib_cm_init(void)
+{
+	int ret;
+
+	memset(&cm, 0, sizeof cm);
+	INIT_LIST_HEAD(&cm.device_list);
+	rwlock_init(&cm.device_lock);
+	spin_lock_init(&cm.lock);
+	cm.listen_service_table = RB_ROOT;
+	cm.listen_service_id = __constant_be64_to_cpu(IB_CM_ASSIGN_SERVICE_ID);
+	cm.remote_id_table = RB_ROOT;
+	cm.remote_qp_table = RB_ROOT;
+	cm.remote_sidr_table = RB_ROOT;
+	idr_init(&cm.local_id_table);
+	idr_pre_get(&cm.local_id_table, GFP_KERNEL);
+
+	cm.wq = create_workqueue("ib_cm");
+	if (!cm.wq)
+		return -ENOMEM;
+
+	ret = ib_register_client(&cm_client);
+	if (ret)
+		goto error;
+
+	return 0;
+error:
+	destroy_workqueue(cm.wq);
+	return ret;
+}
+
+static void __exit ib_cm_cleanup(void)
+{
+	flush_workqueue(cm.wq);
+	destroy_workqueue(cm.wq);
+	ib_unregister_client(&cm_client);
+}
+
+module_init(ib_cm_init);
+module_exit(ib_cm_cleanup);
+
diff --git a/drivers/infiniband/core/cm_msgs.h b/drivers/infiniband/core/cm_msgs.h
new file mode 100644
index 000000000000..15a309a77b2b
--- /dev/null
+++ b/drivers/infiniband/core/cm_msgs.h
@@ -0,0 +1,819 @@
+/*
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  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 the madirectory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use source and binary forms, with or
+ *     withmodification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retathe above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHWARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS THE
+ * SOFTWARE.
+ */
+#if !defined(CM_MSGS_H)
+#define CM_MSGS_H
+
+#include <ib_mad.h>
+
+/*
+ * Parameters to routines below should be in network-byte order, and values
+ * are returned in network-byte order.
+ */
+
+#define IB_CM_CLASS_VERSION	2 /* IB specification 1.2 */
+
+enum cm_msg_attr_id {
+	CM_REQ_ATTR_ID	    = __constant_htons(0x0010),
+	CM_MRA_ATTR_ID	    = __constant_htons(0x0011),
+	CM_REJ_ATTR_ID	    = __constant_htons(0x0012),
+	CM_REP_ATTR_ID	    = __constant_htons(0x0013),
+	CM_RTU_ATTR_ID	    = __constant_htons(0x0014),
+	CM_DREQ_ATTR_ID	    = __constant_htons(0x0015),
+	CM_DREP_ATTR_ID	    = __constant_htons(0x0016),
+	CM_SIDR_REQ_ATTR_ID = __constant_htons(0x0017),
+	CM_SIDR_REP_ATTR_ID = __constant_htons(0x0018),
+	CM_LAP_ATTR_ID      = __constant_htons(0x0019),
+	CM_APR_ATTR_ID      = __constant_htons(0x001A)
+};
+
+enum cm_msg_sequence {
+	CM_MSG_SEQUENCE_REQ,
+	CM_MSG_SEQUENCE_LAP,
+	CM_MSG_SEQUENCE_DREQ,
+	CM_MSG_SEQUENCE_SIDR
+};
+
+struct cm_req_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 rsvd4;
+	u64 service_id;
+	u64 local_ca_guid;
+	u32 rsvd24;
+	u32 local_qkey;
+	/* local QPN:24, responder resources:8 */
+	u32 offset32;
+	/* local EECN:24, initiator depth:8 */
+	u32 offset36;
+	/*
+	 * remote EECN:24, remote CM response timeout:5,
+	 * transport service type:2, end-to-end flow control:1
+	 */
+	u32 offset40;
+	/* starting PSN:24, local CM response timeout:5, retry count:3 */
+	u32 offset44;
+	u16 pkey;
+	/* path MTU:4, RDC exists:1, RNR retry count:3. */
+	u8 offset50;
+	/* max CM Retries:4, SRQ:1, rsvd:3 */
+	u8 offset51;
+
+	u16 primary_local_lid;
+	u16 primary_remote_lid;
+	union ib_gid primary_local_gid;
+	union ib_gid primary_remote_gid;
+	/* flow label:20, rsvd:6, packet rate:6 */
+	u32 primary_offset88;
+	u8 primary_traffic_class;
+	u8 primary_hop_limit;
+	/* SL:4, subnet local:1, rsvd:3 */
+	u8 primary_offset94;
+	/* local ACK timeout:5, rsvd:3 */
+	u8 primary_offset95;
+
+	u16 alt_local_lid;
+	u16 alt_remote_lid;
+	union ib_gid alt_local_gid;
+	union ib_gid alt_remote_gid;
+	/* flow label:20, rsvd:6, packet rate:6 */
+	u32 alt_offset132;
+	u8 alt_traffic_class;
+	u8 alt_hop_limit;
+	/* SL:4, subnet local:1, rsvd:3 */
+	u8 alt_offset138;
+	/* local ACK timeout:5, rsvd:3 */
+	u8 alt_offset139;
+
+	u8 private_data[IB_CM_REQ_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+static inline u32 cm_req_get_local_qpn(struct cm_req_msg *req_msg)
+{
+	return cpu_to_be32(be32_to_cpu(req_msg->offset32) >> 8);
+}
+
+static inline void cm_req_set_local_qpn(struct cm_req_msg *req_msg, u32 qpn)
+{
+	req_msg->offset32 = cpu_to_be32((be32_to_cpu(qpn) << 8) |
+					 (be32_to_cpu(req_msg->offset32) &
+					  0x000000FF));
+}
+
+static inline u8 cm_req_get_resp_res(struct cm_req_msg *req_msg)
+{
+	return (u8) be32_to_cpu(req_msg->offset32);
+}
+
+static inline void cm_req_set_resp_res(struct cm_req_msg *req_msg, u8 resp_res)
+{
+	req_msg->offset32 = cpu_to_be32(resp_res |
+					(be32_to_cpu(req_msg->offset32) &
+					 0xFFFFFF00));
+}
+
+static inline u8 cm_req_get_init_depth(struct cm_req_msg *req_msg)
+{
+	return (u8) be32_to_cpu(req_msg->offset36);
+}
+
+static inline void cm_req_set_init_depth(struct cm_req_msg *req_msg,
+					 u8 init_depth)
+{
+	req_msg->offset36 = cpu_to_be32(init_depth |
+					(be32_to_cpu(req_msg->offset36) &
+					 0xFFFFFF00));
+}
+
+static inline u8 cm_req_get_remote_resp_timeout(struct cm_req_msg *req_msg)
+{
+	return (u8) ((be32_to_cpu(req_msg->offset40) & 0xF8) >> 3);
+}
+
+static inline void cm_req_set_remote_resp_timeout(struct cm_req_msg *req_msg,
+						  u8 resp_timeout)
+{
+	req_msg->offset40 = cpu_to_be32((resp_timeout << 3) |
+					 (be32_to_cpu(req_msg->offset40) &
+					  0xFFFFFF07));
+}
+
+static inline enum ib_qp_type cm_req_get_qp_type(struct cm_req_msg *req_msg)
+{
+	u8 transport_type = (u8) (be32_to_cpu(req_msg->offset40) & 0x06) >> 1;
+	switch(transport_type) {
+	case 0: return IB_QPT_RC;
+	case 1: return IB_QPT_UC;
+	default: return 0;
+	}
+}
+
+static inline void cm_req_set_qp_type(struct cm_req_msg *req_msg,
+				      enum ib_qp_type qp_type)
+{
+	switch(qp_type) {
+	case IB_QPT_UC:
+		req_msg->offset40 = cpu_to_be32((be32_to_cpu(
+						  req_msg->offset40) &
+						   0xFFFFFFF9) | 0x2);
+	default:
+		req_msg->offset40 = cpu_to_be32(be32_to_cpu(
+						 req_msg->offset40) &
+						  0xFFFFFFF9);
+	}
+}
+
+static inline u8 cm_req_get_flow_ctrl(struct cm_req_msg *req_msg)
+{
+	return be32_to_cpu(req_msg->offset40) & 0x1;
+}
+
+static inline void cm_req_set_flow_ctrl(struct cm_req_msg *req_msg,
+					u8 flow_ctrl)
+{
+	req_msg->offset40 = cpu_to_be32((flow_ctrl & 0x1) |
+					 (be32_to_cpu(req_msg->offset40) &
+					  0xFFFFFFFE));
+}
+
+static inline u32 cm_req_get_starting_psn(struct cm_req_msg *req_msg)
+{
+	return cpu_to_be32(be32_to_cpu(req_msg->offset44) >> 8);
+}
+
+static inline void cm_req_set_starting_psn(struct cm_req_msg *req_msg,
+					   u32 starting_psn)
+{
+	req_msg->offset44 = cpu_to_be32((be32_to_cpu(starting_psn) << 8) |
+			    (be32_to_cpu(req_msg->offset44) & 0x000000FF));
+}
+
+static inline u8 cm_req_get_local_resp_timeout(struct cm_req_msg *req_msg)
+{
+	return (u8) ((be32_to_cpu(req_msg->offset44) & 0xF8) >> 3);
+}
+
+static inline void cm_req_set_local_resp_timeout(struct cm_req_msg *req_msg,
+						 u8 resp_timeout)
+{
+	req_msg->offset44 = cpu_to_be32((resp_timeout << 3) |
+			    (be32_to_cpu(req_msg->offset44) & 0xFFFFFF07));
+}
+
+static inline u8 cm_req_get_retry_count(struct cm_req_msg *req_msg)
+{
+	return (u8) (be32_to_cpu(req_msg->offset44) & 0x7);
+}
+
+static inline void cm_req_set_retry_count(struct cm_req_msg *req_msg,
+					  u8 retry_count)
+{
+	req_msg->offset44 = cpu_to_be32((retry_count & 0x7) |
+			    (be32_to_cpu(req_msg->offset44) & 0xFFFFFFF8));
+}
+
+static inline u8 cm_req_get_path_mtu(struct cm_req_msg *req_msg)
+{
+	return req_msg->offset50 >> 4;
+}
+
+static inline void cm_req_set_path_mtu(struct cm_req_msg *req_msg, u8 path_mtu)
+{
+	req_msg->offset50 = (u8) ((req_msg->offset50 & 0xF) | (path_mtu << 4));
+}
+
+static inline u8 cm_req_get_rnr_retry_count(struct cm_req_msg *req_msg)
+{
+	return req_msg->offset50 & 0x7;
+}
+
+static inline void cm_req_set_rnr_retry_count(struct cm_req_msg *req_msg,
+					      u8 rnr_retry_count)
+{
+	req_msg->offset50 = (u8) ((req_msg->offset50 & 0xF8) |
+				  (rnr_retry_count & 0x7));
+}
+
+static inline u8 cm_req_get_max_cm_retries(struct cm_req_msg *req_msg)
+{
+	return req_msg->offset51 >> 4;
+}
+
+static inline void cm_req_set_max_cm_retries(struct cm_req_msg *req_msg,
+					     u8 retries)
+{
+	req_msg->offset51 = (u8) ((req_msg->offset51 & 0xF) | (retries << 4));
+}
+
+static inline u8 cm_req_get_srq(struct cm_req_msg *req_msg)
+{
+	return (req_msg->offset51 & 0x8) >> 3;
+}
+
+static inline void cm_req_set_srq(struct cm_req_msg *req_msg, u8 srq)
+{
+	req_msg->offset51 = (u8) ((req_msg->offset51 & 0xF7) |
+				  ((srq & 0x1) << 3));
+}
+
+static inline u32 cm_req_get_primary_flow_label(struct cm_req_msg *req_msg)
+{
+	return cpu_to_be32((be32_to_cpu(req_msg->primary_offset88) >> 12));
+}
+
+static inline void cm_req_set_primary_flow_label(struct cm_req_msg *req_msg,
+						 u32 flow_label)
+{
+	req_msg->primary_offset88 = cpu_to_be32(
+				    (be32_to_cpu(req_msg->primary_offset88) &
+				     0x00000FFF) |
+				     (be32_to_cpu(flow_label) << 12));
+}
+
+static inline u8 cm_req_get_primary_packet_rate(struct cm_req_msg *req_msg)
+{
+	return (u8) (be32_to_cpu(req_msg->primary_offset88) & 0x3F);
+}
+
+static inline void cm_req_set_primary_packet_rate(struct cm_req_msg *req_msg,
+						  u8 rate)
+{
+	req_msg->primary_offset88 = cpu_to_be32(
+				    (be32_to_cpu(req_msg->primary_offset88) &
+				     0xFFFFFFC0) | (rate & 0x3F));
+}
+
+static inline u8 cm_req_get_primary_sl(struct cm_req_msg *req_msg)
+{
+	return (u8) (req_msg->primary_offset94 >> 4);
+}
+
+static inline void cm_req_set_primary_sl(struct cm_req_msg *req_msg, u8 sl)
+{
+	req_msg->primary_offset94 = (u8) ((req_msg->primary_offset94 & 0x0F) |
+					  (sl << 4));
+}
+
+static inline u8 cm_req_get_primary_subnet_local(struct cm_req_msg *req_msg)
+{
+	return (u8) ((req_msg->primary_offset94 & 0x08) >> 3);
+}
+
+static inline void cm_req_set_primary_subnet_local(struct cm_req_msg *req_msg,
+						   u8 subnet_local)
+{
+	req_msg->primary_offset94 = (u8) ((req_msg->primary_offset94 & 0xF7) |
+					  ((subnet_local & 0x1) << 3));
+}
+
+static inline u8 cm_req_get_primary_local_ack_timeout(struct cm_req_msg *req_msg)
+{
+	return (u8) (req_msg->primary_offset95 >> 3);
+}
+
+static inline void cm_req_set_primary_local_ack_timeout(struct cm_req_msg *req_msg,
+							u8 local_ack_timeout)
+{
+	req_msg->primary_offset95 = (u8) ((req_msg->primary_offset95 & 0x07) |
+					  (local_ack_timeout << 3));
+}
+
+static inline u32 cm_req_get_alt_flow_label(struct cm_req_msg *req_msg)
+{
+	return cpu_to_be32((be32_to_cpu(req_msg->alt_offset132) >> 12));
+}
+
+static inline void cm_req_set_alt_flow_label(struct cm_req_msg *req_msg,
+					     u32 flow_label)
+{
+	req_msg->alt_offset132 = cpu_to_be32(
+				 (be32_to_cpu(req_msg->alt_offset132) &
+				  0x00000FFF) |
+				  (be32_to_cpu(flow_label) << 12));
+}
+
+static inline u8 cm_req_get_alt_packet_rate(struct cm_req_msg *req_msg)
+{
+	return (u8) (be32_to_cpu(req_msg->alt_offset132) & 0x3F);
+}
+
+static inline void cm_req_set_alt_packet_rate(struct cm_req_msg *req_msg,
+					      u8 rate)
+{
+	req_msg->alt_offset132 = cpu_to_be32(
+				 (be32_to_cpu(req_msg->alt_offset132) &
+				  0xFFFFFFC0) | (rate & 0x3F));
+}
+
+static inline u8 cm_req_get_alt_sl(struct cm_req_msg *req_msg)
+{
+	return (u8) (req_msg->alt_offset138 >> 4);
+}
+
+static inline void cm_req_set_alt_sl(struct cm_req_msg *req_msg, u8 sl)
+{
+	req_msg->alt_offset138 = (u8) ((req_msg->alt_offset138 & 0x0F) |
+				       (sl << 4));
+}
+
+static inline u8 cm_req_get_alt_subnet_local(struct cm_req_msg *req_msg)
+{
+	return (u8) ((req_msg->alt_offset138 & 0x08) >> 3);
+}
+
+static inline void cm_req_set_alt_subnet_local(struct cm_req_msg *req_msg,
+					       u8 subnet_local)
+{
+	req_msg->alt_offset138 = (u8) ((req_msg->alt_offset138 & 0xF7) |
+				       ((subnet_local & 0x1) << 3));
+}
+
+static inline u8 cm_req_get_alt_local_ack_timeout(struct cm_req_msg *req_msg)
+{
+	return (u8) (req_msg->alt_offset139 >> 3);
+}
+
+static inline void cm_req_set_alt_local_ack_timeout(struct cm_req_msg *req_msg,
+						    u8 local_ack_timeout)
+{
+	req_msg->alt_offset139 = (u8) ((req_msg->alt_offset139 & 0x07) |
+				       (local_ack_timeout << 3));
+}
+
+/* Message REJected or MRAed */
+enum cm_msg_response {
+	CM_MSG_RESPONSE_REQ = 0x0,
+	CM_MSG_RESPONSE_REP = 0x1,
+	CM_MSG_RESPONSE_OTHER = 0x2
+};
+
+ struct cm_mra_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+	/* message MRAed:2, rsvd:6 */
+	u8 offset8;
+	/* service timeout:5, rsvd:3 */
+	u8 offset9;
+
+	u8 private_data[IB_CM_MRA_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+static inline u8 cm_mra_get_msg_mraed(struct cm_mra_msg *mra_msg)
+{
+	return (u8) (mra_msg->offset8 >> 6);
+}
+
+static inline void cm_mra_set_msg_mraed(struct cm_mra_msg *mra_msg, u8 msg)
+{
+	mra_msg->offset8 = (u8) ((mra_msg->offset8 & 0x3F) | (msg << 6));
+}
+
+static inline u8 cm_mra_get_service_timeout(struct cm_mra_msg *mra_msg)
+{
+	return (u8) (mra_msg->offset9 >> 3);
+}
+
+static inline void cm_mra_set_service_timeout(struct cm_mra_msg *mra_msg,
+					      u8 service_timeout)
+{
+	mra_msg->offset9 = (u8) ((mra_msg->offset9 & 0x07) |
+				 (service_timeout << 3));
+}
+
+struct cm_rej_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+	/* message REJected:2, rsvd:6 */
+	u8 offset8;
+	/* reject info length:7, rsvd:1. */
+	u8 offset9;
+	u16 reason;
+	u8 ari[IB_CM_REJ_ARI_LENGTH];
+
+	u8 private_data[IB_CM_REJ_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+static inline u8 cm_rej_get_msg_rejected(struct cm_rej_msg *rej_msg)
+{
+	return (u8) (rej_msg->offset8 >> 6);
+}
+
+static inline void cm_rej_set_msg_rejected(struct cm_rej_msg *rej_msg, u8 msg)
+{
+	rej_msg->offset8 = (u8) ((rej_msg->offset8 & 0x3F) | (msg << 6));
+}
+
+static inline u8 cm_rej_get_reject_info_len(struct cm_rej_msg *rej_msg)
+{
+	return (u8) (rej_msg->offset9 >> 1);
+}
+
+static inline void cm_rej_set_reject_info_len(struct cm_rej_msg *rej_msg,
+					      u8 len)
+{
+	rej_msg->offset9 = (u8) ((rej_msg->offset9 & 0x1) | (len << 1));
+}
+
+struct cm_rep_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+	u32 local_qkey;
+	/* local QPN:24, rsvd:8 */
+	u32 offset12;
+	/* local EECN:24, rsvd:8 */
+	u32 offset16;
+	/* starting PSN:24 rsvd:8 */
+	u32 offset20;
+	u8 resp_resources;
+	u8 initiator_depth;
+	/* target ACK delay:5, failover accepted:2, end-to-end flow control:1 */
+	u8 offset26;
+	/* RNR retry count:3, SRQ:1, rsvd:5 */
+	u8 offset27;
+	u64 local_ca_guid;
+
+	u8 private_data[IB_CM_REP_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+static inline u32 cm_rep_get_local_qpn(struct cm_rep_msg *rep_msg)
+{
+	return cpu_to_be32(be32_to_cpu(rep_msg->offset12) >> 8);
+}
+
+static inline void cm_rep_set_local_qpn(struct cm_rep_msg *rep_msg, u32 qpn)
+{
+	rep_msg->offset12 = cpu_to_be32((be32_to_cpu(qpn) << 8) |
+			    (be32_to_cpu(rep_msg->offset12) & 0x000000FF));
+}
+
+static inline u32 cm_rep_get_starting_psn(struct cm_rep_msg *rep_msg)
+{
+	return cpu_to_be32(be32_to_cpu(rep_msg->offset20) >> 8);
+}
+
+static inline void cm_rep_set_starting_psn(struct cm_rep_msg *rep_msg,
+					   u32 starting_psn)
+{
+	rep_msg->offset20 = cpu_to_be32((be32_to_cpu(starting_psn) << 8) |
+			    (be32_to_cpu(rep_msg->offset20) & 0x000000FF));
+}
+
+static inline u8 cm_rep_get_target_ack_delay(struct cm_rep_msg *rep_msg)
+{
+	return (u8) (rep_msg->offset26 >> 3);
+}
+
+static inline void cm_rep_set_target_ack_delay(struct cm_rep_msg *rep_msg,
+					       u8 target_ack_delay)
+{
+	rep_msg->offset26 = (u8) ((rep_msg->offset26 & 0x07) |
+				  (target_ack_delay << 3));
+}
+
+static inline u8 cm_rep_get_failover(struct cm_rep_msg *rep_msg)
+{
+	return (u8) ((rep_msg->offset26 & 0x06) >> 1);
+}
+
+static inline void cm_rep_set_failover(struct cm_rep_msg *rep_msg, u8 failover)
+{
+	rep_msg->offset26 = (u8) ((rep_msg->offset26 & 0xF9) |
+				  ((failover & 0x3) << 1));
+}
+
+static inline u8 cm_rep_get_flow_ctrl(struct cm_rep_msg *rep_msg)
+{
+	return (u8) (rep_msg->offset26 & 0x01);
+}
+
+static inline void cm_rep_set_flow_ctrl(struct cm_rep_msg *rep_msg,
+					    u8 flow_ctrl)
+{
+	rep_msg->offset26 = (u8) ((rep_msg->offset26 & 0xFE) |
+				  (flow_ctrl & 0x1));
+}
+
+static inline u8 cm_rep_get_rnr_retry_count(struct cm_rep_msg *rep_msg)
+{
+	return (u8) (rep_msg->offset27 >> 5);
+}
+
+static inline void cm_rep_set_rnr_retry_count(struct cm_rep_msg *rep_msg,
+					      u8 rnr_retry_count)
+{
+	rep_msg->offset27 = (u8) ((rep_msg->offset27 & 0x1F) |
+				  (rnr_retry_count << 5));
+}
+
+static inline u8 cm_rep_get_srq(struct cm_rep_msg *rep_msg)
+{
+	return (u8) ((rep_msg->offset27 >> 4) & 0x1);
+}
+
+static inline void cm_rep_set_srq(struct cm_rep_msg *rep_msg, u8 srq)
+{
+	rep_msg->offset27 = (u8) ((rep_msg->offset27 & 0xEF) |
+				  ((srq & 0x1) << 4));
+}
+
+struct cm_rtu_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+
+	u8 private_data[IB_CM_RTU_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+struct cm_dreq_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+	/* remote QPN/EECN:24, rsvd:8 */
+	u32 offset8;
+
+	u8 private_data[IB_CM_DREQ_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+static inline u32 cm_dreq_get_remote_qpn(struct cm_dreq_msg *dreq_msg)
+{
+	return cpu_to_be32(be32_to_cpu(dreq_msg->offset8) >> 8);
+}
+
+static inline void cm_dreq_set_remote_qpn(struct cm_dreq_msg *dreq_msg, u32 qpn)
+{
+	dreq_msg->offset8 = cpu_to_be32((be32_to_cpu(qpn) << 8) |
+			    (be32_to_cpu(dreq_msg->offset8) & 0x000000FF));
+}
+
+struct cm_drep_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+
+	u8 private_data[IB_CM_DREP_PRIVATE_DATA_SIZE];
+
+} __attribute__ ((packed));
+
+struct cm_lap_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+
+	u32 rsvd8;
+	/* remote QPN/EECN:24, remote CM response timeout:5, rsvd:3 */
+	u32 offset12;
+	u32 rsvd16;
+
+	u16 alt_local_lid;
+	u16 alt_remote_lid;
+	union ib_gid alt_local_gid;
+	union ib_gid alt_remote_gid;
+	/* flow label:20, rsvd:4, traffic class:8 */
+	u32 offset56;
+	u8 alt_hop_limit;
+	/* rsvd:2, packet rate:6 */
+	uint8_t offset61;
+	/* SL:4, subnet local:1, rsvd:3 */
+	uint8_t offset62;
+	/* local ACK timeout:5, rsvd:3 */
+	uint8_t offset63;
+
+	u8 private_data[IB_CM_LAP_PRIVATE_DATA_SIZE];
+} __attribute__  ((packed));
+
+static inline u32 cm_lap_get_remote_qpn(struct cm_lap_msg *lap_msg)
+{
+	return cpu_to_be32(be32_to_cpu(lap_msg->offset12) >> 8);
+}
+
+static inline void cm_lap_set_remote_qpn(struct cm_lap_msg *lap_msg, u32 qpn)
+{
+	lap_msg->offset12 = cpu_to_be32((be32_to_cpu(qpn) << 8) |
+					 (be32_to_cpu(lap_msg->offset12) &
+					  0x000000FF));
+}
+
+static inline u8 cm_lap_get_remote_resp_timeout(struct cm_lap_msg *lap_msg)
+{
+	return (u8) ((be32_to_cpu(lap_msg->offset12) & 0xF8) >> 3);
+}
+
+static inline void cm_lap_set_remote_resp_timeout(struct cm_lap_msg *lap_msg,
+						  u8 resp_timeout)
+{
+	lap_msg->offset12 = cpu_to_be32((resp_timeout << 3) |
+					 (be32_to_cpu(lap_msg->offset12) &
+					  0xFFFFFF07));
+}
+
+static inline u32 cm_lap_get_flow_label(struct cm_lap_msg *lap_msg)
+{
+	return be32_to_cpu(lap_msg->offset56) >> 12;
+}
+
+static inline void cm_lap_set_flow_label(struct cm_lap_msg *lap_msg,
+					 u32 flow_label)
+{
+	lap_msg->offset56 = cpu_to_be32((flow_label << 12) |
+					 (be32_to_cpu(lap_msg->offset56) &
+					  0x00000FFF));
+}
+
+static inline u8 cm_lap_get_traffic_class(struct cm_lap_msg *lap_msg)
+{
+	return (u8) be32_to_cpu(lap_msg->offset56);
+}
+
+static inline void cm_lap_set_traffic_class(struct cm_lap_msg *lap_msg,
+					    u8 traffic_class)
+{
+	lap_msg->offset56 = cpu_to_be32(traffic_class |
+					 (be32_to_cpu(lap_msg->offset56) &
+					  0xFFFFFF00));
+}
+
+static inline u8 cm_lap_get_packet_rate(struct cm_lap_msg *lap_msg)
+{
+	return lap_msg->offset61 & 0x3F;
+}
+
+static inline void cm_lap_set_packet_rate(struct cm_lap_msg *lap_msg,
+					  u8 packet_rate)
+{
+	lap_msg->offset61 = (packet_rate & 0x3F) | (lap_msg->offset61 & 0xC0);
+}
+
+static inline u8 cm_lap_get_sl(struct cm_lap_msg *lap_msg)
+{
+	return lap_msg->offset62 >> 4;
+}
+
+static inline void cm_lap_set_sl(struct cm_lap_msg *lap_msg, u8 sl)
+{
+	lap_msg->offset62 = (sl << 4) | (lap_msg->offset62 & 0x0F);
+}
+
+static inline u8 cm_lap_get_subnet_local(struct cm_lap_msg *lap_msg)
+{
+	return (lap_msg->offset62 >> 3) & 0x1;
+}
+
+static inline void cm_lap_set_subnet_local(struct cm_lap_msg *lap_msg,
+					   u8 subnet_local)
+{
+	lap_msg->offset62 = ((subnet_local & 0x1) << 3) |
+			     (lap_msg->offset61 & 0xF7);
+}
+static inline u8 cm_lap_get_local_ack_timeout(struct cm_lap_msg *lap_msg)
+{
+	return lap_msg->offset63 >> 3;
+}
+
+static inline void cm_lap_set_local_ack_timeout(struct cm_lap_msg *lap_msg,
+						u8 local_ack_timeout)
+{
+	lap_msg->offset63 = (local_ack_timeout << 3) |
+			    (lap_msg->offset63 & 0x07);
+}
+
+struct cm_apr_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 local_comm_id;
+	u32 remote_comm_id;
+
+	u8 info_length;
+	u8 ap_status;
+	u8 info[IB_CM_APR_INFO_LENGTH];
+
+	u8 private_data[IB_CM_APR_PRIVATE_DATA_SIZE];
+} __attribute__ ((packed));
+
+struct cm_sidr_req_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 request_id;
+	u16 pkey;
+	u16 rsvd;
+	u64 service_id;
+
+	u8 private_data[IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE];
+} __attribute__ ((packed));
+
+struct cm_sidr_rep_msg {
+	struct ib_mad_hdr hdr;
+
+	u32 request_id;
+	u8 status;
+	u8 info_length;
+	u16 rsvd;
+	/* QPN:24, rsvd:8 */
+	u32 offset8;
+	u64 service_id;
+	u32 qkey;
+	u8 info[IB_CM_SIDR_REP_INFO_LENGTH];
+
+	u8 private_data[IB_CM_SIDR_REP_PRIVATE_DATA_SIZE];
+} __attribute__ ((packed));
+
+static inline u32 cm_sidr_rep_get_qpn(struct cm_sidr_rep_msg *sidr_rep_msg)
+{
+	return cpu_to_be32(be32_to_cpu(sidr_rep_msg->offset8) >> 8);
+}
+
+static inline void cm_sidr_rep_set_qpn(struct cm_sidr_rep_msg *sidr_rep_msg,
+				       u32 qpn)
+{
+	sidr_rep_msg->offset8 = cpu_to_be32((be32_to_cpu(qpn) << 8) |
+					(be32_to_cpu(sidr_rep_msg->offset8) &
+					 0x000000FF));
+}
+
+#endif /* CM_MSGS_H */
diff --git a/drivers/infiniband/core/fmr_pool.c b/drivers/infiniband/core/fmr_pool.c
index 328feae2a5be..7763b31abba7 100644
--- a/drivers/infiniband/core/fmr_pool.c
+++ b/drivers/infiniband/core/fmr_pool.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, 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
@@ -29,7 +30,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: fmr_pool.c 1349 2004-12-16 21:09:43Z roland $
+ * $Id: fmr_pool.c 2730 2005-06-28 16:43:03Z sean.hefty $
  */
 
 #include <linux/errno.h>
@@ -329,7 +330,7 @@ EXPORT_SYMBOL(ib_create_fmr_pool);
  *
  * Destroy an FMR pool and free all associated resources.
  */
-int ib_destroy_fmr_pool(struct ib_fmr_pool *pool)
+void ib_destroy_fmr_pool(struct ib_fmr_pool *pool)
 {
 	struct ib_pool_fmr *fmr;
 	struct ib_pool_fmr *tmp;
@@ -352,8 +353,6 @@ int ib_destroy_fmr_pool(struct ib_fmr_pool *pool)
 
 	kfree(pool->cache_bucket);
 	kfree(pool);
-
-	return 0;
 }
 EXPORT_SYMBOL(ib_destroy_fmr_pool);
 
diff --git a/drivers/infiniband/core/mad.c b/drivers/infiniband/core/mad.c
index 23628c622a50..b97e210ce9c8 100644
--- a/drivers/infiniband/core/mad.c
+++ b/drivers/infiniband/core/mad.c
@@ -1,5 +1,7 @@
 /*
  * Copyright (c) 2004, 2005 Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2005 Mellanox Technologies Ltd.  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
@@ -29,12 +31,12 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: mad.c 1389 2004-12-27 22:56:47Z roland $
+ * $Id: mad.c 2817 2005-07-07 11:29:26Z halr $
  */
-
 #include <linux/dma-mapping.h>
 
 #include "mad_priv.h"
+#include "mad_rmpp.h"
 #include "smi.h"
 #include "agent.h"
 
@@ -45,6 +47,7 @@ MODULE_AUTHOR("Sean Hefty");
 
 
 kmem_cache_t *ib_mad_cache;
+
 static struct list_head ib_mad_port_list;
 static u32 ib_mad_client_id = 0;
 
@@ -58,16 +61,12 @@ static int method_in_use(struct ib_mad_mgmt_method_table **method,
 static void remove_mad_reg_req(struct ib_mad_agent_private *priv);
 static struct ib_mad_agent_private *find_mad_agent(
 					struct ib_mad_port_private *port_priv,
-					struct ib_mad *mad, int solicited);
+					struct ib_mad *mad);
 static int ib_mad_post_receive_mads(struct ib_mad_qp_info *qp_info,
 				    struct ib_mad_private *mad);
 static void cancel_mads(struct ib_mad_agent_private *mad_agent_priv);
-static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
-				    struct ib_mad_send_wc *mad_send_wc);
 static void timeout_sends(void *data);
-static void cancel_sends(void *data);
 static void local_completions(void *data);
-static int solicited_mad(struct ib_mad *mad);
 static int add_nonoui_reg_req(struct ib_mad_reg_req *mad_reg_req,
 			      struct ib_mad_agent_private *agent_priv,
 			      u8 mgmt_class);
@@ -197,8 +196,8 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
 	if (qpn == -1)
 		goto error1;
 
-	if (rmpp_version)
-		goto error1;	/* XXX: until RMPP implemented */
+	if (rmpp_version && rmpp_version != IB_MGMT_RMPP_VERSION)
+		goto error1;
 
 	/* Validate MAD registration request if supplied */
 	if (mad_reg_req) {
@@ -261,22 +260,29 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
 		ret = ERR_PTR(-ENOMEM);
 		goto error1;
 	}
+	memset(mad_agent_priv, 0, sizeof *mad_agent_priv);
+
+	mad_agent_priv->agent.mr = ib_get_dma_mr(port_priv->qp_info[qpn].qp->pd,
+						 IB_ACCESS_LOCAL_WRITE);
+	if (IS_ERR(mad_agent_priv->agent.mr)) {
+		ret = ERR_PTR(-ENOMEM);
+		goto error2;
+	}
 
 	if (mad_reg_req) {
 		reg_req = kmalloc(sizeof *reg_req, GFP_KERNEL);
 		if (!reg_req) {
 			ret = ERR_PTR(-ENOMEM);
-			goto error2;
+			goto error3;
 		}
 		/* Make a copy of the MAD registration request */
 		memcpy(reg_req, mad_reg_req, sizeof *reg_req);
 	}
 
 	/* Now, fill in the various structures */
-	memset(mad_agent_priv, 0, sizeof *mad_agent_priv);
 	mad_agent_priv->qp_info = &port_priv->qp_info[qpn];
 	mad_agent_priv->reg_req = reg_req;
-	mad_agent_priv->rmpp_version = rmpp_version;
+	mad_agent_priv->agent.rmpp_version = rmpp_version;
 	mad_agent_priv->agent.device = device;
 	mad_agent_priv->agent.recv_handler = recv_handler;
 	mad_agent_priv->agent.send_handler = send_handler;
@@ -301,7 +307,7 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
 				if (method) {
 					if (method_in_use(&method,
 							   mad_reg_req))
-						goto error3;
+						goto error4;
 				}
 			}
 			ret2 = add_nonoui_reg_req(mad_reg_req, mad_agent_priv,
@@ -317,14 +323,14 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
 					if (is_vendor_method_in_use(
 							vendor_class,
 							mad_reg_req))
-						goto error3;
+						goto error4;
 				}
 			}
 			ret2 = add_oui_reg_req(mad_reg_req, mad_agent_priv);
 		}
 		if (ret2) {
 			ret = ERR_PTR(ret2);
-			goto error3;
+			goto error4;
 		}
 	}
 
@@ -335,22 +341,24 @@ struct ib_mad_agent *ib_register_mad_agent(struct ib_device *device,
 	spin_lock_init(&mad_agent_priv->lock);
 	INIT_LIST_HEAD(&mad_agent_priv->send_list);
 	INIT_LIST_HEAD(&mad_agent_priv->wait_list);
+	INIT_LIST_HEAD(&mad_agent_priv->done_list);
+	INIT_LIST_HEAD(&mad_agent_priv->rmpp_list);
 	INIT_WORK(&mad_agent_priv->timed_work, timeout_sends, mad_agent_priv);
 	INIT_LIST_HEAD(&mad_agent_priv->local_list);
 	INIT_WORK(&mad_agent_priv->local_work, local_completions,
 		   mad_agent_priv);
-	INIT_LIST_HEAD(&mad_agent_priv->canceled_list);
-	INIT_WORK(&mad_agent_priv->canceled_work, cancel_sends, mad_agent_priv);
 	atomic_set(&mad_agent_priv->refcount, 1);
 	init_waitqueue_head(&mad_agent_priv->wait);
 
 	return &mad_agent_priv->agent;
 
-error3:
+error4:
 	spin_unlock_irqrestore(&port_priv->reg_lock, flags);
 	kfree(reg_req);
-error2:
+error3:
 	kfree(mad_agent_priv);
+error2:
+	ib_dereg_mr(mad_agent_priv->agent.mr);
 error1:
 	return ret;
 }
@@ -487,18 +495,16 @@ static void unregister_mad_agent(struct ib_mad_agent_private *mad_agent_priv)
 	 * MADs, preventing us from queuing additional work
 	 */
 	cancel_mads(mad_agent_priv);
-
 	port_priv = mad_agent_priv->qp_info->port_priv;
-
 	cancel_delayed_work(&mad_agent_priv->timed_work);
-	flush_workqueue(port_priv->wq);
 
 	spin_lock_irqsave(&port_priv->reg_lock, flags);
 	remove_mad_reg_req(mad_agent_priv);
 	list_del(&mad_agent_priv->agent_list);
 	spin_unlock_irqrestore(&port_priv->reg_lock, flags);
 
-	/* XXX: Cleanup pending RMPP receives for this agent */
+	flush_workqueue(port_priv->wq);
+	ib_cancel_rmpp_recvs(mad_agent_priv);
 
 	atomic_dec(&mad_agent_priv->refcount);
 	wait_event(mad_agent_priv->wait,
@@ -506,6 +512,7 @@ static void unregister_mad_agent(struct ib_mad_agent_private *mad_agent_priv)
 
 	if (mad_agent_priv->reg_req)
 		kfree(mad_agent_priv->reg_req);
+	ib_dereg_mr(mad_agent_priv->agent.mr);
 	kfree(mad_agent_priv);
 }
 
@@ -551,6 +558,13 @@ int ib_unregister_mad_agent(struct ib_mad_agent *mad_agent)
 }
 EXPORT_SYMBOL(ib_unregister_mad_agent);
 
+static inline int response_mad(struct ib_mad *mad)
+{
+	/* Trap represses are responses although response bit is reset */
+	return ((mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS) ||
+		(mad->mad_hdr.method & IB_MGMT_METHOD_RESP));
+}
+
 static void dequeue_mad(struct ib_mad_list_head *mad_list)
 {
 	struct ib_mad_queue *mad_queue;
@@ -643,7 +657,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 				  struct ib_smp *smp,
 				  struct ib_send_wr *send_wr)
 {
-	int ret, solicited;
+	int ret;
 	unsigned long flags;
 	struct ib_mad_local_private *local;
 	struct ib_mad_private *mad_priv;
@@ -689,11 +703,7 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 	switch (ret)
 	{
 	case IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY:
-		/*
-		 * See if response is solicited and
-		 * there is a recv handler
-		 */
-		if (solicited_mad(&mad_priv->mad.mad) &&
+		if (response_mad(&mad_priv->mad.mad) &&
 		    mad_agent_priv->agent.recv_handler) {
 			local->mad_priv = mad_priv;
 			local->recv_mad_agent = mad_agent_priv;
@@ -710,15 +720,13 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 		break;
 	case IB_MAD_RESULT_SUCCESS:
 		/* Treat like an incoming receive MAD */
-		solicited = solicited_mad(&mad_priv->mad.mad);
 		port_priv = ib_get_mad_port(mad_agent_priv->agent.device,
 					    mad_agent_priv->agent.port_num);
 		if (port_priv) {
 			mad_priv->mad.mad.mad_hdr.tid =
 				((struct ib_mad *)smp)->mad_hdr.tid;
 			recv_mad_agent = find_mad_agent(port_priv,
-						       &mad_priv->mad.mad,
-							solicited);
+						        &mad_priv->mad.mad);
 		}
 		if (!port_priv || !recv_mad_agent) {
 			kmem_cache_free(ib_mad_cache, mad_priv);
@@ -750,43 +758,133 @@ static int handle_outgoing_dr_smp(struct ib_mad_agent_private *mad_agent_priv,
 	list_add_tail(&local->completion_list, &mad_agent_priv->local_list);
 	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 	queue_work(mad_agent_priv->qp_info->port_priv->wq,
-		  &mad_agent_priv->local_work);
+		   &mad_agent_priv->local_work);
 	ret = 1;
 out:
 	return ret;
 }
 
-static int ib_send_mad(struct ib_mad_agent_private *mad_agent_priv,
-		       struct ib_mad_send_wr_private *mad_send_wr)
+static int get_buf_length(int hdr_len, int data_len)
+{
+	int seg_size, pad;
+
+	seg_size = sizeof(struct ib_mad) - hdr_len;
+	if (data_len && seg_size) {
+		pad = seg_size - data_len % seg_size;
+		if (pad == seg_size)
+			pad = 0;
+	} else
+		pad = seg_size;
+	return hdr_len + data_len + pad;
+}
+
+struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
+					    u32 remote_qpn, u16 pkey_index,
+					    struct ib_ah *ah, int rmpp_active,
+					    int hdr_len, int data_len,
+					    unsigned int __nocast gfp_mask)
+{
+	struct ib_mad_agent_private *mad_agent_priv;
+	struct ib_mad_send_buf *send_buf;
+	int buf_size;
+	void *buf;
+
+	mad_agent_priv = container_of(mad_agent,
+				      struct ib_mad_agent_private, agent);
+	buf_size = get_buf_length(hdr_len, data_len);
+
+	if ((!mad_agent->rmpp_version &&
+	     (rmpp_active || buf_size > sizeof(struct ib_mad))) ||
+	    (!rmpp_active && buf_size > sizeof(struct ib_mad)))
+		return ERR_PTR(-EINVAL);
+
+	buf = kmalloc(sizeof *send_buf + buf_size, gfp_mask);
+	if (!buf)
+		return ERR_PTR(-ENOMEM);
+	memset(buf, 0, sizeof *send_buf + buf_size);
+
+	send_buf = buf + buf_size;
+	send_buf->mad = buf;
+
+	send_buf->sge.addr = dma_map_single(mad_agent->device->dma_device,
+					    buf, buf_size, DMA_TO_DEVICE);
+	pci_unmap_addr_set(send_buf, mapping, send_buf->sge.addr);
+	send_buf->sge.length = buf_size;
+	send_buf->sge.lkey = mad_agent->mr->lkey;
+
+	send_buf->send_wr.wr_id = (unsigned long) send_buf;
+	send_buf->send_wr.sg_list = &send_buf->sge;
+	send_buf->send_wr.num_sge = 1;
+	send_buf->send_wr.opcode = IB_WR_SEND;
+	send_buf->send_wr.send_flags = IB_SEND_SIGNALED;
+	send_buf->send_wr.wr.ud.ah = ah;
+	send_buf->send_wr.wr.ud.mad_hdr = &send_buf->mad->mad_hdr;
+	send_buf->send_wr.wr.ud.remote_qpn = remote_qpn;
+	send_buf->send_wr.wr.ud.remote_qkey = IB_QP_SET_QKEY;
+	send_buf->send_wr.wr.ud.pkey_index = pkey_index;
+
+	if (rmpp_active) {
+		struct ib_rmpp_mad *rmpp_mad;
+		rmpp_mad = (struct ib_rmpp_mad *)send_buf->mad;
+		rmpp_mad->rmpp_hdr.paylen_newwin = cpu_to_be32(hdr_len -
+			offsetof(struct ib_rmpp_mad, data) + data_len);
+		rmpp_mad->rmpp_hdr.rmpp_version = mad_agent->rmpp_version;
+		rmpp_mad->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_DATA;
+		ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr,
+				  IB_MGMT_RMPP_FLAG_ACTIVE);
+	}
+
+	send_buf->mad_agent = mad_agent;
+	atomic_inc(&mad_agent_priv->refcount);
+	return send_buf;
+}
+EXPORT_SYMBOL(ib_create_send_mad);
+
+void ib_free_send_mad(struct ib_mad_send_buf *send_buf)
+{
+	struct ib_mad_agent_private *mad_agent_priv;
+
+	mad_agent_priv = container_of(send_buf->mad_agent,
+				      struct ib_mad_agent_private, agent);
+
+	dma_unmap_single(send_buf->mad_agent->device->dma_device,
+			 pci_unmap_addr(send_buf, mapping),
+			 send_buf->sge.length, DMA_TO_DEVICE);
+	kfree(send_buf->mad);
+
+	if (atomic_dec_and_test(&mad_agent_priv->refcount))
+		wake_up(&mad_agent_priv->wait);
+}
+EXPORT_SYMBOL(ib_free_send_mad);
+
+int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr)
 {
 	struct ib_mad_qp_info *qp_info;
 	struct ib_send_wr *bad_send_wr;
+	struct list_head *list;
 	unsigned long flags;
 	int ret;
 
-	/* Replace user's WR ID with our own to find WR upon completion */
-	qp_info = mad_agent_priv->qp_info;
-	mad_send_wr->wr_id = mad_send_wr->send_wr.wr_id;
+	/* Set WR ID to find mad_send_wr upon completion */
+	qp_info = mad_send_wr->mad_agent_priv->qp_info;
 	mad_send_wr->send_wr.wr_id = (unsigned long)&mad_send_wr->mad_list;
 	mad_send_wr->mad_list.mad_queue = &qp_info->send_queue;
 
 	spin_lock_irqsave(&qp_info->send_queue.lock, flags);
-	if (qp_info->send_queue.count++ < qp_info->send_queue.max_active) {
-		list_add_tail(&mad_send_wr->mad_list.list,
-			      &qp_info->send_queue.list);
-		spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
-		ret = ib_post_send(mad_agent_priv->agent.qp,
+	if (qp_info->send_queue.count < qp_info->send_queue.max_active) {
+		ret = ib_post_send(mad_send_wr->mad_agent_priv->agent.qp,
 				   &mad_send_wr->send_wr, &bad_send_wr);
-		if (ret) {
-			printk(KERN_ERR PFX "ib_post_send failed: %d\n", ret);
-			dequeue_mad(&mad_send_wr->mad_list);
-		}
+		list = &qp_info->send_queue.list;
 	} else {
-		list_add_tail(&mad_send_wr->mad_list.list,
-			      &qp_info->overflow_list);
-		spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
 		ret = 0;
+		list = &qp_info->overflow_list;
 	}
+
+	if (!ret) {
+		qp_info->send_queue.count++;
+		list_add_tail(&mad_send_wr->mad_list.list, list);
+	}
+	spin_unlock_irqrestore(&qp_info->send_queue.lock, flags);
 	return ret;
 }
 
@@ -860,18 +958,19 @@ int ib_post_send_mad(struct ib_mad_agent *mad_agent,
 			ret = -ENOMEM;
 			goto error2;
 		}
+		memset(mad_send_wr, 0, sizeof *mad_send_wr);
 
 		mad_send_wr->send_wr = *send_wr;
 		mad_send_wr->send_wr.sg_list = mad_send_wr->sg_list;
 		memcpy(mad_send_wr->sg_list, send_wr->sg_list,
 		       sizeof *send_wr->sg_list * send_wr->num_sge);
-		mad_send_wr->send_wr.next = NULL;
+		mad_send_wr->wr_id = send_wr->wr_id;
 		mad_send_wr->tid = send_wr->wr.ud.mad_hdr->tid;
-		mad_send_wr->agent = mad_agent;
+		mad_send_wr->mad_agent_priv = mad_agent_priv;
 		/* Timeout will be updated after send completes */
 		mad_send_wr->timeout = msecs_to_jiffies(send_wr->wr.
 							ud.timeout_ms);
-		mad_send_wr->retry = 0;
+		mad_send_wr->retries = mad_send_wr->send_wr.wr.ud.retries;
 		/* One reference for each work request to QP + response */
 		mad_send_wr->refcount = 1 + (mad_send_wr->timeout > 0);
 		mad_send_wr->status = IB_WC_SUCCESS;
@@ -883,8 +982,13 @@ int ib_post_send_mad(struct ib_mad_agent *mad_agent,
 			      &mad_agent_priv->send_list);
 		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 
-		ret = ib_send_mad(mad_agent_priv, mad_send_wr);
-		if (ret) {
+		if (mad_agent_priv->agent.rmpp_version) {
+			ret = ib_send_rmpp_mad(mad_send_wr);
+			if (ret >= 0 && ret != IB_RMPP_RESULT_CONSUMED)
+				ret = ib_send_mad(mad_send_wr);
+		} else
+			ret = ib_send_mad(mad_send_wr);
+		if (ret < 0) {
 			/* Fail send request */
 			spin_lock_irqsave(&mad_agent_priv->lock, flags);
 			list_del(&mad_send_wr->agent_list);
@@ -910,41 +1014,28 @@ EXPORT_SYMBOL(ib_post_send_mad);
  */
 void ib_free_recv_mad(struct ib_mad_recv_wc *mad_recv_wc)
 {
-	struct ib_mad_recv_buf *entry;
+	struct ib_mad_recv_buf *mad_recv_buf, *temp_recv_buf;
 	struct ib_mad_private_header *mad_priv_hdr;
 	struct ib_mad_private *priv;
+	struct list_head free_list;
 
-	mad_priv_hdr = container_of(mad_recv_wc,
-				    struct ib_mad_private_header,
-				    recv_wc);
-	priv = container_of(mad_priv_hdr, struct ib_mad_private, header);
+	INIT_LIST_HEAD(&free_list);
+	list_splice_init(&mad_recv_wc->rmpp_list, &free_list);
 
-	/*
-	 * Walk receive buffer list associated with this WC
-	 * No need to remove them from list of receive buffers
-	 */
-	list_for_each_entry(entry, &mad_recv_wc->recv_buf.list, list) {
-		/* Free previous receive buffer */
-		kmem_cache_free(ib_mad_cache, priv);
+	list_for_each_entry_safe(mad_recv_buf, temp_recv_buf,
+					&free_list, list) {
+		mad_recv_wc = container_of(mad_recv_buf, struct ib_mad_recv_wc,
+					   recv_buf);
 		mad_priv_hdr = container_of(mad_recv_wc,
 					    struct ib_mad_private_header,
 					    recv_wc);
 		priv = container_of(mad_priv_hdr, struct ib_mad_private,
 				    header);
+		kmem_cache_free(ib_mad_cache, priv);
 	}
-
-	/* Free last buffer */
-	kmem_cache_free(ib_mad_cache, priv);
 }
 EXPORT_SYMBOL(ib_free_recv_mad);
 
-void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc,
-			  void *buf)
-{
-	printk(KERN_ERR PFX "ib_coalesce_recv_mad() not implemented yet\n");
-}
-EXPORT_SYMBOL(ib_coalesce_recv_mad);
-
 struct ib_mad_agent *ib_redirect_mad_qp(struct ib_qp *qp,
 					u8 rmpp_version,
 					ib_mad_send_handler send_handler,
@@ -1338,42 +1429,15 @@ out:
 	return;
 }
 
-static int response_mad(struct ib_mad *mad)
-{
-	/* Trap represses are responses although response bit is reset */
-	return ((mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS) ||
-		(mad->mad_hdr.method & IB_MGMT_METHOD_RESP));
-}
-
-static int solicited_mad(struct ib_mad *mad)
-{
-	/* CM MADs are never solicited */
-	if (mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_CM) {
-		return 0;
-	}
-
-	/* XXX: Determine whether MAD is using RMPP */
-
-	/* Not using RMPP */
-	/* Is this MAD a response to a previous MAD ? */
-	return response_mad(mad);
-}
-
 static struct ib_mad_agent_private *
 find_mad_agent(struct ib_mad_port_private *port_priv,
-	       struct ib_mad *mad,
-	       int solicited)
+	       struct ib_mad *mad)
 {
 	struct ib_mad_agent_private *mad_agent = NULL;
 	unsigned long flags;
 
 	spin_lock_irqsave(&port_priv->reg_lock, flags);
-
-	/*
-	 * Whether MAD was solicited determines type of routing to
-	 * MAD client.
-	 */
-	if (solicited) {
+	if (response_mad(mad)) {
 		u32 hi_tid;
 		struct ib_mad_agent_private *entry;
 
@@ -1477,21 +1541,20 @@ out:
 	return valid;
 }
 
-/*
- * Return start of fully reassembled MAD, or NULL, if MAD isn't assembled yet
- */
-static struct ib_mad_private *
-reassemble_recv(struct ib_mad_agent_private *mad_agent_priv,
-		struct ib_mad_private *recv)
+static int is_data_mad(struct ib_mad_agent_private *mad_agent_priv,
+		       struct ib_mad_hdr *mad_hdr)
 {
-	/* Until we have RMPP, all receives are reassembled!... */
-	INIT_LIST_HEAD(&recv->header.recv_wc.recv_buf.list);
-	return recv;
+	struct ib_rmpp_mad *rmpp_mad;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_hdr;
+	return !mad_agent_priv->agent.rmpp_version ||
+		!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+				    IB_MGMT_RMPP_FLAG_ACTIVE) ||
+		(rmpp_mad->rmpp_hdr.rmpp_type == IB_MGMT_RMPP_TYPE_DATA);
 }
 
-static struct ib_mad_send_wr_private*
-find_send_req(struct ib_mad_agent_private *mad_agent_priv,
-	      u64 tid)
+struct ib_mad_send_wr_private*
+ib_find_send_mad(struct ib_mad_agent_private *mad_agent_priv, u64 tid)
 {
 	struct ib_mad_send_wr_private *mad_send_wr;
 
@@ -1507,7 +1570,9 @@ find_send_req(struct ib_mad_agent_private *mad_agent_priv,
 	 */
 	list_for_each_entry(mad_send_wr, &mad_agent_priv->send_list,
 			    agent_list) {
-		if (mad_send_wr->tid == tid && mad_send_wr->timeout) {
+		if (is_data_mad(mad_agent_priv,
+				mad_send_wr->send_wr.wr.ud.mad_hdr) &&
+		    mad_send_wr->tid == tid && mad_send_wr->timeout) {
 			/* Verify request has not been canceled */
 			return (mad_send_wr->status == IB_WC_SUCCESS) ?
 				mad_send_wr : NULL;
@@ -1516,43 +1581,55 @@ find_send_req(struct ib_mad_agent_private *mad_agent_priv,
 	return NULL;
 }
 
+void ib_mark_mad_done(struct ib_mad_send_wr_private *mad_send_wr)
+{
+	mad_send_wr->timeout = 0;
+	if (mad_send_wr->refcount == 1) {
+		list_del(&mad_send_wr->agent_list);
+		list_add_tail(&mad_send_wr->agent_list,
+			      &mad_send_wr->mad_agent_priv->done_list);
+	}
+}
+
 static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
-				 struct ib_mad_private *recv,
-				 int solicited)
+				 struct ib_mad_recv_wc *mad_recv_wc)
 {
 	struct ib_mad_send_wr_private *mad_send_wr;
 	struct ib_mad_send_wc mad_send_wc;
 	unsigned long flags;
-
-	/* Fully reassemble receive before processing */
-	recv = reassemble_recv(mad_agent_priv, recv);
-	if (!recv) {
-		if (atomic_dec_and_test(&mad_agent_priv->refcount))
-			wake_up(&mad_agent_priv->wait);
-		return;
+	u64 tid;
+
+	INIT_LIST_HEAD(&mad_recv_wc->rmpp_list);
+	list_add(&mad_recv_wc->recv_buf.list, &mad_recv_wc->rmpp_list);
+	if (mad_agent_priv->agent.rmpp_version) {
+		mad_recv_wc = ib_process_rmpp_recv_wc(mad_agent_priv,
+						      mad_recv_wc);
+		if (!mad_recv_wc) {
+			if (atomic_dec_and_test(&mad_agent_priv->refcount))
+				wake_up(&mad_agent_priv->wait);
+			return;
+		}
 	}
 
 	/* Complete corresponding request */
-	if (solicited) {
+	if (response_mad(mad_recv_wc->recv_buf.mad)) {
+		tid = mad_recv_wc->recv_buf.mad->mad_hdr.tid;
 		spin_lock_irqsave(&mad_agent_priv->lock, flags);
-		mad_send_wr = find_send_req(mad_agent_priv,
-					    recv->mad.mad.mad_hdr.tid);
+		mad_send_wr = ib_find_send_mad(mad_agent_priv, tid);
 		if (!mad_send_wr) {
 			spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-			ib_free_recv_mad(&recv->header.recv_wc);
+			ib_free_recv_mad(mad_recv_wc);
 			if (atomic_dec_and_test(&mad_agent_priv->refcount))
 				wake_up(&mad_agent_priv->wait);
 			return;
 		}
-		/* Timeout = 0 means that we won't wait for a response */
-		mad_send_wr->timeout = 0;
+		ib_mark_mad_done(mad_send_wr);
 		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 
 		/* Defined behavior is to complete response before request */
-		recv->header.recv_wc.wc->wr_id = mad_send_wr->wr_id;
-		mad_agent_priv->agent.recv_handler(
-						&mad_agent_priv->agent,
-						&recv->header.recv_wc);
+		mad_recv_wc->wc->wr_id = mad_send_wr->wr_id;
+		mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent,
+						   mad_recv_wc);
 		atomic_dec(&mad_agent_priv->refcount);
 
 		mad_send_wc.status = IB_WC_SUCCESS;
@@ -1560,9 +1637,8 @@ static void ib_mad_complete_recv(struct ib_mad_agent_private *mad_agent_priv,
 		mad_send_wc.wr_id = mad_send_wr->wr_id;
 		ib_mad_complete_send_wr(mad_send_wr, &mad_send_wc);
 	} else {
-		mad_agent_priv->agent.recv_handler(
-						&mad_agent_priv->agent,
-						&recv->header.recv_wc);
+		mad_agent_priv->agent.recv_handler(&mad_agent_priv->agent,
+						   mad_recv_wc);
 		if (atomic_dec_and_test(&mad_agent_priv->refcount))
 			wake_up(&mad_agent_priv->wait);
 	}
@@ -1576,7 +1652,6 @@ static void ib_mad_recv_done_handler(struct ib_mad_port_private *port_priv,
 	struct ib_mad_private *recv, *response;
 	struct ib_mad_list_head *mad_list;
 	struct ib_mad_agent_private *mad_agent;
-	int solicited;
 
 	response = kmem_cache_alloc(ib_mad_cache, GFP_KERNEL);
 	if (!response)
@@ -1662,11 +1737,9 @@ local:
 		}
 	}
 
-	/* Determine corresponding MAD agent for incoming receive MAD */
-	solicited = solicited_mad(&recv->mad.mad);
-	mad_agent = find_mad_agent(port_priv, &recv->mad.mad, solicited);
+	mad_agent = find_mad_agent(port_priv, &recv->mad.mad);
 	if (mad_agent) {
-		ib_mad_complete_recv(mad_agent, recv, solicited);
+		ib_mad_complete_recv(mad_agent, &recv->header.recv_wc);
 		/*
 		 * recv is freed up in error cases in ib_mad_complete_recv
 		 * or via recv_handler in ib_mad_complete_recv()
@@ -1710,26 +1783,31 @@ static void adjust_timeout(struct ib_mad_agent_private *mad_agent_priv)
 	}
 }
 
-static void wait_for_response(struct ib_mad_agent_private *mad_agent_priv,
-			      struct ib_mad_send_wr_private *mad_send_wr )
+static void wait_for_response(struct ib_mad_send_wr_private *mad_send_wr)
 {
+	struct ib_mad_agent_private *mad_agent_priv;
 	struct ib_mad_send_wr_private *temp_mad_send_wr;
 	struct list_head *list_item;
 	unsigned long delay;
 
+	mad_agent_priv = mad_send_wr->mad_agent_priv;
 	list_del(&mad_send_wr->agent_list);
 
 	delay = mad_send_wr->timeout;
 	mad_send_wr->timeout += jiffies;
 
-	list_for_each_prev(list_item, &mad_agent_priv->wait_list) {
-		temp_mad_send_wr = list_entry(list_item,
-					      struct ib_mad_send_wr_private,
-					      agent_list);
-		if (time_after(mad_send_wr->timeout,
-			       temp_mad_send_wr->timeout))
-			break;
+	if (delay) {
+		list_for_each_prev(list_item, &mad_agent_priv->wait_list) {
+			temp_mad_send_wr = list_entry(list_item,
+						struct ib_mad_send_wr_private,
+						agent_list);
+			if (time_after(mad_send_wr->timeout,
+				       temp_mad_send_wr->timeout))
+				break;
+		}
 	}
+	else
+		list_item = &mad_agent_priv->wait_list;
 	list_add(&mad_send_wr->agent_list, list_item);
 
 	/* Reschedule a work item if we have a shorter timeout */
@@ -1740,19 +1818,32 @@ static void wait_for_response(struct ib_mad_agent_private *mad_agent_priv,
 	}
 }
 
+void ib_reset_mad_timeout(struct ib_mad_send_wr_private *mad_send_wr,
+			  int timeout_ms)
+{
+	mad_send_wr->timeout = msecs_to_jiffies(timeout_ms);
+	wait_for_response(mad_send_wr);
+}
+
 /*
  * Process a send work completion
  */
-static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
-				    struct ib_mad_send_wc *mad_send_wc)
+void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
+			     struct ib_mad_send_wc *mad_send_wc)
 {
 	struct ib_mad_agent_private	*mad_agent_priv;
 	unsigned long			flags;
+	int				ret;
 
-	mad_agent_priv = container_of(mad_send_wr->agent,
-				      struct ib_mad_agent_private, agent);
-
+	mad_agent_priv = mad_send_wr->mad_agent_priv;
 	spin_lock_irqsave(&mad_agent_priv->lock, flags);
+	if (mad_agent_priv->agent.rmpp_version) {
+		ret = ib_process_rmpp_send_wc(mad_send_wr, mad_send_wc);
+		if (ret == IB_RMPP_RESULT_CONSUMED)
+			goto done;
+	} else
+		ret = IB_RMPP_RESULT_UNHANDLED;
+
 	if (mad_send_wc->status != IB_WC_SUCCESS &&
 	    mad_send_wr->status == IB_WC_SUCCESS) {
 		mad_send_wr->status = mad_send_wc->status;
@@ -1762,10 +1853,9 @@ static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
 	if (--mad_send_wr->refcount > 0) {
 		if (mad_send_wr->refcount == 1 && mad_send_wr->timeout &&
 		    mad_send_wr->status == IB_WC_SUCCESS) {
-			wait_for_response(mad_agent_priv, mad_send_wr);
+			wait_for_response(mad_send_wr);
 		}
-		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-		return;
+		goto done;
 	}
 
 	/* Remove send from MAD agent and notify client of completion */
@@ -1775,14 +1865,18 @@ static void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
 
 	if (mad_send_wr->status != IB_WC_SUCCESS )
 		mad_send_wc->status = mad_send_wr->status;
-	mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
-					    mad_send_wc);
+	if (ret != IB_RMPP_RESULT_INTERNAL)
+		mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
+						   mad_send_wc);
 
 	/* Release reference on agent taken when sending */
 	if (atomic_dec_and_test(&mad_agent_priv->refcount))
 		wake_up(&mad_agent_priv->wait);
 
 	kfree(mad_send_wr);
+	return;
+done:
+	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 }
 
 static void ib_mad_send_done_handler(struct ib_mad_port_private *port_priv,
@@ -1961,6 +2055,8 @@ static void cancel_mads(struct ib_mad_agent_private *mad_agent_priv)
 
 	/* Empty wait list to prevent receives from finding a request */
 	list_splice_init(&mad_agent_priv->wait_list, &cancel_list);
+	/* Empty local completion list as well */
+	list_splice_init(&mad_agent_priv->local_list, &cancel_list);
 	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 
 	/* Report all cancelled requests */
@@ -1980,8 +2076,7 @@ static void cancel_mads(struct ib_mad_agent_private *mad_agent_priv)
 }
 
 static struct ib_mad_send_wr_private*
-find_send_by_wr_id(struct ib_mad_agent_private *mad_agent_priv,
-		   u64 wr_id)
+find_send_by_wr_id(struct ib_mad_agent_private *mad_agent_priv, u64 wr_id)
 {
 	struct ib_mad_send_wr_private *mad_send_wr;
 
@@ -1993,79 +2088,50 @@ find_send_by_wr_id(struct ib_mad_agent_private *mad_agent_priv,
 
 	list_for_each_entry(mad_send_wr, &mad_agent_priv->send_list,
 			    agent_list) {
-		if (mad_send_wr->wr_id == wr_id)
+		if (is_data_mad(mad_agent_priv,
+				mad_send_wr->send_wr.wr.ud.mad_hdr) &&
+		    mad_send_wr->wr_id == wr_id)
 			return mad_send_wr;
 	}
 	return NULL;
 }
 
-void cancel_sends(void *data)
-{
-	struct ib_mad_agent_private *mad_agent_priv;
-	struct ib_mad_send_wr_private *mad_send_wr;
-	struct ib_mad_send_wc mad_send_wc;
-	unsigned long flags;
-
-	mad_agent_priv = data;
-
-	mad_send_wc.status = IB_WC_WR_FLUSH_ERR;
-	mad_send_wc.vendor_err = 0;
-
-	spin_lock_irqsave(&mad_agent_priv->lock, flags);
-	while (!list_empty(&mad_agent_priv->canceled_list)) {
-		mad_send_wr = list_entry(mad_agent_priv->canceled_list.next,
-					 struct ib_mad_send_wr_private,
-					 agent_list);
-
-		list_del(&mad_send_wr->agent_list);
-		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-
-		mad_send_wc.wr_id = mad_send_wr->wr_id;
-		mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
-						   &mad_send_wc);
-
-		kfree(mad_send_wr);
-		if (atomic_dec_and_test(&mad_agent_priv->refcount))
-			wake_up(&mad_agent_priv->wait);
-		spin_lock_irqsave(&mad_agent_priv->lock, flags);
-	}
-	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-}
-
-void ib_cancel_mad(struct ib_mad_agent *mad_agent,
-		  u64 wr_id)
+int ib_modify_mad(struct ib_mad_agent *mad_agent, u64 wr_id, u32 timeout_ms)
 {
 	struct ib_mad_agent_private *mad_agent_priv;
 	struct ib_mad_send_wr_private *mad_send_wr;
 	unsigned long flags;
+	int active;
 
 	mad_agent_priv = container_of(mad_agent, struct ib_mad_agent_private,
 				      agent);
 	spin_lock_irqsave(&mad_agent_priv->lock, flags);
 	mad_send_wr = find_send_by_wr_id(mad_agent_priv, wr_id);
-	if (!mad_send_wr) {
+	if (!mad_send_wr || mad_send_wr->status != IB_WC_SUCCESS) {
 		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-		goto out;
+		return -EINVAL;
 	}
 
-	if (mad_send_wr->status == IB_WC_SUCCESS)
-		mad_send_wr->refcount -= (mad_send_wr->timeout > 0);
-
-	if (mad_send_wr->refcount != 0) {
+	active = (!mad_send_wr->timeout || mad_send_wr->refcount > 1);
+	if (!timeout_ms) {
 		mad_send_wr->status = IB_WC_WR_FLUSH_ERR;
-		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
-		goto out;
+		mad_send_wr->refcount -= (mad_send_wr->timeout > 0);
 	}
 
-	list_del(&mad_send_wr->agent_list);
-	list_add_tail(&mad_send_wr->agent_list, &mad_agent_priv->canceled_list);
-	adjust_timeout(mad_agent_priv);
+	mad_send_wr->send_wr.wr.ud.timeout_ms = timeout_ms;
+	if (active)
+		mad_send_wr->timeout = msecs_to_jiffies(timeout_ms);
+	else
+		ib_reset_mad_timeout(mad_send_wr, timeout_ms);
+
 	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
+	return 0;
+}
+EXPORT_SYMBOL(ib_modify_mad);
 
-	queue_work(mad_agent_priv->qp_info->port_priv->wq,
-		   &mad_agent_priv->canceled_work);
-out:
-	return;
+void ib_cancel_mad(struct ib_mad_agent *mad_agent, u64 wr_id)
+{
+	ib_modify_mad(mad_agent, wr_id, 0);
 }
 EXPORT_SYMBOL(ib_cancel_mad);
 
@@ -2075,6 +2141,7 @@ static void local_completions(void *data)
 	struct ib_mad_local_private *local;
 	struct ib_mad_agent_private *recv_mad_agent;
 	unsigned long flags;
+	int recv = 0;
 	struct ib_wc wc;
 	struct ib_mad_send_wc mad_send_wc;
 
@@ -2090,10 +2157,10 @@ static void local_completions(void *data)
 			recv_mad_agent = local->recv_mad_agent;
 			if (!recv_mad_agent) {
 				printk(KERN_ERR PFX "No receive MAD agent for local completion\n");
-				kmem_cache_free(ib_mad_cache, local->mad_priv);
 				goto local_send_completion;
 			}
 
+			recv = 1;
 			/*
 			 * Defined behavior is to complete response
 			 * before request
@@ -2105,7 +2172,9 @@ static void local_completions(void *data)
 			local->mad_priv->header.recv_wc.wc = &wc;
 			local->mad_priv->header.recv_wc.mad_len =
 						sizeof(struct ib_mad);
-			INIT_LIST_HEAD(&local->mad_priv->header.recv_wc.recv_buf.list);
+			INIT_LIST_HEAD(&local->mad_priv->header.recv_wc.rmpp_list);
+			list_add(&local->mad_priv->header.recv_wc.recv_buf.list,
+				 &local->mad_priv->header.recv_wc.rmpp_list);
 			local->mad_priv->header.recv_wc.recv_buf.grh = NULL;
 			local->mad_priv->header.recv_wc.recv_buf.mad =
 						&local->mad_priv->mad.mad;
@@ -2136,11 +2205,47 @@ local_send_completion:
 		spin_lock_irqsave(&mad_agent_priv->lock, flags);
 		list_del(&local->completion_list);
 		atomic_dec(&mad_agent_priv->refcount);
+		if (!recv)
+			kmem_cache_free(ib_mad_cache, local->mad_priv);
 		kfree(local);
 	}
 	spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 }
 
+static int retry_send(struct ib_mad_send_wr_private *mad_send_wr)
+{
+	int ret;
+
+	if (!mad_send_wr->retries--)
+		return -ETIMEDOUT;
+
+	mad_send_wr->timeout = msecs_to_jiffies(mad_send_wr->send_wr.
+						wr.ud.timeout_ms);
+
+	if (mad_send_wr->mad_agent_priv->agent.rmpp_version) {
+		ret = ib_retry_rmpp(mad_send_wr);
+		switch (ret) {
+		case IB_RMPP_RESULT_UNHANDLED:
+			ret = ib_send_mad(mad_send_wr);
+			break;
+		case IB_RMPP_RESULT_CONSUMED:
+			ret = 0;
+			break;
+		default:
+			ret = -ECOMM;
+			break;
+		}
+	} else
+		ret = ib_send_mad(mad_send_wr);
+
+	if (!ret) {
+		mad_send_wr->refcount++;
+		list_add_tail(&mad_send_wr->agent_list,
+			      &mad_send_wr->mad_agent_priv->send_list);
+	}
+	return ret;
+}
+
 static void timeout_sends(void *data)
 {
 	struct ib_mad_agent_private *mad_agent_priv;
@@ -2149,8 +2254,6 @@ static void timeout_sends(void *data)
 	unsigned long flags, delay;
 
 	mad_agent_priv = (struct ib_mad_agent_private *)data;
-
-	mad_send_wc.status = IB_WC_RESP_TIMEOUT_ERR;
 	mad_send_wc.vendor_err = 0;
 
 	spin_lock_irqsave(&mad_agent_priv->lock, flags);
@@ -2170,8 +2273,16 @@ static void timeout_sends(void *data)
 		}
 
 		list_del(&mad_send_wr->agent_list);
+		if (mad_send_wr->status == IB_WC_SUCCESS &&
+		    !retry_send(mad_send_wr))
+			continue;
+
 		spin_unlock_irqrestore(&mad_agent_priv->lock, flags);
 
+		if (mad_send_wr->status == IB_WC_SUCCESS)
+			mad_send_wc.status = IB_WC_RESP_TIMEOUT_ERR;
+		else
+			mad_send_wc.status = mad_send_wr->status;
 		mad_send_wc.wr_id = mad_send_wr->wr_id;
 		mad_agent_priv->agent.send_handler(&mad_agent_priv->agent,
 						   &mad_send_wc);
@@ -2447,14 +2558,6 @@ static int ib_mad_port_open(struct ib_device *device,
 	unsigned long flags;
 	char name[sizeof "ib_mad123"];
 
-	/* First, check if port already open at MAD layer */
-	port_priv = ib_get_mad_port(device, port_num);
-	if (port_priv) {
-		printk(KERN_DEBUG PFX "%s port %d already open\n",
-		       device->name, port_num);
-		return 0;
-	}
-
 	/* Create new device info */
 	port_priv = kmalloc(sizeof *port_priv, GFP_KERNEL);
 	if (!port_priv) {
@@ -2579,7 +2682,7 @@ static int ib_mad_port_close(struct ib_device *device, int port_num)
 
 static void ib_mad_init_device(struct ib_device *device)
 {
-	int ret, num_ports, cur_port, i, ret2;
+	int num_ports, cur_port, i;
 
 	if (device->node_type == IB_NODE_SWITCH) {
 		num_ports = 1;
@@ -2589,47 +2692,37 @@ static void ib_mad_init_device(struct ib_device *device)
 		cur_port = 1;
 	}
 	for (i = 0; i < num_ports; i++, cur_port++) {
-		ret = ib_mad_port_open(device, cur_port);
-		if (ret) {
+		if (ib_mad_port_open(device, cur_port)) {
 			printk(KERN_ERR PFX "Couldn't open %s port %d\n",
 			       device->name, cur_port);
 			goto error_device_open;
 		}
-		ret = ib_agent_port_open(device, cur_port);
-		if (ret) {
+		if (ib_agent_port_open(device, cur_port)) {
 			printk(KERN_ERR PFX "Couldn't open %s port %d "
 			       "for agents\n",
 			       device->name, cur_port);
 			goto error_device_open;
 		}
 	}
-
-	goto error_device_query;
+	return;
 
 error_device_open:
 	while (i > 0) {
 		cur_port--;
-		ret2 = ib_agent_port_close(device, cur_port);
-		if (ret2) {
+		if (ib_agent_port_close(device, cur_port))
 			printk(KERN_ERR PFX "Couldn't close %s port %d "
 			       "for agents\n",
 			       device->name, cur_port);
-		}
-		ret2 = ib_mad_port_close(device, cur_port);
-		if (ret2) {
+		if (ib_mad_port_close(device, cur_port))
 			printk(KERN_ERR PFX "Couldn't close %s port %d\n",
 			       device->name, cur_port);
-		}
 		i--;
 	}
-
-error_device_query:
-	return;
 }
 
 static void ib_mad_remove_device(struct ib_device *device)
 {
-	int ret = 0, i, num_ports, cur_port, ret2;
+	int i, num_ports, cur_port;
 
 	if (device->node_type == IB_NODE_SWITCH) {
 		num_ports = 1;
@@ -2639,21 +2732,13 @@ static void ib_mad_remove_device(struct ib_device *device)
 		cur_port = 1;
 	}
 	for (i = 0; i < num_ports; i++, cur_port++) {
-		ret2 = ib_agent_port_close(device, cur_port);
-		if (ret2) {
+		if (ib_agent_port_close(device, cur_port))
 			printk(KERN_ERR PFX "Couldn't close %s port %d "
 			       "for agents\n",
 			       device->name, cur_port);
-			if (!ret)
-				ret = ret2;
-		}
-		ret2 = ib_mad_port_close(device, cur_port);
-		if (ret2) {
+		if (ib_mad_port_close(device, cur_port))
 			printk(KERN_ERR PFX "Couldn't close %s port %d\n",
 			       device->name, cur_port);
-			if (!ret)
-				ret = ret2;
-		}
 	}
 }
 
@@ -2709,3 +2794,4 @@ static void __exit ib_mad_cleanup_module(void)
 
 module_init(ib_mad_init_module);
 module_exit(ib_mad_cleanup_module);
+
diff --git a/drivers/infiniband/core/mad_priv.h b/drivers/infiniband/core/mad_priv.h
index 008cbcb94b15..568da10b05ab 100644
--- a/drivers/infiniband/core/mad_priv.h
+++ b/drivers/infiniband/core/mad_priv.h
@@ -1,5 +1,7 @@
 /*
  * Copyright (c) 2004, 2005, Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005 Intel Corporation. All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, 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
@@ -29,7 +31,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: mad_priv.h 1389 2004-12-27 22:56:47Z roland $
+ * $Id: mad_priv.h 2730 2005-06-28 16:43:03Z sean.hefty $
  */
 
 #ifndef __IB_MAD_PRIV_H__
@@ -92,16 +94,15 @@ struct ib_mad_agent_private {
 	spinlock_t lock;
 	struct list_head send_list;
 	struct list_head wait_list;
+	struct list_head done_list;
 	struct work_struct timed_work;
 	unsigned long timeout;
 	struct list_head local_list;
 	struct work_struct local_work;
-	struct list_head canceled_list;
-	struct work_struct canceled_work;
+	struct list_head rmpp_list;
 
 	atomic_t refcount;
 	wait_queue_head_t wait;
-	u8 rmpp_version;
 };
 
 struct ib_mad_snoop_private {
@@ -116,15 +117,24 @@ struct ib_mad_snoop_private {
 struct ib_mad_send_wr_private {
 	struct ib_mad_list_head mad_list;
 	struct list_head agent_list;
-	struct ib_mad_agent *agent;
+	struct ib_mad_agent_private *mad_agent_priv;
 	struct ib_send_wr send_wr;
 	struct ib_sge sg_list[IB_MAD_SEND_REQ_MAX_SG];
 	u64 wr_id;			/* client WR ID */
 	u64 tid;
 	unsigned long timeout;
+	int retries;
 	int retry;
 	int refcount;
 	enum ib_wc_status status;
+
+	/* RMPP control */
+	int last_ack;
+	int seg_num;
+	int newwin;
+	int total_seg;
+	int data_offset;
+	int pad;
 };
 
 struct ib_mad_local_private {
@@ -197,4 +207,17 @@ struct ib_mad_port_private {
 
 extern kmem_cache_t *ib_mad_cache;
 
+int ib_send_mad(struct ib_mad_send_wr_private *mad_send_wr);
+
+struct ib_mad_send_wr_private *
+ib_find_send_mad(struct ib_mad_agent_private *mad_agent_priv, u64 tid);
+
+void ib_mad_complete_send_wr(struct ib_mad_send_wr_private *mad_send_wr,
+			     struct ib_mad_send_wc *mad_send_wc);
+
+void ib_mark_mad_done(struct ib_mad_send_wr_private *mad_send_wr);
+
+void ib_reset_mad_timeout(struct ib_mad_send_wr_private *mad_send_wr,
+			  int timeout_ms);
+
 #endif	/* __IB_MAD_PRIV_H__ */
diff --git a/drivers/infiniband/core/mad_rmpp.c b/drivers/infiniband/core/mad_rmpp.c
new file mode 100644
index 000000000000..8f1eb80e421f
--- /dev/null
+++ b/drivers/infiniband/core/mad_rmpp.c
@@ -0,0 +1,765 @@
+/*
+ * Copyright (c) 2005 Intel Inc. All rights reserved.
+ * Copyright (c) 2005 Voltaire, 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.
+ *
+ * $Id: mad_rmpp.c 1921 2005-03-02 22:58:44Z sean.hefty $
+ */
+
+#include <linux/dma-mapping.h>
+
+#include "mad_priv.h"
+#include "mad_rmpp.h"
+
+enum rmpp_state {
+	RMPP_STATE_ACTIVE,
+	RMPP_STATE_TIMEOUT,
+	RMPP_STATE_COMPLETE
+};
+
+struct mad_rmpp_recv {
+	struct ib_mad_agent_private *agent;
+	struct list_head list;
+	struct work_struct timeout_work;
+	struct work_struct cleanup_work;
+	wait_queue_head_t wait;
+	enum rmpp_state state;
+	spinlock_t lock;
+	atomic_t refcount;
+
+	struct ib_ah *ah;
+	struct ib_mad_recv_wc *rmpp_wc;
+	struct ib_mad_recv_buf *cur_seg_buf;
+	int last_ack;
+	int seg_num;
+	int newwin;
+
+	u64 tid;
+	u32 src_qp;
+	u16 slid;
+	u8 mgmt_class;
+	u8 class_version;
+	u8 method;
+};
+
+static void destroy_rmpp_recv(struct mad_rmpp_recv *rmpp_recv)
+{
+	atomic_dec(&rmpp_recv->refcount);
+	wait_event(rmpp_recv->wait, !atomic_read(&rmpp_recv->refcount));
+	ib_destroy_ah(rmpp_recv->ah);
+	kfree(rmpp_recv);
+}
+
+void ib_cancel_rmpp_recvs(struct ib_mad_agent_private *agent)
+{
+	struct mad_rmpp_recv *rmpp_recv, *temp_rmpp_recv;
+	unsigned long flags;
+
+	spin_lock_irqsave(&agent->lock, flags);
+	list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
+		cancel_delayed_work(&rmpp_recv->timeout_work);
+		cancel_delayed_work(&rmpp_recv->cleanup_work);
+	}
+	spin_unlock_irqrestore(&agent->lock, flags);
+
+	flush_workqueue(agent->qp_info->port_priv->wq);
+
+	list_for_each_entry_safe(rmpp_recv, temp_rmpp_recv,
+				 &agent->rmpp_list, list) {
+		list_del(&rmpp_recv->list);
+		if (rmpp_recv->state != RMPP_STATE_COMPLETE)
+			ib_free_recv_mad(rmpp_recv->rmpp_wc);
+		destroy_rmpp_recv(rmpp_recv);
+	}
+}
+
+static void recv_timeout_handler(void *data)
+{
+	struct mad_rmpp_recv *rmpp_recv = data;
+	struct ib_mad_recv_wc *rmpp_wc;
+	unsigned long flags;
+
+	spin_lock_irqsave(&rmpp_recv->agent->lock, flags);
+	if (rmpp_recv->state != RMPP_STATE_ACTIVE) {
+		spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
+		return;
+	}
+	rmpp_recv->state = RMPP_STATE_TIMEOUT;
+	list_del(&rmpp_recv->list);
+	spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
+
+	/* TODO: send abort. */
+	rmpp_wc = rmpp_recv->rmpp_wc;
+	destroy_rmpp_recv(rmpp_recv);
+	ib_free_recv_mad(rmpp_wc);
+}
+
+static void recv_cleanup_handler(void *data)
+{
+	struct mad_rmpp_recv *rmpp_recv = data;
+	unsigned long flags;
+
+	spin_lock_irqsave(&rmpp_recv->agent->lock, flags);
+	list_del(&rmpp_recv->list);
+	spin_unlock_irqrestore(&rmpp_recv->agent->lock, flags);
+	destroy_rmpp_recv(rmpp_recv);
+}
+
+static struct mad_rmpp_recv *
+create_rmpp_recv(struct ib_mad_agent_private *agent,
+		 struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct mad_rmpp_recv *rmpp_recv;
+	struct ib_mad_hdr *mad_hdr;
+
+	rmpp_recv = kmalloc(sizeof *rmpp_recv, GFP_KERNEL);
+	if (!rmpp_recv)
+		return NULL;
+
+	rmpp_recv->ah = ib_create_ah_from_wc(agent->agent.qp->pd,
+					     mad_recv_wc->wc,
+					     mad_recv_wc->recv_buf.grh,
+					     agent->agent.port_num);
+	if (IS_ERR(rmpp_recv->ah))
+		goto error;
+
+	rmpp_recv->agent = agent;
+	init_waitqueue_head(&rmpp_recv->wait);
+	INIT_WORK(&rmpp_recv->timeout_work, recv_timeout_handler, rmpp_recv);
+	INIT_WORK(&rmpp_recv->cleanup_work, recv_cleanup_handler, rmpp_recv);
+	spin_lock_init(&rmpp_recv->lock);
+	rmpp_recv->state = RMPP_STATE_ACTIVE;
+	atomic_set(&rmpp_recv->refcount, 1);
+
+	rmpp_recv->rmpp_wc = mad_recv_wc;
+	rmpp_recv->cur_seg_buf = &mad_recv_wc->recv_buf;
+	rmpp_recv->newwin = 1;
+	rmpp_recv->seg_num = 1;
+	rmpp_recv->last_ack = 0;
+
+	mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr;
+	rmpp_recv->tid = mad_hdr->tid;
+	rmpp_recv->src_qp = mad_recv_wc->wc->src_qp;
+	rmpp_recv->slid = mad_recv_wc->wc->slid;
+	rmpp_recv->mgmt_class = mad_hdr->mgmt_class;
+	rmpp_recv->class_version = mad_hdr->class_version;
+	rmpp_recv->method  = mad_hdr->method;
+	return rmpp_recv;
+
+error:	kfree(rmpp_recv);
+	return NULL;
+}
+
+static inline void deref_rmpp_recv(struct mad_rmpp_recv *rmpp_recv)
+{
+	if (atomic_dec_and_test(&rmpp_recv->refcount))
+		wake_up(&rmpp_recv->wait);
+}
+
+static struct mad_rmpp_recv *
+find_rmpp_recv(struct ib_mad_agent_private *agent,
+	       struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct mad_rmpp_recv *rmpp_recv;
+	struct ib_mad_hdr *mad_hdr = &mad_recv_wc->recv_buf.mad->mad_hdr;
+
+	list_for_each_entry(rmpp_recv, &agent->rmpp_list, list) {
+		if (rmpp_recv->tid == mad_hdr->tid &&
+		    rmpp_recv->src_qp == mad_recv_wc->wc->src_qp &&
+		    rmpp_recv->slid == mad_recv_wc->wc->slid &&
+		    rmpp_recv->mgmt_class == mad_hdr->mgmt_class &&
+		    rmpp_recv->class_version == mad_hdr->class_version &&
+		    rmpp_recv->method == mad_hdr->method)
+			return rmpp_recv;
+	}
+	return NULL;
+}
+
+static struct mad_rmpp_recv *
+acquire_rmpp_recv(struct ib_mad_agent_private *agent,
+		  struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct mad_rmpp_recv *rmpp_recv;
+	unsigned long flags;
+
+	spin_lock_irqsave(&agent->lock, flags);
+	rmpp_recv = find_rmpp_recv(agent, mad_recv_wc);
+	if (rmpp_recv)
+		atomic_inc(&rmpp_recv->refcount);
+	spin_unlock_irqrestore(&agent->lock, flags);
+	return rmpp_recv;
+}
+
+static struct mad_rmpp_recv *
+insert_rmpp_recv(struct ib_mad_agent_private *agent,
+		 struct mad_rmpp_recv *rmpp_recv)
+{
+	struct mad_rmpp_recv *cur_rmpp_recv;
+
+	cur_rmpp_recv = find_rmpp_recv(agent, rmpp_recv->rmpp_wc);
+	if (!cur_rmpp_recv)
+		list_add_tail(&rmpp_recv->list, &agent->rmpp_list);
+
+	return cur_rmpp_recv;
+}
+
+static int data_offset(u8 mgmt_class)
+{
+	if (mgmt_class == IB_MGMT_CLASS_SUBN_ADM)
+		return offsetof(struct ib_sa_mad, data);
+	else if ((mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
+		 (mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END))
+		return offsetof(struct ib_vendor_mad, data);
+	else
+		return offsetof(struct ib_rmpp_mad, data);
+}
+
+static void format_ack(struct ib_rmpp_mad *ack,
+		       struct ib_rmpp_mad *data,
+		       struct mad_rmpp_recv *rmpp_recv)
+{
+	unsigned long flags;
+
+	memcpy(&ack->mad_hdr, &data->mad_hdr,
+	       data_offset(data->mad_hdr.mgmt_class));
+
+	ack->mad_hdr.method ^= IB_MGMT_METHOD_RESP;
+	ack->rmpp_hdr.rmpp_type = IB_MGMT_RMPP_TYPE_ACK;
+	ib_set_rmpp_flags(&ack->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
+
+	spin_lock_irqsave(&rmpp_recv->lock, flags);
+	rmpp_recv->last_ack = rmpp_recv->seg_num;
+	ack->rmpp_hdr.seg_num = cpu_to_be32(rmpp_recv->seg_num);
+	ack->rmpp_hdr.paylen_newwin = cpu_to_be32(rmpp_recv->newwin);
+	spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+}
+
+static void ack_recv(struct mad_rmpp_recv *rmpp_recv,
+		     struct ib_mad_recv_wc *recv_wc)
+{
+	struct ib_mad_send_buf *msg;
+	struct ib_send_wr *bad_send_wr;
+	int hdr_len, ret;
+
+	hdr_len = sizeof(struct ib_mad_hdr) + sizeof(struct ib_rmpp_hdr);
+	msg = ib_create_send_mad(&rmpp_recv->agent->agent, recv_wc->wc->src_qp,
+				 recv_wc->wc->pkey_index, rmpp_recv->ah, 1,
+				 hdr_len, sizeof(struct ib_rmpp_mad) - hdr_len,
+				 GFP_KERNEL);
+	if (!msg)
+		return;
+
+	format_ack((struct ib_rmpp_mad *) msg->mad,
+		   (struct ib_rmpp_mad *) recv_wc->recv_buf.mad, rmpp_recv);
+	ret = ib_post_send_mad(&rmpp_recv->agent->agent, &msg->send_wr,
+			       &bad_send_wr);
+	if (ret)
+		ib_free_send_mad(msg);
+}
+
+static inline int get_last_flag(struct ib_mad_recv_buf *seg)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+
+	rmpp_mad = (struct ib_rmpp_mad *) seg->mad;
+	return ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_LAST;
+}
+
+static inline int get_seg_num(struct ib_mad_recv_buf *seg)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+
+	rmpp_mad = (struct ib_rmpp_mad *) seg->mad;
+	return be32_to_cpu(rmpp_mad->rmpp_hdr.seg_num);
+}
+
+static inline struct ib_mad_recv_buf * get_next_seg(struct list_head *rmpp_list,
+						    struct ib_mad_recv_buf *seg)
+{
+	if (seg->list.next == rmpp_list)
+		return NULL;
+
+	return container_of(seg->list.next, struct ib_mad_recv_buf, list);
+}
+
+static inline int window_size(struct ib_mad_agent_private *agent)
+{
+	return max(agent->qp_info->recv_queue.max_active >> 3, 1);
+}
+
+static struct ib_mad_recv_buf * find_seg_location(struct list_head *rmpp_list,
+						  int seg_num)
+{
+        struct ib_mad_recv_buf *seg_buf;
+	int cur_seg_num;
+
+	list_for_each_entry_reverse(seg_buf, rmpp_list, list) {
+		cur_seg_num = get_seg_num(seg_buf);
+		if (seg_num > cur_seg_num)
+			return seg_buf;
+		if (seg_num == cur_seg_num)
+			break;
+	}
+	return NULL;
+}
+
+static void update_seg_num(struct mad_rmpp_recv *rmpp_recv,
+			   struct ib_mad_recv_buf *new_buf)
+{
+	struct list_head *rmpp_list = &rmpp_recv->rmpp_wc->rmpp_list;
+
+	while (new_buf && (get_seg_num(new_buf) == rmpp_recv->seg_num + 1)) {
+		rmpp_recv->cur_seg_buf = new_buf;
+		rmpp_recv->seg_num++;
+		new_buf = get_next_seg(rmpp_list, new_buf);
+	}
+}
+
+static inline int get_mad_len(struct mad_rmpp_recv *rmpp_recv)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+	int hdr_size, data_size, pad;
+
+	rmpp_mad = (struct ib_rmpp_mad *)rmpp_recv->cur_seg_buf->mad;
+
+	hdr_size = data_offset(rmpp_mad->mad_hdr.mgmt_class);
+	data_size = sizeof(struct ib_rmpp_mad) - hdr_size;
+	pad = data_size - be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
+	if (pad > data_size || pad < 0)
+		pad = 0;
+
+	return hdr_size + rmpp_recv->seg_num * data_size - pad;
+}
+
+static struct ib_mad_recv_wc * complete_rmpp(struct mad_rmpp_recv *rmpp_recv)
+{
+	struct ib_mad_recv_wc *rmpp_wc;
+
+	ack_recv(rmpp_recv, rmpp_recv->rmpp_wc);
+	if (rmpp_recv->seg_num > 1)
+		cancel_delayed_work(&rmpp_recv->timeout_work);
+
+	rmpp_wc = rmpp_recv->rmpp_wc;
+	rmpp_wc->mad_len = get_mad_len(rmpp_recv);
+	/* 10 seconds until we can find the packet lifetime */
+	queue_delayed_work(rmpp_recv->agent->qp_info->port_priv->wq,
+			   &rmpp_recv->cleanup_work, msecs_to_jiffies(10000));
+	return rmpp_wc;
+}
+
+void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc, void *buf)
+{
+	struct ib_mad_recv_buf *seg_buf;
+	struct ib_rmpp_mad *rmpp_mad;
+	void *data;
+	int size, len, offset;
+	u8 flags;
+
+	len = mad_recv_wc->mad_len;
+	if (len <= sizeof(struct ib_mad)) {
+		memcpy(buf, mad_recv_wc->recv_buf.mad, len);
+		return;
+	}
+
+	offset = data_offset(mad_recv_wc->recv_buf.mad->mad_hdr.mgmt_class);
+
+	list_for_each_entry(seg_buf, &mad_recv_wc->rmpp_list, list) {
+		rmpp_mad = (struct ib_rmpp_mad *)seg_buf->mad;
+		flags = ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr);
+
+		if (flags & IB_MGMT_RMPP_FLAG_FIRST) {
+			data = rmpp_mad;
+			size = sizeof(*rmpp_mad);
+		} else {
+			data = (void *) rmpp_mad + offset;
+			if (flags & IB_MGMT_RMPP_FLAG_LAST)
+				size = len;
+			else
+				size = sizeof(*rmpp_mad) - offset;
+		}
+
+		memcpy(buf, data, size);
+		len -= size;
+		buf += size;
+	}
+}
+EXPORT_SYMBOL(ib_coalesce_recv_mad);
+
+static struct ib_mad_recv_wc *
+continue_rmpp(struct ib_mad_agent_private *agent,
+	      struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct mad_rmpp_recv *rmpp_recv;
+	struct ib_mad_recv_buf *prev_buf;
+	struct ib_mad_recv_wc *done_wc;
+	int seg_num;
+	unsigned long flags;
+
+	rmpp_recv = acquire_rmpp_recv(agent, mad_recv_wc);
+	if (!rmpp_recv)
+		goto drop1;
+
+	seg_num = get_seg_num(&mad_recv_wc->recv_buf);
+
+	spin_lock_irqsave(&rmpp_recv->lock, flags);
+	if ((rmpp_recv->state == RMPP_STATE_TIMEOUT) ||
+	    (seg_num > rmpp_recv->newwin))
+		goto drop3;
+
+	if ((seg_num <= rmpp_recv->last_ack) ||
+	    (rmpp_recv->state == RMPP_STATE_COMPLETE)) {
+		spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+		ack_recv(rmpp_recv, mad_recv_wc);
+		goto drop2;
+	}
+
+	prev_buf = find_seg_location(&rmpp_recv->rmpp_wc->rmpp_list, seg_num);
+	if (!prev_buf)
+		goto drop3;
+
+	done_wc = NULL;
+	list_add(&mad_recv_wc->recv_buf.list, &prev_buf->list);
+	if (rmpp_recv->cur_seg_buf == prev_buf) {
+		update_seg_num(rmpp_recv, &mad_recv_wc->recv_buf);
+		if (get_last_flag(rmpp_recv->cur_seg_buf)) {
+			rmpp_recv->state = RMPP_STATE_COMPLETE;
+			spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+			done_wc = complete_rmpp(rmpp_recv);
+			goto out;
+		} else if (rmpp_recv->seg_num == rmpp_recv->newwin) {
+			rmpp_recv->newwin += window_size(agent);
+			spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+			ack_recv(rmpp_recv, mad_recv_wc);
+			goto out;
+		}
+	}
+	spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+out:
+	deref_rmpp_recv(rmpp_recv);
+	return done_wc;
+
+drop3:	spin_unlock_irqrestore(&rmpp_recv->lock, flags);
+drop2:	deref_rmpp_recv(rmpp_recv);
+drop1:	ib_free_recv_mad(mad_recv_wc);
+	return NULL;
+}
+
+static struct ib_mad_recv_wc *
+start_rmpp(struct ib_mad_agent_private *agent,
+	   struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct mad_rmpp_recv *rmpp_recv;
+	unsigned long flags;
+
+	rmpp_recv = create_rmpp_recv(agent, mad_recv_wc);
+	if (!rmpp_recv) {
+		ib_free_recv_mad(mad_recv_wc);
+		return NULL;
+	}
+
+	spin_lock_irqsave(&agent->lock, flags);
+	if (insert_rmpp_recv(agent, rmpp_recv)) {
+		spin_unlock_irqrestore(&agent->lock, flags);
+		/* duplicate first MAD */
+		destroy_rmpp_recv(rmpp_recv);
+		return continue_rmpp(agent, mad_recv_wc);
+	}
+	atomic_inc(&rmpp_recv->refcount);
+
+	if (get_last_flag(&mad_recv_wc->recv_buf)) {
+		rmpp_recv->state = RMPP_STATE_COMPLETE;
+		spin_unlock_irqrestore(&agent->lock, flags);
+		complete_rmpp(rmpp_recv);
+	} else {
+		spin_unlock_irqrestore(&agent->lock, flags);
+		/* 40 seconds until we can find the packet lifetimes */
+		queue_delayed_work(agent->qp_info->port_priv->wq,
+				   &rmpp_recv->timeout_work,
+				   msecs_to_jiffies(40000));
+		rmpp_recv->newwin += window_size(agent);
+		ack_recv(rmpp_recv, mad_recv_wc);
+		mad_recv_wc = NULL;
+	}
+	deref_rmpp_recv(rmpp_recv);
+	return mad_recv_wc;
+}
+
+static inline u64 get_seg_addr(struct ib_mad_send_wr_private *mad_send_wr)
+{
+	return mad_send_wr->sg_list[0].addr + mad_send_wr->data_offset +
+	       (sizeof(struct ib_rmpp_mad) - mad_send_wr->data_offset) *
+	       (mad_send_wr->seg_num - 1);
+}
+
+static int send_next_seg(struct ib_mad_send_wr_private *mad_send_wr)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+	int timeout;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+	ib_set_rmpp_flags(&rmpp_mad->rmpp_hdr, IB_MGMT_RMPP_FLAG_ACTIVE);
+	rmpp_mad->rmpp_hdr.seg_num = cpu_to_be32(mad_send_wr->seg_num);
+
+	if (mad_send_wr->seg_num == 1) {
+		rmpp_mad->rmpp_hdr.rmpp_rtime_flags |= IB_MGMT_RMPP_FLAG_FIRST;
+		rmpp_mad->rmpp_hdr.paylen_newwin =
+			cpu_to_be32(mad_send_wr->total_seg *
+				    (sizeof(struct ib_rmpp_mad) -
+				       offsetof(struct ib_rmpp_mad, data)));
+		mad_send_wr->sg_list[0].length = sizeof(struct ib_rmpp_mad);
+	} else {
+		mad_send_wr->send_wr.num_sge = 2;
+		mad_send_wr->sg_list[0].length = mad_send_wr->data_offset;
+		mad_send_wr->sg_list[1].addr = get_seg_addr(mad_send_wr);
+		mad_send_wr->sg_list[1].length = sizeof(struct ib_rmpp_mad) -
+						 mad_send_wr->data_offset;
+		mad_send_wr->sg_list[1].lkey = mad_send_wr->sg_list[0].lkey;
+	}
+
+	if (mad_send_wr->seg_num == mad_send_wr->total_seg) {
+		rmpp_mad->rmpp_hdr.rmpp_rtime_flags |= IB_MGMT_RMPP_FLAG_LAST;
+		rmpp_mad->rmpp_hdr.paylen_newwin =
+			cpu_to_be32(sizeof(struct ib_rmpp_mad) -
+				    offsetof(struct ib_rmpp_mad, data) -
+				    mad_send_wr->pad);
+	}
+
+	/* 2 seconds for an ACK until we can find the packet lifetime */
+	timeout = mad_send_wr->send_wr.wr.ud.timeout_ms;
+	if (!timeout || timeout > 2000)
+		mad_send_wr->timeout = msecs_to_jiffies(2000);
+	mad_send_wr->seg_num++;
+	return ib_send_mad(mad_send_wr);
+}
+
+static void process_rmpp_ack(struct ib_mad_agent_private *agent,
+			     struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct ib_mad_send_wr_private *mad_send_wr;
+	struct ib_rmpp_mad *rmpp_mad;
+	unsigned long flags;
+	int seg_num, newwin, ret;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
+	if (rmpp_mad->rmpp_hdr.rmpp_status)
+		return;
+
+	seg_num = be32_to_cpu(rmpp_mad->rmpp_hdr.seg_num);
+	newwin = be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
+
+	spin_lock_irqsave(&agent->lock, flags);
+	mad_send_wr = ib_find_send_mad(agent, rmpp_mad->mad_hdr.tid);
+	if (!mad_send_wr)
+		goto out;	/* Unmatched ACK */
+
+	if ((mad_send_wr->last_ack == mad_send_wr->total_seg) ||
+	    (!mad_send_wr->timeout) || (mad_send_wr->status != IB_WC_SUCCESS))
+		goto out;	/* Send is already done */
+
+	if (seg_num > mad_send_wr->total_seg)
+		goto out;	/* Bad ACK */
+
+	if (newwin < mad_send_wr->newwin || seg_num < mad_send_wr->last_ack)
+		goto out;	/* Old ACK */
+
+	if (seg_num > mad_send_wr->last_ack) {
+		mad_send_wr->last_ack = seg_num;
+		mad_send_wr->retries = mad_send_wr->send_wr.wr.ud.retries;
+	}
+	mad_send_wr->newwin = newwin;
+	if (mad_send_wr->last_ack == mad_send_wr->total_seg) {
+		/* If no response is expected, the ACK completes the send */
+		if (!mad_send_wr->send_wr.wr.ud.timeout_ms) {
+			struct ib_mad_send_wc wc;
+
+			ib_mark_mad_done(mad_send_wr);
+			spin_unlock_irqrestore(&agent->lock, flags);
+
+			wc.status = IB_WC_SUCCESS;
+			wc.vendor_err = 0;
+			wc.wr_id = mad_send_wr->wr_id;
+			ib_mad_complete_send_wr(mad_send_wr, &wc);
+			return;
+		}
+		if (mad_send_wr->refcount == 1)
+			ib_reset_mad_timeout(mad_send_wr, mad_send_wr->
+					     send_wr.wr.ud.timeout_ms);
+	} else if (mad_send_wr->refcount == 1 &&
+		   mad_send_wr->seg_num < mad_send_wr->newwin &&
+		   mad_send_wr->seg_num <= mad_send_wr->total_seg) {
+		/* Send failure will just result in a timeout/retry */
+		ret = send_next_seg(mad_send_wr);
+		if (ret)
+			goto out;
+
+		mad_send_wr->refcount++;
+		list_del(&mad_send_wr->agent_list);
+		list_add_tail(&mad_send_wr->agent_list,
+			      &mad_send_wr->mad_agent_priv->send_list);
+	}
+out:
+	spin_unlock_irqrestore(&agent->lock, flags);
+}
+
+struct ib_mad_recv_wc *
+ib_process_rmpp_recv_wc(struct ib_mad_agent_private *agent,
+			struct ib_mad_recv_wc *mad_recv_wc)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_recv_wc->recv_buf.mad;
+	if (!(rmpp_mad->rmpp_hdr.rmpp_rtime_flags & IB_MGMT_RMPP_FLAG_ACTIVE))
+		return mad_recv_wc;
+
+	if (rmpp_mad->rmpp_hdr.rmpp_version != IB_MGMT_RMPP_VERSION)
+		goto out;
+
+	switch (rmpp_mad->rmpp_hdr.rmpp_type) {
+	case IB_MGMT_RMPP_TYPE_DATA:
+		if (rmpp_mad->rmpp_hdr.seg_num == __constant_htonl(1))
+			return start_rmpp(agent, mad_recv_wc);
+		else
+			return continue_rmpp(agent, mad_recv_wc);
+	case IB_MGMT_RMPP_TYPE_ACK:
+		process_rmpp_ack(agent, mad_recv_wc);
+		break;
+	case IB_MGMT_RMPP_TYPE_STOP:
+	case IB_MGMT_RMPP_TYPE_ABORT:
+		/* TODO: process_rmpp_nack(agent, mad_recv_wc); */
+		break;
+	default:
+		break;
+	}
+out:
+	ib_free_recv_mad(mad_recv_wc);
+	return NULL;
+}
+
+int ib_send_rmpp_mad(struct ib_mad_send_wr_private *mad_send_wr)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+	int i, total_len, ret;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+	if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+	      IB_MGMT_RMPP_FLAG_ACTIVE))
+		return IB_RMPP_RESULT_UNHANDLED;
+
+	if (rmpp_mad->rmpp_hdr.rmpp_type != IB_MGMT_RMPP_TYPE_DATA)
+		return IB_RMPP_RESULT_INTERNAL;
+
+	if (mad_send_wr->send_wr.num_sge > 1)
+		return -EINVAL;		/* TODO: support num_sge > 1 */
+
+	mad_send_wr->seg_num = 1;
+	mad_send_wr->newwin = 1;
+	mad_send_wr->data_offset = data_offset(rmpp_mad->mad_hdr.mgmt_class);
+
+	total_len = 0;
+	for (i = 0; i < mad_send_wr->send_wr.num_sge; i++)
+		total_len += mad_send_wr->send_wr.sg_list[i].length;
+
+        mad_send_wr->total_seg = (total_len - mad_send_wr->data_offset) /
+			(sizeof(struct ib_rmpp_mad) - mad_send_wr->data_offset);
+	mad_send_wr->pad = total_len - offsetof(struct ib_rmpp_mad, data) -
+			   be32_to_cpu(rmpp_mad->rmpp_hdr.paylen_newwin);
+
+	/* We need to wait for the final ACK even if there isn't a response */
+	mad_send_wr->refcount += (mad_send_wr->timeout == 0);
+	ret = send_next_seg(mad_send_wr);
+	if (!ret)
+		return IB_RMPP_RESULT_CONSUMED;
+	return ret;
+}
+
+int ib_process_rmpp_send_wc(struct ib_mad_send_wr_private *mad_send_wr,
+			    struct ib_mad_send_wc *mad_send_wc)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+	struct ib_mad_send_buf *msg;
+	int ret;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+	if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+	      IB_MGMT_RMPP_FLAG_ACTIVE))
+		return IB_RMPP_RESULT_UNHANDLED; /* RMPP not active */
+
+	if (rmpp_mad->rmpp_hdr.rmpp_type != IB_MGMT_RMPP_TYPE_DATA) {
+		msg = (struct ib_mad_send_buf *) (unsigned long)
+		      mad_send_wc->wr_id;
+		ib_free_send_mad(msg);
+		return IB_RMPP_RESULT_INTERNAL;	 /* ACK, STOP, or ABORT */
+	}
+
+	if (mad_send_wc->status != IB_WC_SUCCESS ||
+	    mad_send_wr->status != IB_WC_SUCCESS)
+		return IB_RMPP_RESULT_PROCESSED; /* Canceled or send error */
+
+	if (!mad_send_wr->timeout)
+		return IB_RMPP_RESULT_PROCESSED; /* Response received */
+
+	if (mad_send_wr->last_ack == mad_send_wr->total_seg) {
+		mad_send_wr->timeout =
+			msecs_to_jiffies(mad_send_wr->send_wr.wr.ud.timeout_ms);
+		return IB_RMPP_RESULT_PROCESSED; /* Send done */
+	}
+
+	if (mad_send_wr->seg_num > mad_send_wr->newwin ||
+	    mad_send_wr->seg_num > mad_send_wr->total_seg)
+		return IB_RMPP_RESULT_PROCESSED; /* Wait for ACK */
+
+	ret = send_next_seg(mad_send_wr);
+	if (ret) {
+		mad_send_wc->status = IB_WC_GENERAL_ERR;
+		return IB_RMPP_RESULT_PROCESSED;
+	}
+	return IB_RMPP_RESULT_CONSUMED;
+}
+
+int ib_retry_rmpp(struct ib_mad_send_wr_private *mad_send_wr)
+{
+	struct ib_rmpp_mad *rmpp_mad;
+	int ret;
+
+	rmpp_mad = (struct ib_rmpp_mad *)mad_send_wr->send_wr.wr.ud.mad_hdr;
+	if (!(ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) &
+	      IB_MGMT_RMPP_FLAG_ACTIVE))
+		return IB_RMPP_RESULT_UNHANDLED; /* RMPP not active */
+
+	if (mad_send_wr->last_ack == mad_send_wr->total_seg)
+		return IB_RMPP_RESULT_PROCESSED;
+
+	mad_send_wr->seg_num = mad_send_wr->last_ack + 1;
+	ret = send_next_seg(mad_send_wr);
+	if (ret)
+		return IB_RMPP_RESULT_PROCESSED;
+
+	return IB_RMPP_RESULT_CONSUMED;
+}
diff --git a/drivers/infiniband/core/mad_rmpp.h b/drivers/infiniband/core/mad_rmpp.h
new file mode 100644
index 000000000000..c4924dfb8e75
--- /dev/null
+++ b/drivers/infiniband/core/mad_rmpp.h
@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2005 Intel 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.
+ *
+ * $Id: mad_rmpp.h 1921 2005-02-25 22:58:44Z sean.hefty $
+ */
+
+#ifndef __MAD_RMPP_H__
+#define __MAD_RMPP_H__
+
+enum {
+	IB_RMPP_RESULT_PROCESSED,
+	IB_RMPP_RESULT_CONSUMED,
+	IB_RMPP_RESULT_INTERNAL,
+	IB_RMPP_RESULT_UNHANDLED
+};
+
+int ib_send_rmpp_mad(struct ib_mad_send_wr_private *mad_send_wr);
+
+struct ib_mad_recv_wc *
+ib_process_rmpp_recv_wc(struct ib_mad_agent_private *agent,
+			struct ib_mad_recv_wc *mad_recv_wc);
+
+int ib_process_rmpp_send_wc(struct ib_mad_send_wr_private *mad_send_wr,
+			    struct ib_mad_send_wc *mad_send_wc);
+
+void ib_cancel_rmpp_recvs(struct ib_mad_agent_private *agent);
+
+int ib_retry_rmpp(struct ib_mad_send_wr_private *mad_send_wr);
+
+#endif	/* __MAD_RMPP_H__ */
diff --git a/drivers/infiniband/core/sa_query.c b/drivers/infiniband/core/sa_query.c
index 5a08e81fa827..795184931c83 100644
--- a/drivers/infiniband/core/sa_query.c
+++ b/drivers/infiniband/core/sa_query.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, 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
@@ -29,7 +30,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: sa_query.c 1389 2004-12-27 22:56:47Z roland $
+ * $Id: sa_query.c 2811 2005-07-06 18:11:43Z halr $
  */
 
 #include <linux/module.h>
@@ -50,26 +51,6 @@ MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("InfiniBand subnet administration query support");
 MODULE_LICENSE("Dual BSD/GPL");
 
-/*
- * These two structures must be packed because they have 64-bit fields
- * that are only 32-bit aligned.  64-bit architectures will lay them
- * out wrong otherwise.  (And unfortunately they are sent on the wire
- * so we can't change the layout)
- */
-struct ib_sa_hdr {
-	u64			sm_key;
-	u16			attr_offset;
-	u16			reserved;
-	ib_sa_comp_mask		comp_mask;
-} __attribute__ ((packed));
-
-struct ib_sa_mad {
-	struct ib_mad_hdr	mad_hdr;
-	struct ib_rmpp_hdr	rmpp_hdr;
-	struct ib_sa_hdr	sa_hdr;
-	u8			data[200];
-} __attribute__ ((packed));
-
 struct ib_sa_sm_ah {
 	struct ib_ah        *ah;
 	struct kref          ref;
@@ -77,7 +58,6 @@ struct ib_sa_sm_ah {
 
 struct ib_sa_port {
 	struct ib_mad_agent *agent;
-	struct ib_mr        *mr;
 	struct ib_sa_sm_ah  *sm_ah;
 	struct work_struct   update_task;
 	spinlock_t           ah_lock;
@@ -100,6 +80,12 @@ struct ib_sa_query {
 	int                 id;
 };
 
+struct ib_sa_service_query {
+	void (*callback)(int, struct ib_sa_service_rec *, void *);
+	void *context;
+	struct ib_sa_query sa_query;
+};
+
 struct ib_sa_path_query {
 	void (*callback)(int, struct ib_sa_path_rec *, void *);
 	void *context;
@@ -341,6 +327,54 @@ static const struct ib_field mcmember_rec_table[] = {
 	  .size_bits    = 23 },
 };
 
+#define SERVICE_REC_FIELD(field) \
+	.struct_offset_bytes = offsetof(struct ib_sa_service_rec, field),	\
+	.struct_size_bytes   = sizeof ((struct ib_sa_service_rec *) 0)->field,	\
+	.field_name          = "sa_service_rec:" #field
+
+static const struct ib_field service_rec_table[] = {
+	{ SERVICE_REC_FIELD(id),
+	  .offset_words = 0,
+	  .offset_bits  = 0,
+	  .size_bits    = 64 },
+	{ SERVICE_REC_FIELD(gid),
+	  .offset_words = 2,
+	  .offset_bits  = 0,
+	  .size_bits    = 128 },
+	{ SERVICE_REC_FIELD(pkey),
+	  .offset_words = 6,
+	  .offset_bits  = 0,
+	  .size_bits    = 16 },
+	{ SERVICE_REC_FIELD(lease),
+	  .offset_words = 7,
+	  .offset_bits  = 0,
+	  .size_bits    = 32 },
+	{ SERVICE_REC_FIELD(key),
+	  .offset_words = 8,
+	  .offset_bits  = 0,
+	  .size_bits    = 128 },
+	{ SERVICE_REC_FIELD(name),
+	  .offset_words = 12,
+	  .offset_bits  = 0,
+	  .size_bits    = 64*8 },
+	{ SERVICE_REC_FIELD(data8),
+	  .offset_words = 28,
+	  .offset_bits  = 0,
+	  .size_bits    = 16*8 },
+	{ SERVICE_REC_FIELD(data16),
+	  .offset_words = 32,
+	  .offset_bits  = 0,
+	  .size_bits    = 8*16 },
+	{ SERVICE_REC_FIELD(data32),
+	  .offset_words = 36,
+	  .offset_bits  = 0,
+	  .size_bits    = 4*32 },
+	{ SERVICE_REC_FIELD(data64),
+	  .offset_words = 40,
+	  .offset_bits  = 0,
+	  .size_bits    = 2*64 },
+};
+
 static void free_sm_ah(struct kref *kref)
 {
 	struct ib_sa_sm_ah *sm_ah = container_of(kref, struct ib_sa_sm_ah, ref);
@@ -463,7 +497,7 @@ static int send_mad(struct ib_sa_query *query, int timeout_ms)
 				 .mad_hdr     = &query->mad->mad_hdr,
 				 .remote_qpn  = 1,
 				 .remote_qkey = IB_QP1_QKEY,
-				 .timeout_ms  = timeout_ms
+				 .timeout_ms  = timeout_ms,
 			 }
 		 }
 	};
@@ -492,7 +526,7 @@ retry:
 					    sizeof (struct ib_sa_mad),
 					    DMA_TO_DEVICE);
 	gather_list.length = sizeof (struct ib_sa_mad);
-	gather_list.lkey   = port->mr->lkey;
+	gather_list.lkey   = port->agent->mr->lkey;
 	pci_unmap_addr_set(query, mapping, gather_list.addr);
 
 	ret = ib_post_send_mad(port->agent, &wr, &bad_wr);
@@ -566,7 +600,7 @@ static void ib_sa_path_rec_release(struct ib_sa_query *sa_query)
 int ib_sa_path_rec_get(struct ib_device *device, u8 port_num,
 		       struct ib_sa_path_rec *rec,
 		       ib_sa_comp_mask comp_mask,
-		       int timeout_ms, int gfp_mask,
+		       int timeout_ms, unsigned int __nocast gfp_mask,
 		       void (*callback)(int status,
 					struct ib_sa_path_rec *resp,
 					void *context),
@@ -616,6 +650,114 @@ int ib_sa_path_rec_get(struct ib_device *device, u8 port_num,
 }
 EXPORT_SYMBOL(ib_sa_path_rec_get);
 
+static void ib_sa_service_rec_callback(struct ib_sa_query *sa_query,
+				    int status,
+				    struct ib_sa_mad *mad)
+{
+	struct ib_sa_service_query *query =
+		container_of(sa_query, struct ib_sa_service_query, sa_query);
+
+	if (mad) {
+		struct ib_sa_service_rec rec;
+
+		ib_unpack(service_rec_table, ARRAY_SIZE(service_rec_table),
+			  mad->data, &rec);
+		query->callback(status, &rec, query->context);
+	} else
+		query->callback(status, NULL, query->context);
+}
+
+static void ib_sa_service_rec_release(struct ib_sa_query *sa_query)
+{
+	kfree(sa_query->mad);
+	kfree(container_of(sa_query, struct ib_sa_service_query, sa_query));
+}
+
+/**
+ * ib_sa_service_rec_query - Start Service Record operation
+ * @device:device to send request on
+ * @port_num: port number to send request on
+ * @method:SA method - should be get, set, or delete
+ * @rec:Service Record to send in request
+ * @comp_mask:component mask to send in request
+ * @timeout_ms:time to wait for response
+ * @gfp_mask:GFP mask to use for internal allocations
+ * @callback:function called when request completes, times out or is
+ * canceled
+ * @context:opaque user context passed to callback
+ * @sa_query:request context, used to cancel request
+ *
+ * Send a Service Record set/get/delete to the SA to register,
+ * unregister or query a service record.
+ * The callback function will be called when the request completes (or
+ * fails); status is 0 for a successful response, -EINTR if the query
+ * is canceled, -ETIMEDOUT is the query timed out, or -EIO if an error
+ * occurred sending the query.  The resp parameter of the callback is
+ * only valid if status is 0.
+ *
+ * If the return value of ib_sa_service_rec_query() is negative, it is an
+ * error code.  Otherwise it is a request ID that can be used to cancel
+ * the query.
+ */
+int ib_sa_service_rec_query(struct ib_device *device, u8 port_num, u8 method,
+			    struct ib_sa_service_rec *rec,
+			    ib_sa_comp_mask comp_mask,
+			    int timeout_ms, unsigned int __nocast gfp_mask,
+			    void (*callback)(int status,
+					     struct ib_sa_service_rec *resp,
+					     void *context),
+			    void *context,
+			    struct ib_sa_query **sa_query)
+{
+	struct ib_sa_service_query *query;
+	struct ib_sa_device *sa_dev = ib_get_client_data(device, &sa_client);
+	struct ib_sa_port   *port   = &sa_dev->port[port_num - sa_dev->start_port];
+	struct ib_mad_agent *agent  = port->agent;
+	int ret;
+
+	if (method != IB_MGMT_METHOD_GET &&
+	    method != IB_MGMT_METHOD_SET &&
+	    method != IB_SA_METHOD_DELETE)
+		return -EINVAL;
+
+	query = kmalloc(sizeof *query, gfp_mask);
+	if (!query)
+		return -ENOMEM;
+	query->sa_query.mad = kmalloc(sizeof *query->sa_query.mad, gfp_mask);
+	if (!query->sa_query.mad) {
+		kfree(query);
+		return -ENOMEM;
+	}
+
+	query->callback = callback;
+	query->context  = context;
+
+	init_mad(query->sa_query.mad, agent);
+
+	query->sa_query.callback              = callback ? ib_sa_service_rec_callback : NULL;
+	query->sa_query.release               = ib_sa_service_rec_release;
+	query->sa_query.port                  = port;
+	query->sa_query.mad->mad_hdr.method   = method;
+	query->sa_query.mad->mad_hdr.attr_id  =
+				cpu_to_be16(IB_SA_ATTR_SERVICE_REC);
+	query->sa_query.mad->sa_hdr.comp_mask = comp_mask;
+
+	ib_pack(service_rec_table, ARRAY_SIZE(service_rec_table),
+		rec, query->sa_query.mad->data);
+
+	*sa_query = &query->sa_query;
+
+	ret = send_mad(&query->sa_query, timeout_ms);
+	if (ret < 0) {
+		*sa_query = NULL;
+		kfree(query->sa_query.mad);
+		kfree(query);
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL(ib_sa_service_rec_query);
+
 static void ib_sa_mcmember_rec_callback(struct ib_sa_query *sa_query,
 					int status,
 					struct ib_sa_mad *mad)
@@ -643,7 +785,7 @@ int ib_sa_mcmember_rec_query(struct ib_device *device, u8 port_num,
 			     u8 method,
 			     struct ib_sa_mcmember_rec *rec,
 			     ib_sa_comp_mask comp_mask,
-			     int timeout_ms, int gfp_mask,
+			     int timeout_ms, unsigned int __nocast gfp_mask,
 			     void (*callback)(int status,
 					      struct ib_sa_mcmember_rec *resp,
 					      void *context),
@@ -780,7 +922,6 @@ static void ib_sa_add_one(struct ib_device *device)
 	sa_dev->end_port   = e;
 
 	for (i = 0; i <= e - s; ++i) {
-		sa_dev->port[i].mr       = NULL;
 		sa_dev->port[i].sm_ah    = NULL;
 		sa_dev->port[i].port_num = i + s;
 		spin_lock_init(&sa_dev->port[i].ah_lock);
@@ -792,13 +933,6 @@ static void ib_sa_add_one(struct ib_device *device)
 		if (IS_ERR(sa_dev->port[i].agent))
 			goto err;
 
-		sa_dev->port[i].mr = ib_get_dma_mr(sa_dev->port[i].agent->qp->pd,
-						   IB_ACCESS_LOCAL_WRITE);
-		if (IS_ERR(sa_dev->port[i].mr)) {
-			ib_unregister_mad_agent(sa_dev->port[i].agent);
-			goto err;
-		}
-
 		INIT_WORK(&sa_dev->port[i].update_task,
 			  update_sm_ah, &sa_dev->port[i]);
 	}
@@ -822,10 +956,8 @@ static void ib_sa_add_one(struct ib_device *device)
 	return;
 
 err:
-	while (--i >= 0) {
-		ib_dereg_mr(sa_dev->port[i].mr);
+	while (--i >= 0)
 		ib_unregister_mad_agent(sa_dev->port[i].agent);
-	}
 
 	kfree(sa_dev);
 
diff --git a/drivers/infiniband/core/ucm.c b/drivers/infiniband/core/ucm.c
new file mode 100644
index 000000000000..546ec61c407f
--- /dev/null
+++ b/drivers/infiniband/core/ucm.c
@@ -0,0 +1,1393 @@
+/*
+ * Copyright (c) 2005 Topspin Communications.  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.
+ *
+ * $Id: ucm.c 2594 2005-06-13 19:46:02Z libor $
+ */
+#include <linux/init.h>
+#include <linux/fs.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/poll.h>
+#include <linux/file.h>
+#include <linux/mount.h>
+#include <linux/cdev.h>
+
+#include <asm/uaccess.h>
+
+#include "ucm.h"
+
+MODULE_AUTHOR("Libor Michalek");
+MODULE_DESCRIPTION("InfiniBand userspace Connection Manager access");
+MODULE_LICENSE("Dual BSD/GPL");
+
+enum {
+	IB_UCM_MAJOR = 231,
+	IB_UCM_MINOR = 255
+};
+
+#define IB_UCM_DEV MKDEV(IB_UCM_MAJOR, IB_UCM_MINOR)
+
+static struct semaphore ctx_id_mutex;
+static struct idr       ctx_id_table;
+static int              ctx_id_rover = 0;
+
+static struct ib_ucm_context *ib_ucm_ctx_get(int id)
+{
+	struct ib_ucm_context *ctx;
+
+	down(&ctx_id_mutex);
+	ctx = idr_find(&ctx_id_table, id);
+	if (ctx)
+		ctx->ref++;
+	up(&ctx_id_mutex);
+
+	return ctx;
+}
+
+static void ib_ucm_ctx_put(struct ib_ucm_context *ctx)
+{
+	struct ib_ucm_event *uevent;
+
+	down(&ctx_id_mutex);
+
+	ctx->ref--;
+	if (!ctx->ref)
+		idr_remove(&ctx_id_table, ctx->id);
+
+	up(&ctx_id_mutex);
+
+	if (ctx->ref)
+		return;
+
+	down(&ctx->file->mutex);
+
+	list_del(&ctx->file_list);
+	while (!list_empty(&ctx->events)) {
+
+		uevent = list_entry(ctx->events.next,
+				    struct ib_ucm_event, ctx_list);
+		list_del(&uevent->file_list);
+		list_del(&uevent->ctx_list);
+
+		/* clear incoming connections. */
+		if (uevent->cm_id)
+			ib_destroy_cm_id(uevent->cm_id);
+
+		kfree(uevent);
+	}
+
+	up(&ctx->file->mutex);
+
+	printk(KERN_ERR "UCM: Destroyed CM ID <%d>\n", ctx->id);
+
+	ib_destroy_cm_id(ctx->cm_id);
+	kfree(ctx);
+}
+
+static struct ib_ucm_context *ib_ucm_ctx_alloc(struct ib_ucm_file *file)
+{
+	struct ib_ucm_context *ctx;
+	int result;
+
+	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return NULL;
+
+	ctx->ref  = 1; /* user reference */
+	ctx->file = file;
+
+	INIT_LIST_HEAD(&ctx->events);
+	init_MUTEX(&ctx->mutex);
+
+	list_add_tail(&ctx->file_list, &file->ctxs);
+
+	ctx_id_rover = (ctx_id_rover + 1) & INT_MAX;
+retry:
+	result = idr_pre_get(&ctx_id_table, GFP_KERNEL);
+	if (!result)
+		goto error;
+
+	down(&ctx_id_mutex);
+	result = idr_get_new_above(&ctx_id_table, ctx, ctx_id_rover, &ctx->id);
+	up(&ctx_id_mutex);
+
+	if (result == -EAGAIN)
+		goto retry;
+	if (result)
+		goto error;
+
+	printk(KERN_ERR "UCM: Allocated CM ID <%d>\n", ctx->id);
+
+	return ctx;
+error:
+	list_del(&ctx->file_list);
+	kfree(ctx);
+
+	return NULL;
+}
+/*
+ * Event portion of the API, handle CM events
+ * and allow event polling.
+ */
+static void ib_ucm_event_path_get(struct ib_ucm_path_rec *upath,
+				  struct ib_sa_path_rec	 *kpath)
+{
+	if (!kpath || !upath)
+		return;
+
+	memcpy(upath->dgid, kpath->dgid.raw, sizeof(union ib_gid));
+	memcpy(upath->sgid, kpath->sgid.raw, sizeof(union ib_gid));
+
+	upath->dlid             = kpath->dlid;
+	upath->slid             = kpath->slid;
+	upath->raw_traffic      = kpath->raw_traffic;
+	upath->flow_label       = kpath->flow_label;
+	upath->hop_limit        = kpath->hop_limit;
+	upath->traffic_class    = kpath->traffic_class;
+	upath->reversible       = kpath->reversible;
+	upath->numb_path        = kpath->numb_path;
+	upath->pkey             = kpath->pkey;
+	upath->sl	        = kpath->sl;
+	upath->mtu_selector     = kpath->mtu_selector;
+	upath->mtu              = kpath->mtu;
+	upath->rate_selector    = kpath->rate_selector;
+	upath->rate             = kpath->rate;
+	upath->packet_life_time = kpath->packet_life_time;
+	upath->preference       = kpath->preference;
+
+	upath->packet_life_time_selector =
+		kpath->packet_life_time_selector;
+}
+
+static void ib_ucm_event_req_get(struct ib_ucm_req_event_resp *ureq,
+				 struct ib_cm_req_event_param *kreq)
+{
+	ureq->listen_id = (long)kreq->listen_id->context;
+
+	ureq->remote_ca_guid             = kreq->remote_ca_guid;
+	ureq->remote_qkey                = kreq->remote_qkey;
+	ureq->remote_qpn                 = kreq->remote_qpn;
+	ureq->qp_type                    = kreq->qp_type;
+	ureq->starting_psn               = kreq->starting_psn;
+	ureq->responder_resources        = kreq->responder_resources;
+	ureq->initiator_depth            = kreq->initiator_depth;
+	ureq->local_cm_response_timeout  = kreq->local_cm_response_timeout;
+	ureq->flow_control               = kreq->flow_control;
+	ureq->remote_cm_response_timeout = kreq->remote_cm_response_timeout;
+	ureq->retry_count                = kreq->retry_count;
+	ureq->rnr_retry_count            = kreq->rnr_retry_count;
+	ureq->srq                        = kreq->srq;
+
+	ib_ucm_event_path_get(&ureq->primary_path, kreq->primary_path);
+	ib_ucm_event_path_get(&ureq->alternate_path, kreq->alternate_path);
+}
+
+static void ib_ucm_event_rep_get(struct ib_ucm_rep_event_resp *urep,
+				 struct ib_cm_rep_event_param *krep)
+{
+	urep->remote_ca_guid      = krep->remote_ca_guid;
+	urep->remote_qkey         = krep->remote_qkey;
+	urep->remote_qpn          = krep->remote_qpn;
+	urep->starting_psn        = krep->starting_psn;
+	urep->responder_resources = krep->responder_resources;
+	urep->initiator_depth     = krep->initiator_depth;
+	urep->target_ack_delay    = krep->target_ack_delay;
+	urep->failover_accepted   = krep->failover_accepted;
+	urep->flow_control        = krep->flow_control;
+	urep->rnr_retry_count     = krep->rnr_retry_count;
+	urep->srq                 = krep->srq;
+}
+
+static void ib_ucm_event_rej_get(struct ib_ucm_rej_event_resp *urej,
+				 struct ib_cm_rej_event_param *krej)
+{
+	urej->reason = krej->reason;
+}
+
+static void ib_ucm_event_mra_get(struct ib_ucm_mra_event_resp *umra,
+				 struct ib_cm_mra_event_param *kmra)
+{
+	umra->timeout = kmra->service_timeout;
+}
+
+static void ib_ucm_event_lap_get(struct ib_ucm_lap_event_resp *ulap,
+				 struct ib_cm_lap_event_param *klap)
+{
+	ib_ucm_event_path_get(&ulap->path, klap->alternate_path);
+}
+
+static void ib_ucm_event_apr_get(struct ib_ucm_apr_event_resp *uapr,
+				 struct ib_cm_apr_event_param *kapr)
+{
+	uapr->status = kapr->ap_status;
+}
+
+static void ib_ucm_event_sidr_req_get(struct ib_ucm_sidr_req_event_resp *ureq,
+				      struct ib_cm_sidr_req_event_param *kreq)
+{
+	ureq->listen_id = (long)kreq->listen_id->context;
+	ureq->pkey      = kreq->pkey;
+}
+
+static void ib_ucm_event_sidr_rep_get(struct ib_ucm_sidr_rep_event_resp *urep,
+				      struct ib_cm_sidr_rep_event_param *krep)
+{
+	urep->status = krep->status;
+	urep->qkey   = krep->qkey;
+	urep->qpn    = krep->qpn;
+};
+
+static int ib_ucm_event_process(struct ib_cm_event *evt,
+				struct ib_ucm_event *uvt)
+{
+	void *info = NULL;
+	int result;
+
+	switch (evt->event) {
+	case IB_CM_REQ_RECEIVED:
+		ib_ucm_event_req_get(&uvt->resp.u.req_resp,
+				     &evt->param.req_rcvd);
+		uvt->data_len      = IB_CM_REQ_PRIVATE_DATA_SIZE;
+		uvt->resp.present |= (evt->param.req_rcvd.primary_path ?
+				      IB_UCM_PRES_PRIMARY : 0);
+		uvt->resp.present |= (evt->param.req_rcvd.alternate_path ?
+				      IB_UCM_PRES_ALTERNATE : 0);
+		break;
+	case IB_CM_REP_RECEIVED:
+		ib_ucm_event_rep_get(&uvt->resp.u.rep_resp,
+				     &evt->param.rep_rcvd);
+		uvt->data_len = IB_CM_REP_PRIVATE_DATA_SIZE;
+
+		break;
+	case IB_CM_RTU_RECEIVED:
+		uvt->data_len = IB_CM_RTU_PRIVATE_DATA_SIZE;
+		uvt->resp.u.send_status = evt->param.send_status;
+
+		break;
+	case IB_CM_DREQ_RECEIVED:
+		uvt->data_len = IB_CM_DREQ_PRIVATE_DATA_SIZE;
+		uvt->resp.u.send_status = evt->param.send_status;
+
+		break;
+	case IB_CM_DREP_RECEIVED:
+		uvt->data_len = IB_CM_DREP_PRIVATE_DATA_SIZE;
+		uvt->resp.u.send_status = evt->param.send_status;
+
+		break;
+	case IB_CM_MRA_RECEIVED:
+		ib_ucm_event_mra_get(&uvt->resp.u.mra_resp,
+				     &evt->param.mra_rcvd);
+		uvt->data_len = IB_CM_MRA_PRIVATE_DATA_SIZE;
+
+		break;
+	case IB_CM_REJ_RECEIVED:
+		ib_ucm_event_rej_get(&uvt->resp.u.rej_resp,
+				     &evt->param.rej_rcvd);
+		uvt->data_len = IB_CM_REJ_PRIVATE_DATA_SIZE;
+		uvt->info_len = evt->param.rej_rcvd.ari_length;
+		info	      = evt->param.rej_rcvd.ari;
+
+		break;
+	case IB_CM_LAP_RECEIVED:
+		ib_ucm_event_lap_get(&uvt->resp.u.lap_resp,
+				     &evt->param.lap_rcvd);
+		uvt->data_len = IB_CM_LAP_PRIVATE_DATA_SIZE;
+		uvt->resp.present |= (evt->param.lap_rcvd.alternate_path ?
+				      IB_UCM_PRES_ALTERNATE : 0);
+		break;
+	case IB_CM_APR_RECEIVED:
+		ib_ucm_event_apr_get(&uvt->resp.u.apr_resp,
+				     &evt->param.apr_rcvd);
+		uvt->data_len = IB_CM_APR_PRIVATE_DATA_SIZE;
+		uvt->info_len = evt->param.apr_rcvd.info_len;
+		info	      = evt->param.apr_rcvd.apr_info;
+
+		break;
+	case IB_CM_SIDR_REQ_RECEIVED:
+		ib_ucm_event_sidr_req_get(&uvt->resp.u.sidr_req_resp,
+					  &evt->param.sidr_req_rcvd);
+		uvt->data_len = IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE;
+
+		break;
+	case IB_CM_SIDR_REP_RECEIVED:
+		ib_ucm_event_sidr_rep_get(&uvt->resp.u.sidr_rep_resp,
+					  &evt->param.sidr_rep_rcvd);
+		uvt->data_len = IB_CM_SIDR_REP_PRIVATE_DATA_SIZE;
+		uvt->info_len = evt->param.sidr_rep_rcvd.info_len;
+		info	      = evt->param.sidr_rep_rcvd.info;
+
+		break;
+	default:
+		uvt->resp.u.send_status = evt->param.send_status;
+
+		break;
+	}
+
+	if (uvt->data_len && evt->private_data) {
+
+		uvt->data = kmalloc(uvt->data_len, GFP_KERNEL);
+		if (!uvt->data) {
+			result = -ENOMEM;
+			goto error;
+		}
+
+		memcpy(uvt->data, evt->private_data, uvt->data_len);
+		uvt->resp.present |= IB_UCM_PRES_DATA;
+	}
+
+	if (uvt->info_len && info) {
+
+		uvt->info = kmalloc(uvt->info_len, GFP_KERNEL);
+		if (!uvt->info) {
+			result = -ENOMEM;
+			goto error;
+		}
+
+		memcpy(uvt->info, info, uvt->info_len);
+		uvt->resp.present |= IB_UCM_PRES_INFO;
+	}
+
+	return 0;
+error:
+	if (uvt->info)
+		kfree(uvt->info);
+	if (uvt->data)
+		kfree(uvt->data);
+	return result;
+}
+
+static int ib_ucm_event_handler(struct ib_cm_id *cm_id,
+				struct ib_cm_event *event)
+{
+	struct ib_ucm_event *uevent;
+	struct ib_ucm_context *ctx;
+	int result = 0;
+	int id;
+	/*
+	 * lookup correct context based on event type.
+	 */
+	switch (event->event) {
+	case IB_CM_REQ_RECEIVED:
+		id = (long)event->param.req_rcvd.listen_id->context;
+		break;
+	case IB_CM_SIDR_REQ_RECEIVED:
+		id = (long)event->param.sidr_req_rcvd.listen_id->context;
+		break;
+	default:
+		id = (long)cm_id->context;
+		break;
+	}
+
+	printk(KERN_ERR "UCM: Event. CM ID <%d> event <%d>\n",
+	       id, event->event);
+
+	ctx = ib_ucm_ctx_get(id);
+	if (!ctx)
+		return -ENOENT;
+
+	if (event->event == IB_CM_REQ_RECEIVED ||
+	    event->event == IB_CM_SIDR_REQ_RECEIVED)
+		id = IB_UCM_CM_ID_INVALID;
+
+	uevent = kmalloc(sizeof(*uevent), GFP_KERNEL);
+	if (!uevent) {
+		result = -ENOMEM;
+		goto done;
+	}
+
+	memset(uevent, 0, sizeof(*uevent));
+
+	uevent->resp.id    = id;
+	uevent->resp.event = event->event;
+
+	result = ib_ucm_event_process(event, uevent);
+	if (result)
+		goto done;
+
+	uevent->ctx   = ctx;
+	uevent->cm_id = ((event->event == IB_CM_REQ_RECEIVED ||
+			  event->event == IB_CM_SIDR_REQ_RECEIVED ) ?
+			 cm_id : NULL);
+
+	down(&ctx->file->mutex);
+
+	list_add_tail(&uevent->file_list, &ctx->file->events);
+	list_add_tail(&uevent->ctx_list, &ctx->events);
+
+	wake_up_interruptible(&ctx->file->poll_wait);
+
+	up(&ctx->file->mutex);
+done:
+	ctx->error = result;
+	ib_ucm_ctx_put(ctx); /* func reference */
+	return result;
+}
+
+static ssize_t ib_ucm_event(struct ib_ucm_file *file,
+			    const char __user *inbuf,
+			    int in_len, int out_len)
+{
+	struct ib_ucm_context *ctx;
+	struct ib_ucm_event_get cmd;
+	struct ib_ucm_event *uevent = NULL;
+	int result = 0;
+	DEFINE_WAIT(wait);
+
+	if (out_len < sizeof(struct ib_ucm_event_resp))
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+	/*
+	 * wait
+	 */
+	down(&file->mutex);
+
+	while (list_empty(&file->events)) {
+
+		if (file->filp->f_flags & O_NONBLOCK) {
+			result = -EAGAIN;
+			break;
+		}
+
+		if (signal_pending(current)) {
+			result = -ERESTARTSYS;
+			break;
+		}
+
+		prepare_to_wait(&file->poll_wait, &wait, TASK_INTERRUPTIBLE);
+
+		up(&file->mutex);
+		schedule();
+		down(&file->mutex);
+
+		finish_wait(&file->poll_wait, &wait);
+	}
+
+	if (result)
+		goto done;
+
+	uevent = list_entry(file->events.next, struct ib_ucm_event, file_list);
+
+	if (!uevent->cm_id)
+		goto user;
+
+	ctx = ib_ucm_ctx_alloc(file);
+	if (!ctx) {
+		result = -ENOMEM;
+		goto done;
+	}
+
+	ctx->cm_id             = uevent->cm_id;
+	ctx->cm_id->cm_handler = ib_ucm_event_handler;
+	ctx->cm_id->context    = (void *)(unsigned long)ctx->id;
+
+	uevent->resp.id = ctx->id;
+
+user:
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &uevent->resp, sizeof(uevent->resp))) {
+		result = -EFAULT;
+		goto done;
+	}
+
+	if (uevent->data) {
+
+		if (cmd.data_len < uevent->data_len) {
+			result = -ENOMEM;
+			goto done;
+		}
+
+		if (copy_to_user((void __user *)(unsigned long)cmd.data,
+				 uevent->data, uevent->data_len)) {
+			result = -EFAULT;
+			goto done;
+		}
+	}
+
+	if (uevent->info) {
+
+		if (cmd.info_len < uevent->info_len) {
+			result = -ENOMEM;
+			goto done;
+		}
+
+		if (copy_to_user((void __user *)(unsigned long)cmd.info,
+				 uevent->info, uevent->info_len)) {
+			result = -EFAULT;
+			goto done;
+		}
+	}
+
+	list_del(&uevent->file_list);
+	list_del(&uevent->ctx_list);
+
+	if (uevent->data)
+		kfree(uevent->data);
+	if (uevent->info)
+		kfree(uevent->info);
+	kfree(uevent);
+done:
+	up(&file->mutex);
+	return result;
+}
+
+
+static ssize_t ib_ucm_create_id(struct ib_ucm_file *file,
+				const char __user *inbuf,
+				int in_len, int out_len)
+{
+	struct ib_ucm_create_id cmd;
+	struct ib_ucm_create_id_resp resp;
+	struct ib_ucm_context *ctx;
+	int result;
+
+	if (out_len < sizeof(resp))
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ib_ucm_ctx_alloc(file);
+	if (!ctx)
+		return -ENOMEM;
+
+	ctx->cm_id = ib_create_cm_id(ib_ucm_event_handler,
+				     (void *)(unsigned long)ctx->id);
+	if (!ctx->cm_id) {
+		result = -ENOMEM;
+		goto err_cm;
+	}
+
+	resp.id = ctx->id;
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &resp, sizeof(resp))) {
+		result = -EFAULT;
+		goto err_ret;
+	}
+
+	return 0;
+err_ret:
+	ib_destroy_cm_id(ctx->cm_id);
+err_cm:
+	ib_ucm_ctx_put(ctx); /* user reference */
+
+	return result;
+}
+
+static ssize_t ib_ucm_destroy_id(struct ib_ucm_file *file,
+				 const char __user *inbuf,
+				 int in_len, int out_len)
+{
+	struct ib_ucm_destroy_id cmd;
+	struct ib_ucm_context *ctx;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx)
+		return -ENOENT;
+
+	ib_ucm_ctx_put(ctx); /* user reference */
+	ib_ucm_ctx_put(ctx); /* func reference */
+
+	return 0;
+}
+
+static ssize_t ib_ucm_attr_id(struct ib_ucm_file *file,
+			      const char __user *inbuf,
+			      int in_len, int out_len)
+{
+	struct ib_ucm_attr_id_resp resp;
+	struct ib_ucm_attr_id cmd;
+	struct ib_ucm_context *ctx;
+	int result = 0;
+
+	if (out_len < sizeof(resp))
+		return -ENOSPC;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx)
+		return -ENOENT;
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file) {
+		result = -EINVAL;
+		goto done;
+	}
+
+	resp.service_id   = ctx->cm_id->service_id;
+	resp.service_mask = ctx->cm_id->service_mask;
+	resp.local_id     = ctx->cm_id->local_id;
+	resp.remote_id    = ctx->cm_id->remote_id;
+
+	if (copy_to_user((void __user *)(unsigned long)cmd.response,
+			 &resp, sizeof(resp)))
+		result = -EFAULT;
+
+done:
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+	return result;
+}
+
+static ssize_t ib_ucm_listen(struct ib_ucm_file *file,
+			     const char __user *inbuf,
+			     int in_len, int out_len)
+{
+	struct ib_ucm_listen cmd;
+	struct ib_ucm_context *ctx;
+	int result;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx)
+		return -ENOENT;
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_cm_listen(ctx->cm_id, cmd.service_id,
+				      cmd.service_mask);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+	return result;
+}
+
+static ssize_t ib_ucm_establish(struct ib_ucm_file *file,
+				const char __user *inbuf,
+				int in_len, int out_len)
+{
+	struct ib_ucm_establish cmd;
+	struct ib_ucm_context *ctx;
+	int result;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx)
+		return -ENOENT;
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_cm_establish(ctx->cm_id);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+	return result;
+}
+
+static int ib_ucm_alloc_data(const void **dest, u64 src, u32 len)
+{
+	void *data;
+
+	*dest = NULL;
+
+	if (!len)
+		return 0;
+
+	data = kmalloc(len, GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	if (copy_from_user(data, (void __user *)(unsigned long)src, len)) {
+		kfree(data);
+		return -EFAULT;
+	}
+
+	*dest = data;
+	return 0;
+}
+
+static int ib_ucm_path_get(struct ib_sa_path_rec **path, u64 src)
+{
+	struct ib_ucm_path_rec ucm_path;
+	struct ib_sa_path_rec  *sa_path;
+
+	*path = NULL;
+
+	if (!src)
+		return 0;
+
+	sa_path = kmalloc(sizeof(*sa_path), GFP_KERNEL);
+	if (!sa_path)
+		return -ENOMEM;
+
+	if (copy_from_user(&ucm_path, (void __user *)(unsigned long)src,
+			   sizeof(ucm_path))) {
+
+		kfree(sa_path);
+		return -EFAULT;
+	}
+
+	memcpy(sa_path->dgid.raw, ucm_path.dgid, sizeof(union ib_gid));
+	memcpy(sa_path->sgid.raw, ucm_path.sgid, sizeof(union ib_gid));
+
+	sa_path->dlid	          = ucm_path.dlid;
+	sa_path->slid	          = ucm_path.slid;
+	sa_path->raw_traffic      = ucm_path.raw_traffic;
+	sa_path->flow_label       = ucm_path.flow_label;
+	sa_path->hop_limit        = ucm_path.hop_limit;
+	sa_path->traffic_class    = ucm_path.traffic_class;
+	sa_path->reversible       = ucm_path.reversible;
+	sa_path->numb_path        = ucm_path.numb_path;
+	sa_path->pkey             = ucm_path.pkey;
+	sa_path->sl               = ucm_path.sl;
+	sa_path->mtu_selector     = ucm_path.mtu_selector;
+	sa_path->mtu              = ucm_path.mtu;
+	sa_path->rate_selector    = ucm_path.rate_selector;
+	sa_path->rate             = ucm_path.rate;
+	sa_path->packet_life_time = ucm_path.packet_life_time;
+	sa_path->preference       = ucm_path.preference;
+
+	sa_path->packet_life_time_selector =
+		ucm_path.packet_life_time_selector;
+
+	*path = sa_path;
+	return 0;
+}
+
+static ssize_t ib_ucm_send_req(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	struct ib_cm_req_param param;
+	struct ib_ucm_context *ctx;
+	struct ib_ucm_req cmd;
+	int result;
+
+	param.private_data   = NULL;
+	param.primary_path   = NULL;
+	param.alternate_path = NULL;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&param.private_data, cmd.data, cmd.len);
+	if (result)
+		goto done;
+
+	result = ib_ucm_path_get(&param.primary_path, cmd.primary_path);
+	if (result)
+		goto done;
+
+	result = ib_ucm_path_get(&param.alternate_path, cmd.alternate_path);
+	if (result)
+		goto done;
+
+	param.private_data_len           = cmd.len;
+	param.service_id                 = cmd.sid;
+	param.qp_num                     = cmd.qpn;
+	param.qp_type                    = cmd.qp_type;
+	param.starting_psn               = cmd.psn;
+	param.peer_to_peer               = cmd.peer_to_peer;
+	param.responder_resources        = cmd.responder_resources;
+	param.initiator_depth            = cmd.initiator_depth;
+	param.remote_cm_response_timeout = cmd.remote_cm_response_timeout;
+	param.flow_control               = cmd.flow_control;
+	param.local_cm_response_timeout  = cmd.local_cm_response_timeout;
+	param.retry_count                = cmd.retry_count;
+	param.rnr_retry_count            = cmd.rnr_retry_count;
+	param.max_cm_retries             = cmd.max_cm_retries;
+	param.srq                        = cmd.srq;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_send_cm_req(ctx->cm_id, &param);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (param.private_data)
+		kfree(param.private_data);
+	if (param.primary_path)
+		kfree(param.primary_path);
+	if (param.alternate_path)
+		kfree(param.alternate_path);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_rep(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	struct ib_cm_rep_param param;
+	struct ib_ucm_context *ctx;
+	struct ib_ucm_rep cmd;
+	int result;
+
+	param.private_data = NULL;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&param.private_data, cmd.data, cmd.len);
+	if (result)
+		return result;
+
+	param.qp_num              = cmd.qpn;
+	param.starting_psn        = cmd.psn;
+	param.private_data_len    = cmd.len;
+	param.responder_resources = cmd.responder_resources;
+	param.initiator_depth     = cmd.initiator_depth;
+	param.target_ack_delay    = cmd.target_ack_delay;
+	param.failover_accepted   = cmd.failover_accepted;
+	param.flow_control        = cmd.flow_control;
+	param.rnr_retry_count     = cmd.rnr_retry_count;
+	param.srq                 = cmd.srq;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_send_cm_rep(ctx->cm_id, &param);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (param.private_data)
+		kfree(param.private_data);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_private_data(struct ib_ucm_file *file,
+					const char __user *inbuf, int in_len,
+					int (*func)(struct ib_cm_id *cm_id,
+						    const void *private_data,
+						    u8 private_data_len))
+{
+	struct ib_ucm_private_data cmd;
+	struct ib_ucm_context *ctx;
+	const void *private_data = NULL;
+	int result;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&private_data, cmd.data, cmd.len);
+	if (result)
+		return result;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = func(ctx->cm_id, private_data, cmd.len);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (private_data)
+		kfree(private_data);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_rtu(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	return ib_ucm_send_private_data(file, inbuf, in_len, ib_send_cm_rtu);
+}
+
+static ssize_t ib_ucm_send_dreq(struct ib_ucm_file *file,
+				const char __user *inbuf,
+				int in_len, int out_len)
+{
+	return ib_ucm_send_private_data(file, inbuf, in_len, ib_send_cm_dreq);
+}
+
+static ssize_t ib_ucm_send_drep(struct ib_ucm_file *file,
+				const char __user *inbuf,
+				int in_len, int out_len)
+{
+	return ib_ucm_send_private_data(file, inbuf, in_len, ib_send_cm_drep);
+}
+
+static ssize_t ib_ucm_send_info(struct ib_ucm_file *file,
+				const char __user *inbuf, int in_len,
+				int (*func)(struct ib_cm_id *cm_id,
+					    int status,
+					    const void *info,
+					    u8 info_len,
+					    const void *data,
+					    u8 data_len))
+{
+	struct ib_ucm_context *ctx;
+	struct ib_ucm_info cmd;
+	const void *data = NULL;
+	const void *info = NULL;
+	int result;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&data, cmd.data, cmd.data_len);
+	if (result)
+		goto done;
+
+	result = ib_ucm_alloc_data(&info, cmd.info, cmd.info_len);
+	if (result)
+		goto done;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = func(ctx->cm_id, cmd.status,
+			      info, cmd.info_len,
+			      data, cmd.data_len);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (data)
+		kfree(data);
+	if (info)
+		kfree(info);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_rej(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	return ib_ucm_send_info(file, inbuf, in_len, (void *)ib_send_cm_rej);
+}
+
+static ssize_t ib_ucm_send_apr(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	return ib_ucm_send_info(file, inbuf, in_len, (void *)ib_send_cm_apr);
+}
+
+static ssize_t ib_ucm_send_mra(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	struct ib_ucm_context *ctx;
+	struct ib_ucm_mra cmd;
+	const void *data = NULL;
+	int result;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&data, cmd.data, cmd.len);
+	if (result)
+		return result;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_send_cm_mra(ctx->cm_id, cmd.timeout,
+					data, cmd.len);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (data)
+		kfree(data);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_lap(struct ib_ucm_file *file,
+			       const char __user *inbuf,
+			       int in_len, int out_len)
+{
+	struct ib_ucm_context *ctx;
+	struct ib_sa_path_rec *path = NULL;
+	struct ib_ucm_lap cmd;
+	const void *data = NULL;
+	int result;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&data, cmd.data, cmd.len);
+	if (result)
+		goto done;
+
+	result = ib_ucm_path_get(&path, cmd.path);
+	if (result)
+		goto done;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_send_cm_lap(ctx->cm_id, path, data, cmd.len);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (data)
+		kfree(data);
+	if (path)
+		kfree(path);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_sidr_req(struct ib_ucm_file *file,
+				    const char __user *inbuf,
+				    int in_len, int out_len)
+{
+	struct ib_cm_sidr_req_param param;
+	struct ib_ucm_context *ctx;
+	struct ib_ucm_sidr_req cmd;
+	int result;
+
+	param.private_data = NULL;
+	param.path = NULL;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&param.private_data, cmd.data, cmd.len);
+	if (result)
+		goto done;
+
+	result = ib_ucm_path_get(&param.path, cmd.path);
+	if (result)
+		goto done;
+
+	param.private_data_len = cmd.len;
+	param.service_id       = cmd.sid;
+	param.timeout_ms       = cmd.timeout;
+	param.max_cm_retries   = cmd.max_cm_retries;
+	param.pkey             = cmd.pkey;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_send_cm_sidr_req(ctx->cm_id, &param);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (param.private_data)
+		kfree(param.private_data);
+	if (param.path)
+		kfree(param.path);
+
+	return result;
+}
+
+static ssize_t ib_ucm_send_sidr_rep(struct ib_ucm_file *file,
+				    const char __user *inbuf,
+				    int in_len, int out_len)
+{
+	struct ib_cm_sidr_rep_param param;
+	struct ib_ucm_sidr_rep cmd;
+	struct ib_ucm_context *ctx;
+	int result;
+
+	param.info = NULL;
+
+	if (copy_from_user(&cmd, inbuf, sizeof(cmd)))
+		return -EFAULT;
+
+	result = ib_ucm_alloc_data(&param.private_data,
+				   cmd.data, cmd.data_len);
+	if (result)
+		goto done;
+
+	result = ib_ucm_alloc_data(&param.info, cmd.info, cmd.info_len);
+	if (result)
+		goto done;
+
+	param.qp_num	   = cmd.qpn;
+	param.qkey	     = cmd.qkey;
+	param.status	   = cmd.status;
+	param.info_length      = cmd.info_len;
+	param.private_data_len = cmd.data_len;
+
+	ctx = ib_ucm_ctx_get(cmd.id);
+	if (!ctx) {
+		result = -ENOENT;
+		goto done;
+	}
+
+ 	down(&ctx->file->mutex);
+	if (ctx->file != file)
+		result = -EINVAL;
+	else
+		result = ib_send_cm_sidr_rep(ctx->cm_id, &param);
+
+	up(&ctx->file->mutex);
+	ib_ucm_ctx_put(ctx); /* func reference */
+done:
+	if (param.private_data)
+		kfree(param.private_data);
+	if (param.info)
+		kfree(param.info);
+
+	return result;
+}
+
+static ssize_t (*ucm_cmd_table[])(struct ib_ucm_file *file,
+				  const char __user *inbuf,
+				  int in_len, int out_len) = {
+	[IB_USER_CM_CMD_CREATE_ID]     = ib_ucm_create_id,
+	[IB_USER_CM_CMD_DESTROY_ID]    = ib_ucm_destroy_id,
+	[IB_USER_CM_CMD_ATTR_ID]       = ib_ucm_attr_id,
+	[IB_USER_CM_CMD_LISTEN]        = ib_ucm_listen,
+	[IB_USER_CM_CMD_ESTABLISH]     = ib_ucm_establish,
+	[IB_USER_CM_CMD_SEND_REQ]      = ib_ucm_send_req,
+	[IB_USER_CM_CMD_SEND_REP]      = ib_ucm_send_rep,
+	[IB_USER_CM_CMD_SEND_RTU]      = ib_ucm_send_rtu,
+	[IB_USER_CM_CMD_SEND_DREQ]     = ib_ucm_send_dreq,
+	[IB_USER_CM_CMD_SEND_DREP]     = ib_ucm_send_drep,
+	[IB_USER_CM_CMD_SEND_REJ]      = ib_ucm_send_rej,
+	[IB_USER_CM_CMD_SEND_MRA]      = ib_ucm_send_mra,
+	[IB_USER_CM_CMD_SEND_LAP]      = ib_ucm_send_lap,
+	[IB_USER_CM_CMD_SEND_APR]      = ib_ucm_send_apr,
+	[IB_USER_CM_CMD_SEND_SIDR_REQ] = ib_ucm_send_sidr_req,
+	[IB_USER_CM_CMD_SEND_SIDR_REP] = ib_ucm_send_sidr_rep,
+	[IB_USER_CM_CMD_EVENT]	       = ib_ucm_event,
+};
+
+static ssize_t ib_ucm_write(struct file *filp, const char __user *buf,
+			    size_t len, loff_t *pos)
+{
+	struct ib_ucm_file *file = filp->private_data;
+	struct ib_ucm_cmd_hdr hdr;
+	ssize_t result;
+
+	if (len < sizeof(hdr))
+		return -EINVAL;
+
+	if (copy_from_user(&hdr, buf, sizeof(hdr)))
+		return -EFAULT;
+
+	printk(KERN_ERR "UCM: Write. cmd <%d> in <%d> out <%d> len <%Zu>\n",
+	       hdr.cmd, hdr.in, hdr.out, len);
+
+	if (hdr.cmd < 0 || hdr.cmd >= ARRAY_SIZE(ucm_cmd_table))
+		return -EINVAL;
+
+	if (hdr.in + sizeof(hdr) > len)
+		return -EINVAL;
+
+	result = ucm_cmd_table[hdr.cmd](file, buf + sizeof(hdr),
+					hdr.in, hdr.out);
+	if (!result)
+		result = len;
+
+	return result;
+}
+
+static unsigned int ib_ucm_poll(struct file *filp,
+				struct poll_table_struct *wait)
+{
+	struct ib_ucm_file *file = filp->private_data;
+	unsigned int mask = 0;
+
+	poll_wait(filp, &file->poll_wait, wait);
+
+	if (!list_empty(&file->events))
+		mask = POLLIN | POLLRDNORM;
+
+	return mask;
+}
+
+static int ib_ucm_open(struct inode *inode, struct file *filp)
+{
+	struct ib_ucm_file *file;
+
+	file = kmalloc(sizeof(*file), GFP_KERNEL);
+	if (!file)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&file->events);
+	INIT_LIST_HEAD(&file->ctxs);
+	init_waitqueue_head(&file->poll_wait);
+
+	init_MUTEX(&file->mutex);
+
+	filp->private_data = file;
+	file->filp = filp;
+
+	printk(KERN_ERR "UCM: Created struct\n");
+
+	return 0;
+}
+
+static int ib_ucm_close(struct inode *inode, struct file *filp)
+{
+	struct ib_ucm_file *file = filp->private_data;
+	struct ib_ucm_context *ctx;
+
+	down(&file->mutex);
+
+	while (!list_empty(&file->ctxs)) {
+
+		ctx = list_entry(file->ctxs.next,
+				 struct ib_ucm_context, file_list);
+
+		up(&ctx->file->mutex);
+		ib_ucm_ctx_put(ctx); /* user reference */
+		down(&file->mutex);
+	}
+
+	up(&file->mutex);
+
+	kfree(file);
+
+	printk(KERN_ERR "UCM: Deleted struct\n");
+	return 0;
+}
+
+static struct file_operations ib_ucm_fops = {
+	.owner 	 = THIS_MODULE,
+	.open 	 = ib_ucm_open,
+	.release = ib_ucm_close,
+	.write 	 = ib_ucm_write,
+	.poll    = ib_ucm_poll,
+};
+
+
+static struct class *ib_ucm_class;
+static struct cdev	  ib_ucm_cdev;
+
+static int __init ib_ucm_init(void)
+{
+	int result;
+
+	result = register_chrdev_region(IB_UCM_DEV, 1, "infiniband_cm");
+	if (result) {
+		printk(KERN_ERR "UCM: Error <%d> registering dev\n", result);
+		goto err_chr;
+	}
+
+	cdev_init(&ib_ucm_cdev, &ib_ucm_fops);
+
+	result = cdev_add(&ib_ucm_cdev, IB_UCM_DEV, 1);
+	if (result) {
+		printk(KERN_ERR "UCM: Error <%d> adding cdev\n", result);
+		goto err_cdev;
+	}
+
+	ib_ucm_class = class_create(THIS_MODULE, "infiniband_cm");
+	if (IS_ERR(ib_ucm_class)) {
+		result = PTR_ERR(ib_ucm_class);
+		printk(KERN_ERR "UCM: Error <%d> creating class\n", result);
+		goto err_class;
+	}
+
+	class_device_create(ib_ucm_class, IB_UCM_DEV, NULL, "ucm");
+
+	idr_init(&ctx_id_table);
+	init_MUTEX(&ctx_id_mutex);
+
+	return 0;
+err_class:
+	cdev_del(&ib_ucm_cdev);
+err_cdev:
+	unregister_chrdev_region(IB_UCM_DEV, 1);
+err_chr:
+	return result;
+}
+
+static void __exit ib_ucm_cleanup(void)
+{
+	class_device_destroy(ib_ucm_class, IB_UCM_DEV);
+	class_destroy(ib_ucm_class);
+	cdev_del(&ib_ucm_cdev);
+	unregister_chrdev_region(IB_UCM_DEV, 1);
+}
+
+module_init(ib_ucm_init);
+module_exit(ib_ucm_cleanup);
diff --git a/drivers/infiniband/core/ucm.h b/drivers/infiniband/core/ucm.h
new file mode 100644
index 000000000000..6d36606151b2
--- /dev/null
+++ b/drivers/infiniband/core/ucm.h
@@ -0,0 +1,89 @@
+/*
+ * Copyright (c) 2005 Topspin Communications.  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.
+ *
+ * $Id: ucm.h 2208 2005-04-22 23:24:31Z libor $
+ */
+
+#ifndef UCM_H
+#define UCM_H
+
+#include <linux/fs.h>
+#include <linux/device.h>
+#include <linux/cdev.h>
+#include <linux/idr.h>
+
+#include <ib_cm.h>
+#include <ib_user_cm.h>
+
+#define IB_UCM_CM_ID_INVALID 0xffffffff
+
+struct ib_ucm_file {
+	struct semaphore mutex;
+	struct file *filp;
+	/*
+	 * list of pending events
+	 */
+	struct list_head  ctxs;   /* list of active connections */
+	struct list_head  events; /* list of pending events */
+	wait_queue_head_t poll_wait;
+};
+
+struct ib_ucm_context {
+	int                 id;
+	int                 ref;
+	int                 error;
+
+	struct ib_ucm_file *file;
+	struct ib_cm_id    *cm_id;
+	struct semaphore    mutex;
+
+	struct list_head    events;    /* list of pending events. */
+	struct list_head    file_list; /* member in file ctx list */
+};
+
+struct ib_ucm_event {
+	struct ib_ucm_context *ctx;
+	struct list_head file_list; /* member in file event list */
+	struct list_head ctx_list;  /* member in ctx event list */
+
+	struct ib_ucm_event_resp resp;
+	void *data;
+	void *info;
+	int data_len;
+	int info_len;
+	/*
+	 * new connection identifiers needs to be saved until
+	 * userspace can get a handle on them.
+	 */
+	struct ib_cm_id *cm_id;
+};
+
+#endif /* UCM_H */
diff --git a/drivers/infiniband/core/user_mad.c b/drivers/infiniband/core/user_mad.c
index 9d912d6877ff..2e38792df533 100644
--- a/drivers/infiniband/core/user_mad.c
+++ b/drivers/infiniband/core/user_mad.c
@@ -1,5 +1,7 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, Inc. All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, 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
@@ -29,7 +31,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: user_mad.c 1389 2004-12-27 22:56:47Z roland $
+ * $Id: user_mad.c 2814 2005-07-06 19:14:09Z halr $
  */
 
 #include <linux/module.h>
@@ -94,10 +96,12 @@ struct ib_umad_file {
 };
 
 struct ib_umad_packet {
-	struct ib_user_mad mad;
 	struct ib_ah      *ah;
+	struct ib_mad_send_buf *msg;
 	struct list_head   list;
+	int		   length;
 	DECLARE_PCI_UNMAP_ADDR(mapping)
+	struct ib_user_mad mad;
 };
 
 static const dev_t base_dev = MKDEV(IB_UMAD_MAJOR, IB_UMAD_MINOR_BASE);
@@ -114,10 +118,10 @@ static int queue_packet(struct ib_umad_file *file,
 	int ret = 1;
 
 	down_read(&file->agent_mutex);
-	for (packet->mad.id = 0;
-	     packet->mad.id < IB_UMAD_MAX_AGENTS;
-	     packet->mad.id++)
-		if (agent == file->agent[packet->mad.id]) {
+	for (packet->mad.hdr.id = 0;
+	     packet->mad.hdr.id < IB_UMAD_MAX_AGENTS;
+	     packet->mad.hdr.id++)
+		if (agent == file->agent[packet->mad.hdr.id]) {
 			spin_lock_irq(&file->recv_lock);
 			list_add_tail(&packet->list, &file->recv_list);
 			spin_unlock_irq(&file->recv_lock);
@@ -135,22 +139,30 @@ static void send_handler(struct ib_mad_agent *agent,
 			 struct ib_mad_send_wc *send_wc)
 {
 	struct ib_umad_file *file = agent->context;
-	struct ib_umad_packet *packet =
+	struct ib_umad_packet *timeout, *packet =
 		(void *) (unsigned long) send_wc->wr_id;
 
-	dma_unmap_single(agent->device->dma_device,
-			 pci_unmap_addr(packet, mapping),
-			 sizeof packet->mad.data,
-			 DMA_TO_DEVICE);
-	ib_destroy_ah(packet->ah);
+	ib_destroy_ah(packet->msg->send_wr.wr.ud.ah);
+	ib_free_send_mad(packet->msg);
 
 	if (send_wc->status == IB_WC_RESP_TIMEOUT_ERR) {
-		packet->mad.status = ETIMEDOUT;
+		timeout = kmalloc(sizeof *timeout + sizeof (struct ib_mad_hdr),
+				  GFP_KERNEL);
+		if (!timeout)
+			goto out;
 
-		if (!queue_packet(file, agent, packet))
-			return;
-	}
+		memset(timeout, 0, sizeof *timeout + sizeof (struct ib_mad_hdr));
 
+		timeout->length = sizeof (struct ib_mad_hdr);
+		timeout->mad.hdr.id = packet->mad.hdr.id;
+		timeout->mad.hdr.status = ETIMEDOUT;
+		memcpy(timeout->mad.data, packet->mad.data,
+		       sizeof (struct ib_mad_hdr));
+
+		if (!queue_packet(file, agent, timeout))
+				return;
+	}
+out:
 	kfree(packet);
 }
 
@@ -159,30 +171,35 @@ static void recv_handler(struct ib_mad_agent *agent,
 {
 	struct ib_umad_file *file = agent->context;
 	struct ib_umad_packet *packet;
+	int length;
 
 	if (mad_recv_wc->wc->status != IB_WC_SUCCESS)
 		goto out;
 
-	packet = kmalloc(sizeof *packet, GFP_KERNEL);
+	length = mad_recv_wc->mad_len;
+	packet = kmalloc(sizeof *packet + length, GFP_KERNEL);
 	if (!packet)
 		goto out;
 
-	memset(packet, 0, sizeof *packet);
+	memset(packet, 0, sizeof *packet + length);
+	packet->length = length;
+
+	ib_coalesce_recv_mad(mad_recv_wc, packet->mad.data);
 
-	memcpy(packet->mad.data, mad_recv_wc->recv_buf.mad, sizeof packet->mad.data);
-	packet->mad.status        = 0;
-	packet->mad.qpn 	  = cpu_to_be32(mad_recv_wc->wc->src_qp);
-	packet->mad.lid 	  = cpu_to_be16(mad_recv_wc->wc->slid);
-	packet->mad.sl  	  = mad_recv_wc->wc->sl;
-	packet->mad.path_bits 	  = mad_recv_wc->wc->dlid_path_bits;
-	packet->mad.grh_present   = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
-	if (packet->mad.grh_present) {
+	packet->mad.hdr.status    = 0;
+	packet->mad.hdr.length    = length + sizeof (struct ib_user_mad);
+	packet->mad.hdr.qpn 	  = cpu_to_be32(mad_recv_wc->wc->src_qp);
+	packet->mad.hdr.lid 	  = cpu_to_be16(mad_recv_wc->wc->slid);
+	packet->mad.hdr.sl  	  = mad_recv_wc->wc->sl;
+	packet->mad.hdr.path_bits = mad_recv_wc->wc->dlid_path_bits;
+	packet->mad.hdr.grh_present = !!(mad_recv_wc->wc->wc_flags & IB_WC_GRH);
+	if (packet->mad.hdr.grh_present) {
 		/* XXX parse GRH */
-		packet->mad.gid_index 	  = 0;
-		packet->mad.hop_limit 	  = 0;
-		packet->mad.traffic_class = 0;
-		memset(packet->mad.gid, 0, 16);
-		packet->mad.flow_label 	  = 0;
+		packet->mad.hdr.gid_index 	= 0;
+		packet->mad.hdr.hop_limit 	= 0;
+		packet->mad.hdr.traffic_class	= 0;
+		memset(packet->mad.hdr.gid, 0, 16);
+		packet->mad.hdr.flow_label	= 0;
 	}
 
 	if (queue_packet(file, agent, packet))
@@ -199,7 +216,7 @@ static ssize_t ib_umad_read(struct file *filp, char __user *buf,
 	struct ib_umad_packet *packet;
 	ssize_t ret;
 
-	if (count < sizeof (struct ib_user_mad))
+	if (count < sizeof (struct ib_user_mad) + sizeof (struct ib_mad))
 		return -EINVAL;
 
 	spin_lock_irq(&file->recv_lock);
@@ -222,12 +239,25 @@ static ssize_t ib_umad_read(struct file *filp, char __user *buf,
 
 	spin_unlock_irq(&file->recv_lock);
 
-	if (copy_to_user(buf, &packet->mad, sizeof packet->mad))
+	if (count < packet->length + sizeof (struct ib_user_mad)) {
+		/* Return length needed (and first RMPP segment) if too small */
+		if (copy_to_user(buf, &packet->mad,
+				 sizeof (struct ib_user_mad) + sizeof (struct ib_mad)))
+			ret = -EFAULT;
+		else
+			ret = -ENOSPC;
+	} else if (copy_to_user(buf, &packet->mad,
+			      packet->length + sizeof (struct ib_user_mad)))
 		ret = -EFAULT;
 	else
-		ret = sizeof packet->mad;
-
-	kfree(packet);
+		ret = packet->length + sizeof (struct ib_user_mad);
+	if (ret < 0) {
+		/* Requeue packet */
+		spin_lock_irq(&file->recv_lock);
+		list_add(&packet->list, &file->recv_list);
+		spin_unlock_irq(&file->recv_lock);
+	} else
+		kfree(packet);
 	return ret;
 }
 
@@ -238,69 +268,57 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
 	struct ib_umad_packet *packet;
 	struct ib_mad_agent *agent;
 	struct ib_ah_attr ah_attr;
-	struct ib_sge      gather_list;
-	struct ib_send_wr *bad_wr, wr = {
-		.opcode      = IB_WR_SEND,
-		.sg_list     = &gather_list,
-		.num_sge     = 1,
-		.send_flags  = IB_SEND_SIGNALED,
-	};
+	struct ib_send_wr *bad_wr;
+	struct ib_rmpp_mad *rmpp_mad;
 	u8 method;
 	u64 *tid;
-	int ret;
+	int ret, length, hdr_len, data_len, rmpp_hdr_size;
+	int rmpp_active = 0;
 
 	if (count < sizeof (struct ib_user_mad))
 		return -EINVAL;
 
-	packet = kmalloc(sizeof *packet, GFP_KERNEL);
+	length = count - sizeof (struct ib_user_mad);
+	packet = kmalloc(sizeof *packet + sizeof(struct ib_mad_hdr) +
+			 sizeof(struct ib_rmpp_hdr), GFP_KERNEL);
 	if (!packet)
 		return -ENOMEM;
 
-	if (copy_from_user(&packet->mad, buf, sizeof packet->mad)) {
-		kfree(packet);
-		return -EFAULT;
+	if (copy_from_user(&packet->mad, buf,
+			    sizeof (struct ib_user_mad) +
+			    sizeof(struct ib_mad_hdr) +
+			    sizeof(struct ib_rmpp_hdr))) {
+		ret = -EFAULT;
+		goto err;
 	}
 
-	if (packet->mad.id < 0 || packet->mad.id >= IB_UMAD_MAX_AGENTS) {
+	if (packet->mad.hdr.id < 0 ||
+	    packet->mad.hdr.id >= IB_UMAD_MAX_AGENTS) {
 		ret = -EINVAL;
 		goto err;
 	}
 
+	packet->length = length;
+
 	down_read(&file->agent_mutex);
 
-	agent = file->agent[packet->mad.id];
+	agent = file->agent[packet->mad.hdr.id];
 	if (!agent) {
 		ret = -EINVAL;
 		goto err_up;
 	}
 
-	/*
-	 * If userspace is generating a request that will generate a
-	 * response, we need to make sure the high-order part of the
-	 * transaction ID matches the agent being used to send the
-	 * MAD.
-	 */
-	method = ((struct ib_mad_hdr *) packet->mad.data)->method;
-
-	if (!(method & IB_MGMT_METHOD_RESP)       &&
-	    method != IB_MGMT_METHOD_TRAP_REPRESS &&
-	    method != IB_MGMT_METHOD_SEND) {
-		tid = &((struct ib_mad_hdr *) packet->mad.data)->tid;
-		*tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
-				   (be64_to_cpup(tid) & 0xffffffff));
-	}
-
 	memset(&ah_attr, 0, sizeof ah_attr);
-	ah_attr.dlid          = be16_to_cpu(packet->mad.lid);
-	ah_attr.sl            = packet->mad.sl;
-	ah_attr.src_path_bits = packet->mad.path_bits;
+	ah_attr.dlid          = be16_to_cpu(packet->mad.hdr.lid);
+	ah_attr.sl            = packet->mad.hdr.sl;
+	ah_attr.src_path_bits = packet->mad.hdr.path_bits;
 	ah_attr.port_num      = file->port->port_num;
-	if (packet->mad.grh_present) {
+	if (packet->mad.hdr.grh_present) {
 		ah_attr.ah_flags = IB_AH_GRH;
-		memcpy(ah_attr.grh.dgid.raw, packet->mad.gid, 16);
-		ah_attr.grh.flow_label 	   = packet->mad.flow_label;
-		ah_attr.grh.hop_limit  	   = packet->mad.hop_limit;
-		ah_attr.grh.traffic_class  = packet->mad.traffic_class;
+		memcpy(ah_attr.grh.dgid.raw, packet->mad.hdr.gid, 16);
+		ah_attr.grh.flow_label 	   = packet->mad.hdr.flow_label;
+		ah_attr.grh.hop_limit  	   = packet->mad.hdr.hop_limit;
+		ah_attr.grh.traffic_class  = packet->mad.hdr.traffic_class;
 	}
 
 	packet->ah = ib_create_ah(agent->qp->pd, &ah_attr);
@@ -309,34 +327,104 @@ static ssize_t ib_umad_write(struct file *filp, const char __user *buf,
 		goto err_up;
 	}
 
-	gather_list.addr = dma_map_single(agent->device->dma_device,
-					  packet->mad.data,
-					  sizeof packet->mad.data,
-					  DMA_TO_DEVICE);
-	gather_list.length = sizeof packet->mad.data;
-	gather_list.lkey   = file->mr[packet->mad.id]->lkey;
-	pci_unmap_addr_set(packet, mapping, gather_list.addr);
+	rmpp_mad = (struct ib_rmpp_mad *) packet->mad.data;
+	if (ib_get_rmpp_flags(&rmpp_mad->rmpp_hdr) & IB_MGMT_RMPP_FLAG_ACTIVE) {
+		/* RMPP active */
+		if (!agent->rmpp_version) {
+			ret = -EINVAL;
+			goto err_ah;
+		}
+		/* Validate that management class can support RMPP */
+		if (rmpp_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_ADM) {
+			hdr_len = offsetof(struct ib_sa_mad, data);
+			data_len = length;
+		} else if ((rmpp_mad->mad_hdr.mgmt_class >= IB_MGMT_CLASS_VENDOR_RANGE2_START) &&
+			    (rmpp_mad->mad_hdr.mgmt_class <= IB_MGMT_CLASS_VENDOR_RANGE2_END)) {
+				hdr_len = offsetof(struct ib_vendor_mad, data);
+				data_len = length - hdr_len;
+		} else {
+			ret = -EINVAL;
+			goto err_ah;
+		}
+		rmpp_active = 1;
+	} else {
+		if (length > sizeof(struct ib_mad)) {
+			ret = -EINVAL;
+			goto err_ah;
+		}
+		hdr_len = offsetof(struct ib_mad, data);
+		data_len = length - hdr_len;
+	}
+
+	packet->msg = ib_create_send_mad(agent,
+					 be32_to_cpu(packet->mad.hdr.qpn),
+					 0, packet->ah, rmpp_active,
+					 hdr_len, data_len,
+					 GFP_KERNEL);
+	if (IS_ERR(packet->msg)) {
+		ret = PTR_ERR(packet->msg);
+		goto err_ah;
+	}
 
-	wr.wr.ud.mad_hdr     = (struct ib_mad_hdr *) packet->mad.data;
-	wr.wr.ud.ah          = packet->ah;
-	wr.wr.ud.remote_qpn  = be32_to_cpu(packet->mad.qpn);
-	wr.wr.ud.remote_qkey = be32_to_cpu(packet->mad.qkey);
-	wr.wr.ud.timeout_ms  = packet->mad.timeout_ms;
+	packet->msg->send_wr.wr.ud.timeout_ms  = packet->mad.hdr.timeout_ms;
+	packet->msg->send_wr.wr.ud.retries = packet->mad.hdr.retries;
 
-	wr.wr_id            = (unsigned long) packet;
+	/* Override send WR WRID initialized in ib_create_send_mad */
+	packet->msg->send_wr.wr_id = (unsigned long) packet;
 
-	ret = ib_post_send_mad(agent, &wr, &bad_wr);
-	if (ret) {
-		dma_unmap_single(agent->device->dma_device,
-				 pci_unmap_addr(packet, mapping),
-				 sizeof packet->mad.data,
-				 DMA_TO_DEVICE);
-		goto err_up;
+	if (!rmpp_active) {
+		/* Copy message from user into send buffer */
+		if (copy_from_user(packet->msg->mad,
+				   buf + sizeof(struct ib_user_mad), length)) {
+			ret = -EFAULT;
+			goto err_msg;
+		}
+	} else {
+		rmpp_hdr_size = sizeof(struct ib_mad_hdr) +
+				sizeof(struct ib_rmpp_hdr);
+
+		/* Only copy MAD headers (RMPP header in place) */
+		memcpy(packet->msg->mad, packet->mad.data,
+		       sizeof(struct ib_mad_hdr));
+
+		/* Now, copy rest of message from user into send buffer */
+		if (copy_from_user(((struct ib_rmpp_mad *) packet->msg->mad)->data,
+				   buf + sizeof (struct ib_user_mad) + rmpp_hdr_size,
+				   length - rmpp_hdr_size)) {
+			ret = -EFAULT;
+			goto err_msg;
+		}
+	}
+
+	/*
+	 * If userspace is generating a request that will generate a
+	 * response, we need to make sure the high-order part of the
+	 * transaction ID matches the agent being used to send the
+	 * MAD.
+	 */
+	method = packet->msg->mad->mad_hdr.method;
+
+	if (!(method & IB_MGMT_METHOD_RESP)       &&
+	    method != IB_MGMT_METHOD_TRAP_REPRESS &&
+	    method != IB_MGMT_METHOD_SEND) {
+		tid = &packet->msg->mad->mad_hdr.tid;
+		*tid = cpu_to_be64(((u64) agent->hi_tid) << 32 |
+				   (be64_to_cpup(tid) & 0xffffffff));
 	}
 
+	ret = ib_post_send_mad(agent, &packet->msg->send_wr, &bad_wr);
+	if (ret)
+		goto err_msg;
+
 	up_read(&file->agent_mutex);
 
-	return sizeof packet->mad;
+	return sizeof (struct ib_user_mad_hdr) + packet->length;
+
+err_msg:
+	ib_free_send_mad(packet->msg);
+
+err_ah:
+	ib_destroy_ah(packet->ah);
 
 err_up:
 	up_read(&file->agent_mutex);
@@ -399,7 +487,8 @@ found:
 	agent = ib_register_mad_agent(file->port->ib_dev, file->port->port_num,
 				      ureq.qpn ? IB_QPT_GSI : IB_QPT_SMI,
 				      ureq.mgmt_class ? &req : NULL,
-				      0, send_handler, recv_handler, file);
+				      ureq.rmpp_version,
+				      send_handler, recv_handler, file);
 	if (IS_ERR(agent)) {
 		ret = PTR_ERR(agent);
 		goto out;
@@ -460,8 +549,8 @@ out:
 	return ret;
 }
 
-static long ib_umad_ioctl(struct file *filp,
-			 unsigned int cmd, unsigned long arg)
+static long ib_umad_ioctl(struct file *filp, unsigned int cmd,
+			  unsigned long arg)
 {
 	switch (cmd) {
 	case IB_USER_MAD_REGISTER_AGENT:
@@ -517,14 +606,14 @@ static int ib_umad_close(struct inode *inode, struct file *filp)
 }
 
 static struct file_operations umad_fops = {
-	.owner 	        = THIS_MODULE,
-	.read 	        = ib_umad_read,
-	.write 	        = ib_umad_write,
-	.poll 	        = ib_umad_poll,
+	.owner 	 	= THIS_MODULE,
+	.read 	 	= ib_umad_read,
+	.write 	 	= ib_umad_write,
+	.poll 	 	= ib_umad_poll,
 	.unlocked_ioctl = ib_umad_ioctl,
-	.compat_ioctl   = ib_umad_ioctl,
-	.open 	        = ib_umad_open,
-	.release        = ib_umad_close
+	.compat_ioctl 	= ib_umad_ioctl,
+	.open 	 	= ib_umad_open,
+	.release 	= ib_umad_close
 };
 
 static int ib_umad_sm_open(struct inode *inode, struct file *filp)
diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
index 2516f9646515..506fdf1f2a26 100644
--- a/drivers/infiniband/core/verbs.c
+++ b/drivers/infiniband/core/verbs.c
@@ -41,6 +41,7 @@
 #include <linux/err.h>
 
 #include <ib_verbs.h>
+#include <ib_cache.h>
 
 /* Protection domains */
 
@@ -88,6 +89,40 @@ struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
 }
 EXPORT_SYMBOL(ib_create_ah);
 
+struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
+				   struct ib_grh *grh, u8 port_num)
+{
+	struct ib_ah_attr ah_attr;
+	u32 flow_class;
+	u16 gid_index;
+	int ret;
+
+	memset(&ah_attr, 0, sizeof ah_attr);
+	ah_attr.dlid = wc->slid;
+	ah_attr.sl = wc->sl;
+	ah_attr.src_path_bits = wc->dlid_path_bits;
+	ah_attr.port_num = port_num;
+
+	if (wc->wc_flags & IB_WC_GRH) {
+		ah_attr.ah_flags = IB_AH_GRH;
+		ah_attr.grh.dgid = grh->dgid;
+
+		ret = ib_find_cached_gid(pd->device, &grh->sgid, &port_num,
+					 &gid_index);
+		if (ret)
+			return ERR_PTR(ret);
+
+		ah_attr.grh.sgid_index = (u8) gid_index;
+		flow_class = be32_to_cpu(grh->version_tclass_flow);
+		ah_attr.grh.flow_label = flow_class & 0xFFFFF;
+		ah_attr.grh.traffic_class = (flow_class >> 20) & 0xFF;
+		ah_attr.grh.hop_limit = grh->hop_limit;
+	}
+
+	return ib_create_ah(pd, &ah_attr);
+}
+EXPORT_SYMBOL(ib_create_ah_from_wc);
+
 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
 {
 	return ah->device->modify_ah ?
diff --git a/drivers/infiniband/include/ib_cm.h b/drivers/infiniband/include/ib_cm.h
new file mode 100644
index 000000000000..e5d74a730a70
--- /dev/null
+++ b/drivers/infiniband/include/ib_cm.h
@@ -0,0 +1,568 @@
+/*
+ * Copyright (c) 2004 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, 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.
+ *
+ * $Id: ib_cm.h 2730 2005-06-28 16:43:03Z sean.hefty $
+ */
+#if !defined(IB_CM_H)
+#define IB_CM_H
+
+#include <ib_mad.h>
+#include <ib_sa.h>
+
+enum ib_cm_state {
+	IB_CM_IDLE,
+	IB_CM_LISTEN,
+	IB_CM_REQ_SENT,
+	IB_CM_REQ_RCVD,
+	IB_CM_MRA_REQ_SENT,
+	IB_CM_MRA_REQ_RCVD,
+	IB_CM_REP_SENT,
+	IB_CM_REP_RCVD,
+	IB_CM_MRA_REP_SENT,
+	IB_CM_MRA_REP_RCVD,
+	IB_CM_ESTABLISHED,
+	IB_CM_DREQ_SENT,
+	IB_CM_DREQ_RCVD,
+	IB_CM_TIMEWAIT,
+	IB_CM_SIDR_REQ_SENT,
+	IB_CM_SIDR_REQ_RCVD
+};
+
+enum ib_cm_lap_state {
+	IB_CM_LAP_IDLE,
+	IB_CM_LAP_SENT,
+	IB_CM_LAP_RCVD,
+	IB_CM_MRA_LAP_SENT,
+	IB_CM_MRA_LAP_RCVD,
+};
+
+enum ib_cm_event_type {
+	IB_CM_REQ_ERROR,
+	IB_CM_REQ_RECEIVED,
+	IB_CM_REP_ERROR,
+	IB_CM_REP_RECEIVED,
+	IB_CM_RTU_RECEIVED,
+	IB_CM_USER_ESTABLISHED,
+	IB_CM_DREQ_ERROR,
+	IB_CM_DREQ_RECEIVED,
+	IB_CM_DREP_RECEIVED,
+	IB_CM_TIMEWAIT_EXIT,
+	IB_CM_MRA_RECEIVED,
+	IB_CM_REJ_RECEIVED,
+	IB_CM_LAP_ERROR,
+	IB_CM_LAP_RECEIVED,
+	IB_CM_APR_RECEIVED,
+	IB_CM_SIDR_REQ_ERROR,
+	IB_CM_SIDR_REQ_RECEIVED,
+	IB_CM_SIDR_REP_RECEIVED
+};
+
+enum ib_cm_data_size {
+	IB_CM_REQ_PRIVATE_DATA_SIZE	 = 92,
+	IB_CM_MRA_PRIVATE_DATA_SIZE	 = 222,
+	IB_CM_REJ_PRIVATE_DATA_SIZE	 = 148,
+	IB_CM_REP_PRIVATE_DATA_SIZE	 = 196,
+	IB_CM_RTU_PRIVATE_DATA_SIZE	 = 224,
+	IB_CM_DREQ_PRIVATE_DATA_SIZE	 = 220,
+	IB_CM_DREP_PRIVATE_DATA_SIZE	 = 224,
+	IB_CM_REJ_ARI_LENGTH		 = 72,
+	IB_CM_LAP_PRIVATE_DATA_SIZE	 = 168,
+	IB_CM_APR_PRIVATE_DATA_SIZE	 = 148,
+	IB_CM_APR_INFO_LENGTH		 = 72,
+	IB_CM_SIDR_REQ_PRIVATE_DATA_SIZE = 216,
+	IB_CM_SIDR_REP_PRIVATE_DATA_SIZE = 136,
+	IB_CM_SIDR_REP_INFO_LENGTH	 = 72
+};
+
+struct ib_cm_id;
+
+struct ib_cm_req_event_param {
+	struct ib_cm_id		*listen_id;
+	struct ib_device	*device;
+	u8			port;
+
+	struct ib_sa_path_rec	*primary_path;
+	struct ib_sa_path_rec	*alternate_path;
+
+	u64			remote_ca_guid;
+	u32			remote_qkey;
+	u32			remote_qpn;
+	enum ib_qp_type		qp_type;
+
+	u32			starting_psn;
+	u8			responder_resources;
+	u8			initiator_depth;
+	unsigned int		local_cm_response_timeout:5;
+	unsigned int		flow_control:1;
+	unsigned int		remote_cm_response_timeout:5;
+	unsigned int		retry_count:3;
+	unsigned int		rnr_retry_count:3;
+	unsigned int		srq:1;
+};
+
+struct ib_cm_rep_event_param {
+	u64			remote_ca_guid;
+	u32			remote_qkey;
+	u32			remote_qpn;
+	u32			starting_psn;
+	u8			responder_resources;
+	u8			initiator_depth;
+	unsigned int		target_ack_delay:5;
+	unsigned int		failover_accepted:2;
+	unsigned int		flow_control:1;
+	unsigned int		rnr_retry_count:3;
+	unsigned int		srq:1;
+};
+
+enum ib_cm_rej_reason {
+	IB_CM_REJ_NO_QP				= __constant_htons(1),
+	IB_CM_REJ_NO_EEC			= __constant_htons(2),
+	IB_CM_REJ_NO_RESOURCES			= __constant_htons(3),
+	IB_CM_REJ_TIMEOUT			= __constant_htons(4),
+	IB_CM_REJ_UNSUPPORTED			= __constant_htons(5),
+	IB_CM_REJ_INVALID_COMM_ID		= __constant_htons(6),
+	IB_CM_REJ_INVALID_COMM_INSTANCE		= __constant_htons(7),
+	IB_CM_REJ_INVALID_SERVICE_ID		= __constant_htons(8),
+	IB_CM_REJ_INVALID_TRANSPORT_TYPE	= __constant_htons(9),
+	IB_CM_REJ_STALE_CONN			= __constant_htons(10),
+	IB_CM_REJ_RDC_NOT_EXIST			= __constant_htons(11),
+	IB_CM_REJ_INVALID_GID			= __constant_htons(12),
+	IB_CM_REJ_INVALID_LID			= __constant_htons(13),
+	IB_CM_REJ_INVALID_SL			= __constant_htons(14),
+	IB_CM_REJ_INVALID_TRAFFIC_CLASS		= __constant_htons(15),
+	IB_CM_REJ_INVALID_HOP_LIMIT		= __constant_htons(16),
+	IB_CM_REJ_INVALID_PACKET_RATE		= __constant_htons(17),
+	IB_CM_REJ_INVALID_ALT_GID		= __constant_htons(18),
+	IB_CM_REJ_INVALID_ALT_LID		= __constant_htons(19),
+	IB_CM_REJ_INVALID_ALT_SL		= __constant_htons(20),
+	IB_CM_REJ_INVALID_ALT_TRAFFIC_CLASS	= __constant_htons(21),
+	IB_CM_REJ_INVALID_ALT_HOP_LIMIT		= __constant_htons(22),
+	IB_CM_REJ_INVALID_ALT_PACKET_RATE	= __constant_htons(23),
+	IB_CM_REJ_PORT_REDIRECT			= __constant_htons(24),
+	IB_CM_REJ_INVALID_MTU			= __constant_htons(26),
+	IB_CM_REJ_INSUFFICIENT_RESP_RESOURCES	= __constant_htons(27),
+	IB_CM_REJ_CONSUMER_DEFINED		= __constant_htons(28),
+	IB_CM_REJ_INVALID_RNR_RETRY		= __constant_htons(29),
+	IB_CM_REJ_DUPLICATE_LOCAL_COMM_ID	= __constant_htons(30),
+	IB_CM_REJ_INVALID_CLASS_VERSION		= __constant_htons(31),
+	IB_CM_REJ_INVALID_FLOW_LABEL		= __constant_htons(32),
+	IB_CM_REJ_INVALID_ALT_FLOW_LABEL	= __constant_htons(33)
+};
+
+struct ib_cm_rej_event_param {
+	enum ib_cm_rej_reason	reason;
+	void			*ari;
+	u8			ari_length;
+};
+
+struct ib_cm_mra_event_param {
+	u8	service_timeout;
+};
+
+struct ib_cm_lap_event_param {
+	struct ib_sa_path_rec	*alternate_path;
+};
+
+enum ib_cm_apr_status {
+	IB_CM_APR_SUCCESS,
+	IB_CM_APR_INVALID_COMM_ID,
+	IB_CM_APR_UNSUPPORTED,
+	IB_CM_APR_REJECT,
+	IB_CM_APR_REDIRECT,
+	IB_CM_APR_IS_CURRENT,
+	IB_CM_APR_INVALID_QPN_EECN,
+	IB_CM_APR_INVALID_LID,
+	IB_CM_APR_INVALID_GID,
+	IB_CM_APR_INVALID_FLOW_LABEL,
+	IB_CM_APR_INVALID_TCLASS,
+	IB_CM_APR_INVALID_HOP_LIMIT,
+	IB_CM_APR_INVALID_PACKET_RATE,
+	IB_CM_APR_INVALID_SL
+};
+
+struct ib_cm_apr_event_param {
+	enum ib_cm_apr_status	ap_status;
+	void			*apr_info;
+	u8			info_len;
+};
+
+struct ib_cm_sidr_req_event_param {
+	struct ib_cm_id		*listen_id;
+	struct ib_device	*device;
+	u8			port;
+
+	u16	pkey;
+};
+
+enum ib_cm_sidr_status {
+	IB_SIDR_SUCCESS,
+	IB_SIDR_UNSUPPORTED,
+	IB_SIDR_REJECT,
+	IB_SIDR_NO_QP,
+	IB_SIDR_REDIRECT,
+	IB_SIDR_UNSUPPORTED_VERSION
+};
+
+struct ib_cm_sidr_rep_event_param {
+	enum ib_cm_sidr_status	status;
+	u32			qkey;
+	u32			qpn;
+	void			*info;
+	u8			info_len;
+
+};
+
+struct ib_cm_event {
+	enum ib_cm_event_type	event;
+	union {
+		struct ib_cm_req_event_param	req_rcvd;
+		struct ib_cm_rep_event_param	rep_rcvd;
+		/* No data for RTU received events. */
+		struct ib_cm_rej_event_param	rej_rcvd;
+		struct ib_cm_mra_event_param	mra_rcvd;
+		struct ib_cm_lap_event_param	lap_rcvd;
+		struct ib_cm_apr_event_param	apr_rcvd;
+		/* No data for DREQ/DREP received events. */
+		struct ib_cm_sidr_req_event_param sidr_req_rcvd;
+		struct ib_cm_sidr_rep_event_param sidr_rep_rcvd;
+		enum ib_wc_status		send_status;
+	} param;
+
+	void			*private_data;
+};
+
+/**
+ * ib_cm_handler - User-defined callback to process communication events.
+ * @cm_id: Communication identifier associated with the reported event.
+ * @event: Information about the communication event.
+ *
+ * IB_CM_REQ_RECEIVED and IB_CM_SIDR_REQ_RECEIVED communication events
+ * generated as a result of listen requests result in the allocation of a
+ * new @cm_id.  The new @cm_id is returned to the user through this callback.
+ * Clients are responsible for destroying the new @cm_id.  For peer-to-peer
+ * IB_CM_REQ_RECEIVED and all other events, the returned @cm_id corresponds
+ * to a user's existing communication identifier.
+ *
+ * Users may not call ib_destroy_cm_id while in the context of this callback;
+ * however, returning a non-zero value instructs the communication manager to
+ * destroy the @cm_id after the callback completes.
+ */
+typedef int (*ib_cm_handler)(struct ib_cm_id *cm_id,
+			     struct ib_cm_event *event);
+
+struct ib_cm_id {
+	ib_cm_handler		cm_handler;
+	void			*context;
+	u64			service_id;
+	u64			service_mask;
+	enum ib_cm_state	state;		/* internal CM/debug use */
+	enum ib_cm_lap_state	lap_state;	/* internal CM/debug use */
+	u32			local_id;
+	u32			remote_id;
+};
+
+/**
+ * ib_create_cm_id - Allocate a communication identifier.
+ * @cm_handler: Callback invoked to notify the user of CM events.
+ * @context: User specified context associated with the communication
+ *   identifier.
+ *
+ * Communication identifiers are used to track connection states, service
+ * ID resolution requests, and listen requests.
+ */
+struct ib_cm_id *ib_create_cm_id(ib_cm_handler cm_handler,
+				 void *context);
+
+/**
+ * ib_destroy_cm_id - Destroy a connection identifier.
+ * @cm_id: Connection identifier to destroy.
+ *
+ * This call blocks until the connection identifier is destroyed.
+ */
+void ib_destroy_cm_id(struct ib_cm_id *cm_id);
+
+#define IB_SERVICE_ID_AGN_MASK	__constant_cpu_to_be64(0xFF00000000000000ULL)
+#define IB_CM_ASSIGN_SERVICE_ID __constant_cpu_to_be64(0x0200000000000000ULL)
+
+/**
+ * ib_cm_listen - Initiates listening on the specified service ID for
+ *   connection and service ID resolution requests.
+ * @cm_id: Connection identifier associated with the listen request.
+ * @service_id: Service identifier matched against incoming connection
+ *   and service ID resolution requests.  The service ID should be specified
+ *   network-byte order.  If set to IB_CM_ASSIGN_SERVICE_ID, the CM will
+ *   assign a service ID to the caller.
+ * @service_mask: Mask applied to service ID used to listen across a
+ *   range of service IDs.  If set to 0, the service ID is matched
+ *   exactly.  This parameter is ignored if %service_id is set to
+ *   IB_CM_ASSIGN_SERVICE_ID.
+ */
+int ib_cm_listen(struct ib_cm_id *cm_id,
+		 u64 service_id,
+		 u64 service_mask);
+
+struct ib_cm_req_param {
+	struct ib_sa_path_rec	*primary_path;
+	struct ib_sa_path_rec	*alternate_path;
+	u64			service_id;
+	u32			qp_num;
+	enum ib_qp_type		qp_type;
+	u32			starting_psn;
+	const void		*private_data;
+	u8			private_data_len;
+	u8			peer_to_peer;
+	u8			responder_resources;
+	u8			initiator_depth;
+	u8			remote_cm_response_timeout;
+	u8			flow_control;
+	u8			local_cm_response_timeout;
+	u8			retry_count;
+	u8			rnr_retry_count;
+	u8			max_cm_retries;
+	u8			srq;
+};
+
+/**
+ * ib_send_cm_req - Sends a connection request to the remote node.
+ * @cm_id: Connection identifier that will be associated with the
+ *   connection request.
+ * @param: Connection request information needed to establish the
+ *   connection.
+ */
+int ib_send_cm_req(struct ib_cm_id *cm_id,
+		   struct ib_cm_req_param *param);
+
+struct ib_cm_rep_param {
+	u32		qp_num;
+	u32		starting_psn;
+	const void	*private_data;
+	u8		private_data_len;
+	u8		responder_resources;
+	u8		initiator_depth;
+	u8		target_ack_delay;
+	u8		failover_accepted;
+	u8		flow_control;
+	u8		rnr_retry_count;
+	u8		srq;
+};
+
+/**
+ * ib_send_cm_rep - Sends a connection reply in response to a connection
+ *   request.
+ * @cm_id: Connection identifier that will be associated with the
+ *   connection request.
+ * @param: Connection reply information needed to establish the
+ *   connection.
+ */
+int ib_send_cm_rep(struct ib_cm_id *cm_id,
+		   struct ib_cm_rep_param *param);
+
+/**
+ * ib_send_cm_rtu - Sends a connection ready to use message in response
+ *   to a connection reply message.
+ * @cm_id: Connection identifier associated with the connection request.
+ * @private_data: Optional user-defined private data sent with the
+ *   ready to use message.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ */
+int ib_send_cm_rtu(struct ib_cm_id *cm_id,
+		   const void *private_data,
+		   u8 private_data_len);
+
+/**
+ * ib_send_cm_dreq - Sends a disconnection request for an existing
+ *   connection.
+ * @cm_id: Connection identifier associated with the connection being
+ *   released.
+ * @private_data: Optional user-defined private data sent with the
+ *   disconnection request message.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ */
+int ib_send_cm_dreq(struct ib_cm_id *cm_id,
+		    const void *private_data,
+		    u8 private_data_len);
+
+/**
+ * ib_send_cm_drep - Sends a disconnection reply to a disconnection request.
+ * @cm_id: Connection identifier associated with the connection being
+ *   released.
+ * @private_data: Optional user-defined private data sent with the
+ *   disconnection reply message.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ *
+ * If the cm_id is in the correct state, the CM will transition the connection
+ * to the timewait state, even if an error occurs sending the DREP message.
+ */
+int ib_send_cm_drep(struct ib_cm_id *cm_id,
+		    const void *private_data,
+		    u8 private_data_len);
+
+/**
+ * ib_cm_establish - Forces a connection state to established.
+ * @cm_id: Connection identifier to transition to established.
+ *
+ * This routine should be invoked by users who receive messages on a
+ * connected QP before an RTU has been received.
+ */
+int ib_cm_establish(struct ib_cm_id *cm_id);
+
+/**
+ * ib_send_cm_rej - Sends a connection rejection message to the
+ *   remote node.
+ * @cm_id: Connection identifier associated with the connection being
+ *   rejected.
+ * @reason: Reason for the connection request rejection.
+ * @ari: Optional additional rejection information.
+ * @ari_length: Size of the additional rejection information, in bytes.
+ * @private_data: Optional user-defined private data sent with the
+ *   rejection message.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ */
+int ib_send_cm_rej(struct ib_cm_id *cm_id,
+		   enum ib_cm_rej_reason reason,
+		   void *ari,
+		   u8 ari_length,
+		   const void *private_data,
+		   u8 private_data_len);
+
+/**
+ * ib_send_cm_mra - Sends a message receipt acknowledgement to a connection
+ *   message.
+ * @cm_id: Connection identifier associated with the connection message.
+ * @service_timeout: The maximum time required for the sender to reply to
+ *   to the connection message.
+ * @private_data: Optional user-defined private data sent with the
+ *   message receipt acknowledgement.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ */
+int ib_send_cm_mra(struct ib_cm_id *cm_id,
+		   u8 service_timeout,
+		   const void *private_data,
+		   u8 private_data_len);
+
+/**
+ * ib_send_cm_lap - Sends a load alternate path request.
+ * @cm_id: Connection identifier associated with the load alternate path
+ *   message.
+ * @alternate_path: A path record that identifies the alternate path to
+ *   load.
+ * @private_data: Optional user-defined private data sent with the
+ *   load alternate path message.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ */
+int ib_send_cm_lap(struct ib_cm_id *cm_id,
+		   struct ib_sa_path_rec *alternate_path,
+		   const void *private_data,
+		   u8 private_data_len);
+
+/**
+ * ib_cm_init_qp_attr - Initializes the QP attributes for use in transitioning
+ *   to a specified QP state.
+ * @cm_id: Communication identifier associated with the QP attributes to
+ *   initialize.
+ * @qp_attr: On input, specifies the desired QP state.  On output, the
+ *   mandatory and desired optional attributes will be set in order to
+ *   modify the QP to the specified state.
+ * @qp_attr_mask: The QP attribute mask that may be used to transition the
+ *   QP to the specified state.
+ *
+ * Users must set the @qp_attr->qp_state to the desired QP state.  This call
+ * will set all required attributes for the given transition, along with
+ * known optional attributes.  Users may override the attributes returned from
+ * this call before calling ib_modify_qp.
+ */
+int ib_cm_init_qp_attr(struct ib_cm_id *cm_id,
+		       struct ib_qp_attr *qp_attr,
+		       int *qp_attr_mask);
+
+/**
+ * ib_send_cm_apr - Sends an alternate path response message in response to
+ *   a load alternate path request.
+ * @cm_id: Connection identifier associated with the alternate path response.
+ * @status: Reply status sent with the alternate path response.
+ * @info: Optional additional information sent with the alternate path
+ *   response.
+ * @info_length: Size of the additional information, in bytes.
+ * @private_data: Optional user-defined private data sent with the
+ *   alternate path response message.
+ * @private_data_len: Size of the private data buffer, in bytes.
+ */
+int ib_send_cm_apr(struct ib_cm_id *cm_id,
+		   enum ib_cm_apr_status status,
+		   void *info,
+		   u8 info_length,
+		   const void *private_data,
+		   u8 private_data_len);
+
+struct ib_cm_sidr_req_param {
+	struct ib_sa_path_rec	*path;
+	u64			service_id;
+	int			timeout_ms;
+	const void		*private_data;
+	u8			private_data_len;
+	u8			max_cm_retries;
+	u16			pkey;
+};
+
+/**
+ * ib_send_cm_sidr_req - Sends a service ID resolution request to the
+ *   remote node.
+ * @cm_id: Communication identifier that will be associated with the
+ *   service ID resolution request.
+ * @param: Service ID resolution request information.
+ */
+int ib_send_cm_sidr_req(struct ib_cm_id *cm_id,
+			struct ib_cm_sidr_req_param *param);
+
+struct ib_cm_sidr_rep_param {
+	u32			qp_num;
+	u32			qkey;
+	enum ib_cm_sidr_status	status;
+	const void		*info;
+	u8			info_length;
+	const void		*private_data;
+	u8			private_data_len;
+};
+
+/**
+ * ib_send_cm_sidr_rep - Sends a service ID resolution request to the
+ *   remote node.
+ * @cm_id: Communication identifier associated with the received service ID
+ *   resolution request.
+ * @param: Service ID resolution reply information.
+ */
+int ib_send_cm_sidr_rep(struct ib_cm_id *cm_id,
+			struct ib_cm_sidr_rep_param *param);
+
+#endif /* IB_CM_H */
diff --git a/drivers/infiniband/include/ib_fmr_pool.h b/drivers/infiniband/include/ib_fmr_pool.h
index e8769657cbbb..6c9e24d6e144 100644
--- a/drivers/infiniband/include/ib_fmr_pool.h
+++ b/drivers/infiniband/include/ib_fmr_pool.h
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, 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
@@ -29,7 +30,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: ib_fmr_pool.h 1349 2004-12-16 21:09:43Z roland $
+ * $Id: ib_fmr_pool.h 2730 2005-06-28 16:43:03Z sean.hefty $
  */
 
 #if !defined(IB_FMR_POOL_H)
@@ -78,7 +79,7 @@ struct ib_pool_fmr {
 struct ib_fmr_pool *ib_create_fmr_pool(struct ib_pd             *pd,
 				       struct ib_fmr_pool_param *params);
 
-int ib_destroy_fmr_pool(struct ib_fmr_pool *pool);
+void ib_destroy_fmr_pool(struct ib_fmr_pool *pool);
 
 int ib_flush_fmr_pool(struct ib_fmr_pool *pool);
 
diff --git a/drivers/infiniband/include/ib_mad.h b/drivers/infiniband/include/ib_mad.h
index 4a6bf6763a97..491b6f25b3b8 100644
--- a/drivers/infiniband/include/ib_mad.h
+++ b/drivers/infiniband/include/ib_mad.h
@@ -33,12 +33,14 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: ib_mad.h 1389 2004-12-27 22:56:47Z roland $
+ * $Id: ib_mad.h 2775 2005-07-02 13:42:12Z halr $
  */
 
 #if !defined( IB_MAD_H )
 #define IB_MAD_H
 
+#include <linux/pci.h>
+
 #include <ib_verbs.h>
 
 /* Management base version */
@@ -56,6 +58,8 @@
 #define IB_MGMT_CLASS_VENDOR_RANGE2_START	0x30
 #define IB_MGMT_CLASS_VENDOR_RANGE2_END		0x4F
 
+#define	IB_OPENIB_OUI				(0x001405)
+
 /* Management methods */
 #define IB_MGMT_METHOD_GET			0x01
 #define IB_MGMT_METHOD_SET			0x02
@@ -70,18 +74,37 @@
 
 #define IB_MGMT_MAX_METHODS			128
 
+/* RMPP information */
+#define IB_MGMT_RMPP_VERSION			1
+
+#define IB_MGMT_RMPP_TYPE_DATA			1
+#define IB_MGMT_RMPP_TYPE_ACK			2
+#define IB_MGMT_RMPP_TYPE_STOP			3
+#define IB_MGMT_RMPP_TYPE_ABORT			4
+
+#define IB_MGMT_RMPP_FLAG_ACTIVE		1
+#define IB_MGMT_RMPP_FLAG_FIRST			(1<<1)
+#define IB_MGMT_RMPP_FLAG_LAST			(1<<2)
+
+#define IB_MGMT_RMPP_NO_RESPTIME		0x1F
+
+#define	IB_MGMT_RMPP_STATUS_SUCCESS		0
+#define	IB_MGMT_RMPP_STATUS_RESX		1
+#define	IB_MGMT_RMPP_STATUS_T2L			118
+#define	IB_MGMT_RMPP_STATUS_BAD_LEN		119
+#define	IB_MGMT_RMPP_STATUS_BAD_SEG		120
+#define	IB_MGMT_RMPP_STATUS_BADT		121
+#define	IB_MGMT_RMPP_STATUS_W2S			122
+#define	IB_MGMT_RMPP_STATUS_S2B			123
+#define	IB_MGMT_RMPP_STATUS_BAD_STATUS		124
+#define	IB_MGMT_RMPP_STATUS_UNV			125
+#define	IB_MGMT_RMPP_STATUS_TMR			126
+#define	IB_MGMT_RMPP_STATUS_UNSPEC		127
+
 #define IB_QP0		0
 #define IB_QP1		__constant_htonl(1)
 #define IB_QP1_QKEY	0x80010000
-
-struct ib_grh {
-	u32		version_tclass_flow;
-	u16		paylen;
-	u8		next_hdr;
-	u8		hop_limit;
-	union ib_gid	sgid;
-	union ib_gid	dgid;
-} __attribute__ ((packed));
+#define IB_QP_SET_QKEY	0x80000000
 
 struct ib_mad_hdr {
 	u8	base_version;
@@ -94,7 +117,7 @@ struct ib_mad_hdr {
 	u16	attr_id;
 	u16	resv;
 	u32	attr_mod;
-} __attribute__ ((packed));
+};
 
 struct ib_rmpp_hdr {
 	u8	rmpp_version;
@@ -103,17 +126,41 @@ struct ib_rmpp_hdr {
 	u8	rmpp_status;
 	u32	seg_num;
 	u32	paylen_newwin;
+};
+
+typedef u64 __bitwise ib_sa_comp_mask;
+
+#define IB_SA_COMP_MASK(n) ((__force ib_sa_comp_mask) cpu_to_be64(1ull << n))
+
+/*
+ * ib_sa_hdr and ib_sa_mad structures must be packed because they have
+ * 64-bit fields that are only 32-bit aligned. 64-bit architectures will
+ * lay them out wrong otherwise.  (And unfortunately they are sent on
+ * the wire so we can't change the layout)
+ */
+struct ib_sa_hdr {
+	u64			sm_key;
+	u16			attr_offset;
+	u16			reserved;
+	ib_sa_comp_mask		comp_mask;
 } __attribute__ ((packed));
 
 struct ib_mad {
 	struct ib_mad_hdr	mad_hdr;
 	u8			data[232];
-} __attribute__ ((packed));
+};
 
 struct ib_rmpp_mad {
 	struct ib_mad_hdr	mad_hdr;
 	struct ib_rmpp_hdr	rmpp_hdr;
 	u8			data[220];
+};
+
+struct ib_sa_mad {
+	struct ib_mad_hdr	mad_hdr;
+	struct ib_rmpp_hdr	rmpp_hdr;
+	struct ib_sa_hdr	sa_hdr;
+	u8			data[200];
 } __attribute__ ((packed));
 
 struct ib_vendor_mad {
@@ -122,7 +169,70 @@ struct ib_vendor_mad {
 	u8			reserved;
 	u8			oui[3];
 	u8			data[216];
-} __attribute__ ((packed));
+};
+
+/**
+ * ib_mad_send_buf - MAD data buffer and work request for sends.
+ * @mad: References an allocated MAD data buffer.  The size of the data
+ *   buffer is specified in the @send_wr.length field.
+ * @mapping: DMA mapping information.
+ * @mad_agent: MAD agent that allocated the buffer.
+ * @context: User-controlled context fields.
+ * @send_wr: An initialized work request structure used when sending the MAD.
+ *   The wr_id field of the work request is initialized to reference this
+ *   data structure.
+ * @sge: A scatter-gather list referenced by the work request.
+ *
+ * Users are responsible for initializing the MAD buffer itself, with the
+ * exception of specifying the payload length field in any RMPP MAD.
+ */
+struct ib_mad_send_buf {
+	struct ib_mad		*mad;
+	DECLARE_PCI_UNMAP_ADDR(mapping)
+	struct ib_mad_agent	*mad_agent;
+	void			*context[2];
+	struct ib_send_wr	send_wr;
+	struct ib_sge		sge;
+};
+
+/**
+ * ib_get_rmpp_resptime - Returns the RMPP response time.
+ * @rmpp_hdr: An RMPP header.
+ */
+static inline u8 ib_get_rmpp_resptime(struct ib_rmpp_hdr *rmpp_hdr)
+{
+	return rmpp_hdr->rmpp_rtime_flags >> 3;
+}
+
+/**
+ * ib_get_rmpp_flags - Returns the RMPP flags.
+ * @rmpp_hdr: An RMPP header.
+ */
+static inline u8 ib_get_rmpp_flags(struct ib_rmpp_hdr *rmpp_hdr)
+{
+	return rmpp_hdr->rmpp_rtime_flags & 0x7;
+}
+
+/**
+ * ib_set_rmpp_resptime - Sets the response time in an RMPP header.
+ * @rmpp_hdr: An RMPP header.
+ * @rtime: The response time to set.
+ */
+static inline void ib_set_rmpp_resptime(struct ib_rmpp_hdr *rmpp_hdr, u8 rtime)
+{
+	rmpp_hdr->rmpp_rtime_flags = ib_get_rmpp_flags(rmpp_hdr) | (rtime << 3);
+}
+
+/**
+ * ib_set_rmpp_flags - Sets the flags in an RMPP header.
+ * @rmpp_hdr: An RMPP header.
+ * @flags: The flags to set.
+ */
+static inline void ib_set_rmpp_flags(struct ib_rmpp_hdr *rmpp_hdr, u8 flags)
+{
+	rmpp_hdr->rmpp_rtime_flags = (rmpp_hdr->rmpp_rtime_flags & 0xF1) |
+				     (flags & 0x7);
+}
 
 struct ib_mad_agent;
 struct ib_mad_send_wc;
@@ -168,6 +278,7 @@ typedef void (*ib_mad_recv_handler)(struct ib_mad_agent *mad_agent,
  * ib_mad_agent - Used to track MAD registration with the access layer.
  * @device: Reference to device registration is on.
  * @qp: Reference to QP used for sending and receiving MADs.
+ * @mr: Memory region for system memory usable for DMA.
  * @recv_handler: Callback handler for a received MAD.
  * @send_handler: Callback handler for a sent MAD.
  * @snoop_handler: Callback handler for snooped sent MADs.
@@ -176,16 +287,19 @@ typedef void (*ib_mad_recv_handler)(struct ib_mad_agent *mad_agent,
  *   Unsolicited MADs sent by this client will have the upper 32-bits
  *   of their TID set to this value.
  * @port_num: Port number on which QP is registered
+ * @rmpp_version: If set, indicates the RMPP version used by this agent.
  */
 struct ib_mad_agent {
 	struct ib_device	*device;
 	struct ib_qp		*qp;
+	struct ib_mr		*mr;
 	ib_mad_recv_handler	recv_handler;
 	ib_mad_send_handler	send_handler;
 	ib_mad_snoop_handler	snoop_handler;
 	void			*context;
 	u32			hi_tid;
 	u8			port_num;
+	u8			rmpp_version;
 };
 
 /**
@@ -219,6 +333,7 @@ struct ib_mad_recv_buf {
  * ib_mad_recv_wc - received MAD information.
  * @wc: Completion information for the received data.
  * @recv_buf: Specifies the location of the received data buffer(s).
+ * @rmpp_list: Specifies a list of RMPP reassembled received MAD buffers.
  * @mad_len: The length of the received MAD, without duplicated headers.
  *
  * For received response, the wr_id field of the wc is set to the wr_id
@@ -227,6 +342,7 @@ struct ib_mad_recv_buf {
 struct ib_mad_recv_wc {
 	struct ib_wc		*wc;
 	struct ib_mad_recv_buf	recv_buf;
+	struct list_head	rmpp_list;
 	int			mad_len;
 };
 
@@ -322,6 +438,16 @@ int ib_unregister_mad_agent(struct ib_mad_agent *mad_agent);
  * @bad_send_wr: Specifies the MAD on which an error was encountered.
  *
  * Sent MADs are not guaranteed to complete in the order that they were posted.
+ *
+ * If the MAD requires RMPP, the data buffer should contain a single copy
+ * of the common MAD, RMPP, and class specific headers, followed by the class
+ * defined data.  If the class defined data would not divide evenly into
+ * RMPP segments, then space must be allocated at the end of the referenced
+ * buffer for any required padding.  To indicate the amount of class defined
+ * data being transferred, the paylen_newwin field in the RMPP header should
+ * be set to the size of the class specific header plus the amount of class
+ * defined data being transferred.  The paylen_newwin field should be
+ * specified in network-byte order.
  */
 int ib_post_send_mad(struct ib_mad_agent *mad_agent,
 		     struct ib_send_wr *send_wr,
@@ -334,15 +460,13 @@ int ib_post_send_mad(struct ib_mad_agent *mad_agent,
  *   referenced buffer should be at least the size of the mad_len specified
  *   by @mad_recv_wc.
  *
- * This call copies a chain of received RMPP MADs into a single data buffer,
+ * This call copies a chain of received MAD segments into a single data buffer,
  * removing duplicated headers.
  */
-void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc,
-			  void *buf);
+void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc, void *buf);
 
 /**
- * ib_free_recv_mad - Returns data buffers used to receive a MAD to the
- *   access layer.
+ * ib_free_recv_mad - Returns data buffers used to receive a MAD.
  * @mad_recv_wc: Work completion information for a received MAD.
  *
  * Clients receiving MADs through their ib_mad_recv_handler must call this
@@ -358,8 +482,18 @@ void ib_free_recv_mad(struct ib_mad_recv_wc *mad_recv_wc);
  * MADs will be returned to the user through the corresponding
  * ib_mad_send_handler.
  */
-void ib_cancel_mad(struct ib_mad_agent *mad_agent,
-		   u64 wr_id);
+void ib_cancel_mad(struct ib_mad_agent *mad_agent, u64 wr_id);
+
+/**
+ * ib_modify_mad - Modifies an outstanding send MAD operation.
+ * @mad_agent: Specifies the registration associated with sent MAD.
+ * @wr_id: Indicates the work request identifier of the MAD to modify.
+ * @timeout_ms: New timeout value for sent MAD.
+ *
+ * This call will reset the timeout value for a sent MAD to the specified
+ * value.
+ */
+int ib_modify_mad(struct ib_mad_agent *mad_agent, u64 wr_id, u32 timeout_ms);
 
 /**
  * ib_redirect_mad_qp - Registers a QP for MAD services.
@@ -401,4 +535,43 @@ struct ib_mad_agent *ib_redirect_mad_qp(struct ib_qp *qp,
 int ib_process_mad_wc(struct ib_mad_agent *mad_agent,
 		      struct ib_wc *wc);
 
+/**
+ * ib_create_send_mad - Allocate and initialize a data buffer and work request
+ *   for sending a MAD.
+ * @mad_agent: Specifies the registered MAD service to associate with the MAD.
+ * @remote_qpn: Specifies the QPN of the receiving node.
+ * @pkey_index: Specifies which PKey the MAD will be sent using.  This field
+ *   is valid only if the remote_qpn is QP 1.
+ * @ah: References the address handle used to transfer to the remote node.
+ * @rmpp_active: Indicates if the send will enable RMPP.
+ * @hdr_len: Indicates the size of the data header of the MAD.  This length
+ *   should include the common MAD header, RMPP header, plus any class
+ *   specific header.
+ * @data_len: Indicates the size of any user-transferred data.  The call will
+ *   automatically adjust the allocated buffer size to account for any
+ *   additional padding that may be necessary.
+ * @gfp_mask: GFP mask used for the memory allocation.
+ *
+ * This is a helper routine that may be used to allocate a MAD.  Users are
+ * not required to allocate outbound MADs using this call.  The returned
+ * MAD send buffer will reference a data buffer usable for sending a MAD, along
+ * with an initialized work request structure.  Users may modify the returned
+ * MAD data buffer or work request before posting the send.
+ *
+ * The returned data buffer will be cleared.  Users are responsible for
+ * initializing the common MAD and any class specific headers.  If @rmpp_active
+ * is set, the RMPP header will be initialized for sending.
+ */
+struct ib_mad_send_buf * ib_create_send_mad(struct ib_mad_agent *mad_agent,
+					    u32 remote_qpn, u16 pkey_index,
+					    struct ib_ah *ah, int rmpp_active,
+					    int hdr_len, int data_len,
+					    unsigned int __nocast gfp_mask);
+
+/**
+ * ib_free_send_mad - Returns data buffers used to send a MAD.
+ * @send_buf: Previously allocated send data buffer.
+ */
+void ib_free_send_mad(struct ib_mad_send_buf *send_buf);
+
 #endif /* IB_MAD_H */
diff --git a/drivers/infiniband/include/ib_sa.h b/drivers/infiniband/include/ib_sa.h
index 00222285eb9a..6d999f7b5d93 100644
--- a/drivers/infiniband/include/ib_sa.h
+++ b/drivers/infiniband/include/ib_sa.h
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, 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
@@ -29,7 +30,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: ib_sa.h 1389 2004-12-27 22:56:47Z roland $
+ * $Id: ib_sa.h 2811 2005-07-06 18:11:43Z halr $
  */
 
 #ifndef IB_SA_H
@@ -41,9 +42,11 @@
 #include <ib_mad.h>
 
 enum {
-	IB_SA_CLASS_VERSION	= 2,	/* IB spec version 1.1/1.2 */
+	IB_SA_CLASS_VERSION		= 2,	/* IB spec version 1.1/1.2 */
 
-	IB_SA_METHOD_DELETE	= 0x15
+	IB_SA_METHOD_GET_TABLE		= 0x12,
+	IB_SA_METHOD_GET_TABLE_RESP	= 0x92,
+	IB_SA_METHOD_DELETE		= 0x15
 };
 
 enum ib_sa_selector {
@@ -87,10 +90,6 @@ static inline int ib_sa_rate_enum_to_int(enum ib_sa_rate rate)
 	}
 }
 
-typedef u64 __bitwise ib_sa_comp_mask;
-
-#define IB_SA_COMP_MASK(n)	((__force ib_sa_comp_mask) cpu_to_be64(1ull << n))
-
 /*
  * Structures for SA records are named "struct ib_sa_xxx_rec."  No
  * attempt is made to pack structures to match the physical layout of
@@ -195,6 +194,61 @@ struct ib_sa_mcmember_rec {
 	int          proxy_join;
 };
 
+/* Service Record Component Mask Sec 15.2.5.14 Ver 1.1	*/
+#define IB_SA_SERVICE_REC_SERVICE_ID			IB_SA_COMP_MASK( 0)
+#define IB_SA_SERVICE_REC_SERVICE_GID			IB_SA_COMP_MASK( 1)
+#define IB_SA_SERVICE_REC_SERVICE_PKEY			IB_SA_COMP_MASK( 2)
+/* reserved:								 3 */
+#define IB_SA_SERVICE_REC_SERVICE_LEASE			IB_SA_COMP_MASK( 4)
+#define IB_SA_SERVICE_REC_SERVICE_KEY			IB_SA_COMP_MASK( 5)
+#define IB_SA_SERVICE_REC_SERVICE_NAME			IB_SA_COMP_MASK( 6)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_0		IB_SA_COMP_MASK( 7)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_1		IB_SA_COMP_MASK( 8)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_2		IB_SA_COMP_MASK( 9)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_3		IB_SA_COMP_MASK(10)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_4		IB_SA_COMP_MASK(11)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_5		IB_SA_COMP_MASK(12)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_6		IB_SA_COMP_MASK(13)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_7		IB_SA_COMP_MASK(14)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_8		IB_SA_COMP_MASK(15)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_9		IB_SA_COMP_MASK(16)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_10		IB_SA_COMP_MASK(17)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_11		IB_SA_COMP_MASK(18)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_12		IB_SA_COMP_MASK(19)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_13		IB_SA_COMP_MASK(20)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_14		IB_SA_COMP_MASK(21)
+#define IB_SA_SERVICE_REC_SERVICE_DATA8_15		IB_SA_COMP_MASK(22)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_0		IB_SA_COMP_MASK(23)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_1		IB_SA_COMP_MASK(24)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_2		IB_SA_COMP_MASK(25)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_3		IB_SA_COMP_MASK(26)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_4		IB_SA_COMP_MASK(27)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_5		IB_SA_COMP_MASK(28)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_6		IB_SA_COMP_MASK(29)
+#define IB_SA_SERVICE_REC_SERVICE_DATA16_7		IB_SA_COMP_MASK(30)
+#define IB_SA_SERVICE_REC_SERVICE_DATA32_0		IB_SA_COMP_MASK(31)
+#define IB_SA_SERVICE_REC_SERVICE_DATA32_1		IB_SA_COMP_MASK(32)
+#define IB_SA_SERVICE_REC_SERVICE_DATA32_2		IB_SA_COMP_MASK(33)
+#define IB_SA_SERVICE_REC_SERVICE_DATA32_3		IB_SA_COMP_MASK(34)
+#define IB_SA_SERVICE_REC_SERVICE_DATA64_0		IB_SA_COMP_MASK(35)
+#define IB_SA_SERVICE_REC_SERVICE_DATA64_1		IB_SA_COMP_MASK(36)
+
+#define IB_DEFAULT_SERVICE_LEASE 	0xFFFFFFFF
+
+struct ib_sa_service_rec {
+	u64		id;
+	union ib_gid	gid;
+	u16 		pkey;
+	/* reserved */
+	u32		lease;
+	u8		key[16];
+	u8		name[64];
+	u8		data8[16];
+	u16		data16[8];
+	u32		data32[4];
+	u64		data64[2];
+};
+
 struct ib_sa_query;
 
 void ib_sa_cancel_query(int id, struct ib_sa_query *query);
@@ -202,7 +256,7 @@ void ib_sa_cancel_query(int id, struct ib_sa_query *query);
 int ib_sa_path_rec_get(struct ib_device *device, u8 port_num,
 		       struct ib_sa_path_rec *rec,
 		       ib_sa_comp_mask comp_mask,
-		       int timeout_ms, int gfp_mask,
+		       int timeout_ms, unsigned int __nocast gfp_mask,
 		       void (*callback)(int status,
 					struct ib_sa_path_rec *resp,
 					void *context),
@@ -213,13 +267,24 @@ int ib_sa_mcmember_rec_query(struct ib_device *device, u8 port_num,
 			     u8 method,
 			     struct ib_sa_mcmember_rec *rec,
 			     ib_sa_comp_mask comp_mask,
-			     int timeout_ms, int gfp_mask,
+			     int timeout_ms, unsigned int __nocast gfp_mask,
 			     void (*callback)(int status,
 					      struct ib_sa_mcmember_rec *resp,
 					      void *context),
 			     void *context,
 			     struct ib_sa_query **query);
 
+int ib_sa_service_rec_query(struct ib_device *device, u8 port_num,
+			 u8 method,
+			 struct ib_sa_service_rec *rec,
+			 ib_sa_comp_mask comp_mask,
+			 int timeout_ms, unsigned int __nocast gfp_mask,
+			 void (*callback)(int status,
+					  struct ib_sa_service_rec *resp,
+					  void *context),
+			 void *context,
+			 struct ib_sa_query **sa_query);
+
 /**
  * ib_sa_mcmember_rec_set - Start an MCMember set query
  * @device:device to send query on
@@ -248,7 +313,7 @@ static inline int
 ib_sa_mcmember_rec_set(struct ib_device *device, u8 port_num,
 		       struct ib_sa_mcmember_rec *rec,
 		       ib_sa_comp_mask comp_mask,
-		       int timeout_ms, int gfp_mask,
+		       int timeout_ms, unsigned int __nocast gfp_mask,
 		       void (*callback)(int status,
 					struct ib_sa_mcmember_rec *resp,
 					void *context),
@@ -290,7 +355,7 @@ static inline int
 ib_sa_mcmember_rec_delete(struct ib_device *device, u8 port_num,
 			  struct ib_sa_mcmember_rec *rec,
 			  ib_sa_comp_mask comp_mask,
-			  int timeout_ms, int gfp_mask,
+			  int timeout_ms, unsigned int __nocast gfp_mask,
 			  void (*callback)(int status,
 					   struct ib_sa_mcmember_rec *resp,
 					   void *context),
diff --git a/drivers/infiniband/include/ib_user_cm.h b/drivers/infiniband/include/ib_user_cm.h
new file mode 100644
index 000000000000..500b1af6ff77
--- /dev/null
+++ b/drivers/infiniband/include/ib_user_cm.h
@@ -0,0 +1,328 @@
+/*
+ * Copyright (c) 2005 Topspin Communications.  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.
+ *
+ * $Id: ib_user_cm.h 2576 2005-06-09 17:00:30Z libor $
+ */
+
+#ifndef IB_USER_CM_H
+#define IB_USER_CM_H
+
+#include <linux/types.h>
+
+#define IB_USER_CM_ABI_VERSION 1
+
+enum {
+	IB_USER_CM_CMD_CREATE_ID,
+	IB_USER_CM_CMD_DESTROY_ID,
+	IB_USER_CM_CMD_ATTR_ID,
+
+	IB_USER_CM_CMD_LISTEN,
+	IB_USER_CM_CMD_ESTABLISH,
+
+	IB_USER_CM_CMD_SEND_REQ,
+	IB_USER_CM_CMD_SEND_REP,
+	IB_USER_CM_CMD_SEND_RTU,
+	IB_USER_CM_CMD_SEND_DREQ,
+	IB_USER_CM_CMD_SEND_DREP,
+	IB_USER_CM_CMD_SEND_REJ,
+	IB_USER_CM_CMD_SEND_MRA,
+	IB_USER_CM_CMD_SEND_LAP,
+	IB_USER_CM_CMD_SEND_APR,
+	IB_USER_CM_CMD_SEND_SIDR_REQ,
+	IB_USER_CM_CMD_SEND_SIDR_REP,
+
+	IB_USER_CM_CMD_EVENT,
+};
+/*
+ * command ABI structures.
+ */
+struct ib_ucm_cmd_hdr {
+	__u32 cmd;
+	__u16 in;
+	__u16 out;
+};
+
+struct ib_ucm_create_id {
+	__u64 response;
+};
+
+struct ib_ucm_create_id_resp {
+	__u32 id;
+};
+
+struct ib_ucm_destroy_id {
+	__u32 id;
+};
+
+struct ib_ucm_attr_id {
+	__u64 response;
+	__u32 id;
+};
+
+struct ib_ucm_attr_id_resp {
+	__u64 service_id;
+	__u64 service_mask;
+	__u32 local_id;
+	__u32 remote_id;
+};
+
+struct ib_ucm_listen {
+	__u64 service_id;
+	__u64 service_mask;
+	__u32 id;
+};
+
+struct ib_ucm_establish {
+	__u32 id;
+};
+
+struct ib_ucm_private_data {
+	__u64 data;
+	__u32 id;
+	__u8  len;
+	__u8  reserved[3];
+};
+
+struct ib_ucm_path_rec {
+	__u8  dgid[16];
+	__u8  sgid[16];
+	__u16 dlid;
+	__u16 slid;
+	__u32 raw_traffic;
+	__u32 flow_label;
+	__u32 reversible;
+	__u32 mtu;
+	__u16 pkey;
+	__u8  hop_limit;
+	__u8  traffic_class;
+	__u8  numb_path;
+	__u8  sl;
+	__u8  mtu_selector;
+	__u8  rate_selector;
+	__u8  rate;
+	__u8  packet_life_time_selector;
+	__u8  packet_life_time;
+	__u8  preference;
+};
+
+struct ib_ucm_req {
+	__u32 id;
+	__u32 qpn;
+	__u32 qp_type;
+	__u32 psn;
+	__u64 sid;
+	__u64 data;
+	__u64 primary_path;
+	__u64 alternate_path;
+	__u8  len;
+	__u8  peer_to_peer;
+	__u8  responder_resources;
+	__u8  initiator_depth;
+	__u8  remote_cm_response_timeout;
+	__u8  flow_control;
+	__u8  local_cm_response_timeout;
+	__u8  retry_count;
+	__u8  rnr_retry_count;
+	__u8  max_cm_retries;
+	__u8  srq;
+	__u8  reserved[1];
+};
+
+struct ib_ucm_rep {
+	__u64 data;
+	__u32 id;
+	__u32 qpn;
+	__u32 psn;
+	__u8  len;
+	__u8  responder_resources;
+	__u8  initiator_depth;
+	__u8  target_ack_delay;
+	__u8  failover_accepted;
+	__u8  flow_control;
+	__u8  rnr_retry_count;
+	__u8  srq;
+};
+
+struct ib_ucm_info {
+	__u32 id;
+	__u32 status;
+	__u64 info;
+	__u64 data;
+	__u8  info_len;
+	__u8  data_len;
+	__u8  reserved[2];
+};
+
+struct ib_ucm_mra {
+	__u64 data;
+	__u32 id;
+	__u8  len;
+	__u8  timeout;
+	__u8  reserved[2];
+};
+
+struct ib_ucm_lap {
+	__u64 path;
+	__u64 data;
+	__u32 id;
+	__u8  len;
+	__u8  reserved[3];
+};
+
+struct ib_ucm_sidr_req {
+	__u32 id;
+	__u32 timeout;
+	__u64 sid;
+	__u64 data;
+	__u64 path;
+	__u16 pkey;
+	__u8  len;
+	__u8  max_cm_retries;
+};
+
+struct ib_ucm_sidr_rep {
+	__u32 id;
+	__u32 qpn;
+	__u32 qkey;
+	__u32 status;
+	__u64 info;
+	__u64 data;
+	__u8  info_len;
+	__u8  data_len;
+	__u8  reserved[2];
+};
+/*
+ * event notification ABI structures.
+ */
+struct ib_ucm_event_get {
+	__u64 response;
+	__u64 data;
+	__u64 info;
+	__u8  data_len;
+	__u8  info_len;
+	__u8  reserved[2];
+};
+
+struct ib_ucm_req_event_resp {
+	__u32                  listen_id;
+	/* device */
+	/* port */
+	struct ib_ucm_path_rec primary_path;
+	struct ib_ucm_path_rec alternate_path;
+	__u64                  remote_ca_guid;
+	__u32                  remote_qkey;
+	__u32                  remote_qpn;
+	__u32                  qp_type;
+	__u32                  starting_psn;
+	__u8  responder_resources;
+	__u8  initiator_depth;
+	__u8  local_cm_response_timeout;
+	__u8  flow_control;
+	__u8  remote_cm_response_timeout;
+	__u8  retry_count;
+	__u8  rnr_retry_count;
+	__u8  srq;
+};
+
+struct ib_ucm_rep_event_resp {
+	__u64 remote_ca_guid;
+	__u32 remote_qkey;
+	__u32 remote_qpn;
+	__u32 starting_psn;
+	__u8  responder_resources;
+	__u8  initiator_depth;
+	__u8  target_ack_delay;
+	__u8  failover_accepted;
+	__u8  flow_control;
+	__u8  rnr_retry_count;
+	__u8  srq;
+	__u8  reserved[1];
+};
+
+struct ib_ucm_rej_event_resp {
+	__u32 reason;
+	/* ari in ib_ucm_event_get info field. */
+};
+
+struct ib_ucm_mra_event_resp {
+	__u8  timeout;
+	__u8  reserved[3];
+};
+
+struct ib_ucm_lap_event_resp {
+	struct ib_ucm_path_rec path;
+};
+
+struct ib_ucm_apr_event_resp {
+	__u32 status;
+	/* apr info in ib_ucm_event_get info field. */
+};
+
+struct ib_ucm_sidr_req_event_resp {
+	__u32 listen_id;
+	/* device */
+	/* port */
+	__u16 pkey;
+	__u8  reserved[2];
+};
+
+struct ib_ucm_sidr_rep_event_resp {
+	__u32 status;
+	__u32 qkey;
+	__u32 qpn;
+	/* info in ib_ucm_event_get info field. */
+};
+
+#define IB_UCM_PRES_DATA      0x01
+#define IB_UCM_PRES_INFO      0x02
+#define IB_UCM_PRES_PRIMARY   0x04
+#define IB_UCM_PRES_ALTERNATE 0x08
+
+struct ib_ucm_event_resp {
+	__u32 id;
+	__u32 event;
+	__u32 present;
+	union {
+		struct ib_ucm_req_event_resp req_resp;
+		struct ib_ucm_rep_event_resp rep_resp;
+		struct ib_ucm_rej_event_resp rej_resp;
+		struct ib_ucm_mra_event_resp mra_resp;
+		struct ib_ucm_lap_event_resp lap_resp;
+		struct ib_ucm_apr_event_resp apr_resp;
+
+		struct ib_ucm_sidr_req_event_resp sidr_req_resp;
+		struct ib_ucm_sidr_rep_event_resp sidr_rep_resp;
+
+		__u32                             send_status;
+	} u;
+};
+
+#endif /* IB_USER_CM_H */
diff --git a/drivers/infiniband/include/ib_user_mad.h b/drivers/infiniband/include/ib_user_mad.h
index 06ad4a6075fa..a9a56b50aacc 100644
--- a/drivers/infiniband/include/ib_user_mad.h
+++ b/drivers/infiniband/include/ib_user_mad.h
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Voltaire, 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
@@ -29,7 +30,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  *
- * $Id: ib_user_mad.h 1389 2004-12-27 22:56:47Z roland $
+ * $Id: ib_user_mad.h 2814 2005-07-06 19:14:09Z halr $
  */
 
 #ifndef IB_USER_MAD_H
@@ -42,7 +43,7 @@
  * Increment this value if any changes that break userspace ABI
  * compatibility are made.
  */
-#define IB_USER_MAD_ABI_VERSION	2
+#define IB_USER_MAD_ABI_VERSION	5
 
 /*
  * Make sure that all structs defined in this file remain laid out so
@@ -51,13 +52,13 @@
  */
 
 /**
- * ib_user_mad - MAD packet
- * @data - Contents of MAD
+ * ib_user_mad_hdr - MAD packet header
  * @id - ID of agent MAD received with/to be sent with
  * @status - 0 on successful receive, ETIMEDOUT if no response
  *   received (transaction ID in data[] will be set to TID of original
  *   request) (ignored on send)
  * @timeout_ms - Milliseconds to wait for response (unset on receive)
+ * @retries - Number of automatic retries to attempt
  * @qpn - Remote QP number received from/to be sent to
  * @qkey - Remote Q_Key to be sent with (unset on receive)
  * @lid - Remote lid received from/to be sent to
@@ -72,11 +73,12 @@
  *
  * All multi-byte quantities are stored in network (big endian) byte order.
  */
-struct ib_user_mad {
-	__u8	data[256];
+struct ib_user_mad_hdr {
 	__u32	id;
 	__u32	status;
 	__u32	timeout_ms;
+	__u32	retries;
+	__u32	length;
 	__u32	qpn;
 	__u32   qkey;
 	__u16	lid;
@@ -91,6 +93,17 @@ struct ib_user_mad {
 };
 
 /**
+ * ib_user_mad - MAD packet
+ * @hdr - MAD packet header
+ * @data - Contents of MAD
+ *
+ */
+struct ib_user_mad {
+	struct ib_user_mad_hdr hdr;
+	__u8	data[0];
+};
+
+/**
  * ib_user_mad_reg_req - MAD registration request
  * @id - Set by the kernel; used to identify agent in future requests.
  * @qpn - Queue pair number; must be 0 or 1.
@@ -103,6 +116,8 @@ struct ib_user_mad {
  *   management class to receive.
  * @oui: Indicates IEEE OUI when mgmt_class is a vendor class
  *   in the range from 0x30 to 0x4f. Otherwise not used.
+ * @rmpp_version: If set, indicates the RMPP version used.
+ *
  */
 struct ib_user_mad_reg_req {
 	__u32	id;
@@ -111,6 +126,7 @@ struct ib_user_mad_reg_req {
 	__u8	mgmt_class;
 	__u8	mgmt_class_version;
 	__u8    oui[3];
+	__u8	rmpp_version;
 };
 
 #define IB_IOCTL_MAGIC		0x1b
diff --git a/drivers/infiniband/include/ib_verbs.h b/drivers/infiniband/include/ib_verbs.h
index e5bd9a10c201..5d24edaa66e6 100644
--- a/drivers/infiniband/include/ib_verbs.h
+++ b/drivers/infiniband/include/ib_verbs.h
@@ -289,6 +289,15 @@ struct ib_global_route {
 	u8		traffic_class;
 };
 
+struct ib_grh {
+	u32		version_tclass_flow;
+	u16		paylen;
+	u8		next_hdr;
+	u8		hop_limit;
+	union ib_gid	sgid;
+	union ib_gid	dgid;
+};
+
 enum {
 	IB_MULTICAST_QPN = 0xffffff
 };
@@ -566,6 +575,7 @@ struct ib_send_wr {
 			u32	remote_qpn;
 			u32	remote_qkey;
 			int	timeout_ms; /* valid for MADs only */
+			int	retries;    /* valid for MADs only */
 			u16	pkey_index; /* valid for GSI only */
 			u8	port_num;   /* valid for DR SMPs on switch only */
 		} ud;
@@ -990,6 +1000,21 @@ int ib_dealloc_pd(struct ib_pd *pd);
 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
 
 /**
+ * ib_create_ah_from_wc - Creates an address handle associated with the
+ *   sender of the specified work completion.
+ * @pd: The protection domain associated with the address handle.
+ * @wc: Work completion information associated with a received message.
+ * @grh: References the received global route header.  This parameter is
+ *   ignored unless the work completion indicates that the GRH is valid.
+ * @port_num: The outbound port number to associate with the address.
+ *
+ * The address handle is used to reference a local or global destination
+ * in all UD QP post sends.
+ */
+struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, struct ib_wc *wc,
+				   struct ib_grh *grh, u8 port_num);
+
+/**
  * ib_modify_ah - Modifies the address vector associated with an address
  *   handle.
  * @ah: The address handle to modify.