Intel Adaptive Virtual Function native device driver plugin

Change-Id: If168a9c54baaa516ecbe78de2141f11c17aa2f53
Signed-off-by: Damjan Marion <damarion@cisco.com>

9 files changed, 2946 insertions, 0 deletions

diff --git a/src/plugins/avf/README.md b/src/plugins/avf/README.md
new file mode 100644
index 00000000000..422a6c07fe8
--- /dev/null
+++ b/src/plugins/avf/README.md
@@ -0,0 +1,94 @@
+# Intel AVF device plugin for VPP   
+
+##Overview
+This plugins provides native device support for intel Adaptive Virtual
+Function (AVF). AVF is driver specification for current and future
+Intel Virtual Function devices. AVF defines communication channel between
+Physical Funciton (PF) and VF.
+In essence, today this driver can be used only with 
+Intel XL710 / X710 / XXV710 adapters.
+
+##Prerequisites
+ * Driver requires newer i40e PF linux driver to be installed on the system,
+which supports virtualchnl interface. This code is tested with i40e driver
+version 2.4.6.
+
+* Driver requires MSI-X interrupt support, which is not supported by
+uio_pci_generic driver, so vfio-pci needs to be used. On systems without IOMMU
+vfio driver can still be used with recent kernels which support no-iommu mode.
+
+##Known issues
+This driver is still in experimental phase, however it shows very good 
+performance numbers. Following items are not implemented (yet).
+
+* Jumbo MTU support
+* Interrupt and adaptive mode
+* NUMA support
+
+## Usage
+### System setup
+
+1. load VFIO driver
+```
+sudo modprobe vfio-pci
+```
+
+2. (systems without IOMMU only) enable unsafe NOIOMMU mode
+```
+echo Y | sudo tee /sys/module/vfio/parameters/enable_unsafe_noiommu_mode
+```
+
+3. Create and bind SR-IOV virtual function(s)
+
+Following script creates VF, assigns MAC address and binds VF to vfio-pci
+```bash
+#!/bin/bash
+
+if [ $USER != "root" ] ; then
+	echo "Restarting script with sudo..."
+	sudo $0 ${*}
+	exit
+fi
+
+setup () {
+  cd /sys/bus/pci/devices/${1}
+  driver=$(basename $(readlink driver))
+  if [ "${driver}" != "i40e" ]; then
+    echo ${1} | tee driver/unbind
+    echo ${1} | tee /sys/bus/pci/drivers/i40e/bind
+  fi
+  ifname=$(basename net/*)
+  echo 0 | tee sriov_numvfs > /dev/null
+  echo 1 | tee sriov_numvfs > /dev/null
+  ip link set dev ${ifname} vf 0 mac ${2}
+  ip link show dev ${ifname}
+  vf=$(basename $(readlink virtfn0))
+  echo ${vf} | tee virtfn0/driver/unbind
+  echo vfio-pci | tee virtfn0/driver_override
+  echo ${vf} | sudo tee /sys/bus/pci/drivers/vfio-pci/bind
+  echo  | tee virtfn0/driver_override
+}
+
+# Setup one VF on PF 0000:3b:00.0 and assign MAC address
+setup 0000:3b:00.0 00:11:22:33:44:00
+# Setup one VF on PF 0000:3b:00.1 and assign MAC address
+setup 0000:3b:00.1 00:11:22:33:44:01
+```
+
+### Interface Cration
+Interfaces can be dynamically created by using following CLI:
+```
+create interface avf 0000:3b:02.0
+set int state AVF0/3b/2/0 up
+```
+
+### Interface Deletion
+Interface can be deleted with following CLI:
+```
+delete interface avf <interface name>
+```
+
+### Interface Statistics
+Interface statistics can be displayed with `sh hardware-interface <if-name>`
+command.
+
diff --git a/src/plugins/avf/avf.h b/src/plugins/avf/avf.h
new file mode 100644
index 00000000000..71e6f09e824
--- /dev/null
+++ b/src/plugins/avf/avf.h
@@ -0,0 +1,265 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <avf/virtchnl.h>
+
+#define foreach_avf_device_flags \
+  _(0, INITIALIZED, "initialized") \
+  _(1, ERROR, "error") \
+  _(2, ADMIN_UP, "admin-up") \
+  _(3, IOVA, "iova") \
+  _(4, LINK_UP, "link-up") \
+  _(5, SHARED_TXQ_LOCK, "shared-txq-lock") \
+  _(6, ELOG, "elog")
+
+enum
+{
+#define _(a, b, c) AVF_DEVICE_F_##b = (1 << a),
+  foreach_avf_device_flags
+#undef _
+};
+
+typedef struct
+{
+  u64 qword[4];
+} avf_rx_desc_t;
+
+STATIC_ASSERT_SIZEOF (avf_rx_desc_t, 32);
+
+typedef struct
+{
+  union
+  {
+    u64 qword[2];
+    u64x2 as_u64x2;
+  };
+} avf_tx_desc_t;
+
+STATIC_ASSERT_SIZEOF (avf_tx_desc_t, 16);
+
+typedef struct
+{
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+  volatile u32 *qrx_tail;
+  u16 next;
+  u16 size;
+  avf_rx_desc_t *descs;
+  u32 *bufs;
+  u16 n_bufs;
+} avf_rxq_t;
+
+typedef struct
+{
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+  volatile u32 *qtx_tail;
+  u16 next;
+  u16 size;
+  clib_spinlock_t lock;
+  avf_tx_desc_t *descs;
+  u32 *bufs;
+  u16 n_bufs;
+} avf_txq_t;
+
+typedef struct
+{
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+  u32 flags;
+  u32 per_interface_next_index;
+
+  u32 dev_instance;
+  u32 sw_if_index;
+  u32 hw_if_index;
+  vlib_pci_dev_handle_t pci_dev_handle;
+  void *bar0;
+
+  /* queues */
+  avf_rxq_t *rxqs;
+  avf_txq_t *txqs;
+
+  /* Admin queues */
+  avf_aq_desc_t *atq;
+  avf_aq_desc_t *arq;
+  void *atq_bufs;
+  void *arq_bufs;
+  u64 atq_bufs_pa;
+  u64 arq_bufs_pa;
+  u16 atq_next_slot;
+  u16 arq_next_slot;
+  virtchnl_pf_event_t *events;
+
+  u16 vsi_id;
+  u32 feature_bitmap;
+  u8 hwaddr[6];
+  u16 num_queue_pairs;
+  u16 max_vectors;
+  u16 max_mtu;
+  u32 rss_key_size;
+  u32 rss_lut_size;
+  virtchnl_link_speed_t link_speed;
+
+  /* stats */
+  virtchnl_eth_stats_t eth_stats;
+
+  /* error */
+  clib_error_t *error;
+} avf_device_t;
+
+typedef struct
+{
+  u32 bi;
+  u32 status;
+  u16 length;
+  u16 tail_length;
+  u8 ptype;
+  u8 error;
+} avf_rx_vector_entry_t;
+
+STATIC_ASSERT_SIZEOF (avf_rx_vector_entry_t, 16);
+
+#define AVF_RX_VECTOR_SZ VLIB_FRAME_SIZE
+
+enum
+{
+  AVF_PROCESS_EVENT_START = 1,
+  AVF_PROCESS_EVENT_STOP = 2,
+  AVF_PROCESS_EVENT_AQ_INT = 3,
+} avf_process_event_t;
+
+typedef struct
+{
+  CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+  avf_rx_vector_entry_t rx_vector[AVF_RX_VECTOR_SZ];
+  u32 *to_free;
+} avf_per_thread_data_t;
+
+typedef struct
+{
+  u8 next_node;
+  i8 buffer_advance;
+  u32 flags;
+} avf_ptype_t;
+
+STATIC_ASSERT (VNET_DEVICE_INPUT_N_NEXT_NODES < 256, "too many next nodes");
+
+typedef struct
+{
+  avf_device_t *devices;
+  avf_per_thread_data_t *per_thread_data;
+  vlib_physmem_region_index_t physmem_region;
+  int physmem_region_alloc;
+
+  /* 256 element array for ptype based lookup */
+  avf_ptype_t *ptypes;
+} avf_main_t;
+
+extern avf_main_t avf_main;
+
+typedef struct
+{
+  vlib_pci_addr_t addr;
+  int enable_elog;
+  /* return */
+  int rv;
+  clib_error_t *error;
+} avf_create_if_args_t;
+
+void avf_create_if (vlib_main_t * vm, avf_create_if_args_t * args);
+void avf_delete_if (vlib_main_t * vm, avf_device_t * ad);
+
+extern vlib_node_registration_t avf_input_node;
+extern vnet_device_class_t avf_device_class;
+uword avf_interface_tx (vlib_main_t * vm, vlib_node_runtime_t * node,
+			vlib_frame_t * frame);
+
+/* format.c */
+format_function_t format_avf_device;
+format_function_t format_avf_device_name;
+format_function_t format_avf_input_trace;
+
+static inline u32
+avf_get_u32 (void *start, int offset)
+{
+  return *(u32 *) (((u8 *) start) + offset);
+}
+
+static inline u64
+avf_get_u64 (void *start, int offset)
+{
+  return *(u64 *) (((u8 *) start) + offset);
+}
+
+static inline u32
+avf_get_u32_bits (void *start, int offset, int first, int last)
+{
+  u32 value = avf_get_u32 (start, offset);
+  if ((last == 0) && (first == 31))
+    return value;
+  value >>= last;
+  value &= (1 << (first - last + 1)) - 1;
+  return value;
+}
+
+static inline u64
+avf_get_u64_bits (void *start, int offset, int first, int last)
+{
+  u64 value = avf_get_u64 (start, offset);
+  if ((last == 0) && (first == 63))
+    return value;
+  value >>= last;
+  value &= (1 << (first - last + 1)) - 1;
+  return value;
+}
+
+static inline void
+avf_set_u32 (void *start, int offset, u32 value)
+{
+  (*(u32 *) (((u8 *) start) + offset)) = value;
+}
+
+static inline void
+avf_reg_write (avf_device_t * ad, u32 addr, u32 val)
+{
+  *(volatile u32 *) ((u8 *) ad->bar0 + addr) = val;
+}
+
+static inline u32
+avf_reg_read (avf_device_t * ad, u32 addr)
+{
+  return *(volatile u32 *) (ad->bar0 + addr);
+}
+
+static inline void
+avf_reg_flush (avf_device_t * ad)
+{
+  avf_reg_read (ad, AVFGEN_RSTAT);
+  asm volatile ("":::"memory");
+}
+
+typedef struct
+{
+  u32 next_index;
+  u32 hw_if_index;
+  avf_rx_vector_entry_t rxve;
+} avf_input_trace_t;
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/cli.c b/src/plugins/avf/cli.c
new file mode 100644
index 00000000000..ba9f5600f99
--- /dev/null
+++ b/src/plugins/avf/cli.c
@@ -0,0 +1,200 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+#include <stdint.h>
+#include <net/if.h>
+#include <sys/ioctl.h>
+#include <inttypes.h>
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vlib/pci/pci.h>
+#include <vnet/ethernet/ethernet.h>
+
+#include <avf/avf.h>
+
+static clib_error_t *
+avf_create_command_fn (vlib_main_t * vm, unformat_input_t * input,
+		       vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  avf_create_if_args_t args;
+
+  memset (&args, 0, sizeof (avf_create_if_args_t));
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return 0;
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "%U", unformat_vlib_pci_addr, &args.addr))
+	;
+      else if (unformat (line_input, "elog"))
+	args.enable_elog = 1;
+      else
+	return clib_error_return (0, "unknown input `%U'",
+				  format_unformat_error, input);
+    }
+  unformat_free (line_input);
+
+
+  avf_create_if (vm, &args);
+
+  return args.error;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (avf_create_command, static) = {
+  .path = "create interface avf",
+  .short_help = "create interface avf <pci-address>",
+  .function = avf_create_command_fn,
+};
+/* *INDENT-ON* */
+
+static clib_error_t *
+avf_delete_command_fn (vlib_main_t * vm, unformat_input_t * input,
+		       vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  u32 sw_if_index = ~0;
+  vnet_hw_interface_t *hw;
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad;
+  vnet_main_t *vnm = vnet_get_main ();
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return 0;
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "sw_if_index %d", &sw_if_index))
+	;
+      else if (unformat (line_input, "%U", unformat_vnet_sw_interface,
+			 vnm, &sw_if_index))
+	;
+      else
+	return clib_error_return (0, "unknown input `%U'",
+				  format_unformat_error, input);
+    }
+  unformat_free (line_input);
+
+  if (sw_if_index == ~0)
+    return clib_error_return (0,
+			      "please specify interface name or sw_if_index");
+
+  hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  if (hw == NULL || avf_device_class.index != hw->dev_class_index)
+    return clib_error_return (0, "not a AVF interface");
+
+  ad = pool_elt_at_index (am->devices, hw->dev_instance);
+
+  avf_delete_if (vm, ad);
+
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (avf_delete_command, static) = {
+  .path = "delete interface avf",
+  .short_help = "delete interface avf "
+    "{<interface> | sw_if_index <sw_idx>}",
+  .function = avf_delete_command_fn,
+};
+/* *INDENT-ON* */
+
+static clib_error_t *
+avf_test_command_fn (vlib_main_t * vm, unformat_input_t * input,
+		     vlib_cli_command_t * cmd)
+{
+  unformat_input_t _line_input, *line_input = &_line_input;
+  u32 sw_if_index = ~0;
+  vnet_hw_interface_t *hw;
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad;
+  vnet_main_t *vnm = vnet_get_main ();
+  int test_irq = 0, enable_elog = 0, disable_elog = 0;
+
+  /* Get a line of input. */
+  if (!unformat_user (input, unformat_line_input, line_input))
+    return 0;
+
+  while (unformat_check_input (line_input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (line_input, "sw_if_index %d", &sw_if_index))
+	;
+      else if (unformat (line_input, "irq"))
+	test_irq = 1;
+      else if (unformat (line_input, "elog-on"))
+	enable_elog = 1;
+      else if (unformat (line_input, "elog-off"))
+	disable_elog = 1;
+      else if (unformat (line_input, "%U", unformat_vnet_sw_interface,
+			 vnm, &sw_if_index))
+	;
+      else
+	return clib_error_return (0, "unknown input `%U'",
+				  format_unformat_error, input);
+    }
+  unformat_free (line_input);
+
+  if (sw_if_index == ~0)
+    return clib_error_return (0,
+			      "please specify interface name or sw_if_index");
+
+  hw = vnet_get_sup_hw_interface (vnm, sw_if_index);
+  if (hw == NULL || avf_device_class.index != hw->dev_class_index)
+    return clib_error_return (0, "not a AVF interface");
+
+  ad = pool_elt_at_index (am->devices, hw->dev_instance);
+
+  if (enable_elog)
+    ad->flags |= AVF_DEVICE_F_ELOG;
+
+  if (disable_elog)
+    ad->flags &= ~AVF_DEVICE_F_ELOG;
+
+  if (test_irq)
+    avf_reg_write (ad, AVFINT_DYN_CTL0, (1 << 0) | (3 << 3) | (1 << 2));
+
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VLIB_CLI_COMMAND (avf_test_command, static) = {
+  .path = "test avf",
+  .short_help = "test avf [<interface> | sw_if_index <sw_idx>] [irq] "
+    "[elog-on] [elog-off]",
+  .function = avf_test_command_fn,
+};
+/* *INDENT-ON* */
+
+clib_error_t *
+avf_cli_init (vlib_main_t * vm)
+{
+  return 0;
+}
+
+VLIB_INIT_FUNCTION (avf_cli_init);
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/device.c b/src/plugins/avf/device.c
new file mode 100644
index 00000000000..4a0fed19d44
--- /dev/null
+++ b/src/plugins/avf/device.c
@@ -0,0 +1,1243 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vlib/pci/pci.h>
+#include <vnet/ethernet/ethernet.h>
+
+#include <avf/avf.h>
+
+#define AVF_MBOX_LEN 64
+#define AVF_MBOX_BUF_SZ 512
+#define AVF_RXQ_SZ 512
+#define AVF_TXQ_SZ 512
+#define AVF_ITR_INT 8160
+
+#define PCI_VENDOR_ID_INTEL			0x8086
+#define PCI_DEVICE_ID_INTEL_AVF			0x1889
+#define PCI_DEVICE_ID_INTEL_X710_VF		0x154c
+#define PCI_DEVICE_ID_INTEL_X722_VF		0x37cd
+
+avf_main_t avf_main;
+
+static pci_device_id_t avf_pci_device_ids[] = {
+  {.vendor_id = PCI_VENDOR_ID_INTEL,.device_id = PCI_DEVICE_ID_INTEL_AVF},
+  {.vendor_id = PCI_VENDOR_ID_INTEL,.device_id = PCI_DEVICE_ID_INTEL_X710_VF},
+  {0},
+};
+
+//#define avf_log_debug(fmt, ...) fformat(stderr, "%s: " fmt "\n", __func__, __VA_ARGS__)
+#define avf_log_debug(fmt, ...)
+
+static inline void
+avf_irq_0_disable (avf_device_t * ad)
+{
+  u32 dyn_ctl0 = 0, icr0_ena = 0;
+
+  dyn_ctl0 |= (3 << 3);		/* 11b = No ITR update */
+
+  avf_reg_write (ad, AVFINT_ICR0_ENA1, icr0_ena);
+  avf_reg_write (ad, AVFINT_DYN_CTL0, dyn_ctl0);
+  avf_reg_flush (ad);
+}
+
+static inline void
+avf_irq_0_enable (avf_device_t * ad)
+{
+  u32 dyn_ctl0 = 0, icr0_ena = 0;
+
+  icr0_ena |= (1 << 30);	/* [30] Admin Queue Enable */
+
+  dyn_ctl0 |= (1 << 0);		/* [0] Interrupt Enable */
+  dyn_ctl0 |= (1 << 1);		/* [1] Clear PBA */
+  //dyn_ctl0 |= (3 << 3);               /* [4:3] ITR Index, 11b = No ITR update */
+  dyn_ctl0 |= ((AVF_ITR_INT / 2) << 5);	/* [16:5] ITR Interval in 2us steps */
+
+  avf_irq_0_disable (ad);
+  avf_reg_write (ad, AVFINT_ICR0_ENA1, icr0_ena);
+  avf_reg_write (ad, AVFINT_DYN_CTL0, dyn_ctl0);
+  avf_reg_flush (ad);
+}
+
+static inline void
+avf_irq_n_disable (avf_device_t * ad, u8 line)
+{
+  u32 dyn_ctln = 0;
+
+  avf_reg_write (ad, AVFINT_DYN_CTLN (line), dyn_ctln);
+  avf_reg_flush (ad);
+}
+
+static inline void
+avf_irq_n_enable (avf_device_t * ad, u8 line)
+{
+  u32 dyn_ctln = 0;
+
+  dyn_ctln |= (1 << 0);		/* [0] Interrupt Enable */
+  dyn_ctln |= (1 << 1);		/* [1] Clear PBA */
+  dyn_ctln |= ((AVF_ITR_INT / 2) << 5);	/* [16:5] ITR Interval in 2us steps */
+
+  avf_irq_n_disable (ad, line);
+  avf_reg_write (ad, AVFINT_DYN_CTLN (line), dyn_ctln);
+  avf_reg_flush (ad);
+}
+
+
+clib_error_t *
+avf_aq_desc_enq (vlib_main_t * vm, avf_device_t * ad, avf_aq_desc_t * dt,
+		 void *data, int len)
+{
+  clib_error_t *err = 0;
+  avf_aq_desc_t *d, dc;
+  int n_retry = 5;
+
+  d = &ad->atq[ad->atq_next_slot];
+  clib_memcpy (d, dt, sizeof (avf_aq_desc_t));
+  d->flags |= AVF_AQ_F_RD | AVF_AQ_F_SI;
+  if (len)
+    d->datalen = len;
+  if (len)
+    {
+      u64 pa;
+      pa = ad->atq_bufs_pa + ad->atq_next_slot * AVF_MBOX_BUF_SZ;
+      d->addr_hi = (u32) (pa >> 32);
+      d->addr_lo = (u32) pa;
+      clib_memcpy (ad->atq_bufs + ad->atq_next_slot * AVF_MBOX_BUF_SZ, data,
+		   len);
+      d->flags |= AVF_AQ_F_BUF;
+    }
+
+  if (ad->flags & AVF_DEVICE_F_ELOG)
+    clib_memcpy (&dc, d, sizeof (avf_aq_desc_t));
+
+  CLIB_MEMORY_BARRIER ();
+  avf_log_debug ("%U", format_hexdump, data, len);
+  ad->atq_next_slot = (ad->atq_next_slot + 1) % AVF_MBOX_LEN;
+  avf_reg_write (ad, AVF_ATQT, ad->atq_next_slot);
+  avf_reg_flush (ad);
+
+retry:
+  vlib_process_suspend (vm, 10e-6);
+
+  if (((d->flags & AVF_AQ_F_DD) == 0) || ((d->flags & AVF_AQ_F_CMP) == 0))
+    {
+      if (--n_retry == 0)
+	{
+	  err = clib_error_return (0, "adminq enqueue timeout [opcode 0x%x]",
+				   d->opcode);
+	  goto done;
+	}
+      goto retry;
+    }
+
+  clib_memcpy (dt, d, sizeof (avf_aq_desc_t));
+  if (d->flags & AVF_AQ_F_ERR)
+    return clib_error_return (0, "adminq enqueue error [opcode 0x%x, retval "
+			      "%d]", d->opcode, d->retval);
+
+done:
+  if (ad->flags & AVF_DEVICE_F_ELOG)
+    {
+      /* *INDENT-OFF* */
+      ELOG_TYPE_DECLARE (el) =
+	{
+	  .format = "avf[%d] aq enq: s_flags 0x%x r_flags 0x%x opcode 0x%x "
+	    "datalen %d retval %d",
+	  .format_args = "i4i2i2i2i2i2",
+	};
+      struct
+	{
+	  u32 dev_instance;
+	  u16 s_flags;
+	  u16 r_flags;
+	  u16 opcode;
+	  u16 datalen;
+	  u16 retval;
+	} *ed;
+      ed = ELOG_DATA (&vm->elog_main, el);
+      ed->dev_instance = ad->dev_instance;
+      ed->s_flags = dc.flags;
+      ed->r_flags = d->flags;
+      ed->opcode = dc.opcode;
+      ed->datalen = dc.datalen;
+      ed->retval = d->retval;
+      /* *INDENT-ON* */
+    }
+
+  return err;
+}
+
+clib_error_t *
+avf_cmd_rx_ctl_reg_write (vlib_main_t * vm, avf_device_t * ad, u32 reg,
+			  u32 val)
+{
+  clib_error_t *err;
+  avf_aq_desc_t d = {.opcode = 0x207,.param1 = reg,.param3 = val };
+  err = avf_aq_desc_enq (vm, ad, &d, 0, 0);
+
+  if (ad->flags & AVF_DEVICE_F_ELOG)
+    {
+      /* *INDENT-OFF* */
+      ELOG_TYPE_DECLARE (el) =
+	{
+	  .format = "avf[%d] rx ctl reg write: reg 0x%x val 0x%x ",
+	  .format_args = "i4i4i4",
+	};
+      struct
+	{
+	  u32 dev_instance;
+	  u32 reg;
+	  u32 val;
+	} *ed;
+      ed = ELOG_DATA (&vm->elog_main, el);
+      ed->dev_instance = ad->dev_instance;
+      ed->reg = reg;
+      ed->val = val;
+      /* *INDENT-ON* */
+    }
+  return err;
+}
+
+clib_error_t *
+avf_rxq_init (vlib_main_t * vm, avf_device_t * ad, u16 qid)
+{
+  avf_main_t *am = &avf_main;
+  avf_rxq_t *rxq;
+  clib_error_t *error = 0;
+  u32 n_alloc, i;
+
+  vec_validate_aligned (ad->rxqs, qid, CLIB_CACHE_LINE_BYTES);
+  rxq = vec_elt_at_index (ad->rxqs, qid);
+  rxq->size = AVF_RXQ_SZ;
+  rxq->next = 0;
+  rxq->descs = vlib_physmem_alloc_aligned (vm, am->physmem_region, &error,
+					   rxq->size * sizeof (avf_rx_desc_t),
+					   2 * CLIB_CACHE_LINE_BYTES);
+  memset (rxq->descs, 0, rxq->size * sizeof (avf_rx_desc_t));
+  vec_validate_aligned (rxq->bufs, rxq->size, CLIB_CACHE_LINE_BYTES);
+  rxq->qrx_tail = ad->bar0 + AVF_QRX_TAIL (qid);
+
+  n_alloc = vlib_buffer_alloc (vm, rxq->bufs, rxq->size - 8);
+
+  if (n_alloc == 0)
+    return clib_error_return (0, "buffer allocation error");
+
+  rxq->n_bufs = n_alloc;
+  avf_rx_desc_t *d = rxq->descs;
+  for (i = 0; i < n_alloc; i++)
+    {
+      if (ad->flags & AVF_DEVICE_F_IOVA)
+	{
+	  vlib_buffer_t *b = vlib_get_buffer (vm, rxq->bufs[i]);
+	  d->qword[0] = pointer_to_uword (b->data);
+	}
+      else
+	d->qword[0] =
+	  vlib_get_buffer_data_physical_address (vm, rxq->bufs[i]);
+      d++;
+    }
+  return 0;
+}
+
+clib_error_t *
+avf_txq_init (vlib_main_t * vm, avf_device_t * ad, u16 qid)
+{
+  avf_main_t *am = &avf_main;
+  avf_txq_t *txq;
+  clib_error_t *error = 0;
+
+  if (qid >= ad->num_queue_pairs)
+    {
+      qid = qid % ad->num_queue_pairs;
+      txq = vec_elt_at_index (ad->txqs, qid);
+      if (txq->lock == 0)
+	clib_spinlock_init (&txq->lock);
+      ad->flags |= AVF_DEVICE_F_SHARED_TXQ_LOCK;
+      return 0;
+    }
+
+  vec_validate_aligned (ad->txqs, qid, CLIB_CACHE_LINE_BYTES);
+  txq = vec_elt_at_index (ad->txqs, qid);
+  txq->size = AVF_TXQ_SZ;
+  txq->next = 0;
+  txq->descs = vlib_physmem_alloc_aligned (vm, am->physmem_region, &error,
+					   txq->size * sizeof (avf_tx_desc_t),
+					   2 * CLIB_CACHE_LINE_BYTES);
+  vec_validate_aligned (txq->bufs, txq->size, CLIB_CACHE_LINE_BYTES);
+  txq->qtx_tail = ad->bar0 + AVF_QTX_TAIL (qid);
+  return 0;
+}
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 flags;
+} virtchnl_promisc_info_t;
+
+void
+avf_arq_slot_init (avf_device_t * ad, u16 slot)
+{
+  avf_aq_desc_t *d;
+  u64 pa = ad->arq_bufs_pa + slot * AVF_MBOX_BUF_SZ;
+  d = &ad->arq[slot];
+  memset (d, 0, sizeof (avf_aq_desc_t));
+  d->flags = AVF_AQ_F_BUF;
+  d->datalen = AVF_MBOX_BUF_SZ;
+  d->addr_hi = (u32) (pa >> 32);
+  d->addr_lo = (u32) pa;
+}
+
+static inline uword
+avf_dma_addr (vlib_main_t * vm, avf_device_t * ad, void *p)
+{
+  avf_main_t *am = &avf_main;
+  return (ad->flags & AVF_DEVICE_F_IOVA) ?
+    pointer_to_uword (p) :
+    vlib_physmem_virtual_to_physical (vm, am->physmem_region, p);
+}
+
+static void
+avf_adminq_init (vlib_main_t * vm, avf_device_t * ad)
+{
+  u64 pa;
+  int i;
+
+  /* VF MailBox Transmit */
+  memset (ad->atq, 0, sizeof (avf_aq_desc_t) * AVF_MBOX_LEN);
+  ad->atq_bufs_pa = avf_dma_addr (vm, ad, ad->atq_bufs);
+
+  pa = avf_dma_addr (vm, ad, ad->atq);
+  avf_reg_write (ad, AVF_ATQT, 0);	/* Tail */
+  avf_reg_write (ad, AVF_ATQH, 0);	/* Head */
+  avf_reg_write (ad, AVF_ATQLEN, AVF_MBOX_LEN | (1 << 31));	/* len & ena */
+  avf_reg_write (ad, AVF_ATQBAL, (u32) pa);	/* Base Address Low */
+  avf_reg_write (ad, AVF_ATQBAH, (u32) (pa >> 32));	/* Base Address High */
+
+  /* VF MailBox Receive */
+  memset (ad->arq, 0, sizeof (avf_aq_desc_t) * AVF_MBOX_LEN);
+  ad->arq_bufs_pa = avf_dma_addr (vm, ad, ad->arq_bufs);
+
+  for (i = 0; i < AVF_MBOX_LEN; i++)
+    avf_arq_slot_init (ad, i);
+
+  pa = avf_dma_addr (vm, ad, ad->arq);
+
+  avf_reg_write (ad, AVF_ARQH, 0);	/* Head */
+  avf_reg_write (ad, AVF_ARQT, 0);	/* Head */
+  avf_reg_write (ad, AVF_ARQLEN, AVF_MBOX_LEN | (1 << 31));	/* len & ena */
+  avf_reg_write (ad, AVF_ARQBAL, (u32) pa);	/* Base Address Low */
+  avf_reg_write (ad, AVF_ARQBAH, (u32) (pa >> 32));	/* Base Address High */
+  avf_reg_write (ad, AVF_ARQT, AVF_MBOX_LEN - 1);	/* Tail */
+
+  ad->atq_next_slot = 0;
+  ad->arq_next_slot = 0;
+}
+
+clib_error_t *
+avf_send_to_pf (vlib_main_t * vm, avf_device_t * ad, virtchnl_ops_t op,
+		void *in, int in_len, void *out, int out_len)
+{
+  clib_error_t *err;
+  avf_aq_desc_t *d, dt = {.opcode = 0x801,.v_opcode = op };
+  u32 head;
+  int n_retry = 5;
+
+
+  /* supppres interrupt in the next adminq receive slot
+     as we are going to wait for response
+     we only need interrupts when event is received */
+  d = &ad->arq[ad->arq_next_slot];
+  d->flags |= AVF_AQ_F_SI;
+
+  if ((err = avf_aq_desc_enq (vm, ad, &dt, in, in_len)))
+    return err;
+
+retry:
+  head = avf_get_u32 (ad->bar0, AVF_ARQH);
+
+  if (ad->arq_next_slot == head)
+    {
+      if (--n_retry == 0)
+	return clib_error_return (0, "timeout");
+      vlib_process_suspend (vm, 10e-3);
+      goto retry;
+    }
+
+  d = &ad->arq[ad->arq_next_slot];
+
+  if (d->v_opcode == VIRTCHNL_OP_EVENT)
+    {
+      void *buf = ad->arq_bufs + ad->arq_next_slot * AVF_MBOX_BUF_SZ;
+      virtchnl_pf_event_t *e;
+
+      if ((d->datalen != sizeof (virtchnl_pf_event_t)) ||
+	  ((d->flags & AVF_AQ_F_BUF) == 0))
+	return clib_error_return (0, "event message error");
+
+      vec_add2 (ad->events, e, 1);
+      clib_memcpy (e, buf, sizeof (virtchnl_pf_event_t));
+      avf_arq_slot_init (ad, ad->arq_next_slot);
+      ad->arq_next_slot++;
+      n_retry = 5;
+      goto retry;
+    }
+
+  if (d->v_opcode != op)
+    {
+      err = clib_error_return (0, "unexpected message receiver [v_opcode = %u"
+			       "expected %u]", d->v_opcode, op);
+      goto done;
+    }
+
+  if (d->v_retval)
+    {
+      err = clib_error_return (0, "error [v_opcode = %u, v_retval %d]",
+			       d->v_opcode, d->v_retval);
+      goto done;
+    }
+
+  if (d->flags & AVF_AQ_F_BUF)
+    {
+      void *buf = ad->arq_bufs + ad->arq_next_slot * AVF_MBOX_BUF_SZ;
+      clib_memcpy (out, buf, out_len);
+    }
+
+  avf_arq_slot_init (ad, ad->arq_next_slot);
+  avf_reg_write (ad, AVF_ARQT, ad->arq_next_slot);
+  avf_reg_flush (ad);
+  ad->arq_next_slot = (ad->arq_next_slot + 1) % AVF_MBOX_LEN;
+
+done:
+
+  if (ad->flags & AVF_DEVICE_F_ELOG)
+    {
+      /* *INDENT-OFF* */
+      ELOG_TYPE_DECLARE (el) =
+	{
+	  .format = "avf[%d] send to pf: v_opcode %s (%d) v_retval 0x%x",
+	  .format_args = "i4t4i4i4",
+	  .n_enum_strings = VIRTCHNL_N_OPS,
+	  .enum_strings = {
+#define _(v, n) [v] = #n,
+	      foreach_virtchnl_op
+#undef _
+	  },
+	};
+      struct
+	{
+	  u32 dev_instance;
+	  u32 v_opcode;
+	  u32 v_opcode_val;
+	  u32 v_retval;
+	} *ed;
+      ed = ELOG_DATA (&vm->elog_main, el);
+      ed->dev_instance = ad->dev_instance;
+      ed->v_opcode = op;
+      ed->v_opcode_val = op;
+      ed->v_retval = d->v_retval;
+      /* *INDENT-ON* */
+    }
+  return err;
+}
+
+clib_error_t *
+avf_op_version (vlib_main_t * vm, avf_device_t * ad,
+		virtchnl_version_info_t * ver)
+{
+  clib_error_t *err = 0;
+  virtchnl_version_info_t myver = {
+    .major = VIRTCHNL_VERSION_MAJOR,
+    .minor = VIRTCHNL_VERSION_MINOR,
+  };
+
+  err = avf_send_to_pf (vm, ad, VIRTCHNL_OP_VERSION, &myver,
+			sizeof (virtchnl_version_info_t), ver,
+			sizeof (virtchnl_version_info_t));
+
+  if (err)
+    return err;
+
+  return err;
+}
+
+clib_error_t *
+avf_op_get_vf_resources (vlib_main_t * vm, avf_device_t * ad,
+			 virtchnl_vf_resource_t * res)
+{
+  clib_error_t *err = 0;
+  u32 bitmap = (VIRTCHNL_VF_OFFLOAD_L2 | VIRTCHNL_VF_OFFLOAD_RSS_AQ |
+		VIRTCHNL_VF_OFFLOAD_RSS_REG | VIRTCHNL_VF_OFFLOAD_WB_ON_ITR |
+		VIRTCHNL_VF_OFFLOAD_VLAN | VIRTCHNL_VF_OFFLOAD_RX_POLLING);
+
+  err = avf_send_to_pf (vm, ad, VIRTCHNL_OP_GET_VF_RESOURCES, &bitmap,
+			sizeof (u32), res, sizeof (virtchnl_vf_resource_t));
+
+  if (err)
+    return err;
+
+  return err;
+}
+
+clib_error_t *
+avf_op_disable_vlan_stripping (vlib_main_t * vm, avf_device_t * ad)
+{
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_DISABLE_VLAN_STRIPPING, 0, 0, 0,
+			 0);
+}
+
+clib_error_t *
+avf_config_promisc_mode (vlib_main_t * vm, avf_device_t * ad)
+{
+  virtchnl_promisc_info_t pi = { 0 };
+
+  pi.vsi_id = ad->vsi_id;
+  pi.flags = 1;
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_CONFIG_PROMISCUOUS_MODE, &pi,
+			 sizeof (virtchnl_promisc_info_t), 0, 0);
+}
+
+
+clib_error_t *
+avf_op_config_vsi_queues (vlib_main_t * vm, avf_device_t * ad)
+{
+  int i;
+  int n_qp = clib_max (vec_len (ad->rxqs), vec_len (ad->txqs));
+  int msg_len = sizeof (virtchnl_vsi_queue_config_info_t) + n_qp *
+    sizeof (virtchnl_queue_pair_info_t);
+  u8 msg[msg_len];
+  virtchnl_vsi_queue_config_info_t *ci;
+
+  memset (msg, 0, msg_len);
+  ci = (virtchnl_vsi_queue_config_info_t *) msg;
+  ci->vsi_id = ad->vsi_id;
+  ci->num_queue_pairs = n_qp;
+
+  for (i = 0; i < n_qp; i++)
+    {
+      virtchnl_txq_info_t *txq = &ci->qpair[i].txq;
+      virtchnl_rxq_info_t *rxq = &ci->qpair[i].rxq;
+
+      rxq->vsi_id = ad->vsi_id;
+      rxq->queue_id = i;
+      rxq->max_pkt_size = 1518;
+      if (i < vec_len (ad->rxqs))
+	{
+	  avf_rxq_t *q = vec_elt_at_index (ad->rxqs, i);
+	  rxq->ring_len = q->size;
+	  rxq->databuffer_size = VLIB_BUFFER_DEFAULT_FREE_LIST_BYTES;
+	  rxq->dma_ring_addr = avf_dma_addr (vm, ad, q->descs);
+	  avf_reg_write (ad, AVF_QRX_TAIL (i), q->size - 1);
+	}
+
+      avf_txq_t *q = vec_elt_at_index (ad->txqs, i);
+      txq->vsi_id = ad->vsi_id;
+      if (i < vec_len (ad->txqs))
+	{
+	  txq->queue_id = i;
+	  txq->ring_len = q->size;
+	  txq->dma_ring_addr = avf_dma_addr (vm, ad, q->descs);
+	}
+    }
+
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_CONFIG_VSI_QUEUES, msg, msg_len,
+			 0, 0);
+}
+
+clib_error_t *
+avf_op_config_irq_map (vlib_main_t * vm, avf_device_t * ad)
+{
+  int count = 1;
+  int msg_len = sizeof (virtchnl_irq_map_info_t) +
+    count * sizeof (virtchnl_vector_map_t);
+  u8 msg[msg_len];
+  virtchnl_irq_map_info_t *imi;
+
+  memset (msg, 0, msg_len);
+  imi = (virtchnl_irq_map_info_t *) msg;
+  imi->num_vectors = count;
+
+  imi->vecmap[0].vector_id = 1;
+  imi->vecmap[0].vsi_id = ad->vsi_id;
+  imi->vecmap[0].rxq_map = 1;
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_CONFIG_IRQ_MAP, msg, msg_len, 0,
+			 0);
+}
+
+clib_error_t *
+avf_op_add_eth_addr (vlib_main_t * vm, avf_device_t * ad, u8 count, u8 * macs)
+{
+  int msg_len =
+    sizeof (virtchnl_ether_addr_list_t) +
+    count * sizeof (virtchnl_ether_addr_t);
+  u8 msg[msg_len];
+  virtchnl_ether_addr_list_t *al;
+  int i;
+
+  memset (msg, 0, msg_len);
+  al = (virtchnl_ether_addr_list_t *) msg;
+  al->vsi_id = ad->vsi_id;
+  al->num_elements = count;
+  for (i = 0; i < count; i++)
+    clib_memcpy (&al->list[i].addr, macs + i * 6, 6);
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_ADD_ETH_ADDR, msg, msg_len, 0,
+			 0);
+}
+
+clib_error_t *
+avf_op_enable_queues (vlib_main_t * vm, avf_device_t * ad, u32 rx, u32 tx)
+{
+  virtchnl_queue_select_t qs = { 0 };
+  qs.vsi_id = ad->vsi_id;
+  qs.rx_queues = rx;
+  qs.tx_queues = tx;
+  avf_rxq_t *rxq = vec_elt_at_index (ad->rxqs, 0);
+  avf_reg_write (ad, AVF_QRX_TAIL (0), rxq->n_bufs);
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_ENABLE_QUEUES, &qs,
+			 sizeof (virtchnl_queue_select_t), 0, 0);
+}
+
+clib_error_t *
+avf_op_get_stats (vlib_main_t * vm, avf_device_t * ad,
+		  virtchnl_eth_stats_t * es)
+{
+  virtchnl_queue_select_t qs = { 0 };
+  qs.vsi_id = ad->vsi_id;
+  return avf_send_to_pf (vm, ad, VIRTCHNL_OP_GET_STATS,
+			 &qs, sizeof (virtchnl_queue_select_t),
+			 es, sizeof (virtchnl_eth_stats_t));
+}
+
+clib_error_t *
+avf_device_reset (vlib_main_t * vm, avf_device_t * ad)
+{
+  avf_aq_desc_t d = { 0 };
+  clib_error_t *error;
+  u32 rstat;
+  int n_retry = 20;
+
+  d.opcode = 0x801;
+  d.v_opcode = VIRTCHNL_OP_RESET_VF;
+  if ((error = avf_aq_desc_enq (vm, ad, &d, 0, 0)))
+    return error;
+
+retry:
+  vlib_process_suspend (vm, 10e-3);
+  rstat = avf_get_u32 (ad->bar0, AVFGEN_RSTAT);
+
+  if (rstat == 2 || rstat == 3)
+    return 0;
+
+  if (--n_retry == 0)
+    return clib_error_return (0, "reset failed (timeout)");
+
+  goto retry;
+}
+
+clib_error_t *
+avf_device_init (vlib_main_t * vm, avf_device_t * ad)
+{
+  virtchnl_version_info_t ver = { 0 };
+  virtchnl_vf_resource_t res = { 0 };
+  clib_error_t *error;
+  vlib_thread_main_t *tm = vlib_get_thread_main ();
+  int i;
+
+  avf_adminq_init (vm, ad);
+
+  if ((error = avf_device_reset (vm, ad)))
+    return error;
+
+  avf_adminq_init (vm, ad);
+
+  /*
+   * OP_VERSION
+   */
+  if ((error = avf_op_version (vm, ad, &ver)))
+    return error;
+
+  if (ver.major != VIRTCHNL_VERSION_MAJOR ||
+      ver.minor != VIRTCHNL_VERSION_MINOR)
+    return clib_error_return (0, "incompatible protocol version "
+			      "(remote %d.%d)", ver.major, ver.minor);
+
+  /*
+   * OP_GET_VF_RESOUCES
+   */
+  if ((error = avf_op_get_vf_resources (vm, ad, &res)))
+    return error;
+
+  if (res.num_vsis != 1 || res.vsi_res[0].vsi_type != VIRTCHNL_VSI_SRIOV)
+    return clib_error_return (0, "unexpected GET_VF_RESOURCE reply received");
+
+  ad->vsi_id = res.vsi_res[0].vsi_id;
+  ad->feature_bitmap = res.vf_offload_flags;
+  ad->num_queue_pairs = res.num_queue_pairs;
+  ad->max_vectors = res.max_vectors;
+  ad->max_mtu = res.max_mtu;
+  ad->rss_key_size = res.rss_key_size;
+  ad->rss_lut_size = res.rss_lut_size;
+
+  clib_memcpy (ad->hwaddr, res.vsi_res[0].default_mac_addr, 6);
+
+  /*
+   * Disable VLAN stripping
+   */
+  if ((error = avf_op_disable_vlan_stripping (vm, ad)))
+    return error;
+
+  if ((error = avf_config_promisc_mode (vm, ad)))
+    return error;
+
+  if ((error = avf_cmd_rx_ctl_reg_write (vm, ad, 0xc400, 0)))
+    return error;
+
+  if ((error = avf_cmd_rx_ctl_reg_write (vm, ad, 0xc404, 0)))
+    return error;
+
+  /*
+   * Init Queues
+   */
+  if ((error = avf_rxq_init (vm, ad, 0)))
+    return error;
+
+  for (i = 0; i < tm->n_vlib_mains; i++)
+    if ((error = avf_txq_init (vm, ad, i)))
+      return error;
+
+  if ((error = avf_op_config_vsi_queues (vm, ad)))
+    return error;
+
+  if ((error = avf_op_config_irq_map (vm, ad)))
+    return error;
+
+  avf_irq_0_enable (ad);
+  avf_irq_n_enable (ad, 0);
+
+  if ((error = avf_op_add_eth_addr (vm, ad, 1, ad->hwaddr)))
+    return error;
+
+  if ((error = avf_op_enable_queues (vm, ad, 1, 0)))
+    return error;
+
+  if ((error = avf_op_enable_queues (vm, ad, 0, 1)))
+    return error;
+
+  ad->flags |= AVF_DEVICE_F_INITIALIZED;
+  return error;
+}
+
+void
+avf_process_one_device (vlib_main_t * vm, avf_device_t * ad, int is_irq)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  virtchnl_pf_event_t *e;
+  u32 r;
+
+  if (ad->flags & AVF_DEVICE_F_ERROR)
+    return;
+
+  if ((ad->flags & AVF_DEVICE_F_INITIALIZED) == 0)
+    return;
+
+  ASSERT (ad->error == 0);
+
+  r = avf_get_u32 (ad->bar0, AVF_ARQLEN);
+  if ((r & 0xf0000000) != (1 << 31))
+    {
+      ad->error = clib_error_return (0, "arq not enabled, arqlen = 0x%x", r);
+      goto error;
+    }
+
+  r = avf_get_u32 (ad->bar0, AVF_ATQLEN);
+  if ((r & 0xf0000000) != (1 << 31))
+    {
+      ad->error = clib_error_return (0, "atq not enabled, atqlen = 0x%x", r);
+      goto error;
+    }
+
+  if (is_irq == 0)
+    avf_op_get_stats (vm, ad, &ad->eth_stats);
+
+  /* *INDENT-OFF* */
+  vec_foreach (e, ad->events)
+    {
+      if (e->event == VIRTCHNL_EVENT_LINK_CHANGE)
+	{
+	  int link_up = e->event_data.link_event.link_status;
+	  virtchnl_link_speed_t speed = e->event_data.link_event.link_speed;
+	  u32 flags = 0;
+
+	  if (link_up && (ad->flags & AVF_DEVICE_F_LINK_UP) == 0)
+	    {
+	      ad->flags |= AVF_DEVICE_F_LINK_UP;
+	      flags |= (VNET_HW_INTERFACE_FLAG_FULL_DUPLEX |
+			VNET_HW_INTERFACE_FLAG_LINK_UP);
+	      if (speed == VIRTCHNL_LINK_SPEED_40GB)
+		flags |= VNET_HW_INTERFACE_FLAG_SPEED_40G;
+	      else if (speed == VIRTCHNL_LINK_SPEED_25GB)
+		flags |= VNET_HW_INTERFACE_FLAG_SPEED_25G;
+	      else if (speed == VIRTCHNL_LINK_SPEED_10GB)
+		flags |= VNET_HW_INTERFACE_FLAG_SPEED_10G;
+	      else if (speed == VIRTCHNL_LINK_SPEED_1GB)
+		flags |= VNET_HW_INTERFACE_FLAG_SPEED_1G;
+	      else if (speed == VIRTCHNL_LINK_SPEED_100MB)
+		flags |= VNET_HW_INTERFACE_FLAG_SPEED_100M;
+	      vnet_hw_interface_set_flags (vnm, ad->hw_if_index, flags);
+	      ad->link_speed = speed;
+	    }
+	  else if (!link_up && (ad->flags & AVF_DEVICE_F_LINK_UP) != 0)
+	    {
+	      ad->flags &= ~AVF_DEVICE_F_LINK_UP;
+	      ad->link_speed = 0;
+	    }
+
+	  if (ad->flags & AVF_DEVICE_F_ELOG)
+	    {
+	      ELOG_TYPE_DECLARE (el) =
+		{
+		  .format = "avf[%d] link change: link_status %d "
+		    "link_speed %d",
+		  .format_args = "i4i1i1",
+		};
+	      struct
+		{
+		  u32 dev_instance;
+		  u8 link_status;
+		  u8 link_speed;
+		} *ed;
+	      ed = ELOG_DATA (&vm->elog_main, el);
+              ed->dev_instance = ad->dev_instance;
+	      ed->link_status = link_up;
+	      ed->link_speed = speed;
+	    }
+	}
+      else
+	{
+	  if (ad->flags & AVF_DEVICE_F_ELOG)
+	    {
+	      ELOG_TYPE_DECLARE (el) =
+		{
+		  .format = "avf[%d] unknown event: event %d severity %d",
+		  .format_args = "i4i4i1i1",
+		};
+	      struct
+		{
+		  u32 dev_instance;
+		  u32 event;
+		  u32 severity;
+		} *ed;
+	      ed = ELOG_DATA (&vm->elog_main, el);
+              ed->dev_instance = ad->dev_instance;
+	      ed->event = e->event;
+	      ed->severity = e->severity;
+	    }
+	}
+    }
+  /* *INDENT-ON* */
+  vec_reset_length (ad->events);
+
+  return;
+
+error:
+  ad->flags |= AVF_DEVICE_F_ERROR;
+  ASSERT (ad->error != 0);
+}
+
+static u32
+avf_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hw, u32 flags)
+{
+  clib_warning ("TODO");
+  return 0;
+}
+
+static uword
+avf_process (vlib_main_t * vm, vlib_node_runtime_t * rt, vlib_frame_t * f)
+{
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad;
+  uword *event_data = 0, event_type;
+  int enabled = 0, irq;
+  f64 last_run_duration = 0;
+  f64 last_periodic_time = 0;
+
+  while (1)
+    {
+      if (enabled)
+	vlib_process_wait_for_event_or_clock (vm, 5.0 - last_run_duration);
+      else
+	vlib_process_wait_for_event (vm);
+
+      event_type = vlib_process_get_events (vm, &event_data);
+      vec_reset_length (event_data);
+      irq = 0;
+
+      switch (event_type)
+	{
+	case ~0:
+	  last_periodic_time = vlib_time_now (vm);
+	  break;
+	case AVF_PROCESS_EVENT_START:
+	  enabled = 1;
+	  break;
+	case AVF_PROCESS_EVENT_STOP:
+	  enabled = 0;
+	  continue;
+	case AVF_PROCESS_EVENT_AQ_INT:
+	  irq = 1;
+	  break;
+	default:
+	  ASSERT (0);
+	}
+
+      /* *INDENT-OFF* */
+      pool_foreach (ad, am->devices,
+        {
+	  avf_process_one_device (vm, ad, irq);
+        });
+      /* *INDENT-ON* */
+      last_run_duration = vlib_time_now (vm) - last_periodic_time;
+    }
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (avf_process_node, static)  = {
+  .function = avf_process,
+  .type = VLIB_NODE_TYPE_PROCESS,
+  .name = "avf-process",
+};
+/* *INDENT-ON* */
+
+static void
+avf_irq_0_handler (vlib_pci_dev_handle_t h, u16 line)
+{
+  vlib_main_t *vm = vlib_get_main ();
+  avf_main_t *am = &avf_main;
+  uword pd = vlib_pci_get_private_data (h);
+  avf_device_t *ad = pool_elt_at_index (am->devices, pd);
+  u32 icr0;
+
+  icr0 = avf_reg_read (ad, AVFINT_ICR0);
+
+  if (ad->flags & AVF_DEVICE_F_ELOG)
+    {
+      /* *INDENT-OFF* */
+      ELOG_TYPE_DECLARE (el) =
+	{
+	  .format = "avf[%d] irq 0: icr0 0x%x",
+	  .format_args = "i4i4",
+	};
+      /* *INDENT-ON* */
+      struct
+      {
+	u32 dev_instance;
+	u32 icr0;
+      } *ed;
+
+      ed = ELOG_DATA (&vm->elog_main, el);
+      ed->dev_instance = ad->dev_instance;
+      ed->icr0 = icr0;
+    }
+
+  avf_irq_0_enable (ad);
+
+  /* bit 30 - Send/Receive Admin queue interrupt indication */
+  if (icr0 & (1 << 30))
+    vlib_process_signal_event (vm, avf_process_node.index,
+			       AVF_PROCESS_EVENT_AQ_INT, 0);
+}
+
+static void
+avf_irq_n_handler (vlib_pci_dev_handle_t h, u16 line)
+{
+  vlib_main_t *vm = vlib_get_main ();
+  avf_main_t *am = &avf_main;
+  uword pd = vlib_pci_get_private_data (h);
+  avf_device_t *ad = pool_elt_at_index (am->devices, pd);
+
+  if (ad->flags & AVF_DEVICE_F_ELOG)
+    {
+      /* *INDENT-OFF* */
+      ELOG_TYPE_DECLARE (el) =
+	{
+	  .format = "avf[%d] irq %d: received",
+	  .format_args = "i4i2",
+	};
+      /* *INDENT-ON* */
+      struct
+      {
+	u32 dev_instance;
+	u16 line;
+      } *ed;
+
+      ed = ELOG_DATA (&vm->elog_main, el);
+      ed->dev_instance = ad->dev_instance;
+      ed->line = line;
+    }
+
+  avf_irq_n_enable (ad, 0);
+}
+
+void
+avf_delete_if (vlib_main_t * vm, avf_device_t * ad)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  avf_main_t *am = &avf_main;
+  int i;
+
+  if (ad->hw_if_index)
+    {
+      vnet_hw_interface_set_flags (vnm, ad->hw_if_index, 0);
+      vnet_hw_interface_unassign_rx_thread (vnm, ad->hw_if_index, 0);
+      ethernet_delete_interface (vnm, ad->hw_if_index);
+    }
+
+  vlib_pci_device_close (ad->pci_dev_handle);
+
+  vlib_physmem_free (vm, am->physmem_region, ad->atq);
+  vlib_physmem_free (vm, am->physmem_region, ad->arq);
+  vlib_physmem_free (vm, am->physmem_region, ad->atq_bufs);
+  vlib_physmem_free (vm, am->physmem_region, ad->arq_bufs);
+
+  /* *INDENT-OFF* */
+  vec_foreach_index (i, ad->rxqs)
+    {
+      avf_rxq_t *rxq = vec_elt_at_index (ad->rxqs, i);
+      vlib_physmem_free (vm, am->physmem_region, rxq->descs);
+      if (rxq->n_bufs)
+	vlib_buffer_free_from_ring (vm, rxq->bufs, rxq->next, rxq->size,
+				    rxq->n_bufs);
+      vec_free (rxq->bufs);
+    }
+  /* *INDENT-ON* */
+  vec_free (ad->rxqs);
+
+  /* *INDENT-OFF* */
+  vec_foreach_index (i, ad->txqs)
+    {
+      avf_txq_t *txq = vec_elt_at_index (ad->txqs, i);
+      vlib_physmem_free (vm, am->physmem_region, txq->descs);
+      if (txq->n_bufs)
+	{
+	  u16 first = (txq->next - txq->n_bufs) & (txq->size -1);
+	  vlib_buffer_free_from_ring (vm, txq->bufs, first, txq->size,
+				      txq->n_bufs);
+	}
+      vec_free (txq->bufs);
+    }
+  /* *INDENT-ON* */
+  vec_free (ad->txqs);
+
+  clib_error_free (ad->error);
+  memset (ad, 0, sizeof (*ad));
+  pool_put (am->devices, ad);
+}
+
+void
+avf_create_if (vlib_main_t * vm, avf_create_if_args_t * args)
+{
+  vnet_main_t *vnm = vnet_get_main ();
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad;
+  vlib_pci_dev_handle_t h;
+  clib_error_t *error = 0;
+
+  pool_get (am->devices, ad);
+  ad->dev_instance = ad - am->devices;
+  ad->per_interface_next_index = ~0;
+
+  if (args->enable_elog)
+    ad->flags |= AVF_DEVICE_F_ELOG;
+
+  if ((error = vlib_pci_device_open (&args->addr, avf_pci_device_ids, &h)))
+    goto error;
+  ad->pci_dev_handle = h;
+
+  vlib_pci_set_private_data (h, ad->dev_instance);
+
+  if ((error = vlib_pci_bus_master_enable (h)))
+    goto error;
+
+  if ((error = vlib_pci_map_region (h, 0, &ad->bar0)))
+    goto error;
+
+  if ((error = vlib_pci_register_msix_handler (h, 0, 1, &avf_irq_0_handler)))
+    goto error;
+
+  if ((error = vlib_pci_register_msix_handler (h, 1, 1, &avf_irq_n_handler)))
+    goto error;
+
+  if ((error = vlib_pci_enable_msix_irq (h, 0, 2)))
+    goto error;
+
+  if (am->physmem_region_alloc == 0)
+    {
+      u32 flags = VLIB_PHYSMEM_F_INIT_MHEAP | VLIB_PHYSMEM_F_HUGETLB;
+      error = vlib_physmem_region_alloc (vm, "avf descriptors", 4 << 20, 0,
+					 flags, &am->physmem_region);
+      if (error)
+	goto error;
+      am->physmem_region_alloc = 1;
+    }
+  ad->atq = vlib_physmem_alloc_aligned (vm, am->physmem_region, &error,
+					sizeof (avf_aq_desc_t) * AVF_MBOX_LEN,
+					64);
+  if (error)
+    goto error;
+
+  ad->arq = vlib_physmem_alloc_aligned (vm, am->physmem_region, &error,
+					sizeof (avf_aq_desc_t) * AVF_MBOX_LEN,
+					64);
+  if (error)
+    goto error;
+
+  ad->atq_bufs = vlib_physmem_alloc_aligned (vm, am->physmem_region, &error,
+					     AVF_MBOX_BUF_SZ * AVF_MBOX_LEN,
+					     64);
+  if (error)
+    goto error;
+
+  ad->arq_bufs = vlib_physmem_alloc_aligned (vm, am->physmem_region, &error,
+					     AVF_MBOX_BUF_SZ * AVF_MBOX_LEN,
+					     64);
+  if (error)
+    goto error;
+
+  if ((error = vlib_pci_intr_enable (h)))
+    goto error;
+
+  /* FIXME detect */
+  ad->flags |= AVF_DEVICE_F_IOVA;
+
+  if ((error = avf_device_init (vm, ad)))
+    goto error;
+
+  /* create interface */
+  error = ethernet_register_interface (vnm, avf_device_class.index,
+				       ad->dev_instance, ad->hwaddr,
+				       &ad->hw_if_index, avf_flag_change);
+
+  if (error)
+    goto error;
+
+  vnet_sw_interface_t *sw = vnet_get_hw_sw_interface (vnm, ad->hw_if_index);
+  ad->sw_if_index = sw->sw_if_index;
+
+  vnet_hw_interface_set_input_node (vnm, ad->hw_if_index,
+				    avf_input_node.index);
+
+  if (pool_elts (am->devices) == 1)
+    vlib_process_signal_event (vm, avf_process_node.index,
+			       AVF_PROCESS_EVENT_START, 0);
+
+  return;
+
+error:
+  avf_delete_if (vm, ad);
+  args->rv = VNET_API_ERROR_INVALID_INTERFACE;
+  args->error = clib_error_return (error, "pci-addr %U",
+				   format_vlib_pci_addr, &args->addr);
+}
+
+static clib_error_t *
+avf_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, u32 flags)
+{
+  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, hw_if_index);
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad = vec_elt_at_index (am->devices, hi->dev_instance);
+  uword is_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0;
+
+  if (ad->flags & AVF_DEVICE_F_ERROR)
+    return clib_error_return (0, "device is in error state");
+
+  if (is_up)
+    {
+      vnet_hw_interface_set_flags (vnm, ad->hw_if_index,
+				   VNET_HW_INTERFACE_FLAG_LINK_UP);
+      ad->flags |= AVF_DEVICE_F_ADMIN_UP;
+      vnet_hw_interface_assign_rx_thread (vnm, ad->hw_if_index, 0, ~0);
+    }
+  else
+    {
+      vnet_hw_interface_set_flags (vnm, ad->hw_if_index, 0);
+      ad->flags &= ~AVF_DEVICE_F_ADMIN_UP;
+    }
+  return 0;
+}
+
+/* *INDENT-OFF* */
+VNET_DEVICE_CLASS (avf_device_class,) =
+{
+  .name = "Adaptive Virtual Function (AVF) interface",
+  .tx_function = avf_interface_tx,
+  .format_device = format_avf_device,
+  .format_device_name = format_avf_device_name,
+  .admin_up_down_function = avf_interface_admin_up_down,
+};
+/* *INDENT-ON* */
+
+clib_error_t *
+avf_init (vlib_main_t * vm)
+{
+  avf_main_t *am = &avf_main;
+  clib_error_t *error;
+  vlib_thread_main_t *tm = vlib_get_thread_main ();
+  int i;
+
+  if ((error = vlib_call_init_function (vm, pci_bus_init)))
+    return error;
+
+  vec_validate_aligned (am->per_thread_data, tm->n_vlib_mains - 1,
+			CLIB_CACHE_LINE_BYTES);
+
+  /* initialize ptype based loopup table */
+  vec_validate_aligned (am->ptypes, 255, CLIB_CACHE_LINE_BYTES);
+
+  /* *INDENT-OFF* */
+  vec_foreach_index (i, am->ptypes)
+    {
+      avf_ptype_t *p = vec_elt_at_index (am->ptypes, i);
+      if ((i >= 22) && (i <= 87))
+	{
+	  p->next_node = VNET_DEVICE_INPUT_NEXT_IP4_NCS_INPUT;
+	  p->flags = VNET_BUFFER_F_IS_IP4;
+	}
+      else if ((i >= 88) && (i <= 153))
+	{
+	  p->next_node = VNET_DEVICE_INPUT_NEXT_IP6_INPUT;
+	  p->flags = VNET_BUFFER_F_IS_IP6;
+	}
+      else
+	p->next_node = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+      p->buffer_advance = device_input_next_node_advance[p->next_node];
+      p->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
+    }
+  /* *INDENT-ON* */
+
+  return 0;
+}
+
+VLIB_INIT_FUNCTION (avf_init);
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/format.c b/src/plugins/avf/format.c
new file mode 100644
index 00000000000..32b9d3f14a6
--- /dev/null
+++ b/src/plugins/avf/format.c
@@ -0,0 +1,145 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vlib/pci/pci.h>
+#include <vnet/ethernet/ethernet.h>
+
+#include <avf/avf.h>
+
+u8 *
+format_avf_device_name (u8 * s, va_list * args)
+{
+  u32 i = va_arg (*args, u32);
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad = vec_elt_at_index (am->devices, i);
+  vlib_pci_addr_t *addr = vlib_pci_get_addr (ad->pci_dev_handle);
+
+  s = format (s, "AVF%x/%x/%x/%x",
+	      addr->domain, addr->bus, addr->slot, addr->function);
+  return s;
+}
+
+u8 *
+format_avf_device_flags (u8 * s, va_list * args)
+{
+  avf_device_t *ad = va_arg (*args, avf_device_t *);
+  u8 *t = 0;
+
+#define _(a, b, c) if (ad->flags & (1 << a)) \
+t = format (t, "%s%s", t ? " ":"", c);
+  foreach_avf_device_flags
+#undef _
+    s = format (s, "%v", t);
+  vec_free (t);
+  return s;
+}
+
+u8 *
+format_avf_vf_cap_flags (u8 * s, va_list * args)
+{
+  u32 flags = va_arg (*args, u32);
+  u8 *t = 0;
+
+#define _(a, b, c) if (flags & (1 << a)) \
+  t = format (t, "%s%s", t ? " ":"", c);
+  foreach_avf_vf_cap_flag;
+#undef _
+  s = format (s, "%v", t);
+  vec_free (t);
+  return s;
+}
+
+static u8 *
+format_virtchnl_link_speed (u8 * s, va_list * args)
+{
+  virtchnl_link_speed_t speed = va_arg (*args, virtchnl_link_speed_t);
+
+  if (speed == 0)
+    return format (s, "unknown");
+#define _(a, b, c) \
+  else if (speed == VIRTCHNL_LINK_SPEED_##b) \
+    return format (s, c);
+  foreach_virtchnl_link_speed;
+#undef _
+  return s;
+}
+
+u8 *
+format_avf_device (u8 * s, va_list * args)
+{
+  u32 i = va_arg (*args, u32);
+  avf_main_t *am = &avf_main;
+  avf_device_t *ad = vec_elt_at_index (am->devices, i);
+  u32 indent = format_get_indent (s);
+  u8 *a = 0;
+
+  s = format (s, "flags: %U", format_avf_device_flags, ad);
+  s = format (s, "\n%Uoffload features: %U", format_white_space, indent,
+	      format_avf_vf_cap_flags, ad->feature_bitmap);
+
+  s = format (s, "\n%Unum-queue-pairs %d max-vectors %u max-mtu %u "
+	      "rss-key-size %u rss-lut-size %u", format_white_space, indent,
+	      ad->num_queue_pairs, ad->max_vectors, ad->max_mtu,
+	      ad->rss_key_size, ad->rss_lut_size);
+  s = format (s, "\n%Uspeed %U", format_white_space, indent,
+	      format_virtchnl_link_speed, ad->link_speed);
+  if (ad->error)
+    s = format (s, "\n%Uerror %U", format_white_space, indent,
+		format_clib_error, ad->error);
+
+#define _(c) if (ad->eth_stats.c) \
+  a = format (a, "\n%U%-20U %u", format_white_space, indent + 2, \
+	      format_c_identifier, #c, ad->eth_stats.c);
+  foreach_virtchnl_eth_stats;
+#undef _
+  if (a)
+    s = format (s, "\n%Ustats:%v", format_white_space, indent, a);
+
+  vec_free (a);
+  return s;
+}
+
+u8 *
+format_avf_input_trace (u8 * s, va_list * args)
+{
+  vlib_main_t *vm = va_arg (*args, vlib_main_t *);
+  vlib_node_t *node = va_arg (*args, vlib_node_t *);
+  avf_input_trace_t *t = va_arg (*args, avf_input_trace_t *);
+  vnet_main_t *vnm = vnet_get_main ();
+  vnet_hw_interface_t *hi = vnet_get_hw_interface (vnm, t->hw_if_index);
+  u32 indent = format_get_indent (s);
+  avf_rx_vector_entry_t *rxve = &t->rxve;
+
+  s = format (s, "avf: %v (%d) next-node %U",
+	      hi->name, t->hw_if_index, format_vlib_next_node_name, vm,
+	      node->index, t->next_index);
+  s = format (s, "\n%Ustatus 0x%x error 0x%x ptype 0x%x length %u",
+	      format_white_space, indent + 2, rxve->status, rxve->error,
+	      rxve->ptype, rxve->length);
+
+  return s;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/input.c b/src/plugins/avf/input.c
new file mode 100644
index 00000000000..a769bf79646
--- /dev/null
+++ b/src/plugins/avf/input.c
@@ -0,0 +1,439 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vlib/pci/pci.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/devices/devices.h>
+
+#include <avf/avf.h>
+
+#define foreach_avf_input_error \
+  _(BUFFER_ALLOC, "buffer alloc error") \
+  _(RX_PACKET_ERROR, "Rx packet errors")
+
+typedef enum
+{
+#define _(f,s) AVF_INPUT_ERROR_##f,
+  foreach_avf_input_error
+#undef _
+    AVF_INPUT_N_ERROR,
+} avf_input_error_t;
+
+static __clib_unused char *avf_input_error_strings[] = {
+#define _(n,s) s,
+  foreach_avf_input_error
+#undef _
+};
+
+#define AVF_RX_DESC_STATUS(x)		(1 << x)
+#define AVF_RX_DESC_STATUS_DD		AVF_RX_DESC_STATUS(0)
+#define AVF_RX_DESC_STATUS_EOP		AVF_RX_DESC_STATUS(1)
+
+static_always_inline void
+avf_input_trace (vlib_main_t * vm, vlib_node_runtime_t * node, u32 next0,
+		 vlib_buffer_t * b0, uword * n_trace, avf_device_t * ad,
+		 avf_rx_vector_entry_t * rxve)
+{
+  avf_input_trace_t *tr;
+  vlib_trace_buffer (vm, node, next0, b0, /* follow_chain */ 0);
+  vlib_set_trace_count (vm, node, --(*n_trace));
+  tr = vlib_add_trace (vm, node, b0, sizeof (*tr));
+  tr->next_index = next0;
+  tr->hw_if_index = ad->hw_if_index;
+  clib_memcpy (&tr->rxve, rxve, sizeof (avf_rx_vector_entry_t));
+}
+
+#define AVF_INPUT_REFILL_TRESHOLD 32
+static_always_inline void
+avf_rxq_refill (vlib_main_t * vm, vlib_node_runtime_t * node, avf_rxq_t * rxq,
+		int use_iova)
+{
+  u16 n_refill, mask, n_alloc, slot;
+  avf_rx_desc_t *d;
+
+  n_refill = rxq->size - 1 - rxq->n_bufs;
+  if (PREDICT_TRUE (n_refill <= AVF_INPUT_REFILL_TRESHOLD))
+    return;
+
+  mask = rxq->size - 1;
+  slot = (rxq->next - n_refill - 1) & mask;
+
+  n_refill &= ~7;		/* round to 8 */
+  n_alloc = vlib_buffer_alloc_to_ring (vm, rxq->bufs, slot, rxq->size,
+				       n_refill);
+
+  if (PREDICT_FALSE (n_alloc != n_refill))
+    {
+      vlib_error_count (vm, node->node_index,
+			AVF_INPUT_ERROR_BUFFER_ALLOC, 1);
+      if (n_alloc)
+	vlib_buffer_free (vm, rxq->bufs + slot, n_alloc);
+      return;
+    }
+
+  rxq->n_bufs += n_alloc;
+
+  while (n_alloc--)
+    {
+      u64 addr;
+      d = ((avf_rx_desc_t *) rxq->descs) + slot;
+      if (use_iova)
+	{
+	  vlib_buffer_t *b = vlib_get_buffer (vm, rxq->bufs[slot]);
+	  addr = pointer_to_uword (b->data);
+	}
+      else
+	addr = vlib_get_buffer_data_physical_address (vm, rxq->bufs[slot]);
+      d->qword[0] = addr;
+      d->qword[1] = 0;
+      slot = (slot + 1) & mask;
+    }
+
+  CLIB_MEMORY_BARRIER ();
+  *(rxq->qrx_tail) = slot;
+}
+
+static_always_inline void
+avf_check_for_error (vlib_node_runtime_t * node, avf_rx_vector_entry_t * rxve,
+		     vlib_buffer_t * b, u32 * next)
+{
+  avf_main_t *am = &avf_main;
+  avf_ptype_t *ptype;
+  if (PREDICT_FALSE (rxve->error))
+    {
+      b->error = node->errors[AVF_INPUT_ERROR_RX_PACKET_ERROR];
+      ptype = am->ptypes + rxve->ptype;
+      /* retract */
+      vlib_buffer_advance (b, --ptype->buffer_advance);
+      *next = VNET_DEVICE_INPUT_NEXT_DROP;
+    }
+}
+
+static_always_inline u32
+avf_find_next (avf_rx_vector_entry_t * rxve, vlib_buffer_t * b,
+	       int maybe_tagged)
+{
+  avf_main_t *am = &avf_main;
+  ethernet_header_t *e = (ethernet_header_t *) b->data;
+  avf_ptype_t *ptype;
+  if (maybe_tagged && ethernet_frame_is_tagged (e->type))
+    return VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+  ptype = am->ptypes + rxve->ptype;
+  vlib_buffer_advance (b, ptype->buffer_advance);
+  b->flags |= ptype->flags;
+  return ptype->next_node;
+}
+
+
+static_always_inline uword
+avf_device_input_inline (vlib_main_t * vm, vlib_node_runtime_t * node,
+			 vlib_frame_t * frame, avf_device_t * ad, u16 qid,
+			 int with_features_or_tracing)
+{
+  avf_main_t *am = &avf_main;
+  vnet_main_t *vnm = vnet_get_main ();
+  u32 thr_idx = vlib_get_thread_index ();
+  avf_per_thread_data_t *ptd =
+    vec_elt_at_index (am->per_thread_data, thr_idx);
+  avf_rxq_t *rxq = vec_elt_at_index (ad->rxqs, qid);
+  avf_rx_vector_entry_t *rxve;
+  uword n_trace = vlib_get_trace_count (vm, node);
+  u32 next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
+  avf_rx_desc_t *d;
+  u32 *to_next = 0;
+  u32 n_rx_packets = 0;
+  u32 n_rx_bytes = 0;
+  u32 sw_if_idx[VLIB_N_RX_TX] = {[VLIB_RX] = ad->sw_if_index,[VLIB_TX] = ~0 };
+  u16 mask = rxq->size - 1;
+  u16 n_rxv = 0;
+  u8 maybe_error = 0;
+
+  /* fetch up to AVF_RX_VECTOR_SZ from the rx ring, unflatten them and
+     copy needed data from descriptor to rx vector */
+  d = rxq->descs + rxq->next;
+  while ((d->qword[1] & AVF_RX_DESC_STATUS_DD) && n_rxv < AVF_RX_VECTOR_SZ)
+    {
+      u16 next_pf = (rxq->next + 8) & mask;
+      CLIB_PREFETCH (rxq->descs + next_pf, CLIB_CACHE_LINE_BYTES, LOAD);
+      rxve = ptd->rx_vector + n_rxv;
+      rxve->bi = rxq->bufs[rxq->next];
+      rxve->status = avf_get_u64_bits (d, 8, 18, 0);
+      rxve->error = avf_get_u64_bits (d, 8, 26, 19);
+      rxve->ptype = avf_get_u64_bits (d, 8, 37, 30);
+      rxve->length = avf_get_u64_bits (d, 8, 63, 38);
+      maybe_error |= rxve->error;
+
+      /* deal with chained buffers */
+      while (PREDICT_FALSE ((d->qword[1] & AVF_RX_DESC_STATUS_EOP) == 0))
+	{
+	  clib_error ("fixme");
+	}
+
+      /* next */
+      rxq->next = (rxq->next + 1) & mask;
+      d = rxq->descs + rxq->next;
+      n_rxv++;
+      rxq->n_bufs--;
+    }
+
+  if (n_rxv == 0)
+    return 0;
+
+  /* refill rx ring */
+  if (ad->flags & AVF_DEVICE_F_IOVA)
+    avf_rxq_refill (vm, node, rxq, 1 /* use_iova */ );
+  else
+    avf_rxq_refill (vm, node, rxq, 0 /* use_iova */ );
+
+  n_rx_packets = n_rxv;
+  rxve = ptd->rx_vector;
+  while (n_rxv)
+    {
+      u32 n_left_to_next;
+      u32 bi0, bi1, bi2, bi3;
+      vlib_buffer_t *b0, *b1, *b2, *b3;
+      u32 next0, next1, next2, next3;
+
+      vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next);
+
+      while (n_rxv >= 12 && n_left_to_next >= 4)
+	{
+	  vlib_buffer_t *p;
+	  p = vlib_get_buffer (vm, rxve[8].bi);
+	  vlib_prefetch_buffer_header (p, LOAD);
+	  CLIB_PREFETCH (p->data, CLIB_CACHE_LINE_BYTES, LOAD);
+
+	  p = vlib_get_buffer (vm, rxve[9].bi);
+	  vlib_prefetch_buffer_header (p, LOAD);
+	  CLIB_PREFETCH (p->data, CLIB_CACHE_LINE_BYTES, LOAD);
+
+	  p = vlib_get_buffer (vm, rxve[10].bi);
+	  vlib_prefetch_buffer_header (p, LOAD);
+	  CLIB_PREFETCH (p->data, CLIB_CACHE_LINE_BYTES, LOAD);
+
+	  p = vlib_get_buffer (vm, rxve[11].bi);
+	  vlib_prefetch_buffer_header (p, LOAD);
+	  CLIB_PREFETCH (p->data, CLIB_CACHE_LINE_BYTES, LOAD);
+
+	  to_next[0] = bi0 = rxve[0].bi;
+	  to_next[1] = bi1 = rxve[1].bi;
+	  to_next[2] = bi2 = rxve[2].bi;
+	  to_next[3] = bi3 = rxve[3].bi;
+
+	  b0 = vlib_get_buffer (vm, bi0);
+	  b1 = vlib_get_buffer (vm, bi1);
+	  b2 = vlib_get_buffer (vm, bi2);
+	  b3 = vlib_get_buffer (vm, bi3);
+
+	  b0->current_length = rxve[0].length;
+	  b1->current_length = rxve[1].length;
+	  b2->current_length = rxve[2].length;
+	  b3->current_length = rxve[3].length;
+
+	  n_rx_bytes += b0->current_length;
+	  n_rx_bytes += b1->current_length;
+	  n_rx_bytes += b2->current_length;
+	  n_rx_bytes += b3->current_length;
+
+	  if (PREDICT_TRUE (ad->per_interface_next_index == ~0))
+	    {
+	      ethernet_header_t *e0, *e1, *e2, *e3;
+
+	      e0 = (ethernet_header_t *) b0->data;
+	      e1 = (ethernet_header_t *) b1->data;
+	      e2 = (ethernet_header_t *) b2->data;
+	      e3 = (ethernet_header_t *) b3->data;
+
+	      if (ethernet_frame_is_any_tagged_x4 (e0->type, e1->type,
+						   e2->type, e3->type))
+		{
+		  next0 = avf_find_next (rxve, b0, 1);
+		  next1 = avf_find_next (rxve + 1, b1, 1);
+		  next2 = avf_find_next (rxve + 2, b2, 1);
+		  next3 = avf_find_next (rxve + 3, b3, 1);
+		}
+	      else
+		{
+		  next0 = avf_find_next (rxve, b0, 0);
+		  next1 = avf_find_next (rxve + 1, b1, 0);
+		  next2 = avf_find_next (rxve + 2, b2, 0);
+		  next3 = avf_find_next (rxve + 3, b3, 0);
+		}
+
+	      if (with_features_or_tracing)
+		vnet_feature_start_device_input_x4 (ad->sw_if_index, &next0,
+						    &next1, &next2, &next3,
+						    b0, b1, b2, b3);
+
+	      if (PREDICT_FALSE (maybe_error))
+		{
+		  avf_check_for_error (node, rxve + 0, b0, &next0);
+		  avf_check_for_error (node, rxve + 1, b1, &next1);
+		  avf_check_for_error (node, rxve + 2, b2, &next2);
+		  avf_check_for_error (node, rxve + 3, b3, &next3);
+		}
+	    }
+	  else
+	    next0 = next1 = next2 = next3 = ad->per_interface_next_index;
+
+	  clib_memcpy (vnet_buffer (b0)->sw_if_index, sw_if_idx,
+		       sizeof (sw_if_idx));
+	  clib_memcpy (vnet_buffer (b1)->sw_if_index, sw_if_idx,
+		       sizeof (sw_if_idx));
+	  clib_memcpy (vnet_buffer (b2)->sw_if_index, sw_if_idx,
+		       sizeof (sw_if_idx));
+	  clib_memcpy (vnet_buffer (b3)->sw_if_index, sw_if_idx,
+		       sizeof (sw_if_idx));
+
+	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
+	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b1);
+	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b2);
+	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b3);
+
+	  if (with_features_or_tracing && PREDICT_FALSE (n_trace))
+	    {
+	      avf_input_trace (vm, node, next0, b0, &n_trace, ad, rxve);
+	      if (n_trace)
+		avf_input_trace (vm, node, next1, b1, &n_trace, ad, rxve + 1);
+	      if (n_trace)
+		avf_input_trace (vm, node, next2, b2, &n_trace, ad, rxve + 2);
+	      if (n_trace)
+		avf_input_trace (vm, node, next3, b3, &n_trace, ad, rxve + 3);
+	    }
+
+	  /* next */
+	  to_next += 4;
+	  n_left_to_next -= 4;
+	  rxve += 4;
+	  n_rxv -= 4;
+
+	  /* enqueue */
+	  vlib_validate_buffer_enqueue_x4 (vm, node, next_index, to_next,
+					   n_left_to_next, bi0, bi1, bi2, bi3,
+					   next0, next1, next2, next3);
+	}
+      while (n_rxv && n_left_to_next)
+	{
+	  bi0 = rxve[0].bi;
+	  to_next[0] = bi0;
+	  b0 = vlib_get_buffer (vm, bi0);
+
+	  b0->current_length = rxve->length;
+	  n_rx_bytes += b0->current_length;
+
+	  if (PREDICT_TRUE (ad->per_interface_next_index == ~0))
+	    {
+	      next0 = avf_find_next (rxve, b0, 1);
+	      if (with_features_or_tracing)
+		vnet_feature_start_device_input_x1 (ad->sw_if_index, &next0,
+						    b0);
+	      avf_check_for_error (node, rxve + 0, b0, &next0);
+	    }
+	  else
+	    next0 = ad->per_interface_next_index;
+
+	  clib_memcpy (vnet_buffer (b0)->sw_if_index, sw_if_idx,
+		       sizeof (sw_if_idx));
+
+	  VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0);
+	  if (with_features_or_tracing && PREDICT_FALSE (n_trace > 0))
+	    avf_input_trace (vm, node, next0, b0, &n_trace, ad, rxve);
+
+	  /* next */
+	  to_next += 1;
+	  n_left_to_next -= 1;
+	  rxve += 1;
+	  n_rxv -= 1;
+
+	  /* enqueue */
+	  vlib_validate_buffer_enqueue_x1 (vm, node, next_index, to_next,
+					   n_left_to_next, bi0, next0);
+	}
+      vlib_put_next_frame (vm, node, next_index, n_left_to_next);
+    }
+
+  vlib_increment_combined_counter (vnm->interface_main.combined_sw_if_counters
+				   + VNET_INTERFACE_COUNTER_RX, thr_idx,
+				   ad->hw_if_index, n_rx_packets, n_rx_bytes);
+
+  return n_rx_packets;
+}
+
+uword
+CLIB_MULTIARCH_FN (avf_input) (vlib_main_t * vm, vlib_node_runtime_t * node,
+			       vlib_frame_t * frame)
+{
+  u32 n_rx = 0;
+  avf_main_t *am = &avf_main;
+  vnet_device_input_runtime_t *rt = (void *) node->runtime_data;
+  vnet_device_and_queue_t *dq;
+
+  foreach_device_and_queue (dq, rt->devices_and_queues)
+  {
+    avf_device_t *ad;
+    ad = vec_elt_at_index (am->devices, dq->dev_instance);
+    if ((ad->flags & AVF_DEVICE_F_ADMIN_UP) == 0)
+      continue;
+    if (vnet_device_input_have_features (ad->sw_if_index) ||
+	vlib_get_trace_count (vm, node))
+      n_rx += avf_device_input_inline (vm, node, frame, ad, dq->queue_id, 1);
+    else
+      n_rx += avf_device_input_inline (vm, node, frame, ad, dq->queue_id, 0);
+  }
+  return n_rx;
+}
+
+#ifndef CLIB_MULTIARCH_VARIANT
+/* *INDENT-OFF* */
+VLIB_REGISTER_NODE (avf_input_node) = {
+  .function = avf_input,
+  .name = "avf-input",
+  .sibling_of = "device-input",
+  .format_trace = format_avf_input_trace,
+  .type = VLIB_NODE_TYPE_INPUT,
+  .state = VLIB_NODE_STATE_POLLING,
+  .n_errors = AVF_INPUT_N_ERROR,
+  .error_strings = avf_input_error_strings,
+};
+
+#if __x86_64__
+vlib_node_function_t __clib_weak avf_input_avx512;
+vlib_node_function_t __clib_weak avf_input_avx2;
+static void __clib_constructor
+avf_input_multiarch_select (void)
+{
+  if (avf_input_avx512 && clib_cpu_supports_avx512f ())
+    avf_input_node.function = avf_input_avx512;
+  else if (avf_input_avx2 && clib_cpu_supports_avx2 ())
+    avf_input_node.function = avf_input_avx2;
+}
+
+#endif
+#endif
+
+/* *INDENT-ON* */
+
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/output.c b/src/plugins/avf/output.c
new file mode 100644
index 00000000000..8e5fa6a43f8
--- /dev/null
+++ b/src/plugins/avf/output.c
@@ -0,0 +1,183 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vlib/unix/unix.h>
+#include <vlib/pci/pci.h>
+#include <vnet/ethernet/ethernet.h>
+#include <vnet/devices/devices.h>
+
+#include <avf/avf.h>
+
+#define AVF_TXQ_DESC_CMD(x)             (1 << (x + 4))
+#define AVF_TXQ_DESC_CMD_EOP		AVF_TXQ_DESC_CMD(0)
+#define AVF_TXQ_DESC_CMD_RS		AVF_TXQ_DESC_CMD(1)
+
+static_always_inline u8
+avf_tx_desc_get_dtyp (avf_tx_desc_t * d)
+{
+  return d->qword[1] & 0x0f;
+}
+
+uword
+CLIB_MULTIARCH_FN (avf_interface_tx) (vlib_main_t * vm,
+				      vlib_node_runtime_t * node,
+				      vlib_frame_t * frame)
+{
+  avf_main_t *am = &avf_main;
+  vnet_interface_output_runtime_t *rd = (void *) node->runtime_data;
+  avf_device_t *ad = pool_elt_at_index (am->devices, rd->dev_instance);
+  u32 thread_index = vlib_get_thread_index ();
+  u8 qid = thread_index;
+  avf_txq_t *txq = vec_elt_at_index (ad->txqs, qid % ad->num_queue_pairs);
+  avf_tx_desc_t *d0, *d1, *d2, *d3;
+  u32 *buffers = vlib_frame_args (frame);
+  u32 bi0, bi1, bi2, bi3;
+  u16 n_left = frame->n_vectors;
+  vlib_buffer_t *b0, *b1, *b2, *b3;
+  u16 mask = txq->size - 1;
+
+  clib_spinlock_lock_if_init (&txq->lock);
+
+  /* release cosumed bufs */
+  if (txq->n_bufs)
+    {
+      u16 first, slot, n_free = 0;
+      first = slot = (txq->next - txq->n_bufs) & mask;
+      d0 = txq->descs + slot;
+      while (n_free < txq->n_bufs && avf_tx_desc_get_dtyp (d0) == 0x0F)
+	{
+	  n_free++;
+	  slot = (slot + 1) & mask;
+	  d0 = txq->descs + slot;
+	}
+
+      if (n_free)
+	{
+	  txq->n_bufs -= n_free;;
+	  vlib_buffer_free_from_ring (vm, txq->bufs, first, txq->size,
+				      n_free);
+	}
+    }
+
+  while (n_left >= 7)
+    {
+      u16 slot0, slot1, slot2, slot3;
+
+      vlib_prefetch_buffer_with_index (vm, buffers[4], LOAD);
+      vlib_prefetch_buffer_with_index (vm, buffers[5], LOAD);
+      vlib_prefetch_buffer_with_index (vm, buffers[6], LOAD);
+      vlib_prefetch_buffer_with_index (vm, buffers[7], LOAD);
+
+      slot0 = txq->next;
+      slot1 = (txq->next + 1) & mask;
+      slot2 = (txq->next + 2) & mask;
+      slot3 = (txq->next + 3) & mask;
+
+      d0 = txq->descs + slot0;
+      d1 = txq->descs + slot1;
+      d2 = txq->descs + slot2;
+      d3 = txq->descs + slot3;
+
+      bi0 = buffers[0];
+      bi1 = buffers[1];
+      bi2 = buffers[2];
+      bi3 = buffers[3];
+
+      txq->bufs[slot0] = bi0;
+      txq->bufs[slot1] = bi1;
+      txq->bufs[slot2] = bi2;
+      txq->bufs[slot3] = bi3;
+      b0 = vlib_get_buffer (vm, bi0);
+      b1 = vlib_get_buffer (vm, bi1);
+      b2 = vlib_get_buffer (vm, bi2);
+      b3 = vlib_get_buffer (vm, bi3);
+
+#if 0
+      d->qword[0] = vlib_get_buffer_data_physical_address (vm, bi0) +
+	b0->current_data;
+#else
+      d0->qword[0] = pointer_to_uword (b0->data);
+      d1->qword[0] = pointer_to_uword (b1->data);
+      d2->qword[0] = pointer_to_uword (b2->data);
+      d3->qword[0] = pointer_to_uword (b3->data);
+
+#endif
+      u64 bits = AVF_TXQ_DESC_CMD_EOP | AVF_TXQ_DESC_CMD_RS;
+      d0->qword[1] = ((u64) b0->current_length) << 34 | bits;
+      d1->qword[1] = ((u64) b1->current_length) << 34 | bits;
+      d2->qword[1] = ((u64) b2->current_length) << 34 | bits;
+      d3->qword[1] = ((u64) b3->current_length) << 34 | bits;
+
+      txq->next = (txq->next + 4) & mask;
+      txq->n_bufs += 4;
+      buffers += 4;
+      n_left -= 4;
+    }
+
+  while (n_left)
+    {
+      d0 = txq->descs + txq->next;
+      bi0 = buffers[0];
+      txq->bufs[txq->next] = bi0;
+      b0 = vlib_get_buffer (vm, bi0);
+
+#if 0
+      d->qword[0] = vlib_get_buffer_data_physical_address (vm, bi0) +
+	b0->current_data;
+#else
+      d0->qword[0] = pointer_to_uword (b0->data);
+
+#endif
+      d0->qword[1] = ((u64) b0->current_length) << 34;
+      d0->qword[1] |= AVF_TXQ_DESC_CMD_EOP | AVF_TXQ_DESC_CMD_RS;
+
+      txq->next = (txq->next + 1) & mask;
+      txq->n_bufs++;
+      buffers++;
+      n_left--;
+    }
+  CLIB_MEMORY_BARRIER ();
+  *(txq->qtx_tail) = txq->next;
+
+  clib_spinlock_unlock_if_init (&txq->lock);
+
+  return frame->n_vectors - n_left;
+}
+
+#ifndef CLIB_MULTIARCH_VARIANT
+#if __x86_64__
+vlib_node_function_t __clib_weak avf_interface_tx_avx512;
+vlib_node_function_t __clib_weak avf_interface_tx_avx2;
+static void __clib_constructor
+avf_interface_tx_multiarch_select (void)
+{
+  if (avf_interface_tx_avx512 && clib_cpu_supports_avx512f ())
+    avf_device_class.tx_function = avf_interface_tx_avx512;
+  else if (avf_interface_tx_avx2 && clib_cpu_supports_avx2 ())
+    avf_device_class.tx_function = avf_interface_tx_avx2;
+}
+#endif
+#endif
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/plugin.c b/src/plugins/avf/plugin.c
new file mode 100644
index 00000000000..d2a043ac2cb
--- /dev/null
+++ b/src/plugins/avf/plugin.c
@@ -0,0 +1,35 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#include <vlib/vlib.h>
+#include <vnet/plugin/plugin.h>
+#include <vpp/app/version.h>
+
+/* *INDENT-OFF* */
+VLIB_PLUGIN_REGISTER () = {
+  .version = VPP_BUILD_VER,
+  .description = "Intel Adaptive Virtual Function (AVF) Device Plugin",
+};
+/* *INDENT-ON* */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/plugins/avf/virtchnl.h b/src/plugins/avf/virtchnl.h
new file mode 100644
index 00000000000..671da959c2b
--- /dev/null
+++ b/src/plugins/avf/virtchnl.h
@@ -0,0 +1,342 @@
+/*
+ *------------------------------------------------------------------
+ * Copyright (c) 2018 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *------------------------------------------------------------------
+ */
+
+#define VIRTCHNL_VERSION_MAJOR 1
+#define VIRTCHNL_VERSION_MINOR 1
+
+
+#define AVFINT_DYN_CTLN(x)  (0x00003800 + (0x4 * x))
+#define AVFINT_ICR0         0x00004800
+#define AVFINT_ICR0_ENA1    0x00005000
+#define AVFINT_DYN_CTL0     0x00005C00
+#define AVF_ARQBAH          0x00006000
+#define AVF_ATQH            0x00006400
+#define AVF_ATQLEN          0x00006800
+#define AVF_ARQBAL          0x00006C00
+#define AVF_ARQT            0x00007000
+#define AVF_ARQH            0x00007400
+#define AVF_ATQBAH          0x00007800
+#define AVF_ATQBAL          0x00007C00
+#define AVF_ARQLEN          0x00008000
+#define AVF_ATQT            0x00008400
+#define AVFGEN_RSTAT        0x00008800
+#define AVF_QTX_TAIL(q)     (0x00000000 + (0x4 * q))
+#define AVF_QRX_TAIL(q)     (0x00002000 + (0x4 * q))
+
+#define AVF_AQ_F_DD  (1 << 0)
+#define AVF_AQ_F_CMP (1 << 1)
+#define AVF_AQ_F_ERR (1 << 2)
+#define AVF_AQ_F_VFE (1 << 3)
+#define AVF_AQ_F_LB  (1 << 9)
+#define AVF_AQ_F_RD  (1 << 10)
+#define AVF_AQ_F_VFC (1 << 11)
+#define AVF_AQ_F_BUF (1 << 12)
+#define AVF_AQ_F_SI  (1 << 13)
+#define AVF_AQ_F_EI  (1 << 14)
+#define AVF_AQ_F_FE  (1 << 15)
+
+
+#define foreach_virtchnl_op \
+  _(0, UNKNOWN)					\
+  _(1, VERSION)					\
+  _(2, RESET_VF)				\
+  _(3, GET_VF_RESOURCES)			\
+  _(4, CONFIG_TX_QUEUE)				\
+  _(5, CONFIG_RX_QUEUE)				\
+  _(6, CONFIG_VSI_QUEUES)			\
+  _(7, CONFIG_IRQ_MAP)				\
+  _(8, ENABLE_QUEUES)				\
+  _(9, DISABLE_QUEUES)				\
+  _(10, ADD_ETH_ADDR)				\
+  _(11, DEL_ETH_ADDR)				\
+  _(12, ADD_VLAN)				\
+  _(13, DEL_VLAN)				\
+  _(14, CONFIG_PROMISCUOUS_MODE)		\
+  _(15, GET_STATS)				\
+  _(16, RSVD)					\
+  _(17, EVENT)					\
+  _(18, UNDEF_18)				\
+  _(19, UNDEF_19)				\
+  _(20, IWARP)					\
+  _(21, CONFIG_IWARP_IRQ_MAP)			\
+  _(22, RELEASE_IWARP_IRQ_MAP)			\
+  _(23, CONFIG_RSS_KEY)				\
+  _(24, CONFIG_RSS_LUT)				\
+  _(25, GET_RSS_HENA_CAPS)			\
+  _(26, SET_RSS_HENA)				\
+  _(27, ENABLE_VLAN_STRIPPING)			\
+  _(28, DISABLE_VLAN_STRIPPING)			\
+  _(29, REQUEST_QUEUES)
+
+typedef enum
+{
+#define _(v,n) VIRTCHNL_OP_##n = v,
+  foreach_virtchnl_op
+#undef _
+    VIRTCHNL_N_OPS,
+} virtchnl_ops_t;
+
+typedef enum
+{
+  VIRTCHNL_STATUS_SUCCESS = 0,
+  VIRTCHNL_ERR_PARAM = -5,
+  VIRTCHNL_STATUS_ERR_OPCODE_MISMATCH = -38,
+  VIRTCHNL_STATUS_ERR_CQP_COMPL_ERROR = -39,
+  VIRTCHNL_STATUS_ERR_INVALID_VF_ID = -40,
+  VIRTCHNL_STATUS_NOT_SUPPORTED = -64,
+} virtchnl_status_code_t;
+
+#define foreach_avf_vf_cap_flag \
+  _( 0, L2, "l2") \
+  _( 1, IWARP, "iwarp") \
+  _( 2, RSVD, "rsvd") \
+  _( 3, RSS_AQ, "rss-aq") \
+  _( 4, RSS_REG, "rss-reg") \
+  _( 5, WB_ON_ITR, "wb-on-itr") \
+  _( 6, REQ_QUEUES, "req-queues") \
+  _(16, VLAN, "vlan") \
+  _(17, RX_POLLING, "rx-polling") \
+  _(18, RSS_PCTYPE_V2, "rss-pctype-v2") \
+  _(19, RSS_PF, "rss-pf") \
+  _(20, ENCAP, "encap") \
+  _(21, ENCAP_CSUM, "encap-csum") \
+  _(22, RX_ENCAP_CSUM, "rx-encap-csum")
+
+typedef enum
+{
+#define _(a, b, c) VIRTCHNL_VF_OFFLOAD_##b = (1 << a),
+  foreach_avf_vf_cap_flag
+#undef _
+} avf_vf_cap_flag_t;
+
+typedef enum
+{
+  VIRTCHNL_VSI_TYPE_INVALID = 0,
+  VIRTCHNL_VSI_SRIOV = 6,
+} virtchnl_vsi_type_t;
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 num_queue_pairs;
+  virtchnl_vsi_type_t vsi_type;
+  u16 qset_handle;
+  u8 default_mac_addr[6];
+} virtchnl_vsi_resource_t;
+
+typedef struct
+{
+  u16 num_vsis;
+  u16 num_queue_pairs;
+  u16 max_vectors;
+  u16 max_mtu;
+  u32 vf_offload_flags;
+  u32 rss_key_size;
+  u32 rss_lut_size;
+  virtchnl_vsi_resource_t vsi_res[1];
+} virtchnl_vf_resource_t;
+
+typedef enum
+{
+  VIRTCHNL_EVENT_UNKNOWN = 0,
+  VIRTCHNL_EVENT_LINK_CHANGE,
+  VIRTCHNL_EVENT_RESET_IMPENDING,
+  VIRTCHNL_EVENT_PF_DRIVER_CLOSE,
+} virtchnl_event_codes_t;
+
+#define foreach_virtchnl_link_speed \
+  _(1, 100MB, "100 Mbps")		\
+  _(2, 1GB, "1 Gbps")			\
+  _(3, 10GB, "10 Gbps")			\
+  _(4, 40GB, "40 Gbps")			\
+  _(5, 20GB, "20 Gbps")			\
+  _(6, 25GB, "25 Gbps")
+
+typedef enum
+{
+  VIRTCHNL_LINK_SPEED_UNKNOWN = 0,
+#define _(a,b,c) VIRTCHNL_LINK_SPEED_##b = (1 << a),
+  foreach_virtchnl_link_speed
+#undef _
+} virtchnl_link_speed_t;
+
+typedef struct
+{
+  virtchnl_event_codes_t event;
+  union
+  {
+    struct
+    {
+      virtchnl_link_speed_t link_speed;
+      _Bool link_status;
+    } link_event;
+  } event_data;
+  int severity;
+} virtchnl_pf_event_t;
+
+STATIC_ASSERT_SIZEOF (virtchnl_pf_event_t, 16);
+
+typedef struct
+{
+  u32 major;
+  u32 minor;
+} virtchnl_version_info_t;
+
+typedef struct
+{
+  u16 flags;
+  u16 opcode;
+  u16 datalen;
+  u16 retval;
+  union
+  {
+    u32 cookie_hi;
+    virtchnl_ops_t v_opcode;
+  };
+  union
+  {
+    u32 cookie_lo;
+    virtchnl_status_code_t v_retval;
+  };
+  u32 param0;
+  u32 param1;
+  union
+  {
+    u32 param2;
+    u32 addr_hi;
+  };
+  union
+  {
+    u32 param3;
+    u32 addr_lo;
+  };
+} avf_aq_desc_t;
+
+STATIC_ASSERT_SIZEOF (avf_aq_desc_t, 32);
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 queue_id;
+  u16 ring_len;
+  u64 dma_ring_addr;
+  u64 dma_headwb_addr;
+} virtchnl_txq_info_t;
+
+STATIC_ASSERT_SIZEOF (virtchnl_txq_info_t, 24);
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 queue_id;
+  u32 ring_len;
+  u16 hdr_size;
+  u16 splithdr_enabled;
+  u32 databuffer_size;
+  u32 max_pkt_size;
+  u32 pad1;
+  u64 dma_ring_addr;
+  int rx_split_pos;
+  u32 pad2;
+} virtchnl_rxq_info_t;
+
+STATIC_ASSERT_SIZEOF (virtchnl_rxq_info_t, 40);
+
+typedef struct
+{
+  virtchnl_txq_info_t txq;
+  virtchnl_rxq_info_t rxq;
+} virtchnl_queue_pair_info_t;
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 num_queue_pairs;
+  u32 pad;
+  virtchnl_queue_pair_info_t qpair[1];
+} virtchnl_vsi_queue_config_info_t;
+
+STATIC_ASSERT_SIZEOF (virtchnl_vsi_queue_config_info_t, 72);
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 pad;
+  u32 rx_queues;
+  u32 tx_queues;
+} virtchnl_queue_select_t;
+
+STATIC_ASSERT_SIZEOF (virtchnl_queue_select_t, 12);
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 vector_id;
+  u16 rxq_map;
+  u16 txq_map;
+  u16 rxitr_idx;
+  u16 txitr_idx;
+} virtchnl_vector_map_t;
+
+typedef struct
+{
+  u16 num_vectors;
+  virtchnl_vector_map_t vecmap[1];
+} virtchnl_irq_map_info_t;
+
+STATIC_ASSERT_SIZEOF (virtchnl_irq_map_info_t, 14);
+
+typedef struct
+{
+  u8 addr[6];
+  u8 pad[2];
+} virtchnl_ether_addr_t;
+
+typedef struct
+{
+  u16 vsi_id;
+  u16 num_elements;
+  virtchnl_ether_addr_t list[1];
+} virtchnl_ether_addr_list_t;
+
+#define foreach_virtchnl_eth_stats \
+  _(rx_bytes)		\
+  _(rx_unicast)		\
+  _(rx_multicast)	\
+  _(rx_broadcast)	\
+  _(rx_discards)	\
+  _(rx_unknown_protocol)\
+  _(tx_bytes)		\
+  _(tx_unicast)		\
+  _(tx_multicast)	\
+  _(tx_broadcast)	\
+  _(tx_discards)	\
+  _(tx_errors)
+
+typedef struct
+{
+#define _(s) u64 s;
+  foreach_virtchnl_eth_stats
+#undef _
+} virtchnl_eth_stats_t;
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */