drivers/net: check process type in close operation
[dpdk.git] / drivers / net / igc / igc_ethdev.c
index 25998db..7f5066d 100644 (file)
@@ -5,14 +5,18 @@
 #include <stdint.h>
 #include <string.h>
 
+#include <rte_string_fns.h>
 #include <rte_pci.h>
 #include <rte_bus_pci.h>
 #include <rte_ethdev_driver.h>
 #include <rte_ethdev_pci.h>
 #include <rte_malloc.h>
+#include <rte_alarm.h>
 
 #include "igc_logs.h"
 #include "igc_txrx.h"
+#include "igc_filter.h"
+#include "igc_flow.h"
 
 #define IGC_INTEL_VENDOR_ID            0x8086
 
 /* External VLAN Enable bit mask */
 #define IGC_CTRL_EXT_EXT_VLAN          (1u << 26)
 
+/* Speed select */
+#define IGC_CTRL_SPEED_MASK            (7u << 8)
+#define IGC_CTRL_SPEED_2500            (6u << 8)
+
+/* External VLAN Ether Type bit mask and shift */
+#define IGC_VET_EXT                    0xFFFF0000
+#define IGC_VET_EXT_SHIFT              16
+
+/* Force EEE Auto-negotiation */
+#define IGC_EEER_EEE_FRC_AN            (1u << 28)
+
+/* Per Queue Good Packets Received Count */
+#define IGC_PQGPRC(idx)                (0x10010 + 0x100 * (idx))
+/* Per Queue Good Octets Received Count */
+#define IGC_PQGORC(idx)                (0x10018 + 0x100 * (idx))
+/* Per Queue Good Octets Transmitted Count */
+#define IGC_PQGOTC(idx)                (0x10034 + 0x100 * (idx))
+/* Per Queue Multicast Packets Received Count */
+#define IGC_PQMPRC(idx)                (0x10038 + 0x100 * (idx))
+/* Transmit Queue Drop Packet Count */
+#define IGC_TQDPC(idx)         (0xe030 + 0x40 * (idx))
+
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+#define U32_0_IN_U64           0       /* lower bytes of u64 */
+#define U32_1_IN_U64           1       /* higher bytes of u64 */
+#else
+#define U32_0_IN_U64           1
+#define U32_1_IN_U64           0
+#endif
+
+#define IGC_ALARM_INTERVAL     8000000u
+/* us, about 13.6s some per-queue registers will wrap around back to 0. */
+
 static const struct rte_eth_desc_lim rx_desc_lim = {
        .nb_max = IGC_MAX_RXD,
        .nb_min = IGC_MIN_RXD,
@@ -70,13 +107,83 @@ static const struct rte_pci_id pci_id_igc_map[] = {
        { .vendor_id = 0, /* sentinel */ },
 };
 
+/* store statistics names and its offset in stats structure */
+struct rte_igc_xstats_name_off {
+       char name[RTE_ETH_XSTATS_NAME_SIZE];
+       unsigned int offset;
+};
+
+static const struct rte_igc_xstats_name_off rte_igc_stats_strings[] = {
+       {"rx_crc_errors", offsetof(struct igc_hw_stats, crcerrs)},
+       {"rx_align_errors", offsetof(struct igc_hw_stats, algnerrc)},
+       {"rx_errors", offsetof(struct igc_hw_stats, rxerrc)},
+       {"rx_missed_packets", offsetof(struct igc_hw_stats, mpc)},
+       {"tx_single_collision_packets", offsetof(struct igc_hw_stats, scc)},
+       {"tx_multiple_collision_packets", offsetof(struct igc_hw_stats, mcc)},
+       {"tx_excessive_collision_packets", offsetof(struct igc_hw_stats,
+               ecol)},
+       {"tx_late_collisions", offsetof(struct igc_hw_stats, latecol)},
+       {"tx_total_collisions", offsetof(struct igc_hw_stats, colc)},
+       {"tx_deferred_packets", offsetof(struct igc_hw_stats, dc)},
+       {"tx_no_carrier_sense_packets", offsetof(struct igc_hw_stats, tncrs)},
+       {"tx_discarded_packets", offsetof(struct igc_hw_stats, htdpmc)},
+       {"rx_length_errors", offsetof(struct igc_hw_stats, rlec)},
+       {"rx_xon_packets", offsetof(struct igc_hw_stats, xonrxc)},
+       {"tx_xon_packets", offsetof(struct igc_hw_stats, xontxc)},
+       {"rx_xoff_packets", offsetof(struct igc_hw_stats, xoffrxc)},
+       {"tx_xoff_packets", offsetof(struct igc_hw_stats, xofftxc)},
+       {"rx_flow_control_unsupported_packets", offsetof(struct igc_hw_stats,
+               fcruc)},
+       {"rx_size_64_packets", offsetof(struct igc_hw_stats, prc64)},
+       {"rx_size_65_to_127_packets", offsetof(struct igc_hw_stats, prc127)},
+       {"rx_size_128_to_255_packets", offsetof(struct igc_hw_stats, prc255)},
+       {"rx_size_256_to_511_packets", offsetof(struct igc_hw_stats, prc511)},
+       {"rx_size_512_to_1023_packets", offsetof(struct igc_hw_stats,
+               prc1023)},
+       {"rx_size_1024_to_max_packets", offsetof(struct igc_hw_stats,
+               prc1522)},
+       {"rx_broadcast_packets", offsetof(struct igc_hw_stats, bprc)},
+       {"rx_multicast_packets", offsetof(struct igc_hw_stats, mprc)},
+       {"rx_undersize_errors", offsetof(struct igc_hw_stats, ruc)},
+       {"rx_fragment_errors", offsetof(struct igc_hw_stats, rfc)},
+       {"rx_oversize_errors", offsetof(struct igc_hw_stats, roc)},
+       {"rx_jabber_errors", offsetof(struct igc_hw_stats, rjc)},
+       {"rx_no_buffers", offsetof(struct igc_hw_stats, rnbc)},
+       {"rx_management_packets", offsetof(struct igc_hw_stats, mgprc)},
+       {"rx_management_dropped", offsetof(struct igc_hw_stats, mgpdc)},
+       {"tx_management_packets", offsetof(struct igc_hw_stats, mgptc)},
+       {"rx_total_packets", offsetof(struct igc_hw_stats, tpr)},
+       {"tx_total_packets", offsetof(struct igc_hw_stats, tpt)},
+       {"rx_total_bytes", offsetof(struct igc_hw_stats, tor)},
+       {"tx_total_bytes", offsetof(struct igc_hw_stats, tot)},
+       {"tx_size_64_packets", offsetof(struct igc_hw_stats, ptc64)},
+       {"tx_size_65_to_127_packets", offsetof(struct igc_hw_stats, ptc127)},
+       {"tx_size_128_to_255_packets", offsetof(struct igc_hw_stats, ptc255)},
+       {"tx_size_256_to_511_packets", offsetof(struct igc_hw_stats, ptc511)},
+       {"tx_size_512_to_1023_packets", offsetof(struct igc_hw_stats,
+               ptc1023)},
+       {"tx_size_1023_to_max_packets", offsetof(struct igc_hw_stats,
+               ptc1522)},
+       {"tx_multicast_packets", offsetof(struct igc_hw_stats, mptc)},
+       {"tx_broadcast_packets", offsetof(struct igc_hw_stats, bptc)},
+       {"tx_tso_packets", offsetof(struct igc_hw_stats, tsctc)},
+       {"rx_sent_to_host_packets", offsetof(struct igc_hw_stats, rpthc)},
+       {"tx_sent_by_host_packets", offsetof(struct igc_hw_stats, hgptc)},
+       {"interrupt_assert_count", offsetof(struct igc_hw_stats, iac)},
+       {"rx_descriptor_lower_threshold",
+               offsetof(struct igc_hw_stats, icrxdmtc)},
+};
+
+#define IGC_NB_XSTATS (sizeof(rte_igc_stats_strings) / \
+               sizeof(rte_igc_stats_strings[0]))
+
 static int eth_igc_configure(struct rte_eth_dev *dev);
 static int eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete);
 static void eth_igc_stop(struct rte_eth_dev *dev);
 static int eth_igc_start(struct rte_eth_dev *dev);
 static int eth_igc_set_link_up(struct rte_eth_dev *dev);
 static int eth_igc_set_link_down(struct rte_eth_dev *dev);
-static void eth_igc_close(struct rte_eth_dev *dev);
+static int eth_igc_close(struct rte_eth_dev *dev);
 static int eth_igc_reset(struct rte_eth_dev *dev);
 static int eth_igc_promiscuous_enable(struct rte_eth_dev *dev);
 static int eth_igc_promiscuous_disable(struct rte_eth_dev *dev);
@@ -98,6 +205,46 @@ static int eth_igc_set_mc_addr_list(struct rte_eth_dev *dev,
 static int eth_igc_allmulticast_enable(struct rte_eth_dev *dev);
 static int eth_igc_allmulticast_disable(struct rte_eth_dev *dev);
 static int eth_igc_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
+static int eth_igc_stats_get(struct rte_eth_dev *dev,
+                       struct rte_eth_stats *rte_stats);
+static int eth_igc_xstats_get(struct rte_eth_dev *dev,
+                       struct rte_eth_xstat *xstats, unsigned int n);
+static int eth_igc_xstats_get_by_id(struct rte_eth_dev *dev,
+                               const uint64_t *ids,
+                               uint64_t *values, unsigned int n);
+static int eth_igc_xstats_get_names(struct rte_eth_dev *dev,
+                               struct rte_eth_xstat_name *xstats_names,
+                               unsigned int size);
+static int eth_igc_xstats_get_names_by_id(struct rte_eth_dev *dev,
+               struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
+               unsigned int limit);
+static int eth_igc_xstats_reset(struct rte_eth_dev *dev);
+static int
+eth_igc_queue_stats_mapping_set(struct rte_eth_dev *dev,
+       uint16_t queue_id, uint8_t stat_idx, uint8_t is_rx);
+static int
+eth_igc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
+static int
+eth_igc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
+static int
+eth_igc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
+static int
+eth_igc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
+static int eth_igc_rss_reta_update(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_reta_entry64 *reta_conf,
+                       uint16_t reta_size);
+static int eth_igc_rss_reta_query(struct rte_eth_dev *dev,
+                      struct rte_eth_rss_reta_entry64 *reta_conf,
+                      uint16_t reta_size);
+static int eth_igc_rss_hash_update(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_conf *rss_conf);
+static int eth_igc_rss_hash_conf_get(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_conf *rss_conf);
+static int
+eth_igc_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
+static int eth_igc_vlan_offload_set(struct rte_eth_dev *dev, int mask);
+static int eth_igc_vlan_tpid_set(struct rte_eth_dev *dev,
+                     enum rte_vlan_type vlan_type, uint16_t tpid);
 
 static const struct eth_dev_ops eth_igc_ops = {
        .dev_configure          = eth_igc_configure,
@@ -125,15 +272,32 @@ static const struct eth_dev_ops eth_igc_ops = {
 
        .rx_queue_setup         = eth_igc_rx_queue_setup,
        .rx_queue_release       = eth_igc_rx_queue_release,
-       .rx_queue_count         = eth_igc_rx_queue_count,
-       .rx_descriptor_done     = eth_igc_rx_descriptor_done,
-       .rx_descriptor_status   = eth_igc_rx_descriptor_status,
-       .tx_descriptor_status   = eth_igc_tx_descriptor_status,
        .tx_queue_setup         = eth_igc_tx_queue_setup,
        .tx_queue_release       = eth_igc_tx_queue_release,
        .tx_done_cleanup        = eth_igc_tx_done_cleanup,
        .rxq_info_get           = eth_igc_rxq_info_get,
        .txq_info_get           = eth_igc_txq_info_get,
+       .stats_get              = eth_igc_stats_get,
+       .xstats_get             = eth_igc_xstats_get,
+       .xstats_get_by_id       = eth_igc_xstats_get_by_id,
+       .xstats_get_names_by_id = eth_igc_xstats_get_names_by_id,
+       .xstats_get_names       = eth_igc_xstats_get_names,
+       .stats_reset            = eth_igc_xstats_reset,
+       .xstats_reset           = eth_igc_xstats_reset,
+       .queue_stats_mapping_set = eth_igc_queue_stats_mapping_set,
+       .rx_queue_intr_enable   = eth_igc_rx_queue_intr_enable,
+       .rx_queue_intr_disable  = eth_igc_rx_queue_intr_disable,
+       .flow_ctrl_get          = eth_igc_flow_ctrl_get,
+       .flow_ctrl_set          = eth_igc_flow_ctrl_set,
+       .reta_update            = eth_igc_rss_reta_update,
+       .reta_query             = eth_igc_rss_reta_query,
+       .rss_hash_update        = eth_igc_rss_hash_update,
+       .rss_hash_conf_get      = eth_igc_rss_hash_conf_get,
+       .vlan_filter_set        = eth_igc_vlan_filter_set,
+       .vlan_offload_set       = eth_igc_vlan_offload_set,
+       .vlan_tpid_set          = eth_igc_vlan_tpid_set,
+       .vlan_strip_queue_set   = eth_igc_vlan_strip_queue_set,
+       .filter_ctrl            = eth_igc_filter_ctrl,
 };
 
 /*
@@ -376,8 +540,7 @@ eth_igc_interrupt_action(struct rte_eth_dev *dev)
                                pci_dev->addr.bus,
                                pci_dev->addr.devid,
                                pci_dev->addr.function);
-               _rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC,
-                               NULL);
+               rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
        }
 }
 
@@ -398,6 +561,22 @@ eth_igc_interrupt_handler(void *param)
        eth_igc_interrupt_action(dev);
 }
 
+static void igc_read_queue_stats_register(struct rte_eth_dev *dev);
+
+/*
+ * Update the queue status every IGC_ALARM_INTERVAL time.
+ * @param
+ *  The address of parameter (struct rte_eth_dev *) registered before.
+ */
+static void
+igc_update_queue_stats_handler(void *param)
+{
+       struct rte_eth_dev *dev = param;
+       igc_read_queue_stats_register(dev);
+       rte_eal_alarm_set(IGC_ALARM_INTERVAL,
+                       igc_update_queue_stats_handler, dev);
+}
+
 /*
  * rx,tx enable/disable
  */
@@ -451,6 +630,8 @@ eth_igc_stop(struct rte_eth_dev *dev)
 
        igc_intr_other_disable(dev);
 
+       rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
+
        /* disable intr eventfd mapping */
        rte_intr_disable(intr_handle);
 
@@ -459,6 +640,9 @@ eth_igc_stop(struct rte_eth_dev *dev)
        /* disable all wake up */
        IGC_WRITE_REG(hw, IGC_WUC, 0);
 
+       /* disable checking EEE operation in MAC loopback mode */
+       igc_read_reg_check_clear_bits(hw, IGC_EEER, IGC_EEER_EEE_FRC_AN);
+
        /* Set bit for Go Link disconnect */
        igc_read_reg_check_set_bits(hw, IGC_82580_PHY_POWER_MGMT,
                        IGC_82580_PM_GO_LINKD);
@@ -480,6 +664,56 @@ eth_igc_stop(struct rte_eth_dev *dev)
 
        /* Clean datapath event and queue/vec mapping */
        rte_intr_efd_disable(intr_handle);
+       if (intr_handle->intr_vec != NULL) {
+               rte_free(intr_handle->intr_vec);
+               intr_handle->intr_vec = NULL;
+       }
+}
+
+/*
+ * write interrupt vector allocation register
+ * @hw
+ *  board private structure
+ * @queue_index
+ *  queue index, valid 0,1,2,3
+ * @tx
+ *  tx:1, rx:0
+ * @msix_vector
+ *  msix-vector, valid 0,1,2,3,4
+ */
+static void
+igc_write_ivar(struct igc_hw *hw, uint8_t queue_index,
+               bool tx, uint8_t msix_vector)
+{
+       uint8_t offset = 0;
+       uint8_t reg_index = queue_index >> 1;
+       uint32_t val;
+
+       /*
+        * IVAR(0)
+        * bit31...24   bit23...16      bit15...8       bit7...0
+        * TX1          RX1             TX0             RX0
+        *
+        * IVAR(1)
+        * bit31...24   bit23...16      bit15...8       bit7...0
+        * TX3          RX3             TX2             RX2
+        */
+
+       if (tx)
+               offset = 8;
+
+       if (queue_index & 1)
+               offset += 16;
+
+       val = IGC_READ_REG_ARRAY(hw, IGC_IVAR0, reg_index);
+
+       /* clear bits */
+       val &= ~((uint32_t)0xFF << offset);
+
+       /* write vector and valid bit */
+       val |= (uint32_t)(msix_vector | IGC_IVAR_VALID) << offset;
+
+       IGC_WRITE_REG_ARRAY(hw, IGC_IVAR0, reg_index, val);
 }
 
 /* Sets up the hardware to generate MSI-X interrupts properly
@@ -494,20 +728,32 @@ igc_configure_msix_intr(struct rte_eth_dev *dev)
        struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 
        uint32_t intr_mask;
+       uint32_t vec = IGC_MISC_VEC_ID;
+       uint32_t base = IGC_MISC_VEC_ID;
+       uint32_t misc_shift = 0;
+       int i;
 
        /* won't configure msix register if no mapping is done
         * between intr vector and event fd
         */
-       if (!rte_intr_dp_is_en(intr_handle) ||
-               !dev->data->dev_conf.intr_conf.lsc)
+       if (!rte_intr_dp_is_en(intr_handle))
                return;
 
+       if (rte_intr_allow_others(intr_handle)) {
+               base = IGC_RX_VEC_START;
+               vec = base;
+               misc_shift = 1;
+       }
+
        /* turn on MSI-X capability first */
        IGC_WRITE_REG(hw, IGC_GPIE, IGC_GPIE_MSIX_MODE |
                                IGC_GPIE_PBA | IGC_GPIE_EIAME |
                                IGC_GPIE_NSICR);
+       intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
+               misc_shift;
 
-       intr_mask = (1u << IGC_MSIX_OTHER_INTR_VEC);
+       if (dev->data->dev_conf.intr_conf.lsc)
+               intr_mask |= (1u << IGC_MSIX_OTHER_INTR_VEC);
 
        /* enable msix auto-clear */
        igc_read_reg_check_set_bits(hw, IGC_EIAC, intr_mask);
@@ -519,6 +765,13 @@ igc_configure_msix_intr(struct rte_eth_dev *dev)
        /* enable auto-mask */
        igc_read_reg_check_set_bits(hw, IGC_EIAM, intr_mask);
 
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               igc_write_ivar(hw, i, 0, vec);
+               intr_handle->intr_vec[i] = vec;
+               if (vec < base + intr_handle->nb_efd - 1)
+                       vec++;
+       }
+
        IGC_WRITE_FLUSH(hw);
 }
 
@@ -541,6 +794,29 @@ igc_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
                intr->mask &= ~IGC_ICR_LSC;
 }
 
+/*
+ * It enables the interrupt.
+ * It will be called once only during nic initialized.
+ */
+static void
+igc_rxq_interrupt_setup(struct rte_eth_dev *dev)
+{
+       uint32_t mask;
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       int misc_shift = rte_intr_allow_others(intr_handle) ? 1 : 0;
+
+       /* won't configure msix register if no mapping is done
+        * between intr vector and event fd
+        */
+       if (!rte_intr_dp_is_en(intr_handle))
+               return;
+
+       mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) << misc_shift;
+       IGC_WRITE_REG(hw, IGC_EIMS, mask);
+}
+
 /*
  *  Get hardware rx-buffer size.
  */
@@ -661,7 +937,26 @@ eth_igc_start(struct rte_eth_dev *dev)
        }
        adapter->stopped = 0;
 
-       /* confiugre msix for rx interrupt */
+       /* check and configure queue intr-vector mapping */
+       if (rte_intr_cap_multiple(intr_handle) &&
+               dev->data->dev_conf.intr_conf.rxq) {
+               uint32_t intr_vector = dev->data->nb_rx_queues;
+               if (rte_intr_efd_enable(intr_handle, intr_vector))
+                       return -1;
+       }
+
+       if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
+               intr_handle->intr_vec = rte_zmalloc("intr_vec",
+                       dev->data->nb_rx_queues * sizeof(int), 0);
+               if (intr_handle->intr_vec == NULL) {
+                       PMD_DRV_LOG(ERR,
+                               "Failed to allocate %d rx_queues intr_vec",
+                               dev->data->nb_rx_queues);
+                       return -ENOMEM;
+               }
+       }
+
+       /* configure msix for rx interrupt */
        igc_configure_msix_intr(dev);
 
        igc_tx_init(dev);
@@ -676,6 +971,11 @@ eth_igc_start(struct rte_eth_dev *dev)
 
        igc_clear_hw_cntrs_base_generic(hw);
 
+       /* VLAN Offload Settings */
+       eth_igc_vlan_offload_set(dev,
+               ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK |
+               ETH_VLAN_EXTEND_MASK);
+
        /* Setup link speed and duplex */
        speeds = &dev->data->dev_conf.link_speeds;
        if (*speeds == ETH_LINK_SPEED_AUTONEG) {
@@ -754,12 +1054,33 @@ eth_igc_start(struct rte_eth_dev *dev)
        /* enable uio/vfio intr/eventfd mapping */
        rte_intr_enable(intr_handle);
 
+       rte_eal_alarm_set(IGC_ALARM_INTERVAL,
+                       igc_update_queue_stats_handler, dev);
+
+       /* check if rxq interrupt is enabled */
+       if (dev->data->dev_conf.intr_conf.rxq &&
+                       rte_intr_dp_is_en(intr_handle))
+               igc_rxq_interrupt_setup(dev);
+
        /* resume enabled intr since hw reset */
        igc_intr_other_enable(dev);
 
        eth_igc_rxtx_control(dev, true);
        eth_igc_link_update(dev, 0);
 
+       /* configure MAC-loopback mode */
+       if (dev->data->dev_conf.lpbk_mode == 1) {
+               uint32_t reg_val;
+
+               reg_val = IGC_READ_REG(hw, IGC_CTRL);
+               reg_val &= ~IGC_CTRL_SPEED_MASK;
+               reg_val |= IGC_CTRL_SLU | IGC_CTRL_FRCSPD |
+                       IGC_CTRL_FRCDPX | IGC_CTRL_FD | IGC_CTRL_SPEED_2500;
+               IGC_WRITE_REG(hw, IGC_CTRL, reg_val);
+
+               igc_read_reg_check_set_bits(hw, IGC_EEER, IGC_EEER_EEE_FRC_AN);
+       }
+
        return 0;
 
 error_invalid_config:
@@ -844,7 +1165,7 @@ igc_dev_free_queues(struct rte_eth_dev *dev)
        dev->data->nb_tx_queues = 0;
 }
 
-static void
+static int
 eth_igc_close(struct rte_eth_dev *dev)
 {
        struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
@@ -854,10 +1175,15 @@ eth_igc_close(struct rte_eth_dev *dev)
        int retry = 0;
 
        PMD_INIT_FUNC_TRACE();
+       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
+               return 0;
 
        if (!adapter->stopped)
                eth_igc_stop(dev);
 
+       igc_flow_flush(dev, NULL);
+       igc_clear_all_filter(dev);
+
        igc_intr_other_disable(dev);
        do {
                int ret = rte_intr_callback_unregister(intr_handle,
@@ -875,6 +1201,8 @@ eth_igc_close(struct rte_eth_dev *dev)
 
        /* Reset any pending lock */
        igc_reset_swfw_lock(hw);
+
+       return 0;
 }
 
 static void
@@ -894,10 +1222,14 @@ eth_igc_dev_init(struct rte_eth_dev *dev)
        struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
        struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
        struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
-       int error = 0;
+       int i, error = 0;
 
        PMD_INIT_FUNC_TRACE();
        dev->dev_ops = &eth_igc_ops;
+       dev->rx_descriptor_done = eth_igc_rx_descriptor_done;
+       dev->rx_queue_count = eth_igc_rx_queue_count;
+       dev->rx_descriptor_status = eth_igc_rx_descriptor_status;
+       dev->tx_descriptor_status = eth_igc_tx_descriptor_status;
 
        /*
         * for secondary processes, we don't initialize any further as primary
@@ -993,11 +1325,6 @@ eth_igc_dev_init(struct rte_eth_dev *dev)
                goto err_late;
        }
 
-       /* Pass the information to the rte_eth_dev_close() that it should also
-        * release the private port resources.
-        */
-       dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
-
        hw->mac.get_link_status = 1;
        igc->stopped = 0;
 
@@ -1019,6 +1346,14 @@ eth_igc_dev_init(struct rte_eth_dev *dev)
        /* enable support intr */
        igc_intr_other_enable(dev);
 
+       /* initiate queue status */
+       for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
+               igc->txq_stats_map[i] = -1;
+               igc->rxq_stats_map[i] = -1;
+       }
+
+       igc_flow_init(dev);
+       igc_clear_all_filter(dev);
        return 0;
 
 err_late:
@@ -1030,10 +1365,6 @@ static int
 eth_igc_dev_uninit(__rte_unused struct rte_eth_dev *eth_dev)
 {
        PMD_INIT_FUNC_TRACE();
-
-       if (rte_eal_process_type() != RTE_PROC_PRIMARY)
-               return 0;
-
        eth_igc_close(eth_dev);
        return 0;
 }
@@ -1155,6 +1486,7 @@ eth_igc_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
        dev_info->max_mac_addrs = hw->mac.rar_entry_count;
        dev_info->rx_offload_capa = IGC_RX_OFFLOAD_ALL;
        dev_info->tx_offload_capa = IGC_TX_OFFLOAD_ALL;
+       dev_info->rx_queue_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
 
        dev_info->max_rx_queues = IGC_QUEUE_PAIRS_NUM;
        dev_info->max_tx_queues = IGC_QUEUE_PAIRS_NUM;
@@ -1323,6 +1655,947 @@ eth_igc_set_mc_addr_list(struct rte_eth_dev *dev,
        return 0;
 }
 
+/*
+ * Read hardware registers
+ */
+static void
+igc_read_stats_registers(struct igc_hw *hw, struct igc_hw_stats *stats)
+{
+       int pause_frames;
+
+       uint64_t old_gprc  = stats->gprc;
+       uint64_t old_gptc  = stats->gptc;
+       uint64_t old_tpr   = stats->tpr;
+       uint64_t old_tpt   = stats->tpt;
+       uint64_t old_rpthc = stats->rpthc;
+       uint64_t old_hgptc = stats->hgptc;
+
+       stats->crcerrs += IGC_READ_REG(hw, IGC_CRCERRS);
+       stats->algnerrc += IGC_READ_REG(hw, IGC_ALGNERRC);
+       stats->rxerrc += IGC_READ_REG(hw, IGC_RXERRC);
+       stats->mpc += IGC_READ_REG(hw, IGC_MPC);
+       stats->scc += IGC_READ_REG(hw, IGC_SCC);
+       stats->ecol += IGC_READ_REG(hw, IGC_ECOL);
+
+       stats->mcc += IGC_READ_REG(hw, IGC_MCC);
+       stats->latecol += IGC_READ_REG(hw, IGC_LATECOL);
+       stats->colc += IGC_READ_REG(hw, IGC_COLC);
+
+       stats->dc += IGC_READ_REG(hw, IGC_DC);
+       stats->tncrs += IGC_READ_REG(hw, IGC_TNCRS);
+       stats->htdpmc += IGC_READ_REG(hw, IGC_HTDPMC);
+       stats->rlec += IGC_READ_REG(hw, IGC_RLEC);
+       stats->xonrxc += IGC_READ_REG(hw, IGC_XONRXC);
+       stats->xontxc += IGC_READ_REG(hw, IGC_XONTXC);
+
+       /*
+        * For watchdog management we need to know if we have been
+        * paused during the last interval, so capture that here.
+        */
+       pause_frames = IGC_READ_REG(hw, IGC_XOFFRXC);
+       stats->xoffrxc += pause_frames;
+       stats->xofftxc += IGC_READ_REG(hw, IGC_XOFFTXC);
+       stats->fcruc += IGC_READ_REG(hw, IGC_FCRUC);
+       stats->prc64 += IGC_READ_REG(hw, IGC_PRC64);
+       stats->prc127 += IGC_READ_REG(hw, IGC_PRC127);
+       stats->prc255 += IGC_READ_REG(hw, IGC_PRC255);
+       stats->prc511 += IGC_READ_REG(hw, IGC_PRC511);
+       stats->prc1023 += IGC_READ_REG(hw, IGC_PRC1023);
+       stats->prc1522 += IGC_READ_REG(hw, IGC_PRC1522);
+       stats->gprc += IGC_READ_REG(hw, IGC_GPRC);
+       stats->bprc += IGC_READ_REG(hw, IGC_BPRC);
+       stats->mprc += IGC_READ_REG(hw, IGC_MPRC);
+       stats->gptc += IGC_READ_REG(hw, IGC_GPTC);
+
+       /* For the 64-bit byte counters the low dword must be read first. */
+       /* Both registers clear on the read of the high dword */
+
+       /* Workaround CRC bytes included in size, take away 4 bytes/packet */
+       stats->gorc += IGC_READ_REG(hw, IGC_GORCL);
+       stats->gorc += ((uint64_t)IGC_READ_REG(hw, IGC_GORCH) << 32);
+       stats->gorc -= (stats->gprc - old_gprc) * RTE_ETHER_CRC_LEN;
+       stats->gotc += IGC_READ_REG(hw, IGC_GOTCL);
+       stats->gotc += ((uint64_t)IGC_READ_REG(hw, IGC_GOTCH) << 32);
+       stats->gotc -= (stats->gptc - old_gptc) * RTE_ETHER_CRC_LEN;
+
+       stats->rnbc += IGC_READ_REG(hw, IGC_RNBC);
+       stats->ruc += IGC_READ_REG(hw, IGC_RUC);
+       stats->rfc += IGC_READ_REG(hw, IGC_RFC);
+       stats->roc += IGC_READ_REG(hw, IGC_ROC);
+       stats->rjc += IGC_READ_REG(hw, IGC_RJC);
+
+       stats->mgprc += IGC_READ_REG(hw, IGC_MGTPRC);
+       stats->mgpdc += IGC_READ_REG(hw, IGC_MGTPDC);
+       stats->mgptc += IGC_READ_REG(hw, IGC_MGTPTC);
+       stats->b2ospc += IGC_READ_REG(hw, IGC_B2OSPC);
+       stats->b2ogprc += IGC_READ_REG(hw, IGC_B2OGPRC);
+       stats->o2bgptc += IGC_READ_REG(hw, IGC_O2BGPTC);
+       stats->o2bspc += IGC_READ_REG(hw, IGC_O2BSPC);
+
+       stats->tpr += IGC_READ_REG(hw, IGC_TPR);
+       stats->tpt += IGC_READ_REG(hw, IGC_TPT);
+
+       stats->tor += IGC_READ_REG(hw, IGC_TORL);
+       stats->tor += ((uint64_t)IGC_READ_REG(hw, IGC_TORH) << 32);
+       stats->tor -= (stats->tpr - old_tpr) * RTE_ETHER_CRC_LEN;
+       stats->tot += IGC_READ_REG(hw, IGC_TOTL);
+       stats->tot += ((uint64_t)IGC_READ_REG(hw, IGC_TOTH) << 32);
+       stats->tot -= (stats->tpt - old_tpt) * RTE_ETHER_CRC_LEN;
+
+       stats->ptc64 += IGC_READ_REG(hw, IGC_PTC64);
+       stats->ptc127 += IGC_READ_REG(hw, IGC_PTC127);
+       stats->ptc255 += IGC_READ_REG(hw, IGC_PTC255);
+       stats->ptc511 += IGC_READ_REG(hw, IGC_PTC511);
+       stats->ptc1023 += IGC_READ_REG(hw, IGC_PTC1023);
+       stats->ptc1522 += IGC_READ_REG(hw, IGC_PTC1522);
+       stats->mptc += IGC_READ_REG(hw, IGC_MPTC);
+       stats->bptc += IGC_READ_REG(hw, IGC_BPTC);
+       stats->tsctc += IGC_READ_REG(hw, IGC_TSCTC);
+
+       stats->iac += IGC_READ_REG(hw, IGC_IAC);
+       stats->rpthc += IGC_READ_REG(hw, IGC_RPTHC);
+       stats->hgptc += IGC_READ_REG(hw, IGC_HGPTC);
+       stats->icrxdmtc += IGC_READ_REG(hw, IGC_ICRXDMTC);
+
+       /* Host to Card Statistics */
+       stats->hgorc += IGC_READ_REG(hw, IGC_HGORCL);
+       stats->hgorc += ((uint64_t)IGC_READ_REG(hw, IGC_HGORCH) << 32);
+       stats->hgorc -= (stats->rpthc - old_rpthc) * RTE_ETHER_CRC_LEN;
+       stats->hgotc += IGC_READ_REG(hw, IGC_HGOTCL);
+       stats->hgotc += ((uint64_t)IGC_READ_REG(hw, IGC_HGOTCH) << 32);
+       stats->hgotc -= (stats->hgptc - old_hgptc) * RTE_ETHER_CRC_LEN;
+       stats->lenerrs += IGC_READ_REG(hw, IGC_LENERRS);
+}
+
+/*
+ * Write 0 to all queue status registers
+ */
+static void
+igc_reset_queue_stats_register(struct igc_hw *hw)
+{
+       int i;
+
+       for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
+               IGC_WRITE_REG(hw, IGC_PQGPRC(i), 0);
+               IGC_WRITE_REG(hw, IGC_PQGPTC(i), 0);
+               IGC_WRITE_REG(hw, IGC_PQGORC(i), 0);
+               IGC_WRITE_REG(hw, IGC_PQGOTC(i), 0);
+               IGC_WRITE_REG(hw, IGC_PQMPRC(i), 0);
+               IGC_WRITE_REG(hw, IGC_RQDPC(i), 0);
+               IGC_WRITE_REG(hw, IGC_TQDPC(i), 0);
+       }
+}
+
+/*
+ * Read all hardware queue status registers
+ */
+static void
+igc_read_queue_stats_register(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_hw_queue_stats *queue_stats =
+                               IGC_DEV_PRIVATE_QUEUE_STATS(dev);
+       int i;
+
+       /*
+        * This register is not cleared on read. Furthermore, the register wraps
+        * around back to 0x00000000 on the next increment when reaching a value
+        * of 0xFFFFFFFF and then continues normal count operation.
+        */
+       for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
+               union {
+                       u64 ddword;
+                       u32 dword[2];
+               } value;
+               u32 tmp;
+
+               /*
+                * Read the register first, if the value is smaller than that
+                * previous read, that mean the register has been overflowed,
+                * then we add the high 4 bytes by 1 and replace the low 4
+                * bytes by the new value.
+                */
+               tmp = IGC_READ_REG(hw, IGC_PQGPRC(i));
+               value.ddword = queue_stats->pqgprc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->pqgprc[i] = value.ddword;
+
+               tmp = IGC_READ_REG(hw, IGC_PQGPTC(i));
+               value.ddword = queue_stats->pqgptc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->pqgptc[i] = value.ddword;
+
+               tmp = IGC_READ_REG(hw, IGC_PQGORC(i));
+               value.ddword = queue_stats->pqgorc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->pqgorc[i] = value.ddword;
+
+               tmp = IGC_READ_REG(hw, IGC_PQGOTC(i));
+               value.ddword = queue_stats->pqgotc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->pqgotc[i] = value.ddword;
+
+               tmp = IGC_READ_REG(hw, IGC_PQMPRC(i));
+               value.ddword = queue_stats->pqmprc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->pqmprc[i] = value.ddword;
+
+               tmp = IGC_READ_REG(hw, IGC_RQDPC(i));
+               value.ddword = queue_stats->rqdpc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->rqdpc[i] = value.ddword;
+
+               tmp = IGC_READ_REG(hw, IGC_TQDPC(i));
+               value.ddword = queue_stats->tqdpc[i];
+               if (value.dword[U32_0_IN_U64] > tmp)
+                       value.dword[U32_1_IN_U64]++;
+               value.dword[U32_0_IN_U64] = tmp;
+               queue_stats->tqdpc[i] = value.ddword;
+       }
+}
+
+static int
+eth_igc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
+{
+       struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_hw_stats *stats = IGC_DEV_PRIVATE_STATS(dev);
+       struct igc_hw_queue_stats *queue_stats =
+                       IGC_DEV_PRIVATE_QUEUE_STATS(dev);
+       int i;
+
+       /*
+        * Cancel status handler since it will read the queue status registers
+        */
+       rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
+
+       /* Read status register */
+       igc_read_queue_stats_register(dev);
+       igc_read_stats_registers(hw, stats);
+
+       if (rte_stats == NULL) {
+               /* Restart queue status handler */
+               rte_eal_alarm_set(IGC_ALARM_INTERVAL,
+                               igc_update_queue_stats_handler, dev);
+               return -EINVAL;
+       }
+
+       /* Rx Errors */
+       rte_stats->imissed = stats->mpc;
+       rte_stats->ierrors = stats->crcerrs +
+                       stats->rlec + stats->ruc + stats->roc +
+                       stats->rxerrc + stats->algnerrc;
+
+       /* 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;
+
+       /* Get per-queue statuses */
+       for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
+               /* GET TX queue statuses */
+               int map_id = igc->txq_stats_map[i];
+               if (map_id >= 0) {
+                       rte_stats->q_opackets[map_id] += queue_stats->pqgptc[i];
+                       rte_stats->q_obytes[map_id] += queue_stats->pqgotc[i];
+               }
+               /* Get RX queue statuses */
+               map_id = igc->rxq_stats_map[i];
+               if (map_id >= 0) {
+                       rte_stats->q_ipackets[map_id] += queue_stats->pqgprc[i];
+                       rte_stats->q_ibytes[map_id] += queue_stats->pqgorc[i];
+                       rte_stats->q_errors[map_id] += queue_stats->rqdpc[i];
+               }
+       }
+
+       /* Restart queue status handler */
+       rte_eal_alarm_set(IGC_ALARM_INTERVAL,
+                       igc_update_queue_stats_handler, dev);
+       return 0;
+}
+
+static int
+eth_igc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
+                  unsigned int n)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_hw_stats *hw_stats =
+                       IGC_DEV_PRIVATE_STATS(dev);
+       unsigned int i;
+
+       igc_read_stats_registers(hw, hw_stats);
+
+       if (n < IGC_NB_XSTATS)
+               return IGC_NB_XSTATS;
+
+       /* If this is a reset xstats is NULL, and we have cleared the
+        * registers by reading them.
+        */
+       if (!xstats)
+               return 0;
+
+       /* Extended stats */
+       for (i = 0; i < IGC_NB_XSTATS; i++) {
+               xstats[i].id = i;
+               xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
+                       rte_igc_stats_strings[i].offset);
+       }
+
+       return IGC_NB_XSTATS;
+}
+
+static int
+eth_igc_xstats_reset(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_hw_stats *hw_stats = IGC_DEV_PRIVATE_STATS(dev);
+       struct igc_hw_queue_stats *queue_stats =
+                       IGC_DEV_PRIVATE_QUEUE_STATS(dev);
+
+       /* Cancel queue status handler for avoid conflict */
+       rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
+
+       /* HW registers are cleared on read */
+       igc_reset_queue_stats_register(hw);
+       igc_read_stats_registers(hw, hw_stats);
+
+       /* Reset software totals */
+       memset(hw_stats, 0, sizeof(*hw_stats));
+       memset(queue_stats, 0, sizeof(*queue_stats));
+
+       /* Restart the queue status handler */
+       rte_eal_alarm_set(IGC_ALARM_INTERVAL, igc_update_queue_stats_handler,
+                       dev);
+
+       return 0;
+}
+
+static int
+eth_igc_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
+       struct rte_eth_xstat_name *xstats_names, unsigned int size)
+{
+       unsigned int i;
+
+       if (xstats_names == NULL)
+               return IGC_NB_XSTATS;
+
+       if (size < IGC_NB_XSTATS) {
+               PMD_DRV_LOG(ERR, "not enough buffers!");
+               return IGC_NB_XSTATS;
+       }
+
+       for (i = 0; i < IGC_NB_XSTATS; i++)
+               strlcpy(xstats_names[i].name, rte_igc_stats_strings[i].name,
+                       sizeof(xstats_names[i].name));
+
+       return IGC_NB_XSTATS;
+}
+
+static int
+eth_igc_xstats_get_names_by_id(struct rte_eth_dev *dev,
+               struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
+               unsigned int limit)
+{
+       unsigned int i;
+
+       if (!ids)
+               return eth_igc_xstats_get_names(dev, xstats_names, limit);
+
+       for (i = 0; i < limit; i++) {
+               if (ids[i] >= IGC_NB_XSTATS) {
+                       PMD_DRV_LOG(ERR, "id value isn't valid");
+                       return -EINVAL;
+               }
+               strlcpy(xstats_names[i].name,
+                       rte_igc_stats_strings[ids[i]].name,
+                       sizeof(xstats_names[i].name));
+       }
+       return limit;
+}
+
+static int
+eth_igc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
+               uint64_t *values, unsigned int n)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_hw_stats *hw_stats = IGC_DEV_PRIVATE_STATS(dev);
+       unsigned int i;
+
+       igc_read_stats_registers(hw, hw_stats);
+
+       if (!ids) {
+               if (n < IGC_NB_XSTATS)
+                       return IGC_NB_XSTATS;
+
+               /* If this is a reset xstats is NULL, and we have cleared the
+                * registers by reading them.
+                */
+               if (!values)
+                       return 0;
+
+               /* Extended stats */
+               for (i = 0; i < IGC_NB_XSTATS; i++)
+                       values[i] = *(uint64_t *)(((char *)hw_stats) +
+                                       rte_igc_stats_strings[i].offset);
+
+               return IGC_NB_XSTATS;
+
+       } else {
+               for (i = 0; i < n; i++) {
+                       if (ids[i] >= IGC_NB_XSTATS) {
+                               PMD_DRV_LOG(ERR, "id value isn't valid");
+                               return -EINVAL;
+                       }
+                       values[i] = *(uint64_t *)(((char *)hw_stats) +
+                                       rte_igc_stats_strings[ids[i]].offset);
+               }
+               return n;
+       }
+}
+
+static int
+eth_igc_queue_stats_mapping_set(struct rte_eth_dev *dev,
+               uint16_t queue_id, uint8_t stat_idx, uint8_t is_rx)
+{
+       struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
+
+       /* check queue id is valid */
+       if (queue_id >= IGC_QUEUE_PAIRS_NUM) {
+               PMD_DRV_LOG(ERR, "queue id(%u) error, max is %u",
+                       queue_id, IGC_QUEUE_PAIRS_NUM - 1);
+               return -EINVAL;
+       }
+
+       /* store the mapping status id */
+       if (is_rx)
+               igc->rxq_stats_map[queue_id] = stat_idx;
+       else
+               igc->txq_stats_map[queue_id] = stat_idx;
+
+       return 0;
+}
+
+static int
+eth_igc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       uint32_t vec = IGC_MISC_VEC_ID;
+
+       if (rte_intr_allow_others(intr_handle))
+               vec = IGC_RX_VEC_START;
+
+       uint32_t mask = 1u << (queue_id + vec);
+
+       IGC_WRITE_REG(hw, IGC_EIMC, mask);
+       IGC_WRITE_FLUSH(hw);
+
+       return 0;
+}
+
+static int
+eth_igc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
+       struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
+       uint32_t vec = IGC_MISC_VEC_ID;
+
+       if (rte_intr_allow_others(intr_handle))
+               vec = IGC_RX_VEC_START;
+
+       uint32_t mask = 1u << (queue_id + vec);
+
+       IGC_WRITE_REG(hw, IGC_EIMS, mask);
+       IGC_WRITE_FLUSH(hw);
+
+       rte_intr_enable(intr_handle);
+
+       return 0;
+}
+
+static int
+eth_igc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint32_t ctrl;
+       int tx_pause;
+       int rx_pause;
+
+       fc_conf->pause_time = hw->fc.pause_time;
+       fc_conf->high_water = hw->fc.high_water;
+       fc_conf->low_water = hw->fc.low_water;
+       fc_conf->send_xon = hw->fc.send_xon;
+       fc_conf->autoneg = hw->mac.autoneg;
+
+       /*
+        * Return rx_pause and tx_pause status according to actual setting of
+        * the TFCE and RFCE bits in the CTRL register.
+        */
+       ctrl = IGC_READ_REG(hw, IGC_CTRL);
+       if (ctrl & IGC_CTRL_TFCE)
+               tx_pause = 1;
+       else
+               tx_pause = 0;
+
+       if (ctrl & IGC_CTRL_RFCE)
+               rx_pause = 1;
+       else
+               rx_pause = 0;
+
+       if (rx_pause && tx_pause)
+               fc_conf->mode = RTE_FC_FULL;
+       else if (rx_pause)
+               fc_conf->mode = RTE_FC_RX_PAUSE;
+       else if (tx_pause)
+               fc_conf->mode = RTE_FC_TX_PAUSE;
+       else
+               fc_conf->mode = RTE_FC_NONE;
+
+       return 0;
+}
+
+static int
+eth_igc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint32_t rx_buf_size;
+       uint32_t max_high_water;
+       uint32_t rctl;
+       int err;
+
+       if (fc_conf->autoneg != hw->mac.autoneg)
+               return -ENOTSUP;
+
+       rx_buf_size = igc_get_rx_buffer_size(hw);
+       PMD_DRV_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
+
+       /* At least reserve one Ethernet frame for watermark */
+       max_high_water = rx_buf_size - RTE_ETHER_MAX_LEN;
+       if (fc_conf->high_water > max_high_water ||
+               fc_conf->high_water < fc_conf->low_water) {
+               PMD_DRV_LOG(ERR,
+                       "Incorrect high(%u)/low(%u) water value, max is %u",
+                       fc_conf->high_water, fc_conf->low_water,
+                       max_high_water);
+               return -EINVAL;
+       }
+
+       switch (fc_conf->mode) {
+       case RTE_FC_NONE:
+               hw->fc.requested_mode = igc_fc_none;
+               break;
+       case RTE_FC_RX_PAUSE:
+               hw->fc.requested_mode = igc_fc_rx_pause;
+               break;
+       case RTE_FC_TX_PAUSE:
+               hw->fc.requested_mode = igc_fc_tx_pause;
+               break;
+       case RTE_FC_FULL:
+               hw->fc.requested_mode = igc_fc_full;
+               break;
+       default:
+               PMD_DRV_LOG(ERR, "unsupported fc mode: %u", fc_conf->mode);
+               return -EINVAL;
+       }
+
+       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 = igc_setup_link_generic(hw);
+       if (err == IGC_SUCCESS) {
+               /**
+                * check if we want to forward MAC frames - driver doesn't have
+                * native capability to do that, so we'll write the registers
+                * ourselves
+                **/
+               rctl = IGC_READ_REG(hw, IGC_RCTL);
+
+               /* set or clear MFLCN.PMCF bit depending on configuration */
+               if (fc_conf->mac_ctrl_frame_fwd != 0)
+                       rctl |= IGC_RCTL_PMCF;
+               else
+                       rctl &= ~IGC_RCTL_PMCF;
+
+               IGC_WRITE_REG(hw, IGC_RCTL, rctl);
+               IGC_WRITE_FLUSH(hw);
+
+               return 0;
+       }
+
+       PMD_DRV_LOG(ERR, "igc_setup_link_generic = 0x%x", err);
+       return -EIO;
+}
+
+static int
+eth_igc_rss_reta_update(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_reta_entry64 *reta_conf,
+                       uint16_t reta_size)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint16_t i;
+
+       if (reta_size != ETH_RSS_RETA_SIZE_128) {
+               PMD_DRV_LOG(ERR,
+                       "The size of RSS redirection table configured(%d) doesn't match the number hardware can supported(%d)",
+                       reta_size, ETH_RSS_RETA_SIZE_128);
+               return -EINVAL;
+       }
+
+       RTE_BUILD_BUG_ON(ETH_RSS_RETA_SIZE_128 % IGC_RSS_RDT_REG_SIZE);
+
+       /* set redirection table */
+       for (i = 0; i < ETH_RSS_RETA_SIZE_128; i += IGC_RSS_RDT_REG_SIZE) {
+               union igc_rss_reta_reg reta, reg;
+               uint16_t idx, shift;
+               uint8_t j, mask;
+
+               idx = i / RTE_RETA_GROUP_SIZE;
+               shift = i % RTE_RETA_GROUP_SIZE;
+               mask = (uint8_t)((reta_conf[idx].mask >> shift) &
+                               IGC_RSS_RDT_REG_SIZE_MASK);
+
+               /* if no need to update the register */
+               if (!mask ||
+                   shift > (RTE_RETA_GROUP_SIZE - IGC_RSS_RDT_REG_SIZE))
+                       continue;
+
+               /* check mask whether need to read the register value first */
+               if (mask == IGC_RSS_RDT_REG_SIZE_MASK)
+                       reg.dword = 0;
+               else
+                       reg.dword = IGC_READ_REG_LE_VALUE(hw,
+                                       IGC_RETA(i / IGC_RSS_RDT_REG_SIZE));
+
+               /* update the register */
+               RTE_BUILD_BUG_ON(sizeof(reta.bytes) != IGC_RSS_RDT_REG_SIZE);
+               for (j = 0; j < IGC_RSS_RDT_REG_SIZE; j++) {
+                       if (mask & (1u << j))
+                               reta.bytes[j] =
+                                       (uint8_t)reta_conf[idx].reta[shift + j];
+                       else
+                               reta.bytes[j] = reg.bytes[j];
+               }
+               IGC_WRITE_REG_LE_VALUE(hw,
+                       IGC_RETA(i / IGC_RSS_RDT_REG_SIZE), reta.dword);
+       }
+
+       return 0;
+}
+
+static int
+eth_igc_rss_reta_query(struct rte_eth_dev *dev,
+                      struct rte_eth_rss_reta_entry64 *reta_conf,
+                      uint16_t reta_size)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint16_t i;
+
+       if (reta_size != ETH_RSS_RETA_SIZE_128) {
+               PMD_DRV_LOG(ERR,
+                       "The size of RSS redirection table configured(%d) doesn't match the number hardware can supported(%d)",
+                       reta_size, ETH_RSS_RETA_SIZE_128);
+               return -EINVAL;
+       }
+
+       RTE_BUILD_BUG_ON(ETH_RSS_RETA_SIZE_128 % IGC_RSS_RDT_REG_SIZE);
+
+       /* read redirection table */
+       for (i = 0; i < ETH_RSS_RETA_SIZE_128; i += IGC_RSS_RDT_REG_SIZE) {
+               union igc_rss_reta_reg reta;
+               uint16_t idx, shift;
+               uint8_t j, mask;
+
+               idx = i / RTE_RETA_GROUP_SIZE;
+               shift = i % RTE_RETA_GROUP_SIZE;
+               mask = (uint8_t)((reta_conf[idx].mask >> shift) &
+                               IGC_RSS_RDT_REG_SIZE_MASK);
+
+               /* if no need to read register */
+               if (!mask ||
+                   shift > (RTE_RETA_GROUP_SIZE - IGC_RSS_RDT_REG_SIZE))
+                       continue;
+
+               /* read register and get the queue index */
+               RTE_BUILD_BUG_ON(sizeof(reta.bytes) != IGC_RSS_RDT_REG_SIZE);
+               reta.dword = IGC_READ_REG_LE_VALUE(hw,
+                               IGC_RETA(i / IGC_RSS_RDT_REG_SIZE));
+               for (j = 0; j < IGC_RSS_RDT_REG_SIZE; j++) {
+                       if (mask & (1u << j))
+                               reta_conf[idx].reta[shift + j] = reta.bytes[j];
+               }
+       }
+
+       return 0;
+}
+
+static int
+eth_igc_rss_hash_update(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_conf *rss_conf)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       igc_hw_rss_hash_set(hw, rss_conf);
+       return 0;
+}
+
+static int
+eth_igc_rss_hash_conf_get(struct rte_eth_dev *dev,
+                       struct rte_eth_rss_conf *rss_conf)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint32_t *hash_key = (uint32_t *)rss_conf->rss_key;
+       uint32_t mrqc;
+       uint64_t rss_hf;
+
+       if (hash_key != NULL) {
+               int i;
+
+               /* if not enough space for store hash key */
+               if (rss_conf->rss_key_len != IGC_HKEY_SIZE) {
+                       PMD_DRV_LOG(ERR,
+                               "RSS hash key size %u in parameter doesn't match the hardware hash key size %u",
+                               rss_conf->rss_key_len, IGC_HKEY_SIZE);
+                       return -EINVAL;
+               }
+
+               /* read RSS key from register */
+               for (i = 0; i < IGC_HKEY_MAX_INDEX; i++)
+                       hash_key[i] = IGC_READ_REG_LE_VALUE(hw, IGC_RSSRK(i));
+       }
+
+       /* get RSS functions configured in MRQC register */
+       mrqc = IGC_READ_REG(hw, IGC_MRQC);
+       if ((mrqc & IGC_MRQC_ENABLE_RSS_4Q) == 0)
+               return 0;
+
+       rss_hf = 0;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV4)
+               rss_hf |= ETH_RSS_IPV4;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV4_TCP)
+               rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV6)
+               rss_hf |= ETH_RSS_IPV6;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_EX)
+               rss_hf |= ETH_RSS_IPV6_EX;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_TCP)
+               rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_TCP_EX)
+               rss_hf |= ETH_RSS_IPV6_TCP_EX;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV4_UDP)
+               rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_UDP)
+               rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
+       if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_UDP_EX)
+               rss_hf |= ETH_RSS_IPV6_UDP_EX;
+
+       rss_conf->rss_hf |= rss_hf;
+       return 0;
+}
+
+static int
+eth_igc_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_vfta *shadow_vfta = IGC_DEV_PRIVATE_VFTA(dev);
+       uint32_t vfta;
+       uint32_t vid_idx;
+       uint32_t vid_bit;
+
+       vid_idx = (vlan_id >> IGC_VFTA_ENTRY_SHIFT) & IGC_VFTA_ENTRY_MASK;
+       vid_bit = 1u << (vlan_id & IGC_VFTA_ENTRY_BIT_SHIFT_MASK);
+       vfta = shadow_vfta->vfta[vid_idx];
+       if (on)
+               vfta |= vid_bit;
+       else
+               vfta &= ~vid_bit;
+       IGC_WRITE_REG_ARRAY(hw, IGC_VFTA, vid_idx, vfta);
+
+       /* update local VFTA copy */
+       shadow_vfta->vfta[vid_idx] = vfta;
+
+       return 0;
+}
+
+static void
+igc_vlan_hw_filter_disable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       igc_read_reg_check_clear_bits(hw, IGC_RCTL,
+                       IGC_RCTL_CFIEN | IGC_RCTL_VFE);
+}
+
+static void
+igc_vlan_hw_filter_enable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       struct igc_vfta *shadow_vfta = IGC_DEV_PRIVATE_VFTA(dev);
+       uint32_t reg_val;
+       int i;
+
+       /* Filter Table Enable, CFI not used for packet acceptance */
+       reg_val = IGC_READ_REG(hw, IGC_RCTL);
+       reg_val &= ~IGC_RCTL_CFIEN;
+       reg_val |= IGC_RCTL_VFE;
+       IGC_WRITE_REG(hw, IGC_RCTL, reg_val);
+
+       /* restore VFTA table */
+       for (i = 0; i < IGC_VFTA_SIZE; i++)
+               IGC_WRITE_REG_ARRAY(hw, IGC_VFTA, i, shadow_vfta->vfta[i]);
+}
+
+static void
+igc_vlan_hw_strip_disable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       igc_read_reg_check_clear_bits(hw, IGC_CTRL, IGC_CTRL_VME);
+}
+
+static void
+igc_vlan_hw_strip_enable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+
+       igc_read_reg_check_set_bits(hw, IGC_CTRL, IGC_CTRL_VME);
+}
+
+static int
+igc_vlan_hw_extend_disable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint32_t ctrl_ext;
+
+       ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
+
+       /* if extend vlan hasn't been enabled */
+       if ((ctrl_ext & IGC_CTRL_EXT_EXT_VLAN) == 0)
+               return 0;
+
+       if ((dev->data->dev_conf.rxmode.offloads &
+                       DEV_RX_OFFLOAD_JUMBO_FRAME) == 0)
+               goto write_ext_vlan;
+
+       /* Update maximum packet length */
+       if (dev->data->dev_conf.rxmode.max_rx_pkt_len <
+               RTE_ETHER_MIN_MTU + VLAN_TAG_SIZE) {
+               PMD_DRV_LOG(ERR, "Maximum packet length %u error, min is %u",
+                       dev->data->dev_conf.rxmode.max_rx_pkt_len,
+                       VLAN_TAG_SIZE + RTE_ETHER_MIN_MTU);
+               return -EINVAL;
+       }
+       dev->data->dev_conf.rxmode.max_rx_pkt_len -= VLAN_TAG_SIZE;
+       IGC_WRITE_REG(hw, IGC_RLPML,
+               dev->data->dev_conf.rxmode.max_rx_pkt_len);
+
+write_ext_vlan:
+       IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext & ~IGC_CTRL_EXT_EXT_VLAN);
+       return 0;
+}
+
+static int
+igc_vlan_hw_extend_enable(struct rte_eth_dev *dev)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint32_t ctrl_ext;
+
+       ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
+
+       /* if extend vlan has been enabled */
+       if (ctrl_ext & IGC_CTRL_EXT_EXT_VLAN)
+               return 0;
+
+       if ((dev->data->dev_conf.rxmode.offloads &
+                       DEV_RX_OFFLOAD_JUMBO_FRAME) == 0)
+               goto write_ext_vlan;
+
+       /* Update maximum packet length */
+       if (dev->data->dev_conf.rxmode.max_rx_pkt_len >
+               MAX_RX_JUMBO_FRAME_SIZE - VLAN_TAG_SIZE) {
+               PMD_DRV_LOG(ERR, "Maximum packet length %u error, max is %u",
+                       dev->data->dev_conf.rxmode.max_rx_pkt_len +
+                       VLAN_TAG_SIZE, MAX_RX_JUMBO_FRAME_SIZE);
+               return -EINVAL;
+       }
+       dev->data->dev_conf.rxmode.max_rx_pkt_len += VLAN_TAG_SIZE;
+       IGC_WRITE_REG(hw, IGC_RLPML,
+               dev->data->dev_conf.rxmode.max_rx_pkt_len);
+
+write_ext_vlan:
+       IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext | IGC_CTRL_EXT_EXT_VLAN);
+       return 0;
+}
+
+static int
+eth_igc_vlan_offload_set(struct rte_eth_dev *dev, int mask)
+{
+       struct rte_eth_rxmode *rxmode;
+
+       rxmode = &dev->data->dev_conf.rxmode;
+       if (mask & ETH_VLAN_STRIP_MASK) {
+               if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
+                       igc_vlan_hw_strip_enable(dev);
+               else
+                       igc_vlan_hw_strip_disable(dev);
+       }
+
+       if (mask & ETH_VLAN_FILTER_MASK) {
+               if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
+                       igc_vlan_hw_filter_enable(dev);
+               else
+                       igc_vlan_hw_filter_disable(dev);
+       }
+
+       if (mask & ETH_VLAN_EXTEND_MASK) {
+               if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
+                       return igc_vlan_hw_extend_enable(dev);
+               else
+                       return igc_vlan_hw_extend_disable(dev);
+       }
+
+       return 0;
+}
+
+static int
+eth_igc_vlan_tpid_set(struct rte_eth_dev *dev,
+                     enum rte_vlan_type vlan_type,
+                     uint16_t tpid)
+{
+       struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
+       uint32_t reg_val;
+
+       /* only outer TPID of double VLAN can be configured*/
+       if (vlan_type == ETH_VLAN_TYPE_OUTER) {
+               reg_val = IGC_READ_REG(hw, IGC_VET);
+               reg_val = (reg_val & (~IGC_VET_EXT)) |
+                       ((uint32_t)tpid << IGC_VET_EXT_SHIFT);
+               IGC_WRITE_REG(hw, IGC_VET, reg_val);
+
+               return 0;
+       }
+
+       /* all other TPID values are read-only*/
+       PMD_DRV_LOG(ERR, "Not supported");
+       return -ENOTSUP;
+}
+
 static int
 eth_igc_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
        struct rte_pci_device *pci_dev)