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_flow_tunnel.h"
  30 #include "sfc_dp.h"
  31 #include "sfc_dp_rx.h"
  32 #include "sfc_repr.h"
  33 #include "sfc_sw_stats.h"
  34 #include "sfc_switch.h"
  35
  36 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
  37 #define SFC_XSTAT_ID_INVALID_NAME '\0'
  38
  39 uint32_t sfc_logtype_driver;
  40
  41 static struct sfc_dp_list sfc_dp_head =
  42         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  43
  44
  45 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  46
  47
  48 static int
  49 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  50 {
  51         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  52         efx_nic_fw_info_t enfi;
  53         int ret;
  54         int rc;
  55
  56         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  57         if (rc != 0)
  58                 return -rc;
  59
  60         ret = snprintf(fw_version, fw_size,
  61                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  62                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  63                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  64         if (ret < 0)
  65                 return ret;
  66
  67         if (enfi.enfi_dpcpu_fw_ids_valid) {
  68                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  69                 int ret_extra;
  70
  71                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  72                                      fw_size - dpcpu_fw_ids_offset,
  73                                      " rx%" PRIx16 " tx%" PRIx16,
  74                                      enfi.enfi_rx_dpcpu_fw_id,
  75                                      enfi.enfi_tx_dpcpu_fw_id);
  76                 if (ret_extra < 0)
  77                         return ret_extra;
  78
  79                 ret += ret_extra;
  80         }
  81
  82         if (fw_size < (size_t)(++ret))
  83                 return ret;
  84         else
  85                 return 0;
  86 }
  87
  88 static int
  89 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  90 {
  91         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  92         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  93         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  94         struct sfc_rss *rss = &sas->rss;
  95         struct sfc_mae *mae = &sa->mae;
  96         uint64_t txq_offloads_def = 0;
  97
  98         sfc_log_init(sa, "entry");
  99
 100         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 101         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 102
 103         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 104
 105         dev_info->max_vfs = sa->sriov.num_vfs;
 106
 107         /* Autonegotiation may be disabled */
 108         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 110                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 112                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 114                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 115         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 116                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 117         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 118                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 119         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 120                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 121
 122         dev_info->max_rx_queues = sa->rxq_max;
 123         dev_info->max_tx_queues = sa->txq_max;
 124
 125         /* By default packets are dropped if no descriptors are available */
 126         dev_info->default_rxconf.rx_drop_en = 1;
 127
 128         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 129
 130         /*
 131          * rx_offload_capa includes both device and queue offloads since
 132          * the latter may be requested on a per device basis which makes
 133          * sense when some offloads are needed to be set on all queues.
 134          */
 135         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 136                                     dev_info->rx_queue_offload_capa;
 137
 138         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 139
 140         /*
 141          * tx_offload_capa includes both device and queue offloads since
 142          * the latter may be requested on a per device basis which makes
 143          * sense when some offloads are needed to be set on all queues.
 144          */
 145         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 146                                     dev_info->tx_queue_offload_capa;
 147
 148         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 149                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 150
 151         dev_info->default_txconf.offloads |= txq_offloads_def;
 152
 153         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 154                 uint64_t rte_hf = 0;
 155                 unsigned int i;
 156
 157                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 158                         rte_hf |= rss->hf_map[i].rte;
 159
 160                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 161                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 162                 dev_info->flow_type_rss_offloads = rte_hf;
 163         }
 164
 165         /* Initialize to hardware limits */
 166         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 167         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 168         /* The RXQ hardware requires that the descriptor count is a power
 169          * of 2, but rx_desc_lim cannot properly describe that constraint.
 170          */
 171         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 172
 173         /* Initialize to hardware limits */
 174         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 175         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 176         /*
 177          * The TXQ hardware requires that the descriptor count is a power
 178          * of 2, but tx_desc_lim cannot properly describe that constraint
 179          */
 180         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 181
 182         if (sap->dp_rx->get_dev_info != NULL)
 183                 sap->dp_rx->get_dev_info(dev_info);
 184         if (sap->dp_tx->get_dev_info != NULL)
 185                 sap->dp_tx->get_dev_info(dev_info);
 186
 187         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 188                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 189
 190         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
 191                 dev_info->switch_info.name = dev->device->driver->name;
 192                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 193                 dev_info->switch_info.port_id = mae->switch_port_id;
 194         }
 195
 196         return 0;
 197 }
 198
 199 static const uint32_t *
 200 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 201 {
 202         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 203
 204         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 205 }
 206
 207 static int
 208 sfc_dev_configure(struct rte_eth_dev *dev)
 209 {
 210         struct rte_eth_dev_data *dev_data = dev->data;
 211         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 212         int rc;
 213
 214         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 215                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 216
 217         sfc_adapter_lock(sa);
 218         switch (sa->state) {
 219         case SFC_ETHDEV_CONFIGURED:
 220                 /* FALLTHROUGH */
 221         case SFC_ETHDEV_INITIALIZED:
 222                 rc = sfc_configure(sa);
 223                 break;
 224         default:
 225                 sfc_err(sa, "unexpected adapter state %u to configure",
 226                         sa->state);
 227                 rc = EINVAL;
 228                 break;
 229         }
 230         sfc_adapter_unlock(sa);
 231
 232         sfc_log_init(sa, "done %d", rc);
 233         SFC_ASSERT(rc >= 0);
 234         return -rc;
 235 }
 236
 237 static int
 238 sfc_dev_start(struct rte_eth_dev *dev)
 239 {
 240         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 241         int rc;
 242
 243         sfc_log_init(sa, "entry");
 244
 245         sfc_adapter_lock(sa);
 246         rc = sfc_start(sa);
 247         sfc_adapter_unlock(sa);
 248
 249         sfc_log_init(sa, "done %d", rc);
 250         SFC_ASSERT(rc >= 0);
 251         return -rc;
 252 }
 253
 254 static int
 255 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 256 {
 257         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 258         struct rte_eth_link current_link;
 259         int ret;
 260
 261         sfc_log_init(sa, "entry");
 262
 263         if (sa->state != SFC_ETHDEV_STARTED) {
 264                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 265         } else if (wait_to_complete) {
 266                 efx_link_mode_t link_mode;
 267
 268                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 269                         link_mode = EFX_LINK_UNKNOWN;
 270                 sfc_port_link_mode_to_info(link_mode, &current_link);
 271
 272         } else {
 273                 sfc_ev_mgmt_qpoll(sa);
 274                 rte_eth_linkstatus_get(dev, &current_link);
 275         }
 276
 277         ret = rte_eth_linkstatus_set(dev, &current_link);
 278         if (ret == 0)
 279                 sfc_notice(sa, "Link status is %s",
 280                            current_link.link_status ? "UP" : "DOWN");
 281
 282         return ret;
 283 }
 284
 285 static int
 286 sfc_dev_stop(struct rte_eth_dev *dev)
 287 {
 288         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 289
 290         sfc_log_init(sa, "entry");
 291
 292         sfc_adapter_lock(sa);
 293         sfc_stop(sa);
 294         sfc_adapter_unlock(sa);
 295
 296         sfc_log_init(sa, "done");
 297
 298         return 0;
 299 }
 300
 301 static int
 302 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 303 {
 304         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 305         int rc;
 306
 307         sfc_log_init(sa, "entry");
 308
 309         sfc_adapter_lock(sa);
 310         rc = sfc_start(sa);
 311         sfc_adapter_unlock(sa);
 312
 313         SFC_ASSERT(rc >= 0);
 314         return -rc;
 315 }
 316
 317 static int
 318 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 319 {
 320         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 321
 322         sfc_log_init(sa, "entry");
 323
 324         sfc_adapter_lock(sa);
 325         sfc_stop(sa);
 326         sfc_adapter_unlock(sa);
 327
 328         return 0;
 329 }
 330
 331 static void
 332 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 333 {
 334         free(dev->process_private);
 335         rte_eth_dev_release_port(dev);
 336 }
 337
 338 static int
 339 sfc_dev_close(struct rte_eth_dev *dev)
 340 {
 341         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 342
 343         sfc_log_init(sa, "entry");
 344
 345         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 346                 sfc_eth_dev_secondary_clear_ops(dev);
 347                 return 0;
 348         }
 349
 350         sfc_pre_detach(sa);
 351
 352         sfc_adapter_lock(sa);
 353         switch (sa->state) {
 354         case SFC_ETHDEV_STARTED:
 355                 sfc_stop(sa);
 356                 SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
 357                 /* FALLTHROUGH */
 358         case SFC_ETHDEV_CONFIGURED:
 359                 sfc_close(sa);
 360                 SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
 361                 /* FALLTHROUGH */
 362         case SFC_ETHDEV_INITIALIZED:
 363                 break;
 364         default:
 365                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 366                 break;
 367         }
 368
 369         /*
 370          * Cleanup all resources.
 371          * Rollback primary process sfc_eth_dev_init() below.
 372          */
 373
 374         sfc_eth_dev_clear_ops(dev);
 375
 376         sfc_detach(sa);
 377         sfc_unprobe(sa);
 378
 379         sfc_kvargs_cleanup(sa);
 380
 381         sfc_adapter_unlock(sa);
 382         sfc_adapter_lock_fini(sa);
 383
 384         sfc_log_init(sa, "done");
 385
 386         /* Required for logging, so cleanup last */
 387         sa->eth_dev = NULL;
 388
 389         free(sa);
 390
 391         return 0;
 392 }
 393
 394 static int
 395 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 396                    boolean_t enabled)
 397 {
 398         struct sfc_port *port;
 399         boolean_t *toggle;
 400         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 401         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 402         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 403         int rc = 0;
 404
 405         sfc_adapter_lock(sa);
 406
 407         port = &sa->port;
 408         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 409
 410         if (*toggle != enabled) {
 411                 *toggle = enabled;
 412
 413                 if (sfc_sa2shared(sa)->isolated) {
 414                         sfc_warn(sa, "isolated mode is active on the port");
 415                         sfc_warn(sa, "the change is to be applied on the next "
 416                                      "start provided that isolated mode is "
 417                                      "disabled prior the next start");
 418                 } else if ((sa->state == SFC_ETHDEV_STARTED) &&
 419                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 420                         *toggle = !(enabled);
 421                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 422                                  ((enabled) ? "enable" : "disable"), desc, rc);
 423
 424                         /*
 425                          * For promiscuous and all-multicast filters a
 426                          * permission failure should be reported as an
 427                          * unsupported filter.
 428                          */
 429                         if (rc == EPERM)
 430                                 rc = ENOTSUP;
 431                 }
 432         }
 433
 434         sfc_adapter_unlock(sa);
 435         return rc;
 436 }
 437
 438 static int
 439 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 440 {
 441         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 442
 443         SFC_ASSERT(rc >= 0);
 444         return -rc;
 445 }
 446
 447 static int
 448 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 449 {
 450         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 451
 452         SFC_ASSERT(rc >= 0);
 453         return -rc;
 454 }
 455
 456 static int
 457 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 458 {
 459         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 460
 461         SFC_ASSERT(rc >= 0);
 462         return -rc;
 463 }
 464
 465 static int
 466 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 467 {
 468         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 469
 470         SFC_ASSERT(rc >= 0);
 471         return -rc;
 472 }
 473
 474 static int
 475 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 476                    uint16_t nb_rx_desc, unsigned int socket_id,
 477                    const struct rte_eth_rxconf *rx_conf,
 478                    struct rte_mempool *mb_pool)
 479 {
 480         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 481         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 482         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
 483         struct sfc_rxq_info *rxq_info;
 484         sfc_sw_index_t sw_index;
 485         int rc;
 486
 487         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 488                      ethdev_qid, nb_rx_desc, socket_id);
 489
 490         sfc_adapter_lock(sa);
 491
 492         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
 493         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
 494                           rx_conf, mb_pool);
 495         if (rc != 0)
 496                 goto fail_rx_qinit;
 497
 498         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
 499         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
 500
 501         sfc_adapter_unlock(sa);
 502
 503         return 0;
 504
 505 fail_rx_qinit:
 506         sfc_adapter_unlock(sa);
 507         SFC_ASSERT(rc > 0);
 508         return -rc;
 509 }
 510
 511 static void
 512 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 513 {
 514         struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
 515         struct sfc_rxq *rxq;
 516         struct sfc_adapter *sa;
 517         sfc_sw_index_t sw_index;
 518
 519         if (dp_rxq == NULL)
 520                 return;
 521
 522         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 523         sa = rxq->evq->sa;
 524         sfc_adapter_lock(sa);
 525
 526         sw_index = dp_rxq->dpq.queue_id;
 527
 528         sfc_log_init(sa, "RxQ=%u", sw_index);
 529
 530         sfc_rx_qfini(sa, sw_index);
 531
 532         sfc_adapter_unlock(sa);
 533 }
 534
 535 static int
 536 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 537                    uint16_t nb_tx_desc, unsigned int socket_id,
 538                    const struct rte_eth_txconf *tx_conf)
 539 {
 540         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 541         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 542         struct sfc_txq_info *txq_info;
 543         sfc_sw_index_t sw_index;
 544         int rc;
 545
 546         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 547                      ethdev_qid, nb_tx_desc, socket_id);
 548
 549         sfc_adapter_lock(sa);
 550
 551         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
 552         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
 553         if (rc != 0)
 554                 goto fail_tx_qinit;
 555
 556         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
 557         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
 558
 559         sfc_adapter_unlock(sa);
 560         return 0;
 561
 562 fail_tx_qinit:
 563         sfc_adapter_unlock(sa);
 564         SFC_ASSERT(rc > 0);
 565         return -rc;
 566 }
 567
 568 static void
 569 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 570 {
 571         struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
 572         struct sfc_txq *txq;
 573         sfc_sw_index_t sw_index;
 574         struct sfc_adapter *sa;
 575
 576         if (dp_txq == NULL)
 577                 return;
 578
 579         txq = sfc_txq_by_dp_txq(dp_txq);
 580         sw_index = dp_txq->dpq.queue_id;
 581
 582         SFC_ASSERT(txq->evq != NULL);
 583         sa = txq->evq->sa;
 584
 585         sfc_log_init(sa, "TxQ = %u", sw_index);
 586
 587         sfc_adapter_lock(sa);
 588
 589         sfc_tx_qfini(sa, sw_index);
 590
 591         sfc_adapter_unlock(sa);
 592 }
 593
 594 static void
 595 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 596 {
 597         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 598         uint64_t pkts_sum = 0;
 599         uint64_t bytes_sum = 0;
 600         unsigned int i;
 601
 602         for (i = 0; i < sas->ethdev_rxq_count; ++i) {
 603                 struct sfc_rxq_info *rxq_info;
 604
 605                 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
 606                 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
 607                         union sfc_pkts_bytes qstats;
 608
 609                         sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
 610                         pkts_sum += qstats.pkts -
 611                                         sa->sw_stats.reset_rx_pkts[i];
 612                         bytes_sum += qstats.bytes -
 613                                         sa->sw_stats.reset_rx_bytes[i];
 614                 }
 615         }
 616
 617         *pkts = pkts_sum;
 618         *bytes = bytes_sum;
 619 }
 620
 621 static void
 622 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 623 {
 624         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 625         uint64_t pkts_sum = 0;
 626         uint64_t bytes_sum = 0;
 627         unsigned int i;
 628
 629         for (i = 0; i < sas->ethdev_txq_count; ++i) {
 630                 struct sfc_txq_info *txq_info;
 631
 632                 txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
 633                 if (txq_info->state & SFC_TXQ_INITIALIZED) {
 634                         union sfc_pkts_bytes qstats;
 635
 636                         sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
 637                         pkts_sum += qstats.pkts -
 638                                         sa->sw_stats.reset_tx_pkts[i];
 639                         bytes_sum += qstats.bytes -
 640                                         sa->sw_stats.reset_tx_bytes[i];
 641                 }
 642         }
 643
 644         *pkts = pkts_sum;
 645         *bytes = bytes_sum;
 646 }
 647
 648 /*
 649  * Some statistics are computed as A - B where A and B each increase
 650  * monotonically with some hardware counter(s) and the counters are read
 651  * asynchronously.
 652  *
 653  * If packet X is counted in A, but not counted in B yet, computed value is
 654  * greater than real.
 655  *
 656  * If packet X is not counted in A at the moment of reading the counter,
 657  * but counted in B at the moment of reading the counter, computed value
 658  * is less than real.
 659  *
 660  * However, counter which grows backward is worse evil than slightly wrong
 661  * value. So, let's try to guarantee that it never happens except may be
 662  * the case when the MAC stats are zeroed as a result of a NIC reset.
 663  */
 664 static void
 665 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 666 {
 667         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 668                 *stat = newval;
 669 }
 670
 671 static int
 672 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 673 {
 674         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 675         bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
 676         bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
 677         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 678         struct sfc_port *port = &sa->port;
 679         uint64_t *mac_stats;
 680         int ret;
 681
 682         sfc_adapter_lock(sa);
 683
 684         if (have_dp_rx_stats)
 685                 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
 686         if (have_dp_tx_stats)
 687                 sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
 688
 689         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 690         if (ret != 0)
 691                 goto unlock;
 692
 693         mac_stats = port->mac_stats_buf;
 694
 695         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 696                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 697                 if (!have_dp_rx_stats) {
 698                         stats->ipackets =
 699                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 700                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 701                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 702                         stats->ibytes =
 703                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 704                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 705                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 706
 707                         /* CRC is included in these stats, but shouldn't be */
 708                         stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 709                 }
 710                 if (!have_dp_tx_stats) {
 711                         stats->opackets =
 712                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 713                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 714                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 715                         stats->obytes =
 716                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 717                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 718                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 719
 720                         /* CRC is included in these stats, but shouldn't be */
 721                         stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 722                 }
 723                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 724                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 725         } else {
 726                 if (!have_dp_tx_stats) {
 727                         stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 728                         stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
 729                                 mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 730                 }
 731
 732                 /*
 733                  * Take into account stats which are whenever supported
 734                  * on EF10. If some stat is not supported by current
 735                  * firmware variant or HW revision, it is guaranteed
 736                  * to be zero in mac_stats.
 737                  */
 738                 stats->imissed =
 739                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 740                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 741                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 742                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 743                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 744                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 745                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 746                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 747                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 748                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 749                 stats->ierrors =
 750                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 751                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 752                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 753                 /* no oerrors counters supported on EF10 */
 754
 755                 if (!have_dp_rx_stats) {
 756                         /* Exclude missed, errors and pauses from Rx packets */
 757                         sfc_update_diff_stat(&port->ipackets,
 758                                 mac_stats[EFX_MAC_RX_PKTS] -
 759                                 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 760                                 stats->imissed - stats->ierrors);
 761                         stats->ipackets = port->ipackets;
 762                         stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
 763                                 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 764                 }
 765         }
 766
 767 unlock:
 768         sfc_adapter_unlock(sa);
 769         SFC_ASSERT(ret >= 0);
 770         return -ret;
 771 }
 772
 773 static int
 774 sfc_stats_reset(struct rte_eth_dev *dev)
 775 {
 776         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 777         struct sfc_port *port = &sa->port;
 778         int rc;
 779
 780         sfc_adapter_lock(sa);
 781
 782         if (sa->state != SFC_ETHDEV_STARTED) {
 783                 /*
 784                  * The operation cannot be done if port is not started; it
 785                  * will be scheduled to be done during the next port start
 786                  */
 787                 port->mac_stats_reset_pending = B_TRUE;
 788                 sfc_adapter_unlock(sa);
 789                 return 0;
 790         }
 791
 792         rc = sfc_port_reset_mac_stats(sa);
 793         if (rc != 0)
 794                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 795
 796         sfc_sw_xstats_reset(sa);
 797
 798         sfc_adapter_unlock(sa);
 799
 800         SFC_ASSERT(rc >= 0);
 801         return -rc;
 802 }
 803
 804 static unsigned int
 805 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 806 {
 807         struct sfc_port *port = &sa->port;
 808         unsigned int nb_supported;
 809
 810         sfc_adapter_lock(sa);
 811         nb_supported = port->mac_stats_nb_supported +
 812                        sfc_sw_xstats_get_nb_supported(sa);
 813         sfc_adapter_unlock(sa);
 814
 815         return nb_supported;
 816 }
 817
 818 static int
 819 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 820                unsigned int xstats_count)
 821 {
 822         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 823         unsigned int nb_written = 0;
 824         unsigned int nb_supported = 0;
 825         int rc;
 826
 827         if (unlikely(xstats == NULL))
 828                 return sfc_xstats_get_nb_supported(sa);
 829
 830         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 831         if (rc < 0)
 832                 return rc;
 833
 834         nb_supported = rc;
 835         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 836                                &nb_supported);
 837
 838         return nb_supported;
 839 }
 840
 841 static int
 842 sfc_xstats_get_names(struct rte_eth_dev *dev,
 843                      struct rte_eth_xstat_name *xstats_names,
 844                      unsigned int xstats_count)
 845 {
 846         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 847         struct sfc_port *port = &sa->port;
 848         unsigned int i;
 849         unsigned int nstats = 0;
 850         unsigned int nb_written = 0;
 851         int ret;
 852
 853         if (unlikely(xstats_names == NULL))
 854                 return sfc_xstats_get_nb_supported(sa);
 855
 856         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 857                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 858                         if (nstats < xstats_count) {
 859                                 strlcpy(xstats_names[nstats].name,
 860                                         efx_mac_stat_name(sa->nic, i),
 861                                         sizeof(xstats_names[0].name));
 862                                 nb_written++;
 863                         }
 864                         nstats++;
 865                 }
 866         }
 867
 868         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 869                                       &nb_written, &nstats);
 870         if (ret != 0) {
 871                 SFC_ASSERT(ret < 0);
 872                 return ret;
 873         }
 874
 875         return nstats;
 876 }
 877
 878 static int
 879 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 880                      uint64_t *values, unsigned int n)
 881 {
 882         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 883         struct sfc_port *port = &sa->port;
 884         unsigned int nb_supported;
 885         unsigned int i;
 886         int rc;
 887
 888         if (unlikely(ids == NULL || values == NULL))
 889                 return -EINVAL;
 890
 891         /*
 892          * Values array could be filled in nonsequential order. Fill values with
 893          * constant indicating invalid ID first.
 894          */
 895         for (i = 0; i < n; i++)
 896                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 897
 898         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 899         if (rc != 0)
 900                 return rc;
 901
 902         nb_supported = port->mac_stats_nb_supported;
 903         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 904
 905         /* Return number of written stats before invalid ID is encountered. */
 906         for (i = 0; i < n; i++) {
 907                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 908                         return i;
 909         }
 910
 911         return n;
 912 }
 913
 914 static int
 915 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 916                            const uint64_t *ids,
 917                            struct rte_eth_xstat_name *xstats_names,
 918                            unsigned int size)
 919 {
 920         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 921         struct sfc_port *port = &sa->port;
 922         unsigned int nb_supported;
 923         unsigned int i;
 924         int ret;
 925
 926         if (unlikely(xstats_names == NULL && ids != NULL) ||
 927             unlikely(xstats_names != NULL && ids == NULL))
 928                 return -EINVAL;
 929
 930         if (unlikely(xstats_names == NULL && ids == NULL))
 931                 return sfc_xstats_get_nb_supported(sa);
 932
 933         /*
 934          * Names array could be filled in nonsequential order. Fill names with
 935          * string indicating invalid ID first.
 936          */
 937         for (i = 0; i < size; i++)
 938                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 939
 940         sfc_adapter_lock(sa);
 941
 942         SFC_ASSERT(port->mac_stats_nb_supported <=
 943                    RTE_DIM(port->mac_stats_by_id));
 944
 945         for (i = 0; i < size; i++) {
 946                 if (ids[i] < port->mac_stats_nb_supported) {
 947                         strlcpy(xstats_names[i].name,
 948                                 efx_mac_stat_name(sa->nic,
 949                                                  port->mac_stats_by_id[ids[i]]),
 950                                 sizeof(xstats_names[0].name));
 951                 }
 952         }
 953
 954         nb_supported = port->mac_stats_nb_supported;
 955
 956         sfc_adapter_unlock(sa);
 957
 958         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 959                                             &nb_supported);
 960         if (ret != 0) {
 961                 SFC_ASSERT(ret < 0);
 962                 return ret;
 963         }
 964
 965         /* Return number of written names before invalid ID is encountered. */
 966         for (i = 0; i < size; i++) {
 967                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 968                         return i;
 969         }
 970
 971         return size;
 972 }
 973
 974 static int
 975 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 976 {
 977         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 978         unsigned int wanted_fc, link_fc;
 979
 980         memset(fc_conf, 0, sizeof(*fc_conf));
 981
 982         sfc_adapter_lock(sa);
 983
 984         if (sa->state == SFC_ETHDEV_STARTED)
 985                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 986         else
 987                 link_fc = sa->port.flow_ctrl;
 988
 989         switch (link_fc) {
 990         case 0:
 991                 fc_conf->mode = RTE_FC_NONE;
 992                 break;
 993         case EFX_FCNTL_RESPOND:
 994                 fc_conf->mode = RTE_FC_RX_PAUSE;
 995                 break;
 996         case EFX_FCNTL_GENERATE:
 997                 fc_conf->mode = RTE_FC_TX_PAUSE;
 998                 break;
 999         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
1000                 fc_conf->mode = RTE_FC_FULL;
1001                 break;
1002         default:
1003                 sfc_err(sa, "%s: unexpected flow control value %#x",
1004                         __func__, link_fc);
1005         }
1006
1007         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1008
1009         sfc_adapter_unlock(sa);
1010
1011         return 0;
1012 }
1013
1014 static int
1015 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1016 {
1017         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1018         struct sfc_port *port = &sa->port;
1019         unsigned int fcntl;
1020         int rc;
1021
1022         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1023             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1024             fc_conf->mac_ctrl_frame_fwd != 0) {
1025                 sfc_err(sa, "unsupported flow control settings specified");
1026                 rc = EINVAL;
1027                 goto fail_inval;
1028         }
1029
1030         switch (fc_conf->mode) {
1031         case RTE_FC_NONE:
1032                 fcntl = 0;
1033                 break;
1034         case RTE_FC_RX_PAUSE:
1035                 fcntl = EFX_FCNTL_RESPOND;
1036                 break;
1037         case RTE_FC_TX_PAUSE:
1038                 fcntl = EFX_FCNTL_GENERATE;
1039                 break;
1040         case RTE_FC_FULL:
1041                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1042                 break;
1043         default:
1044                 rc = EINVAL;
1045                 goto fail_inval;
1046         }
1047
1048         sfc_adapter_lock(sa);
1049
1050         if (sa->state == SFC_ETHDEV_STARTED) {
1051                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1052                 if (rc != 0)
1053                         goto fail_mac_fcntl_set;
1054         }
1055
1056         port->flow_ctrl = fcntl;
1057         port->flow_ctrl_autoneg = fc_conf->autoneg;
1058
1059         sfc_adapter_unlock(sa);
1060
1061         return 0;
1062
1063 fail_mac_fcntl_set:
1064         sfc_adapter_unlock(sa);
1065 fail_inval:
1066         SFC_ASSERT(rc > 0);
1067         return -rc;
1068 }
1069
1070 static int
1071 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1072 {
1073         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1074         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1075         boolean_t scatter_enabled;
1076         const char *error;
1077         unsigned int i;
1078
1079         for (i = 0; i < sas->rxq_count; i++) {
1080                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1081                         continue;
1082
1083                 scatter_enabled = (sas->rxq_info[i].type_flags &
1084                                    EFX_RXQ_FLAG_SCATTER);
1085
1086                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1087                                           encp->enc_rx_prefix_size,
1088                                           scatter_enabled,
1089                                           encp->enc_rx_scatter_max, &error)) {
1090                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1091                                 error);
1092                         return EINVAL;
1093                 }
1094         }
1095
1096         return 0;
1097 }
1098
1099 static int
1100 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1101 {
1102         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1103         size_t pdu = EFX_MAC_PDU(mtu);
1104         size_t old_pdu;
1105         int rc;
1106
1107         sfc_log_init(sa, "mtu=%u", mtu);
1108
1109         rc = EINVAL;
1110         if (pdu < EFX_MAC_PDU_MIN) {
1111                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1112                         (unsigned int)mtu, (unsigned int)pdu,
1113                         EFX_MAC_PDU_MIN);
1114                 goto fail_inval;
1115         }
1116         if (pdu > EFX_MAC_PDU_MAX) {
1117                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1118                         (unsigned int)mtu, (unsigned int)pdu,
1119                         (unsigned int)EFX_MAC_PDU_MAX);
1120                 goto fail_inval;
1121         }
1122
1123         sfc_adapter_lock(sa);
1124
1125         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1126         if (rc != 0)
1127                 goto fail_check_scatter;
1128
1129         if (pdu != sa->port.pdu) {
1130                 if (sa->state == SFC_ETHDEV_STARTED) {
1131                         sfc_stop(sa);
1132
1133                         old_pdu = sa->port.pdu;
1134                         sa->port.pdu = pdu;
1135                         rc = sfc_start(sa);
1136                         if (rc != 0)
1137                                 goto fail_start;
1138                 } else {
1139                         sa->port.pdu = pdu;
1140                 }
1141         }
1142
1143         /*
1144          * The driver does not use it, but other PMDs update jumbo frame
1145          * flag when MTU is set.
1146          */
1147         if (mtu > RTE_ETHER_MTU) {
1148                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1149                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1150         }
1151
1152         sfc_adapter_unlock(sa);
1153
1154         sfc_log_init(sa, "done");
1155         return 0;
1156
1157 fail_start:
1158         sa->port.pdu = old_pdu;
1159         if (sfc_start(sa) != 0)
1160                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1161                         "PDU max size - port is stopped",
1162                         (unsigned int)pdu, (unsigned int)old_pdu);
1163
1164 fail_check_scatter:
1165         sfc_adapter_unlock(sa);
1166
1167 fail_inval:
1168         sfc_log_init(sa, "failed %d", rc);
1169         SFC_ASSERT(rc > 0);
1170         return -rc;
1171 }
1172 static int
1173 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1174 {
1175         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1176         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1177         struct sfc_port *port = &sa->port;
1178         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1179         int rc = 0;
1180
1181         sfc_adapter_lock(sa);
1182
1183         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1184                 goto unlock;
1185
1186         /*
1187          * Copy the address to the device private data so that
1188          * it could be recalled in the case of adapter restart.
1189          */
1190         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1191
1192         /*
1193          * Neither of the two following checks can return
1194          * an error. The new MAC address is preserved in
1195          * the device private data and can be activated
1196          * on the next port start if the user prevents
1197          * isolated mode from being enabled.
1198          */
1199         if (sfc_sa2shared(sa)->isolated) {
1200                 sfc_warn(sa, "isolated mode is active on the port");
1201                 sfc_warn(sa, "will not set MAC address");
1202                 goto unlock;
1203         }
1204
1205         if (sa->state != SFC_ETHDEV_STARTED) {
1206                 sfc_notice(sa, "the port is not started");
1207                 sfc_notice(sa, "the new MAC address will be set on port start");
1208
1209                 goto unlock;
1210         }
1211
1212         if (encp->enc_allow_set_mac_with_installed_filters) {
1213                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1214                 if (rc != 0) {
1215                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1216                         goto unlock;
1217                 }
1218
1219                 /*
1220                  * Changing the MAC address by means of MCDI request
1221                  * has no effect on received traffic, therefore
1222                  * we also need to update unicast filters
1223                  */
1224                 rc = sfc_set_rx_mode_unchecked(sa);
1225                 if (rc != 0) {
1226                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1227                         /* Rollback the old address */
1228                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1229                         (void)sfc_set_rx_mode_unchecked(sa);
1230                 }
1231         } else {
1232                 sfc_warn(sa, "cannot set MAC address with filters installed");
1233                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1234                 sfc_warn(sa, "(some traffic may be dropped)");
1235
1236                 /*
1237                  * Since setting MAC address with filters installed is not
1238                  * allowed on the adapter, the new MAC address will be set
1239                  * by means of adapter restart. sfc_start() shall retrieve
1240                  * the new address from the device private data and set it.
1241                  */
1242                 sfc_stop(sa);
1243                 rc = sfc_start(sa);
1244                 if (rc != 0)
1245                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1246         }
1247
1248 unlock:
1249         if (rc != 0)
1250                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1251
1252         sfc_adapter_unlock(sa);
1253
1254         SFC_ASSERT(rc >= 0);
1255         return -rc;
1256 }
1257
1258
1259 static int
1260 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1261                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1262 {
1263         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1264         struct sfc_port *port = &sa->port;
1265         uint8_t *mc_addrs = port->mcast_addrs;
1266         int rc;
1267         unsigned int i;
1268
1269         if (sfc_sa2shared(sa)->isolated) {
1270                 sfc_err(sa, "isolated mode is active on the port");
1271                 sfc_err(sa, "will not set multicast address list");
1272                 return -ENOTSUP;
1273         }
1274
1275         if (mc_addrs == NULL)
1276                 return -ENOBUFS;
1277
1278         if (nb_mc_addr > port->max_mcast_addrs) {
1279                 sfc_err(sa, "too many multicast addresses: %u > %u",
1280                          nb_mc_addr, port->max_mcast_addrs);
1281                 return -EINVAL;
1282         }
1283
1284         for (i = 0; i < nb_mc_addr; ++i) {
1285                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1286                                  EFX_MAC_ADDR_LEN);
1287                 mc_addrs += EFX_MAC_ADDR_LEN;
1288         }
1289
1290         port->nb_mcast_addrs = nb_mc_addr;
1291
1292         if (sa->state != SFC_ETHDEV_STARTED)
1293                 return 0;
1294
1295         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1296                                         port->nb_mcast_addrs);
1297         if (rc != 0)
1298                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1299
1300         SFC_ASSERT(rc >= 0);
1301         return -rc;
1302 }
1303
1304 /*
1305  * The function is used by the secondary process as well. It must not
1306  * use any process-local pointers from the adapter data.
1307  */
1308 static void
1309 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1310                       struct rte_eth_rxq_info *qinfo)
1311 {
1312         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1313         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1314         struct sfc_rxq_info *rxq_info;
1315
1316         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1317
1318         qinfo->mp = rxq_info->refill_mb_pool;
1319         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1320         qinfo->conf.rx_drop_en = 1;
1321         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1322         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1323         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1324                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1325                 qinfo->scattered_rx = 1;
1326         }
1327         qinfo->nb_desc = rxq_info->entries;
1328 }
1329
1330 /*
1331  * The function is used by the secondary process as well. It must not
1332  * use any process-local pointers from the adapter data.
1333  */
1334 static void
1335 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1336                       struct rte_eth_txq_info *qinfo)
1337 {
1338         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1339         struct sfc_txq_info *txq_info;
1340
1341         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1342
1343         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1344
1345         memset(qinfo, 0, sizeof(*qinfo));
1346
1347         qinfo->conf.offloads = txq_info->offloads;
1348         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1349         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1350         qinfo->nb_desc = txq_info->entries;
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 uint32_t
1358 sfc_rx_queue_count(void *rx_queue)
1359 {
1360         struct sfc_dp_rxq *dp_rxq = rx_queue;
1361         const struct sfc_dp_rx *dp_rx;
1362         struct sfc_rxq_info *rxq_info;
1363
1364         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1365         rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1366
1367         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1368                 return 0;
1369
1370         return dp_rx->qdesc_npending(dp_rxq);
1371 }
1372
1373 /*
1374  * The function is used by the secondary process as well. It must not
1375  * use any process-local pointers from the adapter data.
1376  */
1377 static int
1378 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1379 {
1380         struct sfc_dp_rxq *dp_rxq = queue;
1381         const struct sfc_dp_rx *dp_rx;
1382
1383         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1384
1385         return dp_rx->qdesc_status(dp_rxq, offset);
1386 }
1387
1388 /*
1389  * The function is used by the secondary process as well. It must not
1390  * use any process-local pointers from the adapter data.
1391  */
1392 static int
1393 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1394 {
1395         struct sfc_dp_txq *dp_txq = queue;
1396         const struct sfc_dp_tx *dp_tx;
1397
1398         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1399
1400         return dp_tx->qdesc_status(dp_txq, offset);
1401 }
1402
1403 static int
1404 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1405 {
1406         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1407         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1408         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1409         struct sfc_rxq_info *rxq_info;
1410         sfc_sw_index_t sw_index;
1411         int rc;
1412
1413         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1414
1415         sfc_adapter_lock(sa);
1416
1417         rc = EINVAL;
1418         if (sa->state != SFC_ETHDEV_STARTED)
1419                 goto fail_not_started;
1420
1421         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1422         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1423                 goto fail_not_setup;
1424
1425         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1426         rc = sfc_rx_qstart(sa, sw_index);
1427         if (rc != 0)
1428                 goto fail_rx_qstart;
1429
1430         rxq_info->deferred_started = B_TRUE;
1431
1432         sfc_adapter_unlock(sa);
1433
1434         return 0;
1435
1436 fail_rx_qstart:
1437 fail_not_setup:
1438 fail_not_started:
1439         sfc_adapter_unlock(sa);
1440         SFC_ASSERT(rc > 0);
1441         return -rc;
1442 }
1443
1444 static int
1445 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1446 {
1447         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1448         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1449         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1450         struct sfc_rxq_info *rxq_info;
1451         sfc_sw_index_t sw_index;
1452
1453         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1454
1455         sfc_adapter_lock(sa);
1456
1457         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1458         sfc_rx_qstop(sa, sw_index);
1459
1460         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1461         rxq_info->deferred_started = B_FALSE;
1462
1463         sfc_adapter_unlock(sa);
1464
1465         return 0;
1466 }
1467
1468 static int
1469 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1470 {
1471         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1472         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1473         struct sfc_txq_info *txq_info;
1474         sfc_sw_index_t sw_index;
1475         int rc;
1476
1477         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1478
1479         sfc_adapter_lock(sa);
1480
1481         rc = EINVAL;
1482         if (sa->state != SFC_ETHDEV_STARTED)
1483                 goto fail_not_started;
1484
1485         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1486         if (txq_info->state != SFC_TXQ_INITIALIZED)
1487                 goto fail_not_setup;
1488
1489         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1490         rc = sfc_tx_qstart(sa, sw_index);
1491         if (rc != 0)
1492                 goto fail_tx_qstart;
1493
1494         txq_info->deferred_started = B_TRUE;
1495
1496         sfc_adapter_unlock(sa);
1497         return 0;
1498
1499 fail_tx_qstart:
1500
1501 fail_not_setup:
1502 fail_not_started:
1503         sfc_adapter_unlock(sa);
1504         SFC_ASSERT(rc > 0);
1505         return -rc;
1506 }
1507
1508 static int
1509 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1510 {
1511         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1512         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1513         struct sfc_txq_info *txq_info;
1514         sfc_sw_index_t sw_index;
1515
1516         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1517
1518         sfc_adapter_lock(sa);
1519
1520         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1521         sfc_tx_qstop(sa, sw_index);
1522
1523         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1524         txq_info->deferred_started = B_FALSE;
1525
1526         sfc_adapter_unlock(sa);
1527         return 0;
1528 }
1529
1530 static efx_tunnel_protocol_t
1531 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1532 {
1533         switch (rte_type) {
1534         case RTE_TUNNEL_TYPE_VXLAN:
1535                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1536         case RTE_TUNNEL_TYPE_GENEVE:
1537                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1538         default:
1539                 return EFX_TUNNEL_NPROTOS;
1540         }
1541 }
1542
1543 enum sfc_udp_tunnel_op_e {
1544         SFC_UDP_TUNNEL_ADD_PORT,
1545         SFC_UDP_TUNNEL_DEL_PORT,
1546 };
1547
1548 static int
1549 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1550                       struct rte_eth_udp_tunnel *tunnel_udp,
1551                       enum sfc_udp_tunnel_op_e op)
1552 {
1553         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1554         efx_tunnel_protocol_t tunnel_proto;
1555         int rc;
1556
1557         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1558                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1559                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1560                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1561
1562         tunnel_proto =
1563                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1564         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1565                 rc = ENOTSUP;
1566                 goto fail_bad_proto;
1567         }
1568
1569         sfc_adapter_lock(sa);
1570
1571         switch (op) {
1572         case SFC_UDP_TUNNEL_ADD_PORT:
1573                 rc = efx_tunnel_config_udp_add(sa->nic,
1574                                                tunnel_udp->udp_port,
1575                                                tunnel_proto);
1576                 break;
1577         case SFC_UDP_TUNNEL_DEL_PORT:
1578                 rc = efx_tunnel_config_udp_remove(sa->nic,
1579                                                   tunnel_udp->udp_port,
1580                                                   tunnel_proto);
1581                 break;
1582         default:
1583                 rc = EINVAL;
1584                 goto fail_bad_op;
1585         }
1586
1587         if (rc != 0)
1588                 goto fail_op;
1589
1590         if (sa->state == SFC_ETHDEV_STARTED) {
1591                 rc = efx_tunnel_reconfigure(sa->nic);
1592                 if (rc == EAGAIN) {
1593                         /*
1594                          * Configuration is accepted by FW and MC reboot
1595                          * is initiated to apply the changes. MC reboot
1596                          * will be handled in a usual way (MC reboot
1597                          * event on management event queue and adapter
1598                          * restart).
1599                          */
1600                         rc = 0;
1601                 } else if (rc != 0) {
1602                         goto fail_reconfigure;
1603                 }
1604         }
1605
1606         sfc_adapter_unlock(sa);
1607         return 0;
1608
1609 fail_reconfigure:
1610         /* Remove/restore entry since the change makes the trouble */
1611         switch (op) {
1612         case SFC_UDP_TUNNEL_ADD_PORT:
1613                 (void)efx_tunnel_config_udp_remove(sa->nic,
1614                                                    tunnel_udp->udp_port,
1615                                                    tunnel_proto);
1616                 break;
1617         case SFC_UDP_TUNNEL_DEL_PORT:
1618                 (void)efx_tunnel_config_udp_add(sa->nic,
1619                                                 tunnel_udp->udp_port,
1620                                                 tunnel_proto);
1621                 break;
1622         }
1623
1624 fail_op:
1625 fail_bad_op:
1626         sfc_adapter_unlock(sa);
1627
1628 fail_bad_proto:
1629         SFC_ASSERT(rc > 0);
1630         return -rc;
1631 }
1632
1633 static int
1634 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1635                             struct rte_eth_udp_tunnel *tunnel_udp)
1636 {
1637         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1638 }
1639
1640 static int
1641 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1642                             struct rte_eth_udp_tunnel *tunnel_udp)
1643 {
1644         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1645 }
1646
1647 /*
1648  * The function is used by the secondary process as well. It must not
1649  * use any process-local pointers from the adapter data.
1650  */
1651 static int
1652 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1653                           struct rte_eth_rss_conf *rss_conf)
1654 {
1655         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1656         struct sfc_rss *rss = &sas->rss;
1657
1658         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1659                 return -ENOTSUP;
1660
1661         /*
1662          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1663          * hence, conversion is done here to derive a correct set of ETH_RSS
1664          * flags which corresponds to the active EFX configuration stored
1665          * locally in 'sfc_adapter' and kept up-to-date
1666          */
1667         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1668         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1669         if (rss_conf->rss_key != NULL)
1670                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1671
1672         return 0;
1673 }
1674
1675 static int
1676 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1677                         struct rte_eth_rss_conf *rss_conf)
1678 {
1679         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1680         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1681         unsigned int efx_hash_types;
1682         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1683         unsigned int n_contexts;
1684         unsigned int mode_i = 0;
1685         unsigned int key_i = 0;
1686         unsigned int i = 0;
1687         int rc = 0;
1688
1689         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1690
1691         if (sfc_sa2shared(sa)->isolated)
1692                 return -ENOTSUP;
1693
1694         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1695                 sfc_err(sa, "RSS is not available");
1696                 return -ENOTSUP;
1697         }
1698
1699         if (rss->channels == 0) {
1700                 sfc_err(sa, "RSS is not configured");
1701                 return -EINVAL;
1702         }
1703
1704         if ((rss_conf->rss_key != NULL) &&
1705             (rss_conf->rss_key_len != sizeof(rss->key))) {
1706                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1707                         sizeof(rss->key));
1708                 return -EINVAL;
1709         }
1710
1711         sfc_adapter_lock(sa);
1712
1713         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1714         if (rc != 0)
1715                 goto fail_rx_hf_rte_to_efx;
1716
1717         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1718                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1719                                            rss->hash_alg, efx_hash_types,
1720                                            B_TRUE);
1721                 if (rc != 0)
1722                         goto fail_scale_mode_set;
1723         }
1724
1725         if (rss_conf->rss_key != NULL) {
1726                 if (sa->state == SFC_ETHDEV_STARTED) {
1727                         for (key_i = 0; key_i < n_contexts; key_i++) {
1728                                 rc = efx_rx_scale_key_set(sa->nic,
1729                                                           contexts[key_i],
1730                                                           rss_conf->rss_key,
1731                                                           sizeof(rss->key));
1732                                 if (rc != 0)
1733                                         goto fail_scale_key_set;
1734                         }
1735                 }
1736
1737                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1738         }
1739
1740         rss->hash_types = efx_hash_types;
1741
1742         sfc_adapter_unlock(sa);
1743
1744         return 0;
1745
1746 fail_scale_key_set:
1747         for (i = 0; i < key_i; i++) {
1748                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1749                                          sizeof(rss->key)) != 0)
1750                         sfc_err(sa, "failed to restore RSS key");
1751         }
1752
1753 fail_scale_mode_set:
1754         for (i = 0; i < mode_i; i++) {
1755                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1756                                           EFX_RX_HASHALG_TOEPLITZ,
1757                                           rss->hash_types, B_TRUE) != 0)
1758                         sfc_err(sa, "failed to restore RSS mode");
1759         }
1760
1761 fail_rx_hf_rte_to_efx:
1762         sfc_adapter_unlock(sa);
1763         return -rc;
1764 }
1765
1766 /*
1767  * The function is used by the secondary process as well. It must not
1768  * use any process-local pointers from the adapter data.
1769  */
1770 static int
1771 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1772                        struct rte_eth_rss_reta_entry64 *reta_conf,
1773                        uint16_t reta_size)
1774 {
1775         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1776         struct sfc_rss *rss = &sas->rss;
1777         int entry;
1778
1779         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1780                 return -ENOTSUP;
1781
1782         if (rss->channels == 0)
1783                 return -EINVAL;
1784
1785         if (reta_size != EFX_RSS_TBL_SIZE)
1786                 return -EINVAL;
1787
1788         for (entry = 0; entry < reta_size; entry++) {
1789                 int grp = entry / RTE_RETA_GROUP_SIZE;
1790                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1791
1792                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1793                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1794         }
1795
1796         return 0;
1797 }
1798
1799 static int
1800 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1801                         struct rte_eth_rss_reta_entry64 *reta_conf,
1802                         uint16_t reta_size)
1803 {
1804         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1805         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1806         unsigned int *rss_tbl_new;
1807         uint16_t entry;
1808         int rc = 0;
1809
1810
1811         if (sfc_sa2shared(sa)->isolated)
1812                 return -ENOTSUP;
1813
1814         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1815                 sfc_err(sa, "RSS is not available");
1816                 return -ENOTSUP;
1817         }
1818
1819         if (rss->channels == 0) {
1820                 sfc_err(sa, "RSS is not configured");
1821                 return -EINVAL;
1822         }
1823
1824         if (reta_size != EFX_RSS_TBL_SIZE) {
1825                 sfc_err(sa, "RETA size is wrong (should be %u)",
1826                         EFX_RSS_TBL_SIZE);
1827                 return -EINVAL;
1828         }
1829
1830         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1831         if (rss_tbl_new == NULL)
1832                 return -ENOMEM;
1833
1834         sfc_adapter_lock(sa);
1835
1836         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1837
1838         for (entry = 0; entry < reta_size; entry++) {
1839                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1840                 struct rte_eth_rss_reta_entry64 *grp;
1841
1842                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1843
1844                 if (grp->mask & (1ull << grp_idx)) {
1845                         if (grp->reta[grp_idx] >= rss->channels) {
1846                                 rc = EINVAL;
1847                                 goto bad_reta_entry;
1848                         }
1849                         rss_tbl_new[entry] = grp->reta[grp_idx];
1850                 }
1851         }
1852
1853         if (sa->state == SFC_ETHDEV_STARTED) {
1854                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1855                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1856                 if (rc != 0)
1857                         goto fail_scale_tbl_set;
1858         }
1859
1860         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1861
1862 fail_scale_tbl_set:
1863 bad_reta_entry:
1864         sfc_adapter_unlock(sa);
1865
1866         rte_free(rss_tbl_new);
1867
1868         SFC_ASSERT(rc >= 0);
1869         return -rc;
1870 }
1871
1872 static int
1873 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1874                      const struct rte_flow_ops **ops)
1875 {
1876         *ops = &sfc_flow_ops;
1877         return 0;
1878 }
1879
1880 static int
1881 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1882 {
1883         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1884
1885         /*
1886          * If Rx datapath does not provide callback to check mempool,
1887          * all pools are supported.
1888          */
1889         if (sap->dp_rx->pool_ops_supported == NULL)
1890                 return 1;
1891
1892         return sap->dp_rx->pool_ops_supported(pool);
1893 }
1894
1895 static int
1896 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1897 {
1898         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1899         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1900         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1901         struct sfc_rxq_info *rxq_info;
1902
1903         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1904
1905         return sap->dp_rx->intr_enable(rxq_info->dp);
1906 }
1907
1908 static int
1909 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1910 {
1911         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1912         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1913         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1914         struct sfc_rxq_info *rxq_info;
1915
1916         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1917
1918         return sap->dp_rx->intr_disable(rxq_info->dp);
1919 }
1920
1921 struct sfc_mport_journal_ctx {
1922         struct sfc_adapter              *sa;
1923         uint16_t                        switch_domain_id;
1924         uint32_t                        mcdi_handle;
1925         bool                            controllers_assigned;
1926         efx_pcie_interface_t            *controllers;
1927         size_t                          nb_controllers;
1928 };
1929
1930 static int
1931 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1932                                efx_pcie_interface_t intf)
1933 {
1934         efx_pcie_interface_t *new_controllers;
1935         size_t i, target;
1936         size_t new_size;
1937
1938         if (ctx->controllers == NULL) {
1939                 ctx->controllers = rte_malloc("sfc_controller_mapping",
1940                                               sizeof(ctx->controllers[0]), 0);
1941                 if (ctx->controllers == NULL)
1942                         return ENOMEM;
1943
1944                 ctx->controllers[0] = intf;
1945                 ctx->nb_controllers = 1;
1946
1947                 return 0;
1948         }
1949
1950         for (i = 0; i < ctx->nb_controllers; i++) {
1951                 if (ctx->controllers[i] == intf)
1952                         return 0;
1953                 if (ctx->controllers[i] > intf)
1954                         break;
1955         }
1956         target = i;
1957
1958         ctx->nb_controllers += 1;
1959         new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1960
1961         new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1962         if (new_controllers == NULL) {
1963                 rte_free(ctx->controllers);
1964                 return ENOMEM;
1965         }
1966         ctx->controllers = new_controllers;
1967
1968         for (i = target + 1; i < ctx->nb_controllers; i++)
1969                 ctx->controllers[i] = ctx->controllers[i - 1];
1970
1971         ctx->controllers[target] = intf;
1972
1973         return 0;
1974 }
1975
1976 static efx_rc_t
1977 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1978                                 efx_mport_desc_t *mport)
1979 {
1980         struct sfc_mae_switch_port_request req;
1981         efx_mport_sel_t entity_selector;
1982         efx_mport_sel_t ethdev_mport;
1983         uint16_t switch_port_id;
1984         efx_rc_t efx_rc;
1985         int rc;
1986
1987         sfc_dbg(ctx->sa,
1988                 "processing mport id %u (controller %u pf %u vf %u)",
1989                 mport->emd_id.id, mport->emd_vnic.ev_intf,
1990                 mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1991         efx_mae_mport_invalid(&ethdev_mport);
1992
1993         if (!ctx->controllers_assigned) {
1994                 rc = sfc_journal_ctx_add_controller(ctx,
1995                                                     mport->emd_vnic.ev_intf);
1996                 if (rc != 0)
1997                         return rc;
1998         }
1999
2000         /* Build Mport selector */
2001         efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
2002                                                 mport->emd_vnic.ev_pf,
2003                                                 mport->emd_vnic.ev_vf,
2004                                                 &entity_selector);
2005         if (efx_rc != 0) {
2006                 sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
2007                         mport->emd_vnic.ev_intf,
2008                         mport->emd_vnic.ev_pf,
2009                         mport->emd_vnic.ev_vf);
2010                 return efx_rc;
2011         }
2012
2013         rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2014                                               &entity_selector,
2015                                               SFC_MAE_SWITCH_PORT_REPRESENTOR,
2016                                               &switch_port_id);
2017         switch (rc) {
2018         case 0:
2019                 /* Already registered */
2020                 break;
2021         case ENOENT:
2022                 /*
2023                  * No representor has been created for this entity.
2024                  * Create a dummy switch registry entry with an invalid ethdev
2025                  * mport selector. When a corresponding representor is created,
2026                  * this entry will be updated.
2027                  */
2028                 req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2029                 req.entity_mportp = &entity_selector;
2030                 req.ethdev_mportp = &ethdev_mport;
2031                 req.ethdev_port_id = RTE_MAX_ETHPORTS;
2032                 req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2033                 req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2034                 req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2035
2036                 rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2037                                                 &req, &switch_port_id);
2038                 if (rc != 0) {
2039                         sfc_err(ctx->sa,
2040                                 "failed to assign MAE switch port for c%upf%uvf%u: %s",
2041                                 mport->emd_vnic.ev_intf,
2042                                 mport->emd_vnic.ev_pf,
2043                                 mport->emd_vnic.ev_vf,
2044                                 rte_strerror(rc));
2045                         return rc;
2046                 }
2047                 break;
2048         default:
2049                 sfc_err(ctx->sa, "failed to find MAE switch port for c%upf%uvf%u: %s",
2050                         mport->emd_vnic.ev_intf,
2051                         mport->emd_vnic.ev_pf,
2052                         mport->emd_vnic.ev_vf,
2053                         rte_strerror(rc));
2054                 return rc;
2055         }
2056
2057         return 0;
2058 }
2059
2060 static efx_rc_t
2061 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2062                              size_t mport_len)
2063 {
2064         struct sfc_mport_journal_ctx *ctx = data;
2065
2066         if (ctx == NULL || ctx->sa == NULL) {
2067                 sfc_err(ctx->sa, "received NULL context or SFC adapter");
2068                 return EINVAL;
2069         }
2070
2071         if (mport_len != sizeof(*mport)) {
2072                 sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2073                 return EINVAL;
2074         }
2075
2076         SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2077
2078         /*
2079          * If a zombie flag is set, it means the mport has been marked for
2080          * deletion and cannot be used for any new operations. The mport will
2081          * be destroyed completely once all references to it are released.
2082          */
2083         if (mport->emd_zombie) {
2084                 sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2085                 return 0;
2086         }
2087         if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2088                 sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2089                 return 0;
2090         }
2091         if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2092                 sfc_dbg(ctx->sa, "mport is not a function, skipping");
2093                 return 0;
2094         }
2095         if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2096                 sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2097                 return 0;
2098         }
2099
2100         return sfc_process_mport_journal_entry(ctx, mport);
2101 }
2102
2103 static int
2104 sfc_process_mport_journal(struct sfc_adapter *sa)
2105 {
2106         struct sfc_mport_journal_ctx ctx;
2107         const efx_pcie_interface_t *controllers;
2108         size_t nb_controllers;
2109         efx_rc_t efx_rc;
2110         int rc;
2111
2112         memset(&ctx, 0, sizeof(ctx));
2113         ctx.sa = sa;
2114         ctx.switch_domain_id = sa->mae.switch_domain_id;
2115
2116         efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2117         if (efx_rc != 0) {
2118                 sfc_err(sa, "failed to get own MCDI handle");
2119                 SFC_ASSERT(efx_rc > 0);
2120                 return efx_rc;
2121         }
2122
2123         rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2124                                                &controllers, &nb_controllers);
2125         if (rc != 0) {
2126                 sfc_err(sa, "failed to get controller mapping");
2127                 return rc;
2128         }
2129
2130         ctx.controllers_assigned = controllers != NULL;
2131         ctx.controllers = NULL;
2132         ctx.nb_controllers = 0;
2133
2134         efx_rc = efx_mae_read_mport_journal(sa->nic,
2135                                             sfc_process_mport_journal_cb, &ctx);
2136         if (efx_rc != 0) {
2137                 sfc_err(sa, "failed to process MAE mport journal");
2138                 SFC_ASSERT(efx_rc > 0);
2139                 return efx_rc;
2140         }
2141
2142         if (controllers == NULL) {
2143                 rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2144                                                            ctx.controllers,
2145                                                            ctx.nb_controllers);
2146                 if (rc != 0)
2147                         return rc;
2148         }
2149
2150         return 0;
2151 }
2152
2153 static void
2154 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2155                           const efx_mport_sel_t *ethdev_mportp __rte_unused,
2156                           uint16_t ethdev_port_id __rte_unused,
2157                           const efx_mport_sel_t *entity_mportp __rte_unused,
2158                           uint16_t switch_port_id __rte_unused,
2159                           union sfc_mae_switch_port_data *port_datap
2160                                 __rte_unused,
2161                           void *user_datap)
2162 {
2163         int *counter = user_datap;
2164
2165         SFC_ASSERT(counter != NULL);
2166
2167         if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2168                 (*counter)++;
2169 }
2170
2171 struct sfc_get_representors_ctx {
2172         struct rte_eth_representor_info *info;
2173         struct sfc_adapter              *sa;
2174         uint16_t                        switch_domain_id;
2175         const efx_pcie_interface_t      *controllers;
2176         size_t                          nb_controllers;
2177 };
2178
2179 static void
2180 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2181                         const efx_mport_sel_t *ethdev_mportp __rte_unused,
2182                         uint16_t ethdev_port_id __rte_unused,
2183                         const efx_mport_sel_t *entity_mportp __rte_unused,
2184                         uint16_t switch_port_id,
2185                         union sfc_mae_switch_port_data *port_datap,
2186                         void *user_datap)
2187 {
2188         struct sfc_get_representors_ctx *ctx = user_datap;
2189         struct rte_eth_representor_range *range;
2190         int ret;
2191         int rc;
2192
2193         SFC_ASSERT(ctx != NULL);
2194         SFC_ASSERT(ctx->info != NULL);
2195         SFC_ASSERT(ctx->sa != NULL);
2196
2197         if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2198                 sfc_dbg(ctx->sa, "not a representor, skipping");
2199                 return;
2200         }
2201         if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2202                 sfc_dbg(ctx->sa, "info structure is full already");
2203                 return;
2204         }
2205
2206         range = &ctx->info->ranges[ctx->info->nb_ranges];
2207         rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2208                                                     ctx->nb_controllers,
2209                                                     port_datap->repr.intf,
2210                                                     &range->controller);
2211         if (rc != 0) {
2212                 sfc_err(ctx->sa, "invalid representor controller: %d",
2213                         port_datap->repr.intf);
2214                 range->controller = -1;
2215         }
2216         range->pf = port_datap->repr.pf;
2217         range->id_base = switch_port_id;
2218         range->id_end = switch_port_id;
2219
2220         if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2221                 range->type = RTE_ETH_REPRESENTOR_VF;
2222                 range->vf = port_datap->repr.vf;
2223                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2224                                "c%dpf%dvf%d", range->controller, range->pf,
2225                                range->vf);
2226         } else {
2227                 range->type = RTE_ETH_REPRESENTOR_PF;
2228                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2229                          "c%dpf%d", range->controller, range->pf);
2230         }
2231         if (ret >= RTE_DEV_NAME_MAX_LEN) {
2232                 sfc_err(ctx->sa, "representor name has been truncated: %s",
2233                         range->name);
2234         }
2235
2236         ctx->info->nb_ranges++;
2237 }
2238
2239 static int
2240 sfc_representor_info_get(struct rte_eth_dev *dev,
2241                          struct rte_eth_representor_info *info)
2242 {
2243         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2244         struct sfc_get_representors_ctx get_repr_ctx;
2245         const efx_nic_cfg_t *nic_cfg;
2246         uint16_t switch_domain_id;
2247         uint32_t nb_repr;
2248         int controller;
2249         int rc;
2250
2251         sfc_adapter_lock(sa);
2252
2253         if (sa->mae.status != SFC_MAE_STATUS_SUPPORTED) {
2254                 sfc_adapter_unlock(sa);
2255                 return -ENOTSUP;
2256         }
2257
2258         rc = sfc_process_mport_journal(sa);
2259         if (rc != 0) {
2260                 sfc_adapter_unlock(sa);
2261                 SFC_ASSERT(rc > 0);
2262                 return -rc;
2263         }
2264
2265         switch_domain_id = sa->mae.switch_domain_id;
2266
2267         nb_repr = 0;
2268         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2269                                           sfc_count_representors_cb,
2270                                           &nb_repr);
2271         if (rc != 0) {
2272                 sfc_adapter_unlock(sa);
2273                 SFC_ASSERT(rc > 0);
2274                 return -rc;
2275         }
2276
2277         if (info == NULL) {
2278                 sfc_adapter_unlock(sa);
2279                 return nb_repr;
2280         }
2281
2282         rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2283                                                &get_repr_ctx.controllers,
2284                                                &get_repr_ctx.nb_controllers);
2285         if (rc != 0) {
2286                 sfc_adapter_unlock(sa);
2287                 SFC_ASSERT(rc > 0);
2288                 return -rc;
2289         }
2290
2291         nic_cfg = efx_nic_cfg_get(sa->nic);
2292
2293         rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2294                                                   nic_cfg->enc_intf,
2295                                                   &controller);
2296         if (rc != 0) {
2297                 sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2298                 controller = -1;
2299         }
2300
2301         info->controller = controller;
2302         info->pf = nic_cfg->enc_pf;
2303
2304         get_repr_ctx.info = info;
2305         get_repr_ctx.sa = sa;
2306         get_repr_ctx.switch_domain_id = switch_domain_id;
2307         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2308                                           sfc_get_representors_cb,
2309                                           &get_repr_ctx);
2310         if (rc != 0) {
2311                 sfc_adapter_unlock(sa);
2312                 SFC_ASSERT(rc > 0);
2313                 return -rc;
2314         }
2315
2316         sfc_adapter_unlock(sa);
2317         return nb_repr;
2318 }
2319
2320 static int
2321 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2322 {
2323         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2324         uint64_t supported = 0;
2325
2326         sfc_adapter_lock(sa);
2327
2328         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2329                 supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2330
2331         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2332                 supported |= RTE_ETH_RX_METADATA_USER_MARK;
2333
2334         if (sfc_flow_tunnel_is_supported(sa))
2335                 supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2336
2337         sa->negotiated_rx_metadata = supported & *features;
2338         *features = sa->negotiated_rx_metadata;
2339
2340         sfc_adapter_unlock(sa);
2341
2342         return 0;
2343 }
2344
2345 static const struct eth_dev_ops sfc_eth_dev_ops = {
2346         .dev_configure                  = sfc_dev_configure,
2347         .dev_start                      = sfc_dev_start,
2348         .dev_stop                       = sfc_dev_stop,
2349         .dev_set_link_up                = sfc_dev_set_link_up,
2350         .dev_set_link_down              = sfc_dev_set_link_down,
2351         .dev_close                      = sfc_dev_close,
2352         .promiscuous_enable             = sfc_dev_promisc_enable,
2353         .promiscuous_disable            = sfc_dev_promisc_disable,
2354         .allmulticast_enable            = sfc_dev_allmulti_enable,
2355         .allmulticast_disable           = sfc_dev_allmulti_disable,
2356         .link_update                    = sfc_dev_link_update,
2357         .stats_get                      = sfc_stats_get,
2358         .stats_reset                    = sfc_stats_reset,
2359         .xstats_get                     = sfc_xstats_get,
2360         .xstats_reset                   = sfc_stats_reset,
2361         .xstats_get_names               = sfc_xstats_get_names,
2362         .dev_infos_get                  = sfc_dev_infos_get,
2363         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2364         .mtu_set                        = sfc_dev_set_mtu,
2365         .rx_queue_start                 = sfc_rx_queue_start,
2366         .rx_queue_stop                  = sfc_rx_queue_stop,
2367         .tx_queue_start                 = sfc_tx_queue_start,
2368         .tx_queue_stop                  = sfc_tx_queue_stop,
2369         .rx_queue_setup                 = sfc_rx_queue_setup,
2370         .rx_queue_release               = sfc_rx_queue_release,
2371         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
2372         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
2373         .tx_queue_setup                 = sfc_tx_queue_setup,
2374         .tx_queue_release               = sfc_tx_queue_release,
2375         .flow_ctrl_get                  = sfc_flow_ctrl_get,
2376         .flow_ctrl_set                  = sfc_flow_ctrl_set,
2377         .mac_addr_set                   = sfc_mac_addr_set,
2378         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
2379         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
2380         .reta_update                    = sfc_dev_rss_reta_update,
2381         .reta_query                     = sfc_dev_rss_reta_query,
2382         .rss_hash_update                = sfc_dev_rss_hash_update,
2383         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2384         .flow_ops_get                   = sfc_dev_flow_ops_get,
2385         .set_mc_addr_list               = sfc_set_mc_addr_list,
2386         .rxq_info_get                   = sfc_rx_queue_info_get,
2387         .txq_info_get                   = sfc_tx_queue_info_get,
2388         .fw_version_get                 = sfc_fw_version_get,
2389         .xstats_get_by_id               = sfc_xstats_get_by_id,
2390         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
2391         .pool_ops_supported             = sfc_pool_ops_supported,
2392         .representor_info_get           = sfc_representor_info_get,
2393         .rx_metadata_negotiate          = sfc_rx_metadata_negotiate,
2394 };
2395
2396 struct sfc_ethdev_init_data {
2397         uint16_t                nb_representors;
2398 };
2399
2400 /**
2401  * Duplicate a string in potentially shared memory required for
2402  * multi-process support.
2403  *
2404  * strdup() allocates from process-local heap/memory.
2405  */
2406 static char *
2407 sfc_strdup(const char *str)
2408 {
2409         size_t size;
2410         char *copy;
2411
2412         if (str == NULL)
2413                 return NULL;
2414
2415         size = strlen(str) + 1;
2416         copy = rte_malloc(__func__, size, 0);
2417         if (copy != NULL)
2418                 rte_memcpy(copy, str, size);
2419
2420         return copy;
2421 }
2422
2423 static int
2424 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2425 {
2426         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2427         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2428         const struct sfc_dp_rx *dp_rx;
2429         const struct sfc_dp_tx *dp_tx;
2430         const efx_nic_cfg_t *encp;
2431         unsigned int avail_caps = 0;
2432         const char *rx_name = NULL;
2433         const char *tx_name = NULL;
2434         int rc;
2435
2436         switch (sa->family) {
2437         case EFX_FAMILY_HUNTINGTON:
2438         case EFX_FAMILY_MEDFORD:
2439         case EFX_FAMILY_MEDFORD2:
2440                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2441                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2442                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2443                 break;
2444         case EFX_FAMILY_RIVERHEAD:
2445                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2446                 break;
2447         default:
2448                 break;
2449         }
2450
2451         encp = efx_nic_cfg_get(sa->nic);
2452         if (encp->enc_rx_es_super_buffer_supported)
2453                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2454
2455         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2456                                 sfc_kvarg_string_handler, &rx_name);
2457         if (rc != 0)
2458                 goto fail_kvarg_rx_datapath;
2459
2460         if (rx_name != NULL) {
2461                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2462                 if (dp_rx == NULL) {
2463                         sfc_err(sa, "Rx datapath %s not found", rx_name);
2464                         rc = ENOENT;
2465                         goto fail_dp_rx;
2466                 }
2467                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2468                         sfc_err(sa,
2469                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2470                                 rx_name);
2471                         rc = EINVAL;
2472                         goto fail_dp_rx_caps;
2473                 }
2474         } else {
2475                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2476                 if (dp_rx == NULL) {
2477                         sfc_err(sa, "Rx datapath by caps %#x not found",
2478                                 avail_caps);
2479                         rc = ENOENT;
2480                         goto fail_dp_rx;
2481                 }
2482         }
2483
2484         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2485         if (sas->dp_rx_name == NULL) {
2486                 rc = ENOMEM;
2487                 goto fail_dp_rx_name;
2488         }
2489
2490         if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2491                 /* FLAG and MARK are always available from Rx prefix. */
2492                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2493                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2494         }
2495
2496         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2497
2498         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2499                                 sfc_kvarg_string_handler, &tx_name);
2500         if (rc != 0)
2501                 goto fail_kvarg_tx_datapath;
2502
2503         if (tx_name != NULL) {
2504                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2505                 if (dp_tx == NULL) {
2506                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2507                         rc = ENOENT;
2508                         goto fail_dp_tx;
2509                 }
2510                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2511                         sfc_err(sa,
2512                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2513                                 tx_name);
2514                         rc = EINVAL;
2515                         goto fail_dp_tx_caps;
2516                 }
2517         } else {
2518                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2519                 if (dp_tx == NULL) {
2520                         sfc_err(sa, "Tx datapath by caps %#x not found",
2521                                 avail_caps);
2522                         rc = ENOENT;
2523                         goto fail_dp_tx;
2524                 }
2525         }
2526
2527         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2528         if (sas->dp_tx_name == NULL) {
2529                 rc = ENOMEM;
2530                 goto fail_dp_tx_name;
2531         }
2532
2533         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2534
2535         sa->priv.dp_rx = dp_rx;
2536         sa->priv.dp_tx = dp_tx;
2537
2538         dev->rx_pkt_burst = dp_rx->pkt_burst;
2539         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2540         dev->tx_pkt_burst = dp_tx->pkt_burst;
2541
2542         dev->rx_queue_count = sfc_rx_queue_count;
2543         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2544         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2545         dev->dev_ops = &sfc_eth_dev_ops;
2546
2547         return 0;
2548
2549 fail_dp_tx_name:
2550 fail_dp_tx_caps:
2551 fail_dp_tx:
2552 fail_kvarg_tx_datapath:
2553         rte_free(sas->dp_rx_name);
2554         sas->dp_rx_name = NULL;
2555
2556 fail_dp_rx_name:
2557 fail_dp_rx_caps:
2558 fail_dp_rx:
2559 fail_kvarg_rx_datapath:
2560         return rc;
2561 }
2562
2563 static void
2564 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2565 {
2566         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2567         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2568
2569         dev->dev_ops = NULL;
2570         dev->tx_pkt_prepare = NULL;
2571         dev->rx_pkt_burst = NULL;
2572         dev->tx_pkt_burst = NULL;
2573
2574         rte_free(sas->dp_tx_name);
2575         sas->dp_tx_name = NULL;
2576         sa->priv.dp_tx = NULL;
2577
2578         rte_free(sas->dp_rx_name);
2579         sas->dp_rx_name = NULL;
2580         sa->priv.dp_rx = NULL;
2581 }
2582
2583 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2584         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2585         .reta_query                     = sfc_dev_rss_reta_query,
2586         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2587         .rxq_info_get                   = sfc_rx_queue_info_get,
2588         .txq_info_get                   = sfc_tx_queue_info_get,
2589 };
2590
2591 static int
2592 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2593 {
2594         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2595         struct sfc_adapter_priv *sap;
2596         const struct sfc_dp_rx *dp_rx;
2597         const struct sfc_dp_tx *dp_tx;
2598         int rc;
2599
2600         /*
2601          * Allocate process private data from heap, since it should not
2602          * be located in shared memory allocated using rte_malloc() API.
2603          */
2604         sap = calloc(1, sizeof(*sap));
2605         if (sap == NULL) {
2606                 rc = ENOMEM;
2607                 goto fail_alloc_priv;
2608         }
2609
2610         sap->logtype_main = logtype_main;
2611
2612         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2613         if (dp_rx == NULL) {
2614                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2615                         "cannot find %s Rx datapath", sas->dp_rx_name);
2616                 rc = ENOENT;
2617                 goto fail_dp_rx;
2618         }
2619         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2620                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2621                         "%s Rx datapath does not support multi-process",
2622                         sas->dp_rx_name);
2623                 rc = EINVAL;
2624                 goto fail_dp_rx_multi_process;
2625         }
2626
2627         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2628         if (dp_tx == NULL) {
2629                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2630                         "cannot find %s Tx datapath", sas->dp_tx_name);
2631                 rc = ENOENT;
2632                 goto fail_dp_tx;
2633         }
2634         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2635                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2636                         "%s Tx datapath does not support multi-process",
2637                         sas->dp_tx_name);
2638                 rc = EINVAL;
2639                 goto fail_dp_tx_multi_process;
2640         }
2641
2642         sap->dp_rx = dp_rx;
2643         sap->dp_tx = dp_tx;
2644
2645         dev->process_private = sap;
2646         dev->rx_pkt_burst = dp_rx->pkt_burst;
2647         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2648         dev->tx_pkt_burst = dp_tx->pkt_burst;
2649         dev->rx_queue_count = sfc_rx_queue_count;
2650         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2651         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2652         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2653
2654         return 0;
2655
2656 fail_dp_tx_multi_process:
2657 fail_dp_tx:
2658 fail_dp_rx_multi_process:
2659 fail_dp_rx:
2660         free(sap);
2661
2662 fail_alloc_priv:
2663         return rc;
2664 }
2665
2666 static void
2667 sfc_register_dp(void)
2668 {
2669         /* Register once */
2670         if (TAILQ_EMPTY(&sfc_dp_head)) {
2671                 /* Prefer EF10 datapath */
2672                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2673                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2674                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2675                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2676
2677                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2678                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2679                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2680                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2681         }
2682 }
2683
2684 static int
2685 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
2686 {
2687         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2688         const char *switch_mode = NULL;
2689         int rc;
2690
2691         sfc_log_init(sa, "entry");
2692
2693         rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2694                                 sfc_kvarg_string_handler, &switch_mode);
2695         if (rc != 0)
2696                 goto fail_kvargs;
2697
2698         if (switch_mode == NULL) {
2699                 sa->switchdev = encp->enc_mae_supported &&
2700                                 (!encp->enc_datapath_cap_evb ||
2701                                  has_representors);
2702         } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2703                 sa->switchdev = false;
2704         } else if (strcasecmp(switch_mode,
2705                               SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2706                 sa->switchdev = true;
2707         } else {
2708                 sfc_err(sa, "invalid switch mode device argument '%s'",
2709                         switch_mode);
2710                 rc = EINVAL;
2711                 goto fail_mode;
2712         }
2713
2714         sfc_log_init(sa, "done");
2715
2716         return 0;
2717
2718 fail_mode:
2719 fail_kvargs:
2720         sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2721
2722         return rc;
2723 }
2724
2725 static int
2726 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
2727 {
2728         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2729         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2730         struct sfc_ethdev_init_data *init_data = init_params;
2731         uint32_t logtype_main;
2732         struct sfc_adapter *sa;
2733         int rc;
2734         const efx_nic_cfg_t *encp;
2735         const struct rte_ether_addr *from;
2736         int ret;
2737
2738         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2739                         SFC_EFX_DEV_CLASS_NET) {
2740                 SFC_GENERIC_LOG(DEBUG,
2741                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2742                 return 1;
2743         }
2744
2745         rc = sfc_dp_mport_register();
2746         if (rc != 0)
2747                 return rc;
2748
2749         sfc_register_dp();
2750
2751         logtype_main = sfc_register_logtype(&pci_dev->addr,
2752                                             SFC_LOGTYPE_MAIN_STR,
2753                                             RTE_LOG_NOTICE);
2754
2755         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2756                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2757
2758         /* Required for logging */
2759         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2760                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2761                         pci_dev->addr.domain, pci_dev->addr.bus,
2762                         pci_dev->addr.devid, pci_dev->addr.function,
2763                         dev->data->port_id);
2764         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2765                 SFC_GENERIC_LOG(ERR,
2766                         "reserved log prefix is too short for " PCI_PRI_FMT,
2767                         pci_dev->addr.domain, pci_dev->addr.bus,
2768                         pci_dev->addr.devid, pci_dev->addr.function);
2769                 return -EINVAL;
2770         }
2771         sas->pci_addr = pci_dev->addr;
2772         sas->port_id = dev->data->port_id;
2773
2774         /*
2775          * Allocate process private data from heap, since it should not
2776          * be located in shared memory allocated using rte_malloc() API.
2777          */
2778         sa = calloc(1, sizeof(*sa));
2779         if (sa == NULL) {
2780                 rc = ENOMEM;
2781                 goto fail_alloc_sa;
2782         }
2783
2784         dev->process_private = sa;
2785
2786         /* Required for logging */
2787         sa->priv.shared = sas;
2788         sa->priv.logtype_main = logtype_main;
2789
2790         sa->eth_dev = dev;
2791
2792         /* Copy PCI device info to the dev->data */
2793         rte_eth_copy_pci_info(dev, pci_dev);
2794         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2795
2796         rc = sfc_kvargs_parse(sa);
2797         if (rc != 0)
2798                 goto fail_kvargs_parse;
2799
2800         sfc_log_init(sa, "entry");
2801
2802         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2803         if (dev->data->mac_addrs == NULL) {
2804                 rc = ENOMEM;
2805                 goto fail_mac_addrs;
2806         }
2807
2808         sfc_adapter_lock_init(sa);
2809         sfc_adapter_lock(sa);
2810
2811         sfc_log_init(sa, "probing");
2812         rc = sfc_probe(sa);
2813         if (rc != 0)
2814                 goto fail_probe;
2815
2816         /*
2817          * Selecting a default switch mode requires the NIC to be probed and
2818          * to have its capabilities filled in.
2819          */
2820         rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
2821         if (rc != 0)
2822                 goto fail_switch_mode;
2823
2824         sfc_log_init(sa, "set device ops");
2825         rc = sfc_eth_dev_set_ops(dev);
2826         if (rc != 0)
2827                 goto fail_set_ops;
2828
2829         sfc_log_init(sa, "attaching");
2830         rc = sfc_attach(sa);
2831         if (rc != 0)
2832                 goto fail_attach;
2833
2834         if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_SUPPORTED) {
2835                 sfc_err(sa,
2836                         "failed to enable switchdev mode without MAE support");
2837                 rc = ENOTSUP;
2838                 goto fail_switchdev_no_mae;
2839         }
2840
2841         encp = efx_nic_cfg_get(sa->nic);
2842
2843         /*
2844          * The arguments are really reverse order in comparison to
2845          * Linux kernel. Copy from NIC config to Ethernet device data.
2846          */
2847         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2848         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2849
2850         sfc_adapter_unlock(sa);
2851
2852         sfc_log_init(sa, "done");
2853         return 0;
2854
2855 fail_switchdev_no_mae:
2856         sfc_detach(sa);
2857
2858 fail_attach:
2859         sfc_eth_dev_clear_ops(dev);
2860
2861 fail_set_ops:
2862 fail_switch_mode:
2863         sfc_unprobe(sa);
2864
2865 fail_probe:
2866         sfc_adapter_unlock(sa);
2867         sfc_adapter_lock_fini(sa);
2868         rte_free(dev->data->mac_addrs);
2869         dev->data->mac_addrs = NULL;
2870
2871 fail_mac_addrs:
2872         sfc_kvargs_cleanup(sa);
2873
2874 fail_kvargs_parse:
2875         sfc_log_init(sa, "failed %d", rc);
2876         dev->process_private = NULL;
2877         free(sa);
2878
2879 fail_alloc_sa:
2880         SFC_ASSERT(rc > 0);
2881         return -rc;
2882 }
2883
2884 static int
2885 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2886 {
2887         sfc_dev_close(dev);
2888
2889         return 0;
2890 }
2891
2892 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2893         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2894         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2895         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2896         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2897         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2898         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2899         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2900         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2901         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2902         { .vendor_id = 0 /* sentinel */ }
2903 };
2904
2905 static int
2906 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
2907 {
2908         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
2909         int rc;
2910
2911         if (args != NULL) {
2912                 rc = rte_eth_devargs_parse(args, &eth_da);
2913                 if (rc != 0) {
2914                         SFC_GENERIC_LOG(ERR,
2915                                         "Failed to parse generic devargs '%s'",
2916                                         args);
2917                         return rc;
2918                 }
2919         }
2920
2921         *devargs = eth_da;
2922
2923         return 0;
2924 }
2925
2926 static int
2927 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
2928                            struct sfc_ethdev_init_data *init_data,
2929                            struct rte_eth_dev **devp,
2930                            bool *dev_created)
2931 {
2932         struct rte_eth_dev *dev;
2933         bool created = false;
2934         int rc;
2935
2936         dev = rte_eth_dev_allocated(pci_dev->device.name);
2937         if (dev == NULL) {
2938                 rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
2939                                         sizeof(struct sfc_adapter_shared),
2940                                         eth_dev_pci_specific_init, pci_dev,
2941                                         sfc_eth_dev_init, init_data);
2942                 if (rc != 0) {
2943                         SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
2944                                         pci_dev->device.name);
2945                         return rc;
2946                 }
2947
2948                 created = true;
2949
2950                 dev = rte_eth_dev_allocated(pci_dev->device.name);
2951                 if (dev == NULL) {
2952                         SFC_GENERIC_LOG(ERR,
2953                                 "Failed to find allocated sfc ethdev '%s'",
2954                                 pci_dev->device.name);
2955                         return -ENODEV;
2956                 }
2957         }
2958
2959         *devp = dev;
2960         *dev_created = created;
2961
2962         return 0;
2963 }
2964
2965 static int
2966 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
2967                         efx_pcie_interface_t controller,
2968                         uint16_t port,
2969                         uint16_t repr_port,
2970                         enum rte_eth_representor_type type)
2971 {
2972         struct sfc_repr_entity_info entity;
2973         efx_mport_sel_t mport_sel;
2974         int rc;
2975
2976         switch (type) {
2977         case RTE_ETH_REPRESENTOR_NONE:
2978                 return 0;
2979         case RTE_ETH_REPRESENTOR_VF:
2980         case RTE_ETH_REPRESENTOR_PF:
2981                 break;
2982         case RTE_ETH_REPRESENTOR_SF:
2983                 sfc_err(sa, "SF representors are not supported");
2984                 return ENOTSUP;
2985         default:
2986                 sfc_err(sa, "unknown representor type: %d", type);
2987                 return ENOTSUP;
2988         }
2989
2990         rc = efx_mae_mport_by_pcie_mh_function(controller,
2991                                                port,
2992                                                repr_port,
2993                                                &mport_sel);
2994         if (rc != 0) {
2995                 sfc_err(sa,
2996                         "failed to get m-port selector for controller %u port %u repr_port %u: %s",
2997                         controller, port, repr_port, rte_strerror(-rc));
2998                 return rc;
2999         }
3000
3001         memset(&entity, 0, sizeof(entity));
3002         entity.type = type;
3003         entity.intf = controller;
3004         entity.pf = port;
3005         entity.vf = repr_port;
3006
3007         rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3008                              &mport_sel);
3009         if (rc != 0) {
3010                 sfc_err(sa,
3011                         "failed to create representor for controller %u port %u repr_port %u: %s",
3012                         controller, port, repr_port, rte_strerror(-rc));
3013                 return rc;
3014         }
3015
3016         return 0;
3017 }
3018
3019 static int
3020 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3021                              const struct rte_eth_devargs *eth_da,
3022                              efx_pcie_interface_t controller,
3023                              uint16_t port)
3024 {
3025         int first_error = 0;
3026         uint16_t i;
3027         int rc;
3028
3029         if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3030                 return sfc_eth_dev_create_repr(sa, controller, port,
3031                                                EFX_PCI_VF_INVALID,
3032                                                eth_da->type);
3033         }
3034
3035         for (i = 0; i < eth_da->nb_representor_ports; i++) {
3036                 rc = sfc_eth_dev_create_repr(sa, controller, port,
3037                                              eth_da->representor_ports[i],
3038                                              eth_da->type);
3039                 if (rc != 0 && first_error == 0)
3040                         first_error = rc;
3041         }
3042
3043         return first_error;
3044 }
3045
3046 static int
3047 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3048                                    const struct rte_eth_devargs *eth_da,
3049                                    efx_pcie_interface_t controller)
3050 {
3051         const efx_nic_cfg_t *encp;
3052         int first_error = 0;
3053         uint16_t default_port;
3054         uint16_t i;
3055         int rc;
3056
3057         if (eth_da->nb_ports == 0) {
3058                 encp = efx_nic_cfg_get(sa->nic);
3059                 default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3060                 return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3061                                                     default_port);
3062         }
3063
3064         for (i = 0; i < eth_da->nb_ports; i++) {
3065                 rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3066                                                   eth_da->ports[i]);
3067                 if (rc != 0 && first_error == 0)
3068                         first_error = rc;
3069         }
3070
3071         return first_error;
3072 }
3073
3074 static int
3075 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3076                                 const struct rte_eth_devargs *eth_da)
3077 {
3078         efx_pcie_interface_t intf;
3079         const efx_nic_cfg_t *encp;
3080         struct sfc_adapter *sa;
3081         uint16_t switch_domain_id;
3082         uint16_t i;
3083         int rc;
3084
3085         sa = sfc_adapter_by_eth_dev(dev);
3086         switch_domain_id = sa->mae.switch_domain_id;
3087
3088         switch (eth_da->type) {
3089         case RTE_ETH_REPRESENTOR_NONE:
3090                 return 0;
3091         case RTE_ETH_REPRESENTOR_PF:
3092         case RTE_ETH_REPRESENTOR_VF:
3093                 break;
3094         case RTE_ETH_REPRESENTOR_SF:
3095                 sfc_err(sa, "SF representors are not supported");
3096                 return -ENOTSUP;
3097         default:
3098                 sfc_err(sa, "unknown representor type: %d",
3099                         eth_da->type);
3100                 return -ENOTSUP;
3101         }
3102
3103         if (!sa->switchdev) {
3104                 sfc_err(sa, "cannot create representors in non-switchdev mode");
3105                 return -EINVAL;
3106         }
3107
3108         if (!sfc_repr_available(sfc_sa2shared(sa))) {
3109                 sfc_err(sa, "cannot create representors: unsupported");
3110
3111                 return -ENOTSUP;
3112         }
3113
3114         /*
3115          * This is needed to construct the DPDK controller -> EFX interface
3116          * mapping.
3117          */
3118         sfc_adapter_lock(sa);
3119         rc = sfc_process_mport_journal(sa);
3120         sfc_adapter_unlock(sa);
3121         if (rc != 0) {
3122                 SFC_ASSERT(rc > 0);
3123                 return -rc;
3124         }
3125
3126         if (eth_da->nb_mh_controllers > 0) {
3127                 for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3128                         rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3129                                                 eth_da->mh_controllers[i],
3130                                                 &intf);
3131                         if (rc != 0) {
3132                                 sfc_err(sa, "failed to get representor");
3133                                 continue;
3134                         }
3135                         sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3136                 }
3137         } else {
3138                 encp = efx_nic_cfg_get(sa->nic);
3139                 sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3140         }
3141
3142         return 0;
3143 }
3144
3145 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3146         struct rte_pci_device *pci_dev)
3147 {
3148         struct sfc_ethdev_init_data init_data;
3149         struct rte_eth_devargs eth_da;
3150         struct rte_eth_dev *dev;
3151         bool dev_created;
3152         int rc;
3153
3154         if (pci_dev->device.devargs != NULL) {
3155                 rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3156                                            &eth_da);
3157                 if (rc != 0)
3158                         return rc;
3159         } else {
3160                 memset(&eth_da, 0, sizeof(eth_da));
3161         }
3162
3163         /* If no VF representors specified, check for PF ones */
3164         if (eth_da.nb_representor_ports > 0)
3165                 init_data.nb_representors = eth_da.nb_representor_ports;
3166         else
3167                 init_data.nb_representors = eth_da.nb_ports;
3168
3169         if (init_data.nb_representors > 0 &&
3170             rte_eal_process_type() != RTE_PROC_PRIMARY) {
3171                 SFC_GENERIC_LOG(ERR,
3172                         "Create representors from secondary process not supported, dev '%s'",
3173                         pci_dev->device.name);
3174                 return -ENOTSUP;
3175         }
3176
3177         /*
3178          * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3179          * if it does not already exist. Re-probing an existing device is
3180          * expected to allow additional representors to be configured.
3181          */
3182         rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3183                                         &dev_created);
3184         if (rc != 0)
3185                 return rc;
3186
3187         rc = sfc_eth_dev_create_representors(dev, &eth_da);
3188         if (rc != 0) {
3189                 if (dev_created)
3190                         (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3191
3192                 return rc;
3193         }
3194
3195         return 0;
3196 }
3197
3198 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3199 {
3200         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3201 }
3202
3203 static struct rte_pci_driver sfc_efx_pmd = {
3204         .id_table = pci_id_sfc_efx_map,
3205         .drv_flags =
3206                 RTE_PCI_DRV_INTR_LSC |
3207                 RTE_PCI_DRV_NEED_MAPPING |
3208                 RTE_PCI_DRV_PROBE_AGAIN,
3209         .probe = sfc_eth_dev_pci_probe,
3210         .remove = sfc_eth_dev_pci_remove,
3211 };
3212
3213 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3214 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3215 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3216 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3217         SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3218         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3219         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3220         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3221         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3222         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3223         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3224
3225 RTE_INIT(sfc_driver_register_logtype)
3226 {
3227         int ret;
3228
3229         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
3230                                                    RTE_LOG_NOTICE);
3231         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
3232 }