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 #include "sfc_nic_dma.h"
  36
  37 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
  38 #define SFC_XSTAT_ID_INVALID_NAME '\0'
  39
  40 uint32_t sfc_logtype_driver;
  41
  42 static struct sfc_dp_list sfc_dp_head =
  43         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  44
  45
  46 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  47
  48
  49 static int
  50 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  51 {
  52         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  53         efx_nic_fw_info_t enfi;
  54         int ret;
  55         int rc;
  56
  57         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  58         if (rc != 0)
  59                 return -rc;
  60
  61         ret = snprintf(fw_version, fw_size,
  62                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  63                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  64                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  65         if (ret < 0)
  66                 return ret;
  67
  68         if (enfi.enfi_dpcpu_fw_ids_valid) {
  69                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  70                 int ret_extra;
  71
  72                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  73                                      fw_size - dpcpu_fw_ids_offset,
  74                                      " rx%" PRIx16 " tx%" PRIx16,
  75                                      enfi.enfi_rx_dpcpu_fw_id,
  76                                      enfi.enfi_tx_dpcpu_fw_id);
  77                 if (ret_extra < 0)
  78                         return ret_extra;
  79
  80                 ret += ret_extra;
  81         }
  82
  83         if (fw_size < (size_t)(++ret))
  84                 return ret;
  85         else
  86                 return 0;
  87 }
  88
  89 static int
  90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  91 {
  92         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  93         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  94         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  95         struct sfc_rss *rss = &sas->rss;
  96         struct sfc_mae *mae = &sa->mae;
  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 = RTE_ETH_LINK_SPEED_FIXED;
 109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 110                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_1G;
 111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 112                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_10G;
 113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 114                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_25G;
 115         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 116                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_40G;
 117         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 118                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_50G;
 119         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 120                 dev_info->speed_capa |= RTE_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 (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 149                 uint64_t rte_hf = 0;
 150                 unsigned int i;
 151
 152                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 153                         rte_hf |= rss->hf_map[i].rte;
 154
 155                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 156                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 157                 dev_info->flow_type_rss_offloads = rte_hf;
 158         }
 159
 160         /* Initialize to hardware limits */
 161         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 162         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 163         /* The RXQ hardware requires that the descriptor count is a power
 164          * of 2, but rx_desc_lim cannot properly describe that constraint.
 165          */
 166         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 167
 168         /* Initialize to hardware limits */
 169         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 170         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 171         /*
 172          * The TXQ hardware requires that the descriptor count is a power
 173          * of 2, but tx_desc_lim cannot properly describe that constraint
 174          */
 175         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 176
 177         if (sap->dp_rx->get_dev_info != NULL)
 178                 sap->dp_rx->get_dev_info(dev_info);
 179         if (sap->dp_tx->get_dev_info != NULL)
 180                 sap->dp_tx->get_dev_info(dev_info);
 181
 182         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 183                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 184         dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
 185
 186         if (mae->status == SFC_MAE_STATUS_SUPPORTED ||
 187             mae->status == SFC_MAE_STATUS_ADMIN) {
 188                 dev_info->switch_info.name = dev->device->driver->name;
 189                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 190                 dev_info->switch_info.port_id = mae->switch_port_id;
 191         }
 192
 193         return 0;
 194 }
 195
 196 static const uint32_t *
 197 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 198 {
 199         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 200
 201         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 202 }
 203
 204 static int
 205 sfc_dev_configure(struct rte_eth_dev *dev)
 206 {
 207         struct rte_eth_dev_data *dev_data = dev->data;
 208         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 209         int rc;
 210
 211         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 212                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 213
 214         sfc_adapter_lock(sa);
 215         switch (sa->state) {
 216         case SFC_ETHDEV_CONFIGURED:
 217                 /* FALLTHROUGH */
 218         case SFC_ETHDEV_INITIALIZED:
 219                 rc = sfc_configure(sa);
 220                 break;
 221         default:
 222                 sfc_err(sa, "unexpected adapter state %u to configure",
 223                         sa->state);
 224                 rc = EINVAL;
 225                 break;
 226         }
 227         sfc_adapter_unlock(sa);
 228
 229         sfc_log_init(sa, "done %d", rc);
 230         SFC_ASSERT(rc >= 0);
 231         return -rc;
 232 }
 233
 234 static int
 235 sfc_dev_start(struct rte_eth_dev *dev)
 236 {
 237         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 238         int rc;
 239
 240         sfc_log_init(sa, "entry");
 241
 242         sfc_adapter_lock(sa);
 243         rc = sfc_start(sa);
 244         sfc_adapter_unlock(sa);
 245
 246         sfc_log_init(sa, "done %d", rc);
 247         SFC_ASSERT(rc >= 0);
 248         return -rc;
 249 }
 250
 251 static int
 252 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 253 {
 254         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 255         struct rte_eth_link current_link;
 256         int ret;
 257
 258         sfc_log_init(sa, "entry");
 259
 260         if (sa->state != SFC_ETHDEV_STARTED) {
 261                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 262         } else if (wait_to_complete) {
 263                 efx_link_mode_t link_mode;
 264
 265                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 266                         link_mode = EFX_LINK_UNKNOWN;
 267                 sfc_port_link_mode_to_info(link_mode, &current_link);
 268
 269         } else {
 270                 sfc_ev_mgmt_qpoll(sa);
 271                 rte_eth_linkstatus_get(dev, &current_link);
 272         }
 273
 274         ret = rte_eth_linkstatus_set(dev, &current_link);
 275         if (ret == 0)
 276                 sfc_notice(sa, "Link status is %s",
 277                            current_link.link_status ? "UP" : "DOWN");
 278
 279         return ret;
 280 }
 281
 282 static int
 283 sfc_dev_stop(struct rte_eth_dev *dev)
 284 {
 285         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 286
 287         sfc_log_init(sa, "entry");
 288
 289         sfc_adapter_lock(sa);
 290         sfc_stop(sa);
 291         sfc_adapter_unlock(sa);
 292
 293         sfc_log_init(sa, "done");
 294
 295         return 0;
 296 }
 297
 298 static int
 299 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 300 {
 301         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 302         int rc;
 303
 304         sfc_log_init(sa, "entry");
 305
 306         sfc_adapter_lock(sa);
 307         rc = sfc_start(sa);
 308         sfc_adapter_unlock(sa);
 309
 310         SFC_ASSERT(rc >= 0);
 311         return -rc;
 312 }
 313
 314 static int
 315 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 316 {
 317         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 318
 319         sfc_log_init(sa, "entry");
 320
 321         sfc_adapter_lock(sa);
 322         sfc_stop(sa);
 323         sfc_adapter_unlock(sa);
 324
 325         return 0;
 326 }
 327
 328 static void
 329 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 330 {
 331         free(dev->process_private);
 332         rte_eth_dev_release_port(dev);
 333 }
 334
 335 static int
 336 sfc_dev_close(struct rte_eth_dev *dev)
 337 {
 338         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 339
 340         sfc_log_init(sa, "entry");
 341
 342         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 343                 sfc_eth_dev_secondary_clear_ops(dev);
 344                 return 0;
 345         }
 346
 347         sfc_pre_detach(sa);
 348
 349         sfc_adapter_lock(sa);
 350         switch (sa->state) {
 351         case SFC_ETHDEV_STARTED:
 352                 sfc_stop(sa);
 353                 SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
 354                 /* FALLTHROUGH */
 355         case SFC_ETHDEV_CONFIGURED:
 356                 sfc_close(sa);
 357                 SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
 358                 /* FALLTHROUGH */
 359         case SFC_ETHDEV_INITIALIZED:
 360                 break;
 361         default:
 362                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 363                 break;
 364         }
 365
 366         /*
 367          * Cleanup all resources.
 368          * Rollback primary process sfc_eth_dev_init() below.
 369          */
 370
 371         sfc_eth_dev_clear_ops(dev);
 372
 373         sfc_nic_dma_detach(sa);
 374         sfc_detach(sa);
 375         sfc_unprobe(sa);
 376
 377         sfc_kvargs_cleanup(sa);
 378
 379         sfc_adapter_unlock(sa);
 380         sfc_adapter_lock_fini(sa);
 381
 382         sfc_log_init(sa, "done");
 383
 384         /* Required for logging, so cleanup last */
 385         sa->eth_dev = NULL;
 386
 387         free(sa);
 388
 389         return 0;
 390 }
 391
 392 static int
 393 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 394                    boolean_t enabled)
 395 {
 396         struct sfc_port *port;
 397         boolean_t *toggle;
 398         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 399         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 400         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 401         int rc = 0;
 402
 403         sfc_adapter_lock(sa);
 404
 405         port = &sa->port;
 406         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 407
 408         if (*toggle != enabled) {
 409                 *toggle = enabled;
 410
 411                 if (sfc_sa2shared(sa)->isolated) {
 412                         sfc_warn(sa, "isolated mode is active on the port");
 413                         sfc_warn(sa, "the change is to be applied on the next "
 414                                      "start provided that isolated mode is "
 415                                      "disabled prior the next start");
 416                 } else if ((sa->state == SFC_ETHDEV_STARTED) &&
 417                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 418                         *toggle = !(enabled);
 419                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 420                                  ((enabled) ? "enable" : "disable"), desc, rc);
 421
 422                         /*
 423                          * For promiscuous and all-multicast filters a
 424                          * permission failure should be reported as an
 425                          * unsupported filter.
 426                          */
 427                         if (rc == EPERM)
 428                                 rc = ENOTSUP;
 429                 }
 430         }
 431
 432         sfc_adapter_unlock(sa);
 433         return rc;
 434 }
 435
 436 static int
 437 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 438 {
 439         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 440
 441         SFC_ASSERT(rc >= 0);
 442         return -rc;
 443 }
 444
 445 static int
 446 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 447 {
 448         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 449
 450         SFC_ASSERT(rc >= 0);
 451         return -rc;
 452 }
 453
 454 static int
 455 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 456 {
 457         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 458
 459         SFC_ASSERT(rc >= 0);
 460         return -rc;
 461 }
 462
 463 static int
 464 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 465 {
 466         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 467
 468         SFC_ASSERT(rc >= 0);
 469         return -rc;
 470 }
 471
 472 static int
 473 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 474                    uint16_t nb_rx_desc, unsigned int socket_id,
 475                    const struct rte_eth_rxconf *rx_conf,
 476                    struct rte_mempool *mb_pool)
 477 {
 478         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 479         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 480         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
 481         struct sfc_rxq_info *rxq_info;
 482         sfc_sw_index_t sw_index;
 483         int rc;
 484
 485         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 486                      ethdev_qid, nb_rx_desc, socket_id);
 487
 488         sfc_adapter_lock(sa);
 489
 490         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
 491         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
 492                           rx_conf, mb_pool);
 493         if (rc != 0)
 494                 goto fail_rx_qinit;
 495
 496         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
 497         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
 498
 499         sfc_adapter_unlock(sa);
 500
 501         return 0;
 502
 503 fail_rx_qinit:
 504         sfc_adapter_unlock(sa);
 505         SFC_ASSERT(rc > 0);
 506         return -rc;
 507 }
 508
 509 static void
 510 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 511 {
 512         struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
 513         struct sfc_rxq *rxq;
 514         struct sfc_adapter *sa;
 515         sfc_sw_index_t sw_index;
 516
 517         if (dp_rxq == NULL)
 518                 return;
 519
 520         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 521         sa = rxq->evq->sa;
 522         sfc_adapter_lock(sa);
 523
 524         sw_index = dp_rxq->dpq.queue_id;
 525
 526         sfc_log_init(sa, "RxQ=%u", sw_index);
 527
 528         sfc_rx_qfini(sa, sw_index);
 529
 530         sfc_adapter_unlock(sa);
 531 }
 532
 533 static int
 534 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 535                    uint16_t nb_tx_desc, unsigned int socket_id,
 536                    const struct rte_eth_txconf *tx_conf)
 537 {
 538         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 539         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 540         struct sfc_txq_info *txq_info;
 541         sfc_sw_index_t sw_index;
 542         int rc;
 543
 544         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 545                      ethdev_qid, nb_tx_desc, socket_id);
 546
 547         sfc_adapter_lock(sa);
 548
 549         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
 550         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
 551         if (rc != 0)
 552                 goto fail_tx_qinit;
 553
 554         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
 555         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
 556
 557         sfc_adapter_unlock(sa);
 558         return 0;
 559
 560 fail_tx_qinit:
 561         sfc_adapter_unlock(sa);
 562         SFC_ASSERT(rc > 0);
 563         return -rc;
 564 }
 565
 566 static void
 567 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 568 {
 569         struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
 570         struct sfc_txq *txq;
 571         sfc_sw_index_t sw_index;
 572         struct sfc_adapter *sa;
 573
 574         if (dp_txq == NULL)
 575                 return;
 576
 577         txq = sfc_txq_by_dp_txq(dp_txq);
 578         sw_index = dp_txq->dpq.queue_id;
 579
 580         SFC_ASSERT(txq->evq != NULL);
 581         sa = txq->evq->sa;
 582
 583         sfc_log_init(sa, "TxQ = %u", sw_index);
 584
 585         sfc_adapter_lock(sa);
 586
 587         sfc_tx_qfini(sa, sw_index);
 588
 589         sfc_adapter_unlock(sa);
 590 }
 591
 592 static void
 593 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 594 {
 595         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 596         uint64_t pkts_sum = 0;
 597         uint64_t bytes_sum = 0;
 598         unsigned int i;
 599
 600         for (i = 0; i < sas->ethdev_rxq_count; ++i) {
 601                 struct sfc_rxq_info *rxq_info;
 602
 603                 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
 604                 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
 605                         union sfc_pkts_bytes qstats;
 606
 607                         sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
 608                         pkts_sum += qstats.pkts -
 609                                         sa->sw_stats.reset_rx_pkts[i];
 610                         bytes_sum += qstats.bytes -
 611                                         sa->sw_stats.reset_rx_bytes[i];
 612                 }
 613         }
 614
 615         *pkts = pkts_sum;
 616         *bytes = bytes_sum;
 617 }
 618
 619 static void
 620 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 621 {
 622         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 623         uint64_t pkts_sum = 0;
 624         uint64_t bytes_sum = 0;
 625         unsigned int i;
 626
 627         for (i = 0; i < sas->ethdev_txq_count; ++i) {
 628                 struct sfc_txq_info *txq_info;
 629
 630                 txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
 631                 if (txq_info->state & SFC_TXQ_INITIALIZED) {
 632                         union sfc_pkts_bytes qstats;
 633
 634                         sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
 635                         pkts_sum += qstats.pkts -
 636                                         sa->sw_stats.reset_tx_pkts[i];
 637                         bytes_sum += qstats.bytes -
 638                                         sa->sw_stats.reset_tx_bytes[i];
 639                 }
 640         }
 641
 642         *pkts = pkts_sum;
 643         *bytes = bytes_sum;
 644 }
 645
 646 /*
 647  * Some statistics are computed as A - B where A and B each increase
 648  * monotonically with some hardware counter(s) and the counters are read
 649  * asynchronously.
 650  *
 651  * If packet X is counted in A, but not counted in B yet, computed value is
 652  * greater than real.
 653  *
 654  * If packet X is not counted in A at the moment of reading the counter,
 655  * but counted in B at the moment of reading the counter, computed value
 656  * is less than real.
 657  *
 658  * However, counter which grows backward is worse evil than slightly wrong
 659  * value. So, let's try to guarantee that it never happens except may be
 660  * the case when the MAC stats are zeroed as a result of a NIC reset.
 661  */
 662 static void
 663 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 664 {
 665         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 666                 *stat = newval;
 667 }
 668
 669 static int
 670 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 671 {
 672         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 673         bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
 674         bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
 675         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 676         struct sfc_port *port = &sa->port;
 677         uint64_t *mac_stats;
 678         int ret;
 679
 680         sfc_adapter_lock(sa);
 681
 682         if (have_dp_rx_stats)
 683                 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
 684         if (have_dp_tx_stats)
 685                 sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
 686
 687         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 688         if (ret != 0)
 689                 goto unlock;
 690
 691         mac_stats = port->mac_stats_buf;
 692
 693         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 694                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 695                 if (!have_dp_rx_stats) {
 696                         stats->ipackets =
 697                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 698                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 699                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 700                         stats->ibytes =
 701                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 702                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 703                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 704
 705                         /* CRC is included in these stats, but shouldn't be */
 706                         stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 707                 }
 708                 if (!have_dp_tx_stats) {
 709                         stats->opackets =
 710                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 711                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 712                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 713                         stats->obytes =
 714                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 715                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 716                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 717
 718                         /* CRC is included in these stats, but shouldn't be */
 719                         stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 720                 }
 721                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 722                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 723         } else {
 724                 if (!have_dp_tx_stats) {
 725                         stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 726                         stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
 727                                 mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 728                 }
 729
 730                 /*
 731                  * Take into account stats which are whenever supported
 732                  * on EF10. If some stat is not supported by current
 733                  * firmware variant or HW revision, it is guaranteed
 734                  * to be zero in mac_stats.
 735                  */
 736                 stats->imissed =
 737                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 738                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 739                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 740                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 741                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 742                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 743                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 744                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 745                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 746                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 747                 stats->ierrors =
 748                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 749                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 750                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 751                 /* no oerrors counters supported on EF10 */
 752
 753                 if (!have_dp_rx_stats) {
 754                         /* Exclude missed, errors and pauses from Rx packets */
 755                         sfc_update_diff_stat(&port->ipackets,
 756                                 mac_stats[EFX_MAC_RX_PKTS] -
 757                                 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 758                                 stats->imissed - stats->ierrors);
 759                         stats->ipackets = port->ipackets;
 760                         stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
 761                                 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 762                 }
 763         }
 764
 765 unlock:
 766         sfc_adapter_unlock(sa);
 767         SFC_ASSERT(ret >= 0);
 768         return -ret;
 769 }
 770
 771 static int
 772 sfc_stats_reset(struct rte_eth_dev *dev)
 773 {
 774         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 775         struct sfc_port *port = &sa->port;
 776         int rc;
 777
 778         sfc_adapter_lock(sa);
 779
 780         if (sa->state != SFC_ETHDEV_STARTED) {
 781                 /*
 782                  * The operation cannot be done if port is not started; it
 783                  * will be scheduled to be done during the next port start
 784                  */
 785                 port->mac_stats_reset_pending = B_TRUE;
 786                 sfc_adapter_unlock(sa);
 787                 return 0;
 788         }
 789
 790         rc = sfc_port_reset_mac_stats(sa);
 791         if (rc != 0)
 792                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 793
 794         sfc_sw_xstats_reset(sa);
 795
 796         sfc_adapter_unlock(sa);
 797
 798         SFC_ASSERT(rc >= 0);
 799         return -rc;
 800 }
 801
 802 static unsigned int
 803 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 804 {
 805         struct sfc_port *port = &sa->port;
 806         unsigned int nb_supported;
 807
 808         sfc_adapter_lock(sa);
 809         nb_supported = port->mac_stats_nb_supported +
 810                        sfc_sw_xstats_get_nb_supported(sa);
 811         sfc_adapter_unlock(sa);
 812
 813         return nb_supported;
 814 }
 815
 816 static int
 817 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 818                unsigned int xstats_count)
 819 {
 820         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 821         unsigned int nb_written = 0;
 822         unsigned int nb_supported = 0;
 823         int rc;
 824
 825         if (unlikely(xstats == NULL))
 826                 return sfc_xstats_get_nb_supported(sa);
 827
 828         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 829         if (rc < 0)
 830                 return rc;
 831
 832         nb_supported = rc;
 833         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 834                                &nb_supported);
 835
 836         return nb_supported;
 837 }
 838
 839 static int
 840 sfc_xstats_get_names(struct rte_eth_dev *dev,
 841                      struct rte_eth_xstat_name *xstats_names,
 842                      unsigned int xstats_count)
 843 {
 844         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 845         struct sfc_port *port = &sa->port;
 846         unsigned int i;
 847         unsigned int nstats = 0;
 848         unsigned int nb_written = 0;
 849         int ret;
 850
 851         if (unlikely(xstats_names == NULL))
 852                 return sfc_xstats_get_nb_supported(sa);
 853
 854         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 855                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 856                         if (nstats < xstats_count) {
 857                                 strlcpy(xstats_names[nstats].name,
 858                                         efx_mac_stat_name(sa->nic, i),
 859                                         sizeof(xstats_names[0].name));
 860                                 nb_written++;
 861                         }
 862                         nstats++;
 863                 }
 864         }
 865
 866         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 867                                       &nb_written, &nstats);
 868         if (ret != 0) {
 869                 SFC_ASSERT(ret < 0);
 870                 return ret;
 871         }
 872
 873         return nstats;
 874 }
 875
 876 static int
 877 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 878                      uint64_t *values, unsigned int n)
 879 {
 880         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 881         struct sfc_port *port = &sa->port;
 882         unsigned int nb_supported;
 883         unsigned int i;
 884         int rc;
 885
 886         if (unlikely(ids == NULL || values == NULL))
 887                 return -EINVAL;
 888
 889         /*
 890          * Values array could be filled in nonsequential order. Fill values with
 891          * constant indicating invalid ID first.
 892          */
 893         for (i = 0; i < n; i++)
 894                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 895
 896         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 897         if (rc != 0)
 898                 return rc;
 899
 900         nb_supported = port->mac_stats_nb_supported;
 901         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 902
 903         /* Return number of written stats before invalid ID is encountered. */
 904         for (i = 0; i < n; i++) {
 905                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 906                         return i;
 907         }
 908
 909         return n;
 910 }
 911
 912 static int
 913 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 914                            const uint64_t *ids,
 915                            struct rte_eth_xstat_name *xstats_names,
 916                            unsigned int size)
 917 {
 918         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 919         struct sfc_port *port = &sa->port;
 920         unsigned int nb_supported;
 921         unsigned int i;
 922         int ret;
 923
 924         if (unlikely(xstats_names == NULL && ids != NULL) ||
 925             unlikely(xstats_names != NULL && ids == NULL))
 926                 return -EINVAL;
 927
 928         if (unlikely(xstats_names == NULL && ids == NULL))
 929                 return sfc_xstats_get_nb_supported(sa);
 930
 931         /*
 932          * Names array could be filled in nonsequential order. Fill names with
 933          * string indicating invalid ID first.
 934          */
 935         for (i = 0; i < size; i++)
 936                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 937
 938         sfc_adapter_lock(sa);
 939
 940         SFC_ASSERT(port->mac_stats_nb_supported <=
 941                    RTE_DIM(port->mac_stats_by_id));
 942
 943         for (i = 0; i < size; i++) {
 944                 if (ids[i] < port->mac_stats_nb_supported) {
 945                         strlcpy(xstats_names[i].name,
 946                                 efx_mac_stat_name(sa->nic,
 947                                                  port->mac_stats_by_id[ids[i]]),
 948                                 sizeof(xstats_names[0].name));
 949                 }
 950         }
 951
 952         nb_supported = port->mac_stats_nb_supported;
 953
 954         sfc_adapter_unlock(sa);
 955
 956         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 957                                             &nb_supported);
 958         if (ret != 0) {
 959                 SFC_ASSERT(ret < 0);
 960                 return ret;
 961         }
 962
 963         /* Return number of written names before invalid ID is encountered. */
 964         for (i = 0; i < size; i++) {
 965                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 966                         return i;
 967         }
 968
 969         return size;
 970 }
 971
 972 static int
 973 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 974 {
 975         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 976         unsigned int wanted_fc, link_fc;
 977
 978         memset(fc_conf, 0, sizeof(*fc_conf));
 979
 980         sfc_adapter_lock(sa);
 981
 982         if (sa->state == SFC_ETHDEV_STARTED)
 983                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 984         else
 985                 link_fc = sa->port.flow_ctrl;
 986
 987         switch (link_fc) {
 988         case 0:
 989                 fc_conf->mode = RTE_ETH_FC_NONE;
 990                 break;
 991         case EFX_FCNTL_RESPOND:
 992                 fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
 993                 break;
 994         case EFX_FCNTL_GENERATE:
 995                 fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
 996                 break;
 997         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 998                 fc_conf->mode = RTE_ETH_FC_FULL;
 999                 break;
1000         default:
1001                 sfc_err(sa, "%s: unexpected flow control value %#x",
1002                         __func__, link_fc);
1003         }
1004
1005         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1006
1007         sfc_adapter_unlock(sa);
1008
1009         return 0;
1010 }
1011
1012 static int
1013 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1014 {
1015         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1016         struct sfc_port *port = &sa->port;
1017         unsigned int fcntl;
1018         int rc;
1019
1020         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1021             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1022             fc_conf->mac_ctrl_frame_fwd != 0) {
1023                 sfc_err(sa, "unsupported flow control settings specified");
1024                 rc = EINVAL;
1025                 goto fail_inval;
1026         }
1027
1028         switch (fc_conf->mode) {
1029         case RTE_ETH_FC_NONE:
1030                 fcntl = 0;
1031                 break;
1032         case RTE_ETH_FC_RX_PAUSE:
1033                 fcntl = EFX_FCNTL_RESPOND;
1034                 break;
1035         case RTE_ETH_FC_TX_PAUSE:
1036                 fcntl = EFX_FCNTL_GENERATE;
1037                 break;
1038         case RTE_ETH_FC_FULL:
1039                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1040                 break;
1041         default:
1042                 rc = EINVAL;
1043                 goto fail_inval;
1044         }
1045
1046         sfc_adapter_lock(sa);
1047
1048         if (sa->state == SFC_ETHDEV_STARTED) {
1049                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1050                 if (rc != 0)
1051                         goto fail_mac_fcntl_set;
1052         }
1053
1054         port->flow_ctrl = fcntl;
1055         port->flow_ctrl_autoneg = fc_conf->autoneg;
1056
1057         sfc_adapter_unlock(sa);
1058
1059         return 0;
1060
1061 fail_mac_fcntl_set:
1062         sfc_adapter_unlock(sa);
1063 fail_inval:
1064         SFC_ASSERT(rc > 0);
1065         return -rc;
1066 }
1067
1068 static int
1069 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1070 {
1071         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1072         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1073         boolean_t scatter_enabled;
1074         const char *error;
1075         unsigned int i;
1076
1077         for (i = 0; i < sas->rxq_count; i++) {
1078                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1079                         continue;
1080
1081                 scatter_enabled = (sas->rxq_info[i].type_flags &
1082                                    EFX_RXQ_FLAG_SCATTER);
1083
1084                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1085                                           encp->enc_rx_prefix_size,
1086                                           scatter_enabled,
1087                                           encp->enc_rx_scatter_max, &error)) {
1088                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1089                                 error);
1090                         return EINVAL;
1091                 }
1092         }
1093
1094         return 0;
1095 }
1096
1097 static int
1098 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1099 {
1100         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1101         size_t pdu = EFX_MAC_PDU(mtu);
1102         size_t old_pdu;
1103         int rc;
1104
1105         sfc_log_init(sa, "mtu=%u", mtu);
1106
1107         rc = EINVAL;
1108         if (pdu < EFX_MAC_PDU_MIN) {
1109                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1110                         (unsigned int)mtu, (unsigned int)pdu,
1111                         EFX_MAC_PDU_MIN);
1112                 goto fail_inval;
1113         }
1114         if (pdu > EFX_MAC_PDU_MAX) {
1115                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1116                         (unsigned int)mtu, (unsigned int)pdu,
1117                         (unsigned int)EFX_MAC_PDU_MAX);
1118                 goto fail_inval;
1119         }
1120
1121         sfc_adapter_lock(sa);
1122
1123         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1124         if (rc != 0)
1125                 goto fail_check_scatter;
1126
1127         if (pdu != sa->port.pdu) {
1128                 if (sa->state == SFC_ETHDEV_STARTED) {
1129                         sfc_stop(sa);
1130
1131                         old_pdu = sa->port.pdu;
1132                         sa->port.pdu = pdu;
1133                         rc = sfc_start(sa);
1134                         if (rc != 0)
1135                                 goto fail_start;
1136                 } else {
1137                         sa->port.pdu = pdu;
1138                 }
1139         }
1140
1141         sfc_adapter_unlock(sa);
1142
1143         sfc_log_init(sa, "done");
1144         return 0;
1145
1146 fail_start:
1147         sa->port.pdu = old_pdu;
1148         if (sfc_start(sa) != 0)
1149                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1150                         "PDU max size - port is stopped",
1151                         (unsigned int)pdu, (unsigned int)old_pdu);
1152
1153 fail_check_scatter:
1154         sfc_adapter_unlock(sa);
1155
1156 fail_inval:
1157         sfc_log_init(sa, "failed %d", rc);
1158         SFC_ASSERT(rc > 0);
1159         return -rc;
1160 }
1161 static int
1162 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1163 {
1164         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1165         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1166         struct sfc_port *port = &sa->port;
1167         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1168         int rc = 0;
1169
1170         sfc_adapter_lock(sa);
1171
1172         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1173                 goto unlock;
1174
1175         /*
1176          * Copy the address to the device private data so that
1177          * it could be recalled in the case of adapter restart.
1178          */
1179         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1180
1181         /*
1182          * Neither of the two following checks can return
1183          * an error. The new MAC address is preserved in
1184          * the device private data and can be activated
1185          * on the next port start if the user prevents
1186          * isolated mode from being enabled.
1187          */
1188         if (sfc_sa2shared(sa)->isolated) {
1189                 sfc_warn(sa, "isolated mode is active on the port");
1190                 sfc_warn(sa, "will not set MAC address");
1191                 goto unlock;
1192         }
1193
1194         if (sa->state != SFC_ETHDEV_STARTED) {
1195                 sfc_notice(sa, "the port is not started");
1196                 sfc_notice(sa, "the new MAC address will be set on port start");
1197
1198                 goto unlock;
1199         }
1200
1201         if (encp->enc_allow_set_mac_with_installed_filters) {
1202                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1203                 if (rc != 0) {
1204                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1205                         goto unlock;
1206                 }
1207
1208                 /*
1209                  * Changing the MAC address by means of MCDI request
1210                  * has no effect on received traffic, therefore
1211                  * we also need to update unicast filters
1212                  */
1213                 rc = sfc_set_rx_mode_unchecked(sa);
1214                 if (rc != 0) {
1215                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1216                         /* Rollback the old address */
1217                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1218                         (void)sfc_set_rx_mode_unchecked(sa);
1219                 }
1220         } else {
1221                 sfc_warn(sa, "cannot set MAC address with filters installed");
1222                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1223                 sfc_warn(sa, "(some traffic may be dropped)");
1224
1225                 /*
1226                  * Since setting MAC address with filters installed is not
1227                  * allowed on the adapter, the new MAC address will be set
1228                  * by means of adapter restart. sfc_start() shall retrieve
1229                  * the new address from the device private data and set it.
1230                  */
1231                 sfc_stop(sa);
1232                 rc = sfc_start(sa);
1233                 if (rc != 0)
1234                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1235         }
1236
1237 unlock:
1238         if (rc != 0)
1239                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1240
1241         sfc_adapter_unlock(sa);
1242
1243         SFC_ASSERT(rc >= 0);
1244         return -rc;
1245 }
1246
1247
1248 static int
1249 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1250                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1251 {
1252         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1253         struct sfc_port *port = &sa->port;
1254         uint8_t *mc_addrs = port->mcast_addrs;
1255         int rc;
1256         unsigned int i;
1257
1258         if (sfc_sa2shared(sa)->isolated) {
1259                 sfc_err(sa, "isolated mode is active on the port");
1260                 sfc_err(sa, "will not set multicast address list");
1261                 return -ENOTSUP;
1262         }
1263
1264         if (mc_addrs == NULL)
1265                 return -ENOBUFS;
1266
1267         if (nb_mc_addr > port->max_mcast_addrs) {
1268                 sfc_err(sa, "too many multicast addresses: %u > %u",
1269                          nb_mc_addr, port->max_mcast_addrs);
1270                 return -EINVAL;
1271         }
1272
1273         for (i = 0; i < nb_mc_addr; ++i) {
1274                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1275                                  EFX_MAC_ADDR_LEN);
1276                 mc_addrs += EFX_MAC_ADDR_LEN;
1277         }
1278
1279         port->nb_mcast_addrs = nb_mc_addr;
1280
1281         if (sa->state != SFC_ETHDEV_STARTED)
1282                 return 0;
1283
1284         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1285                                         port->nb_mcast_addrs);
1286         if (rc != 0)
1287                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1288
1289         SFC_ASSERT(rc >= 0);
1290         return -rc;
1291 }
1292
1293 /*
1294  * The function is used by the secondary process as well. It must not
1295  * use any process-local pointers from the adapter data.
1296  */
1297 static void
1298 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1299                       struct rte_eth_rxq_info *qinfo)
1300 {
1301         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1302         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1303         struct sfc_rxq_info *rxq_info;
1304
1305         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1306
1307         qinfo->mp = rxq_info->refill_mb_pool;
1308         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1309         qinfo->conf.rx_drop_en = 1;
1310         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1311         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1312         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1313                 qinfo->conf.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
1314                 qinfo->scattered_rx = 1;
1315         }
1316         qinfo->nb_desc = rxq_info->entries;
1317 }
1318
1319 /*
1320  * The function is used by the secondary process as well. It must not
1321  * use any process-local pointers from the adapter data.
1322  */
1323 static void
1324 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1325                       struct rte_eth_txq_info *qinfo)
1326 {
1327         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1328         struct sfc_txq_info *txq_info;
1329
1330         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1331
1332         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1333
1334         memset(qinfo, 0, sizeof(*qinfo));
1335
1336         qinfo->conf.offloads = txq_info->offloads;
1337         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1338         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1339         qinfo->nb_desc = txq_info->entries;
1340 }
1341
1342 /*
1343  * The function is used by the secondary process as well. It must not
1344  * use any process-local pointers from the adapter data.
1345  */
1346 static uint32_t
1347 sfc_rx_queue_count(void *rx_queue)
1348 {
1349         struct sfc_dp_rxq *dp_rxq = rx_queue;
1350         const struct sfc_dp_rx *dp_rx;
1351         struct sfc_rxq_info *rxq_info;
1352
1353         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1354         rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1355
1356         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1357                 return 0;
1358
1359         return dp_rx->qdesc_npending(dp_rxq);
1360 }
1361
1362 /*
1363  * The function is used by the secondary process as well. It must not
1364  * use any process-local pointers from the adapter data.
1365  */
1366 static int
1367 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1368 {
1369         struct sfc_dp_rxq *dp_rxq = queue;
1370         const struct sfc_dp_rx *dp_rx;
1371
1372         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1373
1374         return dp_rx->qdesc_status(dp_rxq, offset);
1375 }
1376
1377 /*
1378  * The function is used by the secondary process as well. It must not
1379  * use any process-local pointers from the adapter data.
1380  */
1381 static int
1382 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1383 {
1384         struct sfc_dp_txq *dp_txq = queue;
1385         const struct sfc_dp_tx *dp_tx;
1386
1387         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1388
1389         return dp_tx->qdesc_status(dp_txq, offset);
1390 }
1391
1392 static int
1393 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1394 {
1395         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1396         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1397         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1398         struct sfc_rxq_info *rxq_info;
1399         sfc_sw_index_t sw_index;
1400         int rc;
1401
1402         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1403
1404         sfc_adapter_lock(sa);
1405
1406         rc = EINVAL;
1407         if (sa->state != SFC_ETHDEV_STARTED)
1408                 goto fail_not_started;
1409
1410         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1411         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1412                 goto fail_not_setup;
1413
1414         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1415         rc = sfc_rx_qstart(sa, sw_index);
1416         if (rc != 0)
1417                 goto fail_rx_qstart;
1418
1419         rxq_info->deferred_started = B_TRUE;
1420
1421         sfc_adapter_unlock(sa);
1422
1423         return 0;
1424
1425 fail_rx_qstart:
1426 fail_not_setup:
1427 fail_not_started:
1428         sfc_adapter_unlock(sa);
1429         SFC_ASSERT(rc > 0);
1430         return -rc;
1431 }
1432
1433 static int
1434 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1435 {
1436         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1437         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1438         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1439         struct sfc_rxq_info *rxq_info;
1440         sfc_sw_index_t sw_index;
1441
1442         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1443
1444         sfc_adapter_lock(sa);
1445
1446         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1447         sfc_rx_qstop(sa, sw_index);
1448
1449         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1450         rxq_info->deferred_started = B_FALSE;
1451
1452         sfc_adapter_unlock(sa);
1453
1454         return 0;
1455 }
1456
1457 static int
1458 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1459 {
1460         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1461         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1462         struct sfc_txq_info *txq_info;
1463         sfc_sw_index_t sw_index;
1464         int rc;
1465
1466         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1467
1468         sfc_adapter_lock(sa);
1469
1470         rc = EINVAL;
1471         if (sa->state != SFC_ETHDEV_STARTED)
1472                 goto fail_not_started;
1473
1474         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1475         if (txq_info->state != SFC_TXQ_INITIALIZED)
1476                 goto fail_not_setup;
1477
1478         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1479         rc = sfc_tx_qstart(sa, sw_index);
1480         if (rc != 0)
1481                 goto fail_tx_qstart;
1482
1483         txq_info->deferred_started = B_TRUE;
1484
1485         sfc_adapter_unlock(sa);
1486         return 0;
1487
1488 fail_tx_qstart:
1489
1490 fail_not_setup:
1491 fail_not_started:
1492         sfc_adapter_unlock(sa);
1493         SFC_ASSERT(rc > 0);
1494         return -rc;
1495 }
1496
1497 static int
1498 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1499 {
1500         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1501         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1502         struct sfc_txq_info *txq_info;
1503         sfc_sw_index_t sw_index;
1504
1505         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1506
1507         sfc_adapter_lock(sa);
1508
1509         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1510         sfc_tx_qstop(sa, sw_index);
1511
1512         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1513         txq_info->deferred_started = B_FALSE;
1514
1515         sfc_adapter_unlock(sa);
1516         return 0;
1517 }
1518
1519 static efx_tunnel_protocol_t
1520 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1521 {
1522         switch (rte_type) {
1523         case RTE_ETH_TUNNEL_TYPE_VXLAN:
1524                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1525         case RTE_ETH_TUNNEL_TYPE_GENEVE:
1526                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1527         default:
1528                 return EFX_TUNNEL_NPROTOS;
1529         }
1530 }
1531
1532 enum sfc_udp_tunnel_op_e {
1533         SFC_UDP_TUNNEL_ADD_PORT,
1534         SFC_UDP_TUNNEL_DEL_PORT,
1535 };
1536
1537 static int
1538 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1539                       struct rte_eth_udp_tunnel *tunnel_udp,
1540                       enum sfc_udp_tunnel_op_e op)
1541 {
1542         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1543         efx_tunnel_protocol_t tunnel_proto;
1544         int rc;
1545
1546         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1547                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1548                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1549                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1550
1551         tunnel_proto =
1552                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1553         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1554                 rc = ENOTSUP;
1555                 goto fail_bad_proto;
1556         }
1557
1558         sfc_adapter_lock(sa);
1559
1560         switch (op) {
1561         case SFC_UDP_TUNNEL_ADD_PORT:
1562                 rc = efx_tunnel_config_udp_add(sa->nic,
1563                                                tunnel_udp->udp_port,
1564                                                tunnel_proto);
1565                 break;
1566         case SFC_UDP_TUNNEL_DEL_PORT:
1567                 rc = efx_tunnel_config_udp_remove(sa->nic,
1568                                                   tunnel_udp->udp_port,
1569                                                   tunnel_proto);
1570                 break;
1571         default:
1572                 rc = EINVAL;
1573                 goto fail_bad_op;
1574         }
1575
1576         if (rc != 0)
1577                 goto fail_op;
1578
1579         if (sa->state == SFC_ETHDEV_STARTED) {
1580                 rc = efx_tunnel_reconfigure(sa->nic);
1581                 if (rc == EAGAIN) {
1582                         /*
1583                          * Configuration is accepted by FW and MC reboot
1584                          * is initiated to apply the changes. MC reboot
1585                          * will be handled in a usual way (MC reboot
1586                          * event on management event queue and adapter
1587                          * restart).
1588                          */
1589                         rc = 0;
1590                 } else if (rc != 0) {
1591                         goto fail_reconfigure;
1592                 }
1593         }
1594
1595         sfc_adapter_unlock(sa);
1596         return 0;
1597
1598 fail_reconfigure:
1599         /* Remove/restore entry since the change makes the trouble */
1600         switch (op) {
1601         case SFC_UDP_TUNNEL_ADD_PORT:
1602                 (void)efx_tunnel_config_udp_remove(sa->nic,
1603                                                    tunnel_udp->udp_port,
1604                                                    tunnel_proto);
1605                 break;
1606         case SFC_UDP_TUNNEL_DEL_PORT:
1607                 (void)efx_tunnel_config_udp_add(sa->nic,
1608                                                 tunnel_udp->udp_port,
1609                                                 tunnel_proto);
1610                 break;
1611         }
1612
1613 fail_op:
1614 fail_bad_op:
1615         sfc_adapter_unlock(sa);
1616
1617 fail_bad_proto:
1618         SFC_ASSERT(rc > 0);
1619         return -rc;
1620 }
1621
1622 static int
1623 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1624                             struct rte_eth_udp_tunnel *tunnel_udp)
1625 {
1626         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1627 }
1628
1629 static int
1630 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1631                             struct rte_eth_udp_tunnel *tunnel_udp)
1632 {
1633         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1634 }
1635
1636 /*
1637  * The function is used by the secondary process as well. It must not
1638  * use any process-local pointers from the adapter data.
1639  */
1640 static int
1641 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1642                           struct rte_eth_rss_conf *rss_conf)
1643 {
1644         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1645         struct sfc_rss *rss = &sas->rss;
1646
1647         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1648                 return -ENOTSUP;
1649
1650         /*
1651          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1652          * hence, conversion is done here to derive a correct set of RTE_ETH_RSS
1653          * flags which corresponds to the active EFX configuration stored
1654          * locally in 'sfc_adapter' and kept up-to-date
1655          */
1656         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1657         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1658         if (rss_conf->rss_key != NULL)
1659                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1660
1661         return 0;
1662 }
1663
1664 static int
1665 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1666                         struct rte_eth_rss_conf *rss_conf)
1667 {
1668         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1669         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1670         unsigned int efx_hash_types;
1671         unsigned int n_contexts;
1672         unsigned int mode_i = 0;
1673         unsigned int key_i = 0;
1674         uint32_t contexts[2];
1675         unsigned int i = 0;
1676         int rc = 0;
1677
1678         if (sfc_sa2shared(sa)->isolated)
1679                 return -ENOTSUP;
1680
1681         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1682                 sfc_err(sa, "RSS is not available");
1683                 return -ENOTSUP;
1684         }
1685
1686         if (rss->channels == 0) {
1687                 sfc_err(sa, "RSS is not configured");
1688                 return -EINVAL;
1689         }
1690
1691         if ((rss_conf->rss_key != NULL) &&
1692             (rss_conf->rss_key_len != sizeof(rss->key))) {
1693                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1694                         sizeof(rss->key));
1695                 return -EINVAL;
1696         }
1697
1698         sfc_adapter_lock(sa);
1699
1700         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1701         if (rc != 0)
1702                 goto fail_rx_hf_rte_to_efx;
1703
1704         contexts[0] = EFX_RSS_CONTEXT_DEFAULT;
1705         contexts[1] = rss->dummy_ctx.nic_handle;
1706         n_contexts = (rss->dummy_ctx.nic_handle_refcnt == 0) ? 1 : 2;
1707
1708         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1709                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1710                                            rss->hash_alg, efx_hash_types,
1711                                            B_TRUE);
1712                 if (rc != 0)
1713                         goto fail_scale_mode_set;
1714         }
1715
1716         if (rss_conf->rss_key != NULL) {
1717                 if (sa->state == SFC_ETHDEV_STARTED) {
1718                         for (key_i = 0; key_i < n_contexts; key_i++) {
1719                                 rc = efx_rx_scale_key_set(sa->nic,
1720                                                           contexts[key_i],
1721                                                           rss_conf->rss_key,
1722                                                           sizeof(rss->key));
1723                                 if (rc != 0)
1724                                         goto fail_scale_key_set;
1725                         }
1726                 }
1727
1728                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1729         }
1730
1731         rss->hash_types = efx_hash_types;
1732
1733         sfc_adapter_unlock(sa);
1734
1735         return 0;
1736
1737 fail_scale_key_set:
1738         for (i = 0; i < key_i; i++) {
1739                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1740                                          sizeof(rss->key)) != 0)
1741                         sfc_err(sa, "failed to restore RSS key");
1742         }
1743
1744 fail_scale_mode_set:
1745         for (i = 0; i < mode_i; i++) {
1746                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1747                                           EFX_RX_HASHALG_TOEPLITZ,
1748                                           rss->hash_types, B_TRUE) != 0)
1749                         sfc_err(sa, "failed to restore RSS mode");
1750         }
1751
1752 fail_rx_hf_rte_to_efx:
1753         sfc_adapter_unlock(sa);
1754         return -rc;
1755 }
1756
1757 /*
1758  * The function is used by the secondary process as well. It must not
1759  * use any process-local pointers from the adapter data.
1760  */
1761 static int
1762 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1763                        struct rte_eth_rss_reta_entry64 *reta_conf,
1764                        uint16_t reta_size)
1765 {
1766         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1767         struct sfc_rss *rss = &sas->rss;
1768         int entry;
1769
1770         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1771                 return -ENOTSUP;
1772
1773         if (rss->channels == 0)
1774                 return -EINVAL;
1775
1776         if (reta_size != EFX_RSS_TBL_SIZE)
1777                 return -EINVAL;
1778
1779         for (entry = 0; entry < reta_size; entry++) {
1780                 int grp = entry / RTE_ETH_RETA_GROUP_SIZE;
1781                 int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1782
1783                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1784                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1785         }
1786
1787         return 0;
1788 }
1789
1790 static int
1791 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1792                         struct rte_eth_rss_reta_entry64 *reta_conf,
1793                         uint16_t reta_size)
1794 {
1795         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1796         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1797         unsigned int *rss_tbl_new;
1798         uint16_t entry;
1799         int rc = 0;
1800
1801
1802         if (sfc_sa2shared(sa)->isolated)
1803                 return -ENOTSUP;
1804
1805         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1806                 sfc_err(sa, "RSS is not available");
1807                 return -ENOTSUP;
1808         }
1809
1810         if (rss->channels == 0) {
1811                 sfc_err(sa, "RSS is not configured");
1812                 return -EINVAL;
1813         }
1814
1815         if (reta_size != EFX_RSS_TBL_SIZE) {
1816                 sfc_err(sa, "RETA size is wrong (should be %u)",
1817                         EFX_RSS_TBL_SIZE);
1818                 return -EINVAL;
1819         }
1820
1821         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1822         if (rss_tbl_new == NULL)
1823                 return -ENOMEM;
1824
1825         sfc_adapter_lock(sa);
1826
1827         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1828
1829         for (entry = 0; entry < reta_size; entry++) {
1830                 int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1831                 struct rte_eth_rss_reta_entry64 *grp;
1832
1833                 grp = &reta_conf[entry / RTE_ETH_RETA_GROUP_SIZE];
1834
1835                 if (grp->mask & (1ull << grp_idx)) {
1836                         if (grp->reta[grp_idx] >= rss->channels) {
1837                                 rc = EINVAL;
1838                                 goto bad_reta_entry;
1839                         }
1840                         rss_tbl_new[entry] = grp->reta[grp_idx];
1841                 }
1842         }
1843
1844         if (sa->state == SFC_ETHDEV_STARTED) {
1845                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1846                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1847                 if (rc != 0)
1848                         goto fail_scale_tbl_set;
1849         }
1850
1851         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1852
1853 fail_scale_tbl_set:
1854 bad_reta_entry:
1855         sfc_adapter_unlock(sa);
1856
1857         rte_free(rss_tbl_new);
1858
1859         SFC_ASSERT(rc >= 0);
1860         return -rc;
1861 }
1862
1863 static int
1864 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1865                      const struct rte_flow_ops **ops)
1866 {
1867         *ops = &sfc_flow_ops;
1868         return 0;
1869 }
1870
1871 static int
1872 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1873 {
1874         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1875
1876         /*
1877          * If Rx datapath does not provide callback to check mempool,
1878          * all pools are supported.
1879          */
1880         if (sap->dp_rx->pool_ops_supported == NULL)
1881                 return 1;
1882
1883         return sap->dp_rx->pool_ops_supported(pool);
1884 }
1885
1886 static int
1887 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1888 {
1889         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1890         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1891         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1892         struct sfc_rxq_info *rxq_info;
1893
1894         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1895
1896         return sap->dp_rx->intr_enable(rxq_info->dp);
1897 }
1898
1899 static int
1900 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1901 {
1902         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1903         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1904         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1905         struct sfc_rxq_info *rxq_info;
1906
1907         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1908
1909         return sap->dp_rx->intr_disable(rxq_info->dp);
1910 }
1911
1912 struct sfc_mport_journal_ctx {
1913         struct sfc_adapter              *sa;
1914         uint16_t                        switch_domain_id;
1915         uint32_t                        mcdi_handle;
1916         bool                            controllers_assigned;
1917         efx_pcie_interface_t            *controllers;
1918         size_t                          nb_controllers;
1919 };
1920
1921 static int
1922 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1923                                efx_pcie_interface_t intf)
1924 {
1925         efx_pcie_interface_t *new_controllers;
1926         size_t i, target;
1927         size_t new_size;
1928
1929         if (ctx->controllers == NULL) {
1930                 ctx->controllers = rte_malloc("sfc_controller_mapping",
1931                                               sizeof(ctx->controllers[0]), 0);
1932                 if (ctx->controllers == NULL)
1933                         return ENOMEM;
1934
1935                 ctx->controllers[0] = intf;
1936                 ctx->nb_controllers = 1;
1937
1938                 return 0;
1939         }
1940
1941         for (i = 0; i < ctx->nb_controllers; i++) {
1942                 if (ctx->controllers[i] == intf)
1943                         return 0;
1944                 if (ctx->controllers[i] > intf)
1945                         break;
1946         }
1947         target = i;
1948
1949         ctx->nb_controllers += 1;
1950         new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1951
1952         new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1953         if (new_controllers == NULL) {
1954                 rte_free(ctx->controllers);
1955                 return ENOMEM;
1956         }
1957         ctx->controllers = new_controllers;
1958
1959         for (i = target + 1; i < ctx->nb_controllers; i++)
1960                 ctx->controllers[i] = ctx->controllers[i - 1];
1961
1962         ctx->controllers[target] = intf;
1963
1964         return 0;
1965 }
1966
1967 static efx_rc_t
1968 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1969                                 efx_mport_desc_t *mport)
1970 {
1971         struct sfc_mae_switch_port_request req;
1972         efx_mport_sel_t entity_selector;
1973         efx_mport_sel_t ethdev_mport;
1974         uint16_t switch_port_id;
1975         efx_rc_t efx_rc;
1976         int rc;
1977
1978         sfc_dbg(ctx->sa,
1979                 "processing mport id %u (controller %u pf %u vf %u)",
1980                 mport->emd_id.id, mport->emd_vnic.ev_intf,
1981                 mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1982         efx_mae_mport_invalid(&ethdev_mport);
1983
1984         if (!ctx->controllers_assigned) {
1985                 rc = sfc_journal_ctx_add_controller(ctx,
1986                                                     mport->emd_vnic.ev_intf);
1987                 if (rc != 0)
1988                         return rc;
1989         }
1990
1991         /* Build Mport selector */
1992         efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
1993                                                 mport->emd_vnic.ev_pf,
1994                                                 mport->emd_vnic.ev_vf,
1995                                                 &entity_selector);
1996         if (efx_rc != 0) {
1997                 sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
1998                         mport->emd_vnic.ev_intf,
1999                         mport->emd_vnic.ev_pf,
2000                         mport->emd_vnic.ev_vf);
2001                 return efx_rc;
2002         }
2003
2004         rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2005                                               &entity_selector,
2006                                               SFC_MAE_SWITCH_PORT_REPRESENTOR,
2007                                               &switch_port_id);
2008         switch (rc) {
2009         case 0:
2010                 /* Already registered */
2011                 break;
2012         case ENOENT:
2013                 /*
2014                  * No representor has been created for this entity.
2015                  * Create a dummy switch registry entry with an invalid ethdev
2016                  * mport selector. When a corresponding representor is created,
2017                  * this entry will be updated.
2018                  */
2019                 req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2020                 req.entity_mportp = &entity_selector;
2021                 req.ethdev_mportp = &ethdev_mport;
2022                 req.ethdev_port_id = RTE_MAX_ETHPORTS;
2023                 req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2024                 req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2025                 req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2026
2027                 rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2028                                                 &req, &switch_port_id);
2029                 if (rc != 0) {
2030                         sfc_err(ctx->sa,
2031                                 "failed to assign MAE switch port for c%upf%uvf%u: %s",
2032                                 mport->emd_vnic.ev_intf,
2033                                 mport->emd_vnic.ev_pf,
2034                                 mport->emd_vnic.ev_vf,
2035                                 rte_strerror(rc));
2036                         return rc;
2037                 }
2038                 break;
2039         default:
2040                 sfc_err(ctx->sa, "failed to find 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
2048         return 0;
2049 }
2050
2051 static efx_rc_t
2052 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2053                              size_t mport_len)
2054 {
2055         struct sfc_mport_journal_ctx *ctx = data;
2056
2057         if (ctx == NULL || ctx->sa == NULL) {
2058                 sfc_err(ctx->sa, "received NULL context or SFC adapter");
2059                 return EINVAL;
2060         }
2061
2062         if (mport_len != sizeof(*mport)) {
2063                 sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2064                 return EINVAL;
2065         }
2066
2067         SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2068
2069         /*
2070          * If a zombie flag is set, it means the mport has been marked for
2071          * deletion and cannot be used for any new operations. The mport will
2072          * be destroyed completely once all references to it are released.
2073          */
2074         if (mport->emd_zombie) {
2075                 sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2076                 return 0;
2077         }
2078         if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2079                 sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2080                 return 0;
2081         }
2082         if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2083                 sfc_dbg(ctx->sa, "mport is not a function, skipping");
2084                 return 0;
2085         }
2086         if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2087                 sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2088                 return 0;
2089         }
2090
2091         return sfc_process_mport_journal_entry(ctx, mport);
2092 }
2093
2094 static int
2095 sfc_process_mport_journal(struct sfc_adapter *sa)
2096 {
2097         struct sfc_mport_journal_ctx ctx;
2098         const efx_pcie_interface_t *controllers;
2099         size_t nb_controllers;
2100         efx_rc_t efx_rc;
2101         int rc;
2102
2103         memset(&ctx, 0, sizeof(ctx));
2104         ctx.sa = sa;
2105         ctx.switch_domain_id = sa->mae.switch_domain_id;
2106
2107         efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2108         if (efx_rc != 0) {
2109                 sfc_err(sa, "failed to get own MCDI handle");
2110                 SFC_ASSERT(efx_rc > 0);
2111                 return efx_rc;
2112         }
2113
2114         rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2115                                                &controllers, &nb_controllers);
2116         if (rc != 0) {
2117                 sfc_err(sa, "failed to get controller mapping");
2118                 return rc;
2119         }
2120
2121         ctx.controllers_assigned = controllers != NULL;
2122         ctx.controllers = NULL;
2123         ctx.nb_controllers = 0;
2124
2125         efx_rc = efx_mae_read_mport_journal(sa->nic,
2126                                             sfc_process_mport_journal_cb, &ctx);
2127         if (efx_rc != 0) {
2128                 sfc_err(sa, "failed to process MAE mport journal");
2129                 SFC_ASSERT(efx_rc > 0);
2130                 return efx_rc;
2131         }
2132
2133         if (controllers == NULL) {
2134                 rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2135                                                            ctx.controllers,
2136                                                            ctx.nb_controllers);
2137                 if (rc != 0)
2138                         return rc;
2139         }
2140
2141         return 0;
2142 }
2143
2144 static void
2145 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2146                           const efx_mport_sel_t *ethdev_mportp __rte_unused,
2147                           uint16_t ethdev_port_id __rte_unused,
2148                           const efx_mport_sel_t *entity_mportp __rte_unused,
2149                           uint16_t switch_port_id __rte_unused,
2150                           union sfc_mae_switch_port_data *port_datap
2151                                 __rte_unused,
2152                           void *user_datap)
2153 {
2154         int *counter = user_datap;
2155
2156         SFC_ASSERT(counter != NULL);
2157
2158         if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2159                 (*counter)++;
2160 }
2161
2162 struct sfc_get_representors_ctx {
2163         struct rte_eth_representor_info *info;
2164         struct sfc_adapter              *sa;
2165         uint16_t                        switch_domain_id;
2166         const efx_pcie_interface_t      *controllers;
2167         size_t                          nb_controllers;
2168 };
2169
2170 static void
2171 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2172                         const efx_mport_sel_t *ethdev_mportp __rte_unused,
2173                         uint16_t ethdev_port_id __rte_unused,
2174                         const efx_mport_sel_t *entity_mportp __rte_unused,
2175                         uint16_t switch_port_id,
2176                         union sfc_mae_switch_port_data *port_datap,
2177                         void *user_datap)
2178 {
2179         struct sfc_get_representors_ctx *ctx = user_datap;
2180         struct rte_eth_representor_range *range;
2181         int ret;
2182         int rc;
2183
2184         SFC_ASSERT(ctx != NULL);
2185         SFC_ASSERT(ctx->info != NULL);
2186         SFC_ASSERT(ctx->sa != NULL);
2187
2188         if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2189                 sfc_dbg(ctx->sa, "not a representor, skipping");
2190                 return;
2191         }
2192         if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2193                 sfc_dbg(ctx->sa, "info structure is full already");
2194                 return;
2195         }
2196
2197         range = &ctx->info->ranges[ctx->info->nb_ranges];
2198         rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2199                                                     ctx->nb_controllers,
2200                                                     port_datap->repr.intf,
2201                                                     &range->controller);
2202         if (rc != 0) {
2203                 sfc_err(ctx->sa, "invalid representor controller: %d",
2204                         port_datap->repr.intf);
2205                 range->controller = -1;
2206         }
2207         range->pf = port_datap->repr.pf;
2208         range->id_base = switch_port_id;
2209         range->id_end = switch_port_id;
2210
2211         if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2212                 range->type = RTE_ETH_REPRESENTOR_VF;
2213                 range->vf = port_datap->repr.vf;
2214                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2215                                "c%dpf%dvf%d", range->controller, range->pf,
2216                                range->vf);
2217         } else {
2218                 range->type = RTE_ETH_REPRESENTOR_PF;
2219                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2220                          "c%dpf%d", range->controller, range->pf);
2221         }
2222         if (ret >= RTE_DEV_NAME_MAX_LEN) {
2223                 sfc_err(ctx->sa, "representor name has been truncated: %s",
2224                         range->name);
2225         }
2226
2227         ctx->info->nb_ranges++;
2228 }
2229
2230 static int
2231 sfc_representor_info_get(struct rte_eth_dev *dev,
2232                          struct rte_eth_representor_info *info)
2233 {
2234         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2235         struct sfc_get_representors_ctx get_repr_ctx;
2236         const efx_nic_cfg_t *nic_cfg;
2237         uint16_t switch_domain_id;
2238         uint32_t nb_repr;
2239         int controller;
2240         int rc;
2241
2242         sfc_adapter_lock(sa);
2243
2244         if (sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2245                 sfc_adapter_unlock(sa);
2246                 return -ENOTSUP;
2247         }
2248
2249         rc = sfc_process_mport_journal(sa);
2250         if (rc != 0) {
2251                 sfc_adapter_unlock(sa);
2252                 SFC_ASSERT(rc > 0);
2253                 return -rc;
2254         }
2255
2256         switch_domain_id = sa->mae.switch_domain_id;
2257
2258         nb_repr = 0;
2259         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2260                                           sfc_count_representors_cb,
2261                                           &nb_repr);
2262         if (rc != 0) {
2263                 sfc_adapter_unlock(sa);
2264                 SFC_ASSERT(rc > 0);
2265                 return -rc;
2266         }
2267
2268         if (info == NULL) {
2269                 sfc_adapter_unlock(sa);
2270                 return nb_repr;
2271         }
2272
2273         rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2274                                                &get_repr_ctx.controllers,
2275                                                &get_repr_ctx.nb_controllers);
2276         if (rc != 0) {
2277                 sfc_adapter_unlock(sa);
2278                 SFC_ASSERT(rc > 0);
2279                 return -rc;
2280         }
2281
2282         nic_cfg = efx_nic_cfg_get(sa->nic);
2283
2284         rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2285                                                   nic_cfg->enc_intf,
2286                                                   &controller);
2287         if (rc != 0) {
2288                 sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2289                 controller = -1;
2290         }
2291
2292         info->controller = controller;
2293         info->pf = nic_cfg->enc_pf;
2294
2295         get_repr_ctx.info = info;
2296         get_repr_ctx.sa = sa;
2297         get_repr_ctx.switch_domain_id = switch_domain_id;
2298         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2299                                           sfc_get_representors_cb,
2300                                           &get_repr_ctx);
2301         if (rc != 0) {
2302                 sfc_adapter_unlock(sa);
2303                 SFC_ASSERT(rc > 0);
2304                 return -rc;
2305         }
2306
2307         sfc_adapter_unlock(sa);
2308         return nb_repr;
2309 }
2310
2311 static int
2312 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2313 {
2314         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2315         uint64_t supported = 0;
2316
2317         sfc_adapter_lock(sa);
2318
2319         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2320                 supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2321
2322         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2323                 supported |= RTE_ETH_RX_METADATA_USER_MARK;
2324
2325         if (sfc_flow_tunnel_is_supported(sa))
2326                 supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2327
2328         sa->negotiated_rx_metadata = supported & *features;
2329         *features = sa->negotiated_rx_metadata;
2330
2331         sfc_adapter_unlock(sa);
2332
2333         return 0;
2334 }
2335
2336 static const struct eth_dev_ops sfc_eth_dev_ops = {
2337         .dev_configure                  = sfc_dev_configure,
2338         .dev_start                      = sfc_dev_start,
2339         .dev_stop                       = sfc_dev_stop,
2340         .dev_set_link_up                = sfc_dev_set_link_up,
2341         .dev_set_link_down              = sfc_dev_set_link_down,
2342         .dev_close                      = sfc_dev_close,
2343         .promiscuous_enable             = sfc_dev_promisc_enable,
2344         .promiscuous_disable            = sfc_dev_promisc_disable,
2345         .allmulticast_enable            = sfc_dev_allmulti_enable,
2346         .allmulticast_disable           = sfc_dev_allmulti_disable,
2347         .link_update                    = sfc_dev_link_update,
2348         .stats_get                      = sfc_stats_get,
2349         .stats_reset                    = sfc_stats_reset,
2350         .xstats_get                     = sfc_xstats_get,
2351         .xstats_reset                   = sfc_stats_reset,
2352         .xstats_get_names               = sfc_xstats_get_names,
2353         .dev_infos_get                  = sfc_dev_infos_get,
2354         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2355         .mtu_set                        = sfc_dev_set_mtu,
2356         .rx_queue_start                 = sfc_rx_queue_start,
2357         .rx_queue_stop                  = sfc_rx_queue_stop,
2358         .tx_queue_start                 = sfc_tx_queue_start,
2359         .tx_queue_stop                  = sfc_tx_queue_stop,
2360         .rx_queue_setup                 = sfc_rx_queue_setup,
2361         .rx_queue_release               = sfc_rx_queue_release,
2362         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
2363         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
2364         .tx_queue_setup                 = sfc_tx_queue_setup,
2365         .tx_queue_release               = sfc_tx_queue_release,
2366         .flow_ctrl_get                  = sfc_flow_ctrl_get,
2367         .flow_ctrl_set                  = sfc_flow_ctrl_set,
2368         .mac_addr_set                   = sfc_mac_addr_set,
2369         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
2370         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
2371         .reta_update                    = sfc_dev_rss_reta_update,
2372         .reta_query                     = sfc_dev_rss_reta_query,
2373         .rss_hash_update                = sfc_dev_rss_hash_update,
2374         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2375         .flow_ops_get                   = sfc_dev_flow_ops_get,
2376         .set_mc_addr_list               = sfc_set_mc_addr_list,
2377         .rxq_info_get                   = sfc_rx_queue_info_get,
2378         .txq_info_get                   = sfc_tx_queue_info_get,
2379         .fw_version_get                 = sfc_fw_version_get,
2380         .xstats_get_by_id               = sfc_xstats_get_by_id,
2381         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
2382         .pool_ops_supported             = sfc_pool_ops_supported,
2383         .representor_info_get           = sfc_representor_info_get,
2384         .rx_metadata_negotiate          = sfc_rx_metadata_negotiate,
2385 };
2386
2387 struct sfc_ethdev_init_data {
2388         uint16_t                nb_representors;
2389 };
2390
2391 /**
2392  * Duplicate a string in potentially shared memory required for
2393  * multi-process support.
2394  *
2395  * strdup() allocates from process-local heap/memory.
2396  */
2397 static char *
2398 sfc_strdup(const char *str)
2399 {
2400         size_t size;
2401         char *copy;
2402
2403         if (str == NULL)
2404                 return NULL;
2405
2406         size = strlen(str) + 1;
2407         copy = rte_malloc(__func__, size, 0);
2408         if (copy != NULL)
2409                 rte_memcpy(copy, str, size);
2410
2411         return copy;
2412 }
2413
2414 static int
2415 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2416 {
2417         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2418         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2419         const struct sfc_dp_rx *dp_rx;
2420         const struct sfc_dp_tx *dp_tx;
2421         const efx_nic_cfg_t *encp;
2422         unsigned int avail_caps = 0;
2423         const char *rx_name = NULL;
2424         const char *tx_name = NULL;
2425         int rc;
2426
2427         switch (sa->family) {
2428         case EFX_FAMILY_HUNTINGTON:
2429         case EFX_FAMILY_MEDFORD:
2430         case EFX_FAMILY_MEDFORD2:
2431                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2432                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2433                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2434                 break;
2435         case EFX_FAMILY_RIVERHEAD:
2436                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2437                 break;
2438         default:
2439                 break;
2440         }
2441
2442         encp = efx_nic_cfg_get(sa->nic);
2443         if (encp->enc_rx_es_super_buffer_supported)
2444                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2445
2446         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2447                                 sfc_kvarg_string_handler, &rx_name);
2448         if (rc != 0)
2449                 goto fail_kvarg_rx_datapath;
2450
2451         if (rx_name != NULL) {
2452                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2453                 if (dp_rx == NULL) {
2454                         sfc_err(sa, "Rx datapath %s not found", rx_name);
2455                         rc = ENOENT;
2456                         goto fail_dp_rx;
2457                 }
2458                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2459                         sfc_err(sa,
2460                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2461                                 rx_name);
2462                         rc = EINVAL;
2463                         goto fail_dp_rx_caps;
2464                 }
2465         } else {
2466                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2467                 if (dp_rx == NULL) {
2468                         sfc_err(sa, "Rx datapath by caps %#x not found",
2469                                 avail_caps);
2470                         rc = ENOENT;
2471                         goto fail_dp_rx;
2472                 }
2473         }
2474
2475         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2476         if (sas->dp_rx_name == NULL) {
2477                 rc = ENOMEM;
2478                 goto fail_dp_rx_name;
2479         }
2480
2481         if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2482                 /* FLAG and MARK are always available from Rx prefix. */
2483                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2484                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2485         }
2486
2487         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2488
2489         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2490                                 sfc_kvarg_string_handler, &tx_name);
2491         if (rc != 0)
2492                 goto fail_kvarg_tx_datapath;
2493
2494         if (tx_name != NULL) {
2495                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2496                 if (dp_tx == NULL) {
2497                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2498                         rc = ENOENT;
2499                         goto fail_dp_tx;
2500                 }
2501                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2502                         sfc_err(sa,
2503                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2504                                 tx_name);
2505                         rc = EINVAL;
2506                         goto fail_dp_tx_caps;
2507                 }
2508         } else {
2509                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2510                 if (dp_tx == NULL) {
2511                         sfc_err(sa, "Tx datapath by caps %#x not found",
2512                                 avail_caps);
2513                         rc = ENOENT;
2514                         goto fail_dp_tx;
2515                 }
2516         }
2517
2518         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2519         if (sas->dp_tx_name == NULL) {
2520                 rc = ENOMEM;
2521                 goto fail_dp_tx_name;
2522         }
2523
2524         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2525
2526         sa->priv.dp_rx = dp_rx;
2527         sa->priv.dp_tx = dp_tx;
2528
2529         dev->rx_pkt_burst = dp_rx->pkt_burst;
2530         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2531         dev->tx_pkt_burst = dp_tx->pkt_burst;
2532
2533         dev->rx_queue_count = sfc_rx_queue_count;
2534         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2535         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2536         dev->dev_ops = &sfc_eth_dev_ops;
2537
2538         return 0;
2539
2540 fail_dp_tx_name:
2541 fail_dp_tx_caps:
2542 fail_dp_tx:
2543 fail_kvarg_tx_datapath:
2544         rte_free(sas->dp_rx_name);
2545         sas->dp_rx_name = NULL;
2546
2547 fail_dp_rx_name:
2548 fail_dp_rx_caps:
2549 fail_dp_rx:
2550 fail_kvarg_rx_datapath:
2551         return rc;
2552 }
2553
2554 static void
2555 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2556 {
2557         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2558         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2559
2560         dev->dev_ops = NULL;
2561         dev->tx_pkt_prepare = NULL;
2562         dev->rx_pkt_burst = NULL;
2563         dev->tx_pkt_burst = NULL;
2564
2565         rte_free(sas->dp_tx_name);
2566         sas->dp_tx_name = NULL;
2567         sa->priv.dp_tx = NULL;
2568
2569         rte_free(sas->dp_rx_name);
2570         sas->dp_rx_name = NULL;
2571         sa->priv.dp_rx = NULL;
2572 }
2573
2574 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2575         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2576         .reta_query                     = sfc_dev_rss_reta_query,
2577         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2578         .rxq_info_get                   = sfc_rx_queue_info_get,
2579         .txq_info_get                   = sfc_tx_queue_info_get,
2580 };
2581
2582 static int
2583 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2584 {
2585         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2586         struct sfc_adapter_priv *sap;
2587         const struct sfc_dp_rx *dp_rx;
2588         const struct sfc_dp_tx *dp_tx;
2589         int rc;
2590
2591         /*
2592          * Allocate process private data from heap, since it should not
2593          * be located in shared memory allocated using rte_malloc() API.
2594          */
2595         sap = calloc(1, sizeof(*sap));
2596         if (sap == NULL) {
2597                 rc = ENOMEM;
2598                 goto fail_alloc_priv;
2599         }
2600
2601         sap->logtype_main = logtype_main;
2602
2603         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2604         if (dp_rx == NULL) {
2605                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2606                         "cannot find %s Rx datapath", sas->dp_rx_name);
2607                 rc = ENOENT;
2608                 goto fail_dp_rx;
2609         }
2610         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2611                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2612                         "%s Rx datapath does not support multi-process",
2613                         sas->dp_rx_name);
2614                 rc = EINVAL;
2615                 goto fail_dp_rx_multi_process;
2616         }
2617
2618         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2619         if (dp_tx == NULL) {
2620                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2621                         "cannot find %s Tx datapath", sas->dp_tx_name);
2622                 rc = ENOENT;
2623                 goto fail_dp_tx;
2624         }
2625         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2626                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2627                         "%s Tx datapath does not support multi-process",
2628                         sas->dp_tx_name);
2629                 rc = EINVAL;
2630                 goto fail_dp_tx_multi_process;
2631         }
2632
2633         sap->dp_rx = dp_rx;
2634         sap->dp_tx = dp_tx;
2635
2636         dev->process_private = sap;
2637         dev->rx_pkt_burst = dp_rx->pkt_burst;
2638         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2639         dev->tx_pkt_burst = dp_tx->pkt_burst;
2640         dev->rx_queue_count = sfc_rx_queue_count;
2641         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2642         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2643         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2644
2645         return 0;
2646
2647 fail_dp_tx_multi_process:
2648 fail_dp_tx:
2649 fail_dp_rx_multi_process:
2650 fail_dp_rx:
2651         free(sap);
2652
2653 fail_alloc_priv:
2654         return rc;
2655 }
2656
2657 static void
2658 sfc_register_dp(void)
2659 {
2660         /* Register once */
2661         if (TAILQ_EMPTY(&sfc_dp_head)) {
2662                 /* Prefer EF10 datapath */
2663                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2664                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2665                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2666                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2667
2668                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2669                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2670                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2671                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2672         }
2673 }
2674
2675 static int
2676 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
2677 {
2678         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2679         const char *switch_mode = NULL;
2680         int rc;
2681
2682         sfc_log_init(sa, "entry");
2683
2684         rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2685                                 sfc_kvarg_string_handler, &switch_mode);
2686         if (rc != 0)
2687                 goto fail_kvargs;
2688
2689         if (switch_mode == NULL) {
2690                 sa->switchdev = encp->enc_mae_admin &&
2691                                 (!encp->enc_datapath_cap_evb ||
2692                                  has_representors);
2693         } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2694                 sa->switchdev = false;
2695         } else if (strcasecmp(switch_mode,
2696                               SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2697                 sa->switchdev = true;
2698         } else {
2699                 sfc_err(sa, "invalid switch mode device argument '%s'",
2700                         switch_mode);
2701                 rc = EINVAL;
2702                 goto fail_mode;
2703         }
2704
2705         sfc_log_init(sa, "done");
2706
2707         return 0;
2708
2709 fail_mode:
2710 fail_kvargs:
2711         sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2712
2713         return rc;
2714 }
2715
2716 static int
2717 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
2718 {
2719         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2720         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2721         struct sfc_ethdev_init_data *init_data = init_params;
2722         uint32_t logtype_main;
2723         struct sfc_adapter *sa;
2724         int rc;
2725         const efx_nic_cfg_t *encp;
2726         const struct rte_ether_addr *from;
2727         int ret;
2728
2729         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2730                         SFC_EFX_DEV_CLASS_NET) {
2731                 SFC_GENERIC_LOG(DEBUG,
2732                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2733                 return 1;
2734         }
2735
2736         rc = sfc_dp_mport_register();
2737         if (rc != 0)
2738                 return rc;
2739
2740         sfc_register_dp();
2741
2742         logtype_main = sfc_register_logtype(&pci_dev->addr,
2743                                             SFC_LOGTYPE_MAIN_STR,
2744                                             RTE_LOG_NOTICE);
2745
2746         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2747                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2748
2749         /* Required for logging */
2750         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2751                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2752                         pci_dev->addr.domain, pci_dev->addr.bus,
2753                         pci_dev->addr.devid, pci_dev->addr.function,
2754                         dev->data->port_id);
2755         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2756                 SFC_GENERIC_LOG(ERR,
2757                         "reserved log prefix is too short for " PCI_PRI_FMT,
2758                         pci_dev->addr.domain, pci_dev->addr.bus,
2759                         pci_dev->addr.devid, pci_dev->addr.function);
2760                 return -EINVAL;
2761         }
2762         sas->pci_addr = pci_dev->addr;
2763         sas->port_id = dev->data->port_id;
2764
2765         /*
2766          * Allocate process private data from heap, since it should not
2767          * be located in shared memory allocated using rte_malloc() API.
2768          */
2769         sa = calloc(1, sizeof(*sa));
2770         if (sa == NULL) {
2771                 rc = ENOMEM;
2772                 goto fail_alloc_sa;
2773         }
2774
2775         dev->process_private = sa;
2776
2777         /* Required for logging */
2778         sa->priv.shared = sas;
2779         sa->priv.logtype_main = logtype_main;
2780
2781         sa->eth_dev = dev;
2782
2783         /* Copy PCI device info to the dev->data */
2784         rte_eth_copy_pci_info(dev, pci_dev);
2785         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2786
2787         rc = sfc_kvargs_parse(sa);
2788         if (rc != 0)
2789                 goto fail_kvargs_parse;
2790
2791         sfc_log_init(sa, "entry");
2792
2793         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2794         if (dev->data->mac_addrs == NULL) {
2795                 rc = ENOMEM;
2796                 goto fail_mac_addrs;
2797         }
2798
2799         sfc_adapter_lock_init(sa);
2800         sfc_adapter_lock(sa);
2801
2802         sfc_log_init(sa, "probing");
2803         rc = sfc_probe(sa);
2804         if (rc != 0)
2805                 goto fail_probe;
2806
2807         /*
2808          * Selecting a default switch mode requires the NIC to be probed and
2809          * to have its capabilities filled in.
2810          */
2811         rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
2812         if (rc != 0)
2813                 goto fail_switch_mode;
2814
2815         sfc_log_init(sa, "set device ops");
2816         rc = sfc_eth_dev_set_ops(dev);
2817         if (rc != 0)
2818                 goto fail_set_ops;
2819
2820         sfc_log_init(sa, "attaching");
2821         rc = sfc_attach(sa);
2822         if (rc != 0)
2823                 goto fail_attach;
2824
2825         if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2826                 sfc_err(sa,
2827                         "failed to enable switchdev mode without admin MAE privilege");
2828                 rc = ENOTSUP;
2829                 goto fail_switchdev_no_mae;
2830         }
2831
2832         encp = efx_nic_cfg_get(sa->nic);
2833
2834         /*
2835          * The arguments are really reverse order in comparison to
2836          * Linux kernel. Copy from NIC config to Ethernet device data.
2837          */
2838         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2839         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2840
2841         /*
2842          * Setup the NIC DMA mapping handler. All internal mempools
2843          * MUST be created on attach before this point, and the
2844          * adapter MUST NOT create mempools with the adapter lock
2845          * held after this point.
2846          */
2847         rc = sfc_nic_dma_attach(sa);
2848         if (rc != 0)
2849                 goto fail_nic_dma_attach;
2850
2851         sfc_adapter_unlock(sa);
2852
2853         sfc_log_init(sa, "done");
2854         return 0;
2855
2856 fail_nic_dma_attach:
2857 fail_switchdev_no_mae:
2858         sfc_detach(sa);
2859
2860 fail_attach:
2861         sfc_eth_dev_clear_ops(dev);
2862
2863 fail_set_ops:
2864 fail_switch_mode:
2865         sfc_unprobe(sa);
2866
2867 fail_probe:
2868         sfc_adapter_unlock(sa);
2869         sfc_adapter_lock_fini(sa);
2870         rte_free(dev->data->mac_addrs);
2871         dev->data->mac_addrs = NULL;
2872
2873 fail_mac_addrs:
2874         sfc_kvargs_cleanup(sa);
2875
2876 fail_kvargs_parse:
2877         sfc_log_init(sa, "failed %d", rc);
2878         dev->process_private = NULL;
2879         free(sa);
2880
2881 fail_alloc_sa:
2882         SFC_ASSERT(rc > 0);
2883         return -rc;
2884 }
2885
2886 static int
2887 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2888 {
2889         sfc_dev_close(dev);
2890
2891         return 0;
2892 }
2893
2894 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2895         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2896         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2897         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2898         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2899         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2900         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2901         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2902         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2903         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2904         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD_VF) },
2905         { .vendor_id = 0 /* sentinel */ }
2906 };
2907
2908 static int
2909 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
2910 {
2911         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
2912         int rc;
2913
2914         if (args != NULL) {
2915                 rc = rte_eth_devargs_parse(args, &eth_da);
2916                 if (rc != 0) {
2917                         SFC_GENERIC_LOG(ERR,
2918                                         "Failed to parse generic devargs '%s'",
2919                                         args);
2920                         return rc;
2921                 }
2922         }
2923
2924         *devargs = eth_da;
2925
2926         return 0;
2927 }
2928
2929 static int
2930 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
2931                            struct sfc_ethdev_init_data *init_data,
2932                            struct rte_eth_dev **devp,
2933                            bool *dev_created)
2934 {
2935         struct rte_eth_dev *dev;
2936         bool created = false;
2937         int rc;
2938
2939         dev = rte_eth_dev_allocated(pci_dev->device.name);
2940         if (dev == NULL) {
2941                 rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
2942                                         sizeof(struct sfc_adapter_shared),
2943                                         eth_dev_pci_specific_init, pci_dev,
2944                                         sfc_eth_dev_init, init_data);
2945                 if (rc != 0) {
2946                         SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
2947                                         pci_dev->device.name);
2948                         return rc;
2949                 }
2950
2951                 created = true;
2952
2953                 dev = rte_eth_dev_allocated(pci_dev->device.name);
2954                 if (dev == NULL) {
2955                         SFC_GENERIC_LOG(ERR,
2956                                 "Failed to find allocated sfc ethdev '%s'",
2957                                 pci_dev->device.name);
2958                         return -ENODEV;
2959                 }
2960         }
2961
2962         *devp = dev;
2963         *dev_created = created;
2964
2965         return 0;
2966 }
2967
2968 static int
2969 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
2970                         efx_pcie_interface_t controller,
2971                         uint16_t port,
2972                         uint16_t repr_port,
2973                         enum rte_eth_representor_type type)
2974 {
2975         struct sfc_repr_entity_info entity;
2976         efx_mport_sel_t mport_sel;
2977         int rc;
2978
2979         switch (type) {
2980         case RTE_ETH_REPRESENTOR_NONE:
2981                 return 0;
2982         case RTE_ETH_REPRESENTOR_VF:
2983         case RTE_ETH_REPRESENTOR_PF:
2984                 break;
2985         case RTE_ETH_REPRESENTOR_SF:
2986                 sfc_err(sa, "SF representors are not supported");
2987                 return ENOTSUP;
2988         default:
2989                 sfc_err(sa, "unknown representor type: %d", type);
2990                 return ENOTSUP;
2991         }
2992
2993         rc = efx_mae_mport_by_pcie_mh_function(controller,
2994                                                port,
2995                                                repr_port,
2996                                                &mport_sel);
2997         if (rc != 0) {
2998                 sfc_err(sa,
2999                         "failed to get m-port selector for controller %u port %u repr_port %u: %s",
3000                         controller, port, repr_port, rte_strerror(-rc));
3001                 return rc;
3002         }
3003
3004         memset(&entity, 0, sizeof(entity));
3005         entity.type = type;
3006         entity.intf = controller;
3007         entity.pf = port;
3008         entity.vf = repr_port;
3009
3010         rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3011                              &mport_sel);
3012         if (rc != 0) {
3013                 sfc_err(sa,
3014                         "failed to create representor for controller %u port %u repr_port %u: %s",
3015                         controller, port, repr_port, rte_strerror(-rc));
3016                 return rc;
3017         }
3018
3019         return 0;
3020 }
3021
3022 static int
3023 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3024                              const struct rte_eth_devargs *eth_da,
3025                              efx_pcie_interface_t controller,
3026                              uint16_t port)
3027 {
3028         int first_error = 0;
3029         uint16_t i;
3030         int rc;
3031
3032         if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3033                 return sfc_eth_dev_create_repr(sa, controller, port,
3034                                                EFX_PCI_VF_INVALID,
3035                                                eth_da->type);
3036         }
3037
3038         for (i = 0; i < eth_da->nb_representor_ports; i++) {
3039                 rc = sfc_eth_dev_create_repr(sa, controller, port,
3040                                              eth_da->representor_ports[i],
3041                                              eth_da->type);
3042                 if (rc != 0 && first_error == 0)
3043                         first_error = rc;
3044         }
3045
3046         return first_error;
3047 }
3048
3049 static int
3050 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3051                                    const struct rte_eth_devargs *eth_da,
3052                                    efx_pcie_interface_t controller)
3053 {
3054         const efx_nic_cfg_t *encp;
3055         int first_error = 0;
3056         uint16_t default_port;
3057         uint16_t i;
3058         int rc;
3059
3060         if (eth_da->nb_ports == 0) {
3061                 encp = efx_nic_cfg_get(sa->nic);
3062                 default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3063                 return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3064                                                     default_port);
3065         }
3066
3067         for (i = 0; i < eth_da->nb_ports; i++) {
3068                 rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3069                                                   eth_da->ports[i]);
3070                 if (rc != 0 && first_error == 0)
3071                         first_error = rc;
3072         }
3073
3074         return first_error;
3075 }
3076
3077 static int
3078 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3079                                 const struct rte_eth_devargs *eth_da)
3080 {
3081         efx_pcie_interface_t intf;
3082         const efx_nic_cfg_t *encp;
3083         struct sfc_adapter *sa;
3084         uint16_t switch_domain_id;
3085         uint16_t i;
3086         int rc;
3087
3088         sa = sfc_adapter_by_eth_dev(dev);
3089         switch_domain_id = sa->mae.switch_domain_id;
3090
3091         switch (eth_da->type) {
3092         case RTE_ETH_REPRESENTOR_NONE:
3093                 return 0;
3094         case RTE_ETH_REPRESENTOR_PF:
3095         case RTE_ETH_REPRESENTOR_VF:
3096                 break;
3097         case RTE_ETH_REPRESENTOR_SF:
3098                 sfc_err(sa, "SF representors are not supported");
3099                 return -ENOTSUP;
3100         default:
3101                 sfc_err(sa, "unknown representor type: %d",
3102                         eth_da->type);
3103                 return -ENOTSUP;
3104         }
3105
3106         if (!sa->switchdev) {
3107                 sfc_err(sa, "cannot create representors in non-switchdev mode");
3108                 return -EINVAL;
3109         }
3110
3111         if (!sfc_repr_available(sfc_sa2shared(sa))) {
3112                 sfc_err(sa, "cannot create representors: unsupported");
3113
3114                 return -ENOTSUP;
3115         }
3116
3117         /*
3118          * This is needed to construct the DPDK controller -> EFX interface
3119          * mapping.
3120          */
3121         sfc_adapter_lock(sa);
3122         rc = sfc_process_mport_journal(sa);
3123         sfc_adapter_unlock(sa);
3124         if (rc != 0) {
3125                 SFC_ASSERT(rc > 0);
3126                 return -rc;
3127         }
3128
3129         if (eth_da->nb_mh_controllers > 0) {
3130                 for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3131                         rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3132                                                 eth_da->mh_controllers[i],
3133                                                 &intf);
3134                         if (rc != 0) {
3135                                 sfc_err(sa, "failed to get representor");
3136                                 continue;
3137                         }
3138                         sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3139                 }
3140         } else {
3141                 encp = efx_nic_cfg_get(sa->nic);
3142                 sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3143         }
3144
3145         return 0;
3146 }
3147
3148 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3149         struct rte_pci_device *pci_dev)
3150 {
3151         struct sfc_ethdev_init_data init_data;
3152         struct rte_eth_devargs eth_da;
3153         struct rte_eth_dev *dev;
3154         bool dev_created;
3155         int rc;
3156
3157         if (pci_dev->device.devargs != NULL) {
3158                 rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3159                                            &eth_da);
3160                 if (rc != 0)
3161                         return rc;
3162         } else {
3163                 memset(&eth_da, 0, sizeof(eth_da));
3164         }
3165
3166         /* If no VF representors specified, check for PF ones */
3167         if (eth_da.nb_representor_ports > 0)
3168                 init_data.nb_representors = eth_da.nb_representor_ports;
3169         else
3170                 init_data.nb_representors = eth_da.nb_ports;
3171
3172         if (init_data.nb_representors > 0 &&
3173             rte_eal_process_type() != RTE_PROC_PRIMARY) {
3174                 SFC_GENERIC_LOG(ERR,
3175                         "Create representors from secondary process not supported, dev '%s'",
3176                         pci_dev->device.name);
3177                 return -ENOTSUP;
3178         }
3179
3180         /*
3181          * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3182          * if it does not already exist. Re-probing an existing device is
3183          * expected to allow additional representors to be configured.
3184          */
3185         rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3186                                         &dev_created);
3187         if (rc != 0)
3188                 return rc;
3189
3190         rc = sfc_eth_dev_create_representors(dev, &eth_da);
3191         if (rc != 0) {
3192                 if (dev_created)
3193                         (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3194
3195                 return rc;
3196         }
3197
3198         return 0;
3199 }
3200
3201 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3202 {
3203         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3204 }
3205
3206 static struct rte_pci_driver sfc_efx_pmd = {
3207         .id_table = pci_id_sfc_efx_map,
3208         .drv_flags =
3209                 RTE_PCI_DRV_INTR_LSC |
3210                 RTE_PCI_DRV_NEED_MAPPING |
3211                 RTE_PCI_DRV_PROBE_AGAIN,
3212         .probe = sfc_eth_dev_pci_probe,
3213         .remove = sfc_eth_dev_pci_remove,
3214 };
3215
3216 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3217 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3218 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3219 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3220         SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3221         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3222         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3223         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3224         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3225         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3226         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3227
3228 RTE_INIT(sfc_driver_register_logtype)
3229 {
3230         int ret;
3231
3232         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
3233                                                    RTE_LOG_NOTICE);
3234         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
3235 }