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(void *queue)
 508 {
 509         struct sfc_dp_rxq *dp_rxq = queue;
 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(void *queue)
 565 {
 566         struct sfc_dp_txq *dp_txq = queue;
 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 /*
 590  * Some statistics are computed as A - B where A and B each increase
 591  * monotonically with some hardware counter(s) and the counters are read
 592  * asynchronously.
 593  *
 594  * If packet X is counted in A, but not counted in B yet, computed value is
 595  * greater than real.
 596  *
 597  * If packet X is not counted in A at the moment of reading the counter,
 598  * but counted in B at the moment of reading the counter, computed value
 599  * is less than real.
 600  *
 601  * However, counter which grows backward is worse evil than slightly wrong
 602  * value. So, let's try to guarantee that it never happens except may be
 603  * the case when the MAC stats are zeroed as a result of a NIC reset.
 604  */
 605 static void
 606 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 607 {
 608         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 609                 *stat = newval;
 610 }
 611
 612 static int
 613 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 614 {
 615         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 616         struct sfc_port *port = &sa->port;
 617         uint64_t *mac_stats;
 618         int ret;
 619
 620         sfc_adapter_lock(sa);
 621
 622         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 623         if (ret != 0)
 624                 goto unlock;
 625
 626         mac_stats = port->mac_stats_buf;
 627
 628         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 629                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 630                 stats->ipackets =
 631                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 632                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 633                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 634                 stats->opackets =
 635                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 636                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 637                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 638                 stats->ibytes =
 639                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 640                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 641                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 642                 stats->obytes =
 643                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 644                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 645                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 646                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 647                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 648
 649                 /* CRC is included in these stats, but shouldn't be */
 650                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 651                 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 652         } else {
 653                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 654                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 655                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 656
 657                 /* CRC is included in these stats, but shouldn't be */
 658                 stats->ibytes -= mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 659                 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 660
 661                 /*
 662                  * Take into account stats which are whenever supported
 663                  * on EF10. If some stat is not supported by current
 664                  * firmware variant or HW revision, it is guaranteed
 665                  * to be zero in mac_stats.
 666                  */
 667                 stats->imissed =
 668                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 669                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 670                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 671                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 672                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 673                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 674                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 675                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 676                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 677                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 678                 stats->ierrors =
 679                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 680                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 681                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 682                 /* no oerrors counters supported on EF10 */
 683
 684                 /* Exclude missed, errors and pauses from Rx packets */
 685                 sfc_update_diff_stat(&port->ipackets,
 686                         mac_stats[EFX_MAC_RX_PKTS] -
 687                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 688                         stats->imissed - stats->ierrors);
 689                 stats->ipackets = port->ipackets;
 690         }
 691
 692 unlock:
 693         sfc_adapter_unlock(sa);
 694         SFC_ASSERT(ret >= 0);
 695         return -ret;
 696 }
 697
 698 static int
 699 sfc_stats_reset(struct rte_eth_dev *dev)
 700 {
 701         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 702         struct sfc_port *port = &sa->port;
 703         int rc;
 704
 705         sfc_adapter_lock(sa);
 706
 707         if (sa->state != SFC_ADAPTER_STARTED) {
 708                 /*
 709                  * The operation cannot be done if port is not started; it
 710                  * will be scheduled to be done during the next port start
 711                  */
 712                 port->mac_stats_reset_pending = B_TRUE;
 713                 sfc_adapter_unlock(sa);
 714                 return 0;
 715         }
 716
 717         rc = sfc_port_reset_mac_stats(sa);
 718         if (rc != 0)
 719                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 720
 721         sfc_sw_xstats_reset(sa);
 722
 723         sfc_adapter_unlock(sa);
 724
 725         SFC_ASSERT(rc >= 0);
 726         return -rc;
 727 }
 728
 729 static unsigned int
 730 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 731 {
 732         struct sfc_port *port = &sa->port;
 733         unsigned int nb_supported;
 734
 735         sfc_adapter_lock(sa);
 736         nb_supported = port->mac_stats_nb_supported +
 737                        sfc_sw_xstats_get_nb_supported(sa);
 738         sfc_adapter_unlock(sa);
 739
 740         return nb_supported;
 741 }
 742
 743 static int
 744 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 745                unsigned int xstats_count)
 746 {
 747         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 748         unsigned int nb_written = 0;
 749         unsigned int nb_supported = 0;
 750         int rc;
 751
 752         if (unlikely(xstats == NULL))
 753                 return sfc_xstats_get_nb_supported(sa);
 754
 755         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 756         if (rc < 0)
 757                 return rc;
 758
 759         nb_supported = rc;
 760         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 761                                &nb_supported);
 762
 763         return nb_supported;
 764 }
 765
 766 static int
 767 sfc_xstats_get_names(struct rte_eth_dev *dev,
 768                      struct rte_eth_xstat_name *xstats_names,
 769                      unsigned int xstats_count)
 770 {
 771         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 772         struct sfc_port *port = &sa->port;
 773         unsigned int i;
 774         unsigned int nstats = 0;
 775         unsigned int nb_written = 0;
 776         int ret;
 777
 778         if (unlikely(xstats_names == NULL))
 779                 return sfc_xstats_get_nb_supported(sa);
 780
 781         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 782                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 783                         if (nstats < xstats_count) {
 784                                 strlcpy(xstats_names[nstats].name,
 785                                         efx_mac_stat_name(sa->nic, i),
 786                                         sizeof(xstats_names[0].name));
 787                                 nb_written++;
 788                         }
 789                         nstats++;
 790                 }
 791         }
 792
 793         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 794                                       &nb_written, &nstats);
 795         if (ret != 0) {
 796                 SFC_ASSERT(ret < 0);
 797                 return ret;
 798         }
 799
 800         return nstats;
 801 }
 802
 803 static int
 804 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 805                      uint64_t *values, unsigned int n)
 806 {
 807         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 808         struct sfc_port *port = &sa->port;
 809         unsigned int nb_supported;
 810         unsigned int i;
 811         int rc;
 812
 813         if (unlikely(ids == NULL || values == NULL))
 814                 return -EINVAL;
 815
 816         /*
 817          * Values array could be filled in nonsequential order. Fill values with
 818          * constant indicating invalid ID first.
 819          */
 820         for (i = 0; i < n; i++)
 821                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 822
 823         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 824         if (rc != 0)
 825                 return rc;
 826
 827         nb_supported = port->mac_stats_nb_supported;
 828         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 829
 830         /* Return number of written stats before invalid ID is encountered. */
 831         for (i = 0; i < n; i++) {
 832                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 833                         return i;
 834         }
 835
 836         return n;
 837 }
 838
 839 static int
 840 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 841                            struct rte_eth_xstat_name *xstats_names,
 842                            const uint64_t *ids, unsigned int size)
 843 {
 844         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 845         struct sfc_port *port = &sa->port;
 846         unsigned int nb_supported;
 847         unsigned int i;
 848         int ret;
 849
 850         if (unlikely(xstats_names == NULL && ids != NULL) ||
 851             unlikely(xstats_names != NULL && ids == NULL))
 852                 return -EINVAL;
 853
 854         if (unlikely(xstats_names == NULL && ids == NULL))
 855                 return sfc_xstats_get_nb_supported(sa);
 856
 857         /*
 858          * Names array could be filled in nonsequential order. Fill names with
 859          * string indicating invalid ID first.
 860          */
 861         for (i = 0; i < size; i++)
 862                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 863
 864         sfc_adapter_lock(sa);
 865
 866         SFC_ASSERT(port->mac_stats_nb_supported <=
 867                    RTE_DIM(port->mac_stats_by_id));
 868
 869         for (i = 0; i < size; i++) {
 870                 if (ids[i] < port->mac_stats_nb_supported) {
 871                         strlcpy(xstats_names[i].name,
 872                                 efx_mac_stat_name(sa->nic,
 873                                                  port->mac_stats_by_id[ids[i]]),
 874                                 sizeof(xstats_names[0].name));
 875                 }
 876         }
 877
 878         nb_supported = port->mac_stats_nb_supported;
 879
 880         sfc_adapter_unlock(sa);
 881
 882         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 883                                             &nb_supported);
 884         if (ret != 0) {
 885                 SFC_ASSERT(ret < 0);
 886                 return ret;
 887         }
 888
 889         /* Return number of written names before invalid ID is encountered. */
 890         for (i = 0; i < size; i++) {
 891                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 892                         return i;
 893         }
 894
 895         return size;
 896 }
 897
 898 static int
 899 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 900 {
 901         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 902         unsigned int wanted_fc, link_fc;
 903
 904         memset(fc_conf, 0, sizeof(*fc_conf));
 905
 906         sfc_adapter_lock(sa);
 907
 908         if (sa->state == SFC_ADAPTER_STARTED)
 909                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 910         else
 911                 link_fc = sa->port.flow_ctrl;
 912
 913         switch (link_fc) {
 914         case 0:
 915                 fc_conf->mode = RTE_FC_NONE;
 916                 break;
 917         case EFX_FCNTL_RESPOND:
 918                 fc_conf->mode = RTE_FC_RX_PAUSE;
 919                 break;
 920         case EFX_FCNTL_GENERATE:
 921                 fc_conf->mode = RTE_FC_TX_PAUSE;
 922                 break;
 923         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 924                 fc_conf->mode = RTE_FC_FULL;
 925                 break;
 926         default:
 927                 sfc_err(sa, "%s: unexpected flow control value %#x",
 928                         __func__, link_fc);
 929         }
 930
 931         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 932
 933         sfc_adapter_unlock(sa);
 934
 935         return 0;
 936 }
 937
 938 static int
 939 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 940 {
 941         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 942         struct sfc_port *port = &sa->port;
 943         unsigned int fcntl;
 944         int rc;
 945
 946         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 947             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 948             fc_conf->mac_ctrl_frame_fwd != 0) {
 949                 sfc_err(sa, "unsupported flow control settings specified");
 950                 rc = EINVAL;
 951                 goto fail_inval;
 952         }
 953
 954         switch (fc_conf->mode) {
 955         case RTE_FC_NONE:
 956                 fcntl = 0;
 957                 break;
 958         case RTE_FC_RX_PAUSE:
 959                 fcntl = EFX_FCNTL_RESPOND;
 960                 break;
 961         case RTE_FC_TX_PAUSE:
 962                 fcntl = EFX_FCNTL_GENERATE;
 963                 break;
 964         case RTE_FC_FULL:
 965                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 966                 break;
 967         default:
 968                 rc = EINVAL;
 969                 goto fail_inval;
 970         }
 971
 972         sfc_adapter_lock(sa);
 973
 974         if (sa->state == SFC_ADAPTER_STARTED) {
 975                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 976                 if (rc != 0)
 977                         goto fail_mac_fcntl_set;
 978         }
 979
 980         port->flow_ctrl = fcntl;
 981         port->flow_ctrl_autoneg = fc_conf->autoneg;
 982
 983         sfc_adapter_unlock(sa);
 984
 985         return 0;
 986
 987 fail_mac_fcntl_set:
 988         sfc_adapter_unlock(sa);
 989 fail_inval:
 990         SFC_ASSERT(rc > 0);
 991         return -rc;
 992 }
 993
 994 static int
 995 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 996 {
 997         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 998         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 999         boolean_t scatter_enabled;
1000         const char *error;
1001         unsigned int i;
1002
1003         for (i = 0; i < sas->rxq_count; i++) {
1004                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1005                         continue;
1006
1007                 scatter_enabled = (sas->rxq_info[i].type_flags &
1008                                    EFX_RXQ_FLAG_SCATTER);
1009
1010                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1011                                           encp->enc_rx_prefix_size,
1012                                           scatter_enabled,
1013                                           encp->enc_rx_scatter_max, &error)) {
1014                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1015                                 error);
1016                         return EINVAL;
1017                 }
1018         }
1019
1020         return 0;
1021 }
1022
1023 static int
1024 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1025 {
1026         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1027         size_t pdu = EFX_MAC_PDU(mtu);
1028         size_t old_pdu;
1029         int rc;
1030
1031         sfc_log_init(sa, "mtu=%u", mtu);
1032
1033         rc = EINVAL;
1034         if (pdu < EFX_MAC_PDU_MIN) {
1035                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1036                         (unsigned int)mtu, (unsigned int)pdu,
1037                         EFX_MAC_PDU_MIN);
1038                 goto fail_inval;
1039         }
1040         if (pdu > EFX_MAC_PDU_MAX) {
1041                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1042                         (unsigned int)mtu, (unsigned int)pdu,
1043                         (unsigned int)EFX_MAC_PDU_MAX);
1044                 goto fail_inval;
1045         }
1046
1047         sfc_adapter_lock(sa);
1048
1049         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1050         if (rc != 0)
1051                 goto fail_check_scatter;
1052
1053         if (pdu != sa->port.pdu) {
1054                 if (sa->state == SFC_ADAPTER_STARTED) {
1055                         sfc_stop(sa);
1056
1057                         old_pdu = sa->port.pdu;
1058                         sa->port.pdu = pdu;
1059                         rc = sfc_start(sa);
1060                         if (rc != 0)
1061                                 goto fail_start;
1062                 } else {
1063                         sa->port.pdu = pdu;
1064                 }
1065         }
1066
1067         /*
1068          * The driver does not use it, but other PMDs update jumbo frame
1069          * flag and max_rx_pkt_len when MTU is set.
1070          */
1071         if (mtu > RTE_ETHER_MTU) {
1072                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1073                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1074         }
1075
1076         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1077
1078         sfc_adapter_unlock(sa);
1079
1080         sfc_log_init(sa, "done");
1081         return 0;
1082
1083 fail_start:
1084         sa->port.pdu = old_pdu;
1085         if (sfc_start(sa) != 0)
1086                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1087                         "PDU max size - port is stopped",
1088                         (unsigned int)pdu, (unsigned int)old_pdu);
1089
1090 fail_check_scatter:
1091         sfc_adapter_unlock(sa);
1092
1093 fail_inval:
1094         sfc_log_init(sa, "failed %d", rc);
1095         SFC_ASSERT(rc > 0);
1096         return -rc;
1097 }
1098 static int
1099 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1100 {
1101         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1102         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1103         struct sfc_port *port = &sa->port;
1104         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1105         int rc = 0;
1106
1107         sfc_adapter_lock(sa);
1108
1109         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1110                 goto unlock;
1111
1112         /*
1113          * Copy the address to the device private data so that
1114          * it could be recalled in the case of adapter restart.
1115          */
1116         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1117
1118         /*
1119          * Neither of the two following checks can return
1120          * an error. The new MAC address is preserved in
1121          * the device private data and can be activated
1122          * on the next port start if the user prevents
1123          * isolated mode from being enabled.
1124          */
1125         if (sfc_sa2shared(sa)->isolated) {
1126                 sfc_warn(sa, "isolated mode is active on the port");
1127                 sfc_warn(sa, "will not set MAC address");
1128                 goto unlock;
1129         }
1130
1131         if (sa->state != SFC_ADAPTER_STARTED) {
1132                 sfc_notice(sa, "the port is not started");
1133                 sfc_notice(sa, "the new MAC address will be set on port start");
1134
1135                 goto unlock;
1136         }
1137
1138         if (encp->enc_allow_set_mac_with_installed_filters) {
1139                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1140                 if (rc != 0) {
1141                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1142                         goto unlock;
1143                 }
1144
1145                 /*
1146                  * Changing the MAC address by means of MCDI request
1147                  * has no effect on received traffic, therefore
1148                  * we also need to update unicast filters
1149                  */
1150                 rc = sfc_set_rx_mode_unchecked(sa);
1151                 if (rc != 0) {
1152                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1153                         /* Rollback the old address */
1154                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1155                         (void)sfc_set_rx_mode_unchecked(sa);
1156                 }
1157         } else {
1158                 sfc_warn(sa, "cannot set MAC address with filters installed");
1159                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1160                 sfc_warn(sa, "(some traffic may be dropped)");
1161
1162                 /*
1163                  * Since setting MAC address with filters installed is not
1164                  * allowed on the adapter, the new MAC address will be set
1165                  * by means of adapter restart. sfc_start() shall retrieve
1166                  * the new address from the device private data and set it.
1167                  */
1168                 sfc_stop(sa);
1169                 rc = sfc_start(sa);
1170                 if (rc != 0)
1171                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1172         }
1173
1174 unlock:
1175         if (rc != 0)
1176                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1177
1178         sfc_adapter_unlock(sa);
1179
1180         SFC_ASSERT(rc >= 0);
1181         return -rc;
1182 }
1183
1184
1185 static int
1186 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1187                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1188 {
1189         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1190         struct sfc_port *port = &sa->port;
1191         uint8_t *mc_addrs = port->mcast_addrs;
1192         int rc;
1193         unsigned int i;
1194
1195         if (sfc_sa2shared(sa)->isolated) {
1196                 sfc_err(sa, "isolated mode is active on the port");
1197                 sfc_err(sa, "will not set multicast address list");
1198                 return -ENOTSUP;
1199         }
1200
1201         if (mc_addrs == NULL)
1202                 return -ENOBUFS;
1203
1204         if (nb_mc_addr > port->max_mcast_addrs) {
1205                 sfc_err(sa, "too many multicast addresses: %u > %u",
1206                          nb_mc_addr, port->max_mcast_addrs);
1207                 return -EINVAL;
1208         }
1209
1210         for (i = 0; i < nb_mc_addr; ++i) {
1211                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1212                                  EFX_MAC_ADDR_LEN);
1213                 mc_addrs += EFX_MAC_ADDR_LEN;
1214         }
1215
1216         port->nb_mcast_addrs = nb_mc_addr;
1217
1218         if (sa->state != SFC_ADAPTER_STARTED)
1219                 return 0;
1220
1221         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1222                                         port->nb_mcast_addrs);
1223         if (rc != 0)
1224                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1225
1226         SFC_ASSERT(rc >= 0);
1227         return -rc;
1228 }
1229
1230 /*
1231  * The function is used by the secondary process as well. It must not
1232  * use any process-local pointers from the adapter data.
1233  */
1234 static void
1235 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1236                       struct rte_eth_rxq_info *qinfo)
1237 {
1238         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1239         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1240         struct sfc_rxq_info *rxq_info;
1241
1242         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1243
1244         qinfo->mp = rxq_info->refill_mb_pool;
1245         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1246         qinfo->conf.rx_drop_en = 1;
1247         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1248         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1249         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1250                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1251                 qinfo->scattered_rx = 1;
1252         }
1253         qinfo->nb_desc = rxq_info->entries;
1254 }
1255
1256 /*
1257  * The function is used by the secondary process as well. It must not
1258  * use any process-local pointers from the adapter data.
1259  */
1260 static void
1261 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1262                       struct rte_eth_txq_info *qinfo)
1263 {
1264         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1265         struct sfc_txq_info *txq_info;
1266
1267         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1268
1269         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1270
1271         memset(qinfo, 0, sizeof(*qinfo));
1272
1273         qinfo->conf.offloads = txq_info->offloads;
1274         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1275         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1276         qinfo->nb_desc = txq_info->entries;
1277 }
1278
1279 /*
1280  * The function is used by the secondary process as well. It must not
1281  * use any process-local pointers from the adapter data.
1282  */
1283 static uint32_t
1284 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1285 {
1286         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1287         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1288         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1289         struct sfc_rxq_info *rxq_info;
1290
1291         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1292
1293         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1294                 return 0;
1295
1296         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1297 }
1298
1299 /*
1300  * The function is used by the secondary process as well. It must not
1301  * use any process-local pointers from the adapter data.
1302  */
1303 static int
1304 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1305 {
1306         struct sfc_dp_rxq *dp_rxq = queue;
1307         const struct sfc_dp_rx *dp_rx;
1308
1309         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1310
1311         return offset < dp_rx->qdesc_npending(dp_rxq);
1312 }
1313
1314 /*
1315  * The function is used by the secondary process as well. It must not
1316  * use any process-local pointers from the adapter data.
1317  */
1318 static int
1319 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1320 {
1321         struct sfc_dp_rxq *dp_rxq = queue;
1322         const struct sfc_dp_rx *dp_rx;
1323
1324         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1325
1326         return dp_rx->qdesc_status(dp_rxq, offset);
1327 }
1328
1329 /*
1330  * The function is used by the secondary process as well. It must not
1331  * use any process-local pointers from the adapter data.
1332  */
1333 static int
1334 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1335 {
1336         struct sfc_dp_txq *dp_txq = queue;
1337         const struct sfc_dp_tx *dp_tx;
1338
1339         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1340
1341         return dp_tx->qdesc_status(dp_txq, offset);
1342 }
1343
1344 static int
1345 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1346 {
1347         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1348         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1349         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1350         struct sfc_rxq_info *rxq_info;
1351         sfc_sw_index_t sw_index;
1352         int rc;
1353
1354         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1355
1356         sfc_adapter_lock(sa);
1357
1358         rc = EINVAL;
1359         if (sa->state != SFC_ADAPTER_STARTED)
1360                 goto fail_not_started;
1361
1362         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1363         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1364                 goto fail_not_setup;
1365
1366         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1367         rc = sfc_rx_qstart(sa, sw_index);
1368         if (rc != 0)
1369                 goto fail_rx_qstart;
1370
1371         rxq_info->deferred_started = B_TRUE;
1372
1373         sfc_adapter_unlock(sa);
1374
1375         return 0;
1376
1377 fail_rx_qstart:
1378 fail_not_setup:
1379 fail_not_started:
1380         sfc_adapter_unlock(sa);
1381         SFC_ASSERT(rc > 0);
1382         return -rc;
1383 }
1384
1385 static int
1386 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1387 {
1388         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1389         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1390         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1391         struct sfc_rxq_info *rxq_info;
1392         sfc_sw_index_t sw_index;
1393
1394         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1395
1396         sfc_adapter_lock(sa);
1397
1398         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1399         sfc_rx_qstop(sa, sw_index);
1400
1401         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1402         rxq_info->deferred_started = B_FALSE;
1403
1404         sfc_adapter_unlock(sa);
1405
1406         return 0;
1407 }
1408
1409 static int
1410 sfc_tx_queue_start(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         struct sfc_txq_info *txq_info;
1415         sfc_sw_index_t sw_index;
1416         int rc;
1417
1418         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1419
1420         sfc_adapter_lock(sa);
1421
1422         rc = EINVAL;
1423         if (sa->state != SFC_ADAPTER_STARTED)
1424                 goto fail_not_started;
1425
1426         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1427         if (txq_info->state != SFC_TXQ_INITIALIZED)
1428                 goto fail_not_setup;
1429
1430         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1431         rc = sfc_tx_qstart(sa, sw_index);
1432         if (rc != 0)
1433                 goto fail_tx_qstart;
1434
1435         txq_info->deferred_started = B_TRUE;
1436
1437         sfc_adapter_unlock(sa);
1438         return 0;
1439
1440 fail_tx_qstart:
1441
1442 fail_not_setup:
1443 fail_not_started:
1444         sfc_adapter_unlock(sa);
1445         SFC_ASSERT(rc > 0);
1446         return -rc;
1447 }
1448
1449 static int
1450 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1451 {
1452         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1453         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1454         struct sfc_txq_info *txq_info;
1455         sfc_sw_index_t sw_index;
1456
1457         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1458
1459         sfc_adapter_lock(sa);
1460
1461         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1462         sfc_tx_qstop(sa, sw_index);
1463
1464         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1465         txq_info->deferred_started = B_FALSE;
1466
1467         sfc_adapter_unlock(sa);
1468         return 0;
1469 }
1470
1471 static efx_tunnel_protocol_t
1472 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1473 {
1474         switch (rte_type) {
1475         case RTE_TUNNEL_TYPE_VXLAN:
1476                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1477         case RTE_TUNNEL_TYPE_GENEVE:
1478                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1479         default:
1480                 return EFX_TUNNEL_NPROTOS;
1481         }
1482 }
1483
1484 enum sfc_udp_tunnel_op_e {
1485         SFC_UDP_TUNNEL_ADD_PORT,
1486         SFC_UDP_TUNNEL_DEL_PORT,
1487 };
1488
1489 static int
1490 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1491                       struct rte_eth_udp_tunnel *tunnel_udp,
1492                       enum sfc_udp_tunnel_op_e op)
1493 {
1494         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1495         efx_tunnel_protocol_t tunnel_proto;
1496         int rc;
1497
1498         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1499                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1500                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1501                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1502
1503         tunnel_proto =
1504                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1505         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1506                 rc = ENOTSUP;
1507                 goto fail_bad_proto;
1508         }
1509
1510         sfc_adapter_lock(sa);
1511
1512         switch (op) {
1513         case SFC_UDP_TUNNEL_ADD_PORT:
1514                 rc = efx_tunnel_config_udp_add(sa->nic,
1515                                                tunnel_udp->udp_port,
1516                                                tunnel_proto);
1517                 break;
1518         case SFC_UDP_TUNNEL_DEL_PORT:
1519                 rc = efx_tunnel_config_udp_remove(sa->nic,
1520                                                   tunnel_udp->udp_port,
1521                                                   tunnel_proto);
1522                 break;
1523         default:
1524                 rc = EINVAL;
1525                 goto fail_bad_op;
1526         }
1527
1528         if (rc != 0)
1529                 goto fail_op;
1530
1531         if (sa->state == SFC_ADAPTER_STARTED) {
1532                 rc = efx_tunnel_reconfigure(sa->nic);
1533                 if (rc == EAGAIN) {
1534                         /*
1535                          * Configuration is accepted by FW and MC reboot
1536                          * is initiated to apply the changes. MC reboot
1537                          * will be handled in a usual way (MC reboot
1538                          * event on management event queue and adapter
1539                          * restart).
1540                          */
1541                         rc = 0;
1542                 } else if (rc != 0) {
1543                         goto fail_reconfigure;
1544                 }
1545         }
1546
1547         sfc_adapter_unlock(sa);
1548         return 0;
1549
1550 fail_reconfigure:
1551         /* Remove/restore entry since the change makes the trouble */
1552         switch (op) {
1553         case SFC_UDP_TUNNEL_ADD_PORT:
1554                 (void)efx_tunnel_config_udp_remove(sa->nic,
1555                                                    tunnel_udp->udp_port,
1556                                                    tunnel_proto);
1557                 break;
1558         case SFC_UDP_TUNNEL_DEL_PORT:
1559                 (void)efx_tunnel_config_udp_add(sa->nic,
1560                                                 tunnel_udp->udp_port,
1561                                                 tunnel_proto);
1562                 break;
1563         }
1564
1565 fail_op:
1566 fail_bad_op:
1567         sfc_adapter_unlock(sa);
1568
1569 fail_bad_proto:
1570         SFC_ASSERT(rc > 0);
1571         return -rc;
1572 }
1573
1574 static int
1575 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1576                             struct rte_eth_udp_tunnel *tunnel_udp)
1577 {
1578         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1579 }
1580
1581 static int
1582 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1583                             struct rte_eth_udp_tunnel *tunnel_udp)
1584 {
1585         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1586 }
1587
1588 /*
1589  * The function is used by the secondary process as well. It must not
1590  * use any process-local pointers from the adapter data.
1591  */
1592 static int
1593 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1594                           struct rte_eth_rss_conf *rss_conf)
1595 {
1596         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1597         struct sfc_rss *rss = &sas->rss;
1598
1599         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1600                 return -ENOTSUP;
1601
1602         /*
1603          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1604          * hence, conversion is done here to derive a correct set of ETH_RSS
1605          * flags which corresponds to the active EFX configuration stored
1606          * locally in 'sfc_adapter' and kept up-to-date
1607          */
1608         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1609         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1610         if (rss_conf->rss_key != NULL)
1611                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1612
1613         return 0;
1614 }
1615
1616 static int
1617 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1618                         struct rte_eth_rss_conf *rss_conf)
1619 {
1620         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1621         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1622         unsigned int efx_hash_types;
1623         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1624         unsigned int n_contexts;
1625         unsigned int mode_i = 0;
1626         unsigned int key_i = 0;
1627         unsigned int i = 0;
1628         int rc = 0;
1629
1630         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1631
1632         if (sfc_sa2shared(sa)->isolated)
1633                 return -ENOTSUP;
1634
1635         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1636                 sfc_err(sa, "RSS is not available");
1637                 return -ENOTSUP;
1638         }
1639
1640         if (rss->channels == 0) {
1641                 sfc_err(sa, "RSS is not configured");
1642                 return -EINVAL;
1643         }
1644
1645         if ((rss_conf->rss_key != NULL) &&
1646             (rss_conf->rss_key_len != sizeof(rss->key))) {
1647                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1648                         sizeof(rss->key));
1649                 return -EINVAL;
1650         }
1651
1652         sfc_adapter_lock(sa);
1653
1654         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1655         if (rc != 0)
1656                 goto fail_rx_hf_rte_to_efx;
1657
1658         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1659                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1660                                            rss->hash_alg, efx_hash_types,
1661                                            B_TRUE);
1662                 if (rc != 0)
1663                         goto fail_scale_mode_set;
1664         }
1665
1666         if (rss_conf->rss_key != NULL) {
1667                 if (sa->state == SFC_ADAPTER_STARTED) {
1668                         for (key_i = 0; key_i < n_contexts; key_i++) {
1669                                 rc = efx_rx_scale_key_set(sa->nic,
1670                                                           contexts[key_i],
1671                                                           rss_conf->rss_key,
1672                                                           sizeof(rss->key));
1673                                 if (rc != 0)
1674                                         goto fail_scale_key_set;
1675                         }
1676                 }
1677
1678                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1679         }
1680
1681         rss->hash_types = efx_hash_types;
1682
1683         sfc_adapter_unlock(sa);
1684
1685         return 0;
1686
1687 fail_scale_key_set:
1688         for (i = 0; i < key_i; i++) {
1689                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1690                                          sizeof(rss->key)) != 0)
1691                         sfc_err(sa, "failed to restore RSS key");
1692         }
1693
1694 fail_scale_mode_set:
1695         for (i = 0; i < mode_i; i++) {
1696                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1697                                           EFX_RX_HASHALG_TOEPLITZ,
1698                                           rss->hash_types, B_TRUE) != 0)
1699                         sfc_err(sa, "failed to restore RSS mode");
1700         }
1701
1702 fail_rx_hf_rte_to_efx:
1703         sfc_adapter_unlock(sa);
1704         return -rc;
1705 }
1706
1707 /*
1708  * The function is used by the secondary process as well. It must not
1709  * use any process-local pointers from the adapter data.
1710  */
1711 static int
1712 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1713                        struct rte_eth_rss_reta_entry64 *reta_conf,
1714                        uint16_t reta_size)
1715 {
1716         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1717         struct sfc_rss *rss = &sas->rss;
1718         int entry;
1719
1720         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1721                 return -ENOTSUP;
1722
1723         if (rss->channels == 0)
1724                 return -EINVAL;
1725
1726         if (reta_size != EFX_RSS_TBL_SIZE)
1727                 return -EINVAL;
1728
1729         for (entry = 0; entry < reta_size; entry++) {
1730                 int grp = entry / RTE_RETA_GROUP_SIZE;
1731                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1732
1733                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1734                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1735         }
1736
1737         return 0;
1738 }
1739
1740 static int
1741 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1742                         struct rte_eth_rss_reta_entry64 *reta_conf,
1743                         uint16_t reta_size)
1744 {
1745         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1746         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1747         unsigned int *rss_tbl_new;
1748         uint16_t entry;
1749         int rc = 0;
1750
1751
1752         if (sfc_sa2shared(sa)->isolated)
1753                 return -ENOTSUP;
1754
1755         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1756                 sfc_err(sa, "RSS is not available");
1757                 return -ENOTSUP;
1758         }
1759
1760         if (rss->channels == 0) {
1761                 sfc_err(sa, "RSS is not configured");
1762                 return -EINVAL;
1763         }
1764
1765         if (reta_size != EFX_RSS_TBL_SIZE) {
1766                 sfc_err(sa, "RETA size is wrong (should be %u)",
1767                         EFX_RSS_TBL_SIZE);
1768                 return -EINVAL;
1769         }
1770
1771         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1772         if (rss_tbl_new == NULL)
1773                 return -ENOMEM;
1774
1775         sfc_adapter_lock(sa);
1776
1777         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1778
1779         for (entry = 0; entry < reta_size; entry++) {
1780                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1781                 struct rte_eth_rss_reta_entry64 *grp;
1782
1783                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1784
1785                 if (grp->mask & (1ull << grp_idx)) {
1786                         if (grp->reta[grp_idx] >= rss->channels) {
1787                                 rc = EINVAL;
1788                                 goto bad_reta_entry;
1789                         }
1790                         rss_tbl_new[entry] = grp->reta[grp_idx];
1791                 }
1792         }
1793
1794         if (sa->state == SFC_ADAPTER_STARTED) {
1795                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1796                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1797                 if (rc != 0)
1798                         goto fail_scale_tbl_set;
1799         }
1800
1801         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1802
1803 fail_scale_tbl_set:
1804 bad_reta_entry:
1805         sfc_adapter_unlock(sa);
1806
1807         rte_free(rss_tbl_new);
1808
1809         SFC_ASSERT(rc >= 0);
1810         return -rc;
1811 }
1812
1813 static int
1814 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1815                      const struct rte_flow_ops **ops)
1816 {
1817         *ops = &sfc_flow_ops;
1818         return 0;
1819 }
1820
1821 static int
1822 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1823 {
1824         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1825
1826         /*
1827          * If Rx datapath does not provide callback to check mempool,
1828          * all pools are supported.
1829          */
1830         if (sap->dp_rx->pool_ops_supported == NULL)
1831                 return 1;
1832
1833         return sap->dp_rx->pool_ops_supported(pool);
1834 }
1835
1836 static int
1837 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1838 {
1839         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1840         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1841         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1842         struct sfc_rxq_info *rxq_info;
1843
1844         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1845
1846         return sap->dp_rx->intr_enable(rxq_info->dp);
1847 }
1848
1849 static int
1850 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1851 {
1852         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1853         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1854         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1855         struct sfc_rxq_info *rxq_info;
1856
1857         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1858
1859         return sap->dp_rx->intr_disable(rxq_info->dp);
1860 }
1861
1862 static const struct eth_dev_ops sfc_eth_dev_ops = {
1863         .dev_configure                  = sfc_dev_configure,
1864         .dev_start                      = sfc_dev_start,
1865         .dev_stop                       = sfc_dev_stop,
1866         .dev_set_link_up                = sfc_dev_set_link_up,
1867         .dev_set_link_down              = sfc_dev_set_link_down,
1868         .dev_close                      = sfc_dev_close,
1869         .promiscuous_enable             = sfc_dev_promisc_enable,
1870         .promiscuous_disable            = sfc_dev_promisc_disable,
1871         .allmulticast_enable            = sfc_dev_allmulti_enable,
1872         .allmulticast_disable           = sfc_dev_allmulti_disable,
1873         .link_update                    = sfc_dev_link_update,
1874         .stats_get                      = sfc_stats_get,
1875         .stats_reset                    = sfc_stats_reset,
1876         .xstats_get                     = sfc_xstats_get,
1877         .xstats_reset                   = sfc_stats_reset,
1878         .xstats_get_names               = sfc_xstats_get_names,
1879         .dev_infos_get                  = sfc_dev_infos_get,
1880         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1881         .mtu_set                        = sfc_dev_set_mtu,
1882         .rx_queue_start                 = sfc_rx_queue_start,
1883         .rx_queue_stop                  = sfc_rx_queue_stop,
1884         .tx_queue_start                 = sfc_tx_queue_start,
1885         .tx_queue_stop                  = sfc_tx_queue_stop,
1886         .rx_queue_setup                 = sfc_rx_queue_setup,
1887         .rx_queue_release               = sfc_rx_queue_release,
1888         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1889         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1890         .tx_queue_setup                 = sfc_tx_queue_setup,
1891         .tx_queue_release               = sfc_tx_queue_release,
1892         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1893         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1894         .mac_addr_set                   = sfc_mac_addr_set,
1895         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1896         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1897         .reta_update                    = sfc_dev_rss_reta_update,
1898         .reta_query                     = sfc_dev_rss_reta_query,
1899         .rss_hash_update                = sfc_dev_rss_hash_update,
1900         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1901         .flow_ops_get                   = sfc_dev_flow_ops_get,
1902         .set_mc_addr_list               = sfc_set_mc_addr_list,
1903         .rxq_info_get                   = sfc_rx_queue_info_get,
1904         .txq_info_get                   = sfc_tx_queue_info_get,
1905         .fw_version_get                 = sfc_fw_version_get,
1906         .xstats_get_by_id               = sfc_xstats_get_by_id,
1907         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1908         .pool_ops_supported             = sfc_pool_ops_supported,
1909 };
1910
1911 /**
1912  * Duplicate a string in potentially shared memory required for
1913  * multi-process support.
1914  *
1915  * strdup() allocates from process-local heap/memory.
1916  */
1917 static char *
1918 sfc_strdup(const char *str)
1919 {
1920         size_t size;
1921         char *copy;
1922
1923         if (str == NULL)
1924                 return NULL;
1925
1926         size = strlen(str) + 1;
1927         copy = rte_malloc(__func__, size, 0);
1928         if (copy != NULL)
1929                 rte_memcpy(copy, str, size);
1930
1931         return copy;
1932 }
1933
1934 static int
1935 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1936 {
1937         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1938         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1939         const struct sfc_dp_rx *dp_rx;
1940         const struct sfc_dp_tx *dp_tx;
1941         const efx_nic_cfg_t *encp;
1942         unsigned int avail_caps = 0;
1943         const char *rx_name = NULL;
1944         const char *tx_name = NULL;
1945         int rc;
1946
1947         switch (sa->family) {
1948         case EFX_FAMILY_HUNTINGTON:
1949         case EFX_FAMILY_MEDFORD:
1950         case EFX_FAMILY_MEDFORD2:
1951                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1952                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1953                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1954                 break;
1955         case EFX_FAMILY_RIVERHEAD:
1956                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1957                 break;
1958         default:
1959                 break;
1960         }
1961
1962         encp = efx_nic_cfg_get(sa->nic);
1963         if (encp->enc_rx_es_super_buffer_supported)
1964                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1965
1966         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1967                                 sfc_kvarg_string_handler, &rx_name);
1968         if (rc != 0)
1969                 goto fail_kvarg_rx_datapath;
1970
1971         if (rx_name != NULL) {
1972                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1973                 if (dp_rx == NULL) {
1974                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1975                         rc = ENOENT;
1976                         goto fail_dp_rx;
1977                 }
1978                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1979                         sfc_err(sa,
1980                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1981                                 rx_name);
1982                         rc = EINVAL;
1983                         goto fail_dp_rx_caps;
1984                 }
1985         } else {
1986                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1987                 if (dp_rx == NULL) {
1988                         sfc_err(sa, "Rx datapath by caps %#x not found",
1989                                 avail_caps);
1990                         rc = ENOENT;
1991                         goto fail_dp_rx;
1992                 }
1993         }
1994
1995         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1996         if (sas->dp_rx_name == NULL) {
1997                 rc = ENOMEM;
1998                 goto fail_dp_rx_name;
1999         }
2000
2001         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2002
2003         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2004                                 sfc_kvarg_string_handler, &tx_name);
2005         if (rc != 0)
2006                 goto fail_kvarg_tx_datapath;
2007
2008         if (tx_name != NULL) {
2009                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2010                 if (dp_tx == NULL) {
2011                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2012                         rc = ENOENT;
2013                         goto fail_dp_tx;
2014                 }
2015                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2016                         sfc_err(sa,
2017                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2018                                 tx_name);
2019                         rc = EINVAL;
2020                         goto fail_dp_tx_caps;
2021                 }
2022         } else {
2023                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2024                 if (dp_tx == NULL) {
2025                         sfc_err(sa, "Tx datapath by caps %#x not found",
2026                                 avail_caps);
2027                         rc = ENOENT;
2028                         goto fail_dp_tx;
2029                 }
2030         }
2031
2032         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2033         if (sas->dp_tx_name == NULL) {
2034                 rc = ENOMEM;
2035                 goto fail_dp_tx_name;
2036         }
2037
2038         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2039
2040         sa->priv.dp_rx = dp_rx;
2041         sa->priv.dp_tx = dp_tx;
2042
2043         dev->rx_pkt_burst = dp_rx->pkt_burst;
2044         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2045         dev->tx_pkt_burst = dp_tx->pkt_burst;
2046
2047         dev->rx_queue_count = sfc_rx_queue_count;
2048         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2049         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2050         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2051         dev->dev_ops = &sfc_eth_dev_ops;
2052
2053         return 0;
2054
2055 fail_dp_tx_name:
2056 fail_dp_tx_caps:
2057 fail_dp_tx:
2058 fail_kvarg_tx_datapath:
2059         rte_free(sas->dp_rx_name);
2060         sas->dp_rx_name = NULL;
2061
2062 fail_dp_rx_name:
2063 fail_dp_rx_caps:
2064 fail_dp_rx:
2065 fail_kvarg_rx_datapath:
2066         return rc;
2067 }
2068
2069 static void
2070 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2071 {
2072         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2073         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2074
2075         dev->dev_ops = NULL;
2076         dev->tx_pkt_prepare = NULL;
2077         dev->rx_pkt_burst = NULL;
2078         dev->tx_pkt_burst = NULL;
2079
2080         rte_free(sas->dp_tx_name);
2081         sas->dp_tx_name = NULL;
2082         sa->priv.dp_tx = NULL;
2083
2084         rte_free(sas->dp_rx_name);
2085         sas->dp_rx_name = NULL;
2086         sa->priv.dp_rx = NULL;
2087 }
2088
2089 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2090         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2091         .reta_query                     = sfc_dev_rss_reta_query,
2092         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2093         .rxq_info_get                   = sfc_rx_queue_info_get,
2094         .txq_info_get                   = sfc_tx_queue_info_get,
2095 };
2096
2097 static int
2098 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2099 {
2100         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2101         struct sfc_adapter_priv *sap;
2102         const struct sfc_dp_rx *dp_rx;
2103         const struct sfc_dp_tx *dp_tx;
2104         int rc;
2105
2106         /*
2107          * Allocate process private data from heap, since it should not
2108          * be located in shared memory allocated using rte_malloc() API.
2109          */
2110         sap = calloc(1, sizeof(*sap));
2111         if (sap == NULL) {
2112                 rc = ENOMEM;
2113                 goto fail_alloc_priv;
2114         }
2115
2116         sap->logtype_main = logtype_main;
2117
2118         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2119         if (dp_rx == NULL) {
2120                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2121                         "cannot find %s Rx datapath", sas->dp_rx_name);
2122                 rc = ENOENT;
2123                 goto fail_dp_rx;
2124         }
2125         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2126                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2127                         "%s Rx datapath does not support multi-process",
2128                         sas->dp_rx_name);
2129                 rc = EINVAL;
2130                 goto fail_dp_rx_multi_process;
2131         }
2132
2133         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2134         if (dp_tx == NULL) {
2135                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2136                         "cannot find %s Tx datapath", sas->dp_tx_name);
2137                 rc = ENOENT;
2138                 goto fail_dp_tx;
2139         }
2140         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2141                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2142                         "%s Tx datapath does not support multi-process",
2143                         sas->dp_tx_name);
2144                 rc = EINVAL;
2145                 goto fail_dp_tx_multi_process;
2146         }
2147
2148         sap->dp_rx = dp_rx;
2149         sap->dp_tx = dp_tx;
2150
2151         dev->process_private = sap;
2152         dev->rx_pkt_burst = dp_rx->pkt_burst;
2153         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2154         dev->tx_pkt_burst = dp_tx->pkt_burst;
2155         dev->rx_queue_count = sfc_rx_queue_count;
2156         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2157         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2158         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2159         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2160
2161         return 0;
2162
2163 fail_dp_tx_multi_process:
2164 fail_dp_tx:
2165 fail_dp_rx_multi_process:
2166 fail_dp_rx:
2167         free(sap);
2168
2169 fail_alloc_priv:
2170         return rc;
2171 }
2172
2173 static void
2174 sfc_register_dp(void)
2175 {
2176         /* Register once */
2177         if (TAILQ_EMPTY(&sfc_dp_head)) {
2178                 /* Prefer EF10 datapath */
2179                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2180                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2181                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2182                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2183
2184                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2185                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2186                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2187                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2188         }
2189 }
2190
2191 static int
2192 sfc_eth_dev_init(struct rte_eth_dev *dev)
2193 {
2194         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2195         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2196         uint32_t logtype_main;
2197         struct sfc_adapter *sa;
2198         int rc;
2199         const efx_nic_cfg_t *encp;
2200         const struct rte_ether_addr *from;
2201         int ret;
2202
2203         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2204                         SFC_EFX_DEV_CLASS_NET) {
2205                 SFC_GENERIC_LOG(DEBUG,
2206                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2207                 return 1;
2208         }
2209
2210         sfc_register_dp();
2211
2212         logtype_main = sfc_register_logtype(&pci_dev->addr,
2213                                             SFC_LOGTYPE_MAIN_STR,
2214                                             RTE_LOG_NOTICE);
2215
2216         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2217                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2218
2219         /* Required for logging */
2220         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2221                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2222                         pci_dev->addr.domain, pci_dev->addr.bus,
2223                         pci_dev->addr.devid, pci_dev->addr.function,
2224                         dev->data->port_id);
2225         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2226                 SFC_GENERIC_LOG(ERR,
2227                         "reserved log prefix is too short for " PCI_PRI_FMT,
2228                         pci_dev->addr.domain, pci_dev->addr.bus,
2229                         pci_dev->addr.devid, pci_dev->addr.function);
2230                 return -EINVAL;
2231         }
2232         sas->pci_addr = pci_dev->addr;
2233         sas->port_id = dev->data->port_id;
2234
2235         /*
2236          * Allocate process private data from heap, since it should not
2237          * be located in shared memory allocated using rte_malloc() API.
2238          */
2239         sa = calloc(1, sizeof(*sa));
2240         if (sa == NULL) {
2241                 rc = ENOMEM;
2242                 goto fail_alloc_sa;
2243         }
2244
2245         dev->process_private = sa;
2246
2247         /* Required for logging */
2248         sa->priv.shared = sas;
2249         sa->priv.logtype_main = logtype_main;
2250
2251         sa->eth_dev = dev;
2252
2253         /* Copy PCI device info to the dev->data */
2254         rte_eth_copy_pci_info(dev, pci_dev);
2255         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2256         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2257
2258         rc = sfc_kvargs_parse(sa);
2259         if (rc != 0)
2260                 goto fail_kvargs_parse;
2261
2262         sfc_log_init(sa, "entry");
2263
2264         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2265         if (dev->data->mac_addrs == NULL) {
2266                 rc = ENOMEM;
2267                 goto fail_mac_addrs;
2268         }
2269
2270         sfc_adapter_lock_init(sa);
2271         sfc_adapter_lock(sa);
2272
2273         sfc_log_init(sa, "probing");
2274         rc = sfc_probe(sa);
2275         if (rc != 0)
2276                 goto fail_probe;
2277
2278         sfc_log_init(sa, "set device ops");
2279         rc = sfc_eth_dev_set_ops(dev);
2280         if (rc != 0)
2281                 goto fail_set_ops;
2282
2283         sfc_log_init(sa, "attaching");
2284         rc = sfc_attach(sa);
2285         if (rc != 0)
2286                 goto fail_attach;
2287
2288         encp = efx_nic_cfg_get(sa->nic);
2289
2290         /*
2291          * The arguments are really reverse order in comparison to
2292          * Linux kernel. Copy from NIC config to Ethernet device data.
2293          */
2294         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2295         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2296
2297         sfc_adapter_unlock(sa);
2298
2299         sfc_log_init(sa, "done");
2300         return 0;
2301
2302 fail_attach:
2303         sfc_eth_dev_clear_ops(dev);
2304
2305 fail_set_ops:
2306         sfc_unprobe(sa);
2307
2308 fail_probe:
2309         sfc_adapter_unlock(sa);
2310         sfc_adapter_lock_fini(sa);
2311         rte_free(dev->data->mac_addrs);
2312         dev->data->mac_addrs = NULL;
2313
2314 fail_mac_addrs:
2315         sfc_kvargs_cleanup(sa);
2316
2317 fail_kvargs_parse:
2318         sfc_log_init(sa, "failed %d", rc);
2319         dev->process_private = NULL;
2320         free(sa);
2321
2322 fail_alloc_sa:
2323         SFC_ASSERT(rc > 0);
2324         return -rc;
2325 }
2326
2327 static int
2328 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2329 {
2330         sfc_dev_close(dev);
2331
2332         return 0;
2333 }
2334
2335 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2336         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2337         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2338         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2339         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2340         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2341         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2342         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2343         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2344         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2345         { .vendor_id = 0 /* sentinel */ }
2346 };
2347
2348 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2349         struct rte_pci_device *pci_dev)
2350 {
2351         return rte_eth_dev_pci_generic_probe(pci_dev,
2352                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2353 }
2354
2355 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2356 {
2357         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2358 }
2359
2360 static struct rte_pci_driver sfc_efx_pmd = {
2361         .id_table = pci_id_sfc_efx_map,
2362         .drv_flags =
2363                 RTE_PCI_DRV_INTR_LSC |
2364                 RTE_PCI_DRV_NEED_MAPPING,
2365         .probe = sfc_eth_dev_pci_probe,
2366         .remove = sfc_eth_dev_pci_remove,
2367 };
2368
2369 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2370 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2371 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2372 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2373         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2374         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2375         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2376         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2377         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2378         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2379
2380 RTE_INIT(sfc_driver_register_logtype)
2381 {
2382         int ret;
2383
2384         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2385                                                    RTE_LOG_NOTICE);
2386         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2387 }