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 /*
  24  * The overhead from MTU to max frame size.
  25  * Considering VLAN so tag needs to be counted.
  26  */
  27 #define IGC_ETH_OVERHEAD                (RTE_ETHER_HDR_LEN + \
  28                                         RTE_ETHER_CRC_LEN + VLAN_TAG_SIZE)
  29
  30 #define IGC_FC_PAUSE_TIME               0x0680
  31 #define IGC_LINK_UPDATE_CHECK_TIMEOUT   90  /* 9s */
  32 #define IGC_LINK_UPDATE_CHECK_INTERVAL  100 /* ms */
  33
  34 #define IGC_MISC_VEC_ID                 RTE_INTR_VEC_ZERO_OFFSET
  35 #define IGC_RX_VEC_START                RTE_INTR_VEC_RXTX_OFFSET
  36 #define IGC_MSIX_OTHER_INTR_VEC         0   /* MSI-X other interrupt vector */
  37 #define IGC_FLAG_NEED_LINK_UPDATE       (1u << 0)       /* need update link */
  38
  39 #define IGC_DEFAULT_RX_FREE_THRESH      32
  40
  41 #define IGC_DEFAULT_RX_PTHRESH          8
  42 #define IGC_DEFAULT_RX_HTHRESH          8
  43 #define IGC_DEFAULT_RX_WTHRESH          4
  44
  45 #define IGC_DEFAULT_TX_PTHRESH          8
  46 #define IGC_DEFAULT_TX_HTHRESH          1
  47 #define IGC_DEFAULT_TX_WTHRESH          16
  48
  49 /* MSI-X other interrupt vector */
  50 #define IGC_MSIX_OTHER_INTR_VEC         0
  51
  52 /* External VLAN Enable bit mask */
  53 #define IGC_CTRL_EXT_EXT_VLAN           (1u << 26)
  54
  55 /* Speed select */
  56 #define IGC_CTRL_SPEED_MASK             (7u << 8)
  57 #define IGC_CTRL_SPEED_2500             (6u << 8)
  58
  59 /* External VLAN Ether Type bit mask and shift */
  60 #define IGC_VET_EXT                     0xFFFF0000
  61 #define IGC_VET_EXT_SHIFT               16
  62
  63 /* Force EEE Auto-negotiation */
  64 #define IGC_EEER_EEE_FRC_AN             (1u << 28)
  65
  66 /* Per Queue Good Packets Received Count */
  67 #define IGC_PQGPRC(idx)         (0x10010 + 0x100 * (idx))
  68 /* Per Queue Good Octets Received Count */
  69 #define IGC_PQGORC(idx)         (0x10018 + 0x100 * (idx))
  70 /* Per Queue Good Octets Transmitted Count */
  71 #define IGC_PQGOTC(idx)         (0x10034 + 0x100 * (idx))
  72 /* Per Queue Multicast Packets Received Count */
  73 #define IGC_PQMPRC(idx)         (0x10038 + 0x100 * (idx))
  74 /* Transmit Queue Drop Packet Count */
  75 #define IGC_TQDPC(idx)          (0xe030 + 0x40 * (idx))
  76
  77 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
  78 #define U32_0_IN_U64            0       /* lower bytes of u64 */
  79 #define U32_1_IN_U64            1       /* higher bytes of u64 */
  80 #else
  81 #define U32_0_IN_U64            1
  82 #define U32_1_IN_U64            0
  83 #endif
  84
  85 #define IGC_ALARM_INTERVAL      8000000u
  86 /* us, about 13.6s some per-queue registers will wrap around back to 0. */
  87
  88 static const struct rte_eth_desc_lim rx_desc_lim = {
  89         .nb_max = IGC_MAX_RXD,
  90         .nb_min = IGC_MIN_RXD,
  91         .nb_align = IGC_RXD_ALIGN,
  92 };
  93
  94 static const struct rte_eth_desc_lim tx_desc_lim = {
  95         .nb_max = IGC_MAX_TXD,
  96         .nb_min = IGC_MIN_TXD,
  97         .nb_align = IGC_TXD_ALIGN,
  98         .nb_seg_max = IGC_TX_MAX_SEG,
  99         .nb_mtu_seg_max = IGC_TX_MAX_MTU_SEG,
 100 };
 101
 102 static const struct rte_pci_id pci_id_igc_map[] = {
 103         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_LM) },
 104         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_V)  },
 105         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_I)  },
 106         { RTE_PCI_DEVICE(IGC_INTEL_VENDOR_ID, IGC_DEV_ID_I225_K)  },
 107         { .vendor_id = 0, /* sentinel */ },
 108 };
 109
 110 /* store statistics names and its offset in stats structure */
 111 struct rte_igc_xstats_name_off {
 112         char name[RTE_ETH_XSTATS_NAME_SIZE];
 113         unsigned int offset;
 114 };
 115
 116 static const struct rte_igc_xstats_name_off rte_igc_stats_strings[] = {
 117         {"rx_crc_errors", offsetof(struct igc_hw_stats, crcerrs)},
 118         {"rx_align_errors", offsetof(struct igc_hw_stats, algnerrc)},
 119         {"rx_errors", offsetof(struct igc_hw_stats, rxerrc)},
 120         {"rx_missed_packets", offsetof(struct igc_hw_stats, mpc)},
 121         {"tx_single_collision_packets", offsetof(struct igc_hw_stats, scc)},
 122         {"tx_multiple_collision_packets", offsetof(struct igc_hw_stats, mcc)},
 123         {"tx_excessive_collision_packets", offsetof(struct igc_hw_stats,
 124                 ecol)},
 125         {"tx_late_collisions", offsetof(struct igc_hw_stats, latecol)},
 126         {"tx_total_collisions", offsetof(struct igc_hw_stats, colc)},
 127         {"tx_deferred_packets", offsetof(struct igc_hw_stats, dc)},
 128         {"tx_no_carrier_sense_packets", offsetof(struct igc_hw_stats, tncrs)},
 129         {"tx_discarded_packets", offsetof(struct igc_hw_stats, htdpmc)},
 130         {"rx_length_errors", offsetof(struct igc_hw_stats, rlec)},
 131         {"rx_xon_packets", offsetof(struct igc_hw_stats, xonrxc)},
 132         {"tx_xon_packets", offsetof(struct igc_hw_stats, xontxc)},
 133         {"rx_xoff_packets", offsetof(struct igc_hw_stats, xoffrxc)},
 134         {"tx_xoff_packets", offsetof(struct igc_hw_stats, xofftxc)},
 135         {"rx_flow_control_unsupported_packets", offsetof(struct igc_hw_stats,
 136                 fcruc)},
 137         {"rx_size_64_packets", offsetof(struct igc_hw_stats, prc64)},
 138         {"rx_size_65_to_127_packets", offsetof(struct igc_hw_stats, prc127)},
 139         {"rx_size_128_to_255_packets", offsetof(struct igc_hw_stats, prc255)},
 140         {"rx_size_256_to_511_packets", offsetof(struct igc_hw_stats, prc511)},
 141         {"rx_size_512_to_1023_packets", offsetof(struct igc_hw_stats,
 142                 prc1023)},
 143         {"rx_size_1024_to_max_packets", offsetof(struct igc_hw_stats,
 144                 prc1522)},
 145         {"rx_broadcast_packets", offsetof(struct igc_hw_stats, bprc)},
 146         {"rx_multicast_packets", offsetof(struct igc_hw_stats, mprc)},
 147         {"rx_undersize_errors", offsetof(struct igc_hw_stats, ruc)},
 148         {"rx_fragment_errors", offsetof(struct igc_hw_stats, rfc)},
 149         {"rx_oversize_errors", offsetof(struct igc_hw_stats, roc)},
 150         {"rx_jabber_errors", offsetof(struct igc_hw_stats, rjc)},
 151         {"rx_no_buffers", offsetof(struct igc_hw_stats, rnbc)},
 152         {"rx_management_packets", offsetof(struct igc_hw_stats, mgprc)},
 153         {"rx_management_dropped", offsetof(struct igc_hw_stats, mgpdc)},
 154         {"tx_management_packets", offsetof(struct igc_hw_stats, mgptc)},
 155         {"rx_total_packets", offsetof(struct igc_hw_stats, tpr)},
 156         {"tx_total_packets", offsetof(struct igc_hw_stats, tpt)},
 157         {"rx_total_bytes", offsetof(struct igc_hw_stats, tor)},
 158         {"tx_total_bytes", offsetof(struct igc_hw_stats, tot)},
 159         {"tx_size_64_packets", offsetof(struct igc_hw_stats, ptc64)},
 160         {"tx_size_65_to_127_packets", offsetof(struct igc_hw_stats, ptc127)},
 161         {"tx_size_128_to_255_packets", offsetof(struct igc_hw_stats, ptc255)},
 162         {"tx_size_256_to_511_packets", offsetof(struct igc_hw_stats, ptc511)},
 163         {"tx_size_512_to_1023_packets", offsetof(struct igc_hw_stats,
 164                 ptc1023)},
 165         {"tx_size_1023_to_max_packets", offsetof(struct igc_hw_stats,
 166                 ptc1522)},
 167         {"tx_multicast_packets", offsetof(struct igc_hw_stats, mptc)},
 168         {"tx_broadcast_packets", offsetof(struct igc_hw_stats, bptc)},
 169         {"tx_tso_packets", offsetof(struct igc_hw_stats, tsctc)},
 170         {"rx_sent_to_host_packets", offsetof(struct igc_hw_stats, rpthc)},
 171         {"tx_sent_by_host_packets", offsetof(struct igc_hw_stats, hgptc)},
 172         {"interrupt_assert_count", offsetof(struct igc_hw_stats, iac)},
 173         {"rx_descriptor_lower_threshold",
 174                 offsetof(struct igc_hw_stats, icrxdmtc)},
 175 };
 176
 177 #define IGC_NB_XSTATS (sizeof(rte_igc_stats_strings) / \
 178                 sizeof(rte_igc_stats_strings[0]))
 179
 180 static int eth_igc_configure(struct rte_eth_dev *dev);
 181 static int eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete);
 182 static int eth_igc_stop(struct rte_eth_dev *dev);
 183 static int eth_igc_start(struct rte_eth_dev *dev);
 184 static int eth_igc_set_link_up(struct rte_eth_dev *dev);
 185 static int eth_igc_set_link_down(struct rte_eth_dev *dev);
 186 static int eth_igc_close(struct rte_eth_dev *dev);
 187 static int eth_igc_reset(struct rte_eth_dev *dev);
 188 static int eth_igc_promiscuous_enable(struct rte_eth_dev *dev);
 189 static int eth_igc_promiscuous_disable(struct rte_eth_dev *dev);
 190 static int eth_igc_fw_version_get(struct rte_eth_dev *dev,
 191                                 char *fw_version, size_t fw_size);
 192 static int eth_igc_infos_get(struct rte_eth_dev *dev,
 193                         struct rte_eth_dev_info *dev_info);
 194 static int eth_igc_led_on(struct rte_eth_dev *dev);
 195 static int eth_igc_led_off(struct rte_eth_dev *dev);
 196 static const uint32_t *eth_igc_supported_ptypes_get(struct rte_eth_dev *dev);
 197 static int eth_igc_rar_set(struct rte_eth_dev *dev,
 198                 struct rte_ether_addr *mac_addr, uint32_t index, uint32_t pool);
 199 static void eth_igc_rar_clear(struct rte_eth_dev *dev, uint32_t index);
 200 static int eth_igc_default_mac_addr_set(struct rte_eth_dev *dev,
 201                         struct rte_ether_addr *addr);
 202 static int eth_igc_set_mc_addr_list(struct rte_eth_dev *dev,
 203                          struct rte_ether_addr *mc_addr_set,
 204                          uint32_t nb_mc_addr);
 205 static int eth_igc_allmulticast_enable(struct rte_eth_dev *dev);
 206 static int eth_igc_allmulticast_disable(struct rte_eth_dev *dev);
 207 static int eth_igc_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
 208 static int eth_igc_stats_get(struct rte_eth_dev *dev,
 209                         struct rte_eth_stats *rte_stats);
 210 static int eth_igc_xstats_get(struct rte_eth_dev *dev,
 211                         struct rte_eth_xstat *xstats, unsigned int n);
 212 static int eth_igc_xstats_get_by_id(struct rte_eth_dev *dev,
 213                                 const uint64_t *ids,
 214                                 uint64_t *values, unsigned int n);
 215 static int eth_igc_xstats_get_names(struct rte_eth_dev *dev,
 216                                 struct rte_eth_xstat_name *xstats_names,
 217                                 unsigned int size);
 218 static int eth_igc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 219                 struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
 220                 unsigned int limit);
 221 static int eth_igc_xstats_reset(struct rte_eth_dev *dev);
 222 static int
 223 eth_igc_queue_stats_mapping_set(struct rte_eth_dev *dev,
 224         uint16_t queue_id, uint8_t stat_idx, uint8_t is_rx);
 225 static int
 226 eth_igc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id);
 227 static int
 228 eth_igc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id);
 229 static int
 230 eth_igc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
 231 static int
 232 eth_igc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf);
 233 static int eth_igc_rss_reta_update(struct rte_eth_dev *dev,
 234                         struct rte_eth_rss_reta_entry64 *reta_conf,
 235                         uint16_t reta_size);
 236 static int eth_igc_rss_reta_query(struct rte_eth_dev *dev,
 237                        struct rte_eth_rss_reta_entry64 *reta_conf,
 238                        uint16_t reta_size);
 239 static int eth_igc_rss_hash_update(struct rte_eth_dev *dev,
 240                         struct rte_eth_rss_conf *rss_conf);
 241 static int eth_igc_rss_hash_conf_get(struct rte_eth_dev *dev,
 242                         struct rte_eth_rss_conf *rss_conf);
 243 static int
 244 eth_igc_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on);
 245 static int eth_igc_vlan_offload_set(struct rte_eth_dev *dev, int mask);
 246 static int eth_igc_vlan_tpid_set(struct rte_eth_dev *dev,
 247                       enum rte_vlan_type vlan_type, uint16_t tpid);
 248
 249 static const struct eth_dev_ops eth_igc_ops = {
 250         .dev_configure          = eth_igc_configure,
 251         .link_update            = eth_igc_link_update,
 252         .dev_stop               = eth_igc_stop,
 253         .dev_start              = eth_igc_start,
 254         .dev_close              = eth_igc_close,
 255         .dev_reset              = eth_igc_reset,
 256         .dev_set_link_up        = eth_igc_set_link_up,
 257         .dev_set_link_down      = eth_igc_set_link_down,
 258         .promiscuous_enable     = eth_igc_promiscuous_enable,
 259         .promiscuous_disable    = eth_igc_promiscuous_disable,
 260         .allmulticast_enable    = eth_igc_allmulticast_enable,
 261         .allmulticast_disable   = eth_igc_allmulticast_disable,
 262         .fw_version_get         = eth_igc_fw_version_get,
 263         .dev_infos_get          = eth_igc_infos_get,
 264         .dev_led_on             = eth_igc_led_on,
 265         .dev_led_off            = eth_igc_led_off,
 266         .dev_supported_ptypes_get = eth_igc_supported_ptypes_get,
 267         .mtu_set                = eth_igc_mtu_set,
 268         .mac_addr_add           = eth_igc_rar_set,
 269         .mac_addr_remove        = eth_igc_rar_clear,
 270         .mac_addr_set           = eth_igc_default_mac_addr_set,
 271         .set_mc_addr_list       = eth_igc_set_mc_addr_list,
 272
 273         .rx_queue_setup         = eth_igc_rx_queue_setup,
 274         .rx_queue_release       = eth_igc_rx_queue_release,
 275         .tx_queue_setup         = eth_igc_tx_queue_setup,
 276         .tx_queue_release       = eth_igc_tx_queue_release,
 277         .tx_done_cleanup        = eth_igc_tx_done_cleanup,
 278         .rxq_info_get           = eth_igc_rxq_info_get,
 279         .txq_info_get           = eth_igc_txq_info_get,
 280         .stats_get              = eth_igc_stats_get,
 281         .xstats_get             = eth_igc_xstats_get,
 282         .xstats_get_by_id       = eth_igc_xstats_get_by_id,
 283         .xstats_get_names_by_id = eth_igc_xstats_get_names_by_id,
 284         .xstats_get_names       = eth_igc_xstats_get_names,
 285         .stats_reset            = eth_igc_xstats_reset,
 286         .xstats_reset           = eth_igc_xstats_reset,
 287         .queue_stats_mapping_set = eth_igc_queue_stats_mapping_set,
 288         .rx_queue_intr_enable   = eth_igc_rx_queue_intr_enable,
 289         .rx_queue_intr_disable  = eth_igc_rx_queue_intr_disable,
 290         .flow_ctrl_get          = eth_igc_flow_ctrl_get,
 291         .flow_ctrl_set          = eth_igc_flow_ctrl_set,
 292         .reta_update            = eth_igc_rss_reta_update,
 293         .reta_query             = eth_igc_rss_reta_query,
 294         .rss_hash_update        = eth_igc_rss_hash_update,
 295         .rss_hash_conf_get      = eth_igc_rss_hash_conf_get,
 296         .vlan_filter_set        = eth_igc_vlan_filter_set,
 297         .vlan_offload_set       = eth_igc_vlan_offload_set,
 298         .vlan_tpid_set          = eth_igc_vlan_tpid_set,
 299         .vlan_strip_queue_set   = eth_igc_vlan_strip_queue_set,
 300         .flow_ops_get           = eth_igc_flow_ops_get,
 301 };
 302
 303 /*
 304  * multiple queue mode checking
 305  */
 306 static int
 307 igc_check_mq_mode(struct rte_eth_dev *dev)
 308 {
 309         enum rte_eth_rx_mq_mode rx_mq_mode = dev->data->dev_conf.rxmode.mq_mode;
 310         enum rte_eth_tx_mq_mode tx_mq_mode = dev->data->dev_conf.txmode.mq_mode;
 311
 312         if (RTE_ETH_DEV_SRIOV(dev).active != 0) {
 313                 PMD_INIT_LOG(ERR, "SRIOV is not supported.");
 314                 return -EINVAL;
 315         }
 316
 317         if (rx_mq_mode != ETH_MQ_RX_NONE &&
 318                 rx_mq_mode != ETH_MQ_RX_RSS) {
 319                 /* RSS together with VMDq not supported*/
 320                 PMD_INIT_LOG(ERR, "RX mode %d is not supported.",
 321                                 rx_mq_mode);
 322                 return -EINVAL;
 323         }
 324
 325         /* To no break software that set invalid mode, only display
 326          * warning if invalid mode is used.
 327          */
 328         if (tx_mq_mode != ETH_MQ_TX_NONE)
 329                 PMD_INIT_LOG(WARNING,
 330                         "TX mode %d is not supported. Due to meaningless in this driver, just ignore",
 331                         tx_mq_mode);
 332
 333         return 0;
 334 }
 335
 336 static int
 337 eth_igc_configure(struct rte_eth_dev *dev)
 338 {
 339         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 340         int ret;
 341
 342         PMD_INIT_FUNC_TRACE();
 343
 344         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 345                 dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
 346
 347         ret  = igc_check_mq_mode(dev);
 348         if (ret != 0)
 349                 return ret;
 350
 351         intr->flags |= IGC_FLAG_NEED_LINK_UPDATE;
 352         return 0;
 353 }
 354
 355 static int
 356 eth_igc_set_link_up(struct rte_eth_dev *dev)
 357 {
 358         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 359
 360         if (hw->phy.media_type == igc_media_type_copper)
 361                 igc_power_up_phy(hw);
 362         else
 363                 igc_power_up_fiber_serdes_link(hw);
 364         return 0;
 365 }
 366
 367 static int
 368 eth_igc_set_link_down(struct rte_eth_dev *dev)
 369 {
 370         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 371
 372         if (hw->phy.media_type == igc_media_type_copper)
 373                 igc_power_down_phy(hw);
 374         else
 375                 igc_shutdown_fiber_serdes_link(hw);
 376         return 0;
 377 }
 378
 379 /*
 380  * disable other interrupt
 381  */
 382 static void
 383 igc_intr_other_disable(struct rte_eth_dev *dev)
 384 {
 385         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 386         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 387         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 388
 389         if (rte_intr_allow_others(intr_handle) &&
 390                 dev->data->dev_conf.intr_conf.lsc) {
 391                 IGC_WRITE_REG(hw, IGC_EIMC, 1u << IGC_MSIX_OTHER_INTR_VEC);
 392         }
 393
 394         IGC_WRITE_REG(hw, IGC_IMC, ~0);
 395         IGC_WRITE_FLUSH(hw);
 396 }
 397
 398 /*
 399  * enable other interrupt
 400  */
 401 static inline void
 402 igc_intr_other_enable(struct rte_eth_dev *dev)
 403 {
 404         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 405         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 406         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 407         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 408
 409         if (rte_intr_allow_others(intr_handle) &&
 410                 dev->data->dev_conf.intr_conf.lsc) {
 411                 IGC_WRITE_REG(hw, IGC_EIMS, 1u << IGC_MSIX_OTHER_INTR_VEC);
 412         }
 413
 414         IGC_WRITE_REG(hw, IGC_IMS, intr->mask);
 415         IGC_WRITE_FLUSH(hw);
 416 }
 417
 418 /*
 419  * It reads ICR and gets interrupt causes, check it and set a bit flag
 420  * to update link status.
 421  */
 422 static void
 423 eth_igc_interrupt_get_status(struct rte_eth_dev *dev)
 424 {
 425         uint32_t icr;
 426         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 427         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 428
 429         /* read-on-clear nic registers here */
 430         icr = IGC_READ_REG(hw, IGC_ICR);
 431
 432         intr->flags = 0;
 433         if (icr & IGC_ICR_LSC)
 434                 intr->flags |= IGC_FLAG_NEED_LINK_UPDATE;
 435 }
 436
 437 /* return 0 means link status changed, -1 means not changed */
 438 static int
 439 eth_igc_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 440 {
 441         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 442         struct rte_eth_link link;
 443         int link_check, count;
 444
 445         link_check = 0;
 446         hw->mac.get_link_status = 1;
 447
 448         /* possible wait-to-complete in up to 9 seconds */
 449         for (count = 0; count < IGC_LINK_UPDATE_CHECK_TIMEOUT; count++) {
 450                 /* Read the real link status */
 451                 switch (hw->phy.media_type) {
 452                 case igc_media_type_copper:
 453                         /* Do the work to read phy */
 454                         igc_check_for_link(hw);
 455                         link_check = !hw->mac.get_link_status;
 456                         break;
 457
 458                 case igc_media_type_fiber:
 459                         igc_check_for_link(hw);
 460                         link_check = (IGC_READ_REG(hw, IGC_STATUS) &
 461                                       IGC_STATUS_LU);
 462                         break;
 463
 464                 case igc_media_type_internal_serdes:
 465                         igc_check_for_link(hw);
 466                         link_check = hw->mac.serdes_has_link;
 467                         break;
 468
 469                 default:
 470                         break;
 471                 }
 472                 if (link_check || wait_to_complete == 0)
 473                         break;
 474                 rte_delay_ms(IGC_LINK_UPDATE_CHECK_INTERVAL);
 475         }
 476         memset(&link, 0, sizeof(link));
 477
 478         /* Now we check if a transition has happened */
 479         if (link_check) {
 480                 uint16_t duplex, speed;
 481                 hw->mac.ops.get_link_up_info(hw, &speed, &duplex);
 482                 link.link_duplex = (duplex == FULL_DUPLEX) ?
 483                                 ETH_LINK_FULL_DUPLEX :
 484                                 ETH_LINK_HALF_DUPLEX;
 485                 link.link_speed = speed;
 486                 link.link_status = ETH_LINK_UP;
 487                 link.link_autoneg = !(dev->data->dev_conf.link_speeds &
 488                                 ETH_LINK_SPEED_FIXED);
 489
 490                 if (speed == SPEED_2500) {
 491                         uint32_t tipg = IGC_READ_REG(hw, IGC_TIPG);
 492                         if ((tipg & IGC_TIPG_IPGT_MASK) != 0x0b) {
 493                                 tipg &= ~IGC_TIPG_IPGT_MASK;
 494                                 tipg |= 0x0b;
 495                                 IGC_WRITE_REG(hw, IGC_TIPG, tipg);
 496                         }
 497                 }
 498         } else {
 499                 link.link_speed = 0;
 500                 link.link_duplex = ETH_LINK_HALF_DUPLEX;
 501                 link.link_status = ETH_LINK_DOWN;
 502                 link.link_autoneg = ETH_LINK_FIXED;
 503         }
 504
 505         return rte_eth_linkstatus_set(dev, &link);
 506 }
 507
 508 /*
 509  * It executes link_update after knowing an interrupt is present.
 510  */
 511 static void
 512 eth_igc_interrupt_action(struct rte_eth_dev *dev)
 513 {
 514         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 515         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 516         struct rte_eth_link link;
 517         int ret;
 518
 519         if (intr->flags & IGC_FLAG_NEED_LINK_UPDATE) {
 520                 intr->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
 521
 522                 /* set get_link_status to check register later */
 523                 ret = eth_igc_link_update(dev, 0);
 524
 525                 /* check if link has changed */
 526                 if (ret < 0)
 527                         return;
 528
 529                 rte_eth_linkstatus_get(dev, &link);
 530                 if (link.link_status)
 531                         PMD_DRV_LOG(INFO,
 532                                 " Port %d: Link Up - speed %u Mbps - %s",
 533                                 dev->data->port_id,
 534                                 (unsigned int)link.link_speed,
 535                                 link.link_duplex == ETH_LINK_FULL_DUPLEX ?
 536                                 "full-duplex" : "half-duplex");
 537                 else
 538                         PMD_DRV_LOG(INFO, " Port %d: Link Down",
 539                                 dev->data->port_id);
 540
 541                 PMD_DRV_LOG(DEBUG, "PCI Address: " PCI_PRI_FMT,
 542                                 pci_dev->addr.domain,
 543                                 pci_dev->addr.bus,
 544                                 pci_dev->addr.devid,
 545                                 pci_dev->addr.function);
 546                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_LSC, NULL);
 547         }
 548 }
 549
 550 /*
 551  * Interrupt handler which shall be registered at first.
 552  *
 553  * @handle
 554  *  Pointer to interrupt handle.
 555  * @param
 556  *  The address of parameter (struct rte_eth_dev *) registered before.
 557  */
 558 static void
 559 eth_igc_interrupt_handler(void *param)
 560 {
 561         struct rte_eth_dev *dev = (struct rte_eth_dev *)param;
 562
 563         eth_igc_interrupt_get_status(dev);
 564         eth_igc_interrupt_action(dev);
 565 }
 566
 567 static void igc_read_queue_stats_register(struct rte_eth_dev *dev);
 568
 569 /*
 570  * Update the queue status every IGC_ALARM_INTERVAL time.
 571  * @param
 572  *  The address of parameter (struct rte_eth_dev *) registered before.
 573  */
 574 static void
 575 igc_update_queue_stats_handler(void *param)
 576 {
 577         struct rte_eth_dev *dev = param;
 578         igc_read_queue_stats_register(dev);
 579         rte_eal_alarm_set(IGC_ALARM_INTERVAL,
 580                         igc_update_queue_stats_handler, dev);
 581 }
 582
 583 /*
 584  * rx,tx enable/disable
 585  */
 586 static void
 587 eth_igc_rxtx_control(struct rte_eth_dev *dev, bool enable)
 588 {
 589         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 590         uint32_t tctl, rctl;
 591
 592         tctl = IGC_READ_REG(hw, IGC_TCTL);
 593         rctl = IGC_READ_REG(hw, IGC_RCTL);
 594
 595         if (enable) {
 596                 /* enable Tx/Rx */
 597                 tctl |= IGC_TCTL_EN;
 598                 rctl |= IGC_RCTL_EN;
 599         } else {
 600                 /* disable Tx/Rx */
 601                 tctl &= ~IGC_TCTL_EN;
 602                 rctl &= ~IGC_RCTL_EN;
 603         }
 604         IGC_WRITE_REG(hw, IGC_TCTL, tctl);
 605         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
 606         IGC_WRITE_FLUSH(hw);
 607 }
 608
 609 /*
 610  *  This routine disables all traffic on the adapter by issuing a
 611  *  global reset on the MAC.
 612  */
 613 static int
 614 eth_igc_stop(struct rte_eth_dev *dev)
 615 {
 616         struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
 617         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 618         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 619         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 620         struct rte_eth_link link;
 621
 622         dev->data->dev_started = 0;
 623         adapter->stopped = 1;
 624
 625         /* disable receive and transmit */
 626         eth_igc_rxtx_control(dev, false);
 627
 628         /* disable all MSI-X interrupts */
 629         IGC_WRITE_REG(hw, IGC_EIMC, 0x1f);
 630         IGC_WRITE_FLUSH(hw);
 631
 632         /* clear all MSI-X interrupts */
 633         IGC_WRITE_REG(hw, IGC_EICR, 0x1f);
 634
 635         igc_intr_other_disable(dev);
 636
 637         rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
 638
 639         /* disable intr eventfd mapping */
 640         rte_intr_disable(intr_handle);
 641
 642         igc_reset_hw(hw);
 643
 644         /* disable all wake up */
 645         IGC_WRITE_REG(hw, IGC_WUC, 0);
 646
 647         /* disable checking EEE operation in MAC loopback mode */
 648         igc_read_reg_check_clear_bits(hw, IGC_EEER, IGC_EEER_EEE_FRC_AN);
 649
 650         /* Set bit for Go Link disconnect */
 651         igc_read_reg_check_set_bits(hw, IGC_82580_PHY_POWER_MGMT,
 652                         IGC_82580_PM_GO_LINKD);
 653
 654         /* Power down the phy. Needed to make the link go Down */
 655         eth_igc_set_link_down(dev);
 656
 657         igc_dev_clear_queues(dev);
 658
 659         /* clear the recorded link status */
 660         memset(&link, 0, sizeof(link));
 661         rte_eth_linkstatus_set(dev, &link);
 662
 663         if (!rte_intr_allow_others(intr_handle))
 664                 /* resume to the default handler */
 665                 rte_intr_callback_register(intr_handle,
 666                                            eth_igc_interrupt_handler,
 667                                            (void *)dev);
 668
 669         /* Clean datapath event and queue/vec mapping */
 670         rte_intr_efd_disable(intr_handle);
 671         if (intr_handle->intr_vec != NULL) {
 672                 rte_free(intr_handle->intr_vec);
 673                 intr_handle->intr_vec = NULL;
 674         }
 675
 676         return 0;
 677 }
 678
 679 /*
 680  * write interrupt vector allocation register
 681  * @hw
 682  *  board private structure
 683  * @queue_index
 684  *  queue index, valid 0,1,2,3
 685  * @tx
 686  *  tx:1, rx:0
 687  * @msix_vector
 688  *  msix-vector, valid 0,1,2,3,4
 689  */
 690 static void
 691 igc_write_ivar(struct igc_hw *hw, uint8_t queue_index,
 692                 bool tx, uint8_t msix_vector)
 693 {
 694         uint8_t offset = 0;
 695         uint8_t reg_index = queue_index >> 1;
 696         uint32_t val;
 697
 698         /*
 699          * IVAR(0)
 700          * bit31...24   bit23...16      bit15...8       bit7...0
 701          * TX1          RX1             TX0             RX0
 702          *
 703          * IVAR(1)
 704          * bit31...24   bit23...16      bit15...8       bit7...0
 705          * TX3          RX3             TX2             RX2
 706          */
 707
 708         if (tx)
 709                 offset = 8;
 710
 711         if (queue_index & 1)
 712                 offset += 16;
 713
 714         val = IGC_READ_REG_ARRAY(hw, IGC_IVAR0, reg_index);
 715
 716         /* clear bits */
 717         val &= ~((uint32_t)0xFF << offset);
 718
 719         /* write vector and valid bit */
 720         val |= (uint32_t)(msix_vector | IGC_IVAR_VALID) << offset;
 721
 722         IGC_WRITE_REG_ARRAY(hw, IGC_IVAR0, reg_index, val);
 723 }
 724
 725 /* Sets up the hardware to generate MSI-X interrupts properly
 726  * @hw
 727  *  board private structure
 728  */
 729 static void
 730 igc_configure_msix_intr(struct rte_eth_dev *dev)
 731 {
 732         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 733         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 734         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 735
 736         uint32_t intr_mask;
 737         uint32_t vec = IGC_MISC_VEC_ID;
 738         uint32_t base = IGC_MISC_VEC_ID;
 739         uint32_t misc_shift = 0;
 740         int i;
 741
 742         /* won't configure msix register if no mapping is done
 743          * between intr vector and event fd
 744          */
 745         if (!rte_intr_dp_is_en(intr_handle))
 746                 return;
 747
 748         if (rte_intr_allow_others(intr_handle)) {
 749                 base = IGC_RX_VEC_START;
 750                 vec = base;
 751                 misc_shift = 1;
 752         }
 753
 754         /* turn on MSI-X capability first */
 755         IGC_WRITE_REG(hw, IGC_GPIE, IGC_GPIE_MSIX_MODE |
 756                                 IGC_GPIE_PBA | IGC_GPIE_EIAME |
 757                                 IGC_GPIE_NSICR);
 758         intr_mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) <<
 759                 misc_shift;
 760
 761         if (dev->data->dev_conf.intr_conf.lsc)
 762                 intr_mask |= (1u << IGC_MSIX_OTHER_INTR_VEC);
 763
 764         /* enable msix auto-clear */
 765         igc_read_reg_check_set_bits(hw, IGC_EIAC, intr_mask);
 766
 767         /* set other cause interrupt vector */
 768         igc_read_reg_check_set_bits(hw, IGC_IVAR_MISC,
 769                 (uint32_t)(IGC_MSIX_OTHER_INTR_VEC | IGC_IVAR_VALID) << 8);
 770
 771         /* enable auto-mask */
 772         igc_read_reg_check_set_bits(hw, IGC_EIAM, intr_mask);
 773
 774         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 775                 igc_write_ivar(hw, i, 0, vec);
 776                 intr_handle->intr_vec[i] = vec;
 777                 if (vec < base + intr_handle->nb_efd - 1)
 778                         vec++;
 779         }
 780
 781         IGC_WRITE_FLUSH(hw);
 782 }
 783
 784 /**
 785  * It enables the interrupt mask and then enable the interrupt.
 786  *
 787  * @dev
 788  *  Pointer to struct rte_eth_dev.
 789  * @on
 790  *  Enable or Disable
 791  */
 792 static void
 793 igc_lsc_interrupt_setup(struct rte_eth_dev *dev, uint8_t on)
 794 {
 795         struct igc_interrupt *intr = IGC_DEV_PRIVATE_INTR(dev);
 796
 797         if (on)
 798                 intr->mask |= IGC_ICR_LSC;
 799         else
 800                 intr->mask &= ~IGC_ICR_LSC;
 801 }
 802
 803 /*
 804  * It enables the interrupt.
 805  * It will be called once only during nic initialized.
 806  */
 807 static void
 808 igc_rxq_interrupt_setup(struct rte_eth_dev *dev)
 809 {
 810         uint32_t mask;
 811         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 812         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 813         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 814         int misc_shift = rte_intr_allow_others(intr_handle) ? 1 : 0;
 815
 816         /* won't configure msix register if no mapping is done
 817          * between intr vector and event fd
 818          */
 819         if (!rte_intr_dp_is_en(intr_handle))
 820                 return;
 821
 822         mask = RTE_LEN2MASK(intr_handle->nb_efd, uint32_t) << misc_shift;
 823         IGC_WRITE_REG(hw, IGC_EIMS, mask);
 824 }
 825
 826 /*
 827  *  Get hardware rx-buffer size.
 828  */
 829 static inline int
 830 igc_get_rx_buffer_size(struct igc_hw *hw)
 831 {
 832         return (IGC_READ_REG(hw, IGC_RXPBS) & 0x3f) << 10;
 833 }
 834
 835 /*
 836  * igc_hw_control_acquire sets CTRL_EXT:DRV_LOAD bit.
 837  * For ASF and Pass Through versions of f/w this means
 838  * that the driver is loaded.
 839  */
 840 static void
 841 igc_hw_control_acquire(struct igc_hw *hw)
 842 {
 843         uint32_t ctrl_ext;
 844
 845         /* Let firmware know the driver has taken over */
 846         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
 847         IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
 848 }
 849
 850 /*
 851  * igc_hw_control_release resets CTRL_EXT:DRV_LOAD bit.
 852  * For ASF and Pass Through versions of f/w this means that the
 853  * driver is no longer loaded.
 854  */
 855 static void
 856 igc_hw_control_release(struct igc_hw *hw)
 857 {
 858         uint32_t ctrl_ext;
 859
 860         /* Let firmware taken over control of h/w */
 861         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
 862         IGC_WRITE_REG(hw, IGC_CTRL_EXT,
 863                         ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
 864 }
 865
 866 static int
 867 igc_hardware_init(struct igc_hw *hw)
 868 {
 869         uint32_t rx_buf_size;
 870         int diag;
 871
 872         /* Let the firmware know the OS is in control */
 873         igc_hw_control_acquire(hw);
 874
 875         /* Issue a global reset */
 876         igc_reset_hw(hw);
 877
 878         /* disable all wake up */
 879         IGC_WRITE_REG(hw, IGC_WUC, 0);
 880
 881         /*
 882          * Hardware flow control
 883          * - High water mark should allow for at least two standard size (1518)
 884          *   frames to be received after sending an XOFF.
 885          * - Low water mark works best when it is very near the high water mark.
 886          *   This allows the receiver to restart by sending XON when it has
 887          *   drained a bit. Here we use an arbitrary value of 1500 which will
 888          *   restart after one full frame is pulled from the buffer. There
 889          *   could be several smaller frames in the buffer and if so they will
 890          *   not trigger the XON until their total number reduces the buffer
 891          *   by 1500.
 892          */
 893         rx_buf_size = igc_get_rx_buffer_size(hw);
 894         hw->fc.high_water = rx_buf_size - (RTE_ETHER_MAX_LEN * 2);
 895         hw->fc.low_water = hw->fc.high_water - 1500;
 896         hw->fc.pause_time = IGC_FC_PAUSE_TIME;
 897         hw->fc.send_xon = 1;
 898         hw->fc.requested_mode = igc_fc_full;
 899
 900         diag = igc_init_hw(hw);
 901         if (diag < 0)
 902                 return diag;
 903
 904         igc_get_phy_info(hw);
 905         igc_check_for_link(hw);
 906
 907         return 0;
 908 }
 909
 910 static int
 911 eth_igc_start(struct rte_eth_dev *dev)
 912 {
 913         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
 914         struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
 915         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 916         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 917         uint32_t *speeds;
 918         int ret;
 919
 920         PMD_INIT_FUNC_TRACE();
 921
 922         /* disable all MSI-X interrupts */
 923         IGC_WRITE_REG(hw, IGC_EIMC, 0x1f);
 924         IGC_WRITE_FLUSH(hw);
 925
 926         /* clear all MSI-X interrupts */
 927         IGC_WRITE_REG(hw, IGC_EICR, 0x1f);
 928
 929         /* disable uio/vfio intr/eventfd mapping */
 930         if (!adapter->stopped)
 931                 rte_intr_disable(intr_handle);
 932
 933         /* Power up the phy. Needed to make the link go Up */
 934         eth_igc_set_link_up(dev);
 935
 936         /* Put the address into the Receive Address Array */
 937         igc_rar_set(hw, hw->mac.addr, 0);
 938
 939         /* Initialize the hardware */
 940         if (igc_hardware_init(hw)) {
 941                 PMD_DRV_LOG(ERR, "Unable to initialize the hardware");
 942                 return -EIO;
 943         }
 944         adapter->stopped = 0;
 945
 946         /* check and configure queue intr-vector mapping */
 947         if (rte_intr_cap_multiple(intr_handle) &&
 948                 dev->data->dev_conf.intr_conf.rxq) {
 949                 uint32_t intr_vector = dev->data->nb_rx_queues;
 950                 if (rte_intr_efd_enable(intr_handle, intr_vector))
 951                         return -1;
 952         }
 953
 954         if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
 955                 intr_handle->intr_vec = rte_zmalloc("intr_vec",
 956                         dev->data->nb_rx_queues * sizeof(int), 0);
 957                 if (intr_handle->intr_vec == NULL) {
 958                         PMD_DRV_LOG(ERR,
 959                                 "Failed to allocate %d rx_queues intr_vec",
 960                                 dev->data->nb_rx_queues);
 961                         return -ENOMEM;
 962                 }
 963         }
 964
 965         /* configure msix for rx interrupt */
 966         igc_configure_msix_intr(dev);
 967
 968         igc_tx_init(dev);
 969
 970         /* This can fail when allocating mbufs for descriptor rings */
 971         ret = igc_rx_init(dev);
 972         if (ret) {
 973                 PMD_DRV_LOG(ERR, "Unable to initialize RX hardware");
 974                 igc_dev_clear_queues(dev);
 975                 return ret;
 976         }
 977
 978         igc_clear_hw_cntrs_base_generic(hw);
 979
 980         /* VLAN Offload Settings */
 981         eth_igc_vlan_offload_set(dev,
 982                 ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK |
 983                 ETH_VLAN_EXTEND_MASK);
 984
 985         /* Setup link speed and duplex */
 986         speeds = &dev->data->dev_conf.link_speeds;
 987         if (*speeds == ETH_LINK_SPEED_AUTONEG) {
 988                 hw->phy.autoneg_advertised = IGC_ALL_SPEED_DUPLEX_2500;
 989                 hw->mac.autoneg = 1;
 990         } else {
 991                 int num_speeds = 0;
 992
 993                 if (*speeds & ETH_LINK_SPEED_FIXED) {
 994                         PMD_DRV_LOG(ERR,
 995                                     "Force speed mode currently not supported");
 996                         igc_dev_clear_queues(dev);
 997                         return -EINVAL;
 998                 }
 999
1000                 hw->phy.autoneg_advertised = 0;
1001                 hw->mac.autoneg = 1;
1002
1003                 if (*speeds & ~(ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
1004                                 ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
1005                                 ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G)) {
1006                         num_speeds = -1;
1007                         goto error_invalid_config;
1008                 }
1009                 if (*speeds & ETH_LINK_SPEED_10M_HD) {
1010                         hw->phy.autoneg_advertised |= ADVERTISE_10_HALF;
1011                         num_speeds++;
1012                 }
1013                 if (*speeds & ETH_LINK_SPEED_10M) {
1014                         hw->phy.autoneg_advertised |= ADVERTISE_10_FULL;
1015                         num_speeds++;
1016                 }
1017                 if (*speeds & ETH_LINK_SPEED_100M_HD) {
1018                         hw->phy.autoneg_advertised |= ADVERTISE_100_HALF;
1019                         num_speeds++;
1020                 }
1021                 if (*speeds & ETH_LINK_SPEED_100M) {
1022                         hw->phy.autoneg_advertised |= ADVERTISE_100_FULL;
1023                         num_speeds++;
1024                 }
1025                 if (*speeds & ETH_LINK_SPEED_1G) {
1026                         hw->phy.autoneg_advertised |= ADVERTISE_1000_FULL;
1027                         num_speeds++;
1028                 }
1029                 if (*speeds & ETH_LINK_SPEED_2_5G) {
1030                         hw->phy.autoneg_advertised |= ADVERTISE_2500_FULL;
1031                         num_speeds++;
1032                 }
1033                 if (num_speeds == 0)
1034                         goto error_invalid_config;
1035         }
1036
1037         igc_setup_link(hw);
1038
1039         if (rte_intr_allow_others(intr_handle)) {
1040                 /* check if lsc interrupt is enabled */
1041                 if (dev->data->dev_conf.intr_conf.lsc)
1042                         igc_lsc_interrupt_setup(dev, 1);
1043                 else
1044                         igc_lsc_interrupt_setup(dev, 0);
1045         } else {
1046                 rte_intr_callback_unregister(intr_handle,
1047                                              eth_igc_interrupt_handler,
1048                                              (void *)dev);
1049                 if (dev->data->dev_conf.intr_conf.lsc)
1050                         PMD_DRV_LOG(INFO,
1051                                 "LSC won't enable because of no intr multiplex");
1052         }
1053
1054         /* enable uio/vfio intr/eventfd mapping */
1055         rte_intr_enable(intr_handle);
1056
1057         rte_eal_alarm_set(IGC_ALARM_INTERVAL,
1058                         igc_update_queue_stats_handler, dev);
1059
1060         /* check if rxq interrupt is enabled */
1061         if (dev->data->dev_conf.intr_conf.rxq &&
1062                         rte_intr_dp_is_en(intr_handle))
1063                 igc_rxq_interrupt_setup(dev);
1064
1065         /* resume enabled intr since hw reset */
1066         igc_intr_other_enable(dev);
1067
1068         eth_igc_rxtx_control(dev, true);
1069         eth_igc_link_update(dev, 0);
1070
1071         /* configure MAC-loopback mode */
1072         if (dev->data->dev_conf.lpbk_mode == 1) {
1073                 uint32_t reg_val;
1074
1075                 reg_val = IGC_READ_REG(hw, IGC_CTRL);
1076                 reg_val &= ~IGC_CTRL_SPEED_MASK;
1077                 reg_val |= IGC_CTRL_SLU | IGC_CTRL_FRCSPD |
1078                         IGC_CTRL_FRCDPX | IGC_CTRL_FD | IGC_CTRL_SPEED_2500;
1079                 IGC_WRITE_REG(hw, IGC_CTRL, reg_val);
1080
1081                 igc_read_reg_check_set_bits(hw, IGC_EEER, IGC_EEER_EEE_FRC_AN);
1082         }
1083
1084         return 0;
1085
1086 error_invalid_config:
1087         PMD_DRV_LOG(ERR, "Invalid advertised speeds (%u) for port %u",
1088                      dev->data->dev_conf.link_speeds, dev->data->port_id);
1089         igc_dev_clear_queues(dev);
1090         return -EINVAL;
1091 }
1092
1093 static int
1094 igc_reset_swfw_lock(struct igc_hw *hw)
1095 {
1096         int ret_val;
1097
1098         /*
1099          * Do mac ops initialization manually here, since we will need
1100          * some function pointers set by this call.
1101          */
1102         ret_val = igc_init_mac_params(hw);
1103         if (ret_val)
1104                 return ret_val;
1105
1106         /*
1107          * SMBI lock should not fail in this early stage. If this is the case,
1108          * it is due to an improper exit of the application.
1109          * So force the release of the faulty lock.
1110          */
1111         if (igc_get_hw_semaphore_generic(hw) < 0)
1112                 PMD_DRV_LOG(DEBUG, "SMBI lock released");
1113
1114         igc_put_hw_semaphore_generic(hw);
1115
1116         if (hw->mac.ops.acquire_swfw_sync != NULL) {
1117                 uint16_t mask;
1118
1119                 /*
1120                  * Phy lock should not fail in this early stage.
1121                  * If this is the case, it is due to an improper exit of the
1122                  * application. So force the release of the faulty lock.
1123                  */
1124                 mask = IGC_SWFW_PHY0_SM;
1125                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0) {
1126                         PMD_DRV_LOG(DEBUG, "SWFW phy%d lock released",
1127                                     hw->bus.func);
1128                 }
1129                 hw->mac.ops.release_swfw_sync(hw, mask);
1130
1131                 /*
1132                  * This one is more tricky since it is common to all ports; but
1133                  * swfw_sync retries last long enough (1s) to be almost sure
1134                  * that if lock can not be taken it is due to an improper lock
1135                  * of the semaphore.
1136                  */
1137                 mask = IGC_SWFW_EEP_SM;
1138                 if (hw->mac.ops.acquire_swfw_sync(hw, mask) < 0)
1139                         PMD_DRV_LOG(DEBUG, "SWFW common locks released");
1140
1141                 hw->mac.ops.release_swfw_sync(hw, mask);
1142         }
1143
1144         return IGC_SUCCESS;
1145 }
1146
1147 /*
1148  * free all rx/tx queues.
1149  */
1150 static void
1151 igc_dev_free_queues(struct rte_eth_dev *dev)
1152 {
1153         uint16_t i;
1154
1155         for (i = 0; i < dev->data->nb_rx_queues; i++) {
1156                 eth_igc_rx_queue_release(dev->data->rx_queues[i]);
1157                 dev->data->rx_queues[i] = NULL;
1158         }
1159         dev->data->nb_rx_queues = 0;
1160
1161         for (i = 0; i < dev->data->nb_tx_queues; i++) {
1162                 eth_igc_tx_queue_release(dev->data->tx_queues[i]);
1163                 dev->data->tx_queues[i] = NULL;
1164         }
1165         dev->data->nb_tx_queues = 0;
1166 }
1167
1168 static int
1169 eth_igc_close(struct rte_eth_dev *dev)
1170 {
1171         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1172         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
1173         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1174         struct igc_adapter *adapter = IGC_DEV_PRIVATE(dev);
1175         int retry = 0;
1176         int ret = 0;
1177
1178         PMD_INIT_FUNC_TRACE();
1179         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1180                 return 0;
1181
1182         if (!adapter->stopped)
1183                 ret = eth_igc_stop(dev);
1184
1185         igc_flow_flush(dev, NULL);
1186         igc_clear_all_filter(dev);
1187
1188         igc_intr_other_disable(dev);
1189         do {
1190                 int ret = rte_intr_callback_unregister(intr_handle,
1191                                 eth_igc_interrupt_handler, dev);
1192                 if (ret >= 0 || ret == -ENOENT || ret == -EINVAL)
1193                         break;
1194
1195                 PMD_DRV_LOG(ERR, "intr callback unregister failed: %d", ret);
1196                 DELAY(200 * 1000); /* delay 200ms */
1197         } while (retry++ < 5);
1198
1199         igc_phy_hw_reset(hw);
1200         igc_hw_control_release(hw);
1201         igc_dev_free_queues(dev);
1202
1203         /* Reset any pending lock */
1204         igc_reset_swfw_lock(hw);
1205
1206         return ret;
1207 }
1208
1209 static void
1210 igc_identify_hardware(struct rte_eth_dev *dev, struct rte_pci_device *pci_dev)
1211 {
1212         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1213
1214         hw->vendor_id = pci_dev->id.vendor_id;
1215         hw->device_id = pci_dev->id.device_id;
1216         hw->subsystem_vendor_id = pci_dev->id.subsystem_vendor_id;
1217         hw->subsystem_device_id = pci_dev->id.subsystem_device_id;
1218 }
1219
1220 static int
1221 eth_igc_dev_init(struct rte_eth_dev *dev)
1222 {
1223         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1224         struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
1225         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1226         int i, error = 0;
1227
1228         PMD_INIT_FUNC_TRACE();
1229         dev->dev_ops = &eth_igc_ops;
1230         dev->rx_descriptor_done = eth_igc_rx_descriptor_done;
1231         dev->rx_queue_count = eth_igc_rx_queue_count;
1232         dev->rx_descriptor_status = eth_igc_rx_descriptor_status;
1233         dev->tx_descriptor_status = eth_igc_tx_descriptor_status;
1234
1235         /*
1236          * for secondary processes, we don't initialize any further as primary
1237          * has already done this work. Only check we don't need a different
1238          * RX function.
1239          */
1240         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1241                 return 0;
1242
1243         rte_eth_copy_pci_info(dev, pci_dev);
1244         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1245
1246         hw->back = pci_dev;
1247         hw->hw_addr = (void *)pci_dev->mem_resource[0].addr;
1248
1249         igc_identify_hardware(dev, pci_dev);
1250         if (igc_setup_init_funcs(hw, false) != IGC_SUCCESS) {
1251                 error = -EIO;
1252                 goto err_late;
1253         }
1254
1255         igc_get_bus_info(hw);
1256
1257         /* Reset any pending lock */
1258         if (igc_reset_swfw_lock(hw) != IGC_SUCCESS) {
1259                 error = -EIO;
1260                 goto err_late;
1261         }
1262
1263         /* Finish initialization */
1264         if (igc_setup_init_funcs(hw, true) != IGC_SUCCESS) {
1265                 error = -EIO;
1266                 goto err_late;
1267         }
1268
1269         hw->mac.autoneg = 1;
1270         hw->phy.autoneg_wait_to_complete = 0;
1271         hw->phy.autoneg_advertised = IGC_ALL_SPEED_DUPLEX_2500;
1272
1273         /* Copper options */
1274         if (hw->phy.media_type == igc_media_type_copper) {
1275                 hw->phy.mdix = 0; /* AUTO_ALL_MODES */
1276                 hw->phy.disable_polarity_correction = 0;
1277                 hw->phy.ms_type = igc_ms_hw_default;
1278         }
1279
1280         /*
1281          * Start from a known state, this is important in reading the nvm
1282          * and mac from that.
1283          */
1284         igc_reset_hw(hw);
1285
1286         /* Make sure we have a good EEPROM before we read from it */
1287         if (igc_validate_nvm_checksum(hw) < 0) {
1288                 /*
1289                  * Some PCI-E parts fail the first check due to
1290                  * the link being in sleep state, call it again,
1291                  * if it fails a second time its a real issue.
1292                  */
1293                 if (igc_validate_nvm_checksum(hw) < 0) {
1294                         PMD_INIT_LOG(ERR, "EEPROM checksum invalid");
1295                         error = -EIO;
1296                         goto err_late;
1297                 }
1298         }
1299
1300         /* Read the permanent MAC address out of the EEPROM */
1301         if (igc_read_mac_addr(hw) != 0) {
1302                 PMD_INIT_LOG(ERR, "EEPROM error while reading MAC address");
1303                 error = -EIO;
1304                 goto err_late;
1305         }
1306
1307         /* Allocate memory for storing MAC addresses */
1308         dev->data->mac_addrs = rte_zmalloc("igc",
1309                 RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count, 0);
1310         if (dev->data->mac_addrs == NULL) {
1311                 PMD_INIT_LOG(ERR, "Failed to allocate %d bytes for storing MAC",
1312                                 RTE_ETHER_ADDR_LEN * hw->mac.rar_entry_count);
1313                 error = -ENOMEM;
1314                 goto err_late;
1315         }
1316
1317         /* Copy the permanent MAC address */
1318         rte_ether_addr_copy((struct rte_ether_addr *)hw->mac.addr,
1319                         &dev->data->mac_addrs[0]);
1320
1321         /* Now initialize the hardware */
1322         if (igc_hardware_init(hw) != 0) {
1323                 PMD_INIT_LOG(ERR, "Hardware initialization failed");
1324                 rte_free(dev->data->mac_addrs);
1325                 dev->data->mac_addrs = NULL;
1326                 error = -ENODEV;
1327                 goto err_late;
1328         }
1329
1330         hw->mac.get_link_status = 1;
1331         igc->stopped = 0;
1332
1333         /* Indicate SOL/IDER usage */
1334         if (igc_check_reset_block(hw) < 0)
1335                 PMD_INIT_LOG(ERR,
1336                         "PHY reset is blocked due to SOL/IDER session.");
1337
1338         PMD_INIT_LOG(DEBUG, "port_id %d vendorID=0x%x deviceID=0x%x",
1339                         dev->data->port_id, pci_dev->id.vendor_id,
1340                         pci_dev->id.device_id);
1341
1342         rte_intr_callback_register(&pci_dev->intr_handle,
1343                         eth_igc_interrupt_handler, (void *)dev);
1344
1345         /* enable uio/vfio intr/eventfd mapping */
1346         rte_intr_enable(&pci_dev->intr_handle);
1347
1348         /* enable support intr */
1349         igc_intr_other_enable(dev);
1350
1351         /* initiate queue status */
1352         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1353                 igc->txq_stats_map[i] = -1;
1354                 igc->rxq_stats_map[i] = -1;
1355         }
1356
1357         igc_flow_init(dev);
1358         igc_clear_all_filter(dev);
1359         return 0;
1360
1361 err_late:
1362         igc_hw_control_release(hw);
1363         return error;
1364 }
1365
1366 static int
1367 eth_igc_dev_uninit(__rte_unused struct rte_eth_dev *eth_dev)
1368 {
1369         PMD_INIT_FUNC_TRACE();
1370         eth_igc_close(eth_dev);
1371         return 0;
1372 }
1373
1374 static int
1375 eth_igc_reset(struct rte_eth_dev *dev)
1376 {
1377         int ret;
1378
1379         PMD_INIT_FUNC_TRACE();
1380
1381         ret = eth_igc_dev_uninit(dev);
1382         if (ret)
1383                 return ret;
1384
1385         return eth_igc_dev_init(dev);
1386 }
1387
1388 static int
1389 eth_igc_promiscuous_enable(struct rte_eth_dev *dev)
1390 {
1391         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1392         uint32_t rctl;
1393
1394         rctl = IGC_READ_REG(hw, IGC_RCTL);
1395         rctl |= (IGC_RCTL_UPE | IGC_RCTL_MPE);
1396         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1397         return 0;
1398 }
1399
1400 static int
1401 eth_igc_promiscuous_disable(struct rte_eth_dev *dev)
1402 {
1403         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1404         uint32_t rctl;
1405
1406         rctl = IGC_READ_REG(hw, IGC_RCTL);
1407         rctl &= (~IGC_RCTL_UPE);
1408         if (dev->data->all_multicast == 1)
1409                 rctl |= IGC_RCTL_MPE;
1410         else
1411                 rctl &= (~IGC_RCTL_MPE);
1412         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1413         return 0;
1414 }
1415
1416 static int
1417 eth_igc_allmulticast_enable(struct rte_eth_dev *dev)
1418 {
1419         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1420         uint32_t rctl;
1421
1422         rctl = IGC_READ_REG(hw, IGC_RCTL);
1423         rctl |= IGC_RCTL_MPE;
1424         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1425         return 0;
1426 }
1427
1428 static int
1429 eth_igc_allmulticast_disable(struct rte_eth_dev *dev)
1430 {
1431         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1432         uint32_t rctl;
1433
1434         if (dev->data->promiscuous == 1)
1435                 return 0;       /* must remain in all_multicast mode */
1436
1437         rctl = IGC_READ_REG(hw, IGC_RCTL);
1438         rctl &= (~IGC_RCTL_MPE);
1439         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1440         return 0;
1441 }
1442
1443 static int
1444 eth_igc_fw_version_get(struct rte_eth_dev *dev, char *fw_version,
1445                        size_t fw_size)
1446 {
1447         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1448         struct igc_fw_version fw;
1449         int ret;
1450
1451         igc_get_fw_version(hw, &fw);
1452
1453         /* if option rom is valid, display its version too */
1454         if (fw.or_valid) {
1455                 ret = snprintf(fw_version, fw_size,
1456                          "%d.%d, 0x%08x, %d.%d.%d",
1457                          fw.eep_major, fw.eep_minor, fw.etrack_id,
1458                          fw.or_major, fw.or_build, fw.or_patch);
1459         /* no option rom */
1460         } else {
1461                 if (fw.etrack_id != 0X0000) {
1462                         ret = snprintf(fw_version, fw_size,
1463                                  "%d.%d, 0x%08x",
1464                                  fw.eep_major, fw.eep_minor,
1465                                  fw.etrack_id);
1466                 } else {
1467                         ret = snprintf(fw_version, fw_size,
1468                                  "%d.%d.%d",
1469                                  fw.eep_major, fw.eep_minor,
1470                                  fw.eep_build);
1471                 }
1472         }
1473         if (ret < 0)
1474                 return -EINVAL;
1475
1476         ret += 1; /* add the size of '\0' */
1477         if (fw_size < (size_t)ret)
1478                 return ret;
1479         else
1480                 return 0;
1481 }
1482
1483 static int
1484 eth_igc_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
1485 {
1486         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1487
1488         dev_info->min_rx_bufsize = 256; /* See BSIZE field of RCTL register. */
1489         dev_info->max_rx_pktlen = MAX_RX_JUMBO_FRAME_SIZE;
1490         dev_info->max_mac_addrs = hw->mac.rar_entry_count;
1491         dev_info->rx_offload_capa = IGC_RX_OFFLOAD_ALL;
1492         dev_info->tx_offload_capa = IGC_TX_OFFLOAD_ALL;
1493         dev_info->rx_queue_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
1494
1495         dev_info->max_rx_queues = IGC_QUEUE_PAIRS_NUM;
1496         dev_info->max_tx_queues = IGC_QUEUE_PAIRS_NUM;
1497         dev_info->max_vmdq_pools = 0;
1498
1499         dev_info->hash_key_size = IGC_HKEY_MAX_INDEX * sizeof(uint32_t);
1500         dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
1501         dev_info->flow_type_rss_offloads = IGC_RSS_OFFLOAD_ALL;
1502
1503         dev_info->default_rxconf = (struct rte_eth_rxconf) {
1504                 .rx_thresh = {
1505                         .pthresh = IGC_DEFAULT_RX_PTHRESH,
1506                         .hthresh = IGC_DEFAULT_RX_HTHRESH,
1507                         .wthresh = IGC_DEFAULT_RX_WTHRESH,
1508                 },
1509                 .rx_free_thresh = IGC_DEFAULT_RX_FREE_THRESH,
1510                 .rx_drop_en = 0,
1511                 .offloads = 0,
1512         };
1513
1514         dev_info->default_txconf = (struct rte_eth_txconf) {
1515                 .tx_thresh = {
1516                         .pthresh = IGC_DEFAULT_TX_PTHRESH,
1517                         .hthresh = IGC_DEFAULT_TX_HTHRESH,
1518                         .wthresh = IGC_DEFAULT_TX_WTHRESH,
1519                 },
1520                 .offloads = 0,
1521         };
1522
1523         dev_info->rx_desc_lim = rx_desc_lim;
1524         dev_info->tx_desc_lim = tx_desc_lim;
1525
1526         dev_info->speed_capa = ETH_LINK_SPEED_10M_HD | ETH_LINK_SPEED_10M |
1527                         ETH_LINK_SPEED_100M_HD | ETH_LINK_SPEED_100M |
1528                         ETH_LINK_SPEED_1G | ETH_LINK_SPEED_2_5G;
1529
1530         dev_info->max_mtu = dev_info->max_rx_pktlen - IGC_ETH_OVERHEAD;
1531         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
1532         return 0;
1533 }
1534
1535 static int
1536 eth_igc_led_on(struct rte_eth_dev *dev)
1537 {
1538         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1539
1540         return igc_led_on(hw) == IGC_SUCCESS ? 0 : -ENOTSUP;
1541 }
1542
1543 static int
1544 eth_igc_led_off(struct rte_eth_dev *dev)
1545 {
1546         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1547
1548         return igc_led_off(hw) == IGC_SUCCESS ? 0 : -ENOTSUP;
1549 }
1550
1551 static const uint32_t *
1552 eth_igc_supported_ptypes_get(__rte_unused struct rte_eth_dev *dev)
1553 {
1554         static const uint32_t ptypes[] = {
1555                 /* refers to rx_desc_pkt_info_to_pkt_type() */
1556                 RTE_PTYPE_L2_ETHER,
1557                 RTE_PTYPE_L3_IPV4,
1558                 RTE_PTYPE_L3_IPV4_EXT,
1559                 RTE_PTYPE_L3_IPV6,
1560                 RTE_PTYPE_L3_IPV6_EXT,
1561                 RTE_PTYPE_L4_TCP,
1562                 RTE_PTYPE_L4_UDP,
1563                 RTE_PTYPE_L4_SCTP,
1564                 RTE_PTYPE_TUNNEL_IP,
1565                 RTE_PTYPE_INNER_L3_IPV6,
1566                 RTE_PTYPE_INNER_L3_IPV6_EXT,
1567                 RTE_PTYPE_INNER_L4_TCP,
1568                 RTE_PTYPE_INNER_L4_UDP,
1569                 RTE_PTYPE_UNKNOWN
1570         };
1571
1572         return ptypes;
1573 }
1574
1575 static int
1576 eth_igc_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1577 {
1578         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1579         uint32_t frame_size = mtu + IGC_ETH_OVERHEAD;
1580         uint32_t rctl;
1581
1582         /* if extend vlan has been enabled */
1583         if (IGC_READ_REG(hw, IGC_CTRL_EXT) & IGC_CTRL_EXT_EXT_VLAN)
1584                 frame_size += VLAN_TAG_SIZE;
1585
1586         /* check that mtu is within the allowed range */
1587         if (mtu < RTE_ETHER_MIN_MTU ||
1588                 frame_size > MAX_RX_JUMBO_FRAME_SIZE)
1589                 return -EINVAL;
1590
1591         /*
1592          * If device is started, refuse mtu that requires the support of
1593          * scattered packets when this feature has not been enabled before.
1594          */
1595         if (dev->data->dev_started && !dev->data->scattered_rx &&
1596             frame_size > dev->data->min_rx_buf_size - RTE_PKTMBUF_HEADROOM) {
1597                 PMD_INIT_LOG(ERR, "Stop port first.");
1598                 return -EINVAL;
1599         }
1600
1601         rctl = IGC_READ_REG(hw, IGC_RCTL);
1602
1603         /* switch to jumbo mode if needed */
1604         if (mtu > RTE_ETHER_MTU) {
1605                 dev->data->dev_conf.rxmode.offloads |=
1606                         DEV_RX_OFFLOAD_JUMBO_FRAME;
1607                 rctl |= IGC_RCTL_LPE;
1608         } else {
1609                 dev->data->dev_conf.rxmode.offloads &=
1610                         ~DEV_RX_OFFLOAD_JUMBO_FRAME;
1611                 rctl &= ~IGC_RCTL_LPE;
1612         }
1613         IGC_WRITE_REG(hw, IGC_RCTL, rctl);
1614
1615         /* update max frame size */
1616         dev->data->dev_conf.rxmode.max_rx_pkt_len = frame_size;
1617
1618         IGC_WRITE_REG(hw, IGC_RLPML,
1619                         dev->data->dev_conf.rxmode.max_rx_pkt_len);
1620
1621         return 0;
1622 }
1623
1624 static int
1625 eth_igc_rar_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr,
1626                 uint32_t index, uint32_t pool)
1627 {
1628         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1629
1630         igc_rar_set(hw, mac_addr->addr_bytes, index);
1631         RTE_SET_USED(pool);
1632         return 0;
1633 }
1634
1635 static void
1636 eth_igc_rar_clear(struct rte_eth_dev *dev, uint32_t index)
1637 {
1638         uint8_t addr[RTE_ETHER_ADDR_LEN];
1639         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1640
1641         memset(addr, 0, sizeof(addr));
1642         igc_rar_set(hw, addr, index);
1643 }
1644
1645 static int
1646 eth_igc_default_mac_addr_set(struct rte_eth_dev *dev,
1647                         struct rte_ether_addr *addr)
1648 {
1649         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1650         igc_rar_set(hw, addr->addr_bytes, 0);
1651         return 0;
1652 }
1653
1654 static int
1655 eth_igc_set_mc_addr_list(struct rte_eth_dev *dev,
1656                          struct rte_ether_addr *mc_addr_set,
1657                          uint32_t nb_mc_addr)
1658 {
1659         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1660         igc_update_mc_addr_list(hw, (u8 *)mc_addr_set, nb_mc_addr);
1661         return 0;
1662 }
1663
1664 /*
1665  * Read hardware registers
1666  */
1667 static void
1668 igc_read_stats_registers(struct igc_hw *hw, struct igc_hw_stats *stats)
1669 {
1670         int pause_frames;
1671
1672         uint64_t old_gprc  = stats->gprc;
1673         uint64_t old_gptc  = stats->gptc;
1674         uint64_t old_tpr   = stats->tpr;
1675         uint64_t old_tpt   = stats->tpt;
1676         uint64_t old_rpthc = stats->rpthc;
1677         uint64_t old_hgptc = stats->hgptc;
1678
1679         stats->crcerrs += IGC_READ_REG(hw, IGC_CRCERRS);
1680         stats->algnerrc += IGC_READ_REG(hw, IGC_ALGNERRC);
1681         stats->rxerrc += IGC_READ_REG(hw, IGC_RXERRC);
1682         stats->mpc += IGC_READ_REG(hw, IGC_MPC);
1683         stats->scc += IGC_READ_REG(hw, IGC_SCC);
1684         stats->ecol += IGC_READ_REG(hw, IGC_ECOL);
1685
1686         stats->mcc += IGC_READ_REG(hw, IGC_MCC);
1687         stats->latecol += IGC_READ_REG(hw, IGC_LATECOL);
1688         stats->colc += IGC_READ_REG(hw, IGC_COLC);
1689
1690         stats->dc += IGC_READ_REG(hw, IGC_DC);
1691         stats->tncrs += IGC_READ_REG(hw, IGC_TNCRS);
1692         stats->htdpmc += IGC_READ_REG(hw, IGC_HTDPMC);
1693         stats->rlec += IGC_READ_REG(hw, IGC_RLEC);
1694         stats->xonrxc += IGC_READ_REG(hw, IGC_XONRXC);
1695         stats->xontxc += IGC_READ_REG(hw, IGC_XONTXC);
1696
1697         /*
1698          * For watchdog management we need to know if we have been
1699          * paused during the last interval, so capture that here.
1700          */
1701         pause_frames = IGC_READ_REG(hw, IGC_XOFFRXC);
1702         stats->xoffrxc += pause_frames;
1703         stats->xofftxc += IGC_READ_REG(hw, IGC_XOFFTXC);
1704         stats->fcruc += IGC_READ_REG(hw, IGC_FCRUC);
1705         stats->prc64 += IGC_READ_REG(hw, IGC_PRC64);
1706         stats->prc127 += IGC_READ_REG(hw, IGC_PRC127);
1707         stats->prc255 += IGC_READ_REG(hw, IGC_PRC255);
1708         stats->prc511 += IGC_READ_REG(hw, IGC_PRC511);
1709         stats->prc1023 += IGC_READ_REG(hw, IGC_PRC1023);
1710         stats->prc1522 += IGC_READ_REG(hw, IGC_PRC1522);
1711         stats->gprc += IGC_READ_REG(hw, IGC_GPRC);
1712         stats->bprc += IGC_READ_REG(hw, IGC_BPRC);
1713         stats->mprc += IGC_READ_REG(hw, IGC_MPRC);
1714         stats->gptc += IGC_READ_REG(hw, IGC_GPTC);
1715
1716         /* For the 64-bit byte counters the low dword must be read first. */
1717         /* Both registers clear on the read of the high dword */
1718
1719         /* Workaround CRC bytes included in size, take away 4 bytes/packet */
1720         stats->gorc += IGC_READ_REG(hw, IGC_GORCL);
1721         stats->gorc += ((uint64_t)IGC_READ_REG(hw, IGC_GORCH) << 32);
1722         stats->gorc -= (stats->gprc - old_gprc) * RTE_ETHER_CRC_LEN;
1723         stats->gotc += IGC_READ_REG(hw, IGC_GOTCL);
1724         stats->gotc += ((uint64_t)IGC_READ_REG(hw, IGC_GOTCH) << 32);
1725         stats->gotc -= (stats->gptc - old_gptc) * RTE_ETHER_CRC_LEN;
1726
1727         stats->rnbc += IGC_READ_REG(hw, IGC_RNBC);
1728         stats->ruc += IGC_READ_REG(hw, IGC_RUC);
1729         stats->rfc += IGC_READ_REG(hw, IGC_RFC);
1730         stats->roc += IGC_READ_REG(hw, IGC_ROC);
1731         stats->rjc += IGC_READ_REG(hw, IGC_RJC);
1732
1733         stats->mgprc += IGC_READ_REG(hw, IGC_MGTPRC);
1734         stats->mgpdc += IGC_READ_REG(hw, IGC_MGTPDC);
1735         stats->mgptc += IGC_READ_REG(hw, IGC_MGTPTC);
1736         stats->b2ospc += IGC_READ_REG(hw, IGC_B2OSPC);
1737         stats->b2ogprc += IGC_READ_REG(hw, IGC_B2OGPRC);
1738         stats->o2bgptc += IGC_READ_REG(hw, IGC_O2BGPTC);
1739         stats->o2bspc += IGC_READ_REG(hw, IGC_O2BSPC);
1740
1741         stats->tpr += IGC_READ_REG(hw, IGC_TPR);
1742         stats->tpt += IGC_READ_REG(hw, IGC_TPT);
1743
1744         stats->tor += IGC_READ_REG(hw, IGC_TORL);
1745         stats->tor += ((uint64_t)IGC_READ_REG(hw, IGC_TORH) << 32);
1746         stats->tor -= (stats->tpr - old_tpr) * RTE_ETHER_CRC_LEN;
1747         stats->tot += IGC_READ_REG(hw, IGC_TOTL);
1748         stats->tot += ((uint64_t)IGC_READ_REG(hw, IGC_TOTH) << 32);
1749         stats->tot -= (stats->tpt - old_tpt) * RTE_ETHER_CRC_LEN;
1750
1751         stats->ptc64 += IGC_READ_REG(hw, IGC_PTC64);
1752         stats->ptc127 += IGC_READ_REG(hw, IGC_PTC127);
1753         stats->ptc255 += IGC_READ_REG(hw, IGC_PTC255);
1754         stats->ptc511 += IGC_READ_REG(hw, IGC_PTC511);
1755         stats->ptc1023 += IGC_READ_REG(hw, IGC_PTC1023);
1756         stats->ptc1522 += IGC_READ_REG(hw, IGC_PTC1522);
1757         stats->mptc += IGC_READ_REG(hw, IGC_MPTC);
1758         stats->bptc += IGC_READ_REG(hw, IGC_BPTC);
1759         stats->tsctc += IGC_READ_REG(hw, IGC_TSCTC);
1760
1761         stats->iac += IGC_READ_REG(hw, IGC_IAC);
1762         stats->rpthc += IGC_READ_REG(hw, IGC_RPTHC);
1763         stats->hgptc += IGC_READ_REG(hw, IGC_HGPTC);
1764         stats->icrxdmtc += IGC_READ_REG(hw, IGC_ICRXDMTC);
1765
1766         /* Host to Card Statistics */
1767         stats->hgorc += IGC_READ_REG(hw, IGC_HGORCL);
1768         stats->hgorc += ((uint64_t)IGC_READ_REG(hw, IGC_HGORCH) << 32);
1769         stats->hgorc -= (stats->rpthc - old_rpthc) * RTE_ETHER_CRC_LEN;
1770         stats->hgotc += IGC_READ_REG(hw, IGC_HGOTCL);
1771         stats->hgotc += ((uint64_t)IGC_READ_REG(hw, IGC_HGOTCH) << 32);
1772         stats->hgotc -= (stats->hgptc - old_hgptc) * RTE_ETHER_CRC_LEN;
1773         stats->lenerrs += IGC_READ_REG(hw, IGC_LENERRS);
1774 }
1775
1776 /*
1777  * Write 0 to all queue status registers
1778  */
1779 static void
1780 igc_reset_queue_stats_register(struct igc_hw *hw)
1781 {
1782         int i;
1783
1784         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1785                 IGC_WRITE_REG(hw, IGC_PQGPRC(i), 0);
1786                 IGC_WRITE_REG(hw, IGC_PQGPTC(i), 0);
1787                 IGC_WRITE_REG(hw, IGC_PQGORC(i), 0);
1788                 IGC_WRITE_REG(hw, IGC_PQGOTC(i), 0);
1789                 IGC_WRITE_REG(hw, IGC_PQMPRC(i), 0);
1790                 IGC_WRITE_REG(hw, IGC_RQDPC(i), 0);
1791                 IGC_WRITE_REG(hw, IGC_TQDPC(i), 0);
1792         }
1793 }
1794
1795 /*
1796  * Read all hardware queue status registers
1797  */
1798 static void
1799 igc_read_queue_stats_register(struct rte_eth_dev *dev)
1800 {
1801         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1802         struct igc_hw_queue_stats *queue_stats =
1803                                 IGC_DEV_PRIVATE_QUEUE_STATS(dev);
1804         int i;
1805
1806         /*
1807          * This register is not cleared on read. Furthermore, the register wraps
1808          * around back to 0x00000000 on the next increment when reaching a value
1809          * of 0xFFFFFFFF and then continues normal count operation.
1810          */
1811         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1812                 union {
1813                         u64 ddword;
1814                         u32 dword[2];
1815                 } value;
1816                 u32 tmp;
1817
1818                 /*
1819                  * Read the register first, if the value is smaller than that
1820                  * previous read, that mean the register has been overflowed,
1821                  * then we add the high 4 bytes by 1 and replace the low 4
1822                  * bytes by the new value.
1823                  */
1824                 tmp = IGC_READ_REG(hw, IGC_PQGPRC(i));
1825                 value.ddword = queue_stats->pqgprc[i];
1826                 if (value.dword[U32_0_IN_U64] > tmp)
1827                         value.dword[U32_1_IN_U64]++;
1828                 value.dword[U32_0_IN_U64] = tmp;
1829                 queue_stats->pqgprc[i] = value.ddword;
1830
1831                 tmp = IGC_READ_REG(hw, IGC_PQGPTC(i));
1832                 value.ddword = queue_stats->pqgptc[i];
1833                 if (value.dword[U32_0_IN_U64] > tmp)
1834                         value.dword[U32_1_IN_U64]++;
1835                 value.dword[U32_0_IN_U64] = tmp;
1836                 queue_stats->pqgptc[i] = value.ddword;
1837
1838                 tmp = IGC_READ_REG(hw, IGC_PQGORC(i));
1839                 value.ddword = queue_stats->pqgorc[i];
1840                 if (value.dword[U32_0_IN_U64] > tmp)
1841                         value.dword[U32_1_IN_U64]++;
1842                 value.dword[U32_0_IN_U64] = tmp;
1843                 queue_stats->pqgorc[i] = value.ddword;
1844
1845                 tmp = IGC_READ_REG(hw, IGC_PQGOTC(i));
1846                 value.ddword = queue_stats->pqgotc[i];
1847                 if (value.dword[U32_0_IN_U64] > tmp)
1848                         value.dword[U32_1_IN_U64]++;
1849                 value.dword[U32_0_IN_U64] = tmp;
1850                 queue_stats->pqgotc[i] = value.ddword;
1851
1852                 tmp = IGC_READ_REG(hw, IGC_PQMPRC(i));
1853                 value.ddword = queue_stats->pqmprc[i];
1854                 if (value.dword[U32_0_IN_U64] > tmp)
1855                         value.dword[U32_1_IN_U64]++;
1856                 value.dword[U32_0_IN_U64] = tmp;
1857                 queue_stats->pqmprc[i] = value.ddword;
1858
1859                 tmp = IGC_READ_REG(hw, IGC_RQDPC(i));
1860                 value.ddword = queue_stats->rqdpc[i];
1861                 if (value.dword[U32_0_IN_U64] > tmp)
1862                         value.dword[U32_1_IN_U64]++;
1863                 value.dword[U32_0_IN_U64] = tmp;
1864                 queue_stats->rqdpc[i] = value.ddword;
1865
1866                 tmp = IGC_READ_REG(hw, IGC_TQDPC(i));
1867                 value.ddword = queue_stats->tqdpc[i];
1868                 if (value.dword[U32_0_IN_U64] > tmp)
1869                         value.dword[U32_1_IN_U64]++;
1870                 value.dword[U32_0_IN_U64] = tmp;
1871                 queue_stats->tqdpc[i] = value.ddword;
1872         }
1873 }
1874
1875 static int
1876 eth_igc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *rte_stats)
1877 {
1878         struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
1879         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1880         struct igc_hw_stats *stats = IGC_DEV_PRIVATE_STATS(dev);
1881         struct igc_hw_queue_stats *queue_stats =
1882                         IGC_DEV_PRIVATE_QUEUE_STATS(dev);
1883         int i;
1884
1885         /*
1886          * Cancel status handler since it will read the queue status registers
1887          */
1888         rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
1889
1890         /* Read status register */
1891         igc_read_queue_stats_register(dev);
1892         igc_read_stats_registers(hw, stats);
1893
1894         if (rte_stats == NULL) {
1895                 /* Restart queue status handler */
1896                 rte_eal_alarm_set(IGC_ALARM_INTERVAL,
1897                                 igc_update_queue_stats_handler, dev);
1898                 return -EINVAL;
1899         }
1900
1901         /* Rx Errors */
1902         rte_stats->imissed = stats->mpc;
1903         rte_stats->ierrors = stats->crcerrs + stats->rlec +
1904                         stats->rxerrc + stats->algnerrc;
1905
1906         /* Tx Errors */
1907         rte_stats->oerrors = stats->ecol + stats->latecol;
1908
1909         rte_stats->ipackets = stats->gprc;
1910         rte_stats->opackets = stats->gptc;
1911         rte_stats->ibytes   = stats->gorc;
1912         rte_stats->obytes   = stats->gotc;
1913
1914         /* Get per-queue statuses */
1915         for (i = 0; i < IGC_QUEUE_PAIRS_NUM; i++) {
1916                 /* GET TX queue statuses */
1917                 int map_id = igc->txq_stats_map[i];
1918                 if (map_id >= 0) {
1919                         rte_stats->q_opackets[map_id] += queue_stats->pqgptc[i];
1920                         rte_stats->q_obytes[map_id] += queue_stats->pqgotc[i];
1921                 }
1922                 /* Get RX queue statuses */
1923                 map_id = igc->rxq_stats_map[i];
1924                 if (map_id >= 0) {
1925                         rte_stats->q_ipackets[map_id] += queue_stats->pqgprc[i];
1926                         rte_stats->q_ibytes[map_id] += queue_stats->pqgorc[i];
1927                         rte_stats->q_errors[map_id] += queue_stats->rqdpc[i];
1928                 }
1929         }
1930
1931         /* Restart queue status handler */
1932         rte_eal_alarm_set(IGC_ALARM_INTERVAL,
1933                         igc_update_queue_stats_handler, dev);
1934         return 0;
1935 }
1936
1937 static int
1938 eth_igc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1939                    unsigned int n)
1940 {
1941         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1942         struct igc_hw_stats *hw_stats =
1943                         IGC_DEV_PRIVATE_STATS(dev);
1944         unsigned int i;
1945
1946         igc_read_stats_registers(hw, hw_stats);
1947
1948         if (n < IGC_NB_XSTATS)
1949                 return IGC_NB_XSTATS;
1950
1951         /* If this is a reset xstats is NULL, and we have cleared the
1952          * registers by reading them.
1953          */
1954         if (!xstats)
1955                 return 0;
1956
1957         /* Extended stats */
1958         for (i = 0; i < IGC_NB_XSTATS; i++) {
1959                 xstats[i].id = i;
1960                 xstats[i].value = *(uint64_t *)(((char *)hw_stats) +
1961                         rte_igc_stats_strings[i].offset);
1962         }
1963
1964         return IGC_NB_XSTATS;
1965 }
1966
1967 static int
1968 eth_igc_xstats_reset(struct rte_eth_dev *dev)
1969 {
1970         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
1971         struct igc_hw_stats *hw_stats = IGC_DEV_PRIVATE_STATS(dev);
1972         struct igc_hw_queue_stats *queue_stats =
1973                         IGC_DEV_PRIVATE_QUEUE_STATS(dev);
1974
1975         /* Cancel queue status handler for avoid conflict */
1976         rte_eal_alarm_cancel(igc_update_queue_stats_handler, dev);
1977
1978         /* HW registers are cleared on read */
1979         igc_reset_queue_stats_register(hw);
1980         igc_read_stats_registers(hw, hw_stats);
1981
1982         /* Reset software totals */
1983         memset(hw_stats, 0, sizeof(*hw_stats));
1984         memset(queue_stats, 0, sizeof(*queue_stats));
1985
1986         /* Restart the queue status handler */
1987         rte_eal_alarm_set(IGC_ALARM_INTERVAL, igc_update_queue_stats_handler,
1988                         dev);
1989
1990         return 0;
1991 }
1992
1993 static int
1994 eth_igc_xstats_get_names(__rte_unused struct rte_eth_dev *dev,
1995         struct rte_eth_xstat_name *xstats_names, unsigned int size)
1996 {
1997         unsigned int i;
1998
1999         if (xstats_names == NULL)
2000                 return IGC_NB_XSTATS;
2001
2002         if (size < IGC_NB_XSTATS) {
2003                 PMD_DRV_LOG(ERR, "not enough buffers!");
2004                 return IGC_NB_XSTATS;
2005         }
2006
2007         for (i = 0; i < IGC_NB_XSTATS; i++)
2008                 strlcpy(xstats_names[i].name, rte_igc_stats_strings[i].name,
2009                         sizeof(xstats_names[i].name));
2010
2011         return IGC_NB_XSTATS;
2012 }
2013
2014 static int
2015 eth_igc_xstats_get_names_by_id(struct rte_eth_dev *dev,
2016                 struct rte_eth_xstat_name *xstats_names, const uint64_t *ids,
2017                 unsigned int limit)
2018 {
2019         unsigned int i;
2020
2021         if (!ids)
2022                 return eth_igc_xstats_get_names(dev, xstats_names, limit);
2023
2024         for (i = 0; i < limit; i++) {
2025                 if (ids[i] >= IGC_NB_XSTATS) {
2026                         PMD_DRV_LOG(ERR, "id value isn't valid");
2027                         return -EINVAL;
2028                 }
2029                 strlcpy(xstats_names[i].name,
2030                         rte_igc_stats_strings[ids[i]].name,
2031                         sizeof(xstats_names[i].name));
2032         }
2033         return limit;
2034 }
2035
2036 static int
2037 eth_igc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
2038                 uint64_t *values, unsigned int n)
2039 {
2040         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2041         struct igc_hw_stats *hw_stats = IGC_DEV_PRIVATE_STATS(dev);
2042         unsigned int i;
2043
2044         igc_read_stats_registers(hw, hw_stats);
2045
2046         if (!ids) {
2047                 if (n < IGC_NB_XSTATS)
2048                         return IGC_NB_XSTATS;
2049
2050                 /* If this is a reset xstats is NULL, and we have cleared the
2051                  * registers by reading them.
2052                  */
2053                 if (!values)
2054                         return 0;
2055
2056                 /* Extended stats */
2057                 for (i = 0; i < IGC_NB_XSTATS; i++)
2058                         values[i] = *(uint64_t *)(((char *)hw_stats) +
2059                                         rte_igc_stats_strings[i].offset);
2060
2061                 return IGC_NB_XSTATS;
2062
2063         } else {
2064                 for (i = 0; i < n; i++) {
2065                         if (ids[i] >= IGC_NB_XSTATS) {
2066                                 PMD_DRV_LOG(ERR, "id value isn't valid");
2067                                 return -EINVAL;
2068                         }
2069                         values[i] = *(uint64_t *)(((char *)hw_stats) +
2070                                         rte_igc_stats_strings[ids[i]].offset);
2071                 }
2072                 return n;
2073         }
2074 }
2075
2076 static int
2077 eth_igc_queue_stats_mapping_set(struct rte_eth_dev *dev,
2078                 uint16_t queue_id, uint8_t stat_idx, uint8_t is_rx)
2079 {
2080         struct igc_adapter *igc = IGC_DEV_PRIVATE(dev);
2081
2082         /* check queue id is valid */
2083         if (queue_id >= IGC_QUEUE_PAIRS_NUM) {
2084                 PMD_DRV_LOG(ERR, "queue id(%u) error, max is %u",
2085                         queue_id, IGC_QUEUE_PAIRS_NUM - 1);
2086                 return -EINVAL;
2087         }
2088
2089         /* store the mapping status id */
2090         if (is_rx)
2091                 igc->rxq_stats_map[queue_id] = stat_idx;
2092         else
2093                 igc->txq_stats_map[queue_id] = stat_idx;
2094
2095         return 0;
2096 }
2097
2098 static int
2099 eth_igc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
2100 {
2101         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2102         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2103         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2104         uint32_t vec = IGC_MISC_VEC_ID;
2105
2106         if (rte_intr_allow_others(intr_handle))
2107                 vec = IGC_RX_VEC_START;
2108
2109         uint32_t mask = 1u << (queue_id + vec);
2110
2111         IGC_WRITE_REG(hw, IGC_EIMC, mask);
2112         IGC_WRITE_FLUSH(hw);
2113
2114         return 0;
2115 }
2116
2117 static int
2118 eth_igc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
2119 {
2120         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2121         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2122         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
2123         uint32_t vec = IGC_MISC_VEC_ID;
2124
2125         if (rte_intr_allow_others(intr_handle))
2126                 vec = IGC_RX_VEC_START;
2127
2128         uint32_t mask = 1u << (queue_id + vec);
2129
2130         IGC_WRITE_REG(hw, IGC_EIMS, mask);
2131         IGC_WRITE_FLUSH(hw);
2132
2133         rte_intr_enable(intr_handle);
2134
2135         return 0;
2136 }
2137
2138 static int
2139 eth_igc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2140 {
2141         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2142         uint32_t ctrl;
2143         int tx_pause;
2144         int rx_pause;
2145
2146         fc_conf->pause_time = hw->fc.pause_time;
2147         fc_conf->high_water = hw->fc.high_water;
2148         fc_conf->low_water = hw->fc.low_water;
2149         fc_conf->send_xon = hw->fc.send_xon;
2150         fc_conf->autoneg = hw->mac.autoneg;
2151
2152         /*
2153          * Return rx_pause and tx_pause status according to actual setting of
2154          * the TFCE and RFCE bits in the CTRL register.
2155          */
2156         ctrl = IGC_READ_REG(hw, IGC_CTRL);
2157         if (ctrl & IGC_CTRL_TFCE)
2158                 tx_pause = 1;
2159         else
2160                 tx_pause = 0;
2161
2162         if (ctrl & IGC_CTRL_RFCE)
2163                 rx_pause = 1;
2164         else
2165                 rx_pause = 0;
2166
2167         if (rx_pause && tx_pause)
2168                 fc_conf->mode = RTE_FC_FULL;
2169         else if (rx_pause)
2170                 fc_conf->mode = RTE_FC_RX_PAUSE;
2171         else if (tx_pause)
2172                 fc_conf->mode = RTE_FC_TX_PAUSE;
2173         else
2174                 fc_conf->mode = RTE_FC_NONE;
2175
2176         return 0;
2177 }
2178
2179 static int
2180 eth_igc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
2181 {
2182         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2183         uint32_t rx_buf_size;
2184         uint32_t max_high_water;
2185         uint32_t rctl;
2186         int err;
2187
2188         if (fc_conf->autoneg != hw->mac.autoneg)
2189                 return -ENOTSUP;
2190
2191         rx_buf_size = igc_get_rx_buffer_size(hw);
2192         PMD_DRV_LOG(DEBUG, "Rx packet buffer size = 0x%x", rx_buf_size);
2193
2194         /* At least reserve one Ethernet frame for watermark */
2195         max_high_water = rx_buf_size - RTE_ETHER_MAX_LEN;
2196         if (fc_conf->high_water > max_high_water ||
2197                 fc_conf->high_water < fc_conf->low_water) {
2198                 PMD_DRV_LOG(ERR,
2199                         "Incorrect high(%u)/low(%u) water value, max is %u",
2200                         fc_conf->high_water, fc_conf->low_water,
2201                         max_high_water);
2202                 return -EINVAL;
2203         }
2204
2205         switch (fc_conf->mode) {
2206         case RTE_FC_NONE:
2207                 hw->fc.requested_mode = igc_fc_none;
2208                 break;
2209         case RTE_FC_RX_PAUSE:
2210                 hw->fc.requested_mode = igc_fc_rx_pause;
2211                 break;
2212         case RTE_FC_TX_PAUSE:
2213                 hw->fc.requested_mode = igc_fc_tx_pause;
2214                 break;
2215         case RTE_FC_FULL:
2216                 hw->fc.requested_mode = igc_fc_full;
2217                 break;
2218         default:
2219                 PMD_DRV_LOG(ERR, "unsupported fc mode: %u", fc_conf->mode);
2220                 return -EINVAL;
2221         }
2222
2223         hw->fc.pause_time     = fc_conf->pause_time;
2224         hw->fc.high_water     = fc_conf->high_water;
2225         hw->fc.low_water      = fc_conf->low_water;
2226         hw->fc.send_xon       = fc_conf->send_xon;
2227
2228         err = igc_setup_link_generic(hw);
2229         if (err == IGC_SUCCESS) {
2230                 /**
2231                  * check if we want to forward MAC frames - driver doesn't have
2232                  * native capability to do that, so we'll write the registers
2233                  * ourselves
2234                  **/
2235                 rctl = IGC_READ_REG(hw, IGC_RCTL);
2236
2237                 /* set or clear MFLCN.PMCF bit depending on configuration */
2238                 if (fc_conf->mac_ctrl_frame_fwd != 0)
2239                         rctl |= IGC_RCTL_PMCF;
2240                 else
2241                         rctl &= ~IGC_RCTL_PMCF;
2242
2243                 IGC_WRITE_REG(hw, IGC_RCTL, rctl);
2244                 IGC_WRITE_FLUSH(hw);
2245
2246                 return 0;
2247         }
2248
2249         PMD_DRV_LOG(ERR, "igc_setup_link_generic = 0x%x", err);
2250         return -EIO;
2251 }
2252
2253 static int
2254 eth_igc_rss_reta_update(struct rte_eth_dev *dev,
2255                         struct rte_eth_rss_reta_entry64 *reta_conf,
2256                         uint16_t reta_size)
2257 {
2258         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2259         uint16_t i;
2260
2261         if (reta_size != ETH_RSS_RETA_SIZE_128) {
2262                 PMD_DRV_LOG(ERR,
2263                         "The size of RSS redirection table configured(%d) doesn't match the number hardware can supported(%d)",
2264                         reta_size, ETH_RSS_RETA_SIZE_128);
2265                 return -EINVAL;
2266         }
2267
2268         RTE_BUILD_BUG_ON(ETH_RSS_RETA_SIZE_128 % IGC_RSS_RDT_REG_SIZE);
2269
2270         /* set redirection table */
2271         for (i = 0; i < ETH_RSS_RETA_SIZE_128; i += IGC_RSS_RDT_REG_SIZE) {
2272                 union igc_rss_reta_reg reta, reg;
2273                 uint16_t idx, shift;
2274                 uint8_t j, mask;
2275
2276                 idx = i / RTE_RETA_GROUP_SIZE;
2277                 shift = i % RTE_RETA_GROUP_SIZE;
2278                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2279                                 IGC_RSS_RDT_REG_SIZE_MASK);
2280
2281                 /* if no need to update the register */
2282                 if (!mask ||
2283                     shift > (RTE_RETA_GROUP_SIZE - IGC_RSS_RDT_REG_SIZE))
2284                         continue;
2285
2286                 /* check mask whether need to read the register value first */
2287                 if (mask == IGC_RSS_RDT_REG_SIZE_MASK)
2288                         reg.dword = 0;
2289                 else
2290                         reg.dword = IGC_READ_REG_LE_VALUE(hw,
2291                                         IGC_RETA(i / IGC_RSS_RDT_REG_SIZE));
2292
2293                 /* update the register */
2294                 RTE_BUILD_BUG_ON(sizeof(reta.bytes) != IGC_RSS_RDT_REG_SIZE);
2295                 for (j = 0; j < IGC_RSS_RDT_REG_SIZE; j++) {
2296                         if (mask & (1u << j))
2297                                 reta.bytes[j] =
2298                                         (uint8_t)reta_conf[idx].reta[shift + j];
2299                         else
2300                                 reta.bytes[j] = reg.bytes[j];
2301                 }
2302                 IGC_WRITE_REG_LE_VALUE(hw,
2303                         IGC_RETA(i / IGC_RSS_RDT_REG_SIZE), reta.dword);
2304         }
2305
2306         return 0;
2307 }
2308
2309 static int
2310 eth_igc_rss_reta_query(struct rte_eth_dev *dev,
2311                        struct rte_eth_rss_reta_entry64 *reta_conf,
2312                        uint16_t reta_size)
2313 {
2314         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2315         uint16_t i;
2316
2317         if (reta_size != ETH_RSS_RETA_SIZE_128) {
2318                 PMD_DRV_LOG(ERR,
2319                         "The size of RSS redirection table configured(%d) doesn't match the number hardware can supported(%d)",
2320                         reta_size, ETH_RSS_RETA_SIZE_128);
2321                 return -EINVAL;
2322         }
2323
2324         RTE_BUILD_BUG_ON(ETH_RSS_RETA_SIZE_128 % IGC_RSS_RDT_REG_SIZE);
2325
2326         /* read redirection table */
2327         for (i = 0; i < ETH_RSS_RETA_SIZE_128; i += IGC_RSS_RDT_REG_SIZE) {
2328                 union igc_rss_reta_reg reta;
2329                 uint16_t idx, shift;
2330                 uint8_t j, mask;
2331
2332                 idx = i / RTE_RETA_GROUP_SIZE;
2333                 shift = i % RTE_RETA_GROUP_SIZE;
2334                 mask = (uint8_t)((reta_conf[idx].mask >> shift) &
2335                                 IGC_RSS_RDT_REG_SIZE_MASK);
2336
2337                 /* if no need to read register */
2338                 if (!mask ||
2339                     shift > (RTE_RETA_GROUP_SIZE - IGC_RSS_RDT_REG_SIZE))
2340                         continue;
2341
2342                 /* read register and get the queue index */
2343                 RTE_BUILD_BUG_ON(sizeof(reta.bytes) != IGC_RSS_RDT_REG_SIZE);
2344                 reta.dword = IGC_READ_REG_LE_VALUE(hw,
2345                                 IGC_RETA(i / IGC_RSS_RDT_REG_SIZE));
2346                 for (j = 0; j < IGC_RSS_RDT_REG_SIZE; j++) {
2347                         if (mask & (1u << j))
2348                                 reta_conf[idx].reta[shift + j] = reta.bytes[j];
2349                 }
2350         }
2351
2352         return 0;
2353 }
2354
2355 static int
2356 eth_igc_rss_hash_update(struct rte_eth_dev *dev,
2357                         struct rte_eth_rss_conf *rss_conf)
2358 {
2359         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2360         igc_hw_rss_hash_set(hw, rss_conf);
2361         return 0;
2362 }
2363
2364 static int
2365 eth_igc_rss_hash_conf_get(struct rte_eth_dev *dev,
2366                         struct rte_eth_rss_conf *rss_conf)
2367 {
2368         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2369         uint32_t *hash_key = (uint32_t *)rss_conf->rss_key;
2370         uint32_t mrqc;
2371         uint64_t rss_hf;
2372
2373         if (hash_key != NULL) {
2374                 int i;
2375
2376                 /* if not enough space for store hash key */
2377                 if (rss_conf->rss_key_len != IGC_HKEY_SIZE) {
2378                         PMD_DRV_LOG(ERR,
2379                                 "RSS hash key size %u in parameter doesn't match the hardware hash key size %u",
2380                                 rss_conf->rss_key_len, IGC_HKEY_SIZE);
2381                         return -EINVAL;
2382                 }
2383
2384                 /* read RSS key from register */
2385                 for (i = 0; i < IGC_HKEY_MAX_INDEX; i++)
2386                         hash_key[i] = IGC_READ_REG_LE_VALUE(hw, IGC_RSSRK(i));
2387         }
2388
2389         /* get RSS functions configured in MRQC register */
2390         mrqc = IGC_READ_REG(hw, IGC_MRQC);
2391         if ((mrqc & IGC_MRQC_ENABLE_RSS_4Q) == 0)
2392                 return 0;
2393
2394         rss_hf = 0;
2395         if (mrqc & IGC_MRQC_RSS_FIELD_IPV4)
2396                 rss_hf |= ETH_RSS_IPV4;
2397         if (mrqc & IGC_MRQC_RSS_FIELD_IPV4_TCP)
2398                 rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
2399         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6)
2400                 rss_hf |= ETH_RSS_IPV6;
2401         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_EX)
2402                 rss_hf |= ETH_RSS_IPV6_EX;
2403         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_TCP)
2404                 rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
2405         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_TCP_EX)
2406                 rss_hf |= ETH_RSS_IPV6_TCP_EX;
2407         if (mrqc & IGC_MRQC_RSS_FIELD_IPV4_UDP)
2408                 rss_hf |= ETH_RSS_NONFRAG_IPV4_UDP;
2409         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_UDP)
2410                 rss_hf |= ETH_RSS_NONFRAG_IPV6_UDP;
2411         if (mrqc & IGC_MRQC_RSS_FIELD_IPV6_UDP_EX)
2412                 rss_hf |= ETH_RSS_IPV6_UDP_EX;
2413
2414         rss_conf->rss_hf |= rss_hf;
2415         return 0;
2416 }
2417
2418 static int
2419 eth_igc_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vlan_id, int on)
2420 {
2421         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2422         struct igc_vfta *shadow_vfta = IGC_DEV_PRIVATE_VFTA(dev);
2423         uint32_t vfta;
2424         uint32_t vid_idx;
2425         uint32_t vid_bit;
2426
2427         vid_idx = (vlan_id >> IGC_VFTA_ENTRY_SHIFT) & IGC_VFTA_ENTRY_MASK;
2428         vid_bit = 1u << (vlan_id & IGC_VFTA_ENTRY_BIT_SHIFT_MASK);
2429         vfta = shadow_vfta->vfta[vid_idx];
2430         if (on)
2431                 vfta |= vid_bit;
2432         else
2433                 vfta &= ~vid_bit;
2434         IGC_WRITE_REG_ARRAY(hw, IGC_VFTA, vid_idx, vfta);
2435
2436         /* update local VFTA copy */
2437         shadow_vfta->vfta[vid_idx] = vfta;
2438
2439         return 0;
2440 }
2441
2442 static void
2443 igc_vlan_hw_filter_disable(struct rte_eth_dev *dev)
2444 {
2445         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2446         igc_read_reg_check_clear_bits(hw, IGC_RCTL,
2447                         IGC_RCTL_CFIEN | IGC_RCTL_VFE);
2448 }
2449
2450 static void
2451 igc_vlan_hw_filter_enable(struct rte_eth_dev *dev)
2452 {
2453         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2454         struct igc_vfta *shadow_vfta = IGC_DEV_PRIVATE_VFTA(dev);
2455         uint32_t reg_val;
2456         int i;
2457
2458         /* Filter Table Enable, CFI not used for packet acceptance */
2459         reg_val = IGC_READ_REG(hw, IGC_RCTL);
2460         reg_val &= ~IGC_RCTL_CFIEN;
2461         reg_val |= IGC_RCTL_VFE;
2462         IGC_WRITE_REG(hw, IGC_RCTL, reg_val);
2463
2464         /* restore VFTA table */
2465         for (i = 0; i < IGC_VFTA_SIZE; i++)
2466                 IGC_WRITE_REG_ARRAY(hw, IGC_VFTA, i, shadow_vfta->vfta[i]);
2467 }
2468
2469 static void
2470 igc_vlan_hw_strip_disable(struct rte_eth_dev *dev)
2471 {
2472         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2473
2474         igc_read_reg_check_clear_bits(hw, IGC_CTRL, IGC_CTRL_VME);
2475 }
2476
2477 static void
2478 igc_vlan_hw_strip_enable(struct rte_eth_dev *dev)
2479 {
2480         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2481
2482         igc_read_reg_check_set_bits(hw, IGC_CTRL, IGC_CTRL_VME);
2483 }
2484
2485 static int
2486 igc_vlan_hw_extend_disable(struct rte_eth_dev *dev)
2487 {
2488         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2489         uint32_t ctrl_ext;
2490
2491         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
2492
2493         /* if extend vlan hasn't been enabled */
2494         if ((ctrl_ext & IGC_CTRL_EXT_EXT_VLAN) == 0)
2495                 return 0;
2496
2497         if ((dev->data->dev_conf.rxmode.offloads &
2498                         DEV_RX_OFFLOAD_JUMBO_FRAME) == 0)
2499                 goto write_ext_vlan;
2500
2501         /* Update maximum packet length */
2502         if (dev->data->dev_conf.rxmode.max_rx_pkt_len <
2503                 RTE_ETHER_MIN_MTU + VLAN_TAG_SIZE) {
2504                 PMD_DRV_LOG(ERR, "Maximum packet length %u error, min is %u",
2505                         dev->data->dev_conf.rxmode.max_rx_pkt_len,
2506                         VLAN_TAG_SIZE + RTE_ETHER_MIN_MTU);
2507                 return -EINVAL;
2508         }
2509         dev->data->dev_conf.rxmode.max_rx_pkt_len -= VLAN_TAG_SIZE;
2510         IGC_WRITE_REG(hw, IGC_RLPML,
2511                 dev->data->dev_conf.rxmode.max_rx_pkt_len);
2512
2513 write_ext_vlan:
2514         IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext & ~IGC_CTRL_EXT_EXT_VLAN);
2515         return 0;
2516 }
2517
2518 static int
2519 igc_vlan_hw_extend_enable(struct rte_eth_dev *dev)
2520 {
2521         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2522         uint32_t ctrl_ext;
2523
2524         ctrl_ext = IGC_READ_REG(hw, IGC_CTRL_EXT);
2525
2526         /* if extend vlan has been enabled */
2527         if (ctrl_ext & IGC_CTRL_EXT_EXT_VLAN)
2528                 return 0;
2529
2530         if ((dev->data->dev_conf.rxmode.offloads &
2531                         DEV_RX_OFFLOAD_JUMBO_FRAME) == 0)
2532                 goto write_ext_vlan;
2533
2534         /* Update maximum packet length */
2535         if (dev->data->dev_conf.rxmode.max_rx_pkt_len >
2536                 MAX_RX_JUMBO_FRAME_SIZE - VLAN_TAG_SIZE) {
2537                 PMD_DRV_LOG(ERR, "Maximum packet length %u error, max is %u",
2538                         dev->data->dev_conf.rxmode.max_rx_pkt_len +
2539                         VLAN_TAG_SIZE, MAX_RX_JUMBO_FRAME_SIZE);
2540                 return -EINVAL;
2541         }
2542         dev->data->dev_conf.rxmode.max_rx_pkt_len += VLAN_TAG_SIZE;
2543         IGC_WRITE_REG(hw, IGC_RLPML,
2544                 dev->data->dev_conf.rxmode.max_rx_pkt_len);
2545
2546 write_ext_vlan:
2547         IGC_WRITE_REG(hw, IGC_CTRL_EXT, ctrl_ext | IGC_CTRL_EXT_EXT_VLAN);
2548         return 0;
2549 }
2550
2551 static int
2552 eth_igc_vlan_offload_set(struct rte_eth_dev *dev, int mask)
2553 {
2554         struct rte_eth_rxmode *rxmode;
2555
2556         rxmode = &dev->data->dev_conf.rxmode;
2557         if (mask & ETH_VLAN_STRIP_MASK) {
2558                 if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2559                         igc_vlan_hw_strip_enable(dev);
2560                 else
2561                         igc_vlan_hw_strip_disable(dev);
2562         }
2563
2564         if (mask & ETH_VLAN_FILTER_MASK) {
2565                 if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2566                         igc_vlan_hw_filter_enable(dev);
2567                 else
2568                         igc_vlan_hw_filter_disable(dev);
2569         }
2570
2571         if (mask & ETH_VLAN_EXTEND_MASK) {
2572                 if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2573                         return igc_vlan_hw_extend_enable(dev);
2574                 else
2575                         return igc_vlan_hw_extend_disable(dev);
2576         }
2577
2578         return 0;
2579 }
2580
2581 static int
2582 eth_igc_vlan_tpid_set(struct rte_eth_dev *dev,
2583                       enum rte_vlan_type vlan_type,
2584                       uint16_t tpid)
2585 {
2586         struct igc_hw *hw = IGC_DEV_PRIVATE_HW(dev);
2587         uint32_t reg_val;
2588
2589         /* only outer TPID of double VLAN can be configured*/
2590         if (vlan_type == ETH_VLAN_TYPE_OUTER) {
2591                 reg_val = IGC_READ_REG(hw, IGC_VET);
2592                 reg_val = (reg_val & (~IGC_VET_EXT)) |
2593                         ((uint32_t)tpid << IGC_VET_EXT_SHIFT);
2594                 IGC_WRITE_REG(hw, IGC_VET, reg_val);
2595
2596                 return 0;
2597         }
2598
2599         /* all other TPID values are read-only*/
2600         PMD_DRV_LOG(ERR, "Not supported");
2601         return -ENOTSUP;
2602 }
2603
2604 static int
2605 eth_igc_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2606         struct rte_pci_device *pci_dev)
2607 {
2608         PMD_INIT_FUNC_TRACE();
2609         return rte_eth_dev_pci_generic_probe(pci_dev,
2610                 sizeof(struct igc_adapter), eth_igc_dev_init);
2611 }
2612
2613 static int
2614 eth_igc_pci_remove(struct rte_pci_device *pci_dev)
2615 {
2616         PMD_INIT_FUNC_TRACE();
2617         return rte_eth_dev_pci_generic_remove(pci_dev, eth_igc_dev_uninit);
2618 }
2619
2620 static struct rte_pci_driver rte_igc_pmd = {
2621         .id_table = pci_id_igc_map,
2622         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
2623         .probe = eth_igc_pci_probe,
2624         .remove = eth_igc_pci_remove,
2625 };
2626
2627 RTE_PMD_REGISTER_PCI(net_igc, rte_igc_pmd);
2628 RTE_PMD_REGISTER_PCI_TABLE(net_igc, pci_id_igc_map);
2629 RTE_PMD_REGISTER_KMOD_DEP(net_igc, "* igb_uio | uio_pci_generic | vfio-pci");