drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2021 Xilinx, Inc.
   4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17 #include <rte_ether.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_debug.h"
  23 #include "sfc_log.h"
  24 #include "sfc_kvargs.h"
  25 #include "sfc_ev.h"
  26 #include "sfc_rx.h"
  27 #include "sfc_tx.h"
  28 #include "sfc_flow.h"
  29 #include "sfc_dp.h"
  30 #include "sfc_dp_rx.h"
  31 #include "sfc_sw_stats.h"
  32
  33 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
  34 #define SFC_XSTAT_ID_INVALID_NAME '\0'
  35
  36 uint32_t sfc_logtype_driver;
  37
  38 static struct sfc_dp_list sfc_dp_head =
  39         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  40
  41
  42 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  43
  44
  45 static int
  46 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  47 {
  48         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  49         efx_nic_fw_info_t enfi;
  50         int ret;
  51         int rc;
  52
  53         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  54         if (rc != 0)
  55                 return -rc;
  56
  57         ret = snprintf(fw_version, fw_size,
  58                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  59                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  60                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  61         if (ret < 0)
  62                 return ret;
  63
  64         if (enfi.enfi_dpcpu_fw_ids_valid) {
  65                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  66                 int ret_extra;
  67
  68                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  69                                      fw_size - dpcpu_fw_ids_offset,
  70                                      " rx%" PRIx16 " tx%" PRIx16,
  71                                      enfi.enfi_rx_dpcpu_fw_id,
  72                                      enfi.enfi_tx_dpcpu_fw_id);
  73                 if (ret_extra < 0)
  74                         return ret_extra;
  75
  76                 ret += ret_extra;
  77         }
  78
  79         if (fw_size < (size_t)(++ret))
  80                 return ret;
  81         else
  82                 return 0;
  83 }
  84
  85 static int
  86 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  87 {
  88         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  89         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  90         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  91         struct sfc_rss *rss = &sas->rss;
  92         struct sfc_mae *mae = &sa->mae;
  93         uint64_t txq_offloads_def = 0;
  94
  95         sfc_log_init(sa, "entry");
  96
  97         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
  98         dev_info->max_mtu = EFX_MAC_SDU_MAX;
  99
 100         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 101
 102         dev_info->max_vfs = sa->sriov.num_vfs;
 103
 104         /* Autonegotiation may be disabled */
 105         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 106         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 107                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 108         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 109                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 110         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 111                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 112         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 113                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 114         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 115                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 116         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 117                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 118
 119         dev_info->max_rx_queues = sa->rxq_max;
 120         dev_info->max_tx_queues = sa->txq_max;
 121
 122         /* By default packets are dropped if no descriptors are available */
 123         dev_info->default_rxconf.rx_drop_en = 1;
 124
 125         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 126
 127         /*
 128          * rx_offload_capa includes both device and queue offloads since
 129          * the latter may be requested on a per device basis which makes
 130          * sense when some offloads are needed to be set on all queues.
 131          */
 132         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 133                                     dev_info->rx_queue_offload_capa;
 134
 135         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 136
 137         /*
 138          * tx_offload_capa includes both device and queue offloads since
 139          * the latter may be requested on a per device basis which makes
 140          * sense when some offloads are needed to be set on all queues.
 141          */
 142         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 143                                     dev_info->tx_queue_offload_capa;
 144
 145         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 146                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 147
 148         dev_info->default_txconf.offloads |= txq_offloads_def;
 149
 150         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 151                 uint64_t rte_hf = 0;
 152                 unsigned int i;
 153
 154                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 155                         rte_hf |= rss->hf_map[i].rte;
 156
 157                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 158                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 159                 dev_info->flow_type_rss_offloads = rte_hf;
 160         }
 161
 162         /* Initialize to hardware limits */
 163         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 164         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 165         /* The RXQ hardware requires that the descriptor count is a power
 166          * of 2, but rx_desc_lim cannot properly describe that constraint.
 167          */
 168         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 169
 170         /* Initialize to hardware limits */
 171         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 172         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 173         /*
 174          * The TXQ hardware requires that the descriptor count is a power
 175          * of 2, but tx_desc_lim cannot properly describe that constraint
 176          */
 177         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 178
 179         if (sap->dp_rx->get_dev_info != NULL)
 180                 sap->dp_rx->get_dev_info(dev_info);
 181         if (sap->dp_tx->get_dev_info != NULL)
 182                 sap->dp_tx->get_dev_info(dev_info);
 183
 184         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 185                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 186
 187         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
 188                 dev_info->switch_info.name = dev->device->driver->name;
 189                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 190                 dev_info->switch_info.port_id = mae->switch_port_id;
 191         }
 192
 193         return 0;
 194 }
 195
 196 static const uint32_t *
 197 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 198 {
 199         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 200
 201         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 202 }
 203
 204 static int
 205 sfc_dev_configure(struct rte_eth_dev *dev)
 206 {
 207         struct rte_eth_dev_data *dev_data = dev->data;
 208         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 209         int rc;
 210
 211         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 212                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 213
 214         sfc_adapter_lock(sa);
 215         switch (sa->state) {
 216         case SFC_ADAPTER_CONFIGURED:
 217                 /* FALLTHROUGH */
 218         case SFC_ADAPTER_INITIALIZED:
 219                 rc = sfc_configure(sa);
 220                 break;
 221         default:
 222                 sfc_err(sa, "unexpected adapter state %u to configure",
 223                         sa->state);
 224                 rc = EINVAL;
 225                 break;
 226         }
 227         sfc_adapter_unlock(sa);
 228
 229         sfc_log_init(sa, "done %d", rc);
 230         SFC_ASSERT(rc >= 0);
 231         return -rc;
 232 }
 233
 234 static int
 235 sfc_dev_start(struct rte_eth_dev *dev)
 236 {
 237         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 238         int rc;
 239
 240         sfc_log_init(sa, "entry");
 241
 242         sfc_adapter_lock(sa);
 243         rc = sfc_start(sa);
 244         sfc_adapter_unlock(sa);
 245
 246         sfc_log_init(sa, "done %d", rc);
 247         SFC_ASSERT(rc >= 0);
 248         return -rc;
 249 }
 250
 251 static int
 252 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 253 {
 254         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 255         struct rte_eth_link current_link;
 256         int ret;
 257
 258         sfc_log_init(sa, "entry");
 259
 260         if (sa->state != SFC_ADAPTER_STARTED) {
 261                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 262         } else if (wait_to_complete) {
 263                 efx_link_mode_t link_mode;
 264
 265                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 266                         link_mode = EFX_LINK_UNKNOWN;
 267                 sfc_port_link_mode_to_info(link_mode, &current_link);
 268
 269         } else {
 270                 sfc_ev_mgmt_qpoll(sa);
 271                 rte_eth_linkstatus_get(dev, &current_link);
 272         }
 273
 274         ret = rte_eth_linkstatus_set(dev, &current_link);
 275         if (ret == 0)
 276                 sfc_notice(sa, "Link status is %s",
 277                            current_link.link_status ? "UP" : "DOWN");
 278
 279         return ret;
 280 }
 281
 282 static int
 283 sfc_dev_stop(struct rte_eth_dev *dev)
 284 {
 285         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 286
 287         sfc_log_init(sa, "entry");
 288
 289         sfc_adapter_lock(sa);
 290         sfc_stop(sa);
 291         sfc_adapter_unlock(sa);
 292
 293         sfc_log_init(sa, "done");
 294
 295         return 0;
 296 }
 297
 298 static int
 299 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 300 {
 301         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 302         int rc;
 303
 304         sfc_log_init(sa, "entry");
 305
 306         sfc_adapter_lock(sa);
 307         rc = sfc_start(sa);
 308         sfc_adapter_unlock(sa);
 309
 310         SFC_ASSERT(rc >= 0);
 311         return -rc;
 312 }
 313
 314 static int
 315 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 316 {
 317         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 318
 319         sfc_log_init(sa, "entry");
 320
 321         sfc_adapter_lock(sa);
 322         sfc_stop(sa);
 323         sfc_adapter_unlock(sa);
 324
 325         return 0;
 326 }
 327
 328 static void
 329 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 330 {
 331         free(dev->process_private);
 332         rte_eth_dev_release_port(dev);
 333 }
 334
 335 static int
 336 sfc_dev_close(struct rte_eth_dev *dev)
 337 {
 338         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 339
 340         sfc_log_init(sa, "entry");
 341
 342         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 343                 sfc_eth_dev_secondary_clear_ops(dev);
 344                 return 0;
 345         }
 346
 347         sfc_adapter_lock(sa);
 348         switch (sa->state) {
 349         case SFC_ADAPTER_STARTED:
 350                 sfc_stop(sa);
 351                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 352                 /* FALLTHROUGH */
 353         case SFC_ADAPTER_CONFIGURED:
 354                 sfc_close(sa);
 355                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 356                 /* FALLTHROUGH */
 357         case SFC_ADAPTER_INITIALIZED:
 358                 break;
 359         default:
 360                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 361                 break;
 362         }
 363
 364         /*
 365          * Cleanup all resources.
 366          * Rollback primary process sfc_eth_dev_init() below.
 367          */
 368
 369         sfc_eth_dev_clear_ops(dev);
 370
 371         sfc_detach(sa);
 372         sfc_unprobe(sa);
 373
 374         sfc_kvargs_cleanup(sa);
 375
 376         sfc_adapter_unlock(sa);
 377         sfc_adapter_lock_fini(sa);
 378
 379         sfc_log_init(sa, "done");
 380
 381         /* Required for logging, so cleanup last */
 382         sa->eth_dev = NULL;
 383
 384         free(sa);
 385
 386         return 0;
 387 }
 388
 389 static int
 390 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 391                    boolean_t enabled)
 392 {
 393         struct sfc_port *port;
 394         boolean_t *toggle;
 395         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 396         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 397         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 398         int rc = 0;
 399
 400         sfc_adapter_lock(sa);
 401
 402         port = &sa->port;
 403         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 404
 405         if (*toggle != enabled) {
 406                 *toggle = enabled;
 407
 408                 if (sfc_sa2shared(sa)->isolated) {
 409                         sfc_warn(sa, "isolated mode is active on the port");
 410                         sfc_warn(sa, "the change is to be applied on the next "
 411                                      "start provided that isolated mode is "
 412                                      "disabled prior the next start");
 413                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 414                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 415                         *toggle = !(enabled);
 416                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 417                                  ((enabled) ? "enable" : "disable"), desc, rc);
 418
 419                         /*
 420                          * For promiscuous and all-multicast filters a
 421                          * permission failure should be reported as an
 422                          * unsupported filter.
 423                          */
 424                         if (rc == EPERM)
 425                                 rc = ENOTSUP;
 426                 }
 427         }
 428
 429         sfc_adapter_unlock(sa);
 430         return rc;
 431 }
 432
 433 static int
 434 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 435 {
 436         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 437
 438         SFC_ASSERT(rc >= 0);
 439         return -rc;
 440 }
 441
 442 static int
 443 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 444 {
 445         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 446
 447         SFC_ASSERT(rc >= 0);
 448         return -rc;
 449 }
 450
 451 static int
 452 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 453 {
 454         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 455
 456         SFC_ASSERT(rc >= 0);
 457         return -rc;
 458 }
 459
 460 static int
 461 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 462 {
 463         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 464
 465         SFC_ASSERT(rc >= 0);
 466         return -rc;
 467 }
 468
 469 static int
 470 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 471                    uint16_t nb_rx_desc, unsigned int socket_id,
 472                    const struct rte_eth_rxconf *rx_conf,
 473                    struct rte_mempool *mb_pool)
 474 {
 475         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 476         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 477         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
 478         struct sfc_rxq_info *rxq_info;
 479         sfc_sw_index_t sw_index;
 480         int rc;
 481
 482         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 483                      ethdev_qid, nb_rx_desc, socket_id);
 484
 485         sfc_adapter_lock(sa);
 486
 487         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
 488         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
 489                           rx_conf, mb_pool);
 490         if (rc != 0)
 491                 goto fail_rx_qinit;
 492
 493         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
 494         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
 495
 496         sfc_adapter_unlock(sa);
 497
 498         return 0;
 499
 500 fail_rx_qinit:
 501         sfc_adapter_unlock(sa);
 502         SFC_ASSERT(rc > 0);
 503         return -rc;
 504 }
 505
 506 static void
 507 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 508 {
 509         struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
 510         struct sfc_rxq *rxq;
 511         struct sfc_adapter *sa;
 512         sfc_sw_index_t sw_index;
 513
 514         if (dp_rxq == NULL)
 515                 return;
 516
 517         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 518         sa = rxq->evq->sa;
 519         sfc_adapter_lock(sa);
 520
 521         sw_index = dp_rxq->dpq.queue_id;
 522
 523         sfc_log_init(sa, "RxQ=%u", sw_index);
 524
 525         sfc_rx_qfini(sa, sw_index);
 526
 527         sfc_adapter_unlock(sa);
 528 }
 529
 530 static int
 531 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 532                    uint16_t nb_tx_desc, unsigned int socket_id,
 533                    const struct rte_eth_txconf *tx_conf)
 534 {
 535         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 536         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 537         struct sfc_txq_info *txq_info;
 538         sfc_sw_index_t sw_index;
 539         int rc;
 540
 541         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 542                      ethdev_qid, nb_tx_desc, socket_id);
 543
 544         sfc_adapter_lock(sa);
 545
 546         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
 547         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
 548         if (rc != 0)
 549                 goto fail_tx_qinit;
 550
 551         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
 552         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
 553
 554         sfc_adapter_unlock(sa);
 555         return 0;
 556
 557 fail_tx_qinit:
 558         sfc_adapter_unlock(sa);
 559         SFC_ASSERT(rc > 0);
 560         return -rc;
 561 }
 562
 563 static void
 564 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 565 {
 566         struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
 567         struct sfc_txq *txq;
 568         sfc_sw_index_t sw_index;
 569         struct sfc_adapter *sa;
 570
 571         if (dp_txq == NULL)
 572                 return;
 573
 574         txq = sfc_txq_by_dp_txq(dp_txq);
 575         sw_index = dp_txq->dpq.queue_id;
 576
 577         SFC_ASSERT(txq->evq != NULL);
 578         sa = txq->evq->sa;
 579
 580         sfc_log_init(sa, "TxQ = %u", sw_index);
 581
 582         sfc_adapter_lock(sa);
 583
 584         sfc_tx_qfini(sa, sw_index);
 585
 586         sfc_adapter_unlock(sa);
 587 }
 588
 589 static void
 590 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 591 {
 592         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 593         uint64_t pkts_sum = 0;
 594         uint64_t bytes_sum = 0;
 595         unsigned int i;
 596
 597         for (i = 0; i < sas->ethdev_rxq_count; ++i) {
 598                 struct sfc_rxq_info *rxq_info;
 599
 600                 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
 601                 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
 602                         union sfc_pkts_bytes qstats;
 603
 604                         sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
 605                         pkts_sum += qstats.pkts -
 606                                         sa->sw_stats.reset_rx_pkts[i];
 607                         bytes_sum += qstats.bytes -
 608                                         sa->sw_stats.reset_rx_bytes[i];
 609                 }
 610         }
 611
 612         *pkts = pkts_sum;
 613         *bytes = bytes_sum;
 614 }
 615
 616 /*
 617  * Some statistics are computed as A - B where A and B each increase
 618  * monotonically with some hardware counter(s) and the counters are read
 619  * asynchronously.
 620  *
 621  * If packet X is counted in A, but not counted in B yet, computed value is
 622  * greater than real.
 623  *
 624  * If packet X is not counted in A at the moment of reading the counter,
 625  * but counted in B at the moment of reading the counter, computed value
 626  * is less than real.
 627  *
 628  * However, counter which grows backward is worse evil than slightly wrong
 629  * value. So, let's try to guarantee that it never happens except may be
 630  * the case when the MAC stats are zeroed as a result of a NIC reset.
 631  */
 632 static void
 633 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 634 {
 635         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 636                 *stat = newval;
 637 }
 638
 639 static int
 640 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 641 {
 642         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 643         bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
 644         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 645         struct sfc_port *port = &sa->port;
 646         uint64_t *mac_stats;
 647         int ret;
 648
 649         sfc_adapter_lock(sa);
 650
 651         if (have_dp_rx_stats)
 652                 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
 653
 654         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 655         if (ret != 0)
 656                 goto unlock;
 657
 658         mac_stats = port->mac_stats_buf;
 659
 660         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 661                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 662                 if (!have_dp_rx_stats) {
 663                         stats->ipackets =
 664                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 665                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 666                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 667                         stats->ibytes =
 668                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 669                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 670                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 671
 672                         /* CRC is included in these stats, but shouldn't be */
 673                         stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 674                 }
 675                 stats->opackets =
 676                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 677                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 678                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 679                 stats->obytes =
 680                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 681                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 682                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 683                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 684                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 685
 686                 /* CRC is included in these stats, but shouldn't be */
 687                 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 688         } else {
 689                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 690                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 691
 692                 /* CRC is included in these stats, but shouldn't be */
 693                 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 694
 695                 /*
 696                  * Take into account stats which are whenever supported
 697                  * on EF10. If some stat is not supported by current
 698                  * firmware variant or HW revision, it is guaranteed
 699                  * to be zero in mac_stats.
 700                  */
 701                 stats->imissed =
 702                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 703                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 704                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 705                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 706                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 707                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 708                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 709                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 710                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 711                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 712                 stats->ierrors =
 713                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 714                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 715                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 716                 /* no oerrors counters supported on EF10 */
 717
 718                 if (!have_dp_rx_stats) {
 719                         /* Exclude missed, errors and pauses from Rx packets */
 720                         sfc_update_diff_stat(&port->ipackets,
 721                                 mac_stats[EFX_MAC_RX_PKTS] -
 722                                 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 723                                 stats->imissed - stats->ierrors);
 724                         stats->ipackets = port->ipackets;
 725                         stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
 726                                 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 727                 }
 728         }
 729
 730 unlock:
 731         sfc_adapter_unlock(sa);
 732         SFC_ASSERT(ret >= 0);
 733         return -ret;
 734 }
 735
 736 static int
 737 sfc_stats_reset(struct rte_eth_dev *dev)
 738 {
 739         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 740         struct sfc_port *port = &sa->port;
 741         int rc;
 742
 743         sfc_adapter_lock(sa);
 744
 745         if (sa->state != SFC_ADAPTER_STARTED) {
 746                 /*
 747                  * The operation cannot be done if port is not started; it
 748                  * will be scheduled to be done during the next port start
 749                  */
 750                 port->mac_stats_reset_pending = B_TRUE;
 751                 sfc_adapter_unlock(sa);
 752                 return 0;
 753         }
 754
 755         rc = sfc_port_reset_mac_stats(sa);
 756         if (rc != 0)
 757                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 758
 759         sfc_sw_xstats_reset(sa);
 760
 761         sfc_adapter_unlock(sa);
 762
 763         SFC_ASSERT(rc >= 0);
 764         return -rc;
 765 }
 766
 767 static unsigned int
 768 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 769 {
 770         struct sfc_port *port = &sa->port;
 771         unsigned int nb_supported;
 772
 773         sfc_adapter_lock(sa);
 774         nb_supported = port->mac_stats_nb_supported +
 775                        sfc_sw_xstats_get_nb_supported(sa);
 776         sfc_adapter_unlock(sa);
 777
 778         return nb_supported;
 779 }
 780
 781 static int
 782 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 783                unsigned int xstats_count)
 784 {
 785         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 786         unsigned int nb_written = 0;
 787         unsigned int nb_supported = 0;
 788         int rc;
 789
 790         if (unlikely(xstats == NULL))
 791                 return sfc_xstats_get_nb_supported(sa);
 792
 793         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 794         if (rc < 0)
 795                 return rc;
 796
 797         nb_supported = rc;
 798         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 799                                &nb_supported);
 800
 801         return nb_supported;
 802 }
 803
 804 static int
 805 sfc_xstats_get_names(struct rte_eth_dev *dev,
 806                      struct rte_eth_xstat_name *xstats_names,
 807                      unsigned int xstats_count)
 808 {
 809         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 810         struct sfc_port *port = &sa->port;
 811         unsigned int i;
 812         unsigned int nstats = 0;
 813         unsigned int nb_written = 0;
 814         int ret;
 815
 816         if (unlikely(xstats_names == NULL))
 817                 return sfc_xstats_get_nb_supported(sa);
 818
 819         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 820                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 821                         if (nstats < xstats_count) {
 822                                 strlcpy(xstats_names[nstats].name,
 823                                         efx_mac_stat_name(sa->nic, i),
 824                                         sizeof(xstats_names[0].name));
 825                                 nb_written++;
 826                         }
 827                         nstats++;
 828                 }
 829         }
 830
 831         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 832                                       &nb_written, &nstats);
 833         if (ret != 0) {
 834                 SFC_ASSERT(ret < 0);
 835                 return ret;
 836         }
 837
 838         return nstats;
 839 }
 840
 841 static int
 842 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 843                      uint64_t *values, unsigned int n)
 844 {
 845         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 846         struct sfc_port *port = &sa->port;
 847         unsigned int nb_supported;
 848         unsigned int i;
 849         int rc;
 850
 851         if (unlikely(ids == NULL || values == NULL))
 852                 return -EINVAL;
 853
 854         /*
 855          * Values array could be filled in nonsequential order. Fill values with
 856          * constant indicating invalid ID first.
 857          */
 858         for (i = 0; i < n; i++)
 859                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 860
 861         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 862         if (rc != 0)
 863                 return rc;
 864
 865         nb_supported = port->mac_stats_nb_supported;
 866         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 867
 868         /* Return number of written stats before invalid ID is encountered. */
 869         for (i = 0; i < n; i++) {
 870                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 871                         return i;
 872         }
 873
 874         return n;
 875 }
 876
 877 static int
 878 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 879                            const uint64_t *ids,
 880                            struct rte_eth_xstat_name *xstats_names,
 881                            unsigned int size)
 882 {
 883         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 884         struct sfc_port *port = &sa->port;
 885         unsigned int nb_supported;
 886         unsigned int i;
 887         int ret;
 888
 889         if (unlikely(xstats_names == NULL && ids != NULL) ||
 890             unlikely(xstats_names != NULL && ids == NULL))
 891                 return -EINVAL;
 892
 893         if (unlikely(xstats_names == NULL && ids == NULL))
 894                 return sfc_xstats_get_nb_supported(sa);
 895
 896         /*
 897          * Names array could be filled in nonsequential order. Fill names with
 898          * string indicating invalid ID first.
 899          */
 900         for (i = 0; i < size; i++)
 901                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 902
 903         sfc_adapter_lock(sa);
 904
 905         SFC_ASSERT(port->mac_stats_nb_supported <=
 906                    RTE_DIM(port->mac_stats_by_id));
 907
 908         for (i = 0; i < size; i++) {
 909                 if (ids[i] < port->mac_stats_nb_supported) {
 910                         strlcpy(xstats_names[i].name,
 911                                 efx_mac_stat_name(sa->nic,
 912                                                  port->mac_stats_by_id[ids[i]]),
 913                                 sizeof(xstats_names[0].name));
 914                 }
 915         }
 916
 917         nb_supported = port->mac_stats_nb_supported;
 918
 919         sfc_adapter_unlock(sa);
 920
 921         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 922                                             &nb_supported);
 923         if (ret != 0) {
 924                 SFC_ASSERT(ret < 0);
 925                 return ret;
 926         }
 927
 928         /* Return number of written names before invalid ID is encountered. */
 929         for (i = 0; i < size; i++) {
 930                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 931                         return i;
 932         }
 933
 934         return size;
 935 }
 936
 937 static int
 938 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 939 {
 940         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 941         unsigned int wanted_fc, link_fc;
 942
 943         memset(fc_conf, 0, sizeof(*fc_conf));
 944
 945         sfc_adapter_lock(sa);
 946
 947         if (sa->state == SFC_ADAPTER_STARTED)
 948                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 949         else
 950                 link_fc = sa->port.flow_ctrl;
 951
 952         switch (link_fc) {
 953         case 0:
 954                 fc_conf->mode = RTE_FC_NONE;
 955                 break;
 956         case EFX_FCNTL_RESPOND:
 957                 fc_conf->mode = RTE_FC_RX_PAUSE;
 958                 break;
 959         case EFX_FCNTL_GENERATE:
 960                 fc_conf->mode = RTE_FC_TX_PAUSE;
 961                 break;
 962         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 963                 fc_conf->mode = RTE_FC_FULL;
 964                 break;
 965         default:
 966                 sfc_err(sa, "%s: unexpected flow control value %#x",
 967                         __func__, link_fc);
 968         }
 969
 970         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 971
 972         sfc_adapter_unlock(sa);
 973
 974         return 0;
 975 }
 976
 977 static int
 978 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 979 {
 980         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 981         struct sfc_port *port = &sa->port;
 982         unsigned int fcntl;
 983         int rc;
 984
 985         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 986             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 987             fc_conf->mac_ctrl_frame_fwd != 0) {
 988                 sfc_err(sa, "unsupported flow control settings specified");
 989                 rc = EINVAL;
 990                 goto fail_inval;
 991         }
 992
 993         switch (fc_conf->mode) {
 994         case RTE_FC_NONE:
 995                 fcntl = 0;
 996                 break;
 997         case RTE_FC_RX_PAUSE:
 998                 fcntl = EFX_FCNTL_RESPOND;
 999                 break;
1000         case RTE_FC_TX_PAUSE:
1001                 fcntl = EFX_FCNTL_GENERATE;
1002                 break;
1003         case RTE_FC_FULL:
1004                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1005                 break;
1006         default:
1007                 rc = EINVAL;
1008                 goto fail_inval;
1009         }
1010
1011         sfc_adapter_lock(sa);
1012
1013         if (sa->state == SFC_ADAPTER_STARTED) {
1014                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1015                 if (rc != 0)
1016                         goto fail_mac_fcntl_set;
1017         }
1018
1019         port->flow_ctrl = fcntl;
1020         port->flow_ctrl_autoneg = fc_conf->autoneg;
1021
1022         sfc_adapter_unlock(sa);
1023
1024         return 0;
1025
1026 fail_mac_fcntl_set:
1027         sfc_adapter_unlock(sa);
1028 fail_inval:
1029         SFC_ASSERT(rc > 0);
1030         return -rc;
1031 }
1032
1033 static int
1034 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1035 {
1036         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1037         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1038         boolean_t scatter_enabled;
1039         const char *error;
1040         unsigned int i;
1041
1042         for (i = 0; i < sas->rxq_count; i++) {
1043                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1044                         continue;
1045
1046                 scatter_enabled = (sas->rxq_info[i].type_flags &
1047                                    EFX_RXQ_FLAG_SCATTER);
1048
1049                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1050                                           encp->enc_rx_prefix_size,
1051                                           scatter_enabled,
1052                                           encp->enc_rx_scatter_max, &error)) {
1053                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1054                                 error);
1055                         return EINVAL;
1056                 }
1057         }
1058
1059         return 0;
1060 }
1061
1062 static int
1063 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1064 {
1065         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1066         size_t pdu = EFX_MAC_PDU(mtu);
1067         size_t old_pdu;
1068         int rc;
1069
1070         sfc_log_init(sa, "mtu=%u", mtu);
1071
1072         rc = EINVAL;
1073         if (pdu < EFX_MAC_PDU_MIN) {
1074                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1075                         (unsigned int)mtu, (unsigned int)pdu,
1076                         EFX_MAC_PDU_MIN);
1077                 goto fail_inval;
1078         }
1079         if (pdu > EFX_MAC_PDU_MAX) {
1080                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1081                         (unsigned int)mtu, (unsigned int)pdu,
1082                         (unsigned int)EFX_MAC_PDU_MAX);
1083                 goto fail_inval;
1084         }
1085
1086         sfc_adapter_lock(sa);
1087
1088         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1089         if (rc != 0)
1090                 goto fail_check_scatter;
1091
1092         if (pdu != sa->port.pdu) {
1093                 if (sa->state == SFC_ADAPTER_STARTED) {
1094                         sfc_stop(sa);
1095
1096                         old_pdu = sa->port.pdu;
1097                         sa->port.pdu = pdu;
1098                         rc = sfc_start(sa);
1099                         if (rc != 0)
1100                                 goto fail_start;
1101                 } else {
1102                         sa->port.pdu = pdu;
1103                 }
1104         }
1105
1106         /*
1107          * The driver does not use it, but other PMDs update jumbo frame
1108          * flag and max_rx_pkt_len when MTU is set.
1109          */
1110         if (mtu > RTE_ETHER_MTU) {
1111                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1112                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1113         }
1114
1115         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1116
1117         sfc_adapter_unlock(sa);
1118
1119         sfc_log_init(sa, "done");
1120         return 0;
1121
1122 fail_start:
1123         sa->port.pdu = old_pdu;
1124         if (sfc_start(sa) != 0)
1125                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1126                         "PDU max size - port is stopped",
1127                         (unsigned int)pdu, (unsigned int)old_pdu);
1128
1129 fail_check_scatter:
1130         sfc_adapter_unlock(sa);
1131
1132 fail_inval:
1133         sfc_log_init(sa, "failed %d", rc);
1134         SFC_ASSERT(rc > 0);
1135         return -rc;
1136 }
1137 static int
1138 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1139 {
1140         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1141         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1142         struct sfc_port *port = &sa->port;
1143         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1144         int rc = 0;
1145
1146         sfc_adapter_lock(sa);
1147
1148         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1149                 goto unlock;
1150
1151         /*
1152          * Copy the address to the device private data so that
1153          * it could be recalled in the case of adapter restart.
1154          */
1155         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1156
1157         /*
1158          * Neither of the two following checks can return
1159          * an error. The new MAC address is preserved in
1160          * the device private data and can be activated
1161          * on the next port start if the user prevents
1162          * isolated mode from being enabled.
1163          */
1164         if (sfc_sa2shared(sa)->isolated) {
1165                 sfc_warn(sa, "isolated mode is active on the port");
1166                 sfc_warn(sa, "will not set MAC address");
1167                 goto unlock;
1168         }
1169
1170         if (sa->state != SFC_ADAPTER_STARTED) {
1171                 sfc_notice(sa, "the port is not started");
1172                 sfc_notice(sa, "the new MAC address will be set on port start");
1173
1174                 goto unlock;
1175         }
1176
1177         if (encp->enc_allow_set_mac_with_installed_filters) {
1178                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1179                 if (rc != 0) {
1180                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1181                         goto unlock;
1182                 }
1183
1184                 /*
1185                  * Changing the MAC address by means of MCDI request
1186                  * has no effect on received traffic, therefore
1187                  * we also need to update unicast filters
1188                  */
1189                 rc = sfc_set_rx_mode_unchecked(sa);
1190                 if (rc != 0) {
1191                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1192                         /* Rollback the old address */
1193                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1194                         (void)sfc_set_rx_mode_unchecked(sa);
1195                 }
1196         } else {
1197                 sfc_warn(sa, "cannot set MAC address with filters installed");
1198                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1199                 sfc_warn(sa, "(some traffic may be dropped)");
1200
1201                 /*
1202                  * Since setting MAC address with filters installed is not
1203                  * allowed on the adapter, the new MAC address will be set
1204                  * by means of adapter restart. sfc_start() shall retrieve
1205                  * the new address from the device private data and set it.
1206                  */
1207                 sfc_stop(sa);
1208                 rc = sfc_start(sa);
1209                 if (rc != 0)
1210                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1211         }
1212
1213 unlock:
1214         if (rc != 0)
1215                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1216
1217         sfc_adapter_unlock(sa);
1218
1219         SFC_ASSERT(rc >= 0);
1220         return -rc;
1221 }
1222
1223
1224 static int
1225 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1226                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1227 {
1228         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1229         struct sfc_port *port = &sa->port;
1230         uint8_t *mc_addrs = port->mcast_addrs;
1231         int rc;
1232         unsigned int i;
1233
1234         if (sfc_sa2shared(sa)->isolated) {
1235                 sfc_err(sa, "isolated mode is active on the port");
1236                 sfc_err(sa, "will not set multicast address list");
1237                 return -ENOTSUP;
1238         }
1239
1240         if (mc_addrs == NULL)
1241                 return -ENOBUFS;
1242
1243         if (nb_mc_addr > port->max_mcast_addrs) {
1244                 sfc_err(sa, "too many multicast addresses: %u > %u",
1245                          nb_mc_addr, port->max_mcast_addrs);
1246                 return -EINVAL;
1247         }
1248
1249         for (i = 0; i < nb_mc_addr; ++i) {
1250                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1251                                  EFX_MAC_ADDR_LEN);
1252                 mc_addrs += EFX_MAC_ADDR_LEN;
1253         }
1254
1255         port->nb_mcast_addrs = nb_mc_addr;
1256
1257         if (sa->state != SFC_ADAPTER_STARTED)
1258                 return 0;
1259
1260         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1261                                         port->nb_mcast_addrs);
1262         if (rc != 0)
1263                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1264
1265         SFC_ASSERT(rc >= 0);
1266         return -rc;
1267 }
1268
1269 /*
1270  * The function is used by the secondary process as well. It must not
1271  * use any process-local pointers from the adapter data.
1272  */
1273 static void
1274 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1275                       struct rte_eth_rxq_info *qinfo)
1276 {
1277         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1278         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1279         struct sfc_rxq_info *rxq_info;
1280
1281         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1282
1283         qinfo->mp = rxq_info->refill_mb_pool;
1284         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1285         qinfo->conf.rx_drop_en = 1;
1286         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1287         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1288         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1289                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1290                 qinfo->scattered_rx = 1;
1291         }
1292         qinfo->nb_desc = rxq_info->entries;
1293 }
1294
1295 /*
1296  * The function is used by the secondary process as well. It must not
1297  * use any process-local pointers from the adapter data.
1298  */
1299 static void
1300 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1301                       struct rte_eth_txq_info *qinfo)
1302 {
1303         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1304         struct sfc_txq_info *txq_info;
1305
1306         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1307
1308         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1309
1310         memset(qinfo, 0, sizeof(*qinfo));
1311
1312         qinfo->conf.offloads = txq_info->offloads;
1313         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1314         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1315         qinfo->nb_desc = txq_info->entries;
1316 }
1317
1318 /*
1319  * The function is used by the secondary process as well. It must not
1320  * use any process-local pointers from the adapter data.
1321  */
1322 static uint32_t
1323 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1324 {
1325         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1326         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1327         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1328         struct sfc_rxq_info *rxq_info;
1329
1330         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1331
1332         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1333                 return 0;
1334
1335         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1336 }
1337
1338 /*
1339  * The function is used by the secondary process as well. It must not
1340  * use any process-local pointers from the adapter data.
1341  */
1342 static int
1343 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1344 {
1345         struct sfc_dp_rxq *dp_rxq = queue;
1346         const struct sfc_dp_rx *dp_rx;
1347
1348         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1349
1350         return dp_rx->qdesc_status(dp_rxq, offset);
1351 }
1352
1353 /*
1354  * The function is used by the secondary process as well. It must not
1355  * use any process-local pointers from the adapter data.
1356  */
1357 static int
1358 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1359 {
1360         struct sfc_dp_txq *dp_txq = queue;
1361         const struct sfc_dp_tx *dp_tx;
1362
1363         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1364
1365         return dp_tx->qdesc_status(dp_txq, offset);
1366 }
1367
1368 static int
1369 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1370 {
1371         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1372         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1373         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1374         struct sfc_rxq_info *rxq_info;
1375         sfc_sw_index_t sw_index;
1376         int rc;
1377
1378         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1379
1380         sfc_adapter_lock(sa);
1381
1382         rc = EINVAL;
1383         if (sa->state != SFC_ADAPTER_STARTED)
1384                 goto fail_not_started;
1385
1386         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1387         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1388                 goto fail_not_setup;
1389
1390         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1391         rc = sfc_rx_qstart(sa, sw_index);
1392         if (rc != 0)
1393                 goto fail_rx_qstart;
1394
1395         rxq_info->deferred_started = B_TRUE;
1396
1397         sfc_adapter_unlock(sa);
1398
1399         return 0;
1400
1401 fail_rx_qstart:
1402 fail_not_setup:
1403 fail_not_started:
1404         sfc_adapter_unlock(sa);
1405         SFC_ASSERT(rc > 0);
1406         return -rc;
1407 }
1408
1409 static int
1410 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1411 {
1412         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1413         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1414         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1415         struct sfc_rxq_info *rxq_info;
1416         sfc_sw_index_t sw_index;
1417
1418         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1419
1420         sfc_adapter_lock(sa);
1421
1422         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1423         sfc_rx_qstop(sa, sw_index);
1424
1425         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1426         rxq_info->deferred_started = B_FALSE;
1427
1428         sfc_adapter_unlock(sa);
1429
1430         return 0;
1431 }
1432
1433 static int
1434 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1435 {
1436         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1437         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1438         struct sfc_txq_info *txq_info;
1439         sfc_sw_index_t sw_index;
1440         int rc;
1441
1442         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1443
1444         sfc_adapter_lock(sa);
1445
1446         rc = EINVAL;
1447         if (sa->state != SFC_ADAPTER_STARTED)
1448                 goto fail_not_started;
1449
1450         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1451         if (txq_info->state != SFC_TXQ_INITIALIZED)
1452                 goto fail_not_setup;
1453
1454         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1455         rc = sfc_tx_qstart(sa, sw_index);
1456         if (rc != 0)
1457                 goto fail_tx_qstart;
1458
1459         txq_info->deferred_started = B_TRUE;
1460
1461         sfc_adapter_unlock(sa);
1462         return 0;
1463
1464 fail_tx_qstart:
1465
1466 fail_not_setup:
1467 fail_not_started:
1468         sfc_adapter_unlock(sa);
1469         SFC_ASSERT(rc > 0);
1470         return -rc;
1471 }
1472
1473 static int
1474 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1475 {
1476         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1477         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1478         struct sfc_txq_info *txq_info;
1479         sfc_sw_index_t sw_index;
1480
1481         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1482
1483         sfc_adapter_lock(sa);
1484
1485         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1486         sfc_tx_qstop(sa, sw_index);
1487
1488         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1489         txq_info->deferred_started = B_FALSE;
1490
1491         sfc_adapter_unlock(sa);
1492         return 0;
1493 }
1494
1495 static efx_tunnel_protocol_t
1496 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1497 {
1498         switch (rte_type) {
1499         case RTE_TUNNEL_TYPE_VXLAN:
1500                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1501         case RTE_TUNNEL_TYPE_GENEVE:
1502                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1503         default:
1504                 return EFX_TUNNEL_NPROTOS;
1505         }
1506 }
1507
1508 enum sfc_udp_tunnel_op_e {
1509         SFC_UDP_TUNNEL_ADD_PORT,
1510         SFC_UDP_TUNNEL_DEL_PORT,
1511 };
1512
1513 static int
1514 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1515                       struct rte_eth_udp_tunnel *tunnel_udp,
1516                       enum sfc_udp_tunnel_op_e op)
1517 {
1518         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1519         efx_tunnel_protocol_t tunnel_proto;
1520         int rc;
1521
1522         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1523                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1524                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1525                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1526
1527         tunnel_proto =
1528                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1529         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1530                 rc = ENOTSUP;
1531                 goto fail_bad_proto;
1532         }
1533
1534         sfc_adapter_lock(sa);
1535
1536         switch (op) {
1537         case SFC_UDP_TUNNEL_ADD_PORT:
1538                 rc = efx_tunnel_config_udp_add(sa->nic,
1539                                                tunnel_udp->udp_port,
1540                                                tunnel_proto);
1541                 break;
1542         case SFC_UDP_TUNNEL_DEL_PORT:
1543                 rc = efx_tunnel_config_udp_remove(sa->nic,
1544                                                   tunnel_udp->udp_port,
1545                                                   tunnel_proto);
1546                 break;
1547         default:
1548                 rc = EINVAL;
1549                 goto fail_bad_op;
1550         }
1551
1552         if (rc != 0)
1553                 goto fail_op;
1554
1555         if (sa->state == SFC_ADAPTER_STARTED) {
1556                 rc = efx_tunnel_reconfigure(sa->nic);
1557                 if (rc == EAGAIN) {
1558                         /*
1559                          * Configuration is accepted by FW and MC reboot
1560                          * is initiated to apply the changes. MC reboot
1561                          * will be handled in a usual way (MC reboot
1562                          * event on management event queue and adapter
1563                          * restart).
1564                          */
1565                         rc = 0;
1566                 } else if (rc != 0) {
1567                         goto fail_reconfigure;
1568                 }
1569         }
1570
1571         sfc_adapter_unlock(sa);
1572         return 0;
1573
1574 fail_reconfigure:
1575         /* Remove/restore entry since the change makes the trouble */
1576         switch (op) {
1577         case SFC_UDP_TUNNEL_ADD_PORT:
1578                 (void)efx_tunnel_config_udp_remove(sa->nic,
1579                                                    tunnel_udp->udp_port,
1580                                                    tunnel_proto);
1581                 break;
1582         case SFC_UDP_TUNNEL_DEL_PORT:
1583                 (void)efx_tunnel_config_udp_add(sa->nic,
1584                                                 tunnel_udp->udp_port,
1585                                                 tunnel_proto);
1586                 break;
1587         }
1588
1589 fail_op:
1590 fail_bad_op:
1591         sfc_adapter_unlock(sa);
1592
1593 fail_bad_proto:
1594         SFC_ASSERT(rc > 0);
1595         return -rc;
1596 }
1597
1598 static int
1599 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1600                             struct rte_eth_udp_tunnel *tunnel_udp)
1601 {
1602         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1603 }
1604
1605 static int
1606 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1607                             struct rte_eth_udp_tunnel *tunnel_udp)
1608 {
1609         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1610 }
1611
1612 /*
1613  * The function is used by the secondary process as well. It must not
1614  * use any process-local pointers from the adapter data.
1615  */
1616 static int
1617 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1618                           struct rte_eth_rss_conf *rss_conf)
1619 {
1620         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1621         struct sfc_rss *rss = &sas->rss;
1622
1623         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1624                 return -ENOTSUP;
1625
1626         /*
1627          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1628          * hence, conversion is done here to derive a correct set of ETH_RSS
1629          * flags which corresponds to the active EFX configuration stored
1630          * locally in 'sfc_adapter' and kept up-to-date
1631          */
1632         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1633         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1634         if (rss_conf->rss_key != NULL)
1635                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1636
1637         return 0;
1638 }
1639
1640 static int
1641 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1642                         struct rte_eth_rss_conf *rss_conf)
1643 {
1644         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1645         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1646         unsigned int efx_hash_types;
1647         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1648         unsigned int n_contexts;
1649         unsigned int mode_i = 0;
1650         unsigned int key_i = 0;
1651         unsigned int i = 0;
1652         int rc = 0;
1653
1654         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1655
1656         if (sfc_sa2shared(sa)->isolated)
1657                 return -ENOTSUP;
1658
1659         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1660                 sfc_err(sa, "RSS is not available");
1661                 return -ENOTSUP;
1662         }
1663
1664         if (rss->channels == 0) {
1665                 sfc_err(sa, "RSS is not configured");
1666                 return -EINVAL;
1667         }
1668
1669         if ((rss_conf->rss_key != NULL) &&
1670             (rss_conf->rss_key_len != sizeof(rss->key))) {
1671                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1672                         sizeof(rss->key));
1673                 return -EINVAL;
1674         }
1675
1676         sfc_adapter_lock(sa);
1677
1678         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1679         if (rc != 0)
1680                 goto fail_rx_hf_rte_to_efx;
1681
1682         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1683                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1684                                            rss->hash_alg, efx_hash_types,
1685                                            B_TRUE);
1686                 if (rc != 0)
1687                         goto fail_scale_mode_set;
1688         }
1689
1690         if (rss_conf->rss_key != NULL) {
1691                 if (sa->state == SFC_ADAPTER_STARTED) {
1692                         for (key_i = 0; key_i < n_contexts; key_i++) {
1693                                 rc = efx_rx_scale_key_set(sa->nic,
1694                                                           contexts[key_i],
1695                                                           rss_conf->rss_key,
1696                                                           sizeof(rss->key));
1697                                 if (rc != 0)
1698                                         goto fail_scale_key_set;
1699                         }
1700                 }
1701
1702                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1703         }
1704
1705         rss->hash_types = efx_hash_types;
1706
1707         sfc_adapter_unlock(sa);
1708
1709         return 0;
1710
1711 fail_scale_key_set:
1712         for (i = 0; i < key_i; i++) {
1713                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1714                                          sizeof(rss->key)) != 0)
1715                         sfc_err(sa, "failed to restore RSS key");
1716         }
1717
1718 fail_scale_mode_set:
1719         for (i = 0; i < mode_i; i++) {
1720                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1721                                           EFX_RX_HASHALG_TOEPLITZ,
1722                                           rss->hash_types, B_TRUE) != 0)
1723                         sfc_err(sa, "failed to restore RSS mode");
1724         }
1725
1726 fail_rx_hf_rte_to_efx:
1727         sfc_adapter_unlock(sa);
1728         return -rc;
1729 }
1730
1731 /*
1732  * The function is used by the secondary process as well. It must not
1733  * use any process-local pointers from the adapter data.
1734  */
1735 static int
1736 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1737                        struct rte_eth_rss_reta_entry64 *reta_conf,
1738                        uint16_t reta_size)
1739 {
1740         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1741         struct sfc_rss *rss = &sas->rss;
1742         int entry;
1743
1744         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1745                 return -ENOTSUP;
1746
1747         if (rss->channels == 0)
1748                 return -EINVAL;
1749
1750         if (reta_size != EFX_RSS_TBL_SIZE)
1751                 return -EINVAL;
1752
1753         for (entry = 0; entry < reta_size; entry++) {
1754                 int grp = entry / RTE_RETA_GROUP_SIZE;
1755                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1756
1757                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1758                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1759         }
1760
1761         return 0;
1762 }
1763
1764 static int
1765 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1766                         struct rte_eth_rss_reta_entry64 *reta_conf,
1767                         uint16_t reta_size)
1768 {
1769         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1770         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1771         unsigned int *rss_tbl_new;
1772         uint16_t entry;
1773         int rc = 0;
1774
1775
1776         if (sfc_sa2shared(sa)->isolated)
1777                 return -ENOTSUP;
1778
1779         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1780                 sfc_err(sa, "RSS is not available");
1781                 return -ENOTSUP;
1782         }
1783
1784         if (rss->channels == 0) {
1785                 sfc_err(sa, "RSS is not configured");
1786                 return -EINVAL;
1787         }
1788
1789         if (reta_size != EFX_RSS_TBL_SIZE) {
1790                 sfc_err(sa, "RETA size is wrong (should be %u)",
1791                         EFX_RSS_TBL_SIZE);
1792                 return -EINVAL;
1793         }
1794
1795         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1796         if (rss_tbl_new == NULL)
1797                 return -ENOMEM;
1798
1799         sfc_adapter_lock(sa);
1800
1801         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1802
1803         for (entry = 0; entry < reta_size; entry++) {
1804                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1805                 struct rte_eth_rss_reta_entry64 *grp;
1806
1807                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1808
1809                 if (grp->mask & (1ull << grp_idx)) {
1810                         if (grp->reta[grp_idx] >= rss->channels) {
1811                                 rc = EINVAL;
1812                                 goto bad_reta_entry;
1813                         }
1814                         rss_tbl_new[entry] = grp->reta[grp_idx];
1815                 }
1816         }
1817
1818         if (sa->state == SFC_ADAPTER_STARTED) {
1819                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1820                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1821                 if (rc != 0)
1822                         goto fail_scale_tbl_set;
1823         }
1824
1825         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1826
1827 fail_scale_tbl_set:
1828 bad_reta_entry:
1829         sfc_adapter_unlock(sa);
1830
1831         rte_free(rss_tbl_new);
1832
1833         SFC_ASSERT(rc >= 0);
1834         return -rc;
1835 }
1836
1837 static int
1838 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1839                      const struct rte_flow_ops **ops)
1840 {
1841         *ops = &sfc_flow_ops;
1842         return 0;
1843 }
1844
1845 static int
1846 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1847 {
1848         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1849
1850         /*
1851          * If Rx datapath does not provide callback to check mempool,
1852          * all pools are supported.
1853          */
1854         if (sap->dp_rx->pool_ops_supported == NULL)
1855                 return 1;
1856
1857         return sap->dp_rx->pool_ops_supported(pool);
1858 }
1859
1860 static int
1861 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1862 {
1863         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1864         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1865         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1866         struct sfc_rxq_info *rxq_info;
1867
1868         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1869
1870         return sap->dp_rx->intr_enable(rxq_info->dp);
1871 }
1872
1873 static int
1874 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1875 {
1876         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1877         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1878         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1879         struct sfc_rxq_info *rxq_info;
1880
1881         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1882
1883         return sap->dp_rx->intr_disable(rxq_info->dp);
1884 }
1885
1886 static const struct eth_dev_ops sfc_eth_dev_ops = {
1887         .dev_configure                  = sfc_dev_configure,
1888         .dev_start                      = sfc_dev_start,
1889         .dev_stop                       = sfc_dev_stop,
1890         .dev_set_link_up                = sfc_dev_set_link_up,
1891         .dev_set_link_down              = sfc_dev_set_link_down,
1892         .dev_close                      = sfc_dev_close,
1893         .promiscuous_enable             = sfc_dev_promisc_enable,
1894         .promiscuous_disable            = sfc_dev_promisc_disable,
1895         .allmulticast_enable            = sfc_dev_allmulti_enable,
1896         .allmulticast_disable           = sfc_dev_allmulti_disable,
1897         .link_update                    = sfc_dev_link_update,
1898         .stats_get                      = sfc_stats_get,
1899         .stats_reset                    = sfc_stats_reset,
1900         .xstats_get                     = sfc_xstats_get,
1901         .xstats_reset                   = sfc_stats_reset,
1902         .xstats_get_names               = sfc_xstats_get_names,
1903         .dev_infos_get                  = sfc_dev_infos_get,
1904         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1905         .mtu_set                        = sfc_dev_set_mtu,
1906         .rx_queue_start                 = sfc_rx_queue_start,
1907         .rx_queue_stop                  = sfc_rx_queue_stop,
1908         .tx_queue_start                 = sfc_tx_queue_start,
1909         .tx_queue_stop                  = sfc_tx_queue_stop,
1910         .rx_queue_setup                 = sfc_rx_queue_setup,
1911         .rx_queue_release               = sfc_rx_queue_release,
1912         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1913         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1914         .tx_queue_setup                 = sfc_tx_queue_setup,
1915         .tx_queue_release               = sfc_tx_queue_release,
1916         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1917         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1918         .mac_addr_set                   = sfc_mac_addr_set,
1919         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1920         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1921         .reta_update                    = sfc_dev_rss_reta_update,
1922         .reta_query                     = sfc_dev_rss_reta_query,
1923         .rss_hash_update                = sfc_dev_rss_hash_update,
1924         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1925         .flow_ops_get                   = sfc_dev_flow_ops_get,
1926         .set_mc_addr_list               = sfc_set_mc_addr_list,
1927         .rxq_info_get                   = sfc_rx_queue_info_get,
1928         .txq_info_get                   = sfc_tx_queue_info_get,
1929         .fw_version_get                 = sfc_fw_version_get,
1930         .xstats_get_by_id               = sfc_xstats_get_by_id,
1931         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1932         .pool_ops_supported             = sfc_pool_ops_supported,
1933 };
1934
1935 /**
1936  * Duplicate a string in potentially shared memory required for
1937  * multi-process support.
1938  *
1939  * strdup() allocates from process-local heap/memory.
1940  */
1941 static char *
1942 sfc_strdup(const char *str)
1943 {
1944         size_t size;
1945         char *copy;
1946
1947         if (str == NULL)
1948                 return NULL;
1949
1950         size = strlen(str) + 1;
1951         copy = rte_malloc(__func__, size, 0);
1952         if (copy != NULL)
1953                 rte_memcpy(copy, str, size);
1954
1955         return copy;
1956 }
1957
1958 static int
1959 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1960 {
1961         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1962         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1963         const struct sfc_dp_rx *dp_rx;
1964         const struct sfc_dp_tx *dp_tx;
1965         const efx_nic_cfg_t *encp;
1966         unsigned int avail_caps = 0;
1967         const char *rx_name = NULL;
1968         const char *tx_name = NULL;
1969         int rc;
1970
1971         switch (sa->family) {
1972         case EFX_FAMILY_HUNTINGTON:
1973         case EFX_FAMILY_MEDFORD:
1974         case EFX_FAMILY_MEDFORD2:
1975                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1976                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1977                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1978                 break;
1979         case EFX_FAMILY_RIVERHEAD:
1980                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1981                 break;
1982         default:
1983                 break;
1984         }
1985
1986         encp = efx_nic_cfg_get(sa->nic);
1987         if (encp->enc_rx_es_super_buffer_supported)
1988                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1989
1990         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1991                                 sfc_kvarg_string_handler, &rx_name);
1992         if (rc != 0)
1993                 goto fail_kvarg_rx_datapath;
1994
1995         if (rx_name != NULL) {
1996                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1997                 if (dp_rx == NULL) {
1998                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1999                         rc = ENOENT;
2000                         goto fail_dp_rx;
2001                 }
2002                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2003                         sfc_err(sa,
2004                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2005                                 rx_name);
2006                         rc = EINVAL;
2007                         goto fail_dp_rx_caps;
2008                 }
2009         } else {
2010                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2011                 if (dp_rx == NULL) {
2012                         sfc_err(sa, "Rx datapath by caps %#x not found",
2013                                 avail_caps);
2014                         rc = ENOENT;
2015                         goto fail_dp_rx;
2016                 }
2017         }
2018
2019         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2020         if (sas->dp_rx_name == NULL) {
2021                 rc = ENOMEM;
2022                 goto fail_dp_rx_name;
2023         }
2024
2025         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2026
2027         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2028                                 sfc_kvarg_string_handler, &tx_name);
2029         if (rc != 0)
2030                 goto fail_kvarg_tx_datapath;
2031
2032         if (tx_name != NULL) {
2033                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2034                 if (dp_tx == NULL) {
2035                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2036                         rc = ENOENT;
2037                         goto fail_dp_tx;
2038                 }
2039                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2040                         sfc_err(sa,
2041                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2042                                 tx_name);
2043                         rc = EINVAL;
2044                         goto fail_dp_tx_caps;
2045                 }
2046         } else {
2047                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2048                 if (dp_tx == NULL) {
2049                         sfc_err(sa, "Tx datapath by caps %#x not found",
2050                                 avail_caps);
2051                         rc = ENOENT;
2052                         goto fail_dp_tx;
2053                 }
2054         }
2055
2056         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2057         if (sas->dp_tx_name == NULL) {
2058                 rc = ENOMEM;
2059                 goto fail_dp_tx_name;
2060         }
2061
2062         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2063
2064         sa->priv.dp_rx = dp_rx;
2065         sa->priv.dp_tx = dp_tx;
2066
2067         dev->rx_pkt_burst = dp_rx->pkt_burst;
2068         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2069         dev->tx_pkt_burst = dp_tx->pkt_burst;
2070
2071         dev->rx_queue_count = sfc_rx_queue_count;
2072         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2073         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2074         dev->dev_ops = &sfc_eth_dev_ops;
2075
2076         return 0;
2077
2078 fail_dp_tx_name:
2079 fail_dp_tx_caps:
2080 fail_dp_tx:
2081 fail_kvarg_tx_datapath:
2082         rte_free(sas->dp_rx_name);
2083         sas->dp_rx_name = NULL;
2084
2085 fail_dp_rx_name:
2086 fail_dp_rx_caps:
2087 fail_dp_rx:
2088 fail_kvarg_rx_datapath:
2089         return rc;
2090 }
2091
2092 static void
2093 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2094 {
2095         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2096         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2097
2098         dev->dev_ops = NULL;
2099         dev->tx_pkt_prepare = NULL;
2100         dev->rx_pkt_burst = NULL;
2101         dev->tx_pkt_burst = NULL;
2102
2103         rte_free(sas->dp_tx_name);
2104         sas->dp_tx_name = NULL;
2105         sa->priv.dp_tx = NULL;
2106
2107         rte_free(sas->dp_rx_name);
2108         sas->dp_rx_name = NULL;
2109         sa->priv.dp_rx = NULL;
2110 }
2111
2112 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2113         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2114         .reta_query                     = sfc_dev_rss_reta_query,
2115         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2116         .rxq_info_get                   = sfc_rx_queue_info_get,
2117         .txq_info_get                   = sfc_tx_queue_info_get,
2118 };
2119
2120 static int
2121 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2122 {
2123         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2124         struct sfc_adapter_priv *sap;
2125         const struct sfc_dp_rx *dp_rx;
2126         const struct sfc_dp_tx *dp_tx;
2127         int rc;
2128
2129         /*
2130          * Allocate process private data from heap, since it should not
2131          * be located in shared memory allocated using rte_malloc() API.
2132          */
2133         sap = calloc(1, sizeof(*sap));
2134         if (sap == NULL) {
2135                 rc = ENOMEM;
2136                 goto fail_alloc_priv;
2137         }
2138
2139         sap->logtype_main = logtype_main;
2140
2141         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2142         if (dp_rx == NULL) {
2143                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2144                         "cannot find %s Rx datapath", sas->dp_rx_name);
2145                 rc = ENOENT;
2146                 goto fail_dp_rx;
2147         }
2148         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2149                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2150                         "%s Rx datapath does not support multi-process",
2151                         sas->dp_rx_name);
2152                 rc = EINVAL;
2153                 goto fail_dp_rx_multi_process;
2154         }
2155
2156         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2157         if (dp_tx == NULL) {
2158                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2159                         "cannot find %s Tx datapath", sas->dp_tx_name);
2160                 rc = ENOENT;
2161                 goto fail_dp_tx;
2162         }
2163         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2164                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2165                         "%s Tx datapath does not support multi-process",
2166                         sas->dp_tx_name);
2167                 rc = EINVAL;
2168                 goto fail_dp_tx_multi_process;
2169         }
2170
2171         sap->dp_rx = dp_rx;
2172         sap->dp_tx = dp_tx;
2173
2174         dev->process_private = sap;
2175         dev->rx_pkt_burst = dp_rx->pkt_burst;
2176         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2177         dev->tx_pkt_burst = dp_tx->pkt_burst;
2178         dev->rx_queue_count = sfc_rx_queue_count;
2179         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2180         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2181         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2182
2183         return 0;
2184
2185 fail_dp_tx_multi_process:
2186 fail_dp_tx:
2187 fail_dp_rx_multi_process:
2188 fail_dp_rx:
2189         free(sap);
2190
2191 fail_alloc_priv:
2192         return rc;
2193 }
2194
2195 static void
2196 sfc_register_dp(void)
2197 {
2198         /* Register once */
2199         if (TAILQ_EMPTY(&sfc_dp_head)) {
2200                 /* Prefer EF10 datapath */
2201                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2202                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2203                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2204                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2205
2206                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2207                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2208                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2209                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2210         }
2211 }
2212
2213 static int
2214 sfc_eth_dev_init(struct rte_eth_dev *dev)
2215 {
2216         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2217         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2218         uint32_t logtype_main;
2219         struct sfc_adapter *sa;
2220         int rc;
2221         const efx_nic_cfg_t *encp;
2222         const struct rte_ether_addr *from;
2223         int ret;
2224
2225         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2226                         SFC_EFX_DEV_CLASS_NET) {
2227                 SFC_GENERIC_LOG(DEBUG,
2228                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2229                 return 1;
2230         }
2231
2232         sfc_register_dp();
2233
2234         logtype_main = sfc_register_logtype(&pci_dev->addr,
2235                                             SFC_LOGTYPE_MAIN_STR,
2236                                             RTE_LOG_NOTICE);
2237
2238         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2239                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2240
2241         /* Required for logging */
2242         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2243                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2244                         pci_dev->addr.domain, pci_dev->addr.bus,
2245                         pci_dev->addr.devid, pci_dev->addr.function,
2246                         dev->data->port_id);
2247         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2248                 SFC_GENERIC_LOG(ERR,
2249                         "reserved log prefix is too short for " PCI_PRI_FMT,
2250                         pci_dev->addr.domain, pci_dev->addr.bus,
2251                         pci_dev->addr.devid, pci_dev->addr.function);
2252                 return -EINVAL;
2253         }
2254         sas->pci_addr = pci_dev->addr;
2255         sas->port_id = dev->data->port_id;
2256
2257         /*
2258          * Allocate process private data from heap, since it should not
2259          * be located in shared memory allocated using rte_malloc() API.
2260          */
2261         sa = calloc(1, sizeof(*sa));
2262         if (sa == NULL) {
2263                 rc = ENOMEM;
2264                 goto fail_alloc_sa;
2265         }
2266
2267         dev->process_private = sa;
2268
2269         /* Required for logging */
2270         sa->priv.shared = sas;
2271         sa->priv.logtype_main = logtype_main;
2272
2273         sa->eth_dev = dev;
2274
2275         /* Copy PCI device info to the dev->data */
2276         rte_eth_copy_pci_info(dev, pci_dev);
2277         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2278         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2279
2280         rc = sfc_kvargs_parse(sa);
2281         if (rc != 0)
2282                 goto fail_kvargs_parse;
2283
2284         sfc_log_init(sa, "entry");
2285
2286         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2287         if (dev->data->mac_addrs == NULL) {
2288                 rc = ENOMEM;
2289                 goto fail_mac_addrs;
2290         }
2291
2292         sfc_adapter_lock_init(sa);
2293         sfc_adapter_lock(sa);
2294
2295         sfc_log_init(sa, "probing");
2296         rc = sfc_probe(sa);
2297         if (rc != 0)
2298                 goto fail_probe;
2299
2300         sfc_log_init(sa, "set device ops");
2301         rc = sfc_eth_dev_set_ops(dev);
2302         if (rc != 0)
2303                 goto fail_set_ops;
2304
2305         sfc_log_init(sa, "attaching");
2306         rc = sfc_attach(sa);
2307         if (rc != 0)
2308                 goto fail_attach;
2309
2310         encp = efx_nic_cfg_get(sa->nic);
2311
2312         /*
2313          * The arguments are really reverse order in comparison to
2314          * Linux kernel. Copy from NIC config to Ethernet device data.
2315          */
2316         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2317         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2318
2319         sfc_adapter_unlock(sa);
2320
2321         sfc_log_init(sa, "done");
2322         return 0;
2323
2324 fail_attach:
2325         sfc_eth_dev_clear_ops(dev);
2326
2327 fail_set_ops:
2328         sfc_unprobe(sa);
2329
2330 fail_probe:
2331         sfc_adapter_unlock(sa);
2332         sfc_adapter_lock_fini(sa);
2333         rte_free(dev->data->mac_addrs);
2334         dev->data->mac_addrs = NULL;
2335
2336 fail_mac_addrs:
2337         sfc_kvargs_cleanup(sa);
2338
2339 fail_kvargs_parse:
2340         sfc_log_init(sa, "failed %d", rc);
2341         dev->process_private = NULL;
2342         free(sa);
2343
2344 fail_alloc_sa:
2345         SFC_ASSERT(rc > 0);
2346         return -rc;
2347 }
2348
2349 static int
2350 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2351 {
2352         sfc_dev_close(dev);
2353
2354         return 0;
2355 }
2356
2357 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2358         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2359         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2360         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2361         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2362         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2363         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2364         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2365         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2366         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2367         { .vendor_id = 0 /* sentinel */ }
2368 };
2369
2370 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2371         struct rte_pci_device *pci_dev)
2372 {
2373         return rte_eth_dev_pci_generic_probe(pci_dev,
2374                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2375 }
2376
2377 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2378 {
2379         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2380 }
2381
2382 static struct rte_pci_driver sfc_efx_pmd = {
2383         .id_table = pci_id_sfc_efx_map,
2384         .drv_flags =
2385                 RTE_PCI_DRV_INTR_LSC |
2386                 RTE_PCI_DRV_NEED_MAPPING,
2387         .probe = sfc_eth_dev_pci_probe,
2388         .remove = sfc_eth_dev_pci_remove,
2389 };
2390
2391 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2392 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2393 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2394 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2395         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2396         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2397         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2398         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2399         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2400         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2401
2402 RTE_INIT(sfc_driver_register_logtype)
2403 {
2404         int ret;
2405
2406         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2407                                                    RTE_LOG_NOTICE);
2408         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2409 }