[RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Erez Shitrit]

    The IPoIB protocol encapsulates IP packets over Infiniband datagrams.
    As a direct RDMA Upper Layer Protocol (ULP), IPoIB cannot support HW
    features that are specific to the IP protocol stack.

    Nevertheless, RDMA interfaces have been extended to support some of the
    prominent IP offload features, such as TCP/UDP checksum and TSO.
    This provided reasonable performance gain for IPoIB but is still
    insufficient to cope with the increasing network bandwidth demand.

    However, New features are exisiting in common network interfaces that
    are very hard to implement in IPoIB interfaces while it uses the RDMA
    layer, examples include TSS and RSS, tunneling offloads, and XDP.
    Rather than continuously porting IP network interface developments into
    the RDMA stack, we propose adding an abstract network data-path interfaces
    to RDMA devices.

    In order to present a consistent interface to users, the IPoIB ULP
    continues to represent the network device to the IP stack.
    The common code also manages the IPoIB control plane, such as resolving
    path queries and registering to multicast groups.
    Data path operations are forwarded to devices that implement the new
    API, or fallback to the standard implementation otherwise.
    Using the forgoing approach, we show how IPoIB closes the performance
    gap compared to state-of-the-art Ethernet network interfaces.

    The implementation idea is to expose a struct that has data members and set
    of functions that are used for network interfaces, like create, delete, init hw
    resources, send, and attach/detach multicast to qp.
    That set of functions encapsulates in new struct, and this struct can or
    can't be given by the specific HW layer.

    The IPoIB code will be adapted to enable the option of accelerating the
    network interface, but the code will work as before if the HW below
    doesn't support the acceleration.
    Each HW vendor can supply the acceleration for the IPoIB or to leave
    IPoIB to work as before.

   TODO:
        1.change the send api in order to move it to the ndo start_xmit (unless it hurts the performance of the default driver)
        2.Take out the ipoib_ah from the send signature and use ib_ah instead, no need with including ipoib.h
        3.Check if/how to add rdma_netdev layer to the default ipoib
        4. splitting out the bulk rename of ipoib_priv into a single patch
        5. change the name of the header to be ipoib_rn.h
        6. no need to pass qkey, it is in the ah struct.

Changes fron v0:
---------------
1. Use the vnic/hfi API as a base for the new design/impl.
2. Change the low level driver to support the new struct.


Changes fron v1:
---------------
1.Add hca to rdma_netdev
2.Take out qp_num and context from rdma_netdev
3.Move dev_init/dev_cleanup to be part of the ndo's (ndo_init/ndo_uninit)
4.mlid instead of lid in mcast funcs
5.Arrange the code to return ENOTSUPP when needed
6.No dev->ib_dev.free_rdma_netdev while it is empty.
7.No need to pass the size of struct ipoib_rdma_netdev to the low-level driver


Erez Shitrit (6):
  IB/ipoib: Separate control and data related initializations
  IB/ipoib: separate control from HW operation on ipoib_open/stop ndo
  IB/ipoib: Rename qpn to dqpn in ipoib_send and post_send functions
  IB/verb: Add ipoib_options struct and API
  IB/ipoib: Support ipoib acceleration options callbacks
  mlx5_ib: skeleton for mlx5_ib to support ipoib_ops

-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC v2 for accelerated IPoIB 01/6] IB/ipoib: Separate control and data related initializations

[Erez Shitrit]

This patch prepares init and and teardown flows so we can call them
through ipoib_options function pointers.
It arranges that area of code as the following:
All operations which deal with the resource allocation/deletion are done
in one place.
All operations that are control oriented, meaning that they are not
connected to a specific HW beneath, are done in a separate place.

The operations for alloc HW resources are now in the function
ipoib_dev_init_default, the deletion of all the resources are in
ipoib_dev_uninit_default The only exception is the creation of the pd
object which is used both for resource allocation (create QP etc.) and for
control flows like creating ah.

It also does:
Move creation of rx_ring and tx_ring to be in the resources allocation
area.
Move the function ipoib_ib_dev_open that does the open device to the
control area instead of the dev_init which creates resources.

Signed-off-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/ulp/ipoib/ipoib.h       |   5 +-
 drivers/infiniband/ulp/ipoib/ipoib_ib.c    |  39 +++-------
 drivers/infiniband/ulp/ipoib/ipoib_main.c  | 116 ++++++++++++++++++++++++-----
 drivers/infiniband/ulp/ipoib/ipoib_verbs.c |  40 +---------
 4 files changed, 110 insertions(+), 90 deletions(-)

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index bed233bf45c3..7cd9befd7d54 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -491,14 +491,13 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 void ipoib_flush_paths(struct net_device *dev);
 int ipoib_check_sm_sendonly_fullmember_support(struct ipoib_dev_priv *priv);
 struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
-
-int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
+void ipoib_ib_tx_timer_func(unsigned long ctx);
 void ipoib_ib_dev_flush_light(struct work_struct *work);
 void ipoib_ib_dev_flush_normal(struct work_struct *work);
 void ipoib_ib_dev_flush_heavy(struct work_struct *work);
 void ipoib_pkey_event(struct work_struct *work);
 void ipoib_ib_dev_cleanup(struct net_device *dev);
-
+void ipoib_dev_uninit_default(struct net_device *dev);
 int ipoib_ib_dev_open(struct net_device *dev);
 void ipoib_ib_dev_up(struct net_device *dev);
 void ipoib_ib_dev_down(struct net_device *dev);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 12c4f84a6639..3c0a35d883e2 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -40,6 +40,7 @@
 
 #include <linux/ip.h>
 #include <linux/tcp.h>
+#include <linux/vmalloc.h>
 
 #include "ipoib.h"
 
@@ -692,7 +693,7 @@ static void ipoib_stop_ah(struct net_device *dev)
 	ipoib_flush_ah(dev);
 }
 
-static void ipoib_ib_tx_timer_func(unsigned long ctx)
+void ipoib_ib_tx_timer_func(unsigned long ctx)
 {
 	drain_tx_cq((struct net_device *)ctx);
 }
@@ -913,32 +914,6 @@ void ipoib_ib_dev_stop(struct net_device *dev)
 	ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP);
 }
 
-int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
-{
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
-
-	priv->ca = ca;
-	priv->port = port;
-	priv->qp = NULL;
-
-	if (ipoib_transport_dev_init(dev, ca)) {
-		printk(KERN_WARNING "%s: ipoib_transport_dev_init failed\n", ca->name);
-		return -ENODEV;
-	}
-
-	setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func,
-		    (unsigned long) dev);
-
-	if (dev->flags & IFF_UP) {
-		if (ipoib_ib_dev_open(dev)) {
-			ipoib_transport_dev_cleanup(dev);
-			return -ENODEV;
-		}
-	}
-
-	return 0;
-}
-
 /*
  * Takes whatever value which is in pkey index 0 and updates priv->pkey
  * returns 0 if the pkey value was changed.
@@ -1236,7 +1211,13 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
 	 */
 	ipoib_stop_ah(dev);
 
-	ipoib_transport_dev_cleanup(dev);
-}
+	clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+
+	ipoib_dev_uninit_default(dev);
 
+	if (priv->pd) {
+		ib_dealloc_pd(priv->pd);
+		priv->pd = NULL;
+	}
+}
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 259c59f67394..8c644bbc2828 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -1622,8 +1622,23 @@ static void ipoib_neigh_hash_uninit(struct net_device *dev)
 	wait_for_completion(&priv->ntbl.deleted);
 }
 
+void ipoib_dev_uninit_default(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
-int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
+	ipoib_transport_dev_cleanup(dev);
+
+	ipoib_cm_dev_cleanup(dev);
+
+	kfree(priv->rx_ring);
+	vfree(priv->tx_ring);
+
+	priv->rx_ring = NULL;
+	priv->tx_ring = NULL;
+}
+
+static int ipoib_dev_init_default(struct net_device *dev, struct ib_device *ca,
+				  int port)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
@@ -1642,29 +1657,95 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 
 	/* priv->tx_head, tx_tail & tx_outstanding are already 0 */
 
-	if (ipoib_ib_dev_init(dev, ca, port))
+	if (ipoib_transport_dev_init(dev, ca)) {
+		pr_warn("%s: ipoib_transport_dev_init failed\n", ca->name);
 		goto out_tx_ring_cleanup;
+	}
+
+	setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func,
+		    (unsigned long)dev);
+
+	return 0;
+
+out_tx_ring_cleanup:
+	vfree(priv->tx_ring);
+
+out_rx_ring_cleanup:
+	kfree(priv->rx_ring);
+
+out:
+	return -ENOMEM;
+}
+
+int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int ret = -ENOMEM;
+
+	priv->ca = ca;
+	priv->port = port;
+	priv->qp = NULL;
 
 	/*
-	 * Must be after ipoib_ib_dev_init so we can allocate a per
-	 * device wq there and use it here
+	 * the various IPoIB tasks assume they will never race against
+	 * themselves, so always use a single thread workqueue
 	 */
-	if (ipoib_neigh_hash_init(priv) < 0)
+	priv->wq = alloc_ordered_workqueue("ipoib_wq", WQ_MEM_RECLAIM);
+	if (!priv->wq) {
+		pr_warn("%s: failed to allocate device WQ\n", dev->name);
+		goto out;
+	}
+
+	/* create pd, which used both for control and datapath*/
+	priv->pd = ib_alloc_pd(priv->ca, 0);
+	if (IS_ERR(priv->pd)) {
+		pr_warn("%s: failed to allocate PD\n", ca->name);
+		goto clean_wq;
+	}
+
+	ret = ipoib_dev_init_default(dev, ca, port);
+	if (ret) {
+		pr_warn("%s failed to init HW resource\n", dev->name);
+		goto out_free_pd;
+	}
+
+	/* after qp created set dev address */
+	priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
+	priv->dev->dev_addr[2] = (priv->qp->qp_num >>  8) & 0xff;
+	priv->dev->dev_addr[3] = (priv->qp->qp_num) & 0xff;
+
+	if (ipoib_neigh_hash_init(priv) < 0) {
+		pr_warn("%s failed to init neigh hash\n", dev->name);
 		goto out_dev_uninit;
+	}
+
+	if (dev->flags & IFF_UP) {
+		if (ipoib_ib_dev_open(dev)) {
+			pr_warn("%s failed to open device\n", dev->name);
+			ret = -ENODEV;
+			goto out_dev_uninit;
+		}
+	}
 
 	return 0;
 
 out_dev_uninit:
 	ipoib_ib_dev_cleanup(dev);
 
-out_tx_ring_cleanup:
-	vfree(priv->tx_ring);
+out_free_pd:
+	if (priv->pd) {
+		ib_dealloc_pd(priv->pd);
+		priv->pd = NULL;
+	}
 
-out_rx_ring_cleanup:
-	kfree(priv->rx_ring);
+clean_wq:
+	if (priv->wq) {
+		destroy_workqueue(priv->wq);
+		priv->wq = NULL;
+	}
 
 out:
-	return -ENOMEM;
+	return ret;
 }
 
 void ipoib_dev_cleanup(struct net_device *dev)
@@ -1685,19 +1766,16 @@ void ipoib_dev_cleanup(struct net_device *dev)
 	}
 	unregister_netdevice_many(&head);
 
-	/*
-	 * Must be before ipoib_ib_dev_cleanup or we delete an in use
-	 * work queue
-	 */
 	ipoib_neigh_hash_uninit(dev);
 
 	ipoib_ib_dev_cleanup(dev);
 
-	kfree(priv->rx_ring);
-	vfree(priv->tx_ring);
-
-	priv->rx_ring = NULL;
-	priv->tx_ring = NULL;
+	/* no more works over the priv->wq */
+	if (priv->wq) {
+		flush_workqueue(priv->wq);
+		destroy_workqueue(priv->wq);
+		priv->wq = NULL;
+	}
 }
 
 static int ipoib_set_vf_link_state(struct net_device *dev, int vf, int link_state)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
index 189dcd1709d2..529b9509a2c4 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
@@ -147,22 +147,6 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 	int ret, size;
 	int i;
 
-	priv->pd = ib_alloc_pd(priv->ca, 0);
-	if (IS_ERR(priv->pd)) {
-		printk(KERN_WARNING "%s: failed to allocate PD\n", ca->name);
-		return -ENODEV;
-	}
-
-	/*
-	 * the various IPoIB tasks assume they will never race against
-	 * themselves, so always use a single thread workqueue
-	 */
-	priv->wq = alloc_ordered_workqueue("ipoib_wq", WQ_MEM_RECLAIM);
-	if (!priv->wq) {
-		printk(KERN_WARNING "ipoib: failed to allocate device WQ\n");
-		goto out_free_pd;
-	}
-
 	size = ipoib_recvq_size + 1;
 	ret = ipoib_cm_dev_init(dev);
 	if (!ret) {
@@ -173,7 +157,7 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 			size += ipoib_recvq_size * ipoib_max_conn_qp;
 	} else
 		if (ret != -ENOSYS)
-			goto out_free_wq;
+			return -ENODEV;
 
 	cq_attr.cqe = size;
 	priv->recv_cq = ib_create_cq(priv->ca, ipoib_ib_completion, NULL,
@@ -212,10 +196,6 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 		goto out_free_send_cq;
 	}
 
-	priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
-	priv->dev->dev_addr[2] = (priv->qp->qp_num >>  8) & 0xff;
-	priv->dev->dev_addr[3] = (priv->qp->qp_num      ) & 0xff;
-
 	for (i = 0; i < MAX_SKB_FRAGS + 1; ++i)
 		priv->tx_sge[i].lkey = priv->pd->local_dma_lkey;
 
@@ -247,13 +227,6 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 out_cm_dev_cleanup:
 	ipoib_cm_dev_cleanup(dev);
 
-out_free_wq:
-	destroy_workqueue(priv->wq);
-	priv->wq = NULL;
-
-out_free_pd:
-	ib_dealloc_pd(priv->pd);
-
 	return -ENODEV;
 }
 
@@ -266,7 +239,6 @@ void ipoib_transport_dev_cleanup(struct net_device *dev)
 			ipoib_warn(priv, "ib_qp_destroy failed\n");
 
 		priv->qp = NULL;
-		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
 	}
 
 	if (ib_destroy_cq(priv->send_cq))
@@ -274,16 +246,6 @@ void ipoib_transport_dev_cleanup(struct net_device *dev)
 
 	if (ib_destroy_cq(priv->recv_cq))
 		ipoib_warn(priv, "ib_cq_destroy (recv) failed\n");
-
-	ipoib_cm_dev_cleanup(dev);
-
-	if (priv->wq) {
-		flush_workqueue(priv->wq);
-		destroy_workqueue(priv->wq);
-		priv->wq = NULL;
-	}
-
-	ib_dealloc_pd(priv->pd);
 }
 
 void ipoib_event(struct ib_event_handler *handler,
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC v2 for accelerated IPoIB 02/6] IB/ipoib: separate control from HW operation on ipoib_open/stop ndo

[Erez Shitrit]

This patch is preparing the netdev part at the ipoib driver to be able
to use the ipoib_options.
It deals with the two flows from the .ndo: ipoib_open and ipoib_stop.
It arranges that area of code as follows:
All operations which deal with the HW resources, (for example change QP
state, post-receive etc.) are done in one place.
All operations that are control oriented (like restart multicast task,
start the reap_ah etc.) are done in separate place.

The functions that deals with the HW resources now located at
__ipoib_ib_dev_open for the ipoib_open flow and __ipoib_ib_dev_stop for
ipoib_stop.

Signed-off-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/ulp/ipoib/ipoib.h      |   2 +-
 drivers/infiniband/ulp/ipoib/ipoib_ib.c   | 228 +++++++++++++++++-------------
 drivers/infiniband/ulp/ipoib/ipoib_main.c |   2 +-
 3 files changed, 129 insertions(+), 103 deletions(-)

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index 7cd9befd7d54..b3900b253ad5 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -501,7 +501,7 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 int ipoib_ib_dev_open(struct net_device *dev);
 void ipoib_ib_dev_up(struct net_device *dev);
 void ipoib_ib_dev_down(struct net_device *dev);
-void ipoib_ib_dev_stop(struct net_device *dev);
+int ipoib_ib_dev_stop_default(struct net_device *dev);
 void ipoib_pkey_dev_check_presence(struct net_device *dev);
 
 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 3c0a35d883e2..5d732c5f01ee 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -693,24 +693,113 @@ static void ipoib_stop_ah(struct net_device *dev)
 	ipoib_flush_ah(dev);
 }
 
-void ipoib_ib_tx_timer_func(unsigned long ctx)
+static int recvs_pending(struct net_device *dev)
 {
-	drain_tx_cq((struct net_device *)ctx);
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int pending = 0;
+	int i;
+
+	for (i = 0; i < ipoib_recvq_size; ++i)
+		if (priv->rx_ring[i].skb)
+			++pending;
+
+	return pending;
 }
 
-int ipoib_ib_dev_open(struct net_device *dev)
+int ipoib_ib_dev_stop_default(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	int ret;
+	struct ib_qp_attr qp_attr;
+	unsigned long begin;
+	struct ipoib_tx_buf *tx_req;
+	int i;
 
-	ipoib_pkey_dev_check_presence(dev);
+	if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
+		napi_disable(&priv->napi);
 
-	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
-		ipoib_warn(priv, "P_Key 0x%04x is %s\n", priv->pkey,
-			   (!(priv->pkey & 0x7fff) ? "Invalid" : "not found"));
-		return -1;
+	ipoib_cm_dev_stop(dev);
+
+	/*
+	 * Move our QP to the error state and then reinitialize in
+	 * when all work requests have completed or have been flushed.
+	 */
+	qp_attr.qp_state = IB_QPS_ERR;
+	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
+		ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
+
+	/* Wait for all sends and receives to complete */
+	begin = jiffies;
+
+	while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) {
+		if (time_after(jiffies, begin + 5 * HZ)) {
+			ipoib_warn(priv, "timing out; %d sends %d receives not completed\n",
+				   priv->tx_head - priv->tx_tail, recvs_pending(dev));
+
+			/*
+			 * assume the HW is wedged and just free up
+			 * all our pending work requests.
+			 */
+			while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
+				tx_req = &priv->tx_ring[priv->tx_tail &
+							(ipoib_sendq_size - 1)];
+				ipoib_dma_unmap_tx(priv, tx_req);
+				dev_kfree_skb_any(tx_req->skb);
+				++priv->tx_tail;
+				--priv->tx_outstanding;
+			}
+
+			for (i = 0; i < ipoib_recvq_size; ++i) {
+				struct ipoib_rx_buf *rx_req;
+
+				rx_req = &priv->rx_ring[i];
+				if (!rx_req->skb)
+					continue;
+				ipoib_ud_dma_unmap_rx(priv,
+						      priv->rx_ring[i].mapping);
+				dev_kfree_skb_any(rx_req->skb);
+				rx_req->skb = NULL;
+			}
+
+			goto timeout;
+		}
+
+		ipoib_drain_cq(dev);
+
+		msleep(1);
 	}
 
+	ipoib_dbg(priv, "All sends and receives done.\n");
+
+timeout:
+	del_timer_sync(&priv->poll_timer);
+	qp_attr.qp_state = IB_QPS_RESET;
+	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
+		ipoib_warn(priv, "Failed to modify QP to RESET state\n");
+
+	ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP);
+
+	return 0;
+}
+
+int ipoib_ib_dev_stop(struct net_device *dev)
+{
+	ipoib_ib_dev_stop_default(dev);
+
+	ipoib_flush_ah(dev);
+
+	return 0;
+}
+
+void ipoib_ib_tx_timer_func(unsigned long ctx)
+{
+	drain_tx_cq((struct net_device *)ctx);
+}
+
+int ipoib_ib_dev_open_default(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int ret;
+
 	ret = ipoib_init_qp(dev);
 	if (ret) {
 		ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
@@ -729,10 +818,6 @@ int ipoib_ib_dev_open(struct net_device *dev)
 		goto dev_stop;
 	}
 
-	clear_bit(IPOIB_STOP_REAPER, &priv->flags);
-	queue_delayed_work(priv->wq, &priv->ah_reap_task,
-			   round_jiffies_relative(HZ));
-
 	if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
 		napi_enable(&priv->napi);
 
@@ -744,6 +829,35 @@ int ipoib_ib_dev_open(struct net_device *dev)
 	return -1;
 }
 
+int ipoib_ib_dev_open(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	ipoib_pkey_dev_check_presence(dev);
+
+	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
+		ipoib_warn(priv, "P_Key 0x%04x is %s\n", priv->pkey,
+			   (!(priv->pkey & 0x7fff) ? "Invalid" : "not found"));
+		return -1;
+	}
+
+	clear_bit(IPOIB_STOP_REAPER, &priv->flags);
+	queue_delayed_work(priv->wq, &priv->ah_reap_task,
+			   round_jiffies_relative(HZ));
+
+	if (ipoib_ib_dev_open_default(dev)) {
+		pr_warn("%s: Failed to open dev\n", dev->name);
+		goto stop_ah_reap;
+	}
+
+	return 0;
+
+stop_ah_reap:
+	set_bit(IPOIB_STOP_REAPER, &priv->flags);
+	cancel_delayed_work(&priv->ah_reap_task);
+	return -1;
+}
+
 void ipoib_pkey_dev_check_presence(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
@@ -787,19 +901,6 @@ void ipoib_ib_dev_down(struct net_device *dev)
 	ipoib_flush_paths(dev);
 }
 
-static int recvs_pending(struct net_device *dev)
-{
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	int pending = 0;
-	int i;
-
-	for (i = 0; i < ipoib_recvq_size; ++i)
-		if (priv->rx_ring[i].skb)
-			++pending;
-
-	return pending;
-}
-
 void ipoib_drain_cq(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
@@ -839,81 +940,6 @@ void ipoib_drain_cq(struct net_device *dev)
 	local_bh_enable();
 }
 
-void ipoib_ib_dev_stop(struct net_device *dev)
-{
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	struct ib_qp_attr qp_attr;
-	unsigned long begin;
-	struct ipoib_tx_buf *tx_req;
-	int i;
-
-	if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
-		napi_disable(&priv->napi);
-
-	ipoib_cm_dev_stop(dev);
-
-	/*
-	 * Move our QP to the error state and then reinitialize in
-	 * when all work requests have completed or have been flushed.
-	 */
-	qp_attr.qp_state = IB_QPS_ERR;
-	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
-		ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
-
-	/* Wait for all sends and receives to complete */
-	begin = jiffies;
-
-	while (priv->tx_head != priv->tx_tail || recvs_pending(dev)) {
-		if (time_after(jiffies, begin + 5 * HZ)) {
-			ipoib_warn(priv, "timing out; %d sends %d receives not completed\n",
-				   priv->tx_head - priv->tx_tail, recvs_pending(dev));
-
-			/*
-			 * assume the HW is wedged and just free up
-			 * all our pending work requests.
-			 */
-			while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
-				tx_req = &priv->tx_ring[priv->tx_tail &
-							(ipoib_sendq_size - 1)];
-				ipoib_dma_unmap_tx(priv, tx_req);
-				dev_kfree_skb_any(tx_req->skb);
-				++priv->tx_tail;
-				--priv->tx_outstanding;
-			}
-
-			for (i = 0; i < ipoib_recvq_size; ++i) {
-				struct ipoib_rx_buf *rx_req;
-
-				rx_req = &priv->rx_ring[i];
-				if (!rx_req->skb)
-					continue;
-				ipoib_ud_dma_unmap_rx(priv,
-						      priv->rx_ring[i].mapping);
-				dev_kfree_skb_any(rx_req->skb);
-				rx_req->skb = NULL;
-			}
-
-			goto timeout;
-		}
-
-		ipoib_drain_cq(dev);
-
-		msleep(1);
-	}
-
-	ipoib_dbg(priv, "All sends and receives done.\n");
-
-timeout:
-	del_timer_sync(&priv->poll_timer);
-	qp_attr.qp_state = IB_QPS_RESET;
-	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
-		ipoib_warn(priv, "Failed to modify QP to RESET state\n");
-
-	ipoib_flush_ah(dev);
-
-	ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP);
-}
-
 /*
  * Takes whatever value which is in pkey index 0 and updates priv->pkey
  * returns 0 if the pkey value was changed.
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index 8c644bbc2828..be8dcea252f9 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -166,7 +166,7 @@ static int ipoib_stop(struct net_device *dev)
 	netif_stop_queue(dev);
 
 	ipoib_ib_dev_down(dev);
-	ipoib_ib_dev_stop(dev);
+	ipoib_ib_dev_stop_default(dev);
 
 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
 		struct ipoib_dev_priv *cpriv;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC v2 for accelerated IPoIB 03/6] IB/ipoib: Rename qpn to dqpn in ipoib_send and post_send functions

[Erez Shitrit]

Signed-off-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/ulp/ipoib/ipoib.h    | 2 +-
 drivers/infiniband/ulp/ipoib/ipoib_ib.c | 8 ++++----
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index b3900b253ad5..d94a7a953338 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -483,7 +483,7 @@ static inline void ipoib_put_ah(struct ipoib_ah *ah)
 int ipoib_add_umcast_attr(struct net_device *dev);
 
 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
-		struct ipoib_ah *address, u32 qpn);
+		struct ipoib_ah *address, u32 dqpn);
 void ipoib_reap_ah(struct work_struct *work);
 
 struct ipoib_path *__path_find(struct net_device *dev, void *gid);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index 5d732c5f01ee..dd5fb2964e63 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -514,7 +514,7 @@ void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr)
 
 static inline int post_send(struct ipoib_dev_priv *priv,
 			    unsigned int wr_id,
-			    struct ib_ah *address, u32 qpn,
+			    struct ib_ah *address, u32 dqpn,
 			    struct ipoib_tx_buf *tx_req,
 			    void *head, int hlen)
 {
@@ -524,7 +524,7 @@ static inline int post_send(struct ipoib_dev_priv *priv,
 	ipoib_build_sge(priv, tx_req);
 
 	priv->tx_wr.wr.wr_id	= wr_id;
-	priv->tx_wr.remote_qpn	= qpn;
+	priv->tx_wr.remote_qpn	= dqpn;
 	priv->tx_wr.ah		= address;
 
 	if (head) {
@@ -539,7 +539,7 @@ static inline int post_send(struct ipoib_dev_priv *priv,
 }
 
 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
-		struct ipoib_ah *address, u32 qpn)
+		struct ipoib_ah *address, u32 dqpn)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_tx_buf *tx_req;
@@ -621,7 +621,7 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 	skb_dst_drop(skb);
 
 	rc = post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
-		       address->ah, qpn, tx_req, phead, hlen);
+		       address->ah, dqpn, tx_req, phead, hlen);
 	if (unlikely(rc)) {
 		ipoib_warn(priv, "post_send failed, error %d\n", rc);
 		++dev->stats.tx_errors;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC v2 for accelerated IPoIB 04/6] IB/verb: Add ipoib_options struct and API

[Erez Shitrit]

The idea is to allow vendors to optimize IPoIB data path.
New struct that includes functions and data member is exposed.
It exposes set of callback functions for handling data path flows in IPoIB driver.
Each vendor can support these set of functions in order to optimize its
specific data path, and let IPoIB to leverage its data path.
The code of IPoIB driver was changed accordingly, and works in both ways
with vendor specific implementation and without.
There is an assumption, that vendors should give the full set of functions
and not only part of them, in order to work properly.

Signed-off-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 include/rdma/ib_ipoib_accel_ops.h | 61 +++++++++++++++++++++++++++++++++++++++
 include/rdma/ib_verbs.h           | 30 +++++++++++++++++++
 2 files changed, 91 insertions(+)
 create mode 100644 include/rdma/ib_ipoib_accel_ops.h

diff --git a/include/rdma/ib_ipoib_accel_ops.h b/include/rdma/ib_ipoib_accel_ops.h
new file mode 100644
index 000000000000..352672de6edf
--- /dev/null
+++ b/include/rdma/ib_ipoib_accel_ops.h
@@ -0,0 +1,61 @@
+/*
+ * Copyright (c) 2017 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
+ * 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.
+ */
+
+#if !defined(IB_IPOIB_ACCEL_OPS_H)
+#define IB_IPOIB_ACCEL_OPS_H
+
+#include <rdma/ib_verbs.h>
+
+/* ipoib rdma netdev's private data structure */
+struct ipoib_rdma_netdev {
+	struct rdma_netdev rn;  /* keep this first */
+	void *context;
+	u32 qpn;
+	/* followed by device private data */
+	char *dev_priv[0];
+};
+
+static inline void *ipoib_priv(const struct net_device *dev)
+{
+	struct rdma_netdev *rn = netdev_priv(dev);
+
+	return rn->clnt_priv;
+}
+
+static inline void *ipoib_accl_priv(const struct net_device *dev)
+{
+	struct ipoib_rdma_netdev *ipoib_rn = netdev_priv(dev);
+
+	return ipoib_rn->dev_priv;
+}
+
+#endif /* IB_IPOIB_ACCEL_OPS_H */
diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index 85b9034c8cfc..94203ba07562 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -1901,6 +1901,36 @@ struct ib_port_immutable {
 	u32                           max_mad_size;
 };
 
+/* rdma netdev type - specifies protocol type */
+enum rdma_netdev_t {
+	RDMA_NETDEV_OPA_VNIC,
+	RDMA_NETDEV_IPOIB
+};
+
+struct ipoib_ah;
+
+/**
+ * struct rdma_netdev - rdma netdev
+ * For cases where netstack interfacing is required.
+ */
+struct rdma_netdev {
+	void *clnt_priv;
+
+	/* control functions */
+	void (*set_id)(struct net_device *netdev, int id);
+	/* send packet */
+	void (*send)(struct net_device *dev, struct sk_buff *skb,
+		     struct ipoib_ah *address, u32 dqpn, u32 dqkey);
+
+	/* multicast */
+	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
+			    union ib_gid *gid, u16 mlid, int set_qkey);
+	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
+			    union ib_gid *gid, u16 mlid);
+	/* rdma-netdev private members */
+	struct ib_device *hca;
+};
+
 struct ib_device {
 	struct device                *dma_device;
 
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC v2 for accelerated IPoIB 05/6] IB/ipoib: Support ipoib acceleration options callbacks

[Erez Shitrit]

IPoIB driver now uses the new set of callback functions.
If the HW provider supports the new ipoib_options implementation, the
driver uses the callbacks in its datapath flows, otherwise it uses the
driver default implementation for all data flows in its code.
The default implementation is exactly the driver implementation as it
was without HW vendor support.

TODO: We added remote qkey to ipoib_send in order to match send op
signature.
In accel mode this param will be used but in regular mode this param is
redundant. Need to think about better solution.

Signed-off-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/ulp/ipoib/ipoib.h           |  29 ++--
 drivers/infiniband/ulp/ipoib/ipoib_cm.c        |  66 ++++----
 drivers/infiniband/ulp/ipoib/ipoib_ethtool.c   |   6 +-
 drivers/infiniband/ulp/ipoib/ipoib_fs.c        |   4 +-
 drivers/infiniband/ulp/ipoib/ipoib_ib.c        |  56 ++++---
 drivers/infiniband/ulp/ipoib/ipoib_main.c      | 224 +++++++++++++++++--------
 drivers/infiniband/ulp/ipoib/ipoib_multicast.c |  39 +++--
 drivers/infiniband/ulp/ipoib/ipoib_netlink.c   |  12 +-
 drivers/infiniband/ulp/ipoib/ipoib_verbs.c     |  24 ++-
 drivers/infiniband/ulp/ipoib/ipoib_vlan.c      |   9 +-
 10 files changed, 289 insertions(+), 180 deletions(-)

diff --git a/drivers/infiniband/ulp/ipoib/ipoib.h b/drivers/infiniband/ulp/ipoib/ipoib.h
index d94a7a953338..3d883df5794e 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib.h
+++ b/drivers/infiniband/ulp/ipoib/ipoib.h
@@ -51,8 +51,8 @@
 #include <rdma/ib_verbs.h>
 #include <rdma/ib_pack.h>
 #include <rdma/ib_sa.h>
+#include <rdma/ib_ipoib_accel_ops.h>
 #include <linux/sched.h>
-
 /* constants */
 
 enum ipoib_flush_level {
@@ -357,6 +357,7 @@ struct ipoib_dev_priv {
 	struct ib_cq	 *recv_cq;
 	struct ib_cq	 *send_cq;
 	struct ib_qp	 *qp;
+	u32		  qp_num;
 	u32		  qkey;
 
 	union ib_gid local_gid;
@@ -404,6 +405,7 @@ struct ipoib_dev_priv {
 	struct timer_list poll_timer;
 	unsigned max_send_sge;
 	bool sm_fullmember_sendonly_support;
+	const struct net_device_ops	*rn_ops;
 };
 
 struct ipoib_ah {
@@ -483,22 +485,25 @@ static inline void ipoib_put_ah(struct ipoib_ah *ah)
 int ipoib_add_umcast_attr(struct net_device *dev);
 
 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
-		struct ipoib_ah *address, u32 dqpn);
+		struct ipoib_ah *address, u32 dqpn, u32 dqkey);
 void ipoib_reap_ah(struct work_struct *work);
 
 struct ipoib_path *__path_find(struct net_device *dev, void *gid);
 void ipoib_mark_paths_invalid(struct net_device *dev);
 void ipoib_flush_paths(struct net_device *dev);
 int ipoib_check_sm_sendonly_fullmember_support(struct ipoib_dev_priv *priv);
-struct ipoib_dev_priv *ipoib_intf_alloc(const char *format);
+struct ipoib_dev_priv *ipoib_intf_alloc(struct ib_device *hca, u8 port,
+					const char *format);
 void ipoib_ib_tx_timer_func(unsigned long ctx);
 void ipoib_ib_dev_flush_light(struct work_struct *work);
 void ipoib_ib_dev_flush_normal(struct work_struct *work);
 void ipoib_ib_dev_flush_heavy(struct work_struct *work);
 void ipoib_pkey_event(struct work_struct *work);
 void ipoib_ib_dev_cleanup(struct net_device *dev);
-void ipoib_dev_uninit_default(struct net_device *dev);
+int ipoib_ib_dev_open_default(struct net_device *dev);
+int ipoib_ib_dev_stop_default(struct net_device *dev);
 int ipoib_ib_dev_open(struct net_device *dev);
+int ipoib_ib_dev_stop(struct net_device *dev);
 void ipoib_ib_dev_up(struct net_device *dev);
 void ipoib_ib_dev_down(struct net_device *dev);
 int ipoib_ib_dev_stop_default(struct net_device *dev);
@@ -561,8 +566,10 @@ void ipoib_path_iter_read(struct ipoib_path_iter *iter,
 			  struct ipoib_path *path);
 #endif
 
-int ipoib_mcast_attach(struct net_device *dev, u16 mlid,
-		       union ib_gid *mgid, int set_qkey);
+int ipoib_mcast_attach(struct net_device *dev, struct ib_device *hca,
+		       union ib_gid *mgid, u16 mlid, int set_qkey);
+int ipoib_mcast_detach(struct net_device *dev, struct ib_device *hca,
+		       union ib_gid *mgid, u16 mlid);
 void ipoib_mcast_remove_list(struct list_head *remove_list);
 void ipoib_check_and_add_mcast_sendonly(struct ipoib_dev_priv *priv, u8 *mgid,
 				struct list_head *remove_list);
@@ -586,7 +593,7 @@ int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
 void ipoib_set_umcast(struct net_device *ndev, int umcast_val);
 int  ipoib_set_mode(struct net_device *dev, const char *buf);
 
-void ipoib_setup(struct net_device *dev);
+void ipoib_setup_common(struct net_device *dev);
 
 void ipoib_pkey_open(struct ipoib_dev_priv *priv);
 void ipoib_drain_cq(struct net_device *dev);
@@ -606,14 +613,14 @@ int __ipoib_vlan_add(struct ipoib_dev_priv *ppriv, struct ipoib_dev_priv *priv,
 
 static inline int ipoib_cm_admin_enabled(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	return IPOIB_CM_SUPPORTED(dev->dev_addr) &&
 		test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
 }
 
 static inline int ipoib_cm_enabled(struct net_device *dev, u8 *hwaddr)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	return IPOIB_CM_SUPPORTED(hwaddr) &&
 		test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags);
 }
@@ -636,13 +643,13 @@ static inline void ipoib_cm_set(struct ipoib_neigh *neigh, struct ipoib_cm_tx *t
 
 static inline int ipoib_cm_has_srq(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	return !!priv->cm.srq;
 }
 
 static inline unsigned int ipoib_cm_max_mtu(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	return priv->cm.max_cm_mtu;
 }
 
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
index 53d69d7d9ad4..9593930454ef 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
@@ -91,7 +91,7 @@ static void ipoib_cm_dma_unmap_rx(struct ipoib_dev_priv *priv, int frags,
 
 static int ipoib_cm_post_receive_srq(struct net_device *dev, int id)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_recv_wr *bad_wr;
 	int i, ret;
 
@@ -117,7 +117,7 @@ static int ipoib_cm_post_receive_nonsrq(struct net_device *dev,
 					struct ib_recv_wr *wr,
 					struct ib_sge *sge, int id)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_recv_wr *bad_wr;
 	int i, ret;
 
@@ -144,7 +144,7 @@ static struct sk_buff *ipoib_cm_alloc_rx_skb(struct net_device *dev,
 					     u64 mapping[IPOIB_CM_RX_SG],
 					     gfp_t gfp)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct sk_buff *skb;
 	int i;
 
@@ -195,7 +195,7 @@ static struct sk_buff *ipoib_cm_alloc_rx_skb(struct net_device *dev,
 static void ipoib_cm_free_rx_ring(struct net_device *dev,
 				  struct ipoib_cm_rx_buf *rx_ring)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int i;
 
 	for (i = 0; i < ipoib_recvq_size; ++i)
@@ -234,7 +234,7 @@ static void ipoib_cm_start_rx_drain(struct ipoib_dev_priv *priv)
 static void ipoib_cm_rx_event_handler(struct ib_event *event, void *ctx)
 {
 	struct ipoib_cm_rx *p = ctx;
-	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(p->dev);
 	unsigned long flags;
 
 	if (event->event != IB_EVENT_QP_LAST_WQE_REACHED)
@@ -250,7 +250,7 @@ static void ipoib_cm_rx_event_handler(struct ib_event *event, void *ctx)
 static struct ib_qp *ipoib_cm_create_rx_qp(struct net_device *dev,
 					   struct ipoib_cm_rx *p)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_init_attr attr = {
 		.event_handler = ipoib_cm_rx_event_handler,
 		.send_cq = priv->recv_cq, /* For drain WR */
@@ -275,7 +275,7 @@ static int ipoib_cm_modify_rx_qp(struct net_device *dev,
 				 struct ib_cm_id *cm_id, struct ib_qp *qp,
 				 unsigned psn)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_attr qp_attr;
 	int qp_attr_mask, ret;
 
@@ -330,7 +330,7 @@ static void ipoib_cm_init_rx_wr(struct net_device *dev,
 				struct ib_recv_wr *wr,
 				struct ib_sge *sge)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int i;
 
 	for (i = 0; i < priv->cm.num_frags; ++i)
@@ -348,7 +348,7 @@ static void ipoib_cm_init_rx_wr(struct net_device *dev,
 static int ipoib_cm_nonsrq_init_rx(struct net_device *dev, struct ib_cm_id *cm_id,
 				   struct ipoib_cm_rx *rx)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct {
 		struct ib_recv_wr wr;
 		struct ib_sge sge[IPOIB_CM_RX_SG];
@@ -419,7 +419,7 @@ static int ipoib_cm_send_rep(struct net_device *dev, struct ib_cm_id *cm_id,
 			     struct ib_qp *qp, struct ib_cm_req_event_param *req,
 			     unsigned psn)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_data data = {};
 	struct ib_cm_rep_param rep = {};
 
@@ -439,7 +439,7 @@ static int ipoib_cm_send_rep(struct net_device *dev, struct ib_cm_id *cm_id,
 static int ipoib_cm_req_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
 {
 	struct net_device *dev = cm_id->context;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_rx *p;
 	unsigned psn;
 	int ret;
@@ -512,7 +512,7 @@ static int ipoib_cm_rx_handler(struct ib_cm_id *cm_id,
 		/* Fall through */
 	case IB_CM_REJ_RECEIVED:
 		p = cm_id->context;
-		priv = netdev_priv(p->dev);
+		priv = ipoib_priv(p->dev);
 		if (ib_modify_qp(p->qp, &ipoib_cm_err_attr, IB_QP_STATE))
 			ipoib_warn(priv, "unable to move qp to error state\n");
 		/* Fall through */
@@ -556,7 +556,7 @@ static void skb_put_frags(struct sk_buff *skb, unsigned int hdr_space,
 
 void ipoib_cm_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_rx_buf *rx_ring;
 	unsigned int wr_id = wc->wr_id & ~(IPOIB_OP_CM | IPOIB_OP_RECV);
 	struct sk_buff *skb, *newskb;
@@ -705,7 +705,7 @@ static inline int post_send(struct ipoib_dev_priv *priv,
 
 void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_tx *tx)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_tx_buf *tx_req;
 	int rc;
 	unsigned usable_sge = tx->max_send_sge - !!skb_headlen(skb);
@@ -783,7 +783,7 @@ void ipoib_cm_send(struct net_device *dev, struct sk_buff *skb, struct ipoib_cm_
 
 void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_tx *tx = wc->qp->qp_context;
 	unsigned int wr_id = wc->wr_id & ~IPOIB_OP_CM;
 	struct ipoib_tx_buf *tx_req;
@@ -852,7 +852,7 @@ void ipoib_cm_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 
 int ipoib_cm_dev_open(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int ret;
 
 	if (!IPOIB_CM_SUPPORTED(dev->dev_addr))
@@ -884,7 +884,7 @@ int ipoib_cm_dev_open(struct net_device *dev)
 
 static void ipoib_cm_free_rx_reap_list(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_rx *rx, *n;
 	LIST_HEAD(list);
 
@@ -907,7 +907,7 @@ static void ipoib_cm_free_rx_reap_list(struct net_device *dev)
 
 void ipoib_cm_dev_stop(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_rx *p;
 	unsigned long begin;
 	int ret;
@@ -966,7 +966,7 @@ void ipoib_cm_dev_stop(struct net_device *dev)
 static int ipoib_cm_rep_handler(struct ib_cm_id *cm_id, struct ib_cm_event *event)
 {
 	struct ipoib_cm_tx *p = cm_id->context;
-	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(p->dev);
 	struct ipoib_cm_data *data = event->private_data;
 	struct sk_buff_head skqueue;
 	struct ib_qp_attr qp_attr;
@@ -1034,7 +1034,7 @@ static int ipoib_cm_rep_handler(struct ib_cm_id *cm_id, struct ib_cm_event *even
 
 static struct ib_qp *ipoib_cm_create_tx_qp(struct net_device *dev, struct ipoib_cm_tx *tx)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_init_attr attr = {
 		.send_cq		= priv->recv_cq,
 		.recv_cq		= priv->recv_cq,
@@ -1067,7 +1067,7 @@ static int ipoib_cm_send_req(struct net_device *dev,
 			     u32 qpn,
 			     struct ib_sa_path_rec *pathrec)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_data data = {};
 	struct ib_cm_req_param req = {};
 
@@ -1102,7 +1102,7 @@ static int ipoib_cm_send_req(struct net_device *dev,
 static int ipoib_cm_modify_tx_init(struct net_device *dev,
 				  struct ib_cm_id *cm_id, struct ib_qp *qp)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_attr qp_attr;
 	int qp_attr_mask, ret;
 	ret = ib_find_pkey(priv->ca, priv->port, priv->pkey, &qp_attr.pkey_index);
@@ -1127,7 +1127,7 @@ static int ipoib_cm_modify_tx_init(struct net_device *dev,
 static int ipoib_cm_tx_init(struct ipoib_cm_tx *p, u32 qpn,
 			    struct ib_sa_path_rec *pathrec)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(p->dev);
 	int ret;
 
 	p->tx_ring = __vmalloc(ipoib_sendq_size * sizeof *p->tx_ring,
@@ -1184,7 +1184,7 @@ static int ipoib_cm_tx_init(struct ipoib_cm_tx *p, u32 qpn,
 
 static void ipoib_cm_tx_destroy(struct ipoib_cm_tx *p)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(p->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(p->dev);
 	struct ipoib_tx_buf *tx_req;
 	unsigned long begin;
 
@@ -1234,7 +1234,7 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 			       struct ib_cm_event *event)
 {
 	struct ipoib_cm_tx *tx = cm_id->context;
-	struct ipoib_dev_priv *priv = netdev_priv(tx->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(tx->dev);
 	struct net_device *dev = priv->dev;
 	struct ipoib_neigh *neigh;
 	unsigned long flags;
@@ -1285,7 +1285,7 @@ static int ipoib_cm_tx_handler(struct ib_cm_id *cm_id,
 struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path *path,
 				       struct ipoib_neigh *neigh)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_cm_tx *tx;
 
 	tx = kzalloc(sizeof *tx, GFP_ATOMIC);
@@ -1304,7 +1304,7 @@ struct ipoib_cm_tx *ipoib_cm_create_tx(struct net_device *dev, struct ipoib_path
 
 void ipoib_cm_destroy_tx(struct ipoib_cm_tx *tx)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(tx->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(tx->dev);
 	unsigned long flags;
 	if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &tx->flags)) {
 		spin_lock_irqsave(&priv->lock, flags);
@@ -1439,7 +1439,7 @@ static void ipoib_cm_skb_reap(struct work_struct *work)
 void ipoib_cm_skb_too_long(struct net_device *dev, struct sk_buff *skb,
 			   unsigned int mtu)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int e = skb_queue_empty(&priv->cm.skb_queue);
 
 	if (skb_dst(skb))
@@ -1488,7 +1488,7 @@ static void ipoib_cm_stale_task(struct work_struct *work)
 static ssize_t show_mode(struct device *d, struct device_attribute *attr,
 			 char *buf)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(d));
+	struct ipoib_dev_priv *priv = ipoib_priv(to_net_dev(d));
 
 	if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
 		return sprintf(buf, "connected\n");
@@ -1501,7 +1501,7 @@ static ssize_t set_mode(struct device *d, struct device_attribute *attr,
 {
 	struct net_device *dev = to_net_dev(d);
 	int ret;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if (test_bit(IPOIB_FLAG_GOING_DOWN, &priv->flags))
 		return -EPERM;
@@ -1530,7 +1530,7 @@ int ipoib_cm_add_mode_attr(struct net_device *dev)
 
 static void ipoib_cm_create_srq(struct net_device *dev, int max_sge)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_srq_init_attr srq_init_attr = {
 		.srq_type = IB_SRQT_BASIC,
 		.attr = {
@@ -1559,7 +1559,7 @@ static void ipoib_cm_create_srq(struct net_device *dev, int max_sge)
 
 int ipoib_cm_dev_init(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int max_srq_sge, i;
 
 	INIT_LIST_HEAD(&priv->cm.passive_ids);
@@ -1620,7 +1620,7 @@ int ipoib_cm_dev_init(struct net_device *dev)
 
 void ipoib_cm_dev_cleanup(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int ret;
 
 	if (!priv->cm.srq)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c b/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c
index 7b6d40ff1acf..32ad805fc721 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ethtool.c
@@ -60,7 +60,7 @@ struct ipoib_stats {
 static void ipoib_get_drvinfo(struct net_device *netdev,
 			      struct ethtool_drvinfo *drvinfo)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(netdev);
+	struct ipoib_dev_priv *priv = ipoib_priv(netdev);
 
 	ib_get_device_fw_str(priv->ca, drvinfo->fw_version,
 			     sizeof(drvinfo->fw_version));
@@ -77,7 +77,7 @@ static void ipoib_get_drvinfo(struct net_device *netdev,
 static int ipoib_get_coalesce(struct net_device *dev,
 			      struct ethtool_coalesce *coal)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	coal->rx_coalesce_usecs = priv->ethtool.coalesce_usecs;
 	coal->rx_max_coalesced_frames = priv->ethtool.max_coalesced_frames;
@@ -88,7 +88,7 @@ static int ipoib_get_coalesce(struct net_device *dev,
 static int ipoib_set_coalesce(struct net_device *dev,
 			      struct ethtool_coalesce *coal)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int ret;
 
 	/*
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_fs.c b/drivers/infiniband/ulp/ipoib/ipoib_fs.c
index 6bd5740e2691..6e73450241bf 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_fs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_fs.c
@@ -261,7 +261,7 @@ static int ipoib_path_open(struct inode *inode, struct file *file)
 
 void ipoib_create_debug_files(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	char name[IFNAMSIZ + sizeof "_path"];
 
 	snprintf(name, sizeof name, "%s_mcg", dev->name);
@@ -279,7 +279,7 @@ void ipoib_create_debug_files(struct net_device *dev)
 
 void ipoib_delete_debug_files(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	debugfs_remove(priv->mcg_dentry);
 	debugfs_remove(priv->path_dentry);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index dd5fb2964e63..300dc4335cf3 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -72,7 +72,7 @@ struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
 		ah = (struct ipoib_ah *)vah;
 	} else {
 		ah->ah = vah;
-		ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah);
+		ipoib_dbg(ipoib_priv(dev), "Created ah %p\n", ah->ah);
 	}
 
 	return ah;
@@ -81,7 +81,7 @@ struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
 void ipoib_free_ah(struct kref *kref)
 {
 	struct ipoib_ah *ah = container_of(kref, struct ipoib_ah, ref);
-	struct ipoib_dev_priv *priv = netdev_priv(ah->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(ah->dev);
 
 	unsigned long flags;
 
@@ -100,7 +100,7 @@ static void ipoib_ud_dma_unmap_rx(struct ipoib_dev_priv *priv,
 
 static int ipoib_ib_post_receive(struct net_device *dev, int id)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_recv_wr *bad_wr;
 	int ret;
 
@@ -122,7 +122,7 @@ static int ipoib_ib_post_receive(struct net_device *dev, int id)
 
 static struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct sk_buff *skb;
 	int buf_size;
 	u64 *mapping;
@@ -154,7 +154,7 @@ static struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id)
 
 static int ipoib_ib_post_receives(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int i;
 
 	for (i = 0; i < ipoib_recvq_size; ++i) {
@@ -173,7 +173,7 @@ static int ipoib_ib_post_receives(struct net_device *dev)
 
 static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV;
 	struct sk_buff *skb;
 	u64 mapping[IPOIB_UD_RX_SG];
@@ -382,7 +382,7 @@ static void ipoib_qp_state_validate_work(struct work_struct *work)
 
 static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	unsigned int wr_id = wc->wr_id;
 	struct ipoib_tx_buf *tx_req;
 
@@ -486,14 +486,14 @@ int ipoib_poll(struct napi_struct *napi, int budget)
 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
 {
 	struct net_device *dev = dev_ptr;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	napi_schedule(&priv->napi);
 }
 
 static void drain_tx_cq(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	netif_tx_lock(dev);
 	while (poll_tx(priv))
@@ -507,7 +507,7 @@ static void drain_tx_cq(struct net_device *dev)
 
 void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev_ptr);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev_ptr);
 
 	mod_timer(&priv->poll_timer, jiffies);
 }
@@ -539,9 +539,9 @@ static inline int post_send(struct ipoib_dev_priv *priv,
 }
 
 void ipoib_send(struct net_device *dev, struct sk_buff *skb,
-		struct ipoib_ah *address, u32 dqpn)
+		struct ipoib_ah *address, u32 dqpn, u32 dqkey)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_tx_buf *tx_req;
 	int hlen, rc;
 	void *phead;
@@ -644,7 +644,7 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 
 static void __ipoib_reap_ah(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_ah *ah, *tah;
 	LIST_HEAD(remove_list);
 	unsigned long flags;
@@ -678,7 +678,7 @@ void ipoib_reap_ah(struct work_struct *work)
 
 static void ipoib_flush_ah(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	cancel_delayed_work(&priv->ah_reap_task);
 	flush_workqueue(priv->wq);
@@ -687,7 +687,7 @@ static void ipoib_flush_ah(struct net_device *dev)
 
 static void ipoib_stop_ah(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	set_bit(IPOIB_STOP_REAPER, &priv->flags);
 	ipoib_flush_ah(dev);
@@ -695,7 +695,7 @@ static void ipoib_stop_ah(struct net_device *dev)
 
 static int recvs_pending(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int pending = 0;
 	int i;
 
@@ -708,7 +708,7 @@ static int recvs_pending(struct net_device *dev)
 
 int ipoib_ib_dev_stop_default(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_attr qp_attr;
 	unsigned long begin;
 	struct ipoib_tx_buf *tx_req;
@@ -783,7 +783,9 @@ int ipoib_ib_dev_stop_default(struct net_device *dev)
 
 int ipoib_ib_dev_stop(struct net_device *dev)
 {
-	ipoib_ib_dev_stop_default(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+
+	priv->rn_ops->ndo_stop(dev);
 
 	ipoib_flush_ah(dev);
 
@@ -797,7 +799,7 @@ void ipoib_ib_tx_timer_func(unsigned long ctx)
 
 int ipoib_ib_dev_open_default(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int ret;
 
 	ret = ipoib_init_qp(dev);
@@ -831,7 +833,7 @@ int ipoib_ib_dev_open_default(struct net_device *dev)
 
 int ipoib_ib_dev_open(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_pkey_dev_check_presence(dev);
 
@@ -845,7 +847,7 @@ int ipoib_ib_dev_open(struct net_device *dev)
 	queue_delayed_work(priv->wq, &priv->ah_reap_task,
 			   round_jiffies_relative(HZ));
 
-	if (ipoib_ib_dev_open_default(dev)) {
+	if (priv->rn_ops->ndo_open(dev)) {
 		pr_warn("%s: Failed to open dev\n", dev->name);
 		goto stop_ah_reap;
 	}
@@ -860,7 +862,7 @@ int ipoib_ib_dev_open(struct net_device *dev)
 
 void ipoib_pkey_dev_check_presence(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if (!(priv->pkey & 0x7fff) ||
 	    ib_find_pkey(priv->ca, priv->port, priv->pkey,
@@ -872,7 +874,7 @@ void ipoib_pkey_dev_check_presence(struct net_device *dev)
 
 void ipoib_ib_dev_up(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_pkey_dev_check_presence(dev);
 
@@ -888,7 +890,7 @@ void ipoib_ib_dev_up(struct net_device *dev)
 
 void ipoib_ib_dev_down(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_dbg(priv, "downing ib_dev\n");
 
@@ -903,7 +905,7 @@ void ipoib_ib_dev_down(struct net_device *dev)
 
 void ipoib_drain_cq(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int i, n;
 
 	/*
@@ -1217,7 +1219,7 @@ void ipoib_ib_dev_flush_heavy(struct work_struct *work)
 
 void ipoib_ib_dev_cleanup(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_dbg(priv, "cleaning up ib_dev\n");
 	/*
@@ -1239,7 +1241,7 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
 
 	clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
 
-	ipoib_dev_uninit_default(dev);
+	priv->rn_ops->ndo_uninit(dev);
 
 	if (priv->pd) {
 		ib_dealloc_pd(priv->pd);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_main.c b/drivers/infiniband/ulp/ipoib/ipoib_main.c
index be8dcea252f9..957630ed2c0f 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_main.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_main.c
@@ -110,7 +110,7 @@ static struct net_device *ipoib_get_net_dev_by_params(
 
 int ipoib_open(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_dbg(priv, "bringing up interface\n");
 
@@ -157,7 +157,7 @@ int ipoib_open(struct net_device *dev)
 
 static int ipoib_stop(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_dbg(priv, "stopping interface\n");
 
@@ -166,7 +166,7 @@ static int ipoib_stop(struct net_device *dev)
 	netif_stop_queue(dev);
 
 	ipoib_ib_dev_down(dev);
-	ipoib_ib_dev_stop_default(dev);
+	ipoib_ib_dev_stop(dev);
 
 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
 		struct ipoib_dev_priv *cpriv;
@@ -195,7 +195,7 @@ static void ipoib_uninit(struct net_device *dev)
 
 static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_features_t features)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if (test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags))
 		features &= ~(NETIF_F_IP_CSUM | NETIF_F_TSO);
@@ -205,7 +205,7 @@ static netdev_features_t ipoib_fix_features(struct net_device *dev, netdev_featu
 
 static int ipoib_change_mtu(struct net_device *dev, int new_mtu)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	/* dev->mtu > 2K ==> connected mode */
 	if (ipoib_cm_admin_enabled(dev)) {
@@ -468,7 +468,7 @@ static struct net_device *ipoib_get_net_dev_by_params(
 
 int ipoib_set_mode(struct net_device *dev, const char *buf)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if ((test_bit(IPOIB_FLAG_ADMIN_CM, &priv->flags) &&
 	     !strcmp(buf, "connected\n")) ||
@@ -505,7 +505,7 @@ int ipoib_set_mode(struct net_device *dev, const char *buf)
 
 struct ipoib_path *__path_find(struct net_device *dev, void *gid)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct rb_node *n = priv->path_tree.rb_node;
 	struct ipoib_path *path;
 	int ret;
@@ -529,7 +529,7 @@ struct ipoib_path *__path_find(struct net_device *dev, void *gid)
 
 static int __path_add(struct net_device *dev, struct ipoib_path *path)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct rb_node **n = &priv->path_tree.rb_node;
 	struct rb_node *pn = NULL;
 	struct ipoib_path *tpath;
@@ -564,7 +564,7 @@ static void path_free(struct net_device *dev, struct ipoib_path *path)
 	while ((skb = __skb_dequeue(&path->queue)))
 		dev_kfree_skb_irq(skb);
 
-	ipoib_dbg(netdev_priv(dev), "path_free\n");
+	ipoib_dbg(ipoib_priv(dev), "path_free\n");
 
 	/* remove all neigh connected to this path */
 	ipoib_del_neighs_by_gid(dev, path->pathrec.dgid.raw);
@@ -598,7 +598,7 @@ struct ipoib_path_iter *ipoib_path_iter_init(struct net_device *dev)
 
 int ipoib_path_iter_next(struct ipoib_path_iter *iter)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(iter->dev);
 	struct rb_node *n;
 	struct ipoib_path *path;
 	int ret = 1;
@@ -635,7 +635,7 @@ void ipoib_path_iter_read(struct ipoib_path_iter *iter,
 
 void ipoib_mark_paths_invalid(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_path *path, *tp;
 
 	spin_lock_irq(&priv->lock);
@@ -731,7 +731,7 @@ static void push_pseudo_header(struct sk_buff *skb, const char *daddr)
 
 void ipoib_flush_paths(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_path *path, *tp;
 	LIST_HEAD(remove_list);
 	unsigned long flags;
@@ -765,7 +765,7 @@ static void path_rec_completion(int status,
 {
 	struct ipoib_path *path = path_ptr;
 	struct net_device *dev = path->dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_ah *ah = NULL;
 	struct ipoib_ah *old_ah = NULL;
 	struct ipoib_neigh *neigh, *tn;
@@ -858,7 +858,7 @@ static void path_rec_completion(int status,
 
 static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_path *path;
 
 	if (!priv->broadcast)
@@ -886,7 +886,7 @@ static struct ipoib_path *path_rec_create(struct net_device *dev, void *gid)
 static int path_rec_start(struct net_device *dev,
 			  struct ipoib_path *path)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_dbg(priv, "Start path record lookup for %pI6\n",
 		  path->pathrec.dgid.raw);
@@ -917,7 +917,8 @@ static int path_rec_start(struct net_device *dev,
 static void neigh_add_path(struct sk_buff *skb, u8 *daddr,
 			   struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
 	struct ipoib_path *path;
 	struct ipoib_neigh *neigh;
 	unsigned long flags;
@@ -964,7 +965,9 @@ static void neigh_add_path(struct sk_buff *skb, u8 *daddr,
 			}
 		} else {
 			spin_unlock_irqrestore(&priv->lock, flags);
-			ipoib_send(dev, skb, path->ah, IPOIB_QPN(daddr));
+			rn->send(dev, skb, path->ah,
+						    IPOIB_QPN(daddr),
+						    priv->tx_wr.remote_qkey);
 			ipoib_neigh_put(neigh);
 			return;
 		}
@@ -998,7 +1001,8 @@ static void neigh_add_path(struct sk_buff *skb, u8 *daddr,
 static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 			     struct ipoib_pseudo_header *phdr)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
 	struct ipoib_path *path;
 	unsigned long flags;
 
@@ -1042,7 +1046,9 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 			  be16_to_cpu(path->pathrec.dlid));
 
 		spin_unlock_irqrestore(&priv->lock, flags);
-		ipoib_send(dev, skb, path->ah, IPOIB_QPN(phdr->hwaddr));
+		rn->send(dev, skb, path->ah,
+			 IPOIB_QPN(phdr->hwaddr),
+			 priv->tx_wr.remote_qkey);
 		return;
 	} else if ((path->query || !path_rec_start(dev, path)) &&
 		   skb_queue_len(&path->queue) < IPOIB_MAX_PATH_REC_QUEUE) {
@@ -1058,7 +1064,8 @@ static void unicast_arp_send(struct sk_buff *skb, struct net_device *dev,
 
 static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
 	struct ipoib_neigh *neigh;
 	struct ipoib_pseudo_header *phdr;
 	struct ipoib_header *header;
@@ -1122,7 +1129,9 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 			goto unref;
 		}
 	} else if (neigh->ah) {
-		ipoib_send(dev, skb, neigh->ah, IPOIB_QPN(phdr->hwaddr));
+		rn->send(dev, skb, neigh->ah,
+			 IPOIB_QPN(phdr->hwaddr),
+			 priv->tx_wr.remote_qkey);
 		goto unref;
 	}
 
@@ -1144,7 +1153,7 @@ static int ipoib_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 static void ipoib_timeout(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_warn(priv, "transmit timeout: latency %d msecs\n",
 		   jiffies_to_msecs(jiffies - dev_trans_start(dev)));
@@ -1178,7 +1187,7 @@ static int ipoib_hard_header(struct sk_buff *skb,
 
 static void ipoib_set_mcast_list(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if (!test_bit(IPOIB_FLAG_OPER_UP, &priv->flags)) {
 		ipoib_dbg(priv, "IPOIB_FLAG_OPER_UP not set");
@@ -1190,7 +1199,7 @@ static void ipoib_set_mcast_list(struct net_device *dev)
 
 static int ipoib_get_iflink(const struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	/* parent interface */
 	if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
@@ -1218,7 +1227,7 @@ static u32 ipoib_addr_hash(struct ipoib_neigh_hash *htbl, u8 *daddr)
 
 struct ipoib_neigh *ipoib_neigh_get(struct net_device *dev, u8 *daddr)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_neigh_table *ntbl = &priv->ntbl;
 	struct ipoib_neigh_hash *htbl;
 	struct ipoib_neigh *neigh = NULL;
@@ -1347,7 +1356,7 @@ static struct ipoib_neigh *ipoib_neigh_ctor(u8 *daddr,
 struct ipoib_neigh *ipoib_neigh_alloc(u8 *daddr,
 				      struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_neigh_table *ntbl = &priv->ntbl;
 	struct ipoib_neigh_hash *htbl;
 	struct ipoib_neigh *neigh;
@@ -1404,7 +1413,7 @@ void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
 {
 	/* neigh reference count was dropprd to zero */
 	struct net_device *dev = neigh->dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct sk_buff *skb;
 	if (neigh->ah)
 		ipoib_put_ah(neigh->ah);
@@ -1414,7 +1423,7 @@ void ipoib_neigh_dtor(struct ipoib_neigh *neigh)
 	}
 	if (ipoib_cm_get(neigh))
 		ipoib_cm_destroy_tx(ipoib_cm_get(neigh));
-	ipoib_dbg(netdev_priv(dev),
+	ipoib_dbg(ipoib_priv(dev),
 		  "neigh free for %06x %pI6\n",
 		  IPOIB_QPN(neigh->daddr),
 		  neigh->daddr + 4);
@@ -1436,7 +1445,7 @@ static void ipoib_neigh_reclaim(struct rcu_head *rp)
 void ipoib_neigh_free(struct ipoib_neigh *neigh)
 {
 	struct net_device *dev = neigh->dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_neigh_table *ntbl = &priv->ntbl;
 	struct ipoib_neigh_hash *htbl;
 	struct ipoib_neigh __rcu **np;
@@ -1519,7 +1528,7 @@ static void neigh_hash_free_rcu(struct rcu_head *head)
 
 void ipoib_del_neighs_by_gid(struct net_device *dev, u8 *gid)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ipoib_neigh_table *ntbl = &priv->ntbl;
 	struct ipoib_neigh_hash *htbl;
 	unsigned long flags;
@@ -1605,7 +1614,7 @@ static void ipoib_flush_neighs(struct ipoib_dev_priv *priv)
 
 static void ipoib_neigh_hash_uninit(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int stopped;
 
 	ipoib_dbg(priv, "ipoib_neigh_hash_uninit\n");
@@ -1624,7 +1633,7 @@ static void ipoib_neigh_hash_uninit(struct net_device *dev)
 
 void ipoib_dev_uninit_default(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_transport_dev_cleanup(dev);
 
@@ -1637,10 +1646,11 @@ void ipoib_dev_uninit_default(struct net_device *dev)
 	priv->tx_ring = NULL;
 }
 
-static int ipoib_dev_init_default(struct net_device *dev, struct ib_device *ca,
-				  int port)
+static int ipoib_dev_init_default(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+
+	netif_napi_add(dev, &priv->napi, ipoib_poll, NAPI_POLL_WEIGHT);
 
 	/* Allocate RX/TX "rings" to hold queued skbs */
 	priv->rx_ring =	kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
@@ -1651,17 +1661,22 @@ static int ipoib_dev_init_default(struct net_device *dev, struct ib_device *ca,
 	priv->tx_ring = vzalloc(ipoib_sendq_size * sizeof *priv->tx_ring);
 	if (!priv->tx_ring) {
 		printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
-		       ca->name, ipoib_sendq_size);
+		       priv->ca->name, ipoib_sendq_size);
 		goto out_rx_ring_cleanup;
 	}
 
 	/* priv->tx_head, tx_tail & tx_outstanding are already 0 */
 
-	if (ipoib_transport_dev_init(dev, ca)) {
-		pr_warn("%s: ipoib_transport_dev_init failed\n", ca->name);
+	if (ipoib_transport_dev_init(dev, priv->ca)) {
+		pr_warn("%s: ipoib_transport_dev_init failed\n", priv->ca->name);
 		goto out_tx_ring_cleanup;
 	}
 
+	/* after qp created set dev address */
+	priv->dev->dev_addr[1] = (priv->qp_num >> 16) & 0xff;
+	priv->dev->dev_addr[2] = (priv->qp_num >>  8) & 0xff;
+	priv->dev->dev_addr[3] = (priv->qp_num) & 0xff;
+
 	setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func,
 		    (unsigned long)dev);
 
@@ -1679,7 +1694,7 @@ static int ipoib_dev_init_default(struct net_device *dev, struct ib_device *ca,
 
 int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int ret = -ENOMEM;
 
 	priv->ca = ca;
@@ -1703,17 +1718,12 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 		goto clean_wq;
 	}
 
-	ret = ipoib_dev_init_default(dev, ca, port);
+	ret = priv->rn_ops->ndo_init(dev);
 	if (ret) {
 		pr_warn("%s failed to init HW resource\n", dev->name);
 		goto out_free_pd;
 	}
 
-	/* after qp created set dev address */
-	priv->dev->dev_addr[1] = (priv->qp->qp_num >> 16) & 0xff;
-	priv->dev->dev_addr[2] = (priv->qp->qp_num >>  8) & 0xff;
-	priv->dev->dev_addr[3] = (priv->qp->qp_num) & 0xff;
-
 	if (ipoib_neigh_hash_init(priv) < 0) {
 		pr_warn("%s failed to init neigh hash\n", dev->name);
 		goto out_dev_uninit;
@@ -1750,7 +1760,7 @@ int ipoib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 
 void ipoib_dev_cleanup(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv, *tcpriv;
+	struct ipoib_dev_priv *priv = ipoib_priv(dev), *cpriv, *tcpriv;
 	LIST_HEAD(head);
 
 	ASSERT_RTNL();
@@ -1780,7 +1790,7 @@ void ipoib_dev_cleanup(struct net_device *dev)
 
 static int ipoib_set_vf_link_state(struct net_device *dev, int vf, int link_state)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	return ib_set_vf_link_state(priv->ca, vf, priv->port, link_state);
 }
@@ -1788,7 +1798,7 @@ static int ipoib_set_vf_link_state(struct net_device *dev, int vf, int link_stat
 static int ipoib_get_vf_config(struct net_device *dev, int vf,
 			       struct ifla_vf_info *ivf)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int err;
 
 	err = ib_get_vf_config(priv->ca, vf, priv->port, ivf);
@@ -1802,7 +1812,7 @@ static int ipoib_get_vf_config(struct net_device *dev, int vf,
 
 static int ipoib_set_vf_guid(struct net_device *dev, int vf, u64 guid, int type)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if (type != IFLA_VF_IB_NODE_GUID && type != IFLA_VF_IB_PORT_GUID)
 		return -EINVAL;
@@ -1813,7 +1823,7 @@ static int ipoib_set_vf_guid(struct net_device *dev, int vf, u64 guid, int type)
 static int ipoib_get_vf_stats(struct net_device *dev, int vf,
 			      struct ifla_vf_stats *vf_stats)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	return ib_get_vf_stats(priv->ca, vf, priv->port, vf_stats);
 }
@@ -1851,21 +1861,12 @@ static int ipoib_get_vf_stats(struct net_device *dev, int vf,
 	.ndo_get_iflink		 = ipoib_get_iflink,
 };
 
-void ipoib_setup(struct net_device *dev)
+void ipoib_setup_common(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
-
-	if (priv->hca_caps & IB_DEVICE_VIRTUAL_FUNCTION)
-		dev->netdev_ops	= &ipoib_netdev_ops_vf;
-	else
-		dev->netdev_ops	= &ipoib_netdev_ops_pf;
-
 	dev->header_ops		 = &ipoib_header_ops;
 
 	ipoib_set_ethtool_ops(dev);
 
-	netif_napi_add(dev, &priv->napi, ipoib_poll, NAPI_POLL_WEIGHT);
-
 	dev->watchdog_timeo	 = HZ;
 
 	dev->flags		|= IFF_BROADCAST | IFF_MULTICAST;
@@ -1879,11 +1880,14 @@ void ipoib_setup(struct net_device *dev)
 	netif_keep_dst(dev);
 
 	memcpy(dev->broadcast, ipv4_bcast_addr, INFINIBAND_ALEN);
+}
 
-	priv->dev = dev;
+void ipoib_build_priv(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
+	priv->dev = dev;
 	spin_lock_init(&priv->lock);
-
 	init_rwsem(&priv->vlan_rwsem);
 
 	INIT_LIST_HEAD(&priv->path_list);
@@ -1901,22 +1905,97 @@ void ipoib_setup(struct net_device *dev)
 	INIT_DELAYED_WORK(&priv->neigh_reap_task, ipoib_reap_neigh);
 }
 
-struct ipoib_dev_priv *ipoib_intf_alloc(const char *name)
+static const struct net_device_ops ipoib_netdev_default_pf = {
+	.ndo_init		 = ipoib_dev_init_default,
+	.ndo_uninit		 = ipoib_dev_uninit_default,
+	.ndo_open		 = ipoib_ib_dev_open_default,
+	.ndo_stop		 = ipoib_ib_dev_stop_default,
+};
+
+static struct net_device *ipoib_create_netdev_default(struct ib_device *hca,
+						      const char *name,
+						      void (*setup)(struct net_device *))
 {
 	struct net_device *dev;
+	struct rdma_netdev *rn;
 
-	dev = alloc_netdev((int)sizeof(struct ipoib_dev_priv), name,
-			   NET_NAME_UNKNOWN, ipoib_setup);
+	dev = alloc_netdev((int)sizeof(struct ipoib_rdma_netdev),
+			   name,
+			   NET_NAME_UNKNOWN, setup);
 	if (!dev)
 		return NULL;
 
-	return netdev_priv(dev);
+	rn = netdev_priv(dev);
+
+	rn->send = ipoib_send;
+	rn->attach_mcast = ipoib_mcast_attach;
+	rn->detach_mcast = ipoib_mcast_detach;
+
+	dev->netdev_ops = &ipoib_netdev_default_pf;
+
+	return dev;
+}
+
+static struct net_device *ipoib_get_netdev(struct ib_device *hca, u8 port,
+					   const char *name)
+{
+	struct net_device *dev;
+
+	if (hca->alloc_rdma_netdev) {
+		dev = hca->alloc_rdma_netdev(hca, port,
+					     RDMA_NETDEV_IPOIB, name,
+					     NET_NAME_UNKNOWN,
+					     ipoib_setup_common);
+		if (IS_ERR_OR_NULL(dev) && PTR_ERR(dev) != ENOTSUPP)
+			return NULL;
+	}
+
+	if (!hca->alloc_rdma_netdev || PTR_ERR(dev) == ENOTSUPP)
+		dev = ipoib_create_netdev_default(hca, name,
+						  ipoib_setup_common);
+
+	return dev;
+}
+
+struct ipoib_dev_priv *ipoib_intf_alloc(struct ib_device *hca, u8 port,
+					const char *name)
+{
+	struct net_device *dev;
+	struct ipoib_dev_priv *priv;
+	struct rdma_netdev *rn;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv) {
+		pr_err("%s failed allocting priv\n", __func__);
+		return NULL;
+	}
+
+	dev = ipoib_get_netdev(hca, port, name);
+	if (!dev)
+		goto free_priv;
+
+	priv->rn_ops = dev->netdev_ops;
+
+	/* fixme : should be after the query_cap */
+	if (priv->hca_caps & IB_DEVICE_VIRTUAL_FUNCTION)
+		dev->netdev_ops	= &ipoib_netdev_ops_vf;
+	else
+		dev->netdev_ops	= &ipoib_netdev_ops_pf;
+
+	rn = netdev_priv(dev);
+	rn->clnt_priv = priv;
+	ipoib_build_priv(dev);
+
+	return priv;
+free_priv:
+	kfree(priv);
+	return NULL;
 }
 
 static ssize_t show_pkey(struct device *dev,
 			 struct device_attribute *attr, char *buf)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
+	struct ipoib_dev_priv *priv = ipoib_priv(to_net_dev(dev));
 
 	return sprintf(buf, "0x%04x\n", priv->pkey);
 }
@@ -1925,14 +2004,14 @@ static ssize_t show_pkey(struct device *dev,
 static ssize_t show_umcast(struct device *dev,
 			   struct device_attribute *attr, char *buf)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(to_net_dev(dev));
+	struct ipoib_dev_priv *priv = ipoib_priv(to_net_dev(dev));
 
 	return sprintf(buf, "%d\n", test_bit(IPOIB_FLAG_UMCAST, &priv->flags));
 }
 
 void ipoib_set_umcast(struct net_device *ndev, int umcast_val)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(ndev);
+	struct ipoib_dev_priv *priv = ipoib_priv(ndev);
 
 	if (umcast_val > 0) {
 		set_bit(IPOIB_FLAG_UMCAST, &priv->flags);
@@ -2005,7 +2084,7 @@ static int ipoib_check_lladdr(struct net_device *dev,
 
 static int ipoib_set_mac(struct net_device *dev, void *addr)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct sockaddr_storage *ss = addr;
 	int ret;
 
@@ -2094,7 +2173,7 @@ static struct net_device *ipoib_add_port(const char *format,
 	struct ib_port_attr attr;
 	int result = -ENOMEM;
 
-	priv = ipoib_intf_alloc(format);
+	priv = ipoib_intf_alloc(hca, port, format);
 	if (!priv)
 		goto alloc_mem_failed;
 
@@ -2224,7 +2303,7 @@ static void ipoib_add_one(struct ib_device *device)
 			continue;
 		dev = ipoib_add_port("ib%d", device, p);
 		if (!IS_ERR(dev)) {
-			priv = netdev_priv(dev);
+			priv = ipoib_priv(dev);
 			list_add_tail(&priv->list, dev_list);
 			count++;
 		}
@@ -2264,6 +2343,7 @@ static void ipoib_remove_one(struct ib_device *device, void *client_data)
 
 		unregister_netdev(priv->dev);
 		free_netdev(priv->dev);
+		kfree(priv);
 	}
 
 	kfree(dev_list);
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
index 69e146cdc306..a557a5e993fd 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_multicast.c
@@ -114,7 +114,7 @@ static void ipoib_mcast_free(struct ipoib_mcast *mcast)
 	struct net_device *dev = mcast->dev;
 	int tx_dropped = 0;
 
-	ipoib_dbg_mcast(netdev_priv(dev), "deleting multicast group %pI6\n",
+	ipoib_dbg_mcast(ipoib_priv(dev), "deleting multicast group %pI6\n",
 			mcast->mcmember.mgid.raw);
 
 	/* remove all neigh connected to this mcast */
@@ -158,7 +158,7 @@ static struct ipoib_mcast *ipoib_mcast_alloc(struct net_device *dev,
 
 static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct rb_node *n = priv->multicast_tree.rb_node;
 
 	while (n) {
@@ -182,7 +182,7 @@ static struct ipoib_mcast *__ipoib_mcast_find(struct net_device *dev, void *mgid
 
 static int __ipoib_mcast_add(struct net_device *dev, struct ipoib_mcast *mcast)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct rb_node **n = &priv->multicast_tree.rb_node, *pn = NULL;
 
 	while (*n) {
@@ -212,7 +212,8 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
 				   struct ib_sa_mcmember_rec *mcmember)
 {
 	struct net_device *dev = mcast->dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
 	struct ipoib_ah *ah;
 	int ret;
 	int set_qkey = 0;
@@ -260,8 +261,9 @@ static int ipoib_mcast_join_finish(struct ipoib_mcast *mcast,
 			return 0;
 		}
 
-		ret = ipoib_mcast_attach(dev, be16_to_cpu(mcast->mcmember.mlid),
-					 &mcast->mcmember.mgid, set_qkey);
+		ret = rn->attach_mcast(dev, priv->ca, &mcast->mcmember.mgid,
+				       be16_to_cpu(mcast->mcmember.mlid),
+				       set_qkey);
 		if (ret < 0) {
 			ipoib_warn(priv, "couldn't attach QP to multicast group %pI6\n",
 				   mcast->mcmember.mgid.raw);
@@ -375,7 +377,7 @@ static int ipoib_mcast_join_complete(int status,
 {
 	struct ipoib_mcast *mcast = multicast->context;
 	struct net_device *dev = mcast->dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	ipoib_dbg_mcast(priv, "%sjoin completion for %pI6 (status %d)\n",
 			test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags) ?
@@ -477,7 +479,7 @@ static int ipoib_mcast_join_complete(int status,
  */
 static int ipoib_mcast_join(struct net_device *dev, struct ipoib_mcast *mcast)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_sa_multicast *multicast;
 	struct ib_sa_mcmember_rec rec = {
 		.join_state = 1
@@ -678,7 +680,7 @@ void ipoib_mcast_join_task(struct work_struct *work)
 
 void ipoib_mcast_start_thread(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	unsigned long flags;
 
 	ipoib_dbg_mcast(priv, "starting multicast thread\n");
@@ -690,7 +692,7 @@ void ipoib_mcast_start_thread(struct net_device *dev)
 
 int ipoib_mcast_stop_thread(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	unsigned long flags;
 
 	ipoib_dbg_mcast(priv, "stopping multicast thread\n");
@@ -706,7 +708,8 @@ int ipoib_mcast_stop_thread(struct net_device *dev)
 
 static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
 	int ret = 0;
 
 	if (test_and_clear_bit(IPOIB_MCAST_FLAG_BUSY, &mcast->flags))
@@ -720,8 +723,8 @@ static int ipoib_mcast_leave(struct net_device *dev, struct ipoib_mcast *mcast)
 				mcast->mcmember.mgid.raw);
 
 		/* Remove ourselves from the multicast group */
-		ret = ib_detach_mcast(priv->qp, &mcast->mcmember.mgid,
-				      be16_to_cpu(mcast->mcmember.mlid));
+		ret = rn->detach_mcast(dev, priv->ca, &mcast->mcmember.mgid,
+				       be16_to_cpu(mcast->mcmember.mlid));
 		if (ret)
 			ipoib_warn(priv, "ib_detach_mcast failed (result = %d)\n", ret);
 	} else if (!test_bit(IPOIB_MCAST_FLAG_SENDONLY, &mcast->flags))
@@ -762,7 +765,8 @@ void ipoib_mcast_remove_list(struct list_head *remove_list)
 
 void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
 	struct ipoib_mcast *mcast;
 	unsigned long flags;
 	void *mgid = daddr + 4;
@@ -825,7 +829,8 @@ void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
 			}
 		}
 		spin_unlock_irqrestore(&priv->lock, flags);
-		ipoib_send(dev, skb, mcast->ah, IB_MULTICAST_QPN);
+		rn->send(dev, skb, mcast->ah, IB_MULTICAST_QPN,
+			 priv->tx_wr.remote_qkey);
 		if (neigh)
 			ipoib_neigh_put(neigh);
 		return;
@@ -837,7 +842,7 @@ void ipoib_mcast_send(struct net_device *dev, u8 *daddr, struct sk_buff *skb)
 
 void ipoib_mcast_dev_flush(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	LIST_HEAD(remove_list);
 	struct ipoib_mcast *mcast, *tmcast;
 	unsigned long flags;
@@ -1029,7 +1034,7 @@ struct ipoib_mcast_iter *ipoib_mcast_iter_init(struct net_device *dev)
 
 int ipoib_mcast_iter_next(struct ipoib_mcast_iter *iter)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(iter->dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(iter->dev);
 	struct rb_node *n;
 	struct ipoib_mcast *mcast;
 	int ret = 1;
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_netlink.c b/drivers/infiniband/ulp/ipoib/ipoib_netlink.c
index cdc7df4fdb8a..5d795aaea92c 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_netlink.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_netlink.c
@@ -44,7 +44,7 @@
 
 static int ipoib_fill_info(struct sk_buff *skb, const struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	u16 val;
 
 	if (nla_put_u16(skb, IFLA_IPOIB_PKEY, priv->pkey))
@@ -107,7 +107,7 @@ static int ipoib_new_child_link(struct net *src_net, struct net_device *dev,
 	if (!pdev || pdev->type != ARPHRD_INFINIBAND)
 		return -ENODEV;
 
-	ppriv = netdev_priv(pdev);
+	ppriv = ipoib_priv(pdev);
 
 	if (test_bit(IPOIB_FLAG_SUBINTERFACE, &ppriv->flags)) {
 		ipoib_warn(ppriv, "child creation disallowed for child devices\n");
@@ -129,7 +129,7 @@ static int ipoib_new_child_link(struct net *src_net, struct net_device *dev,
 	 */
 	child_pkey |= 0x8000;
 
-	err = __ipoib_vlan_add(ppriv, netdev_priv(dev), child_pkey, IPOIB_RTNL_CHILD);
+	err = __ipoib_vlan_add(ppriv, ipoib_priv(dev), child_pkey, IPOIB_RTNL_CHILD);
 
 	if (!err && data)
 		err = ipoib_changelink(dev, tb, data);
@@ -140,8 +140,8 @@ static void ipoib_unregister_child_dev(struct net_device *dev, struct list_head
 {
 	struct ipoib_dev_priv *priv, *ppriv;
 
-	priv = netdev_priv(dev);
-	ppriv = netdev_priv(priv->parent);
+	priv = ipoib_priv(dev);
+	ppriv = ipoib_priv(priv->parent);
 
 	down_write(&ppriv->vlan_rwsem);
 	unregister_netdevice_queue(dev, head);
@@ -161,7 +161,7 @@ static size_t ipoib_get_size(const struct net_device *dev)
 	.maxtype	= IFLA_IPOIB_MAX,
 	.policy		= ipoib_policy,
 	.priv_size	= sizeof(struct ipoib_dev_priv),
-	.setup		= ipoib_setup,
+	.setup		= ipoib_setup_common,
 	.newlink	= ipoib_new_child_link,
 	.changelink	= ipoib_changelink,
 	.dellink	= ipoib_unregister_child_dev,
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
index 529b9509a2c4..5f3dba9041e0 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_verbs.c
@@ -35,9 +35,10 @@
 
 #include "ipoib.h"
 
-int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid, int set_qkey)
+int ipoib_mcast_attach(struct net_device *dev, struct ib_device *hca,
+		       union ib_gid *mgid, u16 mlid, int set_qkey)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_attr *qp_attr = NULL;
 	int ret;
 	u16 pkey_index;
@@ -74,9 +75,20 @@ int ipoib_mcast_attach(struct net_device *dev, u16 mlid, union ib_gid *mgid, int
 	return ret;
 }
 
+int ipoib_mcast_detach(struct net_device *dev, struct ib_device *hca,
+		       union ib_gid *mgid, u16 mlid)
+{
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
+	int ret;
+
+	ret = ib_detach_mcast(priv->qp, mgid, mlid);
+
+	return ret;
+}
+
 int ipoib_init_qp(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	int ret;
 	struct ib_qp_attr qp_attr;
 	int attr_mask;
@@ -130,7 +142,7 @@ int ipoib_init_qp(struct net_device *dev)
 
 int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 	struct ib_qp_init_attr init_attr = {
 		.cap = {
 			.max_send_wr  = ipoib_sendq_size,
@@ -196,6 +208,8 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 		goto out_free_send_cq;
 	}
 
+	priv->qp_num = priv->qp->qp_num;
+
 	for (i = 0; i < MAX_SKB_FRAGS + 1; ++i)
 		priv->tx_sge[i].lkey = priv->pd->local_dma_lkey;
 
@@ -232,7 +246,7 @@ int ipoib_transport_dev_init(struct net_device *dev, struct ib_device *ca)
 
 void ipoib_transport_dev_cleanup(struct net_device *dev)
 {
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	if (priv->qp) {
 		if (ib_destroy_qp(priv->qp))
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
index deedb6fc1b05..ce1a12ba2fd9 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_vlan.c
@@ -43,7 +43,7 @@ static ssize_t show_parent(struct device *d, struct device_attribute *attr,
 			   char *buf)
 {
 	struct net_device *dev = to_net_dev(d);
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv = ipoib_priv(dev);
 
 	return sprintf(buf, "%s\n", priv->parent->name);
 }
@@ -127,14 +127,15 @@ int ipoib_vlan_add(struct net_device *pdev, unsigned short pkey)
 	if (!capable(CAP_NET_ADMIN))
 		return -EPERM;
 
-	ppriv = netdev_priv(pdev);
+	ppriv = ipoib_priv(pdev);
 
 	if (test_bit(IPOIB_FLAG_GOING_DOWN, &ppriv->flags))
 		return -EPERM;
 
 	snprintf(intf_name, sizeof intf_name, "%s.%04x",
 		 ppriv->dev->name, pkey);
-	priv = ipoib_intf_alloc(intf_name);
+
+	priv = ipoib_intf_alloc(ppriv->ca, ppriv->port, intf_name);
 	if (!priv)
 		return -ENOMEM;
 
@@ -182,7 +183,7 @@ int ipoib_vlan_delete(struct net_device *pdev, unsigned short pkey)
 	if (!capable(CAP_NET_ADMIN))
 		return -EPERM;
 
-	ppriv = netdev_priv(pdev);
+	ppriv = ipoib_priv(pdev);
 
 	if (test_bit(IPOIB_FLAG_GOING_DOWN, &ppriv->flags))
 		return -EPERM;
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[RFC v2 for accelerated IPoIB 6/6] mlx5_ib: skeleton for mlx5_ib to support ipoib_ops

[Erez Shitrit]

Signed-off-by: Erez Shitrit <erezsh-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
---
 drivers/infiniband/hw/mlx5/Makefile         |   2 +-
 drivers/infiniband/hw/mlx5/main.c           |   8 +
 drivers/infiniband/hw/mlx5/mlx5_ipoib_ops.c | 301 ++++++++++++++++++++++++++++
 3 files changed, 310 insertions(+), 1 deletion(-)
 create mode 100644 drivers/infiniband/hw/mlx5/mlx5_ipoib_ops.c

diff --git a/drivers/infiniband/hw/mlx5/Makefile b/drivers/infiniband/hw/mlx5/Makefile
index 90ad2adc752f..0c4caa339565 100644
--- a/drivers/infiniband/hw/mlx5/Makefile
+++ b/drivers/infiniband/hw/mlx5/Makefile
@@ -1,4 +1,4 @@
 obj-$(CONFIG_MLX5_INFINIBAND)	+= mlx5_ib.o
 
-mlx5_ib-y :=	main.o cq.o doorbell.o qp.o mem.o srq.o mr.o ah.o mad.o gsi.o ib_virt.o cmd.o
+mlx5_ib-y :=	main.o cq.o doorbell.o qp.o mem.o srq.o mr.o ah.o mad.o gsi.o ib_virt.o cmd.o mlx5_ipoib_ops.o
 mlx5_ib-$(CONFIG_INFINIBAND_ON_DEMAND_PAGING) += odp.o
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 4a043cf35b9a..cde29d64cc43 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -72,6 +72,13 @@ enum {
 	MLX5_ATOMIC_SIZE_QP_8BYTES = 1 << 3,
 };
 
+struct net_device *mlx5_alloc_rdma_netdev(struct ib_device *hca,
+					  u8 port_num,
+					  enum rdma_netdev_t type,
+					  const char *name,
+					  unsigned char name_assign_type,
+					  void (*setup)(struct net_device *));
+
 static enum rdma_link_layer
 mlx5_port_type_cap_to_rdma_ll(int port_type_cap)
 {
@@ -3436,6 +3443,7 @@ static void *mlx5_ib_add(struct mlx5_core_dev *mdev)
 	dev->ib_dev.alloc_mr		= mlx5_ib_alloc_mr;
 	dev->ib_dev.map_mr_sg		= mlx5_ib_map_mr_sg;
 	dev->ib_dev.check_mr_status	= mlx5_ib_check_mr_status;
+	dev->ib_dev.alloc_rdma_netdev	= mlx5_alloc_rdma_netdev;
 	dev->ib_dev.get_port_immutable  = mlx5_port_immutable;
 	dev->ib_dev.get_dev_fw_str      = get_dev_fw_str;
 	if (mlx5_core_is_pf(mdev)) {
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ipoib_ops.c b/drivers/infiniband/hw/mlx5/mlx5_ipoib_ops.c
new file mode 100644
index 000000000000..35debab63611
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/mlx5_ipoib_ops.c
@@ -0,0 +1,301 @@
+/*
+ * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/netdevice.h>
+#include <rdma/ib_ipoib_accel_ops.h>
+#include "mlx5_ib.h"
+#include <linux/mlx5/driver.h>
+#include <linux/mlx5/fs.h>
+
+/*FIX ME*/
+#include "../../ulp/ipoib/ipoib.h"
+
+#define IB_DEFAULT_Q_KEY   0xb1b
+
+int mlx5_ib_config_ipoib_qp(struct mlx5_ib_dev *ib_dev, struct mlx5_core_qp *qp)
+{
+	u32 *in;
+	struct mlx5_qp_context *context;
+	int inlen;
+	void *addr_path;
+	void *qpc;
+	int ret;
+
+	inlen = MLX5_ST_SZ_BYTES(create_qp_in);
+	in = mlx5_vzalloc(inlen);
+	if (!in)
+		return -ENOMEM;
+
+	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
+	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_UD);
+	MLX5_SET(qpc, qpc, pm_state, MLX5_QP_PM_MIGRATED);
+	MLX5_SET(qpc, qpc, ulp_stateless_offload_mode,
+		 MLX5_QP_ENHANCED_ULP_STATELESS_MODE);
+
+	addr_path = MLX5_ADDR_OF(qpc, qpc, primary_address_path);
+	MLX5_SET(ads, addr_path, port, 1);
+	MLX5_SET(ads, addr_path, grh, 1);
+
+	ret = mlx5_core_create_qp(ib_dev->mdev, qp, in, inlen);
+	if (ret) {
+		pr_err("Failed creating IPoIB QP err : %d\n", ret);
+		goto clean_inbox;
+	}
+
+	/* QP states */
+	context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (!context)
+		goto clean_inbox;
+
+	context->flags = cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
+	context->pri_path.port = 1;
+	context->qkey = cpu_to_be32(IB_DEFAULT_Q_KEY);
+
+	ret = mlx5_core_qp_modify(ib_dev->mdev,
+				  MLX5_CMD_OP_RST2INIT_QP, 0, context, qp);
+	if (ret) {
+		pr_warn("Failed to modify qp RST2INIT, err: %d\n", ret);
+		goto clean_in_modify;
+	}
+	memset(context, 0, sizeof(*context));
+
+	ret = mlx5_core_qp_modify(ib_dev->mdev,
+				  MLX5_CMD_OP_INIT2RTR_QP, 0, context,
+				  qp);
+	if (ret) {
+		pr_warn("Failed to modify qp INIT2RTR, err: %d\n", ret);
+		goto clean_in_modify;
+	}
+
+	ret = mlx5_core_qp_modify(ib_dev->mdev,
+				  MLX5_CMD_OP_RTR2RTS_QP, 0, context,
+				  qp);
+	if (ret) {
+		pr_warn("Failed to modify qp RTR2RTS, err: %d\n", ret);
+		goto clean_in_modify;
+	}
+
+	kvfree(in);
+	kfree(context);
+
+	return 0;
+
+clean_in_modify:
+	kvfree(in);
+clean_inbox:
+	kvfree(in);
+	return ret;
+}
+
+static void mlx5_ib_clean_qp(struct mlx5_ib_dev *ib_dev,
+			     struct mlx5_core_qp *qp)
+{
+	mlx5_core_destroy_qp(ib_dev->mdev, qp);
+	kfree(qp);
+}
+
+int mlx5_ib_dev_init(struct net_device *dev)
+{
+	void *next_priv = ipoib_accl_priv(dev);
+	struct ipoib_rdma_netdev *ipoib_accl_priv = netdev_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
+	struct mlx5_ib_dev *ib_dev;
+	int ret;
+
+	ib_dev = to_mdev(rn->hca);
+
+	ret = mlx5i_attach(ib_dev->mdev, next_priv);
+	if (ret) {
+		pr_err("Failed resources allocation for device: %s ret: %d\n",
+		       dev->name, ret);
+		return ret;
+	}
+
+	/* qp created set dev address */
+	dev->dev_addr[1] = (ipoib_accl_priv->qpn >> 16) & 0xff;
+	dev->dev_addr[2] = (ipoib_accl_priv->qpn >>  8) & 0xff;
+	dev->dev_addr[3] = (ipoib_accl_priv->qpn) & 0xff;
+
+	pr_debug("resources allocated for device: %s\n", dev->name);
+
+	return 0;
+}
+
+void mlx5_ib_dev_cleanup(struct net_device *dev)
+{
+	void *next_priv = ipoib_accl_priv(dev);
+	struct ipoib_rdma_netdev *ipoib_accl_priv = netdev_priv(dev);
+	struct rdma_netdev *rn = netdev_priv(dev);
+	struct mlx5_ib_dev *ib_dev;
+	struct mlx5_qp_context context;
+	int ret;
+
+	ib_dev = to_mdev(rn->hca);
+	/* detach qp from flow-steering by reset it */
+	ret = mlx5_core_qp_modify(ib_dev->mdev,
+				  MLX5_CMD_OP_2RST_QP, 0, &context,
+				  (struct mlx5_core_qp *)ipoib_accl_priv->context);
+	if (ret)
+		pr_err("%s failed (ret: %d) to reset QP\n", __func__, ret);
+
+	mlx5i_detach(ib_dev->mdev, next_priv);
+	ipoib_accl_priv = netdev_priv(dev);
+
+	mlx5_ib_clean_qp(ib_dev, (struct mlx5_core_qp *)ipoib_accl_priv->context/*rn->context*/);
+}
+
+int mlx5_ib_dev_open(struct net_device *dev)
+{
+	void *next_priv = ipoib_accl_priv(dev);
+
+	return mlx5i_open(next_priv);
+}
+
+int mlx5_ib_dev_stop(struct net_device *dev)
+{
+	void *next_priv = ipoib_accl_priv(dev);
+
+	return mlx5i_close(next_priv);
+}
+
+void mlx5_ib_send(struct net_device *dev, struct sk_buff *skb,
+		  struct ipoib_ah *address, u32 dqpn, u32 dqkey)
+{
+	void *next_priv = ipoib_accl_priv(dev);
+
+	mlx5i_xmit(skb, next_priv, &to_mah(address->ah)->av, dqpn, dqkey);
+}
+
+int mlx5_ib_attach_mcast(struct net_device *dev, struct ib_device *hca,
+			 union ib_gid *gid, u16 lid, int set_qkey)
+{
+	struct ipoib_rdma_netdev *ipoib_accl_priv = netdev_priv(dev);
+	struct mlx5_ib_dev *ib_dev = to_mdev(hca);
+	int err;
+
+	pr_debug("%s attaching QPN 0x%x, MGID %pI6\n",
+		 dev->name, ipoib_accl_priv->qpn, gid->raw);
+
+	err = mlx5_core_attach_mcg(ib_dev->mdev, gid, ipoib_accl_priv->qpn);
+	if (err)
+		pr_err("%s failed attaching QPN 0x%x, MGID %pI6\n",
+		       dev->name, ipoib_accl_priv->qpn, gid->raw);
+
+	return err;
+}
+
+int mlx5_ib_detach_mcast(struct net_device *dev, struct ib_device *hca,
+			 union ib_gid *gid, u16 lid)
+{
+	struct ipoib_rdma_netdev *ipoib_accl_priv = netdev_priv(dev);
+	struct mlx5_ib_dev *ib_dev = to_mdev(hca);
+	int err;
+
+	pr_debug("%s de-attaching QPN 0x%x, MGID %pI6\n",
+		 dev->name, ipoib_accl_priv->qpn, gid->raw);
+
+	err = mlx5_core_detach_mcg(ib_dev->mdev, gid, ipoib_accl_priv->qpn);
+	if (err)
+		pr_err("%s failed dettaching QPN 0x%x, MGID %pI6\n",
+		       dev->name, ipoib_accl_priv->qpn, gid->raw);
+
+	return err;
+}
+
+static const struct net_device_ops ipoib_accel_netdev_ops = {
+	.ndo_open		 = mlx5_ib_dev_open,
+	.ndo_stop		 = mlx5_ib_dev_stop,
+	.ndo_init		 = mlx5_ib_dev_init,
+	.ndo_uninit		 = mlx5_ib_dev_cleanup,
+};
+
+struct net_device *mlx5_alloc_rdma_netdev(struct ib_device *hca,
+					  u8 port_num,
+					  enum rdma_netdev_t type,
+					  const char *name,
+					  unsigned char name_assign_type,
+					  void (*setup)(struct net_device *))
+{
+	struct net_device *dev;
+	struct mlx5_ib_dev *ib_dev = to_mdev(hca);
+	struct mlx5_core_qp *qp;
+	struct mlx5i_create_ext_param param;
+	struct rdma_netdev *rn;
+	struct ipoib_rdma_netdev *ipoib_accl_priv;
+	int ret;
+
+	if (type != RDMA_NETDEV_IPOIB)
+		return ERR_PTR(ENOTSUPP);
+
+	/* new IPoIB QP */
+	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+	if (!qp) {
+		pr_warn("Failed allocate memroy for QP\n");
+		return NULL;
+	}
+
+	/* config and move qp to RTS */
+	ret = mlx5_ib_config_ipoib_qp(ib_dev, qp);
+	if (ret) {
+		pr_warn("Failed config IPoIB QP ret: %d\n", ret);
+		goto clean_qp;
+	}
+
+	param.qpn = qp->qpn;
+	dev = mlx5i_create_netdev(ib_dev->mdev, name, setup, &param);
+	if (!dev) {
+		pr_err("%s: Failed to create net device\n", __func__);
+		goto clean_qp;
+	}
+
+	pr_debug("%s qpn: %d created\n", __func__, qp->qpn);
+
+	/* set func pointers */
+	rn = netdev_priv(dev);
+	rn->hca = hca;
+	rn->send = mlx5_ib_send;
+	rn->attach_mcast = mlx5_ib_attach_mcast;
+	rn->detach_mcast = mlx5_ib_detach_mcast;
+
+	dev->netdev_ops = &ipoib_accel_netdev_ops;
+
+	/* private members */
+	ipoib_accl_priv = netdev_priv(dev);
+	ipoib_accl_priv->context = qp;
+	ipoib_accl_priv->qpn = qp->qpn;
+
+	return dev;
+
+clean_qp:
+	mlx5_ib_clean_qp(ib_dev, qp);
+	return NULL;
+}
+
-- 
1.8.3.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Jason Gunthorpe]

On Mon, Mar 20, 2017 at 09:08:22PM +0200, Erez Shitrit wrote:

>    TODO:
>         1.change the send api in order to move it to the ndo start_xmit (unless it hurts the performance of the default driver)
>         2.Take out the ipoib_ah from the send signature and use ib_ah instead, no need with including ipoib.h
>         3.Check if/how to add rdma_netdev layer to the default ipoib
>         4. splitting out the bulk rename of ipoib_priv into a single patch
>         5. change the name of the header to be ipoib_rn.h
>         6. no need to pass qkey, it is in the ah struct.

So.. why are you resending the series? Are these going to get fixed?

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1495 bytes --]

On Mon, Mar 20, 2017 at 02:59:27PM -0600, Jason Gunthorpe wrote:
> On Mon, Mar 20, 2017 at 09:08:22PM +0200, Erez Shitrit wrote:
>
> >    TODO:
> >         1.change the send api in order to move it to the ndo start_xmit (unless it hurts the performance of the default driver)
> >         2.Take out the ipoib_ah from the send signature and use ib_ah instead, no need with including ipoib.h
> >         3.Check if/how to add rdma_netdev layer to the default ipoib
> >         4. splitting out the bulk rename of ipoib_priv into a single patch
> >         5. change the name of the header to be ipoib_rn.h
> >         6. no need to pass qkey, it is in the ah struct.
>
> So.. why are you resending the series? Are these going to get fixed?

Sure, he will. This is why he added them to TODO so nothing will be lost.

The purpose of this RFC is to get clear view on the rdma_netdev design, understand
the API limitations and provide solid ground for actual IPoIB/HFI-VNIC patches.

It is important to mention, that this RFC is working and curious readers
can start playing with it and/or legacy IPoIB too.

So, the question is really simple. Are the interfaces properly designed
and Erez needs to start working on conversion this RFCs to actual patches?

Thanks

>
> Jason
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Jason Gunthorpe]

On Tue, Mar 21, 2017 at 01:49:55PM +0200, Leon Romanovsky wrote:
> On Mon, Mar 20, 2017 at 02:59:27PM -0600, Jason Gunthorpe wrote:
> > On Mon, Mar 20, 2017 at 09:08:22PM +0200, Erez Shitrit wrote:
> >
> > >    TODO:
> > >         1.change the send api in order to move it to the ndo start_xmit (unless it hurts the performance of the default driver)
> > >         2.Take out the ipoib_ah from the send signature and use ib_ah instead, no need with including ipoib.h
> > >         3.Check if/how to add rdma_netdev layer to the default ipoib
> > >         4. splitting out the bulk rename of ipoib_priv into a single patch
> > >         5. change the name of the header to be ipoib_rn.h
> > >         6. no need to pass qkey, it is in the ah struct.
> >
> > So.. why are you resending the series? Are these going to get fixed?
> 
> Sure, he will. This is why he added them to TODO so nothing will be lost.

Well.. if I assume everything done wrong in this series is fixed, then
it seems fine :)

It looks like the only difference from what Intel started with was the
attach/detach_mcast which is basically how I expected this to look.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Leon Romanovsky]

[-- Attachment #1: Type: text/plain, Size: 1614 bytes --]

On Tue, Mar 21, 2017 at 10:15:30AM -0600, Jason Gunthorpe wrote:
> On Tue, Mar 21, 2017 at 01:49:55PM +0200, Leon Romanovsky wrote:
> > On Mon, Mar 20, 2017 at 02:59:27PM -0600, Jason Gunthorpe wrote:
> > > On Mon, Mar 20, 2017 at 09:08:22PM +0200, Erez Shitrit wrote:
> > >
> > > >    TODO:
> > > >         1.change the send api in order to move it to the ndo start_xmit (unless it hurts the performance of the default driver)
> > > >         2.Take out the ipoib_ah from the send signature and use ib_ah instead, no need with including ipoib.h
> > > >         3.Check if/how to add rdma_netdev layer to the default ipoib
> > > >         4. splitting out the bulk rename of ipoib_priv into a single patch
> > > >         5. change the name of the header to be ipoib_rn.h
> > > >         6. no need to pass qkey, it is in the ah struct.
> > >
> > > So.. why are you resending the series? Are these going to get fixed?
> >
> > Sure, he will. This is why he added them to TODO so nothing will be lost.
>
> Well.. if I assume everything done wrong in this series is fixed, then
> it seems fine :)

At least, it is written as TODO :)

>
> It looks like the only difference from what Intel started with was the
> attach/detach_mcast which is basically how I expected this to look.

Glad to hear,

Erez,
It seems that the next step is to fix TODO and to send actual patches.

Thanks

>
> Jason
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Erez Shitrit]

On Tue, Mar 21, 2017 at 6:49 PM, Leon Romanovsky <leonro-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org> wrote:
> On Tue, Mar 21, 2017 at 10:15:30AM -0600, Jason Gunthorpe wrote:
>> On Tue, Mar 21, 2017 at 01:49:55PM +0200, Leon Romanovsky wrote:
>> > On Mon, Mar 20, 2017 at 02:59:27PM -0600, Jason Gunthorpe wrote:
>> > > On Mon, Mar 20, 2017 at 09:08:22PM +0200, Erez Shitrit wrote:
>> > >
>> > > >    TODO:
>> > > >         1.change the send api in order to move it to the ndo start_xmit (unless it hurts the performance of the default driver)
>> > > >         2.Take out the ipoib_ah from the send signature and use ib_ah instead, no need with including ipoib.h
>> > > >         3.Check if/how to add rdma_netdev layer to the default ipoib
>> > > >         4. splitting out the bulk rename of ipoib_priv into a single patch
>> > > >         5. change the name of the header to be ipoib_rn.h
>> > > >         6. no need to pass qkey, it is in the ah struct.
>> > >
>> > > So.. why are you resending the series? Are these going to get fixed?
>> >
>> > Sure, he will. This is why he added them to TODO so nothing will be lost.
>>
>> Well.. if I assume everything done wrong in this series is fixed, then
>> it seems fine :)
>
> At least, it is written as TODO :)
>
>>
>> It looks like the only difference from what Intel started with was the
>> attach/detach_mcast which is basically how I expected this to look.
>
> Glad to hear,
>
> Erez,
> It seems that the next step is to fix TODO and to send actual patches.

Hi Jason,

As Leon wrote above, these TODO items are going to be addressed in the
actual IPoIB offload patch series. While most of the items are doable,
there are two items, which I would like to discuss more deeply.

1. The qkey is not assigned at the initialization stage and the value
in ah struct is not correct. It looks like, there is no way to avoid
forwarding qkey (the item #6 - “no need to pass qkey, it is in the ah
struct”).

2. In order to use the send from the ndo, there is need to pass the
pointer to ib_av struct. The immediate solution is to embed it into
the skb->data (we have 20 bytes free from the pseudo header).

It can be something similar to the code below. It will work, but it
doesn’t look clean: (keeping pointer in the skb data..)

struct send_data
{
        struct ib_ah *av;
        u32 dqpn;
        u32 dqkey;
};

void push_send_data(struct sk_buff *skb, struct ib_ah *av, u32 dqpn,
                    u32 dqkey)
{
                struct send_data *sd;
                sd = (struct send_data *)skb_push(skb, sizeof(*sd));
                sd->av = av;
                sd->dqpn = dqpn;
                sd->dqkey = dqkey;

}

And while sending it do something like:

struct send_data *sd;
sd = (struct send_data *) skb->data;
skb_pull(skb, sizeof(*sd));
mah = to_mah(sd->av);
return mlx5i_xmit(skb, next_priv, &mah->av, sd->dqpn, sd->dqkey);


Another valuable option is just keep the send API as it is now.

Any thoughts on which implementation is better?

Thanks, Erez


>
> Thanks
>
>>
>> Jason
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
>> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Jason Gunthorpe]

On Wed, Mar 22, 2017 at 01:18:30PM +0200, Erez Shitrit wrote:

> 1. The qkey is not assigned at the initialization stage and the value
> in ah struct is not correct. It looks like, there is no way to avoid
> forwarding qkey (the item #6 - “no need to pass qkey, it is in the ah
> struct”).

It does make some sense that the qkey in the AH is not right, but it
still does not really make sense to forward it..

Just have the driver capture the qkey during 'attach_mcast'..

Looking at it more, there is something wrong with the implementation
in mlx5 since attach_mcast does not even use the qkey. I have no idea
how this works since the rx side must use a filter that only allows
packets that match the qkey for security.

Likely you will want to add a qkey argument to 'attach_mcast' to sort
this out..

> 2. In order to use the send from the ndo, there is need to pass the
> pointer to ib_av struct. The immediate solution is to embed it into
> the skb->data (we have 20 bytes free from the pseudo header).

Well, there are two solutions, one is to encode the ah in the pseudo
header during ipoib_hard_header. This is actually an optimization
because ipoib_hard_header is called bfore GSO splitting, so the ah can
be computed once instead of computed for every split packet. The
downside here is that it may be very complicated to store a kref in a
skb pseudo header and properly unref it at all the right times..

The other is to factor out the AH extraction from ipoib_start_xmit
into a helper function and have all the rn drivers call it to get the
AH.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Erez Shitrit]

On Wed, Mar 22, 2017 at 7:52 PM, Jason Gunthorpe
<jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org> wrote:
> On Wed, Mar 22, 2017 at 01:18:30PM +0200, Erez Shitrit wrote:
>
>> 1. The qkey is not assigned at the initialization stage and the value
>> in ah struct is not correct. It looks like, there is no way to avoid
>> forwarding qkey (the item #6 - “no need to pass qkey, it is in the ah
>> struct”).
>
> It does make some sense that the qkey in the AH is not right, but it
> still does not really make sense to forward it..
>
> Just have the driver capture the qkey during 'attach_mcast'..

I agree, will move it via the attach over the broadcast mcast.

>
> Looking at it more, there is something wrong with the implementation
> in mlx5 since attach_mcast does not even use the qkey. I have no idea
> how this works since the rx side must use a filter that only allows
> packets that match the qkey for security.

It works since the sm only generates 0xb1b  q_key value, and the mlx5 used that.
This was for the RFC only, I will update the q_key in the qp after the
join to the broadcast.

>
> Likely you will want to add a qkey argument to 'attach_mcast' to sort
> this out..

Yes. I will. thanks.

>
>> 2. In order to use the send from the ndo, there is need to pass the
>> pointer to ib_av struct. The immediate solution is to embed it into
>> the skb->data (we have 20 bytes free from the pseudo header).
>
> Well, there are two solutions, one is to encode the ah in the pseudo
> header during ipoib_hard_header. This is actually an optimization
> because ipoib_hard_header is called bfore GSO splitting, so the ah can
> be computed once instead of computed for every split packet. The
> downside here is that it may be very complicated to store a kref in a
> skb pseudo header and properly unref it at all the right times..

It is an optimization, and still we need to get the ah even at the
middle of GSO, for each skb, or to keep it somewhere, also will need
to keep more space in the skb, more processing.

>
> The other is to factor out the AH extraction from ipoib_start_xmit
> into a helper function and have all the rn drivers call it to get the
> AH.

We will pay with performance degradation for each send.

I think that right thing to do is to keep it that way, specific call
for send and not via the ndo.
All the others that we can do doesn't feel right.

>
> Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Jason Gunthorpe]

On Sun, Mar 26, 2017 at 07:17:13PM +0300, Erez Shitrit wrote:

> > be computed once instead of computed for every split packet. The
> > downside here is that it may be very complicated to store a kref in a
> > skb pseudo header and properly unref it at all the right times..
> 
> It is an optimization, and still we need to get the ah even at the
> middle of GSO, for each skb, or to keep it somewhere, also will need
> to keep more space in the skb, more processing.

GSO is done in a way where the pseudo header is retained but not
recomputed for every fragment, much like for ethernet.

Does mlx5 even need a 'AH'? Maybe allowing the driver to provide some
bytes in the pseudo header that match the HW SQE is an overall more
efficient direction.

> > The other is to factor out the AH extraction from ipoib_start_xmit
> > into a helper function and have all the rn drivers call it to get the
> > AH.
> 
> We will pay with performance degradation for each send.

Eh? Having ndo->ipoib->driver or ndo->driver->ipoib as a call chain
doesn't really matter one bit.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Erez Shitrit]

On Sun, Mar 26, 2017 at 11:42 PM, Jason Gunthorpe
<jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org> wrote:
> On Sun, Mar 26, 2017 at 07:17:13PM +0300, Erez Shitrit wrote:
>
>> > be computed once instead of computed for every split packet. The
>> > downside here is that it may be very complicated to store a kref in a
>> > skb pseudo header and properly unref it at all the right times..
>>
>> It is an optimization, and still we need to get the ah even at the
>> middle of GSO, for each skb, or to keep it somewhere, also will need
>> to keep more space in the skb, more processing.
>
> GSO is done in a way where the pseudo header is retained but not
> recomputed for every fragment, much like for ethernet.
>
> Does mlx5 even need a 'AH'? Maybe allowing the driver to provide some
> bytes in the pseudo header that match the HW SQE is an overall more
> efficient direction.

For mlx5/mlx4, the details from the ah are required, (in order to
build the datagram segment in the wqe(, but the ref count is seamless
since the ah is not a hw resource,  I can put that instead of the
pseudo header in the driver start_xmit
need to check if it pays with performance degradation.
will check and update.

>
>> > The other is to factor out the AH extraction from ipoib_start_xmit
>> > into a helper function and have all the rn drivers call it to get the
>> > AH.
>>
>> We will pay with performance degradation for each send.
>
> Eh? Having ndo->ipoib->driver or ndo->driver->ipoib as a call chain
> doesn't really matter one bit.

I meant that the extra call is time consuming, if we need one more
function (to extract the ah) that needs to be exported.

Anyway, will try with the first option.

>
> Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [RFC v2 for accelerated IPoIB 00/6] Enhanced mode for IPoIB driver

[Jason Gunthorpe]

On Tue, Mar 28, 2017 at 02:40:21PM +0300, Erez Shitrit wrote:
> > Does mlx5 even need a 'AH'? Maybe allowing the driver to provide some
> > bytes in the pseudo header that match the HW SQE is an overall more
> > efficient direction.
> 
> For mlx5/mlx4, the details from the ah are required, (in order to
> build the datagram segment in the wqe(,

I mean you can extract the details from the AH and store them in the
pseudo header specifically for mlx5. This avoids the need to store a
pointer.

> but the ref count is seamless since the ah is not a hw resource, I
> can put that instead of the pseudo header in the driver start_xmit

You can't just stuff a pointer into a skb, it needs to be kref'd
properly, which means when the skb is finally done someone kref_put's
that pointer. That is the hard part, IMHO.

The best option might be to store some kind of lookup hint that allows
the AH pointer to be found O(1) in normal cases, and does not require
kref trickery.

But optimizing this for GSO is not really related to this netdev
rework..

> > Eh? Having ndo->ipoib->driver or ndo->driver->ipoib as a call chain
> > doesn't really matter one bit.
> 
> I meant that the extra call is time consuming, if we need one more
> function (to extract the ah) that needs to be exported.

It isn't really extra and there isn't an issue exporting functions,
IHMO.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html