5 files changed, 291 insertions, 84 deletions

diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index 287ad0564acd..82a7dd87089b 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -891,7 +891,7 @@ int mlx4_ib_mad_init(struct mlx4_ib_dev *dev)
 				agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
 							      q ? IB_QPT_GSI : IB_QPT_SMI,
 							      NULL, 0, send_handler,
-							      NULL, NULL);
+							      NULL, NULL, 0);
 				if (IS_ERR(agent)) {
 					ret = PTR_ERR(agent);
 					goto err;
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index 0f7027e7db13..bda5994ceb68 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -59,6 +59,7 @@
 
 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
+#define MLX4_IB_CARD_REV_A0   0xA0
 
 MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
@@ -119,6 +120,17 @@ static int check_flow_steering_support(struct mlx4_dev *dev)
 	return dmfs;
 }
 
+static int num_ib_ports(struct mlx4_dev *dev)
+{
+	int ib_ports = 0;
+	int i;
+
+	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
+		ib_ports++;
+
+	return ib_ports;
+}
+
 static int mlx4_ib_query_device(struct ib_device *ibdev,
 				struct ib_device_attr *props)
 {
@@ -126,6 +138,7 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
 	int err = -ENOMEM;
+	int have_ib_ports;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -142,6 +155,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 
 	memset(props, 0, sizeof *props);
 
+	have_ib_ports = num_ib_ports(dev->dev);
+
 	props->fw_ver = dev->dev->caps.fw_ver;
 	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 		IB_DEVICE_PORT_ACTIVE_EVENT		|
@@ -152,13 +167,15 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
-	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
+	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
-	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
+	if (dev->dev->caps.max_gso_sz &&
+	    (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
+	    (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
 	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
@@ -357,7 +374,7 @@ static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 	props->state		= IB_PORT_DOWN;
 	props->phys_state	= state_to_phys_state(props->state);
 	props->active_mtu	= IB_MTU_256;
-	spin_lock(&iboe->lock);
+	spin_lock_bh(&iboe->lock);
 	ndev = iboe->netdevs[port - 1];
 	if (!ndev)
 		goto out_unlock;
@@ -369,7 +386,7 @@ static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 					IB_PORT_ACTIVE : IB_PORT_DOWN;
 	props->phys_state	= state_to_phys_state(props->state);
 out_unlock:
-	spin_unlock(&iboe->lock);
+	spin_unlock_bh(&iboe->lock);
 out:
 	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 	return err;
@@ -811,11 +828,11 @@ int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 	if (!mqp->port)
 		return 0;
 
-	spin_lock(&mdev->iboe.lock);
+	spin_lock_bh(&mdev->iboe.lock);
 	ndev = mdev->iboe.netdevs[mqp->port - 1];
 	if (ndev)
 		dev_hold(ndev);
-	spin_unlock(&mdev->iboe.lock);
+	spin_unlock_bh(&mdev->iboe.lock);
 
 	if (ndev) {
 		ret = 1;
@@ -910,8 +927,7 @@ static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
 	const struct default_rules *pdefault_rules = default_table;
 	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
 
-	for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
-	     pdefault_rules++) {
+	for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
 		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
 		memset(&field_types, 0, sizeof(field_types));
 
@@ -965,8 +981,7 @@ static int __mlx4_ib_create_default_rules(
 	int size = 0;
 	int i;
 
-	for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
-			sizeof(pdefault_rules->rules_create_list[0]); i++) {
+	for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
 		int ret;
 		union ib_flow_spec ib_spec;
 		switch (pdefault_rules->rules_create_list[i]) {
@@ -1091,6 +1106,30 @@ static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
 	return err;
 }
 
+static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
+				    u64 *reg_id)
+{
+	void *ib_flow;
+	union ib_flow_spec *ib_spec;
+	struct mlx4_dev	*dev = to_mdev(qp->device)->dev;
+	int err = 0;
+
+	if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
+		return 0; /* do nothing */
+
+	ib_flow = flow_attr + 1;
+	ib_spec = (union ib_flow_spec *)ib_flow;
+
+	if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
+		return 0; /* do nothing */
+
+	err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
+				    flow_attr->port, qp->qp_num,
+				    MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
+				    reg_id);
+	return err;
+}
+
 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 				    struct ib_flow_attr *flow_attr,
 				    int domain)
@@ -1138,6 +1177,12 @@ static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 		i++;
 	}
 
+	if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
+		err = mlx4_ib_tunnel_steer_add(qp, flow_attr, &mflow->reg_id[i]);
+		if (err)
+			goto err_free;
+	}
+
 	return &mflow->ibflow;
 
 err_free:
@@ -1264,11 +1309,11 @@ static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	mutex_lock(&mqp->mutex);
 	ge = find_gid_entry(mqp, gid->raw);
 	if (ge) {
-		spin_lock(&mdev->iboe.lock);
+		spin_lock_bh(&mdev->iboe.lock);
 		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
 		if (ndev)
 			dev_hold(ndev);
-		spin_unlock(&mdev->iboe.lock);
+		spin_unlock_bh(&mdev->iboe.lock);
 		if (ndev)
 			dev_put(ndev);
 		list_del(&ge->list);
@@ -1389,6 +1434,9 @@ static void update_gids_task(struct work_struct *work)
 	int err;
 	struct mlx4_dev	*dev = gw->dev->dev;
 
+	if (!gw->dev->ib_active)
+		return;
+
 	mailbox = mlx4_alloc_cmd_mailbox(dev);
 	if (IS_ERR(mailbox)) {
 		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
@@ -1419,6 +1467,9 @@ static void reset_gids_task(struct work_struct *work)
 	int err;
 	struct mlx4_dev	*dev = gw->dev->dev;
 
+	if (!gw->dev->ib_active)
+		return;
+
 	mailbox = mlx4_alloc_cmd_mailbox(dev);
 	if (IS_ERR(mailbox)) {
 		pr_warn("reset gid table failed\n");
@@ -1553,7 +1604,7 @@ static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
 		return 0;
 
 	iboe = &ibdev->iboe;
-	spin_lock(&iboe->lock);
+	spin_lock_bh(&iboe->lock);
 
 	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
 		if ((netif_is_bond_master(real_dev) &&
@@ -1563,7 +1614,7 @@ static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
 			update_gid_table(ibdev, port, gid,
 					 event == NETDEV_DOWN, 0);
 
-	spin_unlock(&iboe->lock);
+	spin_unlock_bh(&iboe->lock);
 	return 0;
 
 }
@@ -1636,13 +1687,21 @@ static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
 	new_smac = mlx4_mac_to_u64(dev->dev_addr);
 	read_unlock(&dev_base_lock);
 
+	atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
+
+	/* no need for update QP1 and mac registration in non-SRIOV */
+	if (!mlx4_is_mfunc(ibdev->dev))
+		return;
+
 	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
 	qp = ibdev->qp1_proxy[port - 1];
 	if (qp) {
 		int new_smac_index;
-		u64 old_smac = qp->pri.smac;
+		u64 old_smac;
 		struct mlx4_update_qp_params update_params;
 
+		mutex_lock(&qp->mutex);
+		old_smac = qp->pri.smac;
 		if (new_smac == old_smac)
 			goto unlock;
 
@@ -1652,22 +1711,25 @@ static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
 			goto unlock;
 
 		update_params.smac_index = new_smac_index;
-		if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC,
+		if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
 				   &update_params)) {
 			release_mac = new_smac;
 			goto unlock;
 		}
-
+		/* if old port was zero, no mac was yet registered for this QP */
+		if (qp->pri.smac_port)
+			release_mac = old_smac;
 		qp->pri.smac = new_smac;
+		qp->pri.smac_port = port;
 		qp->pri.smac_index = new_smac_index;
-
-		release_mac = old_smac;
 	}
 
 unlock:
-	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
 	if (release_mac != MLX4_IB_INVALID_MAC)
 		mlx4_unregister_mac(ibdev->dev, port, release_mac);
+	if (qp)
+		mutex_unlock(&qp->mutex);
+	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
 }
 
 static void mlx4_ib_get_dev_addr(struct net_device *dev,
@@ -1678,6 +1740,7 @@ static void mlx4_ib_get_dev_addr(struct net_device *dev,
 	struct inet6_dev *in6_dev;
 	union ib_gid  *pgid;
 	struct inet6_ifaddr *ifp;
+	union ib_gid default_gid;
 #endif
 	union ib_gid gid;
 
@@ -1698,12 +1761,15 @@ static void mlx4_ib_get_dev_addr(struct net_device *dev,
 		in_dev_put(in_dev);
 	}
 #if IS_ENABLED(CONFIG_IPV6)
+	mlx4_make_default_gid(dev, &default_gid);
 	/* IPv6 gids */
 	in6_dev = in6_dev_get(dev);
 	if (in6_dev) {
 		read_lock_bh(&in6_dev->lock);
 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
 			pgid = (union ib_gid *)&ifp->addr;
+			if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
+				continue;
 			update_gid_table(ibdev, port, pgid, 0, 0);
 		}
 		read_unlock_bh(&in6_dev->lock);
@@ -1725,24 +1791,33 @@ static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
 	struct	net_device *dev;
 	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 	int i;
+	int err = 0;
 
-	for (i = 1; i <= ibdev->num_ports; ++i)
-		if (reset_gid_table(ibdev, i))
-			return -1;
+	for (i = 1; i <= ibdev->num_ports; ++i) {
+		if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
+		    IB_LINK_LAYER_ETHERNET) {
+			err = reset_gid_table(ibdev, i);
+			if (err)
+				goto out;
+		}
+	}
 
 	read_lock(&dev_base_lock);
-	spin_lock(&iboe->lock);
+	spin_lock_bh(&iboe->lock);
 
 	for_each_netdev(&init_net, dev) {
 		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
-		if (port)
+		/* port will be non-zero only for ETH ports */
+		if (port) {
+			mlx4_ib_set_default_gid(ibdev, dev, port);
 			mlx4_ib_get_dev_addr(dev, ibdev, port);
+		}
 	}
 
-	spin_unlock(&iboe->lock);
+	spin_unlock_bh(&iboe->lock);
 	read_unlock(&dev_base_lock);
-
-	return 0;
+out:
+	return err;
 }
 
 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
@@ -1756,7 +1831,7 @@ static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
 
 	iboe = &ibdev->iboe;
 
-	spin_lock(&iboe->lock);
+	spin_lock_bh(&iboe->lock);
 	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
 		enum ib_port_state	port_state = IB_PORT_NOP;
 		struct net_device *old_master = iboe->masters[port - 1];
@@ -1788,35 +1863,47 @@ static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
 			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
 						IB_PORT_ACTIVE : IB_PORT_DOWN;
 			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
-		} else {
-			reset_gid_table(ibdev, port);
-		}
-		/* if using bonding/team and a slave port is down, we don't the bond IP
-		 * based gids in the table since flows that select port by gid may get
-		 * the down port.
-		 */
-		if (curr_master && (port_state == IB_PORT_DOWN)) {
-			reset_gid_table(ibdev, port);
-			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
-		}
-		/* if bonding is used it is possible that we add it to masters
-		 * only after IP address is assigned to the net bonding
-		 * interface.
-		*/
-		if (curr_master && (old_master != curr_master)) {
-			reset_gid_table(ibdev, port);
-			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
-			mlx4_ib_get_dev_addr(curr_master, ibdev, port);
-		}
+			if (curr_master) {
+				/* if using bonding/team and a slave port is down, we
+				 * don't want the bond IP based gids in the table since
+				 * flows that select port by gid may get the down port.
+				*/
+				if (port_state == IB_PORT_DOWN) {
+					reset_gid_table(ibdev, port);
+					mlx4_ib_set_default_gid(ibdev,
+								curr_netdev,
+								port);
+				} else {
+					/* gids from the upper dev (bond/team)
+					 * should appear in port's gid table
+					*/
+					mlx4_ib_get_dev_addr(curr_master,
+							     ibdev, port);
+				}
+			}
+			/* if bonding is used it is possible that we add it to
+			 * masters only after IP address is assigned to the
+			 * net bonding interface.
+			*/
+			if (curr_master && (old_master != curr_master)) {
+				reset_gid_table(ibdev, port);
+				mlx4_ib_set_default_gid(ibdev,
+							curr_netdev, port);
+				mlx4_ib_get_dev_addr(curr_master, ibdev, port);
+			}
 
-		if (!curr_master && (old_master != curr_master)) {
+			if (!curr_master && (old_master != curr_master)) {
+				reset_gid_table(ibdev, port);
+				mlx4_ib_set_default_gid(ibdev,
+							curr_netdev, port);
+				mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
+			}
+		} else {
 			reset_gid_table(ibdev, port);
-			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
-			mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
 		}
 	}
 
-	spin_unlock(&iboe->lock);
+	spin_unlock_bh(&iboe->lock);
 
 	if (update_qps_port > 0)
 		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
@@ -2007,6 +2094,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
 		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
 		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
+		(1ull << IB_USER_VERBS_CMD_REREG_MR)		|
 		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
 		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
 		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
@@ -2059,6 +2147,7 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->ib_dev.req_notify_cq	= mlx4_ib_arm_cq;
 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
+	ibdev->ib_dev.rereg_user_mr	= mlx4_ib_rereg_user_mr;
 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
 	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
 	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
@@ -2156,6 +2245,9 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 			goto err_steer_free_bitmap;
 	}
 
+	for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
+		atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
+
 	if (ib_register_device(&ibdev->ib_dev, NULL))
 		goto err_steer_free_bitmap;
 
@@ -2192,12 +2284,8 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 			}
 		}
 #endif
-		for (i = 1 ; i <= ibdev->num_ports ; ++i)
-			reset_gid_table(ibdev, i);
-		rtnl_lock();
-		mlx4_ib_scan_netdevs(ibdev, NULL, 0);
-		rtnl_unlock();
-		mlx4_ib_init_gid_table(ibdev);
+		if (mlx4_ib_init_gid_table(ibdev))
+			goto err_notif;
 	}
 
 	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
@@ -2345,6 +2433,9 @@ static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
 	int p;
 
+	ibdev->ib_active = false;
+	flush_workqueue(wq);
+
 	mlx4_ib_close_sriov(ibdev);
 	mlx4_ib_mad_cleanup(ibdev);
 	ib_unregister_device(&ibdev->ib_dev);
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index 369da3ca5d64..6eb743f65f6f 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -451,6 +451,7 @@ struct mlx4_ib_iboe {
 	spinlock_t		lock;
 	struct net_device      *netdevs[MLX4_MAX_PORTS];
 	struct net_device      *masters[MLX4_MAX_PORTS];
+	atomic64_t		mac[MLX4_MAX_PORTS];
 	struct notifier_block 	nb;
 	struct notifier_block	nb_inet;
 	struct notifier_block	nb_inet6;
@@ -788,5 +789,9 @@ int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn);
 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count);
 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 			 int is_attach);
+int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags,
+			  u64 start, u64 length, u64 virt_addr,
+			  int mr_access_flags, struct ib_pd *pd,
+			  struct ib_udata *udata);
 
 #endif /* MLX4_IB_H */
diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index cb2a8727f3fb..8f9325cfc85d 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -144,8 +144,10 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	if (!mr)
 		return ERR_PTR(-ENOMEM);
 
+	/* Force registering the memory as writable. */
+	/* Used for memory re-registeration. HCA protects the access */
 	mr->umem = ib_umem_get(pd->uobject->context, start, length,
-			       access_flags, 0);
+			       access_flags | IB_ACCESS_LOCAL_WRITE, 0);
 	if (IS_ERR(mr->umem)) {
 		err = PTR_ERR(mr->umem);
 		goto err_free;
@@ -183,6 +185,93 @@ err_free:
 	return ERR_PTR(err);
 }
 
+int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags,
+			  u64 start, u64 length, u64 virt_addr,
+			  int mr_access_flags, struct ib_pd *pd,
+			  struct ib_udata *udata)
+{
+	struct mlx4_ib_dev *dev = to_mdev(mr->device);
+	struct mlx4_ib_mr *mmr = to_mmr(mr);
+	struct mlx4_mpt_entry *mpt_entry;
+	struct mlx4_mpt_entry **pmpt_entry = &mpt_entry;
+	int err;
+
+	/* Since we synchronize this call and mlx4_ib_dereg_mr via uverbs,
+	 * we assume that the calls can't run concurrently. Otherwise, a
+	 * race exists.
+	 */
+	err =  mlx4_mr_hw_get_mpt(dev->dev, &mmr->mmr, &pmpt_entry);
+
+	if (err)
+		return err;
+
+	if (flags & IB_MR_REREG_PD) {
+		err = mlx4_mr_hw_change_pd(dev->dev, *pmpt_entry,
+					   to_mpd(pd)->pdn);
+
+		if (err)
+			goto release_mpt_entry;
+	}
+
+	if (flags & IB_MR_REREG_ACCESS) {
+		err = mlx4_mr_hw_change_access(dev->dev, *pmpt_entry,
+					       convert_access(mr_access_flags));
+
+		if (err)
+			goto release_mpt_entry;
+	}
+
+	if (flags & IB_MR_REREG_TRANS) {
+		int shift;
+		int err;
+		int n;
+
+		mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
+		ib_umem_release(mmr->umem);
+		mmr->umem = ib_umem_get(mr->uobject->context, start, length,
+					mr_access_flags |
+					IB_ACCESS_LOCAL_WRITE,
+					0);
+		if (IS_ERR(mmr->umem)) {
+			err = PTR_ERR(mmr->umem);
+			/* Prevent mlx4_ib_dereg_mr from free'ing invalid pointer */
+			mmr->umem = NULL;
+			goto release_mpt_entry;
+		}
+		n = ib_umem_page_count(mmr->umem);
+		shift = ilog2(mmr->umem->page_size);
+
+		err = mlx4_mr_rereg_mem_write(dev->dev, &mmr->mmr,
+					      virt_addr, length, n, shift,
+					      *pmpt_entry);
+		if (err) {
+			ib_umem_release(mmr->umem);
+			goto release_mpt_entry;
+		}
+		mmr->mmr.iova       = virt_addr;
+		mmr->mmr.size       = length;
+
+		err = mlx4_ib_umem_write_mtt(dev, &mmr->mmr.mtt, mmr->umem);
+		if (err) {
+			mlx4_mr_rereg_mem_cleanup(dev->dev, &mmr->mmr);
+			ib_umem_release(mmr->umem);
+			goto release_mpt_entry;
+		}
+	}
+
+	/* If we couldn't transfer the MR to the HCA, just remember to
+	 * return a failure. But dereg_mr will free the resources.
+	 */
+	err = mlx4_mr_hw_write_mpt(dev->dev, &mmr->mmr, pmpt_entry);
+	if (!err && flags & IB_MR_REREG_ACCESS)
+		mmr->mmr.access = mr_access_flags;
+
+release_mpt_entry:
+	mlx4_mr_hw_put_mpt(dev->dev, pmpt_entry);
+
+	return err;
+}
+
 int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
 {
 	struct mlx4_ib_mr *mr = to_mmr(ibmr);
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 67780452f0cf..9c5150c3cb31 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -964,9 +964,10 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 				   MLX4_QP_STATE_RST, NULL, 0, 0, &qp->mqp))
 			pr_warn("modify QP %06x to RESET failed.\n",
 			       qp->mqp.qpn);
-		if (qp->pri.smac) {
+		if (qp->pri.smac || (!qp->pri.smac && qp->pri.smac_port)) {
 			mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
 			qp->pri.smac = 0;
+			qp->pri.smac_port = 0;
 		}
 		if (qp->alt.smac) {
 			mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
@@ -1325,7 +1326,8 @@ static int _mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah,
 		 * If one was already assigned, but the new mac differs,
 		 * unregister the old one and register the new one.
 		*/
-		if (!smac_info->smac || smac_info->smac != smac) {
+		if ((!smac_info->smac && !smac_info->smac_port) ||
+		    smac_info->smac != smac) {
 			/* register candidate now, unreg if needed, after success */
 			smac_index = mlx4_register_mac(dev->dev, port, smac);
 			if (smac_index >= 0) {
@@ -1390,21 +1392,13 @@ static void update_mcg_macs(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 static int handle_eth_ud_smac_index(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, u8 *smac,
 				    struct mlx4_qp_context *context)
 {
-	struct net_device *ndev;
 	u64 u64_mac;
 	int smac_index;
 
-
-	ndev = dev->iboe.netdevs[qp->port - 1];
-	if (ndev) {
-		smac = ndev->dev_addr;
-		u64_mac = mlx4_mac_to_u64(smac);
-	} else {
-		u64_mac = dev->dev->caps.def_mac[qp->port];
-	}
+	u64_mac = atomic64_read(&dev->iboe.mac[qp->port - 1]);
 
 	context->pri_path.sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE | ((qp->port - 1) << 6);
-	if (!qp->pri.smac) {
+	if (!qp->pri.smac && !qp->pri.smac_port) {
 		smac_index = mlx4_register_mac(dev->dev, qp->port, u64_mac);
 		if (smac_index >= 0) {
 			qp->pri.candidate_smac_index = smac_index;
@@ -1432,6 +1426,12 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	int steer_qp = 0;
 	int err = -EINVAL;
 
+	/* APM is not supported under RoCE */
+	if (attr_mask & IB_QP_ALT_PATH &&
+	    rdma_port_get_link_layer(&dev->ib_dev, qp->port) ==
+	    IB_LINK_LAYER_ETHERNET)
+		return -ENOTSUPP;
+
 	context = kzalloc(sizeof *context, GFP_KERNEL);
 	if (!context)
 		return -ENOMEM;
@@ -1677,9 +1677,15 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		}
 	}
 
-	if (qp->ibqp.qp_type == IB_QPT_RAW_PACKET)
+	if (qp->ibqp.qp_type == IB_QPT_RAW_PACKET) {
 		context->pri_path.ackto = (context->pri_path.ackto & 0xf8) |
 					MLX4_IB_LINK_TYPE_ETH;
+		if (dev->dev->caps.tunnel_offload_mode ==  MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
+			/* set QP to receive both tunneled & non-tunneled packets */
+			if (!(context->flags & cpu_to_be32(1 << MLX4_RSS_QPC_FLAG_OFFSET)))
+				context->srqn = cpu_to_be32(7 << 28);
+		}
+	}
 
 	if (ibqp->qp_type == IB_QPT_UD && (new_state == IB_QPS_RTR)) {
 		int is_eth = rdma_port_get_link_layer(
@@ -1780,9 +1786,10 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 			if (qp->flags & MLX4_IB_QP_NETIF)
 				mlx4_ib_steer_qp_reg(dev, qp, 0);
 		}
-		if (qp->pri.smac) {
+		if (qp->pri.smac || (!qp->pri.smac && qp->pri.smac_port)) {
 			mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
 			qp->pri.smac = 0;
+			qp->pri.smac_port = 0;
 		}
 		if (qp->alt.smac) {
 			mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
@@ -1806,11 +1813,12 @@ out:
 	if (err && steer_qp)
 		mlx4_ib_steer_qp_reg(dev, qp, 0);
 	kfree(context);
-	if (qp->pri.candidate_smac) {
+	if (qp->pri.candidate_smac ||
+	    (!qp->pri.candidate_smac && qp->pri.candidate_smac_port)) {
 		if (err) {
 			mlx4_unregister_mac(dev->dev, qp->pri.candidate_smac_port, qp->pri.candidate_smac);
 		} else {
-			if (qp->pri.smac)
+			if (qp->pri.smac || (!qp->pri.smac && qp->pri.smac_port))
 				mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
 			qp->pri.smac = qp->pri.candidate_smac;
 			qp->pri.smac_index = qp->pri.candidate_smac_index;
@@ -2083,6 +2091,16 @@ static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
 	return 0;
 }
 
+static void mlx4_u64_to_smac(u8 *dst_mac, u64 src_mac)
+{
+	int i;
+
+	for (i = ETH_ALEN; i; i--) {
+		dst_mac[i - 1] = src_mac & 0xff;
+		src_mac >>= 8;
+	}
+}
+
 static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 			    void *wqe, unsigned *mlx_seg_len)
 {
@@ -2197,7 +2215,6 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 	}
 
 	if (is_eth) {
-		u8 *smac;
 		struct in6_addr in6;
 
 		u16 pcp = (be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 29) << 13;
@@ -2210,12 +2227,17 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 		memcpy(&ctrl->imm, ah->av.eth.mac + 2, 4);
 		memcpy(&in6, sgid.raw, sizeof(in6));
 
-		if (!mlx4_is_mfunc(to_mdev(ib_dev)->dev))
-			smac = to_mdev(sqp->qp.ibqp.device)->
-				iboe.netdevs[sqp->qp.port - 1]->dev_addr;
-		else	/* use the src mac of the tunnel */
-			smac = ah->av.eth.s_mac;
-		memcpy(sqp->ud_header.eth.smac_h, smac, 6);
+		if (!mlx4_is_mfunc(to_mdev(ib_dev)->dev)) {
+			u64 mac = atomic64_read(&to_mdev(ib_dev)->iboe.mac[sqp->qp.port - 1]);
+			u8 smac[ETH_ALEN];
+
+			mlx4_u64_to_smac(smac, mac);
+			memcpy(sqp->ud_header.eth.smac_h, smac, ETH_ALEN);
+		} else {
+			/* use the src mac of the tunnel */
+			memcpy(sqp->ud_header.eth.smac_h, ah->av.eth.s_mac, ETH_ALEN);
+		}
+
 		if (!memcmp(sqp->ud_header.eth.smac_h, sqp->ud_header.eth.dmac_h, 6))
 			mlx->flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
 		if (!is_vlan) {