[5/5] net/netvsc: integrated VF support
diff mbox series

Message ID 20180830223512.21297-6-stephen@networkplumber.org
State Accepted, archived
Delegated to: Ferruh Yigit
Headers show
Series
  • netvsc changes for 18.11
Related show

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK

Commit Message

Stephen Hemminger Aug. 30, 2018, 10:35 p.m. UTC
From: Stephen Hemminger <sthemmin@microsoft.com>

Integrate accelerated networking support into netvsc PMD.
This allows netvsc to manage VF without using failsafe or vdev_netvsc.
For the exception vswitch path some tests like transmit
get a 22% increase in packets/sec.
For the VF path, the code is slightly shorter but has no
real change in performance.

Pro:
   * using netvsc is more like other DPDK NIC's
   * the exception packet uses less CPU
   * much smaller code size
   * no locking required on VF transmit/receive path
   * no legacy Linux network device to get mangled by userspace
   * much simpler (1K vs 9K) LOC
   * unified extended statistics

Con:
   * using netvsc has more complex startup model
   * no bifurcated driver support
   * no flow support (since host does not have flow API).
   * no tunnel offload support
   * no receive interrupt support

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
---
 devtools/checkpatches.sh               |   2 +-
 doc/guides/nics/netvsc.rst             |  12 +-
 doc/guides/rel_notes/release_18_11.rst |  14 +-
 drivers/net/netvsc/Makefile            |   1 +
 drivers/net/netvsc/hn_ethdev.c         | 109 +++++-
 drivers/net/netvsc/hn_nvs.c            |  11 +-
 drivers/net/netvsc/hn_nvs.h            |   9 +
 drivers/net/netvsc/hn_rndis.c          |  31 ++
 drivers/net/netvsc/hn_rndis.h          |   1 +
 drivers/net/netvsc/hn_rxtx.c           | 136 ++++---
 drivers/net/netvsc/hn_var.h            |  43 ++-
 drivers/net/netvsc/hn_vf.c             | 512 +++++++++++++++++++++++++
 drivers/net/netvsc/meson.build         |   2 +-
 13 files changed, 799 insertions(+), 84 deletions(-)
 create mode 100644 drivers/net/netvsc/hn_vf.c

Comments

Ferruh Yigit Sept. 14, 2018, 12:57 p.m. UTC | #1
On 8/30/2018 11:35 PM, Stephen Hemminger wrote:
> From: Stephen Hemminger <sthemmin@microsoft.com>
> 
> Integrate accelerated networking support into netvsc PMD.
> This allows netvsc to manage VF without using failsafe or vdev_netvsc.
> For the exception vswitch path some tests like transmit
> get a 22% increase in packets/sec.
> For the VF path, the code is slightly shorter but has no
> real change in performance.
> 
> Pro:
>    * using netvsc is more like other DPDK NIC's
>    * the exception packet uses less CPU
>    * much smaller code size
>    * no locking required on VF transmit/receive path
>    * no legacy Linux network device to get mangled by userspace
>    * much simpler (1K vs 9K) LOC
>    * unified extended statistics
> 
> Con:
>    * using netvsc has more complex startup model
>    * no bifurcated driver support
>    * no flow support (since host does not have flow API).
>    * no tunnel offload support
>    * no receive interrupt support
> 
> Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
> ---
>  devtools/checkpatches.sh               |   2 +-
>  doc/guides/nics/netvsc.rst             |  12 +-
>  doc/guides/rel_notes/release_18_11.rst |  14 +-
>  drivers/net/netvsc/Makefile            |   1 +
>  drivers/net/netvsc/hn_ethdev.c         | 109 +++++-
>  drivers/net/netvsc/hn_nvs.c            |  11 +-
>  drivers/net/netvsc/hn_nvs.h            |   9 +
>  drivers/net/netvsc/hn_rndis.c          |  31 ++
>  drivers/net/netvsc/hn_rndis.h          |   1 +
>  drivers/net/netvsc/hn_rxtx.c           | 136 ++++---
>  drivers/net/netvsc/hn_var.h            |  43 ++-
>  drivers/net/netvsc/hn_vf.c             | 512 +++++++++++++++++++++++++
>  drivers/net/netvsc/meson.build         |   2 +-
>  13 files changed, 799 insertions(+), 84 deletions(-)
>  create mode 100644 drivers/net/netvsc/hn_vf.c
> 
> diff --git a/devtools/checkpatches.sh b/devtools/checkpatches.sh
> index ba795ad1dc36..db7afc43d4c0 100755
> --- a/devtools/checkpatches.sh
> +++ b/devtools/checkpatches.sh
> @@ -1,4 +1,4 @@
> -#! /bin/sh
> +#! /bin/bash
>  # SPDX-License-Identifier: BSD-3-Clause
>  # Copyright 2015 6WIND S.A.

This tool patch shouldn't be part of driver update, would you mind sending a
separate patch for it?
Stephen Hemminger Sept. 14, 2018, 3:22 p.m. UTC | #2
On Fri, 14 Sep 2018 13:57:58 +0100
Ferruh Yigit <ferruh.yigit@intel.com> wrote:

> On 8/30/2018 11:35 PM, Stephen Hemminger wrote:
> > From: Stephen Hemminger <sthemmin@microsoft.com>
> > 
> > Integrate accelerated networking support into netvsc PMD.
> > This allows netvsc to manage VF without using failsafe or vdev_netvsc.
> > For the exception vswitch path some tests like transmit
> > get a 22% increase in packets/sec.
> > For the VF path, the code is slightly shorter but has no
> > real change in performance.
> > 
> > Pro:
> >    * using netvsc is more like other DPDK NIC's
> >    * the exception packet uses less CPU
> >    * much smaller code size
> >    * no locking required on VF transmit/receive path
> >    * no legacy Linux network device to get mangled by userspace
> >    * much simpler (1K vs 9K) LOC
> >    * unified extended statistics
> > 
> > Con:
> >    * using netvsc has more complex startup model
> >    * no bifurcated driver support
> >    * no flow support (since host does not have flow API).
> >    * no tunnel offload support
> >    * no receive interrupt support
> > 
> > Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
> > ---
> >  devtools/checkpatches.sh               |   2 +-
> >  doc/guides/nics/netvsc.rst             |  12 +-
> >  doc/guides/rel_notes/release_18_11.rst |  14 +-
> >  drivers/net/netvsc/Makefile            |   1 +
> >  drivers/net/netvsc/hn_ethdev.c         | 109 +++++-
> >  drivers/net/netvsc/hn_nvs.c            |  11 +-
> >  drivers/net/netvsc/hn_nvs.h            |   9 +
> >  drivers/net/netvsc/hn_rndis.c          |  31 ++
> >  drivers/net/netvsc/hn_rndis.h          |   1 +
> >  drivers/net/netvsc/hn_rxtx.c           | 136 ++++---
> >  drivers/net/netvsc/hn_var.h            |  43 ++-
> >  drivers/net/netvsc/hn_vf.c             | 512 +++++++++++++++++++++++++
> >  drivers/net/netvsc/meson.build         |   2 +-
> >  13 files changed, 799 insertions(+), 84 deletions(-)
> >  create mode 100644 drivers/net/netvsc/hn_vf.c
> > 
> > diff --git a/devtools/checkpatches.sh b/devtools/checkpatches.sh
> > index ba795ad1dc36..db7afc43d4c0 100755
> > --- a/devtools/checkpatches.sh
> > +++ b/devtools/checkpatches.sh
> > @@ -1,4 +1,4 @@
> > -#! /bin/sh
> > +#! /bin/bash
> >  # SPDX-License-Identifier: BSD-3-Clause
> >  # Copyright 2015 6WIND S.A.  
> 
> This tool patch shouldn't be part of driver update, would you mind sending a
> separate patch for it?


Sorry, that slipped in. Will drop it.

Patch
diff mbox series

diff --git a/devtools/checkpatches.sh b/devtools/checkpatches.sh
index ba795ad1dc36..db7afc43d4c0 100755
--- a/devtools/checkpatches.sh
+++ b/devtools/checkpatches.sh
@@ -1,4 +1,4 @@ 
-#! /bin/sh
+#! /bin/bash
 # SPDX-License-Identifier: BSD-3-Clause
 # Copyright 2015 6WIND S.A.
 
diff --git a/doc/guides/nics/netvsc.rst b/doc/guides/nics/netvsc.rst
index c5f9b7c6fa1e..87fabf5b8c80 100644
--- a/doc/guides/nics/netvsc.rst
+++ b/doc/guides/nics/netvsc.rst
@@ -34,14 +34,10 @@  In this release, the hyper PMD driver provides the basic functionality of packet
 *   The maximum number of queues is limited by the host (currently 64).
     When used with 4.16 kernel only a single queue is available.
 
-.. note::
-   This driver is intended for use with **Hyper-V only** and is
-   not recommended for use on Azure because accelerated Networking
-   (SR-IOV) is not supported.
-
-   On Azure, use the :doc:`vdev_netvsc` which
-   automatically configures the necessary TAP and failsave drivers.
-
+*   This driver supports SR-IOV network acceleration.
+    If SR-IOV is enabled then the driver will transparently manage the interface,
+    and send and receive packets using the VF path.
+    The VDEV_NETVSC and FAILSAFE drivers are *not* used when using netvsc PMD.
 
 Installation
 ------------
diff --git a/doc/guides/rel_notes/release_18_11.rst b/doc/guides/rel_notes/release_18_11.rst
index 3ae6b3f58b46..95c0801d0e6e 100644
--- a/doc/guides/rel_notes/release_18_11.rst
+++ b/doc/guides/rel_notes/release_18_11.rst
@@ -54,6 +54,12 @@  New Features
      Also, make sure to start the actual text at the margin.
      =========================================================
 
+* **Added support for SR-IOV in netvsc PMD.**
+
+  The ``netvsc`` poll mode driver now supports the Accelerated Networking
+  SR-IOV option in Hyper-V and Azure. This is an alternative to the previous
+  vdev_netvsc, tap, and failsafe drivers combination.
+
 
 API Changes
 -----------
@@ -122,7 +128,7 @@  The libraries prepended with a plus sign were incremented in this version.
      librte_bus_fslmc.so.1
      librte_bus_pci.so.1
      librte_bus_vdev.so.1
-     librte_bus_vmbus.so.1
+   + librte_bus_vmbus.so.1
      librte_cfgfile.so.2
      librte_cmdline.so.2
      librte_common_octeontx.so.1
@@ -159,6 +165,7 @@  The libraries prepended with a plus sign were incremented in this version.
      librte_pmd_ring.so.2
      librte_pmd_softnic.so.1
      librte_pmd_vhost.so.2
+   + librte_pmd_netvsc.so.1
      librte_port.so.3
      librte_power.so.1
      librte_rawdev.so.1
@@ -185,6 +192,10 @@  Known Issues
    Also, make sure to start the actual text at the margin.
    =========================================================
 
+* When using SR-IOV (VF) support with netvsc PMD and the Mellanox mlx5 bifurcated
+  driver; the Linux netvsc device must be brought up before the netvsc device is
+  unbound and passed to the DPDK.
+
 
 Tested Platforms
 ----------------
@@ -204,4 +215,3 @@  Tested Platforms
    This section is a comment. Do not overwrite or remove it.
    Also, make sure to start the actual text at the margin.
    =========================================================
-
diff --git a/drivers/net/netvsc/Makefile b/drivers/net/netvsc/Makefile
index 3c713af3c8fc..71482591a9c1 100644
--- a/drivers/net/netvsc/Makefile
+++ b/drivers/net/netvsc/Makefile
@@ -15,6 +15,7 @@  SRCS-$(CONFIG_RTE_LIBRTE_NETVSC_PMD) += hn_ethdev.c
 SRCS-$(CONFIG_RTE_LIBRTE_NETVSC_PMD) += hn_rxtx.c
 SRCS-$(CONFIG_RTE_LIBRTE_NETVSC_PMD) += hn_rndis.c
 SRCS-$(CONFIG_RTE_LIBRTE_NETVSC_PMD) += hn_nvs.c
+SRCS-$(CONFIG_RTE_LIBRTE_NETVSC_PMD) += hn_vf.c
 
 LDLIBS += -lrte_eal -lrte_mbuf -lrte_mempool -lrte_ring
 LDLIBS += -lrte_ethdev -lrte_net -lrte_kvargs
diff --git a/drivers/net/netvsc/hn_ethdev.c b/drivers/net/netvsc/hn_ethdev.c
index 63e04a453546..0aa5dab1b4b4 100644
--- a/drivers/net/netvsc/hn_ethdev.c
+++ b/drivers/net/netvsc/hn_ethdev.c
@@ -193,7 +193,7 @@  static int hn_parse_args(const struct rte_eth_dev *dev)
  */
 int
 hn_dev_link_update(struct rte_eth_dev *dev,
-		   __rte_unused int wait_to_complete)
+		   int wait_to_complete)
 {
 	struct hn_data *hv = dev->data->dev_private;
 	struct rte_eth_link link, old;
@@ -207,6 +207,8 @@  hn_dev_link_update(struct rte_eth_dev *dev,
 
 	hn_rndis_get_linkspeed(hv);
 
+	hn_vf_link_update(dev, wait_to_complete);
+
 	link = (struct rte_eth_link) {
 		.link_duplex = ETH_LINK_FULL_DUPLEX,
 		.link_autoneg = ETH_LINK_SPEED_FIXED,
@@ -245,6 +247,7 @@  static void hn_dev_info_get(struct rte_eth_dev *dev,
 	dev_info->max_tx_queues = hv->max_queues;
 
 	hn_rndis_get_offload(hv, dev_info);
+	hn_vf_info_get(hv, dev_info);
 }
 
 static void
@@ -395,7 +398,7 @@  static int hn_dev_configure(struct rte_eth_dev *dev)
 		}
 	}
 
-	return 0;
+	return hn_vf_configure(dev, dev_conf);
 }
 
 static int hn_dev_stats_get(struct rte_eth_dev *dev,
@@ -403,6 +406,8 @@  static int hn_dev_stats_get(struct rte_eth_dev *dev,
 {
 	unsigned int i;
 
+	hn_vf_stats_get(dev, stats);
+
 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
 		const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 
@@ -465,18 +470,38 @@  hn_dev_stats_reset(struct rte_eth_dev *dev)
 	}
 }
 
+static void
+hn_dev_xstats_reset(struct rte_eth_dev *dev)
+{
+	hn_dev_stats_reset(dev);
+	hn_vf_xstats_reset(dev);
+}
+
+static int
+hn_dev_xstats_count(struct rte_eth_dev *dev)
+{
+	int ret, count;
+
+	count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
+	count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
+
+	ret = hn_vf_xstats_get_names(dev, NULL, 0);
+	if (ret < 0)
+		return ret;
+
+	return count + ret;
+}
+
 static int
 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 			struct rte_eth_xstat_name *xstats_names,
-			__rte_unused unsigned int limit)
+			unsigned int limit)
 {
 	unsigned int i, t, count = 0;
-
-	PMD_INIT_FUNC_TRACE();
+	int ret;
 
 	if (!xstats_names)
-		return dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings)
-			+ dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
+		return hn_dev_xstats_count(dev);
 
 	/* Note: limit checked in rte_eth_xstats_names() */
 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
@@ -485,6 +510,9 @@  hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 		if (!txq)
 			continue;
 
+		if (count >= limit)
+			break;
+
 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 			snprintf(xstats_names[count++].name,
 				 RTE_ETH_XSTATS_NAME_SIZE,
@@ -497,6 +525,9 @@  hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 		if (!rxq)
 			continue;
 
+		if (count >= limit)
+			break;
+
 		for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 			snprintf(xstats_names[count++].name,
 				 RTE_ETH_XSTATS_NAME_SIZE,
@@ -504,7 +535,12 @@  hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 				 hn_stat_strings[t].name);
 	}
 
-	return count;
+	ret = hn_vf_xstats_get_names(dev, xstats_names + count,
+				     limit - count);
+	if (ret < 0)
+		return ret;
+
+	return count + ret;
 }
 
 static int
@@ -513,11 +549,9 @@  hn_dev_xstats_get(struct rte_eth_dev *dev,
 		  unsigned int n)
 {
 	unsigned int i, t, count = 0;
-
-	const unsigned int nstats =
-		dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings)
-		+ dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
+	const unsigned int nstats = hn_dev_xstats_count(dev);
 	const char *stats;
+	int ret;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -548,20 +582,33 @@  hn_dev_xstats_get(struct rte_eth_dev *dev,
 				(stats + hn_stat_strings[t].offset);
 	}
 
-	return count;
+	ret = hn_vf_xstats_get(dev, xstats + count, n - count);
+	if (ret < 0)
+		return ret;
+
+	return count + ret;
 }
 
 static int
 hn_dev_start(struct rte_eth_dev *dev)
 {
 	struct hn_data *hv = dev->data->dev_private;
+	int error;
 
 	PMD_INIT_FUNC_TRACE();
 
-	return hn_rndis_set_rxfilter(hv,
-				     NDIS_PACKET_TYPE_BROADCAST |
-				     NDIS_PACKET_TYPE_ALL_MULTICAST |
-				     NDIS_PACKET_TYPE_DIRECTED);
+	error = hn_rndis_set_rxfilter(hv,
+				      NDIS_PACKET_TYPE_BROADCAST |
+				      NDIS_PACKET_TYPE_ALL_MULTICAST |
+				      NDIS_PACKET_TYPE_DIRECTED);
+	if (error)
+		return error;
+
+	error = hn_vf_start(dev);
+	if (error)
+		hn_rndis_set_rxfilter(hv, 0);
+
+	return error;
 }
 
 static void
@@ -572,12 +619,15 @@  hn_dev_stop(struct rte_eth_dev *dev)
 	PMD_INIT_FUNC_TRACE();
 
 	hn_rndis_set_rxfilter(hv, 0);
+	hn_vf_stop(dev);
 }
 
 static void
 hn_dev_close(struct rte_eth_dev *dev __rte_unused)
 {
 	PMD_INIT_LOG(DEBUG, "close");
+
+	hn_vf_close(dev);
 }
 
 static const struct eth_dev_ops hn_eth_dev_ops = {
@@ -586,8 +636,7 @@  static const struct eth_dev_ops hn_eth_dev_ops = {
 	.dev_stop		= hn_dev_stop,
 	.dev_close		= hn_dev_close,
 	.dev_infos_get		= hn_dev_info_get,
-	.txq_info_get		= hn_dev_tx_queue_info,
-	.rxq_info_get		= hn_dev_rx_queue_info,
+	.dev_supported_ptypes_get = hn_vf_supported_ptypes,
 	.promiscuous_enable     = hn_dev_promiscuous_enable,
 	.promiscuous_disable    = hn_dev_promiscuous_disable,
 	.allmulticast_enable    = hn_dev_allmulticast_enable,
@@ -599,10 +648,10 @@  static const struct eth_dev_ops hn_eth_dev_ops = {
 	.rx_queue_release	= hn_dev_rx_queue_release,
 	.link_update		= hn_dev_link_update,
 	.stats_get		= hn_dev_stats_get,
+	.stats_reset            = hn_dev_stats_reset,
 	.xstats_get		= hn_dev_xstats_get,
 	.xstats_get_names	= hn_dev_xstats_get_names,
-	.stats_reset            = hn_dev_stats_reset,
-	.xstats_reset		= hn_dev_stats_reset,
+	.xstats_reset		= hn_dev_xstats_reset,
 };
 
 /*
@@ -680,6 +729,14 @@  eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 	if (err)
 		return err;
 
+	strlcpy(hv->owner.name, eth_dev->device->name,
+		RTE_ETH_MAX_OWNER_NAME_LEN);
+	err = rte_eth_dev_owner_new(&hv->owner.id);
+	if (err) {
+		PMD_INIT_LOG(ERR, "Can not get owner id");
+		return err;
+	}
+
 	/* Initialize primary channel input for control operations */
 	err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 	if (err)
@@ -715,6 +772,15 @@  eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 
 	hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
 
+	/* If VF was reported but not added, do it now */
+	if (hv->vf_present && !hv->vf_dev) {
+		PMD_INIT_LOG(DEBUG, "Adding VF device");
+
+		err = hn_vf_add(eth_dev, hv);
+		if (err)
+			goto failed;
+	}
+
 	return 0;
 
 failed:
@@ -744,6 +810,7 @@  eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
 	hn_detach(hv);
 	rte_vmbus_chan_close(hv->primary->chan);
 	rte_free(hv->primary);
+	rte_eth_dev_owner_delete(hv->owner.id);
 
 	eth_dev->data->mac_addrs = NULL;
 
diff --git a/drivers/net/netvsc/hn_nvs.c b/drivers/net/netvsc/hn_nvs.c
index a458bb720f82..9690c5f8a3e7 100644
--- a/drivers/net/netvsc/hn_nvs.c
+++ b/drivers/net/netvsc/hn_nvs.c
@@ -532,10 +532,19 @@  void
 hn_nvs_set_datapath(struct hn_data *hv, uint32_t path)
 {
 	struct hn_nvs_datapath dp;
+	int error;
+
+	PMD_DRV_LOG(DEBUG, "set datapath %s",
+		    path ? "VF" : "Synthetic");
 
 	memset(&dp, 0, sizeof(dp));
 	dp.type = NVS_TYPE_SET_DATAPATH;
 	dp.active_path = path;
 
-	hn_nvs_req_send(hv, &dp, sizeof(dp));
+	error = hn_nvs_req_send(hv, &dp, sizeof(dp));
+	if (error) {
+		PMD_DRV_LOG(ERR,
+			    "send set datapath failed: %d",
+			    error);
+	}
 }
diff --git a/drivers/net/netvsc/hn_nvs.h b/drivers/net/netvsc/hn_nvs.h
index 984a9c11c51d..2563fd8d8669 100644
--- a/drivers/net/netvsc/hn_nvs.h
+++ b/drivers/net/netvsc/hn_nvs.h
@@ -105,6 +105,12 @@  struct hn_nvs_ndis_init {
 	uint8_t		rsvd[28];
 } __rte_packed;
 
+struct hn_nvs_vf_association {
+	uint32_t	type;	/* NVS_TYPE_VFASSOC_NOTE */
+	uint32_t	allocated;
+	uint32_t	serial;
+} __rte_packed;
+
 #define NVS_DATAPATH_SYNTHETIC	0
 #define NVS_DATAPATH_VF		1
 
@@ -207,6 +213,9 @@  void	hn_nvs_detach(struct hn_data *hv);
 void	hn_nvs_ack_rxbuf(struct vmbus_channel *chan, uint64_t tid);
 int	hn_nvs_alloc_subchans(struct hn_data *hv, uint32_t *nsubch);
 void	hn_nvs_set_datapath(struct hn_data *hv, uint32_t path);
+void	hn_nvs_handle_vfassoc(struct rte_eth_dev *dev,
+			      const struct vmbus_chanpkt_hdr *hdr,
+			      const void *data);
 
 static inline int
 hn_nvs_send(struct vmbus_channel *chan, uint16_t flags,
diff --git a/drivers/net/netvsc/hn_rndis.c b/drivers/net/netvsc/hn_rndis.c
index a5d205ce2636..0610da3e139d 100644
--- a/drivers/net/netvsc/hn_rndis.c
+++ b/drivers/net/netvsc/hn_rndis.c
@@ -914,6 +914,37 @@  int hn_rndis_get_offload(struct hn_data *hv,
 	return 0;
 }
 
+uint32_t
+hn_rndis_get_ptypes(struct hn_data *hv)
+{
+	struct ndis_offload hwcaps;
+	uint32_t ptypes;
+	int error;
+
+	memset(&hwcaps, 0, sizeof(hwcaps));
+
+	error = hn_rndis_query_hwcaps(hv, &hwcaps);
+	if (error) {
+		PMD_DRV_LOG(ERR, "hwcaps query failed: %d", error);
+		return RTE_PTYPE_L2_ETHER;
+	}
+
+	ptypes = RTE_PTYPE_L2_ETHER;
+
+	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_IP4)
+		ptypes |= RTE_PTYPE_L3_IPV4;
+
+	if ((hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_TCP4) ||
+	    (hwcaps.ndis_csum.ndis_ip6_rxcsum & NDIS_RXCSUM_CAP_TCP6))
+		ptypes |= RTE_PTYPE_L4_TCP;
+
+	if ((hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_UDP4) ||
+	    (hwcaps.ndis_csum.ndis_ip6_rxcsum & NDIS_RXCSUM_CAP_UDP6))
+		ptypes |= RTE_PTYPE_L4_UDP;
+
+	return ptypes;
+}
+
 int
 hn_rndis_set_rxfilter(struct hn_data *hv, uint32_t filter)
 {
diff --git a/drivers/net/netvsc/hn_rndis.h b/drivers/net/netvsc/hn_rndis.h
index 01b5120631b1..319b497a7a70 100644
--- a/drivers/net/netvsc/hn_rndis.h
+++ b/drivers/net/netvsc/hn_rndis.h
@@ -24,6 +24,7 @@  int	hn_rndis_query_rsscaps(struct hn_data *hv,
 			       unsigned int *rxr_cnt0);
 int	hn_rndis_conf_rss(struct hn_data *hv,
 			  const struct rte_eth_rss_conf *rss_conf);
+uint32_t hn_rndis_get_ptypes(struct hn_data *hv);
 
 #ifdef RTE_LIBRTE_NETVSC_DEBUG_DUMP
 void hn_rndis_dump(const void *buf);
diff --git a/drivers/net/netvsc/hn_rxtx.c b/drivers/net/netvsc/hn_rxtx.c
index ad22a95f6c27..f4a36641b6c3 100644
--- a/drivers/net/netvsc/hn_rxtx.c
+++ b/drivers/net/netvsc/hn_rxtx.c
@@ -217,6 +217,7 @@  hn_dev_tx_queue_setup(struct rte_eth_dev *dev,
 	struct hn_data *hv = dev->data->dev_private;
 	struct hn_tx_queue *txq;
 	uint32_t tx_free_thresh;
+	int err;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -246,8 +247,14 @@  hn_dev_tx_queue_setup(struct rte_eth_dev *dev,
 
 	hn_reset_txagg(txq);
 
-	dev->data->tx_queues[queue_idx] = txq;
+	err = hn_vf_tx_queue_setup(dev, queue_idx, nb_desc,
+				     socket_id, tx_conf);
+	if (err) {
+		rte_free(txq);
+		return err;
+	}
 
+	dev->data->tx_queues[queue_idx] = txq;
 	return 0;
 }
 
@@ -270,17 +277,6 @@  hn_dev_tx_queue_release(void *arg)
 	rte_free(txq);
 }
 
-void
-hn_dev_tx_queue_info(struct rte_eth_dev *dev, uint16_t queue_idx,
-		     struct rte_eth_txq_info *qinfo)
-{
-	struct hn_data *hv = dev->data->dev_private;
-	struct hn_tx_queue *txq = dev->data->rx_queues[queue_idx];
-
-	qinfo->conf.tx_free_thresh = txq->free_thresh;
-	qinfo->nb_desc = hv->tx_pool->size;
-}
-
 static void
 hn_nvs_send_completed(struct rte_eth_dev *dev, uint16_t queue_id,
 		      unsigned long xactid, const struct hn_nvs_rndis_ack *ack)
@@ -713,6 +709,35 @@  hn_nvs_handle_rxbuf(struct rte_eth_dev *dev,
 	hn_rx_buf_release(rxb);
 }
 
+/*
+ * Called when NVS inband events are received.
+ * Send up a two part message with port_id and the NVS message
+ * to the pipe to the netvsc-vf-event control thread.
+ */
+static void hn_nvs_handle_notify(struct rte_eth_dev *dev,
+				 const struct vmbus_chanpkt_hdr *pkt,
+				 const void *data)
+{
+	const struct hn_nvs_hdr *hdr = data;
+
+	switch (hdr->type) {
+	case NVS_TYPE_TXTBL_NOTE:
+		/* Transmit indirection table has locking problems
+		 * in DPDK and therefore not implemented
+		 */
+		PMD_DRV_LOG(DEBUG, "host notify of transmit indirection table");
+		break;
+
+	case NVS_TYPE_VFASSOC_NOTE:
+		hn_nvs_handle_vfassoc(dev, pkt, data);
+		break;
+
+	default:
+		PMD_DRV_LOG(INFO,
+			    "got notify, nvs type %u", hdr->type);
+	}
+}
+
 struct hn_rx_queue *hn_rx_queue_alloc(struct hn_data *hv,
 				      uint16_t queue_id,
 				      unsigned int socket_id)
@@ -744,13 +769,14 @@  int
 hn_dev_rx_queue_setup(struct rte_eth_dev *dev,
 		      uint16_t queue_idx, uint16_t nb_desc,
 		      unsigned int socket_id,
-		      const struct rte_eth_rxconf *rx_conf __rte_unused,
+		      const struct rte_eth_rxconf *rx_conf,
 		      struct rte_mempool *mp)
 {
 	struct hn_data *hv = dev->data->dev_private;
 	char ring_name[RTE_RING_NAMESIZE];
 	struct hn_rx_queue *rxq;
 	unsigned int count;
+	int error = -ENOMEM;
 
 	PMD_INIT_FUNC_TRACE();
 
@@ -780,6 +806,11 @@  hn_dev_rx_queue_setup(struct rte_eth_dev *dev,
 	if (!rxq->rx_ring)
 		goto fail;
 
+	error = hn_vf_rx_queue_setup(dev, queue_idx, nb_desc,
+				     socket_id, rx_conf, mp);
+	if (error)
+		goto fail;
+
 	dev->data->rx_queues[queue_idx] = rxq;
 	return 0;
 
@@ -787,7 +818,7 @@  hn_dev_rx_queue_setup(struct rte_eth_dev *dev,
 	rte_ring_free(rxq->rx_ring);
 	rte_free(rxq->event_buf);
 	rte_free(rxq);
-	return -ENOMEM;
+	return error;
 }
 
 void
@@ -804,6 +835,9 @@  hn_dev_rx_queue_release(void *arg)
 	rxq->rx_ring = NULL;
 	rxq->mb_pool = NULL;
 
+	hn_vf_rx_queue_release(rxq->hv, rxq->queue_id);
+
+	/* Keep primary queue to allow for control operations */
 	if (rxq != rxq->hv->primary) {
 		rte_free(rxq->event_buf);
 		rte_free(rxq);
@@ -818,32 +852,6 @@  hn_dev_tx_done_cleanup(void *arg, uint32_t free_cnt)
 	return hn_process_events(txq->hv, txq->queue_id, free_cnt);
 }
 
-void
-hn_dev_rx_queue_info(struct rte_eth_dev *dev, uint16_t queue_idx,
-		     struct rte_eth_rxq_info *qinfo)
-{
-	struct hn_rx_queue *rxq = dev->data->rx_queues[queue_idx];
-
-	qinfo->mp = rxq->mb_pool;
-	qinfo->scattered_rx = 1;
-	qinfo->nb_desc = rte_ring_get_capacity(rxq->rx_ring);
-}
-
-static void
-hn_nvs_handle_notify(const struct vmbus_chanpkt_hdr *pkthdr,
-		     const void *data)
-{
-	const struct hn_nvs_hdr *hdr = data;
-
-	if (unlikely(vmbus_chanpkt_datalen(pkthdr) < sizeof(*hdr))) {
-		PMD_DRV_LOG(ERR, "invalid nvs notify");
-		return;
-	}
-
-	PMD_DRV_LOG(INFO,
-		    "got notify, nvs type %u", hdr->type);
-}
-
 /*
  * Process pending events on the channel.
  * Called from both Rx queue poll and Tx cleanup
@@ -916,7 +924,7 @@  uint32_t hn_process_events(struct hn_data *hv, uint16_t queue_id,
 			break;
 
 		case VMBUS_CHANPKT_TYPE_INBAND:
-			hn_nvs_handle_notify(pkt, data);
+			hn_nvs_handle_notify(dev, pkt, data);
 			break;
 
 		default:
@@ -1275,7 +1283,9 @@  uint16_t
 hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 {
 	struct hn_tx_queue *txq = ptxq;
+	uint16_t queue_id = txq->queue_id;
 	struct hn_data *hv = txq->hv;
+	struct rte_eth_dev *vf_dev;
 	bool need_sig = false;
 	uint16_t nb_tx;
 	int ret;
@@ -1283,6 +1293,15 @@  hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 	if (unlikely(hv->closed))
 		return 0;
 
+	/* Transmit over VF if present and up */
+	vf_dev = hv->vf_dev;
+	rte_compiler_barrier();
+	if (vf_dev && vf_dev->data->dev_started) {
+		void *sub_q = vf_dev->data->tx_queues[queue_id];
+
+		return (*vf_dev->tx_pkt_burst)(sub_q, tx_pkts, nb_pkts);
+	}
+
 	if (rte_mempool_avail_count(hv->tx_pool) <= txq->free_thresh)
 		hn_process_events(hv, txq->queue_id, 0);
 
@@ -1304,7 +1323,7 @@  hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 			if (unlikely(!pkt))
 				break;
 
-			hn_encap(pkt, txq->queue_id, m);
+			hn_encap(pkt, queue_id, m);
 			hn_append_to_chim(txq, pkt, m);
 
 			rte_pktmbuf_free(m);
@@ -1331,7 +1350,7 @@  hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 			txd->data_size += m->pkt_len;
 			++txd->packets;
 
-			hn_encap(pkt, txq->queue_id, m);
+			hn_encap(pkt, queue_id, m);
 
 			ret = hn_xmit_sg(txq, txd, m, &need_sig);
 			if (unlikely(ret != 0)) {
@@ -1360,15 +1379,36 @@  hn_recv_pkts(void *prxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 {
 	struct hn_rx_queue *rxq = prxq;
 	struct hn_data *hv = rxq->hv;
+	struct rte_eth_dev *vf_dev;
+	uint16_t nb_rcv;
 
 	if (unlikely(hv->closed))
 		return 0;
 
-	/* If ring is empty then process more */
-	if (rte_ring_count(rxq->rx_ring) < nb_pkts)
+	vf_dev = hv->vf_dev;
+	rte_compiler_barrier();
+
+	if (vf_dev && vf_dev->data->dev_started) {
+		/* Normally, with SR-IOV the ring buffer will be empty */
 		hn_process_events(hv, rxq->queue_id, 0);
 
-	/* Get mbufs off staging ring */
-	return rte_ring_sc_dequeue_burst(rxq->rx_ring, (void **)rx_pkts,
-					 nb_pkts, NULL);
+		/* Get mbufs some bufs off of staging ring */
+		nb_rcv = rte_ring_sc_dequeue_burst(rxq->rx_ring,
+						   (void **)rx_pkts,
+						   nb_pkts / 2, NULL);
+		/* And rest off of VF */
+		nb_rcv += rte_eth_rx_burst(vf_dev->data->port_id,
+					   rxq->queue_id,
+					   rx_pkts + nb_rcv, nb_pkts - nb_rcv);
+	} else {
+		/* If receive ring is not full then get more */
+		if (rte_ring_count(rxq->rx_ring) < nb_pkts)
+			hn_process_events(hv, rxq->queue_id, 0);
+
+		nb_rcv = rte_ring_sc_dequeue_burst(rxq->rx_ring,
+						   (void **)rx_pkts,
+						   nb_pkts, NULL);
+	}
+
+	return nb_rcv;
 }
diff --git a/drivers/net/netvsc/hn_var.h b/drivers/net/netvsc/hn_var.h
index 17b67941dc83..b8d9e5d5c470 100644
--- a/drivers/net/netvsc/hn_var.h
+++ b/drivers/net/netvsc/hn_var.h
@@ -94,8 +94,11 @@  struct hn_rx_bufinfo {
 struct hn_data {
 	struct rte_vmbus_device *vmbus;
 	struct hn_rx_queue *primary;
+	struct rte_eth_dev *vf_dev;		/* Subordinate device */
+	rte_spinlock_t  vf_lock;
 	uint16_t	port_id;
 	bool		closed;
+	bool		vf_present;
 	uint32_t	link_status;
 	uint32_t	link_speed;
 
@@ -124,6 +127,10 @@  struct hn_data {
 	uint8_t		rndis_resp[256];
 
 	struct ether_addr mac_addr;
+
+	struct rte_eth_dev_owner owner;
+	struct rte_intr_handle vf_intr;
+
 	struct vmbus_channel *channels[HN_MAX_CHANNELS];
 };
 
@@ -160,5 +167,37 @@  int	hn_dev_rx_queue_setup(struct rte_eth_dev *dev,
 			      const struct rte_eth_rxconf *rx_conf,
 			      struct rte_mempool *mp);
 void	hn_dev_rx_queue_release(void *arg);
-void	hn_dev_rx_queue_info(struct rte_eth_dev *dev, uint16_t queue_idx,
-			     struct rte_eth_rxq_info *qinfo);
+
+void	hn_vf_info_get(struct hn_data *hv,
+		       struct rte_eth_dev_info *info);
+int	hn_vf_add(struct rte_eth_dev *dev, struct hn_data *hv);
+int	hn_vf_configure(struct rte_eth_dev *dev,
+			const struct rte_eth_conf *dev_conf);
+const uint32_t *hn_vf_supported_ptypes(struct rte_eth_dev *dev);
+int	hn_vf_start(struct rte_eth_dev *dev);
+void	hn_vf_reset(struct rte_eth_dev *dev);
+void	hn_vf_stop(struct rte_eth_dev *dev);
+void	hn_vf_close(struct rte_eth_dev *dev);
+int	hn_vf_link_update(struct rte_eth_dev *dev,
+			  int wait_to_complete);
+int	hn_vf_tx_queue_setup(struct rte_eth_dev *dev,
+			     uint16_t queue_idx, uint16_t nb_desc,
+			     unsigned int socket_id,
+			     const struct rte_eth_txconf *tx_conf);
+void	hn_vf_tx_queue_release(struct hn_data *hv, uint16_t queue_id);
+int	hn_vf_rx_queue_setup(struct rte_eth_dev *dev,
+			     uint16_t queue_idx, uint16_t nb_desc,
+			     unsigned int socket_id,
+			     const struct rte_eth_rxconf *rx_conf,
+			     struct rte_mempool *mp);
+void	hn_vf_rx_queue_release(struct hn_data *hv, uint16_t queue_id);
+
+int	hn_vf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
+void	hn_vf_stats_reset(struct rte_eth_dev *dev);
+int	hn_vf_xstats_get_names(struct rte_eth_dev *dev,
+			       struct rte_eth_xstat_name *xstats_names,
+			       unsigned int size);
+int	hn_vf_xstats_get(struct rte_eth_dev *dev,
+			 struct rte_eth_xstat *xstats,
+			 unsigned int n);
+void	hn_vf_xstats_reset(struct rte_eth_dev *dev);
diff --git a/drivers/net/netvsc/hn_vf.c b/drivers/net/netvsc/hn_vf.c
new file mode 100644
index 000000000000..c68d180fd894
--- /dev/null
+++ b/drivers/net/netvsc/hn_vf.c
@@ -0,0 +1,512 @@ 
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2018 Microsoft Corp.
+ * All rights reserved.
+ */
+
+#include <stdio.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdbool.h>
+#include <errno.h>
+#include <unistd.h>
+#include <dirent.h>
+#include <sys/types.h>
+#include <sys/fcntl.h>
+#include <sys/uio.h>
+
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_ethdev_driver.h>
+#include <rte_lcore.h>
+#include <rte_memory.h>
+#include <rte_bus_vmbus.h>
+#include <rte_pci.h>
+#include <rte_bus_pci.h>
+#include <rte_log.h>
+#include <rte_string_fns.h>
+
+#include "hn_logs.h"
+#include "hn_var.h"
+#include "hn_nvs.h"
+
+/* Search for VF with matching MAC address, return port id */
+static int hn_vf_match(const struct rte_eth_dev *dev)
+{
+	const struct ether_addr *mac = dev->data->mac_addrs;
+	char buf[32];
+	int i;
+
+	ether_format_addr(buf, sizeof(buf), mac);
+	RTE_ETH_FOREACH_DEV(i) {
+		const struct rte_eth_dev *vf_dev = &rte_eth_devices[i];
+		const struct ether_addr *vf_mac = vf_dev->data->mac_addrs;
+
+		if (vf_dev == dev)
+			continue;
+
+		ether_format_addr(buf, sizeof(buf), vf_mac);
+		if (is_same_ether_addr(mac, vf_mac))
+			return i;
+	}
+	return -ENOENT;
+}
+
+/*
+ * Attach new PCI VF device and return the port_id
+ */
+static int hn_vf_attach(struct hn_data *hv, uint16_t port_id,
+			struct rte_eth_dev **vf_dev)
+{
+	struct rte_eth_dev_owner owner = { .id = RTE_ETH_DEV_NO_OWNER };
+	int ret;
+
+	ret = rte_eth_dev_owner_get(port_id, &owner);
+	if (ret < 0) {
+		PMD_DRV_LOG(ERR, "Can not find owner for port %d", port_id);
+		return ret;
+	}
+
+	if (owner.id != RTE_ETH_DEV_NO_OWNER) {
+		PMD_DRV_LOG(ERR, "Port %u already owned by other device %s",
+			    port_id, owner.name);
+		return -EBUSY;
+	}
+
+	ret = rte_eth_dev_owner_set(port_id, &hv->owner);
+	if (ret < 0) {
+		PMD_DRV_LOG(ERR, "Can set owner for port %d", port_id);
+		return ret;
+	}
+
+	PMD_DRV_LOG(DEBUG, "Attach VF device %u", port_id);
+	rte_smp_wmb();
+	*vf_dev = &rte_eth_devices[port_id];
+	return 0;
+}
+
+/* Add new VF device to synthetic device */
+int hn_vf_add(struct rte_eth_dev *dev, struct hn_data *hv)
+{
+	int port, err;
+
+	port = hn_vf_match(dev);
+	if (port < 0) {
+		PMD_DRV_LOG(NOTICE, "No matching MAC found");
+		return port;
+	}
+
+	rte_spinlock_lock(&hv->vf_lock);
+	if (hv->vf_dev) {
+		PMD_DRV_LOG(ERR, "VF already attached");
+		err = -EBUSY;
+	} else {
+		err = hn_vf_attach(hv, port, &hv->vf_dev);
+	}
+
+	if (err == 0) {
+		dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
+		hv->vf_intr = (struct rte_intr_handle) {
+			.fd = -1,
+			.type = RTE_INTR_HANDLE_EXT,
+		};
+		dev->intr_handle = &hv->vf_intr;
+		hn_nvs_set_datapath(hv, NVS_DATAPATH_VF);
+	}
+	rte_spinlock_unlock(&hv->vf_lock);
+
+	return err;
+}
+
+/* Remove new VF device */
+static void hn_vf_remove(struct hn_data *hv)
+{
+	struct rte_eth_dev *vf_dev;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (!vf_dev) {
+		PMD_DRV_LOG(ERR, "VF path not active");
+		rte_spinlock_unlock(&hv->vf_lock);
+		return;
+	}
+
+	/* Stop incoming packets from arriving on VF */
+	hn_nvs_set_datapath(hv, NVS_DATAPATH_SYNTHETIC);
+	hv->vf_dev = NULL;
+
+	/* Give back ownership */
+	rte_eth_dev_owner_unset(vf_dev->data->port_id, hv->owner.id);
+	rte_spinlock_unlock(&hv->vf_lock);
+}
+
+/* Handle VF association message from host */
+void
+hn_nvs_handle_vfassoc(struct rte_eth_dev *dev,
+		      const struct vmbus_chanpkt_hdr *hdr,
+		      const void *data)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	const struct hn_nvs_vf_association *vf_assoc = data;
+
+	if (unlikely(vmbus_chanpkt_datalen(hdr) < sizeof(*vf_assoc))) {
+		PMD_DRV_LOG(ERR, "invalid vf association NVS");
+		return;
+	}
+
+	PMD_DRV_LOG(DEBUG, "VF serial %u %s port %u",
+		    vf_assoc->serial,
+		    vf_assoc->allocated ? "add to" : "remove from",
+		    dev->data->port_id);
+
+	hv->vf_present = vf_assoc->allocated;
+
+	if (dev->state != RTE_ETH_DEV_ATTACHED)
+		return;
+
+	if (vf_assoc->allocated)
+		hn_vf_add(dev, hv);
+	else
+		hn_vf_remove(hv);
+}
+
+/*
+ * Merge the info from the VF and synthetic path.
+ * use the default config of the VF
+ * and the minimum number of queues and buffer sizes.
+ */
+static void hn_vf_info_merge(struct rte_eth_dev *vf_dev,
+			     struct rte_eth_dev_info *info)
+{
+	struct rte_eth_dev_info vf_info;
+
+	rte_eth_dev_info_get(vf_dev->data->port_id, &vf_info);
+
+	info->speed_capa = vf_info.speed_capa;
+	info->default_rxportconf = vf_info.default_rxportconf;
+	info->default_txportconf = vf_info.default_txportconf;
+
+	info->max_rx_queues = RTE_MIN(vf_info.max_rx_queues,
+				      info->max_rx_queues);
+	info->rx_offload_capa &= vf_info.rx_offload_capa;
+	info->rx_queue_offload_capa &= vf_info.rx_queue_offload_capa;
+	info->flow_type_rss_offloads &= vf_info.flow_type_rss_offloads;
+
+	info->max_tx_queues = RTE_MIN(vf_info.max_tx_queues,
+				      info->max_tx_queues);
+	info->tx_offload_capa &= vf_info.tx_offload_capa;
+	info->tx_queue_offload_capa &= vf_info.tx_queue_offload_capa;
+
+	info->min_rx_bufsize = RTE_MAX(vf_info.min_rx_bufsize,
+				       info->min_rx_bufsize);
+	info->max_rx_pktlen  = RTE_MAX(vf_info.max_rx_pktlen,
+				       info->max_rx_pktlen);
+}
+
+void hn_vf_info_get(struct hn_data *hv, struct rte_eth_dev_info *info)
+{
+	struct rte_eth_dev *vf_dev;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		hn_vf_info_merge(vf_dev, info);
+	rte_spinlock_unlock(&hv->vf_lock);
+}
+
+int hn_vf_link_update(struct rte_eth_dev *dev,
+		      int wait_to_complete)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int ret = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->link_update)
+		ret = (*vf_dev->dev_ops->link_update)(dev, wait_to_complete);
+	rte_spinlock_unlock(&hv->vf_lock);
+
+	return ret;
+}
+
+/* called when VF has link state interrupts enabled */
+static int hn_vf_lsc_event(uint16_t port_id __rte_unused,
+			   enum rte_eth_event_type event,
+			   void *cb_arg, void *out __rte_unused)
+{
+	struct rte_eth_dev *dev = cb_arg;
+
+	if (event != RTE_ETH_EVENT_INTR_LSC)
+		return 0;
+
+	/* if link state has changed pass on */
+	if (hn_dev_link_update(dev, 0) == 0)
+		return 0; /* no change */
+
+	return _rte_eth_dev_callback_process(dev,
+					     RTE_ETH_EVENT_INTR_LSC,
+					     NULL);
+}
+
+static int _hn_vf_configure(struct rte_eth_dev *dev,
+			    struct rte_eth_dev *vf_dev,
+			    const struct rte_eth_conf *dev_conf)
+{
+	struct rte_eth_conf vf_conf = *dev_conf;
+	uint16_t vf_port = vf_dev->data->port_id;
+	int ret;
+
+	if (dev_conf->intr_conf.lsc &&
+	    (vf_dev->data->dev_flags & RTE_ETH_DEV_INTR_LSC)) {
+		PMD_DRV_LOG(DEBUG, "enabling LSC for VF %u",
+			    vf_port);
+		vf_conf.intr_conf.lsc = 1;
+	} else {
+		PMD_DRV_LOG(DEBUG, "disabling LSC for VF %u",
+			    vf_port);
+		vf_conf.intr_conf.lsc = 0;
+	}
+
+	ret = rte_eth_dev_configure(vf_port,
+				    dev->data->nb_rx_queues,
+				    dev->data->nb_tx_queues,
+				    &vf_conf);
+	if (ret) {
+		PMD_DRV_LOG(ERR,
+			    "VF configuration failed: %d", ret);
+	} else if (vf_conf.intr_conf.lsc) {
+		ret = rte_eth_dev_callback_register(vf_port,
+						    RTE_ETH_DEV_INTR_LSC,
+						    hn_vf_lsc_event, dev);
+		if (ret)
+			PMD_DRV_LOG(ERR,
+				    "Failed to register LSC callback for VF %u",
+				    vf_port);
+	}
+	return ret;
+}
+
+/*
+ * Configure VF if present.
+ * Force VF to have same number of queues as synthetic device
+ */
+int hn_vf_configure(struct rte_eth_dev *dev,
+		    const struct rte_eth_conf *dev_conf)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int ret = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		ret = _hn_vf_configure(dev, vf_dev, dev_conf);
+	rte_spinlock_unlock(&hv->vf_lock);
+	return ret;
+}
+
+const uint32_t *hn_vf_supported_ptypes(struct rte_eth_dev *dev)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	const uint32_t *ptypes = NULL;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->dev_supported_ptypes_get)
+		ptypes = (*vf_dev->dev_ops->dev_supported_ptypes_get)(vf_dev);
+	rte_spinlock_unlock(&hv->vf_lock);
+
+	return ptypes;
+}
+
+int hn_vf_start(struct rte_eth_dev *dev)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int ret = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		ret = rte_eth_dev_start(vf_dev->data->port_id);
+	rte_spinlock_unlock(&hv->vf_lock);
+	return ret;
+}
+
+void hn_vf_stop(struct rte_eth_dev *dev)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		rte_eth_dev_stop(vf_dev->data->port_id);
+	rte_spinlock_unlock(&hv->vf_lock);
+}
+
+/* If VF is present, then cascade configuration down */
+#define VF_ETHDEV_FUNC(dev, func)				\
+	{							\
+		struct hn_data *hv = (dev)->data->dev_private;	\
+		struct rte_eth_dev *vf_dev;			\
+		rte_spinlock_lock(&hv->vf_lock);		\
+		vf_dev = hv->vf_dev;				\
+		if (vf_dev)					\
+			func(vf_dev->data->port_id);		\
+		rte_spinlock_unlock(&hv->vf_lock);		\
+	}
+
+void hn_vf_reset(struct rte_eth_dev *dev)
+{
+	VF_ETHDEV_FUNC(dev, rte_eth_dev_reset);
+}
+
+void hn_vf_close(struct rte_eth_dev *dev)
+{
+	VF_ETHDEV_FUNC(dev, rte_eth_dev_close);
+}
+
+void hn_vf_stats_reset(struct rte_eth_dev *dev)
+{
+	VF_ETHDEV_FUNC(dev, rte_eth_stats_reset);
+}
+
+int hn_vf_tx_queue_setup(struct rte_eth_dev *dev,
+			 uint16_t queue_idx, uint16_t nb_desc,
+			 unsigned int socket_id,
+			 const struct rte_eth_txconf *tx_conf)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int ret = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		ret = rte_eth_tx_queue_setup(vf_dev->data->port_id,
+					     queue_idx, nb_desc,
+					     socket_id, tx_conf);
+	rte_spinlock_unlock(&hv->vf_lock);
+	return ret;
+}
+
+void hn_vf_tx_queue_release(struct hn_data *hv, uint16_t queue_id)
+{
+	struct rte_eth_dev *vf_dev;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->tx_queue_release) {
+		void *subq = vf_dev->data->tx_queues[queue_id];
+
+		(*vf_dev->dev_ops->tx_queue_release)(subq);
+	}
+
+	rte_spinlock_unlock(&hv->vf_lock);
+}
+
+int hn_vf_rx_queue_setup(struct rte_eth_dev *dev,
+			 uint16_t queue_idx, uint16_t nb_desc,
+			 unsigned int socket_id,
+			 const struct rte_eth_rxconf *rx_conf,
+			 struct rte_mempool *mp)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int ret = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		ret = rte_eth_rx_queue_setup(vf_dev->data->port_id,
+					     queue_idx, nb_desc,
+					     socket_id, rx_conf, mp);
+	rte_spinlock_unlock(&hv->vf_lock);
+	return ret;
+}
+
+void hn_vf_rx_queue_release(struct hn_data *hv, uint16_t queue_id)
+{
+	struct rte_eth_dev *vf_dev;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->rx_queue_release) {
+		void *subq = vf_dev->data->rx_queues[queue_id];
+
+		(*vf_dev->dev_ops->rx_queue_release)(subq);
+	}
+	rte_spinlock_unlock(&hv->vf_lock);
+}
+
+int hn_vf_stats_get(struct rte_eth_dev *dev,
+		    struct rte_eth_stats *stats)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int ret = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev)
+		ret = rte_eth_stats_get(vf_dev->data->port_id, stats);
+	rte_spinlock_unlock(&hv->vf_lock);
+	return ret;
+}
+
+int hn_vf_xstats_get_names(struct rte_eth_dev *dev,
+			   struct rte_eth_xstat_name *names,
+			   unsigned int n)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int i, count = 0;
+	char tmp[RTE_ETH_XSTATS_NAME_SIZE];
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->xstats_get_names)
+		count = vf_dev->dev_ops->xstats_get_names(vf_dev, names, n);
+	rte_spinlock_unlock(&hv->vf_lock);
+
+	/* add vf_ prefix to xstat names */
+	if (names) {
+		for (i = 0; i < count; i++) {
+			snprintf(tmp, sizeof(tmp), "vf_%s", names[i].name);
+			strlcpy(names[i].name, tmp, sizeof(names[i].name));
+		}
+	}
+
+	return count;
+}
+
+int hn_vf_xstats_get(struct rte_eth_dev *dev,
+		     struct rte_eth_xstat *xstats,
+		     unsigned int n)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+	int count = 0;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->xstats_get)
+		count = vf_dev->dev_ops->xstats_get(vf_dev, xstats, n);
+	rte_spinlock_unlock(&hv->vf_lock);
+
+	return count;
+}
+
+void hn_vf_xstats_reset(struct rte_eth_dev *dev)
+{
+	struct hn_data *hv = dev->data->dev_private;
+	struct rte_eth_dev *vf_dev;
+
+	rte_spinlock_lock(&hv->vf_lock);
+	vf_dev = hv->vf_dev;
+	if (vf_dev && vf_dev->dev_ops->xstats_reset)
+		vf_dev->dev_ops->xstats_reset(vf_dev);
+	rte_spinlock_unlock(&hv->vf_lock);
+}
diff --git a/drivers/net/netvsc/meson.build b/drivers/net/netvsc/meson.build
index a717cdd48365..c8426971624f 100644
--- a/drivers/net/netvsc/meson.build
+++ b/drivers/net/netvsc/meson.build
@@ -3,7 +3,7 @@ 
 
 build = dpdk_conf.has('RTE_LIBRTE_VMBUS_BUS')
 version = 2
-sources = files('hn_ethdev.c', 'hn_rxtx.c', 'hn_rndis.c', 'hn_nvs.c')
+sources = files('hn_ethdev.c', 'hn_rxtx.c', 'hn_rndis.c', 'hn_nvs.c', 'hn_vf.c')
 
 deps += ['bus_vmbus' ]