first public release
[dpdk.git] / lib / librte_pmd_igb / e1000_ethdev.c
diff --git a/lib/librte_pmd_igb/e1000_ethdev.c b/lib/librte_pmd_igb/e1000_ethdev.c
new file mode 100644 (file)
index 0000000..a984428
--- /dev/null
@@ -0,0 +1,1319 @@
+/*-
+ *   BSD LICENSE
+ * 
+ *   Copyright(c) 2010-2012 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ * 
+ *   Redistribution and use in source and binary forms, with or without 
+ *   modification, are permitted provided that the following conditions 
+ *   are met:
+ * 
+ *     * Redistributions of source code must retain the above copyright 
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright 
+ *       notice, this list of conditions and the following disclaimer in 
+ *       the documentation and/or other materials provided with the 
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its 
+ *       contributors may be used to endorse or promote products derived 
+ *       from this software without specific prior written permission.
+ * 
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ * 
+ *  version: DPDK.L.1.2.3-3
+ */
+
+#include <sys/queue.h>
+#include <stdio.h>
+#include <errno.h>
+#include <stdint.h>
+#include <stdarg.h>
+
+#include <rte_common.h>
+#include <rte_interrupts.h>
+#include <rte_byteorder.h>
+#include <rte_log.h>
+#include <rte_debug.h>
+#include <rte_pci.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_tailq.h>
+#include <rte_eal.h>
+#include <rte_atomic.h>
+#include <rte_malloc.h>
+
+#include "e1000_logs.h"
+#include "igb/e1000_api.h"
+#include "igb/e1000_hw.h"
+#include "e1000_ethdev.h"
+
+static int  eth_igb_configure(struct rte_eth_dev *dev, uint16_t nb_rx_q,
+               uint16_t nb_tx_q);
+static int  eth_igb_start(struct rte_eth_dev *dev);
+static void eth_igb_stop(struct rte_eth_dev *dev);
+static void eth_igb_close(struct rte_eth_dev *dev);
+static void eth_igb_promiscuous_enable(struct rte_eth_dev *dev);
+static void eth_igb_promiscuous_disable(struct rte_eth_dev *dev);
+static void eth_igb_allmulticast_enable(struct rte_eth_dev *dev);
+static void eth_igb_allmulticast_disable(struct rte_eth_dev *dev);
+static int  eth_igb_link_update(struct rte_eth_dev *dev,
+                               int wait_to_complete);
+static void eth_igb_stats_get(struct rte_eth_dev *dev,
+                               struct rte_eth_stats *rte_stats);
+static void eth_igb_stats_reset(struct rte_eth_dev *dev);
+static void eth_igb_infos_get(struct rte_eth_dev *dev,
+                               struct rte_eth_dev_info *dev_info);
+static int  eth_igb_flow_ctrl_set(struct rte_eth_dev *dev,
+                               struct rte_eth_fc_conf *fc_conf);
+static int eth_igb_interrupt_setup(struct rte_eth_dev *dev);
+static int eth_igb_interrupt_get_status(struct rte_eth_dev *dev);
+static int eth_igb_interrupt_action(struct rte_eth_dev *dev);
+static void eth_igb_interrupt_handler(struct rte_intr_handle *handle,
+                                                       void *param);
+static int  igb_hardware_init(struct e1000_hw *hw);
+static void igb_hw_control_acquire(struct e1000_hw *hw);
+static void igb_hw_control_release(struct e1000_hw *hw);
+static void igb_init_manageability(struct e1000_hw *hw);
+static void igb_release_manageability(struct e1000_hw *hw);
+static void igb_vlan_hw_support_enable(struct rte_eth_dev *dev);
+static void igb_vlan_hw_support_disable(struct rte_eth_dev *dev);
+static void eth_igb_vlan_filter_set(struct rte_eth_dev *dev,
+                                     uint16_t vlan_id,
+                                     int on);
+static int eth_igb_led_on(struct rte_eth_dev *dev);
+static int eth_igb_led_off(struct rte_eth_dev *dev);
+
+static void igb_intr_disable(struct e1000_hw *hw);
+static int  igb_get_rx_buffer_size(struct e1000_hw *hw);
+static void eth_igb_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
+               uint32_t index, uint32_t pool);
+static void eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index);
+
+#define IGB_FC_PAUSE_TIME 0x0680
+#define IGB_LINK_UPDATE_CHECK_TIMEOUT  90  /* 9s */
+#define IGB_LINK_UPDATE_CHECK_INTERVAL 100 /* ms */
+
+static enum e1000_fc_mode igb_fc_setting = e1000_fc_full;
+
+/*
+ * The set of PCI devices this driver supports
+ */
+static struct rte_pci_id pci_id_igb_map[] = {
+
+#undef RTE_LIBRTE_IXGBE_PMD
+#define RTE_PCI_DEV_ID_DECL(vend, dev) {RTE_PCI_DEVICE(vend, dev)},
+#include "rte_pci_dev_ids.h"
+
+{.device_id = 0},
+};
+
+static struct eth_dev_ops eth_igb_ops = {
+       .dev_configure        = eth_igb_configure,
+       .dev_start            = eth_igb_start,
+       .dev_stop             = eth_igb_stop,
+       .dev_close            = eth_igb_close,
+       .promiscuous_enable   = eth_igb_promiscuous_enable,
+       .promiscuous_disable  = eth_igb_promiscuous_disable,
+       .allmulticast_enable  = eth_igb_allmulticast_enable,
+       .allmulticast_disable = eth_igb_allmulticast_disable,
+       .link_update          = eth_igb_link_update,
+       .stats_get            = eth_igb_stats_get,
+       .stats_reset          = eth_igb_stats_reset,
+       .dev_infos_get        = eth_igb_infos_get,
+       .vlan_filter_set      = eth_igb_vlan_filter_set,
+       .rx_queue_setup       = eth_igb_rx_queue_setup,
+       .tx_queue_setup       = eth_igb_tx_queue_setup,
+       .dev_led_on           = eth_igb_led_on,
+       .dev_led_off          = eth_igb_led_off,
+       .flow_ctrl_set        = eth_igb_flow_ctrl_set,
+       .mac_addr_add         = eth_igb_rar_set,
+       .mac_addr_remove      = eth_igb_rar_clear,
+};
+
+/**
+ * Atomically reads the link status information from global
+ * structure rte_eth_dev.
+ *
+ * @param dev
+ *   - Pointer to the structure rte_eth_dev to read from.
+ *   - Pointer to the buffer to be saved with the link status.
+ *
+ * @return
+ *   - On success, zero.
+ *   - On failure, negative value.
+ */
+static inline int
+rte_igb_dev_atomic_read_link_status(struct rte_eth_dev *dev,
+                               struct rte_eth_link *link)
+{
+       struct rte_eth_link *dst = link;
+       struct rte_eth_link *src = &(dev->data->dev_link);
+
+       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
+                                       *(uint64_t *)src) == 0)
+               return -1;
+
+       return 0;
+}
+
+/**
+ * Atomically writes the link status information into global
+ * structure rte_eth_dev.
+ *
+ * @param dev
+ *   - Pointer to the structure rte_eth_dev to read from.
+ *   - Pointer to the buffer to be saved with the link status.
+ *
+ * @return
+ *   - On success, zero.
+ *   - On failure, negative value.
+ */
+static inline int
+rte_igb_dev_atomic_write_link_status(struct rte_eth_dev *dev,
+                               struct rte_eth_link *link)
+{
+       struct rte_eth_link *dst = &(dev->data->dev_link);
+       struct rte_eth_link *src = link;
+
+       if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
+                                       *(uint64_t *)src) == 0)
+               return -1;
+
+       return 0;
+}
+
+static void
+igb_identify_hardware(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       hw->vendor_id = dev->pci_dev->id.vendor_id;
+       hw->device_id = dev->pci_dev->id.device_id;
+       hw->subsystem_vendor_id = dev->pci_dev->id.subsystem_vendor_id;
+       hw->subsystem_device_id = dev->pci_dev->id.subsystem_device_id;
+
+       e1000_set_mac_type(hw);
+
+       /* need to check if it is a vf device below */
+}
+
+static int
+eth_igb_dev_init(__attribute__((unused)) struct eth_driver *eth_drv,
+                  struct rte_eth_dev *eth_dev)
+{
+       int error = 0;
+       struct rte_pci_device *pci_dev;
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
+       struct e1000_vfta * shadow_vfta =
+               E1000_DEV_PRIVATE_TO_VFTA(eth_dev->data->dev_private);
+
+       pci_dev = eth_dev->pci_dev;
+       eth_dev->dev_ops = &eth_igb_ops;
+       eth_dev->rx_pkt_burst = &eth_igb_recv_pkts;
+       eth_dev->tx_pkt_burst = &eth_igb_xmit_pkts;
+
+       /* for secondary processes, we don't initialise any further as primary
+        * has already done this work. Only check we don't need a different
+        * RX function */
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY){
+               if (eth_dev->data->scattered_rx)
+                       eth_dev->rx_pkt_burst = &eth_igb_recv_scattered_pkts;
+               return 0;
+       }
+
+       hw->hw_addr= (void *)pci_dev->mem_resource.addr;
+
+       igb_identify_hardware(eth_dev);
+
+       if (e1000_setup_init_funcs(hw, TRUE) != E1000_SUCCESS) {
+               error = -EIO;
+               goto err_late;
+       }
+
+       e1000_get_bus_info(hw);
+
+       hw->mac.autoneg = 1;
+       hw->phy.autoneg_wait_to_complete = 0;
+       hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
+
+       /* Copper options */
+       if (hw->phy.media_type == e1000_media_type_copper) {
+               hw->phy.mdix = 0; /* AUTO_ALL_MODES */
+               hw->phy.disable_polarity_correction = 0;
+               hw->phy.ms_type = e1000_ms_hw_default;
+       }
+
+       /*
+        * Start from a known state, this is important in reading the nvm
+        * and mac from that.
+        */
+       e1000_reset_hw(hw);
+
+       /* Make sure we have a good EEPROM before we read from it */
+       if (e1000_validate_nvm_checksum(hw) < 0) {
+               /*
+                * Some PCI-E parts fail the first check due to
+                * the link being in sleep state, call it again,
+                * if it fails a second time its a real issue.
+                */
+               if (e1000_validate_nvm_checksum(hw) < 0) {
+                       PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
+                       error = -EIO;
+                       goto err_late;
+               }
+       }
+
+       /* Read the permanent MAC address out of the EEPROM */
+       if (e1000_read_mac_addr(hw) != 0) {
+               PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
+               error = -EIO;
+               goto err_late;
+       }
+
+       /* Allocate memory for storing MAC addresses */
+       eth_dev->data->mac_addrs = rte_zmalloc("e1000",
+               ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
+       if (eth_dev->data->mac_addrs == NULL) {
+               PMD_INIT_LOG(ERR, "Failed to allocate %d bytes needed to "
+                                               "store MAC addresses",
+                               ETHER_ADDR_LEN * hw->mac.rar_entry_count);
+               error = -ENOMEM;
+               goto err_late;
+       }
+
+       /* Copy the permanent MAC address */
+       ether_addr_copy((struct ether_addr *)hw->mac.addr, &eth_dev->data->mac_addrs[0]);
+
+       /* initialize the vfta */
+       memset(shadow_vfta, 0, sizeof(*shadow_vfta));
+
+       /* Now initialize the hardware */
+       if (igb_hardware_init(hw) != 0) {
+               PMD_INIT_LOG(ERR, "Hardware initialization failed");
+               rte_free(eth_dev->data->mac_addrs);
+               eth_dev->data->mac_addrs = NULL;
+               error = -ENODEV;
+               goto err_late;
+       }
+       hw->mac.get_link_status = 1;
+
+       /* Indicate SOL/IDER usage */
+       if (e1000_check_reset_block(hw) < 0) {
+               PMD_INIT_LOG(ERR, "PHY reset is blocked due to"
+                                       "SOL/IDER session");
+       }
+
+       PMD_INIT_LOG(INFO, "port_id %d vendorID=0x%x deviceID=0x%x\n",
+                    eth_dev->data->port_id, pci_dev->id.vendor_id,
+                    pci_dev->id.device_id);
+
+       rte_intr_callback_register(&(pci_dev->intr_handle),
+               eth_igb_interrupt_handler, (void *)eth_dev);
+
+       return 0;
+
+err_late:
+       igb_hw_control_release(hw);
+
+       return (error);
+}
+
+static struct eth_driver rte_igb_pmd = {
+       {
+               .name = "rte_igb_pmd",
+               .id_table = pci_id_igb_map,
+               .drv_flags = RTE_PCI_DRV_NEED_IGB_UIO,
+       },
+       .eth_dev_init = eth_igb_dev_init,
+       .dev_private_size = sizeof(struct e1000_adapter),
+};
+
+int
+rte_igb_pmd_init(void)
+{
+       rte_eth_driver_register(&rte_igb_pmd);
+       return 0;
+}
+
+static int
+eth_igb_configure(struct rte_eth_dev *dev, uint16_t nb_rx_q, uint16_t nb_tx_q)
+{
+       struct e1000_interrupt *intr =
+               E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
+       int diag;
+
+       PMD_INIT_LOG(DEBUG, ">>");
+
+       intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
+
+       /* Allocate the array of pointers to RX structures */
+       diag = igb_dev_rx_queue_alloc(dev, nb_rx_q);
+       if (diag != 0) {
+               PMD_INIT_LOG(ERR, "ethdev port_id=%u allocation of array of %u"
+                                       " pointers to RX queues failed",
+                                       dev->data->port_id, nb_rx_q);
+               return diag;
+       }
+
+       /* Allocate the array of pointers to TX structures */
+       diag = igb_dev_tx_queue_alloc(dev, nb_tx_q);
+       if (diag != 0) {
+               PMD_INIT_LOG(ERR, "ethdev port_id=%u allocation of array of %u"
+                                       " pointers to TX queues failed",
+                                       dev->data->port_id, nb_tx_q);
+
+               return diag;
+       }
+
+       PMD_INIT_LOG(DEBUG, "<<");
+
+       return (0);
+}
+
+static int
+eth_igb_start(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       int ret, i;
+
+       PMD_INIT_LOG(DEBUG, ">>");
+
+       igb_intr_disable(hw);
+
+       /* Power up the phy. Needed to make the link go Up */
+       e1000_power_up_phy(hw);
+
+       /*
+        * Packet Buffer Allocation (PBA)
+        * Writing PBA sets the receive portion of the buffer
+        * the remainder is used for the transmit buffer.
+        */
+       if (hw->mac.type == e1000_82575) {
+               uint32_t pba;
+
+               pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
+               E1000_WRITE_REG(hw, E1000_PBA, pba);
+       }
+
+       /* Put the address into the Receive Address Array */
+       e1000_rar_set(hw, hw->mac.addr, 0);
+
+       /* Initialize the hardware */
+       if (igb_hardware_init(hw)) {
+               PMD_INIT_LOG(ERR, "Unable to initialize the hardware");
+               return (-1);
+       }
+
+       E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN);
+
+       /* Configure for OS presence */
+       igb_init_manageability(hw);
+
+       eth_igb_tx_init(dev);
+
+       /* This can fail when allocating mbufs for descriptor rings */
+       ret = eth_igb_rx_init(dev);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "Unable to initialize RX hardware");
+               return ret;
+       }
+
+       e1000_clear_hw_cntrs_base_generic(hw);
+
+       /*
+        * If VLAN filtering is enabled, set up VLAN tag offload and filtering
+        * and restore the VFTA.
+        */
+       if (dev->data->dev_conf.rxmode.hw_vlan_filter)
+               igb_vlan_hw_support_enable(dev);
+       else
+               igb_vlan_hw_support_disable(dev);
+
+       /*
+        * Configure the Interrupt Moderation register (EITR) with the maximum
+        * possible value (0xFFFF) to minimize "System Partial Write" issued by
+        * spurious [DMA] memory updates of RX and TX ring descriptors.
+        *
+        * With a EITR granularity of 2 microseconds in the 82576, only 7/8
+        * spurious memory updates per second should be expected.
+        * ((65535 * 2) / 1000.1000 ~= 0.131 second).
+        *
+        * Because interrupts are not used at all, the MSI-X is not activated
+        * and interrupt moderation is controlled by EITR[0].
+        *
+        * Note that having [almost] disabled memory updates of RX and TX ring
+        * descriptors through the Interrupt Moderation mechanism, memory
+        * updates of ring descriptors are now moderated by the configurable
+        * value of Write-Back Threshold registers.
+        */
+       if ((hw->mac.type == e1000_82576) || (hw->mac.type == e1000_82580) ||
+                       (hw->mac.type == e1000_i350)) {
+               uint32_t ivar;
+
+               /* Enable all RX & TX queues in the IVAR registers */
+               ivar = (uint32_t) ((E1000_IVAR_VALID << 16) | E1000_IVAR_VALID);
+               for (i = 0; i < 8; i++)
+                       E1000_WRITE_REG_ARRAY(hw, E1000_IVAR0, i, ivar);
+
+               /* Configure EITR with the maximum possible value (0xFFFF) */
+               E1000_WRITE_REG(hw, E1000_EITR(0), 0xFFFF);
+       }
+
+       /* Don't reset the phy next time init gets called */
+       hw->phy.reset_disable = 1;
+
+       /* Setup link speed and duplex */
+       switch (dev->data->dev_conf.link_speed) {
+       case ETH_LINK_SPEED_AUTONEG:
+               if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
+                       hw->phy.autoneg_advertised = E1000_ALL_SPEED_DUPLEX;
+               else if (dev->data->dev_conf.link_duplex == ETH_LINK_HALF_DUPLEX)
+                       hw->phy.autoneg_advertised = E1000_ALL_HALF_DUPLEX;
+               else if (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX)
+                       hw->phy.autoneg_advertised = E1000_ALL_FULL_DUPLEX;
+               else
+                       goto error_invalid_config;
+               break;
+       case ETH_LINK_SPEED_10:
+               if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
+                       hw->phy.autoneg_advertised = E1000_ALL_10_SPEED;
+               else if (dev->data->dev_conf.link_duplex == ETH_LINK_HALF_DUPLEX)
+                       hw->phy.autoneg_advertised = ADVERTISE_10_HALF;
+               else if (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX)
+                       hw->phy.autoneg_advertised = ADVERTISE_10_FULL;
+               else
+                       goto error_invalid_config;
+               break;
+       case ETH_LINK_SPEED_100:
+               if (dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX)
+                       hw->phy.autoneg_advertised = E1000_ALL_100_SPEED;
+               else if (dev->data->dev_conf.link_duplex == ETH_LINK_HALF_DUPLEX)
+                       hw->phy.autoneg_advertised = ADVERTISE_100_HALF;
+               else if (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX)
+                       hw->phy.autoneg_advertised = ADVERTISE_100_FULL;
+               else
+                       goto error_invalid_config;
+               break;
+       case ETH_LINK_SPEED_1000:
+               if ((dev->data->dev_conf.link_duplex == ETH_LINK_AUTONEG_DUPLEX) ||
+                               (dev->data->dev_conf.link_duplex == ETH_LINK_FULL_DUPLEX))
+                       hw->phy.autoneg_advertised = ADVERTISE_1000_FULL;
+               else
+                       goto error_invalid_config;
+               break;
+       case ETH_LINK_SPEED_10000:
+       default:
+               goto error_invalid_config;
+       }
+       e1000_setup_link(hw);
+
+       PMD_INIT_LOG(DEBUG, "<<");
+
+       /* check if lsc interrupt feature is enabled */
+       if (dev->data->dev_conf.intr_conf.lsc != 0)
+               return eth_igb_interrupt_setup(dev);
+
+       return (0);
+
+error_invalid_config:
+       PMD_INIT_LOG(ERR, "Invalid link_speed/link_duplex (%u/%u) for port %u\n",
+                       dev->data->dev_conf.link_speed,
+                       dev->data->dev_conf.link_duplex, dev->data->port_id);
+       return -1;
+}
+
+/*********************************************************************
+ *
+ *  This routine disables all traffic on the adapter by issuing a
+ *  global reset on the MAC.
+ *
+ **********************************************************************/
+static void
+eth_igb_stop(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_eth_link link;
+
+       igb_intr_disable(hw);
+       e1000_reset_hw(hw);
+       E1000_WRITE_REG(hw, E1000_WUC, 0);
+
+       /* Power down the phy. Needed to make the link go Down */
+       e1000_power_down_phy(hw);
+
+       igb_dev_clear_queues(dev);
+
+       /* clear the recorded link status */
+       memset(&link, 0, sizeof(link));
+       rte_igb_dev_atomic_write_link_status(dev, &link);
+}
+
+static void
+eth_igb_close(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_eth_link link;
+
+       eth_igb_stop(dev);
+       e1000_phy_hw_reset(hw);
+       igb_release_manageability(hw);
+       igb_hw_control_release(hw);
+
+       igb_dev_clear_queues(dev);
+
+       memset(&link, 0, sizeof(link));
+       rte_igb_dev_atomic_write_link_status(dev, &link);
+}
+
+static int
+igb_get_rx_buffer_size(struct e1000_hw *hw)
+{
+       uint32_t rx_buf_size;
+       if (hw->mac.type == e1000_82576) {
+               rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xffff) << 10;
+       } else if (hw->mac.type == e1000_82580) {
+               /* PBS needs to be translated according to a lookup table */
+               rx_buf_size = (E1000_READ_REG(hw, E1000_RXPBS) & 0xf);
+               rx_buf_size = (uint32_t) e1000_rxpbs_adjust_82580(rx_buf_size);
+               rx_buf_size = (rx_buf_size << 10);
+       } else {
+               rx_buf_size = (E1000_READ_REG(hw, E1000_PBA) & 0xffff) << 10;
+       }
+
+       return rx_buf_size;
+}
+
+/*********************************************************************
+ *
+ *  Initialize the hardware
+ *
+ **********************************************************************/
+static int
+igb_hardware_init(struct e1000_hw *hw)
+{
+       uint32_t rx_buf_size;
+       int diag;
+
+       /* Let the firmware know the OS is in control */
+       igb_hw_control_acquire(hw);
+
+       /*
+        * These parameters control the automatic generation (Tx) and
+        * response (Rx) to Ethernet PAUSE frames.
+        * - High water mark should allow for at least two standard size (1518)
+        *   frames to be received after sending an XOFF.
+        * - Low water mark works best when it is very near the high water mark.
+        *   This allows the receiver to restart by sending XON when it has
+        *   drained a bit. Here we use an arbitary value of 1500 which will
+        *   restart after one full frame is pulled from the buffer. There
+        *   could be several smaller frames in the buffer and if so they will
+        *   not trigger the XON until their total number reduces the buffer
+        *   by 1500.
+        * - The pause time is fairly large at 1000 x 512ns = 512 usec.
+        */
+       rx_buf_size = igb_get_rx_buffer_size(hw);
+
+       hw->fc.high_water = rx_buf_size - (ETHER_MAX_LEN * 2);
+       hw->fc.low_water = hw->fc.high_water - 1500;
+       hw->fc.pause_time = IGB_FC_PAUSE_TIME;
+       hw->fc.send_xon = 1;
+
+       /* Set Flow control, use the tunable location if sane */
+       if ((igb_fc_setting != e1000_fc_none) && (igb_fc_setting < 4))
+               hw->fc.requested_mode = igb_fc_setting;
+       else
+               hw->fc.requested_mode = e1000_fc_none;
+
+       /* Issue a global reset */
+       e1000_reset_hw(hw);
+       E1000_WRITE_REG(hw, E1000_WUC, 0);
+
+       diag = e1000_init_hw(hw);
+       if (diag < 0)
+               return (diag);
+
+       E1000_WRITE_REG(hw, E1000_VET, ETHER_TYPE_VLAN);
+       e1000_get_phy_info(hw);
+       e1000_check_for_link(hw);
+
+       return (0);
+}
+
+/* This function is based on igb_update_stats_counters() in igb/if_igb.c */
+static void
+eth_igb_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_hw_stats *stats =
+                       E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
+       int pause_frames;
+
+       if(hw->phy.media_type == e1000_media_type_copper ||
+           (E1000_READ_REG(hw, E1000_STATUS) & E1000_STATUS_LU)) {
+               stats->symerrs +=
+                   E1000_READ_REG(hw,E1000_SYMERRS);
+               stats->sec += E1000_READ_REG(hw, E1000_SEC);
+       }
+
+       stats->crcerrs += E1000_READ_REG(hw, E1000_CRCERRS);
+       stats->mpc += E1000_READ_REG(hw, E1000_MPC);
+       stats->scc += E1000_READ_REG(hw, E1000_SCC);
+       stats->ecol += E1000_READ_REG(hw, E1000_ECOL);
+
+       stats->mcc += E1000_READ_REG(hw, E1000_MCC);
+       stats->latecol += E1000_READ_REG(hw, E1000_LATECOL);
+       stats->colc += E1000_READ_REG(hw, E1000_COLC);
+       stats->dc += E1000_READ_REG(hw, E1000_DC);
+       stats->rlec += E1000_READ_REG(hw, E1000_RLEC);
+       stats->xonrxc += E1000_READ_REG(hw, E1000_XONRXC);
+       stats->xontxc += E1000_READ_REG(hw, E1000_XONTXC);
+       /*
+       ** For watchdog management we need to know if we have been
+       ** paused during the last interval, so capture that here.
+       */
+       pause_frames = E1000_READ_REG(hw, E1000_XOFFRXC);
+       stats->xoffrxc += pause_frames;
+       stats->xofftxc += E1000_READ_REG(hw, E1000_XOFFTXC);
+       stats->fcruc += E1000_READ_REG(hw, E1000_FCRUC);
+       stats->prc64 += E1000_READ_REG(hw, E1000_PRC64);
+       stats->prc127 += E1000_READ_REG(hw, E1000_PRC127);
+       stats->prc255 += E1000_READ_REG(hw, E1000_PRC255);
+       stats->prc511 += E1000_READ_REG(hw, E1000_PRC511);
+       stats->prc1023 += E1000_READ_REG(hw, E1000_PRC1023);
+       stats->prc1522 += E1000_READ_REG(hw, E1000_PRC1522);
+       stats->gprc += E1000_READ_REG(hw, E1000_GPRC);
+       stats->bprc += E1000_READ_REG(hw, E1000_BPRC);
+       stats->mprc += E1000_READ_REG(hw, E1000_MPRC);
+       stats->gptc += E1000_READ_REG(hw, E1000_GPTC);
+
+       /* For the 64-bit byte counters the low dword must be read first. */
+       /* Both registers clear on the read of the high dword */
+
+       stats->gorc += E1000_READ_REG(hw, E1000_GORCL);
+       stats->gorc += ((uint64_t)E1000_READ_REG(hw, E1000_GORCH) << 32);
+       stats->gotc += E1000_READ_REG(hw, E1000_GOTCL);
+       stats->gotc += ((uint64_t)E1000_READ_REG(hw, E1000_GOTCH) << 32);
+
+       stats->rnbc += E1000_READ_REG(hw, E1000_RNBC);
+       stats->ruc += E1000_READ_REG(hw, E1000_RUC);
+       stats->rfc += E1000_READ_REG(hw, E1000_RFC);
+       stats->roc += E1000_READ_REG(hw, E1000_ROC);
+       stats->rjc += E1000_READ_REG(hw, E1000_RJC);
+
+       stats->tor += E1000_READ_REG(hw, E1000_TORH);
+       stats->tot += E1000_READ_REG(hw, E1000_TOTH);
+
+       stats->tpr += E1000_READ_REG(hw, E1000_TPR);
+       stats->tpt += E1000_READ_REG(hw, E1000_TPT);
+       stats->ptc64 += E1000_READ_REG(hw, E1000_PTC64);
+       stats->ptc127 += E1000_READ_REG(hw, E1000_PTC127);
+       stats->ptc255 += E1000_READ_REG(hw, E1000_PTC255);
+       stats->ptc511 += E1000_READ_REG(hw, E1000_PTC511);
+       stats->ptc1023 += E1000_READ_REG(hw, E1000_PTC1023);
+       stats->ptc1522 += E1000_READ_REG(hw, E1000_PTC1522);
+       stats->mptc += E1000_READ_REG(hw, E1000_MPTC);
+       stats->bptc += E1000_READ_REG(hw, E1000_BPTC);
+
+       /* Interrupt Counts */
+
+       stats->iac += E1000_READ_REG(hw, E1000_IAC);
+       stats->icrxptc += E1000_READ_REG(hw, E1000_ICRXPTC);
+       stats->icrxatc += E1000_READ_REG(hw, E1000_ICRXATC);
+       stats->ictxptc += E1000_READ_REG(hw, E1000_ICTXPTC);
+       stats->ictxatc += E1000_READ_REG(hw, E1000_ICTXATC);
+       stats->ictxqec += E1000_READ_REG(hw, E1000_ICTXQEC);
+       stats->ictxqmtc += E1000_READ_REG(hw, E1000_ICTXQMTC);
+       stats->icrxdmtc += E1000_READ_REG(hw, E1000_ICRXDMTC);
+       stats->icrxoc += E1000_READ_REG(hw, E1000_ICRXOC);
+
+       /* Host to Card Statistics */
+
+       stats->cbtmpc += E1000_READ_REG(hw, E1000_CBTMPC);
+       stats->htdpmc += E1000_READ_REG(hw, E1000_HTDPMC);
+       stats->cbrdpc += E1000_READ_REG(hw, E1000_CBRDPC);
+       stats->cbrmpc += E1000_READ_REG(hw, E1000_CBRMPC);
+       stats->rpthc += E1000_READ_REG(hw, E1000_RPTHC);
+       stats->hgptc += E1000_READ_REG(hw, E1000_HGPTC);
+       stats->htcbdpc += E1000_READ_REG(hw, E1000_HTCBDPC);
+       stats->hgorc += E1000_READ_REG(hw, E1000_HGORCL);
+       stats->hgorc += ((uint64_t)E1000_READ_REG(hw, E1000_HGORCH) << 32);
+       stats->hgotc += E1000_READ_REG(hw, E1000_HGOTCL);
+       stats->hgotc += ((uint64_t)E1000_READ_REG(hw, E1000_HGOTCH) << 32);
+       stats->lenerrs += E1000_READ_REG(hw, E1000_LENERRS);
+       stats->scvpc += E1000_READ_REG(hw, E1000_SCVPC);
+       stats->hrmpc += E1000_READ_REG(hw, E1000_HRMPC);
+
+       stats->algnerrc += E1000_READ_REG(hw, E1000_ALGNERRC);
+       stats->rxerrc += E1000_READ_REG(hw, E1000_RXERRC);
+       stats->tncrs += E1000_READ_REG(hw, E1000_TNCRS);
+       stats->cexterr += E1000_READ_REG(hw, E1000_CEXTERR);
+       stats->tsctc += E1000_READ_REG(hw, E1000_TSCTC);
+       stats->tsctfc += E1000_READ_REG(hw, E1000_TSCTFC);
+
+       if (rte_stats == NULL)
+               return;
+
+       /* Rx Errors */
+       rte_stats->ierrors = stats->rxerrc + stats->crcerrs + stats->algnerrc +
+           stats->ruc + stats->roc + stats->mpc + stats->cexterr;
+
+       /* Tx Errors */
+       rte_stats->oerrors = stats->ecol + stats->latecol;
+
+       rte_stats->ipackets = stats->gprc;
+       rte_stats->opackets = stats->gptc;
+       rte_stats->ibytes   = stats->gorc;
+       rte_stats->obytes   = stats->gotc;
+}
+
+static void
+eth_igb_stats_reset(struct rte_eth_dev *dev)
+{
+       struct e1000_hw_stats *hw_stats =
+                       E1000_DEV_PRIVATE_TO_STATS(dev->data->dev_private);
+
+       /* HW registers are cleared on read */
+       eth_igb_stats_get(dev, NULL);
+
+       /* Reset software totals */
+       memset(hw_stats, 0, sizeof(*hw_stats));
+}
+
+static void
+eth_igb_infos_get(struct rte_eth_dev *dev,
+                   struct rte_eth_dev_info *dev_info)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
+       dev_info->max_rx_pktlen  = 0x3FFF; /* See RLPML register. */
+       dev_info->max_mac_addrs = hw->mac.rar_entry_count;
+
+       switch (hw->mac.type) {
+       case e1000_82575:
+               dev_info->max_rx_queues = 4;
+               dev_info->max_tx_queues = 4;
+               break;
+
+       case e1000_82576:
+               dev_info->max_rx_queues = 16;
+               dev_info->max_tx_queues = 16;
+               break;
+
+       case e1000_82580:
+               dev_info->max_rx_queues = 8;
+               dev_info->max_tx_queues = 8;
+               break;
+
+       case e1000_i350:
+               dev_info->max_rx_queues = 8;
+               dev_info->max_tx_queues = 8;
+               break;
+
+       default:
+               /* Should not happen */
+               dev_info->max_rx_queues = 0;
+               dev_info->max_tx_queues = 0;
+       }
+}
+
+/* return 0 means link status changed, -1 means not changed */
+static int
+eth_igb_link_update(struct rte_eth_dev *dev, int wait_to_complete)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct rte_eth_link link, old;
+       int link_check, count;
+
+       link_check = 0;
+       hw->mac.get_link_status = 1;
+
+       /* possible wait-to-complete in up to 9 seconds */
+       for (count = 0; count < IGB_LINK_UPDATE_CHECK_TIMEOUT; count ++) {
+               /* Read the real link status */
+               switch (hw->phy.media_type) {
+               case e1000_media_type_copper:
+                       /* Do the work to read phy */
+                       e1000_check_for_link(hw);
+                       link_check = !hw->mac.get_link_status;
+                       break;
+
+               case e1000_media_type_fiber:
+                       e1000_check_for_link(hw);
+                       link_check = (E1000_READ_REG(hw, E1000_STATUS) &
+                                     E1000_STATUS_LU);
+                       break;
+
+               case e1000_media_type_internal_serdes:
+                       e1000_check_for_link(hw);
+                       link_check = hw->mac.serdes_has_link;
+                       break;
+
+               default:
+               case e1000_media_type_unknown:
+                       break;
+               }
+               if (link_check || wait_to_complete == 0)
+                       break;
+               rte_delay_ms(IGB_LINK_UPDATE_CHECK_INTERVAL);
+       }
+       memset(&link, 0, sizeof(link));
+       rte_igb_dev_atomic_read_link_status(dev, &link);
+       old = link;
+
+       /* Now we check if a transition has happened */
+       if (link_check) {
+               hw->mac.ops.get_link_up_info(hw, &link.link_speed,
+                                         &link.link_duplex);
+               link.link_status = 1;
+       } else if (!link_check) {
+               link.link_speed = 0;
+               link.link_duplex = 0;
+               link.link_status = 0;
+       }
+       rte_igb_dev_atomic_write_link_status(dev, &link);
+
+       /* not changed */
+       if (old.link_status == link.link_status)
+               return -1;
+
+       /* changed */
+       return 0;
+}
+
+/*
+ * igb_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
+ * For ASF and Pass Through versions of f/w this means
+ * that the driver is loaded.
+ */
+static void
+igb_hw_control_acquire(struct e1000_hw *hw)
+{
+       uint32_t ctrl_ext;
+
+       /* Let firmware know the driver has taken over */
+       ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
+       E1000_WRITE_REG(hw, E1000_CTRL_EXT, ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
+}
+
+/*
+ * igb_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
+ * For ASF and Pass Through versions of f/w this means that the
+ * driver is no longer loaded.
+ */
+static void
+igb_hw_control_release(struct e1000_hw *hw)
+{
+       uint32_t ctrl_ext;
+
+       /* Let firmware taken over control of h/w */
+       ctrl_ext = E1000_READ_REG(hw, E1000_CTRL_EXT);
+       E1000_WRITE_REG(hw, E1000_CTRL_EXT,
+                       ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
+}
+
+/*
+ * Bit of a misnomer, what this really means is
+ * to enable OS management of the system... aka
+ * to disable special hardware management features.
+ */
+static void
+igb_init_manageability(struct e1000_hw *hw)
+{
+       if (e1000_enable_mng_pass_thru(hw)) {
+               uint32_t manc2h = E1000_READ_REG(hw, E1000_MANC2H);
+               uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
+
+               /* disable hardware interception of ARP */
+               manc &= ~(E1000_MANC_ARP_EN);
+
+               /* enable receiving management packets to the host */
+               manc |= E1000_MANC_EN_MNG2HOST;
+               manc2h |= 1 << 5;  /* Mng Port 623 */
+               manc2h |= 1 << 6;  /* Mng Port 664 */
+               E1000_WRITE_REG(hw, E1000_MANC2H, manc2h);
+               E1000_WRITE_REG(hw, E1000_MANC, manc);
+       }
+}
+
+static void
+igb_release_manageability(struct e1000_hw *hw)
+{
+       if (e1000_enable_mng_pass_thru(hw)) {
+               uint32_t manc = E1000_READ_REG(hw, E1000_MANC);
+
+               manc |= E1000_MANC_ARP_EN;
+               manc &= ~E1000_MANC_EN_MNG2HOST;
+
+               E1000_WRITE_REG(hw, E1000_MANC, manc);
+       }
+}
+
+static void
+eth_igb_promiscuous_enable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t rctl;
+
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+}
+
+static void
+eth_igb_promiscuous_disable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t rctl;
+
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       rctl &= (~E1000_RCTL_UPE);
+       if (dev->data->all_multicast == 1)
+               rctl |= E1000_RCTL_MPE;
+       else
+               rctl &= (~E1000_RCTL_MPE);
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+}
+
+static void
+eth_igb_allmulticast_enable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t rctl;
+
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       rctl |= E1000_RCTL_MPE;
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+}
+
+static void
+eth_igb_allmulticast_disable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t rctl;
+
+       if (dev->data->promiscuous == 1)
+               return; /* must remain in all_multicast mode */
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       rctl &= (~E1000_RCTL_MPE);
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+}
+
+static void
+eth_igb_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vfta * shadow_vfta =
+               E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
+       uint32_t vfta;
+       uint32_t vid_idx;
+       uint32_t vid_bit;
+
+       vid_idx = (uint32_t) ((vlan_id >> E1000_VFTA_ENTRY_SHIFT) &
+                             E1000_VFTA_ENTRY_MASK);
+       vid_bit = (uint32_t) (1 << (vlan_id & E1000_VFTA_ENTRY_BIT_SHIFT_MASK));
+       vfta = E1000_READ_REG_ARRAY(hw, E1000_VFTA, vid_idx);
+       if (on)
+               vfta |= vid_bit;
+       else
+               vfta &= ~vid_bit;
+       E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, vid_idx, vfta);
+
+       /* update local VFTA copy */
+       shadow_vfta->vfta[vid_idx] = vfta;
+}
+
+static void
+igb_vlan_hw_support_enable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_vfta * shadow_vfta =
+               E1000_DEV_PRIVATE_TO_VFTA(dev->data->dev_private);
+       uint32_t reg;
+       int i;
+
+       /* VLAN Mode Enable */
+       reg = E1000_READ_REG(hw, E1000_CTRL);
+       reg |= E1000_CTRL_VME;
+       E1000_WRITE_REG(hw, E1000_CTRL, reg);
+
+       /* Filter Table Enable */
+       reg = E1000_READ_REG(hw, E1000_RCTL);
+       reg &= ~E1000_RCTL_CFIEN;
+       reg |= E1000_RCTL_VFE;
+       E1000_WRITE_REG(hw, E1000_RCTL, reg);
+
+       /* Update maximum frame size */
+       reg = E1000_READ_REG(hw, E1000_RLPML);
+       reg += VLAN_TAG_SIZE;
+       E1000_WRITE_REG(hw, E1000_RLPML, reg);
+
+       /* restore VFTA table */
+       for (i = 0; i < E1000_VFTA_SIZE; i++)
+               E1000_WRITE_REG_ARRAY(hw, E1000_VFTA, i, shadow_vfta->vfta[i]);
+}
+
+static void
+igb_vlan_hw_support_disable(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       uint32_t reg;
+
+       /* VLAN Mode disable */
+       reg = E1000_READ_REG(hw, E1000_CTRL);
+       reg &= ~E1000_CTRL_VME;
+       E1000_WRITE_REG(hw, E1000_CTRL, reg);
+}
+
+static void
+igb_intr_disable(struct e1000_hw *hw)
+{
+       E1000_WRITE_REG(hw, E1000_IMC, ~0);
+       E1000_WRITE_FLUSH(hw);
+}
+
+/**
+ * It enables the interrupt mask and then enable the interrupt.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+eth_igb_interrupt_setup(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       E1000_WRITE_REG(hw, E1000_IMS, E1000_ICR_LSC);
+       E1000_WRITE_FLUSH(hw);
+       rte_intr_enable(&(dev->pci_dev->intr_handle));
+
+       return 0;
+}
+
+/*
+ * It reads ICR and gets interrupt causes, check it and set a bit flag
+ * to update link status.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+eth_igb_interrupt_get_status(struct rte_eth_dev *dev)
+{
+       uint32_t icr;
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_interrupt *intr =
+               E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
+
+       /* read-on-clear nic registers here */
+       icr = E1000_READ_REG(hw, E1000_ICR);
+       if (icr & E1000_ICR_LSC) {
+               intr->flags |= E1000_FLAG_NEED_LINK_UPDATE;
+       }
+
+       return 0;
+}
+
+/*
+ * It executes link_update after knowing an interrupt is prsent.
+ *
+ * @param dev
+ *  Pointer to struct rte_eth_dev.
+ *
+ * @return
+ *  - On success, zero.
+ *  - On failure, a negative value.
+ */
+static int
+eth_igb_interrupt_action(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw =
+               E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct e1000_interrupt *intr =
+               E1000_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
+       uint32_t tctl, rctl;
+       struct rte_eth_link link;
+       int ret;
+
+       if (!(intr->flags & E1000_FLAG_NEED_LINK_UPDATE))
+               return -1;
+
+       intr->flags &= ~E1000_FLAG_NEED_LINK_UPDATE;
+       rte_intr_enable(&(dev->pci_dev->intr_handle));
+
+       /* set get_link_status to check register later */
+       hw->mac.get_link_status = 1;
+       ret = eth_igb_link_update(dev, 0);
+
+       /* check if link has changed */
+       if (ret < 0)
+               return 0;
+
+       memset(&link, 0, sizeof(link));
+       rte_igb_dev_atomic_read_link_status(dev, &link);
+       if (link.link_status) {
+               PMD_INIT_LOG(INFO,
+                       " Port %d: Link Up - speed %u Mbps - %s\n",
+                       dev->data->port_id, (unsigned)link.link_speed,
+                       link.link_duplex == ETH_LINK_FULL_DUPLEX ?
+                               "full-duplex" : "half-duplex");
+       } else {
+               PMD_INIT_LOG(INFO, " Port %d: Link Down\n",
+                                       dev->data->port_id);
+       }
+       PMD_INIT_LOG(INFO, "PCI Address: %04d:%02d:%02d:%d",
+                               dev->pci_dev->addr.domain,
+                               dev->pci_dev->addr.bus,
+                               dev->pci_dev->addr.devid,
+                               dev->pci_dev->addr.function);
+       tctl = E1000_READ_REG(hw, E1000_TCTL);
+       rctl = E1000_READ_REG(hw, E1000_RCTL);
+       if (link.link_status) {
+               /* enable Tx/Rx */
+               tctl |= E1000_TCTL_EN;
+               rctl |= E1000_RCTL_EN;
+       } else {
+               /* disable Tx/Rx */
+               tctl &= ~E1000_TCTL_EN;
+               rctl &= ~E1000_RCTL_EN;
+       }
+       E1000_WRITE_REG(hw, E1000_TCTL, tctl);
+       E1000_WRITE_REG(hw, E1000_RCTL, rctl);
+       E1000_WRITE_FLUSH(hw);
+
+       return 0;
+}
+
+/**
+ * Interrupt handler which shall be registered at first.
+ *
+ * @param handle
+ *  Pointer to interrupt handle.
+ * @param param
+ *  The address of parameter (struct rte_eth_dev *) regsitered before.
+ *
+ * @return
+ *  void
+ */
+static void
+eth_igb_interrupt_handler(struct rte_intr_handle *handle, void *param)
+{
+       struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
+
+       eth_igb_interrupt_get_status(dev);
+       eth_igb_interrupt_action(dev);
+       _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC);
+}
+
+static int
+eth_igb_led_on(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw;
+
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       return (e1000_led_on(hw) == E1000_SUCCESS ? 0 : -ENOTSUP);
+}
+
+static int
+eth_igb_led_off(struct rte_eth_dev *dev)
+{
+       struct e1000_hw *hw;
+
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       return (e1000_led_off(hw) == E1000_SUCCESS ? 0 : -ENOTSUP);
+}
+
+static int
+eth_igb_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+       struct e1000_hw *hw;
+       int err;
+       enum e1000_fc_mode rte_fcmode_2_e1000_fcmode[] = {
+               e1000_fc_none,
+               e1000_fc_rx_pause,
+               e1000_fc_tx_pause,
+               e1000_fc_full
+       };
+       uint32_t rx_buf_size;
+       uint32_t max_high_water;
+
+       hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       rx_buf_size = igb_get_rx_buffer_size(hw);
+       PMD_INIT_LOG(DEBUG, "Rx packet buffer size = 0x%x \n", rx_buf_size);
+
+       /* At least reserve one Ethernet frame for watermark */
+       max_high_water = rx_buf_size - ETHER_MAX_LEN;
+       if ((fc_conf->high_water > max_high_water) ||
+               (fc_conf->high_water < fc_conf->low_water)) {
+               PMD_INIT_LOG(ERR, "e1000 incorrect high/low water value \n");
+               PMD_INIT_LOG(ERR, "high water must <=  0x%x \n", max_high_water);
+               return (-EINVAL);
+       }
+
+       hw->fc.requested_mode = rte_fcmode_2_e1000_fcmode[fc_conf->mode];
+       hw->fc.pause_time     = fc_conf->pause_time;
+       hw->fc.high_water     = fc_conf->high_water;
+       hw->fc.low_water      = fc_conf->low_water;
+       hw->fc.send_xon       = fc_conf->send_xon;
+
+       err = e1000_setup_link_generic(hw);
+       if (err == E1000_SUCCESS) {
+               return 0;
+       }
+
+       PMD_INIT_LOG(ERR, "e1000_setup_link_generic = 0x%x \n", err);
+       return (-EIO);
+}
+
+static void
+eth_igb_rar_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr,
+               uint32_t index, __rte_unused uint32_t pool)
+{
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       e1000_rar_set(hw, mac_addr->addr_bytes, index);
+}
+
+static void
+eth_igb_rar_clear(struct rte_eth_dev *dev, uint32_t index)
+{
+       uint8_t addr[ETHER_ADDR_LEN];
+       struct e1000_hw *hw = E1000_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       memset(addr, 0, sizeof(addr));
+
+       e1000_rar_set(hw, addr, index);
+}