drivers/net/igc/igc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2019-2020 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <string.h>
   7
   8 #include <rte_string_fns.h>
   9 #include <rte_pci.h>
  10 #include <rte_bus_pci.h>
  11 #include <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_malloc.h>
  14 #include <rte_alarm.h>
  15
  16 #include "igc_logs.h"
  17 #include "igc_txrx.h"
  18 #include "igc_filter.h"
  19 #include "igc_flow.h"
  20
  21 #define IGC_INTEL_VENDOR_ID             0x8086
  22
  23 #define IGC_FC_PAUSE_TIME               0x0680
  24 #define IGC_LINK_UPDATE_CHECK_TIMEOUT   90  /* 9s */
  25 #define IGC_LINK_UPDATE_CHECK_INTERVAL  100 /* ms */
  26
  27 #define IGC_MISC_VEC_ID                 RTE_INTR_VEC_ZERO_OFFSET
  28 #define IGC_RX_VEC_START                RTE_INTR_VEC_RXTX_OFFSET
  29 #define IGC_MSIX_OTHER_INTR_VEC         0   /* MSI-X other interrupt vector */
  30 #define IGC_FLAG_NEED_LINK_UPDATE       (1u << 0)       /* need update link */
  31
  32 #define IGC_DEFAULT_RX_FREE_THRESH      32
  33
  34 #define IGC_DEFAULT_RX_PTHRESH          8
  35 #define IGC_DEFAULT_RX_HTHRESH          8
  36 #define IGC_DEFAULT_RX_WTHRESH          4
  37
  38 #define IGC_DEFAULT_TX_PTHRESH          8
  39 #define IGC_DEFAULT_TX_HTHRESH          1
  40 #define IGC_DEFAULT_TX_WTHRESH          16
  41
  42 /* MSI-X other interrupt vector */
  43 #define IGC_MSIX_OTHER_INTR_VEC         0
  44
  45 /* External VLAN Enable bit mask */
  46 #define IGC_CTRL_EXT_EXT_VLAN           (1u << 26)
  47
  48 /* Speed select */
  49 #define IGC_CTRL_SPEED_MASK             (7u << 8)
  50 #define IGC_CTRL_SPEED_2500             (6u << 8)
  51
  52 /* External VLAN Ether Type bit mask and shift */
  53 #define IGC_VET_EXT                     0xFFFF0000
  54 #define IGC_VET_EXT_SHIFT               16
  55
  56 /* Force EEE Auto-negotiation */
  57 #define IGC_EEER_EEE_FRC_AN             (1u << 28)
  58
  59 /* Per Queue Good Packets Received Count */
  60 #define IGC_PQGPRC(idx)         (0x10010 + 0x100 * (idx))
  61 /* Per Queue Good Octets Received Count */
  62 #define IGC_PQGORC(idx)         (0x10018 + 0x100 * (idx))
  63 /* Per Queue Good Octets Transmitted Count */
  64 #define IGC_PQGOTC(idx)         (0x10034 + 0x100 * (idx))
  65 /* Per Queue Multicast Packets Received Count */
  66 #define IGC_PQMPRC(idx)         (0x10038 + 0x100 * (idx))
  67 /* Transmit Queue Drop Packet Count */
  68 #define IGC_TQDPC(idx)          (0xe030 + 0x40 * (idx))
  69
  70 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
  71 #define U32_0_IN_U64            0       /* lower bytes of u64 */
  72 #define U32_1_IN_U64            1       /* higher bytes of u64 */
  73 #else
  74 #define U32_0_IN_U64            1
  75 #define U32_1_IN_U64            0
  76 #endif
  77
  78 #define IGC_ALARM_INTERVAL      8000000u
  79 /* us, about 13.6s some per-queue registers will wrap around back to 0. */
  80
  81 static const struct rte_eth_desc_lim rx_desc_lim = {
  82         .nb_max = IGC_MAX_RXD,
  83         .nb_min = IGC_MIN_RXD,
  84         .nb_align = IGC_RXD_ALIGN,
  85 };
  86
  87 static const struct rte_eth_desc_lim tx_desc_lim = {
  88         .nb_max = IGC_MAX_TXD,
  89         .nb_min = IGC_MIN_TXD,
  90         .nb_align = IGC_TXD_ALIGN,
  91         .nb_seg_max = IGC_TX_MAX_SEG,
  92         .nb_mtu_seg_max = IGC_TX_MAX_MTU_SEG,
  93 };
  94
  95 static const struct rte_pci_id pci_id_igc_map[] = {
  96         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_LM) },
  97         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_V)  },
  98         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_I)  },
  99         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_K)  },
 100         { .vendor_id = 0, /* sentinel */ },
 101 };
 102
 103 /* store statistics names and its offset in stats structure */
 104 struct rte_igc_xstats_name_off {
 105         char name[RTE_ETH_XSTATS_NAME_SIZE];
 106         unsigned int offset;
 107 };
 108
 109 static const struct rte_igc_xstats_name_off rte_igc_stats_strings[] = {
 110         {"rx_crc_errors", offsetof(struct igc_hw_stats, crcerrs)},
 111         {"rx_align_errors", offsetof(struct igc_hw_stats, algnerrc)},
 112         {"rx_errors", offsetof(struct igc_hw_stats, rxerrc)},
 113         {"rx_missed_packets", offsetof(struct igc_hw_stats, mpc)},
 114         {"tx_single_collision_packets", offsetof(struct igc_hw_stats, scc)},
 115         {"tx_multiple_collision_packets", offsetof(struct igc_hw_stats, mcc)},
 116         {"tx_excessive_collision_packets", offsetof(struct igc_hw_stats,
 117                 ecol)},
 118         {"tx_late_collisions", offsetof(struct igc_hw_stats, latecol)},
 119         {"tx_total_collisions", offsetof(struct igc_hw_stats, colc)},
 120         {"tx_deferred_packets", offsetof(struct igc_hw_stats, dc)},
 121         {"tx_no_carrier_sense_packets", offsetof(struct igc_hw_stats, tncrs)},
 122         {"tx_discarded_packets", offsetof(struct igc_hw_stats, htdpmc)},
 123         {"rx_length_errors", offsetof(struct igc_hw_stats, rlec)},
 124         {"rx_xon_packets", offsetof(struct igc_hw_stats, xonrxc)},
 125         {"tx_xon_packets", offsetof(struct igc_hw_stats, xontxc)},
 126         {"rx_xoff_packets", offsetof(struct igc_hw_stats, xoffrxc)},
 127         {"tx_xoff_packets", offsetof(struct igc_hw_stats, xofftxc)},
 128         {"rx_flow_control_unsupported_packets", offsetof(struct igc_hw_stats,
 129                 fcruc)},
 130         {"rx_size_64_packets", offsetof(struct igc_hw_stats, prc64)},
 131         {"rx_size_65_to_127_packets", offsetof(struct igc_hw_stats, prc127)},
 132         {"rx_size_128_to_255_packets", offsetof(struct igc_hw_stats, prc255)},
 133         {"rx_size_256_to_511_packets", offsetof(struct igc_hw_stats, prc511)},
 134         {"rx_size_512_to_1023_packets", offsetof(struct igc_hw_stats,
 135                 prc1023)},
 136         {"rx_size_1024_to_max_packets", offsetof(struct igc_hw_stats,
 137                 prc1522)},
 138         {"rx_broadcast_packets", offsetof(struct igc_hw_stats, bprc)},
 139         {"rx_multicast_packets", offsetof(struct igc_hw_stats, mprc)},
 140         {"rx_undersize_errors", offsetof(struct igc_hw_stats, ruc)},
 141         {"rx_fragment_errors", offsetof(struct igc_hw_stats, rfc)},
 142         {"rx_oversize_errors", offsetof(struct igc_hw_stats, roc)},
 143         {"rx_jabber_errors", offsetof(struct igc_hw_stats, rjc)},
 144         {"rx_no_buffers", offsetof(struct igc_hw_stats, rnbc)},
 145         {"rx_management_packets", offsetof(struct igc_hw_stats, mgprc)},
 146         {"rx_management_dropped", offsetof(struct igc_hw_stats, mgpdc)},
 147         {"tx_management_packets", offsetof(struct igc_hw_stats, mgptc)},
 148         {"rx_total_packets", offsetof(struct igc_hw_stats, tpr)},
 149         {"tx_total_packets", offsetof(struct igc_hw_stats, tpt)},
 150         {"rx_total_bytes", offsetof(struct igc_hw_stats, tor)},
 151         {"tx_total_bytes", offsetof(struct igc_hw_stats, tot)},
 152         {"tx_size_64_packets", offsetof(struct igc_hw_stats, ptc64)},
 153         {"tx_size_65_to_127_packets", offsetof(struct igc_hw_stats, ptc127)},
 154         {"tx_size_128_to_255_packets", offsetof(struct igc_hw_stats, ptc255)},
 155         {"tx_size_256_to_511_packets", offsetof(struct igc_hw_stats, ptc511)},
 156         {"tx_size_512_to_1023_packets", offsetof(struct igc_hw_stats,
 157                 ptc1023)},
 158         {"tx_size_1023_to_max_packets", offsetof(struct igc_hw_stats,
 159                 ptc1522)},
 160         {"tx_multicast_packets", offsetof(struct igc_hw_stats, mptc)},
 161         {"tx_broadcast_packets", offsetof(struct igc_hw_stats, bptc)},
 162         {"tx_tso_packets", offsetof(struct igc_hw_stats, tsctc)},
 163         {"rx_sent_to_host_packets", offsetof(struct igc_hw_stats, rpthc)},
 164         {"tx_sent_by_host_packets", offsetof(struct igc_hw_stats, hgptc)},
 165         {"interrupt_assert_count", offsetof(struct igc_hw_stats, iac)},
 166         {"rx_descriptor_lower_threshold",
 167                 offsetof(struct igc_hw_stats, icrxdmtc)},
 168 };
 169
 170 #define IGC_NB_XSTATS (sizeof(rte_igc_stats_strings) / \
 171                 sizeof(rte_igc_stats_strings[0]))
 172
 173 static int eth_igc_configure(struct rte_eth_dev *dev);
 174 static int eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete);
 175 static int eth_igc_stop(struct rte_eth_dev *dev);
 176 static int eth_igc_start(struct rte_eth_dev *dev);
 177 static int eth_igc_set_link_up(struct rte_eth_dev *dev);
 178 static int eth_igc_set_link_down(struct rte_eth_dev *dev);
 179 static int eth_igc_close(struct rte_eth_dev *dev);
 180 static int eth_igc_reset(struct rte_eth_dev *dev);
 181 static int eth_igc_promiscuous_enable(struct rte_eth_dev *dev);
 182 static int eth_igc_promiscuous_disable(struct rte_eth_dev *dev);
 183 static int eth_igc_fw_version_get(struct rte_eth_dev *dev,
 184                                 char *fw_version, size_t fw_size);
 185 static int eth_igc_infos_get(struct rte_eth_dev *dev,
 186                         struct rte_eth_dev_info *dev_info);
 187 static int eth_igc_led_on(struct rte_eth_dev *dev);
 188 static int eth_igc_led_off(struct rte_eth_dev *dev);
 189 static const uint32_t *eth_igc_supported_ptypes_get(struct rte_eth_dev *dev);
 190 static int eth_igc_rar_set(struct rte_eth_dev *dev,
 191                 struct rte_ether_addr *mac_addr, uint32_t index, uint32_t pool);
 192 static void eth_igc_rar_clear(struct rte_eth_dev *dev, uint32_t index);
 193 static int eth_igc_default_mac_addr_set(struct rte_eth_dev *dev,
 194                         struct rte_ether_addr *addr);
 195 static int eth_igc_set_mc_addr_list(struct rte_eth_dev *dev,
 196                          struct rte_ether_addr *mc_addr_set,
 197                          uint32_t nb_mc_addr);
 198 static int eth_igc_allmulticast_enable(struct rte_eth_dev *dev);
 199 static int eth_igc_allmulticast_disable(struct rte_eth_dev *dev);
 200 static int eth_igc_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
 201 static int eth_igc_stats_get(struct rte_eth_dev *dev,
 202                         struct rte_eth_stats *rte_stats);
 203 static int eth_igc_xstats_get(struct rte_eth_dev *dev,
 204                         struct rte_eth_xstat *xstats, unsigned int n);
 205 static int eth_igc_xstats_get_by_id(struct rte_eth_dev *dev,
 206                                 const uint64_t *ids,
 207                                 uint64_t *values, unsigned int n);
 208 static int eth_igc_xstats_get_names(struct rte_eth_dev *dev,
 209                                 struct rte_eth_xstat_name *xstats_names,
 210                                 unsigned int size);
 211 static int eth_igc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 212                 const uint64_t *ids, struct rte_eth_xstat_name *xstats_names,
 213                 unsigned int limit);
 214 static int eth_igc_xstats_reset(struct rte_eth_dev *dev);
 215 static int
 216 eth_igc_queue_stats_mapping_set(struct rte_eth_dev *dev,
 217         uint16_t queue_id, uint8_t stat_idx, uint8_t is_rx);
 218 static int
 219 eth_igc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
 220 static int
 221 eth_igc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
 222 static int
 223 eth_igc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
 224 static int
 225 eth_igc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
 226 static int eth_igc_rss_reta_update(struct rte_eth_dev *dev,
 227                         struct rte_eth_rss_reta_entry64 *reta_conf,
 228                         uint16_t reta_size);
 229 static int eth_igc_rss_reta_query(struct rte_eth_dev *dev,
 230                        struct rte_eth_rss_reta_entry64 *reta_conf,
 231                        uint16_t reta_size);
 232 static int eth_igc_rss_hash_update(struct rte_eth_dev *dev,
 233                         struct rte_eth_rss_conf *rss_conf);
 234 static int eth_igc_rss_hash_conf_get(struct rte_eth_dev *dev,
 235                         struct rte_eth_rss_conf *rss_conf);
 236 static int
 237 eth_igc_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
 238 static int eth_igc_vlan_offload_set(struct rte_eth_dev *dev, int mask);
 239 static int eth_igc_vlan_tpid_set(struct rte_eth_dev *dev,
 240                       enum rte_vlan_type vlan_type, uint16_t tpid);
 241
 242 static const struct eth_dev_ops eth_igc_ops = {
 243         .dev_configure          = eth_igc_configure,
 244         .link_update            = eth_igc_link_update,
 245         .dev_stop               = eth_igc_stop,
 246         .dev_start              = eth_igc_start,
 247         .dev_close              = eth_igc_close,
 248         .dev_reset              = eth_igc_reset,
 249         .dev_set_link_up        = eth_igc_set_link_up,
 250         .dev_set_link_down      = eth_igc_set_link_down,
 251         .promiscuous_enable     = eth_igc_promiscuous_enable,
 252         .promiscuous_disable    = eth_igc_promiscuous_disable,
 253         .allmulticast_enable    = eth_igc_allmulticast_enable,
 254         .allmulticast_disable   = eth_igc_allmulticast_disable,
 255         .fw_version_get         = eth_igc_fw_version_get,
 256         .dev_infos_get          = eth_igc_infos_get,
 257         .dev_led_on             = eth_igc_led_on,
 258         .dev_led_off            = eth_igc_led_off,
 259         .dev_supported_ptypes_get = eth_igc_supported_ptypes_get,
 260         .mtu_set                = eth_igc_mtu_set,
 261         .mac_addr_add           = eth_igc_rar_set,
 262         .mac_addr_remove        = eth_igc_rar_clear,
 263         .mac_addr_set           = eth_igc_default_mac_addr_set,
 264         .set_mc_addr_list       = eth_igc_set_mc_addr_list,
 265
 266         .rx_queue_setup         = eth_igc_rx_queue_setup,
 267         .rx_queue_release       = eth_igc_rx_queue_release,
 268         .tx_queue_setup         = eth_igc_tx_queue_setup,
 269         .tx_queue_release       = eth_igc_tx_queue_release,
 270         .tx_done_cleanup        = eth_igc_tx_done_cleanup,
 271         .rxq_info_get           = eth_igc_rxq_info_get,
 272         .txq_info_get           = eth_igc_txq_info_get,
 273         .stats_get              = eth_igc_stats_get,
 274         .xstats_get             = eth_igc_xstats_get,
 275         .xstats_get_by_id       = eth_igc_xstats_get_by_id,
 276         .xstats_get_names_by_id = eth_igc_xstats_get_names_by_id,
 277         .xstats_get_names       = eth_igc_xstats_get_names,
 278         .stats_reset            = eth_igc_xstats_reset,
 279         .xstats_reset           = eth_igc_xstats_reset,
 280         .queue_stats_mapping_set = eth_igc_queue_stats_mapping_set,
 281         .rx_queue_intr_enable   = eth_igc_rx_queue_intr_enable,
 282         .rx_queue_intr_disable  = eth_igc_rx_queue_intr_disable,
 283         .flow_ctrl_get          = eth_igc_flow_ctrl_get,
 284         .flow_ctrl_set          = eth_igc_flow_ctrl_set,
 285         .reta_update            = eth_igc_rss_reta_update,
 286         .reta_query             = eth_igc_rss_reta_query,
 287         .rss_hash_update        = eth_igc_rss_hash_update,
 288         .rss_hash_conf_get      = eth_igc_rss_hash_conf_get,
 289         .vlan_filter_set        = eth_igc_vlan_filter_set,
 290         .vlan_offload_set       = eth_igc_vlan_offload_set,
 291         .vlan_tpid_set          = eth_igc_vlan_tpid_set,
 292         .vlan_strip_queue_set   = eth_igc_vlan_strip_queue_set,
 293         .flow_ops_get           = eth_igc_flow_ops_get,
 294 };
 295
 296 /*
 297  * multiple queue mode checking
 298  */
 299 static int
 300 igc_check_mq_mode(struct rte_eth_dev *dev)
 301 {
 302         enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
 303         enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
 304
 305         if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
 306                 PMD_INIT_LOG(ERR, "SRIOV is not supported.");
 307                 return -EINVAL;
 308         }
 309
 310         if (rx_mq_mode != RTE_ETH_MQ_RX_NONE &&
 311                 rx_mq_mode != RTE_ETH_MQ_RX_RSS) {
 312                 /* RSS together with VMDq not supported*/
 313                 PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
 314                                 rx_mq_mode);
 315                 return -EINVAL;
 316         }
 317
 318         /* To no break software that set invalid mode, only display
 319          * warning if invalid mode is used.
 320          */
 321         if (tx_mq_mode != RTE_ETH_MQ_TX_NONE)
 322                 PMD_INIT_LOG(WARNING,
 323                         "TX mode %d is not supported. Due to meaningless in this driver, just ignore",
 324                         tx_mq_mode);
 325
 326         return 0;
 327 }
 328
 329 static int
 330 eth_igc_configure(struct rte_eth_dev *dev)
 331 {
 332         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 333         int ret;
 334
 335         PMD_INIT_FUNC_TRACE();
 336
 337         if (dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
 338                 dev->data->dev_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
 339
 340         ret  = igc_check_mq_mode(dev);
 341         if (ret != 0)
 342                 return ret;
 343
 344         intr->flags |= IGC_FLAG_NEED_LINK_UPDATE;
 345         return 0;
 346 }
 347
 348 static int
 349 eth_igc_set_link_up(struct rte_eth_dev *dev)
 350 {
 351         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 352
 353         if (hw->phy.media_type == igc_media_type_copper)
 354                 igc_power_up_phy(hw);
 355         else
 356                 igc_power_up_fiber_serdes_link(hw);
 357         return 0;
 358 }
 359
 360 static int
 361 eth_igc_set_link_down(struct rte_eth_dev *dev)
 362 {
 363         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 364
 365         if (hw->phy.media_type == igc_media_type_copper)
 366                 igc_power_down_phy(hw);
 367         else
 368                 igc_shutdown_fiber_serdes_link(hw);
 369         return 0;
 370 }
 371
 372 /*
 373  * disable other interrupt
 374  */
 375 static void
 376 igc_intr_other_disable(struct rte_eth_dev *dev)
 377 {
 378         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 379         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 380         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
 381
 382         if (rte_intr_allow_others(intr_handle) &&
 383                 dev->data->dev_conf.intr_conf.lsc) {
 384                 IGC_WRITE_REG(hw, IGC_EIMC, 1u << IGC_MSIX_OTHER_INTR_VEC);
 385         }
 386
 387         IGC_WRITE_REG(hw, IGC_IMC, ~0);
 388         IGC_WRITE_FLUSH(hw);
 389 }
 390
 391 /*
 392  * enable other interrupt
 393  */
 394 static inline void
 395 igc_intr_other_enable(struct rte_eth_dev *dev)
 396 {
 397         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 398         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 399         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 400         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
 401
 402         if (rte_intr_allow_others(intr_handle) &&
 403                 dev->data->dev_conf.intr_conf.lsc) {
 404                 IGC_WRITE_REG(hw, IGC_EIMS, 1u << IGC_MSIX_OTHER_INTR_VEC);
 405         }
 406
 407         IGC_WRITE_REG(hw, IGC_IMS, intr->mask);
 408         IGC_WRITE_FLUSH(hw);
 409 }
 410
 411 /*
 412  * It reads ICR and gets interrupt causes, check it and set a bit flag
 413  * to update link status.
 414  */
 415 static void
 416 eth_igc_interrupt_get_status(struct rte_eth_dev *dev)
 417 {
 418         uint32_t icr;
 419         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 420         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 421
 422         /* read-on-clear nic registers here */
 423         icr = IGC_READ_REG(hw, IGC_ICR);
 424
 425         intr->flags = 0;
 426         if (icr & IGC_ICR_LSC)
 427                 intr->flags |= IGC_FLAG_NEED_LINK_UPDATE;
 428 }
 429
 430 /* return 0 means link status changed, -1 means not changed */
 431 static int
 432 eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 433 {
 434         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 435         struct rte_eth_link link;
 436         int link_check, count;
 437
 438         link_check = 0;
 439         hw->mac.get_link_status = 1;
 440
 441         /* possible wait-to-complete in up to 9 seconds */
 442         for (count = 0; count < IGC_LINK_UPDATE_CHECK_TIMEOUT; count++) {
 443                 /* Read the real link status */
 444                 switch (hw->phy.media_type) {
 445                 case igc_media_type_copper:
 446                         /* Do the work to read phy */
 447                         igc_check_for_link(hw);
 448                         link_check = !hw->mac.get_link_status;
 449                         break;
 450
 451                 case igc_media_type_fiber:
 452                         igc_check_for_link(hw);
 453                         link_check = (IGC_READ_REG(hw, IGC_STATUS) &
 454                                       IGC_STATUS_LU);
 455                         break;
 456
 457                 case igc_media_type_internal_serdes:
 458                         igc_check_for_link(hw);
 459                         link_check = hw->mac.serdes_has_link;
 460                         break;
 461
 462                 default:
 463                         break;
 464                 }
 465                 if (link_check || wait_to_complete == 0)
 466                         break;
 467                 rte_delay_ms(IGC_LINK_UPDATE_CHECK_INTERVAL);
 468         }
 469         memset(&link, 0, sizeof(link));
 470
 471         /* Now we check if a transition has happened */
 472         if (link_check) {
 473                 uint16_t duplex, speed;
 474                 hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
 475                 link.link_duplex = (duplex == FULL_DUPLEX) ?
 476                                 RTE_ETH_LINK_FULL_DUPLEX :
 477                                 RTE_ETH_LINK_HALF_DUPLEX;
 478                 link.link_speed = speed;
 479                 link.link_status = RTE_ETH_LINK_UP;
 480                 link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 481                                 RTE_ETH_LINK_SPEED_FIXED);
 482
 483                 if (speed == SPEED_2500) {
 484                         uint32_t tipg = IGC_READ_REG(hw, IGC_TIPG);
 485                         if ((tipg & IGC_TIPG_IPGT_MASK) != 0x0b) {
 486                                 tipg &= ~IGC_TIPG_IPGT_MASK;
 487                                 tipg |= 0x0b;
 488                                 IGC_WRITE_REG(hw, IGC_TIPG, tipg);
 489                         }
 490                 }
 491         } else {
 492                 link.link_speed = 0;
 493                 link.link_duplex = RTE_ETH_LINK_HALF_DUPLEX;
 494                 link.link_status = RTE_ETH_LINK_DOWN;
 495                 link.link_autoneg = RTE_ETH_LINK_FIXED;
 496         }
 497
 498         return rte_eth_linkstatus_set(dev, &link);
 499 }
 500
 501 /*
 502  * It executes link_update after knowing an interrupt is present.
 503  */
 504 static void
 505 eth_igc_interrupt_action(struct rte_eth_dev *dev)
 506 {
 507         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 508         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 509         struct rte_eth_link link;
 510         int ret;
 511
 512         if (intr->flags & IGC_FLAG_NEED_LINK_UPDATE) {
 513                 intr->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
 514
 515                 /* set get_link_status to check register later */
 516                 ret = eth_igc_link_update(dev, 0);
 517
 518                 /* check if link has changed */
 519                 if (ret < 0)
 520                         return;
 521
 522                 rte_eth_linkstatus_get(dev, &link);
 523                 if (link.link_status)
 524                         PMD_DRV_LOG(INFO,
 525                                 " Port %d: Link Up - speed %u Mbps - %s",
 526                                 dev->data->port_id,
 527                                 (unsigned int)link.link_speed,
 528                                 link.link_duplex == RTE_ETH_LINK_FULL_DUPLEX ?
 529                                 "full-duplex" : "half-duplex");
 530                 else
 531                         PMD_DRV_LOG(INFO, " Port %d: Link Down",
 532                                 dev->data->port_id);
 533
 534                 PMD_DRV_LOG(DEBUG, "PCI Address: " PCI_PRI_FMT,
 535                                 pci_dev->addr.domain,
 536                                 pci_dev->addr.bus,
 537                                 pci_dev->addr.devid,
 538                                 pci_dev->addr.function);
 539                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 540         }
 541 }
 542
 543 /*
 544  * Interrupt handler which shall be registered at first.
 545  *
 546  * @handle
 547  *  Pointer to interrupt handle.
 548  * @param
 549  *  The address of parameter (struct rte_eth_dev *) registered before.
 550  */
 551 static void
 552 eth_igc_interrupt_handler(void *param)
 553 {
 554         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
 555
 556         eth_igc_interrupt_get_status(dev);
 557         eth_igc_interrupt_action(dev);
 558 }
 559
 560 static void igc_read_queue_stats_register(struct rte_eth_dev *dev);
 561
 562 /*
 563  * Update the queue status every IGC_ALARM_INTERVAL time.
 564  * @param
 565  *  The address of parameter (struct rte_eth_dev *) registered before.
 566  */
 567 static void
 568 igc_update_queue_stats_handler(void *param)
 569 {
 570         struct rte_eth_dev *dev = param;
 571         igc_read_queue_stats_register(dev);
 572         rte_eal_alarm_set(IGC_ALARM_INTERVAL,
 573                         igc_update_queue_stats_handler, dev);
 574 }
 575
 576 /*
 577  * rx,tx enable/disable
 578  */
 579 static void
 580 eth_igc_rxtx_control(struct rte_eth_dev *dev, bool enable)
 581 {
 582         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 583         uint32_t tctl, rctl;
 584
 585         tctl = IGC_READ_REG(hw, IGC_TCTL);
 586         rctl = IGC_READ_REG(hw, IGC_RCTL);
 587
 588         if (enable) {
 589                 /* enable Tx/Rx */
 590                 tctl |= IGC_TCTL_EN;
 591                 rctl |= IGC_RCTL_EN;
 592         } else {
 593                 /* disable Tx/Rx */
 594                 tctl &= ~IGC_TCTL_EN;
 595                 rctl &= ~IGC_RCTL_EN;
 596         }
 597         IGC_WRITE_REG(hw, IGC_TCTL, tctl);
 598         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
 599         IGC_WRITE_FLUSH(hw);
 600 }
 601
 602 /*
 603  *  This routine disables all traffic on the adapter by issuing a
 604  *  global reset on the MAC.
 605  */
 606 static int
 607 eth_igc_stop(struct rte_eth_dev *dev)
 608 {
 609         struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
 610         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 611         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 612         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
 613         struct rte_eth_link link;
 614
 615         dev->data->dev_started = 0;
 616         adapter->stopped = 1;
 617
 618         /* disable receive and transmit */
 619         eth_igc_rxtx_control(dev, false);
 620
 621         /* disable all MSI-X interrupts */
 622         IGC_WRITE_REG(hw, IGC_EIMC, 0x1f);
 623         IGC_WRITE_FLUSH(hw);
 624
 625         /* clear all MSI-X interrupts */
 626         IGC_WRITE_REG(hw, IGC_EICR, 0x1f);
 627
 628         igc_intr_other_disable(dev);
 629
 630         rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
 631
 632         /* disable intr eventfd mapping */
 633         rte_intr_disable(intr_handle);
 634
 635         igc_reset_hw(hw);
 636
 637         /* disable all wake up */
 638         IGC_WRITE_REG(hw, IGC_WUC, 0);
 639
 640         /* disable checking EEE operation in MAC loopback mode */
 641         igc_read_reg_check_clear_bits(hw, IGC_EEER, IGC_EEER_EEE_FRC_AN);
 642
 643         /* Set bit for Go Link disconnect */
 644         igc_read_reg_check_set_bits(hw, IGC_82580_PHY_POWER_MGMT,
 645                         IGC_82580_PM_GO_LINKD);
 646
 647         /* Power down the phy. Needed to make the link go Down */
 648         eth_igc_set_link_down(dev);
 649
 650         igc_dev_clear_queues(dev);
 651
 652         /* clear the recorded link status */
 653         memset(&link, 0, sizeof(link));
 654         rte_eth_linkstatus_set(dev, &link);
 655
 656         if (!rte_intr_allow_others(intr_handle))
 657                 /* resume to the default handler */
 658                 rte_intr_callback_register(intr_handle,
 659                                            eth_igc_interrupt_handler,
 660                                            (void *)dev);
 661
 662         /* Clean datapath event and queue/vec mapping */
 663         rte_intr_efd_disable(intr_handle);
 664         rte_intr_vec_list_free(intr_handle);
 665
 666         return 0;
 667 }
 668
 669 /*
 670  * write interrupt vector allocation register
 671  * @hw
 672  *  board private structure
 673  * @queue_index
 674  *  queue index, valid 0,1,2,3
 675  * @tx
 676  *  tx:1, rx:0
 677  * @msix_vector
 678  *  msix-vector, valid 0,1,2,3,4
 679  */
 680 static void
 681 igc_write_ivar(struct igc_hw *hw, uint8_t queue_index,
 682                 bool tx, uint8_t msix_vector)
 683 {
 684         uint8_t offset = 0;
 685         uint8_t reg_index = queue_index >> 1;
 686         uint32_t val;
 687
 688         /*
 689          * IVAR(0)
 690          * bit31...24   bit23...16      bit15...8       bit7...0
 691          * TX1          RX1             TX0             RX0
 692          *
 693          * IVAR(1)
 694          * bit31...24   bit23...16      bit15...8       bit7...0
 695          * TX3          RX3             TX2             RX2
 696          */
 697
 698         if (tx)
 699                 offset = 8;
 700
 701         if (queue_index & 1)
 702                 offset += 16;
 703
 704         val = IGC_READ_REG_ARRAY(hw, IGC_IVAR0, reg_index);
 705
 706         /* clear bits */
 707         val &= ~((uint32_t)0xFF << offset);
 708
 709         /* write vector and valid bit */
 710         val |= (uint32_t)(msix_vector | IGC_IVAR_VALID) << offset;
 711
 712         IGC_WRITE_REG_ARRAY(hw, IGC_IVAR0, reg_index, val);
 713 }
 714
 715 /* Sets up the hardware to generate MSI-X interrupts properly
 716  * @hw
 717  *  board private structure
 718  */
 719 static void
 720 igc_configure_msix_intr(struct rte_eth_dev *dev)
 721 {
 722         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 723         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 724         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
 725
 726         uint32_t intr_mask;
 727         uint32_t vec = IGC_MISC_VEC_ID;
 728         uint32_t base = IGC_MISC_VEC_ID;
 729         uint32_t misc_shift = 0;
 730         int i;
 731
 732         /* won't configure msix register if no mapping is done
 733          * between intr vector and event fd
 734          */
 735         if (!rte_intr_dp_is_en(intr_handle))
 736                 return;
 737
 738         if (rte_intr_allow_others(intr_handle)) {
 739                 base = IGC_RX_VEC_START;
 740                 vec = base;
 741                 misc_shift = 1;
 742         }
 743
 744         /* turn on MSI-X capability first */
 745         IGC_WRITE_REG(hw, IGC_GPIE, IGC_GPIE_MSIX_MODE |
 746                                 IGC_GPIE_PBA | IGC_GPIE_EIAME |
 747                                 IGC_GPIE_NSICR);
 748         intr_mask = RTE_LEN2MASK(rte_intr_nb_efd_get(intr_handle),
 749                                  uint32_t) << misc_shift;
 750
 751         if (dev->data->dev_conf.intr_conf.lsc)
 752                 intr_mask |= (1u << IGC_MSIX_OTHER_INTR_VEC);
 753
 754         /* enable msix auto-clear */
 755         igc_read_reg_check_set_bits(hw, IGC_EIAC, intr_mask);
 756
 757         /* set other cause interrupt vector */
 758         igc_read_reg_check_set_bits(hw, IGC_IVAR_MISC,
 759                 (uint32_t)(IGC_MSIX_OTHER_INTR_VEC | IGC_IVAR_VALID) << 8);
 760
 761         /* enable auto-mask */
 762         igc_read_reg_check_set_bits(hw, IGC_EIAM, intr_mask);
 763
 764         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 765                 igc_write_ivar(hw, i, 0, vec);
 766                 rte_intr_vec_list_index_set(intr_handle, i, vec);
 767                 if (vec < base + rte_intr_nb_efd_get(intr_handle) - 1)
 768                         vec++;
 769         }
 770
 771         IGC_WRITE_FLUSH(hw);
 772 }
 773
 774 /**
 775  * It enables the interrupt mask and then enable the interrupt.
 776  *
 777  * @dev
 778  *  Pointer to struct rte_eth_dev.
 779  * @on
 780  *  Enable or Disable
 781  */
 782 static void
 783 igc_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
 784 {
 785         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 786
 787         if (on)
 788                 intr->mask |= IGC_ICR_LSC;
 789         else
 790                 intr->mask &= ~IGC_ICR_LSC;
 791 }
 792
 793 /*
 794  * It enables the interrupt.
 795  * It will be called once only during nic initialized.
 796  */
 797 static void
 798 igc_rxq_interrupt_setup(struct rte_eth_dev *dev)
 799 {
 800         uint32_t mask;
 801         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 802         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 803         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
 804         int misc_shift = rte_intr_allow_others(intr_handle) ? 1 : 0;
 805
 806         /* won't configure msix register if no mapping is done
 807          * between intr vector and event fd
 808          */
 809         if (!rte_intr_dp_is_en(intr_handle))
 810                 return;
 811
 812         mask = RTE_LEN2MASK(rte_intr_nb_efd_get(intr_handle), uint32_t)
 813                 << misc_shift;
 814         IGC_WRITE_REG(hw, IGC_EIMS, mask);
 815 }
 816
 817 /*
 818  *  Get hardware rx-buffer size.
 819  */
 820 static inline int
 821 igc_get_rx_buffer_size(struct igc_hw *hw)
 822 {
 823         return (IGC_READ_REG(hw, IGC_RXPBS) & 0x3f) << 10;
 824 }
 825
 826 /*
 827  * igc_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
 828  * For ASF and Pass Through versions of f/w this means
 829  * that the driver is loaded.
 830  */
 831 static void
 832 igc_hw_control_acquire(struct igc_hw *hw)
 833 {
 834         uint32_t ctrl_ext;
 835
 836         /* Let firmware know the driver has taken over */
 837         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
 838         IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
 839 }
 840
 841 /*
 842  * igc_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
 843  * For ASF and Pass Through versions of f/w this means that the
 844  * driver is no longer loaded.
 845  */
 846 static void
 847 igc_hw_control_release(struct igc_hw *hw)
 848 {
 849         uint32_t ctrl_ext;
 850
 851         /* Let firmware taken over control of h/w */
 852         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
 853         IGC_WRITE_REG(hw, IGC_CTRL_EXT,
 854                         ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
 855 }
 856
 857 static int
 858 igc_hardware_init(struct igc_hw *hw)
 859 {
 860         uint32_t rx_buf_size;
 861         int diag;
 862
 863         /* Let the firmware know the OS is in control */
 864         igc_hw_control_acquire(hw);
 865
 866         /* Issue a global reset */
 867         igc_reset_hw(hw);
 868
 869         /* disable all wake up */
 870         IGC_WRITE_REG(hw, IGC_WUC, 0);
 871
 872         /*
 873          * Hardware flow control
 874          * - High water mark should allow for at least two standard size (1518)
 875          *   frames to be received after sending an XOFF.
 876          * - Low water mark works best when it is very near the high water mark.
 877          *   This allows the receiver to restart by sending XON when it has
 878          *   drained a bit. Here we use an arbitrary value of 1500 which will
 879          *   restart after one full frame is pulled from the buffer. There
 880          *   could be several smaller frames in the buffer and if so they will
 881          *   not trigger the XON until their total number reduces the buffer
 882          *   by 1500.
 883          */
 884         rx_buf_size = igc_get_rx_buffer_size(hw);
 885         hw->fc.high_water = rx_buf_size - (RTE_ETHER_MAX_LEN * 2);
 886         hw->fc.low_water = hw->fc.high_water - 1500;
 887         hw->fc.pause_time = IGC_FC_PAUSE_TIME;
 888         hw->fc.send_xon = 1;
 889         hw->fc.requested_mode = igc_fc_full;
 890
 891         diag = igc_init_hw(hw);
 892         if (diag < 0)
 893                 return diag;
 894
 895         igc_get_phy_info(hw);
 896         igc_check_for_link(hw);
 897
 898         return 0;
 899 }
 900
 901 static int
 902 eth_igc_start(struct rte_eth_dev *dev)
 903 {
 904         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 905         struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
 906         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 907         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
 908         uint32_t *speeds;
 909         int ret;
 910
 911         PMD_INIT_FUNC_TRACE();
 912
 913         /* disable all MSI-X interrupts */
 914         IGC_WRITE_REG(hw, IGC_EIMC, 0x1f);
 915         IGC_WRITE_FLUSH(hw);
 916
 917         /* clear all MSI-X interrupts */
 918         IGC_WRITE_REG(hw, IGC_EICR, 0x1f);
 919
 920         /* disable uio/vfio intr/eventfd mapping */
 921         if (!adapter->stopped)
 922                 rte_intr_disable(intr_handle);
 923
 924         /* Power up the phy. Needed to make the link go Up */
 925         eth_igc_set_link_up(dev);
 926
 927         /* Put the address into the Receive Address Array */
 928         igc_rar_set(hw, hw->mac.addr, 0);
 929
 930         /* Initialize the hardware */
 931         if (igc_hardware_init(hw)) {
 932                 PMD_DRV_LOG(ERR, "Unable to initialize the hardware");
 933                 return -EIO;
 934         }
 935         adapter->stopped = 0;
 936
 937         /* check and configure queue intr-vector mapping */
 938         if (rte_intr_cap_multiple(intr_handle) &&
 939                 dev->data->dev_conf.intr_conf.rxq) {
 940                 uint32_t intr_vector = dev->data->nb_rx_queues;
 941                 if (rte_intr_efd_enable(intr_handle, intr_vector))
 942                         return -1;
 943         }
 944
 945         if (rte_intr_dp_is_en(intr_handle)) {
 946                 if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
 947                                                   dev->data->nb_rx_queues)) {
 948                         PMD_DRV_LOG(ERR,
 949                                 "Failed to allocate %d rx_queues intr_vec",
 950                                 dev->data->nb_rx_queues);
 951                         return -ENOMEM;
 952                 }
 953         }
 954
 955         /* configure msix for rx interrupt */
 956         igc_configure_msix_intr(dev);
 957
 958         igc_tx_init(dev);
 959
 960         /* This can fail when allocating mbufs for descriptor rings */
 961         ret = igc_rx_init(dev);
 962         if (ret) {
 963                 PMD_DRV_LOG(ERR, "Unable to initialize RX hardware");
 964                 igc_dev_clear_queues(dev);
 965                 return ret;
 966         }
 967
 968         igc_clear_hw_cntrs_base_generic(hw);
 969
 970         /* VLAN Offload Settings */
 971         eth_igc_vlan_offload_set(dev,
 972                 RTE_ETH_VLAN_STRIP_MASK | RTE_ETH_VLAN_FILTER_MASK |
 973                 RTE_ETH_VLAN_EXTEND_MASK);
 974
 975         /* Setup link speed and duplex */
 976         speeds = &dev->data->dev_conf.link_speeds;
 977         if (*speeds == RTE_ETH_LINK_SPEED_AUTONEG) {
 978                 hw->phy.autoneg_advertised = IGC_ALL_SPEED_DUPLEX_2500;
 979                 hw->mac.autoneg = 1;
 980         } else {
 981                 int num_speeds = 0;
 982
 983                 if (*speeds & RTE_ETH_LINK_SPEED_FIXED) {
 984                         PMD_DRV_LOG(ERR,
 985                                     "Force speed mode currently not supported");
 986                         igc_dev_clear_queues(dev);
 987                         return -EINVAL;
 988                 }
 989
 990                 hw->phy.autoneg_advertised = 0;
 991                 hw->mac.autoneg = 1;
 992
 993                 if (*speeds & ~(RTE_ETH_LINK_SPEED_10M_HD | RTE_ETH_LINK_SPEED_10M |
 994                                 RTE_ETH_LINK_SPEED_100M_HD | RTE_ETH_LINK_SPEED_100M |
 995                                 RTE_ETH_LINK_SPEED_1G | RTE_ETH_LINK_SPEED_2_5G)) {
 996                         num_speeds = -1;
 997                         goto error_invalid_config;
 998                 }
 999                 if (*speeds & RTE_ETH_LINK_SPEED_10M_HD) {
1000                         hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
1001                         num_speeds++;
1002                 }
1003                 if (*speeds & RTE_ETH_LINK_SPEED_10M) {
1004                         hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
1005                         num_speeds++;
1006                 }
1007                 if (*speeds & RTE_ETH_LINK_SPEED_100M_HD) {
1008                         hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
1009                         num_speeds++;
1010                 }
1011                 if (*speeds & RTE_ETH_LINK_SPEED_100M) {
1012                         hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
1013                         num_speeds++;
1014                 }
1015                 if (*speeds & RTE_ETH_LINK_SPEED_1G) {
1016                         hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
1017                         num_speeds++;
1018                 }
1019                 if (*speeds & RTE_ETH_LINK_SPEED_2_5G) {
1020                         hw->phy.autoneg_advertised |= ADVERTISE_2500_FULL;
1021                         num_speeds++;
1022                 }
1023                 if (num_speeds == 0)
1024                         goto error_invalid_config;
1025         }
1026
1027         igc_setup_link(hw);
1028
1029         if (rte_intr_allow_others(intr_handle)) {
1030                 /* check if lsc interrupt is enabled */
1031                 if (dev->data->dev_conf.intr_conf.lsc)
1032                         igc_lsc_interrupt_setup(dev, 1);
1033                 else
1034                         igc_lsc_interrupt_setup(dev, 0);
1035         } else {
1036                 rte_intr_callback_unregister(intr_handle,
1037                                              eth_igc_interrupt_handler,
1038                                              (void *)dev);
1039                 if (dev->data->dev_conf.intr_conf.lsc)
1040                         PMD_DRV_LOG(INFO,
1041                                 "LSC won't enable because of no intr multiplex");
1042         }
1043
1044         /* enable uio/vfio intr/eventfd mapping */
1045         rte_intr_enable(intr_handle);
1046
1047         rte_eal_alarm_set(IGC_ALARM_INTERVAL,
1048                         igc_update_queue_stats_handler, dev);
1049
1050         /* check if rxq interrupt is enabled */
1051         if (dev->data->dev_conf.intr_conf.rxq &&
1052                         rte_intr_dp_is_en(intr_handle))
1053                 igc_rxq_interrupt_setup(dev);
1054
1055         /* resume enabled intr since hw reset */
1056         igc_intr_other_enable(dev);
1057
1058         eth_igc_rxtx_control(dev, true);
1059         eth_igc_link_update(dev, 0);
1060
1061         /* configure MAC-loopback mode */
1062         if (dev->data->dev_conf.lpbk_mode == 1) {
1063                 uint32_t reg_val;
1064
1065                 reg_val = IGC_READ_REG(hw, IGC_CTRL);
1066                 reg_val &= ~IGC_CTRL_SPEED_MASK;
1067                 reg_val |= IGC_CTRL_SLU | IGC_CTRL_FRCSPD |
1068                         IGC_CTRL_FRCDPX | IGC_CTRL_FD | IGC_CTRL_SPEED_2500;
1069                 IGC_WRITE_REG(hw, IGC_CTRL, reg_val);
1070
1071                 igc_read_reg_check_set_bits(hw, IGC_EEER, IGC_EEER_EEE_FRC_AN);
1072         }
1073
1074         return 0;
1075
1076 error_invalid_config:
1077         PMD_DRV_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
1078                      dev->data->dev_conf.link_speeds, dev->data->port_id);
1079         igc_dev_clear_queues(dev);
1080         return -EINVAL;
1081 }
1082
1083 static int
1084 igc_reset_swfw_lock(struct igc_hw *hw)
1085 {
1086         int ret_val;
1087
1088         /*
1089          * Do mac ops initialization manually here, since we will need
1090          * some function pointers set by this call.
1091          */
1092         ret_val = igc_init_mac_params(hw);
1093         if (ret_val)
1094                 return ret_val;
1095
1096         /*
1097          * SMBI lock should not fail in this early stage. If this is the case,
1098          * it is due to an improper exit of the application.
1099          * So force the release of the faulty lock.
1100          */
1101         if (igc_get_hw_semaphore_generic(hw) < 0)
1102                 PMD_DRV_LOG(DEBUG, "SMBI lock released");
1103
1104         igc_put_hw_semaphore_generic(hw);
1105
1106         if (hw->mac.ops.acquire_swfw_sync != NULL) {
1107                 uint16_t mask;
1108
1109                 /*
1110                  * Phy lock should not fail in this early stage.
1111                  * If this is the case, it is due to an improper exit of the
1112                  * application. So force the release of the faulty lock.
1113                  */
1114                 mask = IGC_SWFW_PHY0_SM;
1115                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
1116                         PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
1117                                     hw->bus.func);
1118                 }
1119                 hw->mac.ops.release_swfw_sync(hw, mask);
1120
1121                 /*
1122                  * This one is more tricky since it is common to all ports; but
1123                  * swfw_sync retries last long enough (1s) to be almost sure
1124                  * that if lock can not be taken it is due to an improper lock
1125                  * of the semaphore.
1126                  */
1127                 mask = IGC_SWFW_EEP_SM;
1128                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0)
1129                         PMD_DRV_LOG(DEBUG, "SWFW common locks released");
1130
1131                 hw->mac.ops.release_swfw_sync(hw, mask);
1132         }
1133
1134         return IGC_SUCCESS;
1135 }
1136
1137 /*
1138  * free all rx/tx queues.
1139  */
1140 static void
1141 igc_dev_free_queues(struct rte_eth_dev *dev)
1142 {
1143         uint16_t i;
1144
1145         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1146                 eth_igc_rx_queue_release(dev, i);
1147                 dev->data->rx_queues[i] = NULL;
1148         }
1149         dev->data->nb_rx_queues = 0;
1150
1151         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1152                 eth_igc_tx_queue_release(dev, i);
1153                 dev->data->tx_queues[i] = NULL;
1154         }
1155         dev->data->nb_tx_queues = 0;
1156 }
1157
1158 static int
1159 eth_igc_close(struct rte_eth_dev *dev)
1160 {
1161         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1162         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
1163         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1164         struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
1165         int retry = 0;
1166         int ret = 0;
1167
1168         PMD_INIT_FUNC_TRACE();
1169         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1170                 return 0;
1171
1172         if (!adapter->stopped)
1173                 ret = eth_igc_stop(dev);
1174
1175         igc_flow_flush(dev, NULL);
1176         igc_clear_all_filter(dev);
1177
1178         igc_intr_other_disable(dev);
1179         do {
1180                 int ret = rte_intr_callback_unregister(intr_handle,
1181                                 eth_igc_interrupt_handler, dev);
1182                 if (ret >= 0 || ret == -ENOENT || ret == -EINVAL)
1183                         break;
1184
1185                 PMD_DRV_LOG(ERR, "intr callback unregister failed: %d", ret);
1186                 DELAY(200 * 1000); /* delay 200ms */
1187         } while (retry++ < 5);
1188
1189         igc_phy_hw_reset(hw);
1190         igc_hw_control_release(hw);
1191         igc_dev_free_queues(dev);
1192
1193         /* Reset any pending lock */
1194         igc_reset_swfw_lock(hw);
1195
1196         return ret;
1197 }
1198
1199 static void
1200 igc_identify_hardware(struct rte_eth_dev *dev, struct rte_pci_device *pci_dev)
1201 {
1202         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1203
1204         hw->vendor_id = pci_dev->id.vendor_id;
1205         hw->device_id = pci_dev->id.device_id;
1206         hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1207         hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
1208 }
1209
1210 static int
1211 eth_igc_dev_init(struct rte_eth_dev *dev)
1212 {
1213         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1214         struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
1215         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1216         int i, error = 0;
1217
1218         PMD_INIT_FUNC_TRACE();
1219         dev->dev_ops = &eth_igc_ops;
1220         dev->rx_queue_count = eth_igc_rx_queue_count;
1221         dev->rx_descriptor_status = eth_igc_rx_descriptor_status;
1222         dev->tx_descriptor_status = eth_igc_tx_descriptor_status;
1223
1224         /*
1225          * for secondary processes, we don't initialize any further as primary
1226          * has already done this work. Only check we don't need a different
1227          * RX function.
1228          */
1229         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1230                 return 0;
1231
1232         rte_eth_copy_pci_info(dev, pci_dev);
1233         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1234
1235         hw->back = pci_dev;
1236         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
1237
1238         igc_identify_hardware(dev, pci_dev);
1239         if (igc_setup_init_funcs(hw, false) != IGC_SUCCESS) {
1240                 error = -EIO;
1241                 goto err_late;
1242         }
1243
1244         igc_get_bus_info(hw);
1245
1246         /* Reset any pending lock */
1247         if (igc_reset_swfw_lock(hw) != IGC_SUCCESS) {
1248                 error = -EIO;
1249                 goto err_late;
1250         }
1251
1252         /* Finish initialization */
1253         if (igc_setup_init_funcs(hw, true) != IGC_SUCCESS) {
1254                 error = -EIO;
1255                 goto err_late;
1256         }
1257
1258         hw->mac.autoneg = 1;
1259         hw->phy.autoneg_wait_to_complete = 0;
1260         hw->phy.autoneg_advertised = IGC_ALL_SPEED_DUPLEX_2500;
1261
1262         /* Copper options */
1263         if (hw->phy.media_type == igc_media_type_copper) {
1264                 hw->phy.mdix = 0; /* AUTO_ALL_MODES */
1265                 hw->phy.disable_polarity_correction = 0;
1266                 hw->phy.ms_type = igc_ms_hw_default;
1267         }
1268
1269         /*
1270          * Start from a known state, this is important in reading the nvm
1271          * and mac from that.
1272          */
1273         igc_reset_hw(hw);
1274
1275         /* Make sure we have a good EEPROM before we read from it */
1276         if (igc_validate_nvm_checksum(hw) < 0) {
1277                 /*
1278                  * Some PCI-E parts fail the first check due to
1279                  * the link being in sleep state, call it again,
1280                  * if it fails a second time its a real issue.
1281                  */
1282                 if (igc_validate_nvm_checksum(hw) < 0) {
1283                         PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
1284                         error = -EIO;
1285                         goto err_late;
1286                 }
1287         }
1288
1289         /* Read the permanent MAC address out of the EEPROM */
1290         if (igc_read_mac_addr(hw) != 0) {
1291                 PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
1292                 error = -EIO;
1293                 goto err_late;
1294         }
1295
1296         /* Allocate memory for storing MAC addresses */
1297         dev->data->mac_addrs = rte_zmalloc("igc",
1298                 RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
1299         if (dev->data->mac_addrs == NULL) {
1300                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes for storing MAC",
1301                                 RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count);
1302                 error = -ENOMEM;
1303                 goto err_late;
1304         }
1305
1306         /* Copy the permanent MAC address */
1307         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.addr,
1308                         &dev->data->mac_addrs[0]);
1309
1310         /* Now initialize the hardware */
1311         if (igc_hardware_init(hw) != 0) {
1312                 PMD_INIT_LOG(ERR, "Hardware initialization failed");
1313                 rte_free(dev->data->mac_addrs);
1314                 dev->data->mac_addrs = NULL;
1315                 error = -ENODEV;
1316                 goto err_late;
1317         }
1318
1319         hw->mac.get_link_status = 1;
1320         igc->stopped = 0;
1321
1322         /* Indicate SOL/IDER usage */
1323         if (igc_check_reset_block(hw) < 0)
1324                 PMD_INIT_LOG(ERR,
1325                         "PHY reset is blocked due to SOL/IDER session.");
1326
1327         PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
1328                         dev->data->port_id, pci_dev->id.vendor_id,
1329                         pci_dev->id.device_id);
1330
1331         rte_intr_callback_register(pci_dev->intr_handle,
1332                         eth_igc_interrupt_handler, (void *)dev);
1333
1334         /* enable uio/vfio intr/eventfd mapping */
1335         rte_intr_enable(pci_dev->intr_handle);
1336
1337         /* enable support intr */
1338         igc_intr_other_enable(dev);
1339
1340         /* initiate queue status */
1341         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1342                 igc->txq_stats_map[i] = -1;
1343                 igc->rxq_stats_map[i] = -1;
1344         }
1345
1346         igc_flow_init(dev);
1347         igc_clear_all_filter(dev);
1348         return 0;
1349
1350 err_late:
1351         igc_hw_control_release(hw);
1352         return error;
1353 }
1354
1355 static int
1356 eth_igc_dev_uninit(__rte_unused struct rte_eth_dev *eth_dev)
1357 {
1358         PMD_INIT_FUNC_TRACE();
1359         eth_igc_close(eth_dev);
1360         return 0;
1361 }
1362
1363 static int
1364 eth_igc_reset(struct rte_eth_dev *dev)
1365 {
1366         int ret;
1367
1368         PMD_INIT_FUNC_TRACE();
1369
1370         ret = eth_igc_dev_uninit(dev);
1371         if (ret)
1372                 return ret;
1373
1374         return eth_igc_dev_init(dev);
1375 }
1376
1377 static int
1378 eth_igc_promiscuous_enable(struct rte_eth_dev *dev)
1379 {
1380         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1381         uint32_t rctl;
1382
1383         rctl = IGC_READ_REG(hw, IGC_RCTL);
1384         rctl |= (IGC_RCTL_UPE | IGC_RCTL_MPE);
1385         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1386         return 0;
1387 }
1388
1389 static int
1390 eth_igc_promiscuous_disable(struct rte_eth_dev *dev)
1391 {
1392         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1393         uint32_t rctl;
1394
1395         rctl = IGC_READ_REG(hw, IGC_RCTL);
1396         rctl &= (~IGC_RCTL_UPE);
1397         if (dev->data->all_multicast == 1)
1398                 rctl |= IGC_RCTL_MPE;
1399         else
1400                 rctl &= (~IGC_RCTL_MPE);
1401         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1402         return 0;
1403 }
1404
1405 static int
1406 eth_igc_allmulticast_enable(struct rte_eth_dev *dev)
1407 {
1408         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1409         uint32_t rctl;
1410
1411         rctl = IGC_READ_REG(hw, IGC_RCTL);
1412         rctl |= IGC_RCTL_MPE;
1413         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1414         return 0;
1415 }
1416
1417 static int
1418 eth_igc_allmulticast_disable(struct rte_eth_dev *dev)
1419 {
1420         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1421         uint32_t rctl;
1422
1423         if (dev->data->promiscuous == 1)
1424                 return 0;       /* must remain in all_multicast mode */
1425
1426         rctl = IGC_READ_REG(hw, IGC_RCTL);
1427         rctl &= (~IGC_RCTL_MPE);
1428         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1429         return 0;
1430 }
1431
1432 static int
1433 eth_igc_fw_version_get(struct rte_eth_dev *dev, char *fw_version,
1434                        size_t fw_size)
1435 {
1436         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1437         struct igc_fw_version fw;
1438         int ret;
1439
1440         igc_get_fw_version(hw, &fw);
1441
1442         /* if option rom is valid, display its version too */
1443         if (fw.or_valid) {
1444                 ret = snprintf(fw_version, fw_size,
1445                          "%d.%d, 0x%08x, %d.%d.%d",
1446                          fw.eep_major, fw.eep_minor, fw.etrack_id,
1447                          fw.or_major, fw.or_build, fw.or_patch);
1448         /* no option rom */
1449         } else {
1450                 if (fw.etrack_id != 0X0000) {
1451                         ret = snprintf(fw_version, fw_size,
1452                                  "%d.%d, 0x%08x",
1453                                  fw.eep_major, fw.eep_minor,
1454                                  fw.etrack_id);
1455                 } else {
1456                         ret = snprintf(fw_version, fw_size,
1457                                  "%d.%d.%d",
1458                                  fw.eep_major, fw.eep_minor,
1459                                  fw.eep_build);
1460                 }
1461         }
1462         if (ret < 0)
1463                 return -EINVAL;
1464
1465         ret += 1; /* add the size of '\0' */
1466         if (fw_size < (size_t)ret)
1467                 return ret;
1468         else
1469                 return 0;
1470 }
1471
1472 static int
1473 eth_igc_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1474 {
1475         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1476
1477         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
1478         dev_info->max_rx_pktlen = MAX_RX_JUMBO_FRAME_SIZE;
1479         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
1480         dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
1481         dev_info->rx_offload_capa = IGC_RX_OFFLOAD_ALL;
1482         dev_info->tx_offload_capa = IGC_TX_OFFLOAD_ALL;
1483         dev_info->rx_queue_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
1484
1485         dev_info->max_rx_queues = IGC_QUEUE_PAIRS_NUM;
1486         dev_info->max_tx_queues = IGC_QUEUE_PAIRS_NUM;
1487         dev_info->max_vmdq_pools = 0;
1488
1489         dev_info->hash_key_size = IGC_HKEY_MAX_INDEX * sizeof(uint32_t);
1490         dev_info->reta_size = RTE_ETH_RSS_RETA_SIZE_128;
1491         dev_info->flow_type_rss_offloads = IGC_RSS_OFFLOAD_ALL;
1492
1493         dev_info->default_rxconf = (struct rte_eth_rxconf) {
1494                 .rx_thresh = {
1495                         .pthresh = IGC_DEFAULT_RX_PTHRESH,
1496                         .hthresh = IGC_DEFAULT_RX_HTHRESH,
1497                         .wthresh = IGC_DEFAULT_RX_WTHRESH,
1498                 },
1499                 .rx_free_thresh = IGC_DEFAULT_RX_FREE_THRESH,
1500                 .rx_drop_en = 0,
1501                 .offloads = 0,
1502         };
1503
1504         dev_info->default_txconf = (struct rte_eth_txconf) {
1505                 .tx_thresh = {
1506                         .pthresh = IGC_DEFAULT_TX_PTHRESH,
1507                         .hthresh = IGC_DEFAULT_TX_HTHRESH,
1508                         .wthresh = IGC_DEFAULT_TX_WTHRESH,
1509                 },
1510                 .offloads = 0,
1511         };
1512
1513         dev_info->rx_desc_lim = rx_desc_lim;
1514         dev_info->tx_desc_lim = tx_desc_lim;
1515
1516         dev_info->speed_capa = RTE_ETH_LINK_SPEED_10M_HD | RTE_ETH_LINK_SPEED_10M |
1517                         RTE_ETH_LINK_SPEED_100M_HD | RTE_ETH_LINK_SPEED_100M |
1518                         RTE_ETH_LINK_SPEED_1G | RTE_ETH_LINK_SPEED_2_5G;
1519
1520         dev_info->max_mtu = dev_info->max_rx_pktlen - IGC_ETH_OVERHEAD;
1521         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
1522         return 0;
1523 }
1524
1525 static int
1526 eth_igc_led_on(struct rte_eth_dev *dev)
1527 {
1528         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1529
1530         return igc_led_on(hw) == IGC_SUCCESS ? 0 : -ENOTSUP;
1531 }
1532
1533 static int
1534 eth_igc_led_off(struct rte_eth_dev *dev)
1535 {
1536         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1537
1538         return igc_led_off(hw) == IGC_SUCCESS ? 0 : -ENOTSUP;
1539 }
1540
1541 static const uint32_t *
1542 eth_igc_supported_ptypes_get(__rte_unused struct rte_eth_dev *dev)
1543 {
1544         static const uint32_t ptypes[] = {
1545                 /* refers to rx_desc_pkt_info_to_pkt_type() */
1546                 RTE_PTYPE_L2_ETHER,
1547                 RTE_PTYPE_L3_IPV4,
1548                 RTE_PTYPE_L3_IPV4_EXT,
1549                 RTE_PTYPE_L3_IPV6,
1550                 RTE_PTYPE_L3_IPV6_EXT,
1551                 RTE_PTYPE_L4_TCP,
1552                 RTE_PTYPE_L4_UDP,
1553                 RTE_PTYPE_L4_SCTP,
1554                 RTE_PTYPE_TUNNEL_IP,
1555                 RTE_PTYPE_INNER_L3_IPV6,
1556                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1557                 RTE_PTYPE_INNER_L4_TCP,
1558                 RTE_PTYPE_INNER_L4_UDP,
1559                 RTE_PTYPE_UNKNOWN
1560         };
1561
1562         return ptypes;
1563 }
1564
1565 static int
1566 eth_igc_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1567 {
1568         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1569         uint32_t frame_size = mtu + IGC_ETH_OVERHEAD;
1570         uint32_t rctl;
1571
1572         /* if extend vlan has been enabled */
1573         if (IGC_READ_REG(hw, IGC_CTRL_EXT) & IGC_CTRL_EXT_EXT_VLAN)
1574                 frame_size += VLAN_TAG_SIZE;
1575
1576         /*
1577          * If device is started, refuse mtu that requires the support of
1578          * scattered packets when this feature has not been enabled before.
1579          */
1580         if (dev->data->dev_started && !dev->data->scattered_rx &&
1581             frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM) {
1582                 PMD_INIT_LOG(ERR, "Stop port first.");
1583                 return -EINVAL;
1584         }
1585
1586         rctl = IGC_READ_REG(hw, IGC_RCTL);
1587         if (mtu > RTE_ETHER_MTU)
1588                 rctl |= IGC_RCTL_LPE;
1589         else
1590                 rctl &= ~IGC_RCTL_LPE;
1591         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1592
1593         IGC_WRITE_REG(hw, IGC_RLPML, frame_size);
1594
1595         return 0;
1596 }
1597
1598 static int
1599 eth_igc_rar_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1600                 uint32_t index, uint32_t pool)
1601 {
1602         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1603
1604         igc_rar_set(hw, mac_addr->addr_bytes, index);
1605         RTE_SET_USED(pool);
1606         return 0;
1607 }
1608
1609 static void
1610 eth_igc_rar_clear(struct rte_eth_dev *dev, uint32_t index)
1611 {
1612         uint8_t addr[RTE_ETHER_ADDR_LEN];
1613         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1614
1615         memset(addr, 0, sizeof(addr));
1616         igc_rar_set(hw, addr, index);
1617 }
1618
1619 static int
1620 eth_igc_default_mac_addr_set(struct rte_eth_dev *dev,
1621                         struct rte_ether_addr *addr)
1622 {
1623         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1624         igc_rar_set(hw, addr->addr_bytes, 0);
1625         return 0;
1626 }
1627
1628 static int
1629 eth_igc_set_mc_addr_list(struct rte_eth_dev *dev,
1630                          struct rte_ether_addr *mc_addr_set,
1631                          uint32_t nb_mc_addr)
1632 {
1633         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1634         igc_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
1635         return 0;
1636 }
1637
1638 /*
1639  * Read hardware registers
1640  */
1641 static void
1642 igc_read_stats_registers(struct igc_hw *hw, struct igc_hw_stats *stats)
1643 {
1644         int pause_frames;
1645
1646         uint64_t old_gprc  = stats->gprc;
1647         uint64_t old_gptc  = stats->gptc;
1648         uint64_t old_tpr   = stats->tpr;
1649         uint64_t old_tpt   = stats->tpt;
1650         uint64_t old_rpthc = stats->rpthc;
1651         uint64_t old_hgptc = stats->hgptc;
1652
1653         stats->crcerrs += IGC_READ_REG(hw, IGC_CRCERRS);
1654         stats->algnerrc += IGC_READ_REG(hw, IGC_ALGNERRC);
1655         stats->rxerrc += IGC_READ_REG(hw, IGC_RXERRC);
1656         stats->mpc += IGC_READ_REG(hw, IGC_MPC);
1657         stats->scc += IGC_READ_REG(hw, IGC_SCC);
1658         stats->ecol += IGC_READ_REG(hw, IGC_ECOL);
1659
1660         stats->mcc += IGC_READ_REG(hw, IGC_MCC);
1661         stats->latecol += IGC_READ_REG(hw, IGC_LATECOL);
1662         stats->colc += IGC_READ_REG(hw, IGC_COLC);
1663
1664         stats->dc += IGC_READ_REG(hw, IGC_DC);
1665         stats->tncrs += IGC_READ_REG(hw, IGC_TNCRS);
1666         stats->htdpmc += IGC_READ_REG(hw, IGC_HTDPMC);
1667         stats->rlec += IGC_READ_REG(hw, IGC_RLEC);
1668         stats->xonrxc += IGC_READ_REG(hw, IGC_XONRXC);
1669         stats->xontxc += IGC_READ_REG(hw, IGC_XONTXC);
1670
1671         /*
1672          * For watchdog management we need to know if we have been
1673          * paused during the last interval, so capture that here.
1674          */
1675         pause_frames = IGC_READ_REG(hw, IGC_XOFFRXC);
1676         stats->xoffrxc += pause_frames;
1677         stats->xofftxc += IGC_READ_REG(hw, IGC_XOFFTXC);
1678         stats->fcruc += IGC_READ_REG(hw, IGC_FCRUC);
1679         stats->prc64 += IGC_READ_REG(hw, IGC_PRC64);
1680         stats->prc127 += IGC_READ_REG(hw, IGC_PRC127);
1681         stats->prc255 += IGC_READ_REG(hw, IGC_PRC255);
1682         stats->prc511 += IGC_READ_REG(hw, IGC_PRC511);
1683         stats->prc1023 += IGC_READ_REG(hw, IGC_PRC1023);
1684         stats->prc1522 += IGC_READ_REG(hw, IGC_PRC1522);
1685         stats->gprc += IGC_READ_REG(hw, IGC_GPRC);
1686         stats->bprc += IGC_READ_REG(hw, IGC_BPRC);
1687         stats->mprc += IGC_READ_REG(hw, IGC_MPRC);
1688         stats->gptc += IGC_READ_REG(hw, IGC_GPTC);
1689
1690         /* For the 64-bit byte counters the low dword must be read first. */
1691         /* Both registers clear on the read of the high dword */
1692
1693         /* Workaround CRC bytes included in size, take away 4 bytes/packet */
1694         stats->gorc += IGC_READ_REG(hw, IGC_GORCL);
1695         stats->gorc += ((uint64_t)IGC_READ_REG(hw, IGC_GORCH) << 32);
1696         stats->gorc -= (stats->gprc - old_gprc) * RTE_ETHER_CRC_LEN;
1697         stats->gotc += IGC_READ_REG(hw, IGC_GOTCL);
1698         stats->gotc += ((uint64_t)IGC_READ_REG(hw, IGC_GOTCH) << 32);
1699         stats->gotc -= (stats->gptc - old_gptc) * RTE_ETHER_CRC_LEN;
1700
1701         stats->rnbc += IGC_READ_REG(hw, IGC_RNBC);
1702         stats->ruc += IGC_READ_REG(hw, IGC_RUC);
1703         stats->rfc += IGC_READ_REG(hw, IGC_RFC);
1704         stats->roc += IGC_READ_REG(hw, IGC_ROC);
1705         stats->rjc += IGC_READ_REG(hw, IGC_RJC);
1706
1707         stats->mgprc += IGC_READ_REG(hw, IGC_MGTPRC);
1708         stats->mgpdc += IGC_READ_REG(hw, IGC_MGTPDC);
1709         stats->mgptc += IGC_READ_REG(hw, IGC_MGTPTC);
1710         stats->b2ospc += IGC_READ_REG(hw, IGC_B2OSPC);
1711         stats->b2ogprc += IGC_READ_REG(hw, IGC_B2OGPRC);
1712         stats->o2bgptc += IGC_READ_REG(hw, IGC_O2BGPTC);
1713         stats->o2bspc += IGC_READ_REG(hw, IGC_O2BSPC);
1714
1715         stats->tpr += IGC_READ_REG(hw, IGC_TPR);
1716         stats->tpt += IGC_READ_REG(hw, IGC_TPT);
1717
1718         stats->tor += IGC_READ_REG(hw, IGC_TORL);
1719         stats->tor += ((uint64_t)IGC_READ_REG(hw, IGC_TORH) << 32);
1720         stats->tor -= (stats->tpr - old_tpr) * RTE_ETHER_CRC_LEN;
1721         stats->tot += IGC_READ_REG(hw, IGC_TOTL);
1722         stats->tot += ((uint64_t)IGC_READ_REG(hw, IGC_TOTH) << 32);
1723         stats->tot -= (stats->tpt - old_tpt) * RTE_ETHER_CRC_LEN;
1724
1725         stats->ptc64 += IGC_READ_REG(hw, IGC_PTC64);
1726         stats->ptc127 += IGC_READ_REG(hw, IGC_PTC127);
1727         stats->ptc255 += IGC_READ_REG(hw, IGC_PTC255);
1728         stats->ptc511 += IGC_READ_REG(hw, IGC_PTC511);
1729         stats->ptc1023 += IGC_READ_REG(hw, IGC_PTC1023);
1730         stats->ptc1522 += IGC_READ_REG(hw, IGC_PTC1522);
1731         stats->mptc += IGC_READ_REG(hw, IGC_MPTC);
1732         stats->bptc += IGC_READ_REG(hw, IGC_BPTC);
1733         stats->tsctc += IGC_READ_REG(hw, IGC_TSCTC);
1734
1735         stats->iac += IGC_READ_REG(hw, IGC_IAC);
1736         stats->rpthc += IGC_READ_REG(hw, IGC_RPTHC);
1737         stats->hgptc += IGC_READ_REG(hw, IGC_HGPTC);
1738         stats->icrxdmtc += IGC_READ_REG(hw, IGC_ICRXDMTC);
1739
1740         /* Host to Card Statistics */
1741         stats->hgorc += IGC_READ_REG(hw, IGC_HGORCL);
1742         stats->hgorc += ((uint64_t)IGC_READ_REG(hw, IGC_HGORCH) << 32);
1743         stats->hgorc -= (stats->rpthc - old_rpthc) * RTE_ETHER_CRC_LEN;
1744         stats->hgotc += IGC_READ_REG(hw, IGC_HGOTCL);
1745         stats->hgotc += ((uint64_t)IGC_READ_REG(hw, IGC_HGOTCH) << 32);
1746         stats->hgotc -= (stats->hgptc - old_hgptc) * RTE_ETHER_CRC_LEN;
1747         stats->lenerrs += IGC_READ_REG(hw, IGC_LENERRS);
1748 }
1749
1750 /*
1751  * Write 0 to all queue status registers
1752  */
1753 static void
1754 igc_reset_queue_stats_register(struct igc_hw *hw)
1755 {
1756         int i;
1757
1758         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1759                 IGC_WRITE_REG(hw, IGC_PQGPRC(i), 0);
1760                 IGC_WRITE_REG(hw, IGC_PQGPTC(i), 0);
1761                 IGC_WRITE_REG(hw, IGC_PQGORC(i), 0);
1762                 IGC_WRITE_REG(hw, IGC_PQGOTC(i), 0);
1763                 IGC_WRITE_REG(hw, IGC_PQMPRC(i), 0);
1764                 IGC_WRITE_REG(hw, IGC_RQDPC(i), 0);
1765                 IGC_WRITE_REG(hw, IGC_TQDPC(i), 0);
1766         }
1767 }
1768
1769 /*
1770  * Read all hardware queue status registers
1771  */
1772 static void
1773 igc_read_queue_stats_register(struct rte_eth_dev *dev)
1774 {
1775         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1776         struct igc_hw_queue_stats *queue_stats =
1777                                 IGC_DEV_PRIVATE_QUEUE_STATS(dev);
1778         int i;
1779
1780         /*
1781          * This register is not cleared on read. Furthermore, the register wraps
1782          * around back to 0x00000000 on the next increment when reaching a value
1783          * of 0xFFFFFFFF and then continues normal count operation.
1784          */
1785         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1786                 union {
1787                         u64 ddword;
1788                         u32 dword[2];
1789                 } value;
1790                 u32 tmp;
1791
1792                 /*
1793                  * Read the register first, if the value is smaller than that
1794                  * previous read, that mean the register has been overflowed,
1795                  * then we add the high 4 bytes by 1 and replace the low 4
1796                  * bytes by the new value.
1797                  */
1798                 tmp = IGC_READ_REG(hw, IGC_PQGPRC(i));
1799                 value.ddword = queue_stats->pqgprc[i];
1800                 if (value.dword[U32_0_IN_U64] > tmp)
1801                         value.dword[U32_1_IN_U64]++;
1802                 value.dword[U32_0_IN_U64] = tmp;
1803                 queue_stats->pqgprc[i] = value.ddword;
1804
1805                 tmp = IGC_READ_REG(hw, IGC_PQGPTC(i));
1806                 value.ddword = queue_stats->pqgptc[i];
1807                 if (value.dword[U32_0_IN_U64] > tmp)
1808                         value.dword[U32_1_IN_U64]++;
1809                 value.dword[U32_0_IN_U64] = tmp;
1810                 queue_stats->pqgptc[i] = value.ddword;
1811
1812                 tmp = IGC_READ_REG(hw, IGC_PQGORC(i));
1813                 value.ddword = queue_stats->pqgorc[i];
1814                 if (value.dword[U32_0_IN_U64] > tmp)
1815                         value.dword[U32_1_IN_U64]++;
1816                 value.dword[U32_0_IN_U64] = tmp;
1817                 queue_stats->pqgorc[i] = value.ddword;
1818
1819                 tmp = IGC_READ_REG(hw, IGC_PQGOTC(i));
1820                 value.ddword = queue_stats->pqgotc[i];
1821                 if (value.dword[U32_0_IN_U64] > tmp)
1822                         value.dword[U32_1_IN_U64]++;
1823                 value.dword[U32_0_IN_U64] = tmp;
1824                 queue_stats->pqgotc[i] = value.ddword;
1825
1826                 tmp = IGC_READ_REG(hw, IGC_PQMPRC(i));
1827                 value.ddword = queue_stats->pqmprc[i];
1828                 if (value.dword[U32_0_IN_U64] > tmp)
1829                         value.dword[U32_1_IN_U64]++;
1830                 value.dword[U32_0_IN_U64] = tmp;
1831                 queue_stats->pqmprc[i] = value.ddword;
1832
1833                 tmp = IGC_READ_REG(hw, IGC_RQDPC(i));
1834                 value.ddword = queue_stats->rqdpc[i];
1835                 if (value.dword[U32_0_IN_U64] > tmp)
1836                         value.dword[U32_1_IN_U64]++;
1837                 value.dword[U32_0_IN_U64] = tmp;
1838                 queue_stats->rqdpc[i] = value.ddword;
1839
1840                 tmp = IGC_READ_REG(hw, IGC_TQDPC(i));
1841                 value.ddword = queue_stats->tqdpc[i];
1842                 if (value.dword[U32_0_IN_U64] > tmp)
1843                         value.dword[U32_1_IN_U64]++;
1844                 value.dword[U32_0_IN_U64] = tmp;
1845                 queue_stats->tqdpc[i] = value.ddword;
1846         }
1847 }
1848
1849 static int
1850 eth_igc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1851 {
1852         struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
1853         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1854         struct igc_hw_stats *stats = IGC_DEV_PRIVATE_STATS(dev);
1855         struct igc_hw_queue_stats *queue_stats =
1856                         IGC_DEV_PRIVATE_QUEUE_STATS(dev);
1857         int i;
1858
1859         /*
1860          * Cancel status handler since it will read the queue status registers
1861          */
1862         rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
1863
1864         /* Read status register */
1865         igc_read_queue_stats_register(dev);
1866         igc_read_stats_registers(hw, stats);
1867
1868         if (rte_stats == NULL) {
1869                 /* Restart queue status handler */
1870                 rte_eal_alarm_set(IGC_ALARM_INTERVAL,
1871                                 igc_update_queue_stats_handler, dev);
1872                 return -EINVAL;
1873         }
1874
1875         /* Rx Errors */
1876         rte_stats->imissed = stats->mpc;
1877         rte_stats->ierrors = stats->crcerrs + stats->rlec +
1878                         stats->rxerrc + stats->algnerrc;
1879
1880         /* Tx Errors */
1881         rte_stats->oerrors = stats->ecol + stats->latecol;
1882
1883         rte_stats->ipackets = stats->gprc;
1884         rte_stats->opackets = stats->gptc;
1885         rte_stats->ibytes   = stats->gorc;
1886         rte_stats->obytes   = stats->gotc;
1887
1888         /* Get per-queue statuses */
1889         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1890                 /* GET TX queue statuses */
1891                 int map_id = igc->txq_stats_map[i];
1892                 if (map_id >= 0) {
1893                         rte_stats->q_opackets[map_id] += queue_stats->pqgptc[i];
1894                         rte_stats->q_obytes[map_id] += queue_stats->pqgotc[i];
1895                 }
1896                 /* Get RX queue statuses */
1897                 map_id = igc->rxq_stats_map[i];
1898                 if (map_id >= 0) {
1899                         rte_stats->q_ipackets[map_id] += queue_stats->pqgprc[i];
1900                         rte_stats->q_ibytes[map_id] += queue_stats->pqgorc[i];
1901                         rte_stats->q_errors[map_id] += queue_stats->rqdpc[i];
1902                 }
1903         }
1904
1905         /* Restart queue status handler */
1906         rte_eal_alarm_set(IGC_ALARM_INTERVAL,
1907                         igc_update_queue_stats_handler, dev);
1908         return 0;
1909 }
1910
1911 static int
1912 eth_igc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1913                    unsigned int n)
1914 {
1915         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1916         struct igc_hw_stats *hw_stats =
1917                         IGC_DEV_PRIVATE_STATS(dev);
1918         unsigned int i;
1919
1920         igc_read_stats_registers(hw, hw_stats);
1921
1922         if (n < IGC_NB_XSTATS)
1923                 return IGC_NB_XSTATS;
1924
1925         /* If this is a reset xstats is NULL, and we have cleared the
1926          * registers by reading them.
1927          */
1928         if (!xstats)
1929                 return 0;
1930
1931         /* Extended stats */
1932         for (i = 0; i < IGC_NB_XSTATS; i++) {
1933                 xstats[i].id = i;
1934                 xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1935                         rte_igc_stats_strings[i].offset);
1936         }
1937
1938         return IGC_NB_XSTATS;
1939 }
1940
1941 static int
1942 eth_igc_xstats_reset(struct rte_eth_dev *dev)
1943 {
1944         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1945         struct igc_hw_stats *hw_stats = IGC_DEV_PRIVATE_STATS(dev);
1946         struct igc_hw_queue_stats *queue_stats =
1947                         IGC_DEV_PRIVATE_QUEUE_STATS(dev);
1948
1949         /* Cancel queue status handler for avoid conflict */
1950         rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
1951
1952         /* HW registers are cleared on read */
1953         igc_reset_queue_stats_register(hw);
1954         igc_read_stats_registers(hw, hw_stats);
1955
1956         /* Reset software totals */
1957         memset(hw_stats, 0, sizeof(*hw_stats));
1958         memset(queue_stats, 0, sizeof(*queue_stats));
1959
1960         /* Restart the queue status handler */
1961         rte_eal_alarm_set(IGC_ALARM_INTERVAL, igc_update_queue_stats_handler,
1962                         dev);
1963
1964         return 0;
1965 }
1966
1967 static int
1968 eth_igc_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1969         struct rte_eth_xstat_name *xstats_names, unsigned int size)
1970 {
1971         unsigned int i;
1972
1973         if (xstats_names == NULL)
1974                 return IGC_NB_XSTATS;
1975
1976         if (size < IGC_NB_XSTATS) {
1977                 PMD_DRV_LOG(ERR, "not enough buffers!");
1978                 return IGC_NB_XSTATS;
1979         }
1980
1981         for (i = 0; i < IGC_NB_XSTATS; i++)
1982                 strlcpy(xstats_names[i].name, rte_igc_stats_strings[i].name,
1983                         sizeof(xstats_names[i].name));
1984
1985         return IGC_NB_XSTATS;
1986 }
1987
1988 static int
1989 eth_igc_xstats_get_names_by_id(struct rte_eth_dev *dev,
1990                 const uint64_t *ids, struct rte_eth_xstat_name *xstats_names,
1991                 unsigned int limit)
1992 {
1993         unsigned int i;
1994
1995         if (!ids)
1996                 return eth_igc_xstats_get_names(dev, xstats_names, limit);
1997
1998         for (i = 0; i < limit; i++) {
1999                 if (ids[i] >= IGC_NB_XSTATS) {
2000                         PMD_DRV_LOG(ERR, "id value isn't valid");
2001                         return -EINVAL;
2002                 }
2003                 strlcpy(xstats_names[i].name,
2004                         rte_igc_stats_strings[ids[i]].name,
2005                         sizeof(xstats_names[i].name));
2006         }
2007         return limit;
2008 }
2009
2010 static int
2011 eth_igc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
2012                 uint64_t *values, unsigned int n)
2013 {
2014         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2015         struct igc_hw_stats *hw_stats = IGC_DEV_PRIVATE_STATS(dev);
2016         unsigned int i;
2017
2018         igc_read_stats_registers(hw, hw_stats);
2019
2020         if (!ids) {
2021                 if (n < IGC_NB_XSTATS)
2022                         return IGC_NB_XSTATS;
2023
2024                 /* If this is a reset xstats is NULL, and we have cleared the
2025                  * registers by reading them.
2026                  */
2027                 if (!values)
2028                         return 0;
2029
2030                 /* Extended stats */
2031                 for (i = 0; i < IGC_NB_XSTATS; i++)
2032                         values[i] = *(uint64_t *)(((char *)hw_stats) +
2033                                         rte_igc_stats_strings[i].offset);
2034
2035                 return IGC_NB_XSTATS;
2036
2037         } else {
2038                 for (i = 0; i < n; i++) {
2039                         if (ids[i] >= IGC_NB_XSTATS) {
2040                                 PMD_DRV_LOG(ERR, "id value isn't valid");
2041                                 return -EINVAL;
2042                         }
2043                         values[i] = *(uint64_t *)(((char *)hw_stats) +
2044                                         rte_igc_stats_strings[ids[i]].offset);
2045                 }
2046                 return n;
2047         }
2048 }
2049
2050 static int
2051 eth_igc_queue_stats_mapping_set(struct rte_eth_dev *dev,
2052                 uint16_t queue_id, uint8_t stat_idx, uint8_t is_rx)
2053 {
2054         struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
2055
2056         /* check queue id is valid */
2057         if (queue_id >= IGC_QUEUE_PAIRS_NUM) {
2058                 PMD_DRV_LOG(ERR, "queue id(%u) error, max is %u",
2059                         queue_id, IGC_QUEUE_PAIRS_NUM - 1);
2060                 return -EINVAL;
2061         }
2062
2063         /* store the mapping status id */
2064         if (is_rx)
2065                 igc->rxq_stats_map[queue_id] = stat_idx;
2066         else
2067                 igc->txq_stats_map[queue_id] = stat_idx;
2068
2069         return 0;
2070 }
2071
2072 static int
2073 eth_igc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
2074 {
2075         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2076         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2077         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
2078         uint32_t vec = IGC_MISC_VEC_ID;
2079
2080         if (rte_intr_allow_others(intr_handle))
2081                 vec = IGC_RX_VEC_START;
2082
2083         uint32_t mask = 1u << (queue_id + vec);
2084
2085         IGC_WRITE_REG(hw, IGC_EIMC, mask);
2086         IGC_WRITE_FLUSH(hw);
2087
2088         return 0;
2089 }
2090
2091 static int
2092 eth_igc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
2093 {
2094         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2095         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2096         struct rte_intr_handle *intr_handle = pci_dev->intr_handle;
2097         uint32_t vec = IGC_MISC_VEC_ID;
2098
2099         if (rte_intr_allow_others(intr_handle))
2100                 vec = IGC_RX_VEC_START;
2101
2102         uint32_t mask = 1u << (queue_id + vec);
2103
2104         IGC_WRITE_REG(hw, IGC_EIMS, mask);
2105         IGC_WRITE_FLUSH(hw);
2106
2107         rte_intr_enable(intr_handle);
2108
2109         return 0;
2110 }
2111
2112 static int
2113 eth_igc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2114 {
2115         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2116         uint32_t ctrl;
2117         int tx_pause;
2118         int rx_pause;
2119
2120         fc_conf->pause_time = hw->fc.pause_time;
2121         fc_conf->high_water = hw->fc.high_water;
2122         fc_conf->low_water = hw->fc.low_water;
2123         fc_conf->send_xon = hw->fc.send_xon;
2124         fc_conf->autoneg = hw->mac.autoneg;
2125
2126         /*
2127          * Return rx_pause and tx_pause status according to actual setting of
2128          * the TFCE and RFCE bits in the CTRL register.
2129          */
2130         ctrl = IGC_READ_REG(hw, IGC_CTRL);
2131         if (ctrl & IGC_CTRL_TFCE)
2132                 tx_pause = 1;
2133         else
2134                 tx_pause = 0;
2135
2136         if (ctrl & IGC_CTRL_RFCE)
2137                 rx_pause = 1;
2138         else
2139                 rx_pause = 0;
2140
2141         if (rx_pause && tx_pause)
2142                 fc_conf->mode = RTE_ETH_FC_FULL;
2143         else if (rx_pause)
2144                 fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
2145         else if (tx_pause)
2146                 fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
2147         else
2148                 fc_conf->mode = RTE_ETH_FC_NONE;
2149
2150         return 0;
2151 }
2152
2153 static int
2154 eth_igc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2155 {
2156         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2157         uint32_t rx_buf_size;
2158         uint32_t max_high_water;
2159         uint32_t rctl;
2160         int err;
2161
2162         if (fc_conf->autoneg != hw->mac.autoneg)
2163                 return -ENOTSUP;
2164
2165         rx_buf_size = igc_get_rx_buffer_size(hw);
2166         PMD_DRV_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
2167
2168         /* At least reserve one Ethernet frame for watermark */
2169         max_high_water = rx_buf_size - RTE_ETHER_MAX_LEN;
2170         if (fc_conf->high_water > max_high_water ||
2171                 fc_conf->high_water < fc_conf->low_water) {
2172                 PMD_DRV_LOG(ERR,
2173                         "Incorrect high(%u)/low(%u) water value, max is %u",
2174                         fc_conf->high_water, fc_conf->low_water,
2175                         max_high_water);
2176                 return -EINVAL;
2177         }
2178
2179         switch (fc_conf->mode) {
2180         case RTE_ETH_FC_NONE:
2181                 hw->fc.requested_mode = igc_fc_none;
2182                 break;
2183         case RTE_ETH_FC_RX_PAUSE:
2184                 hw->fc.requested_mode = igc_fc_rx_pause;
2185                 break;
2186         case RTE_ETH_FC_TX_PAUSE:
2187                 hw->fc.requested_mode = igc_fc_tx_pause;
2188                 break;
2189         case RTE_ETH_FC_FULL:
2190                 hw->fc.requested_mode = igc_fc_full;
2191                 break;
2192         default:
2193                 PMD_DRV_LOG(ERR, "unsupported fc mode: %u", fc_conf->mode);
2194                 return -EINVAL;
2195         }
2196
2197         hw->fc.pause_time     = fc_conf->pause_time;
2198         hw->fc.high_water     = fc_conf->high_water;
2199         hw->fc.low_water      = fc_conf->low_water;
2200         hw->fc.send_xon       = fc_conf->send_xon;
2201
2202         err = igc_setup_link_generic(hw);
2203         if (err == IGC_SUCCESS) {
2204                 /**
2205                  * check if we want to forward MAC frames - driver doesn't have
2206                  * native capability to do that, so we'll write the registers
2207                  * ourselves
2208                  **/
2209                 rctl = IGC_READ_REG(hw, IGC_RCTL);
2210
2211                 /* set or clear MFLCN.PMCF bit depending on configuration */
2212                 if (fc_conf->mac_ctrl_frame_fwd != 0)
2213                         rctl |= IGC_RCTL_PMCF;
2214                 else
2215                         rctl &= ~IGC_RCTL_PMCF;
2216
2217                 IGC_WRITE_REG(hw, IGC_RCTL, rctl);
2218                 IGC_WRITE_FLUSH(hw);
2219
2220                 return 0;
2221         }
2222
2223         PMD_DRV_LOG(ERR, "igc_setup_link_generic = 0x%x", err);
2224         return -EIO;
2225 }
2226
2227 static int
2228 eth_igc_rss_reta_update(struct rte_eth_dev *dev,
2229                         struct rte_eth_rss_reta_entry64 *reta_conf,
2230                         uint16_t reta_size)
2231 {
2232         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2233         uint16_t i;
2234
2235         if (reta_size != RTE_ETH_RSS_RETA_SIZE_128) {
2236                 PMD_DRV_LOG(ERR,
2237                         "The size of RSS redirection table configured(%d) doesn't match the number hardware can supported(%d)",
2238                         reta_size, RTE_ETH_RSS_RETA_SIZE_128);
2239                 return -EINVAL;
2240         }
2241
2242         RTE_BUILD_BUG_ON(RTE_ETH_RSS_RETA_SIZE_128 % IGC_RSS_RDT_REG_SIZE);
2243
2244         /* set redirection table */
2245         for (i = 0; i < RTE_ETH_RSS_RETA_SIZE_128; i += IGC_RSS_RDT_REG_SIZE) {
2246                 union igc_rss_reta_reg reta, reg;
2247                 uint16_t idx, shift;
2248                 uint8_t j, mask;
2249
2250                 idx = i / RTE_ETH_RETA_GROUP_SIZE;
2251                 shift = i % RTE_ETH_RETA_GROUP_SIZE;
2252                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2253                                 IGC_RSS_RDT_REG_SIZE_MASK);
2254
2255                 /* if no need to update the register */
2256                 if (!mask ||
2257                     shift > (RTE_ETH_RETA_GROUP_SIZE - IGC_RSS_RDT_REG_SIZE))
2258                         continue;
2259
2260                 /* check mask whether need to read the register value first */
2261                 if (mask == IGC_RSS_RDT_REG_SIZE_MASK)
2262                         reg.dword = 0;
2263                 else
2264                         reg.dword = IGC_READ_REG_LE_VALUE(hw,
2265                                         IGC_RETA(i / IGC_RSS_RDT_REG_SIZE));
2266
2267                 /* update the register */
2268                 RTE_BUILD_BUG_ON(sizeof(reta.bytes) != IGC_RSS_RDT_REG_SIZE);
2269                 for (j = 0; j < IGC_RSS_RDT_REG_SIZE; j++) {
2270                         if (mask & (1u << j))
2271                                 reta.bytes[j] =
2272                                         (uint8_t)reta_conf[idx].reta[shift + j];
2273                         else
2274                                 reta.bytes[j] = reg.bytes[j];
2275                 }
2276                 IGC_WRITE_REG_LE_VALUE(hw,
2277                         IGC_RETA(i / IGC_RSS_RDT_REG_SIZE), reta.dword);
2278         }
2279
2280         return 0;
2281 }
2282
2283 static int
2284 eth_igc_rss_reta_query(struct rte_eth_dev *dev,
2285                        struct rte_eth_rss_reta_entry64 *reta_conf,
2286                        uint16_t reta_size)
2287 {
2288         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2289         uint16_t i;
2290
2291         if (reta_size != RTE_ETH_RSS_RETA_SIZE_128) {
2292                 PMD_DRV_LOG(ERR,
2293                         "The size of RSS redirection table configured(%d) doesn't match the number hardware can supported(%d)",
2294                         reta_size, RTE_ETH_RSS_RETA_SIZE_128);
2295                 return -EINVAL;
2296         }
2297
2298         RTE_BUILD_BUG_ON(RTE_ETH_RSS_RETA_SIZE_128 % IGC_RSS_RDT_REG_SIZE);
2299
2300         /* read redirection table */
2301         for (i = 0; i < RTE_ETH_RSS_RETA_SIZE_128; i += IGC_RSS_RDT_REG_SIZE) {
2302                 union igc_rss_reta_reg reta;
2303                 uint16_t idx, shift;
2304                 uint8_t j, mask;
2305
2306                 idx = i / RTE_ETH_RETA_GROUP_SIZE;
2307                 shift = i % RTE_ETH_RETA_GROUP_SIZE;
2308                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2309                                 IGC_RSS_RDT_REG_SIZE_MASK);
2310
2311                 /* if no need to read register */
2312                 if (!mask ||
2313                     shift > (RTE_ETH_RETA_GROUP_SIZE - IGC_RSS_RDT_REG_SIZE))
2314                         continue;
2315
2316                 /* read register and get the queue index */
2317                 RTE_BUILD_BUG_ON(sizeof(reta.bytes) != IGC_RSS_RDT_REG_SIZE);
2318                 reta.dword = IGC_READ_REG_LE_VALUE(hw,
2319                                 IGC_RETA(i / IGC_RSS_RDT_REG_SIZE));
2320                 for (j = 0; j < IGC_RSS_RDT_REG_SIZE; j++) {
2321                         if (mask & (1u << j))
2322                                 reta_conf[idx].reta[shift + j] = reta.bytes[j];
2323                 }
2324         }
2325
2326         return 0;
2327 }
2328
2329 static int
2330 eth_igc_rss_hash_update(struct rte_eth_dev *dev,
2331                         struct rte_eth_rss_conf *rss_conf)
2332 {
2333         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2334         igc_hw_rss_hash_set(hw, rss_conf);
2335         return 0;
2336 }
2337
2338 static int
2339 eth_igc_rss_hash_conf_get(struct rte_eth_dev *dev,
2340                         struct rte_eth_rss_conf *rss_conf)
2341 {
2342         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2343         uint32_t *hash_key = (uint32_t *)rss_conf->rss_key;
2344         uint32_t mrqc;
2345         uint64_t rss_hf;
2346
2347         if (hash_key != NULL) {
2348                 int i;
2349
2350                 /* if not enough space for store hash key */
2351                 if (rss_conf->rss_key_len != IGC_HKEY_SIZE) {
2352                         PMD_DRV_LOG(ERR,
2353                                 "RSS hash key size %u in parameter doesn't match the hardware hash key size %u",
2354                                 rss_conf->rss_key_len, IGC_HKEY_SIZE);
2355                         return -EINVAL;
2356                 }
2357
2358                 /* read RSS key from register */
2359                 for (i = 0; i < IGC_HKEY_MAX_INDEX; i++)
2360                         hash_key[i] = IGC_READ_REG_LE_VALUE(hw, IGC_RSSRK(i));
2361         }
2362
2363         /* get RSS functions configured in MRQC register */
2364         mrqc = IGC_READ_REG(hw, IGC_MRQC);
2365         if ((mrqc & IGC_MRQC_ENABLE_RSS_4Q) == 0)
2366                 return 0;
2367
2368         rss_hf = 0;
2369         if (mrqc & IGC_MRQC_RSS_FIELD_IPV4)
2370                 rss_hf |= RTE_ETH_RSS_IPV4;
2371         if (mrqc & IGC_MRQC_RSS_FIELD_IPV4_TCP)
2372                 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_TCP;
2373         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6)
2374                 rss_hf |= RTE_ETH_RSS_IPV6;
2375         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_EX)
2376                 rss_hf |= RTE_ETH_RSS_IPV6_EX;
2377         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_TCP)
2378                 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_TCP;
2379         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_TCP_EX)
2380                 rss_hf |= RTE_ETH_RSS_IPV6_TCP_EX;
2381         if (mrqc & IGC_MRQC_RSS_FIELD_IPV4_UDP)
2382                 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV4_UDP;
2383         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_UDP)
2384                 rss_hf |= RTE_ETH_RSS_NONFRAG_IPV6_UDP;
2385         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_UDP_EX)
2386                 rss_hf |= RTE_ETH_RSS_IPV6_UDP_EX;
2387
2388         rss_conf->rss_hf |= rss_hf;
2389         return 0;
2390 }
2391
2392 static int
2393 eth_igc_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2394 {
2395         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2396         struct igc_vfta *shadow_vfta = IGC_DEV_PRIVATE_VFTA(dev);
2397         uint32_t vfta;
2398         uint32_t vid_idx;
2399         uint32_t vid_bit;
2400
2401         vid_idx = (vlan_id >> IGC_VFTA_ENTRY_SHIFT) & IGC_VFTA_ENTRY_MASK;
2402         vid_bit = 1u << (vlan_id & IGC_VFTA_ENTRY_BIT_SHIFT_MASK);
2403         vfta = shadow_vfta->vfta[vid_idx];
2404         if (on)
2405                 vfta |= vid_bit;
2406         else
2407                 vfta &= ~vid_bit;
2408         IGC_WRITE_REG_ARRAY(hw, IGC_VFTA, vid_idx, vfta);
2409
2410         /* update local VFTA copy */
2411         shadow_vfta->vfta[vid_idx] = vfta;
2412
2413         return 0;
2414 }
2415
2416 static void
2417 igc_vlan_hw_filter_disable(struct rte_eth_dev *dev)
2418 {
2419         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2420         igc_read_reg_check_clear_bits(hw, IGC_RCTL,
2421                         IGC_RCTL_CFIEN | IGC_RCTL_VFE);
2422 }
2423
2424 static void
2425 igc_vlan_hw_filter_enable(struct rte_eth_dev *dev)
2426 {
2427         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2428         struct igc_vfta *shadow_vfta = IGC_DEV_PRIVATE_VFTA(dev);
2429         uint32_t reg_val;
2430         int i;
2431
2432         /* Filter Table Enable, CFI not used for packet acceptance */
2433         reg_val = IGC_READ_REG(hw, IGC_RCTL);
2434         reg_val &= ~IGC_RCTL_CFIEN;
2435         reg_val |= IGC_RCTL_VFE;
2436         IGC_WRITE_REG(hw, IGC_RCTL, reg_val);
2437
2438         /* restore VFTA table */
2439         for (i = 0; i < IGC_VFTA_SIZE; i++)
2440                 IGC_WRITE_REG_ARRAY(hw, IGC_VFTA, i, shadow_vfta->vfta[i]);
2441 }
2442
2443 static void
2444 igc_vlan_hw_strip_disable(struct rte_eth_dev *dev)
2445 {
2446         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2447
2448         igc_read_reg_check_clear_bits(hw, IGC_CTRL, IGC_CTRL_VME);
2449 }
2450
2451 static void
2452 igc_vlan_hw_strip_enable(struct rte_eth_dev *dev)
2453 {
2454         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2455
2456         igc_read_reg_check_set_bits(hw, IGC_CTRL, IGC_CTRL_VME);
2457 }
2458
2459 static int
2460 igc_vlan_hw_extend_disable(struct rte_eth_dev *dev)
2461 {
2462         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2463         uint32_t frame_size = dev->data->mtu + IGC_ETH_OVERHEAD;
2464         uint32_t ctrl_ext;
2465
2466         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
2467
2468         /* if extend vlan hasn't been enabled */
2469         if ((ctrl_ext & IGC_CTRL_EXT_EXT_VLAN) == 0)
2470                 return 0;
2471
2472         /* Update maximum packet length */
2473         if (frame_size < RTE_ETHER_MIN_MTU + VLAN_TAG_SIZE) {
2474                 PMD_DRV_LOG(ERR, "Maximum packet length %u error, min is %u",
2475                         frame_size, VLAN_TAG_SIZE + RTE_ETHER_MIN_MTU);
2476                 return -EINVAL;
2477         }
2478         IGC_WRITE_REG(hw, IGC_RLPML, frame_size - VLAN_TAG_SIZE);
2479
2480         IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext & ~IGC_CTRL_EXT_EXT_VLAN);
2481         return 0;
2482 }
2483
2484 static int
2485 igc_vlan_hw_extend_enable(struct rte_eth_dev *dev)
2486 {
2487         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2488         uint32_t frame_size = dev->data->mtu + IGC_ETH_OVERHEAD;
2489         uint32_t ctrl_ext;
2490
2491         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
2492
2493         /* if extend vlan has been enabled */
2494         if (ctrl_ext & IGC_CTRL_EXT_EXT_VLAN)
2495                 return 0;
2496
2497         /* Update maximum packet length */
2498         if (frame_size > MAX_RX_JUMBO_FRAME_SIZE) {
2499                 PMD_DRV_LOG(ERR, "Maximum packet length %u error, max is %u",
2500                         frame_size, MAX_RX_JUMBO_FRAME_SIZE);
2501                 return -EINVAL;
2502         }
2503         IGC_WRITE_REG(hw, IGC_RLPML, frame_size);
2504
2505         IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext | IGC_CTRL_EXT_EXT_VLAN);
2506         return 0;
2507 }
2508
2509 static int
2510 eth_igc_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2511 {
2512         struct rte_eth_rxmode *rxmode;
2513
2514         rxmode = &dev->data->dev_conf.rxmode;
2515         if (mask & RTE_ETH_VLAN_STRIP_MASK) {
2516                 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
2517                         igc_vlan_hw_strip_enable(dev);
2518                 else
2519                         igc_vlan_hw_strip_disable(dev);
2520         }
2521
2522         if (mask & RTE_ETH_VLAN_FILTER_MASK) {
2523                 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
2524                         igc_vlan_hw_filter_enable(dev);
2525                 else
2526                         igc_vlan_hw_filter_disable(dev);
2527         }
2528
2529         if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
2530                 if (rxmode->offloads & RTE_ETH_RX_OFFLOAD_VLAN_EXTEND)
2531                         return igc_vlan_hw_extend_enable(dev);
2532                 else
2533                         return igc_vlan_hw_extend_disable(dev);
2534         }
2535
2536         return 0;
2537 }
2538
2539 static int
2540 eth_igc_vlan_tpid_set(struct rte_eth_dev *dev,
2541                       enum rte_vlan_type vlan_type,
2542                       uint16_t tpid)
2543 {
2544         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2545         uint32_t reg_val;
2546
2547         /* only outer TPID of double VLAN can be configured*/
2548         if (vlan_type == RTE_ETH_VLAN_TYPE_OUTER) {
2549                 reg_val = IGC_READ_REG(hw, IGC_VET);
2550                 reg_val = (reg_val & (~IGC_VET_EXT)) |
2551                         ((uint32_t)tpid << IGC_VET_EXT_SHIFT);
2552                 IGC_WRITE_REG(hw, IGC_VET, reg_val);
2553
2554                 return 0;
2555         }
2556
2557         /* all other TPID values are read-only*/
2558         PMD_DRV_LOG(ERR, "Not supported");
2559         return -ENOTSUP;
2560 }
2561
2562 static int
2563 eth_igc_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2564         struct rte_pci_device *pci_dev)
2565 {
2566         PMD_INIT_FUNC_TRACE();
2567         return rte_eth_dev_pci_generic_probe(pci_dev,
2568                 sizeof(struct igc_adapter), eth_igc_dev_init);
2569 }
2570
2571 static int
2572 eth_igc_pci_remove(struct rte_pci_device *pci_dev)
2573 {
2574         PMD_INIT_FUNC_TRACE();
2575         return rte_eth_dev_pci_generic_remove(pci_dev, eth_igc_dev_uninit);
2576 }
2577
2578 static struct rte_pci_driver rte_igc_pmd = {
2579         .id_table = pci_id_igc_map,
2580         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
2581         .probe = eth_igc_pci_probe,
2582         .remove = eth_igc_pci_remove,
2583 };
2584
2585 RTE_PMD_REGISTER_PCI(net_igc, rte_igc_pmd);
2586 RTE_PMD_REGISTER_PCI_TABLE(net_igc, pci_id_igc_map);
2587 RTE_PMD_REGISTER_KMOD_DEP(net_igc, "* igb_uio | uio_pci_generic | vfio-pci");