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         uint64_t txq_offloads_def = 0;
  98
  99         sfc_log_init(sa, "entry");
 100
 101         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 102         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 103
 104         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 105
 106         dev_info->max_vfs = sa->sriov.num_vfs;
 107
 108         /* Autonegotiation may be disabled */
 109         dev_info->speed_capa = RTE_ETH_LINK_SPEED_FIXED;
 110         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 111                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_1G;
 112         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 113                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_10G;
 114         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 115                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_25G;
 116         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 117                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_40G;
 118         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 119                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_50G;
 120         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 121                 dev_info->speed_capa |= RTE_ETH_LINK_SPEED_100G;
 122
 123         dev_info->max_rx_queues = sa->rxq_max;
 124         dev_info->max_tx_queues = sa->txq_max;
 125
 126         /* By default packets are dropped if no descriptors are available */
 127         dev_info->default_rxconf.rx_drop_en = 1;
 128
 129         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 130
 131         /*
 132          * rx_offload_capa includes both device and queue offloads since
 133          * the latter may be requested on a per device basis which makes
 134          * sense when some offloads are needed to be set on all queues.
 135          */
 136         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 137                                     dev_info->rx_queue_offload_capa;
 138
 139         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 140
 141         /*
 142          * tx_offload_capa includes both device and queue offloads since
 143          * the latter may be requested on a per device basis which makes
 144          * sense when some offloads are needed to be set on all queues.
 145          */
 146         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 147                                     dev_info->tx_queue_offload_capa;
 148
 149         if (dev_info->tx_offload_capa & RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE)
 150                 txq_offloads_def |= RTE_ETH_TX_OFFLOAD_MBUF_FAST_FREE;
 151
 152         dev_info->default_txconf.offloads |= txq_offloads_def;
 153
 154         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 155                 uint64_t rte_hf = 0;
 156                 unsigned int i;
 157
 158                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 159                         rte_hf |= rss->hf_map[i].rte;
 160
 161                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 162                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 163                 dev_info->flow_type_rss_offloads = rte_hf;
 164         }
 165
 166         /* Initialize to hardware limits */
 167         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 168         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 169         /* The RXQ hardware requires that the descriptor count is a power
 170          * of 2, but rx_desc_lim cannot properly describe that constraint.
 171          */
 172         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 173
 174         /* Initialize to hardware limits */
 175         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 176         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 177         /*
 178          * The TXQ hardware requires that the descriptor count is a power
 179          * of 2, but tx_desc_lim cannot properly describe that constraint
 180          */
 181         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 182
 183         if (sap->dp_rx->get_dev_info != NULL)
 184                 sap->dp_rx->get_dev_info(dev_info);
 185         if (sap->dp_tx->get_dev_info != NULL)
 186                 sap->dp_tx->get_dev_info(dev_info);
 187
 188         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 189                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 190         dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
 191
 192         if (mae->status == SFC_MAE_STATUS_SUPPORTED ||
 193             mae->status == SFC_MAE_STATUS_ADMIN) {
 194                 dev_info->switch_info.name = dev->device->driver->name;
 195                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 196                 dev_info->switch_info.port_id = mae->switch_port_id;
 197         }
 198
 199         return 0;
 200 }
 201
 202 static const uint32_t *
 203 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 204 {
 205         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 206
 207         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 208 }
 209
 210 static int
 211 sfc_dev_configure(struct rte_eth_dev *dev)
 212 {
 213         struct rte_eth_dev_data *dev_data = dev->data;
 214         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 215         int rc;
 216
 217         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 218                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 219
 220         sfc_adapter_lock(sa);
 221         switch (sa->state) {
 222         case SFC_ETHDEV_CONFIGURED:
 223                 /* FALLTHROUGH */
 224         case SFC_ETHDEV_INITIALIZED:
 225                 rc = sfc_configure(sa);
 226                 break;
 227         default:
 228                 sfc_err(sa, "unexpected adapter state %u to configure",
 229                         sa->state);
 230                 rc = EINVAL;
 231                 break;
 232         }
 233         sfc_adapter_unlock(sa);
 234
 235         sfc_log_init(sa, "done %d", rc);
 236         SFC_ASSERT(rc >= 0);
 237         return -rc;
 238 }
 239
 240 static int
 241 sfc_dev_start(struct rte_eth_dev *dev)
 242 {
 243         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 244         int rc;
 245
 246         sfc_log_init(sa, "entry");
 247
 248         sfc_adapter_lock(sa);
 249         rc = sfc_start(sa);
 250         sfc_adapter_unlock(sa);
 251
 252         sfc_log_init(sa, "done %d", rc);
 253         SFC_ASSERT(rc >= 0);
 254         return -rc;
 255 }
 256
 257 static int
 258 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 259 {
 260         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 261         struct rte_eth_link current_link;
 262         int ret;
 263
 264         sfc_log_init(sa, "entry");
 265
 266         if (sa->state != SFC_ETHDEV_STARTED) {
 267                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 268         } else if (wait_to_complete) {
 269                 efx_link_mode_t link_mode;
 270
 271                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 272                         link_mode = EFX_LINK_UNKNOWN;
 273                 sfc_port_link_mode_to_info(link_mode, &current_link);
 274
 275         } else {
 276                 sfc_ev_mgmt_qpoll(sa);
 277                 rte_eth_linkstatus_get(dev, &current_link);
 278         }
 279
 280         ret = rte_eth_linkstatus_set(dev, &current_link);
 281         if (ret == 0)
 282                 sfc_notice(sa, "Link status is %s",
 283                            current_link.link_status ? "UP" : "DOWN");
 284
 285         return ret;
 286 }
 287
 288 static int
 289 sfc_dev_stop(struct rte_eth_dev *dev)
 290 {
 291         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 292
 293         sfc_log_init(sa, "entry");
 294
 295         sfc_adapter_lock(sa);
 296         sfc_stop(sa);
 297         sfc_adapter_unlock(sa);
 298
 299         sfc_log_init(sa, "done");
 300
 301         return 0;
 302 }
 303
 304 static int
 305 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 306 {
 307         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 308         int rc;
 309
 310         sfc_log_init(sa, "entry");
 311
 312         sfc_adapter_lock(sa);
 313         rc = sfc_start(sa);
 314         sfc_adapter_unlock(sa);
 315
 316         SFC_ASSERT(rc >= 0);
 317         return -rc;
 318 }
 319
 320 static int
 321 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 322 {
 323         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 324
 325         sfc_log_init(sa, "entry");
 326
 327         sfc_adapter_lock(sa);
 328         sfc_stop(sa);
 329         sfc_adapter_unlock(sa);
 330
 331         return 0;
 332 }
 333
 334 static void
 335 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 336 {
 337         free(dev->process_private);
 338         rte_eth_dev_release_port(dev);
 339 }
 340
 341 static int
 342 sfc_dev_close(struct rte_eth_dev *dev)
 343 {
 344         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 345
 346         sfc_log_init(sa, "entry");
 347
 348         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 349                 sfc_eth_dev_secondary_clear_ops(dev);
 350                 return 0;
 351         }
 352
 353         sfc_pre_detach(sa);
 354
 355         sfc_adapter_lock(sa);
 356         switch (sa->state) {
 357         case SFC_ETHDEV_STARTED:
 358                 sfc_stop(sa);
 359                 SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
 360                 /* FALLTHROUGH */
 361         case SFC_ETHDEV_CONFIGURED:
 362                 sfc_close(sa);
 363                 SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
 364                 /* FALLTHROUGH */
 365         case SFC_ETHDEV_INITIALIZED:
 366                 break;
 367         default:
 368                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 369                 break;
 370         }
 371
 372         /*
 373          * Cleanup all resources.
 374          * Rollback primary process sfc_eth_dev_init() below.
 375          */
 376
 377         sfc_eth_dev_clear_ops(dev);
 378
 379         sfc_nic_dma_detach(sa);
 380         sfc_detach(sa);
 381         sfc_unprobe(sa);
 382
 383         sfc_kvargs_cleanup(sa);
 384
 385         sfc_adapter_unlock(sa);
 386         sfc_adapter_lock_fini(sa);
 387
 388         sfc_log_init(sa, "done");
 389
 390         /* Required for logging, so cleanup last */
 391         sa->eth_dev = NULL;
 392
 393         free(sa);
 394
 395         return 0;
 396 }
 397
 398 static int
 399 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 400                    boolean_t enabled)
 401 {
 402         struct sfc_port *port;
 403         boolean_t *toggle;
 404         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 405         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 406         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 407         int rc = 0;
 408
 409         sfc_adapter_lock(sa);
 410
 411         port = &sa->port;
 412         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 413
 414         if (*toggle != enabled) {
 415                 *toggle = enabled;
 416
 417                 if (sfc_sa2shared(sa)->isolated) {
 418                         sfc_warn(sa, "isolated mode is active on the port");
 419                         sfc_warn(sa, "the change is to be applied on the next "
 420                                      "start provided that isolated mode is "
 421                                      "disabled prior the next start");
 422                 } else if ((sa->state == SFC_ETHDEV_STARTED) &&
 423                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 424                         *toggle = !(enabled);
 425                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 426                                  ((enabled) ? "enable" : "disable"), desc, rc);
 427
 428                         /*
 429                          * For promiscuous and all-multicast filters a
 430                          * permission failure should be reported as an
 431                          * unsupported filter.
 432                          */
 433                         if (rc == EPERM)
 434                                 rc = ENOTSUP;
 435                 }
 436         }
 437
 438         sfc_adapter_unlock(sa);
 439         return rc;
 440 }
 441
 442 static int
 443 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 444 {
 445         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 446
 447         SFC_ASSERT(rc >= 0);
 448         return -rc;
 449 }
 450
 451 static int
 452 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 453 {
 454         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 455
 456         SFC_ASSERT(rc >= 0);
 457         return -rc;
 458 }
 459
 460 static int
 461 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 462 {
 463         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 464
 465         SFC_ASSERT(rc >= 0);
 466         return -rc;
 467 }
 468
 469 static int
 470 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 471 {
 472         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 473
 474         SFC_ASSERT(rc >= 0);
 475         return -rc;
 476 }
 477
 478 static int
 479 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 480                    uint16_t nb_rx_desc, unsigned int socket_id,
 481                    const struct rte_eth_rxconf *rx_conf,
 482                    struct rte_mempool *mb_pool)
 483 {
 484         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 485         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 486         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
 487         struct sfc_rxq_info *rxq_info;
 488         sfc_sw_index_t sw_index;
 489         int rc;
 490
 491         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 492                      ethdev_qid, nb_rx_desc, socket_id);
 493
 494         sfc_adapter_lock(sa);
 495
 496         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
 497         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
 498                           rx_conf, mb_pool);
 499         if (rc != 0)
 500                 goto fail_rx_qinit;
 501
 502         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
 503         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
 504
 505         sfc_adapter_unlock(sa);
 506
 507         return 0;
 508
 509 fail_rx_qinit:
 510         sfc_adapter_unlock(sa);
 511         SFC_ASSERT(rc > 0);
 512         return -rc;
 513 }
 514
 515 static void
 516 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 517 {
 518         struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
 519         struct sfc_rxq *rxq;
 520         struct sfc_adapter *sa;
 521         sfc_sw_index_t sw_index;
 522
 523         if (dp_rxq == NULL)
 524                 return;
 525
 526         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 527         sa = rxq->evq->sa;
 528         sfc_adapter_lock(sa);
 529
 530         sw_index = dp_rxq->dpq.queue_id;
 531
 532         sfc_log_init(sa, "RxQ=%u", sw_index);
 533
 534         sfc_rx_qfini(sa, sw_index);
 535
 536         sfc_adapter_unlock(sa);
 537 }
 538
 539 static int
 540 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 541                    uint16_t nb_tx_desc, unsigned int socket_id,
 542                    const struct rte_eth_txconf *tx_conf)
 543 {
 544         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 545         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 546         struct sfc_txq_info *txq_info;
 547         sfc_sw_index_t sw_index;
 548         int rc;
 549
 550         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 551                      ethdev_qid, nb_tx_desc, socket_id);
 552
 553         sfc_adapter_lock(sa);
 554
 555         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
 556         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
 557         if (rc != 0)
 558                 goto fail_tx_qinit;
 559
 560         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
 561         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
 562
 563         sfc_adapter_unlock(sa);
 564         return 0;
 565
 566 fail_tx_qinit:
 567         sfc_adapter_unlock(sa);
 568         SFC_ASSERT(rc > 0);
 569         return -rc;
 570 }
 571
 572 static void
 573 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 574 {
 575         struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
 576         struct sfc_txq *txq;
 577         sfc_sw_index_t sw_index;
 578         struct sfc_adapter *sa;
 579
 580         if (dp_txq == NULL)
 581                 return;
 582
 583         txq = sfc_txq_by_dp_txq(dp_txq);
 584         sw_index = dp_txq->dpq.queue_id;
 585
 586         SFC_ASSERT(txq->evq != NULL);
 587         sa = txq->evq->sa;
 588
 589         sfc_log_init(sa, "TxQ = %u", sw_index);
 590
 591         sfc_adapter_lock(sa);
 592
 593         sfc_tx_qfini(sa, sw_index);
 594
 595         sfc_adapter_unlock(sa);
 596 }
 597
 598 static void
 599 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 600 {
 601         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 602         uint64_t pkts_sum = 0;
 603         uint64_t bytes_sum = 0;
 604         unsigned int i;
 605
 606         for (i = 0; i < sas->ethdev_rxq_count; ++i) {
 607                 struct sfc_rxq_info *rxq_info;
 608
 609                 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
 610                 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
 611                         union sfc_pkts_bytes qstats;
 612
 613                         sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
 614                         pkts_sum += qstats.pkts -
 615                                         sa->sw_stats.reset_rx_pkts[i];
 616                         bytes_sum += qstats.bytes -
 617                                         sa->sw_stats.reset_rx_bytes[i];
 618                 }
 619         }
 620
 621         *pkts = pkts_sum;
 622         *bytes = bytes_sum;
 623 }
 624
 625 static void
 626 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 627 {
 628         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 629         uint64_t pkts_sum = 0;
 630         uint64_t bytes_sum = 0;
 631         unsigned int i;
 632
 633         for (i = 0; i < sas->ethdev_txq_count; ++i) {
 634                 struct sfc_txq_info *txq_info;
 635
 636                 txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
 637                 if (txq_info->state & SFC_TXQ_INITIALIZED) {
 638                         union sfc_pkts_bytes qstats;
 639
 640                         sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
 641                         pkts_sum += qstats.pkts -
 642                                         sa->sw_stats.reset_tx_pkts[i];
 643                         bytes_sum += qstats.bytes -
 644                                         sa->sw_stats.reset_tx_bytes[i];
 645                 }
 646         }
 647
 648         *pkts = pkts_sum;
 649         *bytes = bytes_sum;
 650 }
 651
 652 /*
 653  * Some statistics are computed as A - B where A and B each increase
 654  * monotonically with some hardware counter(s) and the counters are read
 655  * asynchronously.
 656  *
 657  * If packet X is counted in A, but not counted in B yet, computed value is
 658  * greater than real.
 659  *
 660  * If packet X is not counted in A at the moment of reading the counter,
 661  * but counted in B at the moment of reading the counter, computed value
 662  * is less than real.
 663  *
 664  * However, counter which grows backward is worse evil than slightly wrong
 665  * value. So, let's try to guarantee that it never happens except may be
 666  * the case when the MAC stats are zeroed as a result of a NIC reset.
 667  */
 668 static void
 669 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 670 {
 671         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 672                 *stat = newval;
 673 }
 674
 675 static int
 676 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 677 {
 678         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 679         bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
 680         bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
 681         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 682         struct sfc_port *port = &sa->port;
 683         uint64_t *mac_stats;
 684         int ret;
 685
 686         sfc_adapter_lock(sa);
 687
 688         if (have_dp_rx_stats)
 689                 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
 690         if (have_dp_tx_stats)
 691                 sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
 692
 693         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 694         if (ret != 0)
 695                 goto unlock;
 696
 697         mac_stats = port->mac_stats_buf;
 698
 699         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 700                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 701                 if (!have_dp_rx_stats) {
 702                         stats->ipackets =
 703                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 704                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 705                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 706                         stats->ibytes =
 707                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 708                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 709                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 710
 711                         /* CRC is included in these stats, but shouldn't be */
 712                         stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 713                 }
 714                 if (!have_dp_tx_stats) {
 715                         stats->opackets =
 716                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 717                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 718                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 719                         stats->obytes =
 720                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 721                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 722                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 723
 724                         /* CRC is included in these stats, but shouldn't be */
 725                         stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 726                 }
 727                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 728                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 729         } else {
 730                 if (!have_dp_tx_stats) {
 731                         stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 732                         stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
 733                                 mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 734                 }
 735
 736                 /*
 737                  * Take into account stats which are whenever supported
 738                  * on EF10. If some stat is not supported by current
 739                  * firmware variant or HW revision, it is guaranteed
 740                  * to be zero in mac_stats.
 741                  */
 742                 stats->imissed =
 743                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 744                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 745                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 746                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 747                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 748                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 749                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 750                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 751                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 752                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 753                 stats->ierrors =
 754                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 755                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 756                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 757                 /* no oerrors counters supported on EF10 */
 758
 759                 if (!have_dp_rx_stats) {
 760                         /* Exclude missed, errors and pauses from Rx packets */
 761                         sfc_update_diff_stat(&port->ipackets,
 762                                 mac_stats[EFX_MAC_RX_PKTS] -
 763                                 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 764                                 stats->imissed - stats->ierrors);
 765                         stats->ipackets = port->ipackets;
 766                         stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
 767                                 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 768                 }
 769         }
 770
 771 unlock:
 772         sfc_adapter_unlock(sa);
 773         SFC_ASSERT(ret >= 0);
 774         return -ret;
 775 }
 776
 777 static int
 778 sfc_stats_reset(struct rte_eth_dev *dev)
 779 {
 780         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 781         struct sfc_port *port = &sa->port;
 782         int rc;
 783
 784         sfc_adapter_lock(sa);
 785
 786         if (sa->state != SFC_ETHDEV_STARTED) {
 787                 /*
 788                  * The operation cannot be done if port is not started; it
 789                  * will be scheduled to be done during the next port start
 790                  */
 791                 port->mac_stats_reset_pending = B_TRUE;
 792                 sfc_adapter_unlock(sa);
 793                 return 0;
 794         }
 795
 796         rc = sfc_port_reset_mac_stats(sa);
 797         if (rc != 0)
 798                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 799
 800         sfc_sw_xstats_reset(sa);
 801
 802         sfc_adapter_unlock(sa);
 803
 804         SFC_ASSERT(rc >= 0);
 805         return -rc;
 806 }
 807
 808 static unsigned int
 809 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 810 {
 811         struct sfc_port *port = &sa->port;
 812         unsigned int nb_supported;
 813
 814         sfc_adapter_lock(sa);
 815         nb_supported = port->mac_stats_nb_supported +
 816                        sfc_sw_xstats_get_nb_supported(sa);
 817         sfc_adapter_unlock(sa);
 818
 819         return nb_supported;
 820 }
 821
 822 static int
 823 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 824                unsigned int xstats_count)
 825 {
 826         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 827         unsigned int nb_written = 0;
 828         unsigned int nb_supported = 0;
 829         int rc;
 830
 831         if (unlikely(xstats == NULL))
 832                 return sfc_xstats_get_nb_supported(sa);
 833
 834         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 835         if (rc < 0)
 836                 return rc;
 837
 838         nb_supported = rc;
 839         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 840                                &nb_supported);
 841
 842         return nb_supported;
 843 }
 844
 845 static int
 846 sfc_xstats_get_names(struct rte_eth_dev *dev,
 847                      struct rte_eth_xstat_name *xstats_names,
 848                      unsigned int xstats_count)
 849 {
 850         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 851         struct sfc_port *port = &sa->port;
 852         unsigned int i;
 853         unsigned int nstats = 0;
 854         unsigned int nb_written = 0;
 855         int ret;
 856
 857         if (unlikely(xstats_names == NULL))
 858                 return sfc_xstats_get_nb_supported(sa);
 859
 860         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 861                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 862                         if (nstats < xstats_count) {
 863                                 strlcpy(xstats_names[nstats].name,
 864                                         efx_mac_stat_name(sa->nic, i),
 865                                         sizeof(xstats_names[0].name));
 866                                 nb_written++;
 867                         }
 868                         nstats++;
 869                 }
 870         }
 871
 872         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 873                                       &nb_written, &nstats);
 874         if (ret != 0) {
 875                 SFC_ASSERT(ret < 0);
 876                 return ret;
 877         }
 878
 879         return nstats;
 880 }
 881
 882 static int
 883 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 884                      uint64_t *values, unsigned int n)
 885 {
 886         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 887         struct sfc_port *port = &sa->port;
 888         unsigned int nb_supported;
 889         unsigned int i;
 890         int rc;
 891
 892         if (unlikely(ids == NULL || values == NULL))
 893                 return -EINVAL;
 894
 895         /*
 896          * Values array could be filled in nonsequential order. Fill values with
 897          * constant indicating invalid ID first.
 898          */
 899         for (i = 0; i < n; i++)
 900                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 901
 902         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 903         if (rc != 0)
 904                 return rc;
 905
 906         nb_supported = port->mac_stats_nb_supported;
 907         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 908
 909         /* Return number of written stats before invalid ID is encountered. */
 910         for (i = 0; i < n; i++) {
 911                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 912                         return i;
 913         }
 914
 915         return n;
 916 }
 917
 918 static int
 919 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 920                            const uint64_t *ids,
 921                            struct rte_eth_xstat_name *xstats_names,
 922                            unsigned int size)
 923 {
 924         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 925         struct sfc_port *port = &sa->port;
 926         unsigned int nb_supported;
 927         unsigned int i;
 928         int ret;
 929
 930         if (unlikely(xstats_names == NULL && ids != NULL) ||
 931             unlikely(xstats_names != NULL && ids == NULL))
 932                 return -EINVAL;
 933
 934         if (unlikely(xstats_names == NULL && ids == NULL))
 935                 return sfc_xstats_get_nb_supported(sa);
 936
 937         /*
 938          * Names array could be filled in nonsequential order. Fill names with
 939          * string indicating invalid ID first.
 940          */
 941         for (i = 0; i < size; i++)
 942                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 943
 944         sfc_adapter_lock(sa);
 945
 946         SFC_ASSERT(port->mac_stats_nb_supported <=
 947                    RTE_DIM(port->mac_stats_by_id));
 948
 949         for (i = 0; i < size; i++) {
 950                 if (ids[i] < port->mac_stats_nb_supported) {
 951                         strlcpy(xstats_names[i].name,
 952                                 efx_mac_stat_name(sa->nic,
 953                                                  port->mac_stats_by_id[ids[i]]),
 954                                 sizeof(xstats_names[0].name));
 955                 }
 956         }
 957
 958         nb_supported = port->mac_stats_nb_supported;
 959
 960         sfc_adapter_unlock(sa);
 961
 962         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 963                                             &nb_supported);
 964         if (ret != 0) {
 965                 SFC_ASSERT(ret < 0);
 966                 return ret;
 967         }
 968
 969         /* Return number of written names before invalid ID is encountered. */
 970         for (i = 0; i < size; i++) {
 971                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 972                         return i;
 973         }
 974
 975         return size;
 976 }
 977
 978 static int
 979 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 980 {
 981         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 982         unsigned int wanted_fc, link_fc;
 983
 984         memset(fc_conf, 0, sizeof(*fc_conf));
 985
 986         sfc_adapter_lock(sa);
 987
 988         if (sa->state == SFC_ETHDEV_STARTED)
 989                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 990         else
 991                 link_fc = sa->port.flow_ctrl;
 992
 993         switch (link_fc) {
 994         case 0:
 995                 fc_conf->mode = RTE_ETH_FC_NONE;
 996                 break;
 997         case EFX_FCNTL_RESPOND:
 998                 fc_conf->mode = RTE_ETH_FC_RX_PAUSE;
 999                 break;
1000         case EFX_FCNTL_GENERATE:
1001                 fc_conf->mode = RTE_ETH_FC_TX_PAUSE;
1002                 break;
1003         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
1004                 fc_conf->mode = RTE_ETH_FC_FULL;
1005                 break;
1006         default:
1007                 sfc_err(sa, "%s: unexpected flow control value %#x",
1008                         __func__, link_fc);
1009         }
1010
1011         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1012
1013         sfc_adapter_unlock(sa);
1014
1015         return 0;
1016 }
1017
1018 static int
1019 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1020 {
1021         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1022         struct sfc_port *port = &sa->port;
1023         unsigned int fcntl;
1024         int rc;
1025
1026         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1027             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1028             fc_conf->mac_ctrl_frame_fwd != 0) {
1029                 sfc_err(sa, "unsupported flow control settings specified");
1030                 rc = EINVAL;
1031                 goto fail_inval;
1032         }
1033
1034         switch (fc_conf->mode) {
1035         case RTE_ETH_FC_NONE:
1036                 fcntl = 0;
1037                 break;
1038         case RTE_ETH_FC_RX_PAUSE:
1039                 fcntl = EFX_FCNTL_RESPOND;
1040                 break;
1041         case RTE_ETH_FC_TX_PAUSE:
1042                 fcntl = EFX_FCNTL_GENERATE;
1043                 break;
1044         case RTE_ETH_FC_FULL:
1045                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1046                 break;
1047         default:
1048                 rc = EINVAL;
1049                 goto fail_inval;
1050         }
1051
1052         sfc_adapter_lock(sa);
1053
1054         if (sa->state == SFC_ETHDEV_STARTED) {
1055                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1056                 if (rc != 0)
1057                         goto fail_mac_fcntl_set;
1058         }
1059
1060         port->flow_ctrl = fcntl;
1061         port->flow_ctrl_autoneg = fc_conf->autoneg;
1062
1063         sfc_adapter_unlock(sa);
1064
1065         return 0;
1066
1067 fail_mac_fcntl_set:
1068         sfc_adapter_unlock(sa);
1069 fail_inval:
1070         SFC_ASSERT(rc > 0);
1071         return -rc;
1072 }
1073
1074 static int
1075 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1076 {
1077         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1078         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1079         boolean_t scatter_enabled;
1080         const char *error;
1081         unsigned int i;
1082
1083         for (i = 0; i < sas->rxq_count; i++) {
1084                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1085                         continue;
1086
1087                 scatter_enabled = (sas->rxq_info[i].type_flags &
1088                                    EFX_RXQ_FLAG_SCATTER);
1089
1090                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1091                                           encp->enc_rx_prefix_size,
1092                                           scatter_enabled,
1093                                           encp->enc_rx_scatter_max, &error)) {
1094                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1095                                 error);
1096                         return EINVAL;
1097                 }
1098         }
1099
1100         return 0;
1101 }
1102
1103 static int
1104 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1105 {
1106         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1107         size_t pdu = EFX_MAC_PDU(mtu);
1108         size_t old_pdu;
1109         int rc;
1110
1111         sfc_log_init(sa, "mtu=%u", mtu);
1112
1113         rc = EINVAL;
1114         if (pdu < EFX_MAC_PDU_MIN) {
1115                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1116                         (unsigned int)mtu, (unsigned int)pdu,
1117                         EFX_MAC_PDU_MIN);
1118                 goto fail_inval;
1119         }
1120         if (pdu > EFX_MAC_PDU_MAX) {
1121                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1122                         (unsigned int)mtu, (unsigned int)pdu,
1123                         (unsigned int)EFX_MAC_PDU_MAX);
1124                 goto fail_inval;
1125         }
1126
1127         sfc_adapter_lock(sa);
1128
1129         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1130         if (rc != 0)
1131                 goto fail_check_scatter;
1132
1133         if (pdu != sa->port.pdu) {
1134                 if (sa->state == SFC_ETHDEV_STARTED) {
1135                         sfc_stop(sa);
1136
1137                         old_pdu = sa->port.pdu;
1138                         sa->port.pdu = pdu;
1139                         rc = sfc_start(sa);
1140                         if (rc != 0)
1141                                 goto fail_start;
1142                 } else {
1143                         sa->port.pdu = pdu;
1144                 }
1145         }
1146
1147         sfc_adapter_unlock(sa);
1148
1149         sfc_log_init(sa, "done");
1150         return 0;
1151
1152 fail_start:
1153         sa->port.pdu = old_pdu;
1154         if (sfc_start(sa) != 0)
1155                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1156                         "PDU max size - port is stopped",
1157                         (unsigned int)pdu, (unsigned int)old_pdu);
1158
1159 fail_check_scatter:
1160         sfc_adapter_unlock(sa);
1161
1162 fail_inval:
1163         sfc_log_init(sa, "failed %d", rc);
1164         SFC_ASSERT(rc > 0);
1165         return -rc;
1166 }
1167 static int
1168 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1169 {
1170         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1171         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1172         struct sfc_port *port = &sa->port;
1173         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1174         int rc = 0;
1175
1176         sfc_adapter_lock(sa);
1177
1178         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1179                 goto unlock;
1180
1181         /*
1182          * Copy the address to the device private data so that
1183          * it could be recalled in the case of adapter restart.
1184          */
1185         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1186
1187         /*
1188          * Neither of the two following checks can return
1189          * an error. The new MAC address is preserved in
1190          * the device private data and can be activated
1191          * on the next port start if the user prevents
1192          * isolated mode from being enabled.
1193          */
1194         if (sfc_sa2shared(sa)->isolated) {
1195                 sfc_warn(sa, "isolated mode is active on the port");
1196                 sfc_warn(sa, "will not set MAC address");
1197                 goto unlock;
1198         }
1199
1200         if (sa->state != SFC_ETHDEV_STARTED) {
1201                 sfc_notice(sa, "the port is not started");
1202                 sfc_notice(sa, "the new MAC address will be set on port start");
1203
1204                 goto unlock;
1205         }
1206
1207         if (encp->enc_allow_set_mac_with_installed_filters) {
1208                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1209                 if (rc != 0) {
1210                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1211                         goto unlock;
1212                 }
1213
1214                 /*
1215                  * Changing the MAC address by means of MCDI request
1216                  * has no effect on received traffic, therefore
1217                  * we also need to update unicast filters
1218                  */
1219                 rc = sfc_set_rx_mode_unchecked(sa);
1220                 if (rc != 0) {
1221                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1222                         /* Rollback the old address */
1223                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1224                         (void)sfc_set_rx_mode_unchecked(sa);
1225                 }
1226         } else {
1227                 sfc_warn(sa, "cannot set MAC address with filters installed");
1228                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1229                 sfc_warn(sa, "(some traffic may be dropped)");
1230
1231                 /*
1232                  * Since setting MAC address with filters installed is not
1233                  * allowed on the adapter, the new MAC address will be set
1234                  * by means of adapter restart. sfc_start() shall retrieve
1235                  * the new address from the device private data and set it.
1236                  */
1237                 sfc_stop(sa);
1238                 rc = sfc_start(sa);
1239                 if (rc != 0)
1240                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1241         }
1242
1243 unlock:
1244         if (rc != 0)
1245                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1246
1247         sfc_adapter_unlock(sa);
1248
1249         SFC_ASSERT(rc >= 0);
1250         return -rc;
1251 }
1252
1253
1254 static int
1255 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1256                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1257 {
1258         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1259         struct sfc_port *port = &sa->port;
1260         uint8_t *mc_addrs = port->mcast_addrs;
1261         int rc;
1262         unsigned int i;
1263
1264         if (sfc_sa2shared(sa)->isolated) {
1265                 sfc_err(sa, "isolated mode is active on the port");
1266                 sfc_err(sa, "will not set multicast address list");
1267                 return -ENOTSUP;
1268         }
1269
1270         if (mc_addrs == NULL)
1271                 return -ENOBUFS;
1272
1273         if (nb_mc_addr > port->max_mcast_addrs) {
1274                 sfc_err(sa, "too many multicast addresses: %u > %u",
1275                          nb_mc_addr, port->max_mcast_addrs);
1276                 return -EINVAL;
1277         }
1278
1279         for (i = 0; i < nb_mc_addr; ++i) {
1280                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1281                                  EFX_MAC_ADDR_LEN);
1282                 mc_addrs += EFX_MAC_ADDR_LEN;
1283         }
1284
1285         port->nb_mcast_addrs = nb_mc_addr;
1286
1287         if (sa->state != SFC_ETHDEV_STARTED)
1288                 return 0;
1289
1290         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1291                                         port->nb_mcast_addrs);
1292         if (rc != 0)
1293                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1294
1295         SFC_ASSERT(rc >= 0);
1296         return -rc;
1297 }
1298
1299 /*
1300  * The function is used by the secondary process as well. It must not
1301  * use any process-local pointers from the adapter data.
1302  */
1303 static void
1304 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1305                       struct rte_eth_rxq_info *qinfo)
1306 {
1307         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1308         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1309         struct sfc_rxq_info *rxq_info;
1310
1311         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1312
1313         qinfo->mp = rxq_info->refill_mb_pool;
1314         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1315         qinfo->conf.rx_drop_en = 1;
1316         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1317         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1318         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1319                 qinfo->conf.offloads |= RTE_ETH_RX_OFFLOAD_SCATTER;
1320                 qinfo->scattered_rx = 1;
1321         }
1322         qinfo->nb_desc = rxq_info->entries;
1323 }
1324
1325 /*
1326  * The function is used by the secondary process as well. It must not
1327  * use any process-local pointers from the adapter data.
1328  */
1329 static void
1330 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1331                       struct rte_eth_txq_info *qinfo)
1332 {
1333         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1334         struct sfc_txq_info *txq_info;
1335
1336         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1337
1338         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1339
1340         memset(qinfo, 0, sizeof(*qinfo));
1341
1342         qinfo->conf.offloads = txq_info->offloads;
1343         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1344         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1345         qinfo->nb_desc = txq_info->entries;
1346 }
1347
1348 /*
1349  * The function is used by the secondary process as well. It must not
1350  * use any process-local pointers from the adapter data.
1351  */
1352 static uint32_t
1353 sfc_rx_queue_count(void *rx_queue)
1354 {
1355         struct sfc_dp_rxq *dp_rxq = rx_queue;
1356         const struct sfc_dp_rx *dp_rx;
1357         struct sfc_rxq_info *rxq_info;
1358
1359         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1360         rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1361
1362         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1363                 return 0;
1364
1365         return dp_rx->qdesc_npending(dp_rxq);
1366 }
1367
1368 /*
1369  * The function is used by the secondary process as well. It must not
1370  * use any process-local pointers from the adapter data.
1371  */
1372 static int
1373 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1374 {
1375         struct sfc_dp_rxq *dp_rxq = queue;
1376         const struct sfc_dp_rx *dp_rx;
1377
1378         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1379
1380         return dp_rx->qdesc_status(dp_rxq, offset);
1381 }
1382
1383 /*
1384  * The function is used by the secondary process as well. It must not
1385  * use any process-local pointers from the adapter data.
1386  */
1387 static int
1388 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1389 {
1390         struct sfc_dp_txq *dp_txq = queue;
1391         const struct sfc_dp_tx *dp_tx;
1392
1393         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1394
1395         return dp_tx->qdesc_status(dp_txq, offset);
1396 }
1397
1398 static int
1399 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1400 {
1401         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1402         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1403         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1404         struct sfc_rxq_info *rxq_info;
1405         sfc_sw_index_t sw_index;
1406         int rc;
1407
1408         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1409
1410         sfc_adapter_lock(sa);
1411
1412         rc = EINVAL;
1413         if (sa->state != SFC_ETHDEV_STARTED)
1414                 goto fail_not_started;
1415
1416         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1417         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1418                 goto fail_not_setup;
1419
1420         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1421         rc = sfc_rx_qstart(sa, sw_index);
1422         if (rc != 0)
1423                 goto fail_rx_qstart;
1424
1425         rxq_info->deferred_started = B_TRUE;
1426
1427         sfc_adapter_unlock(sa);
1428
1429         return 0;
1430
1431 fail_rx_qstart:
1432 fail_not_setup:
1433 fail_not_started:
1434         sfc_adapter_unlock(sa);
1435         SFC_ASSERT(rc > 0);
1436         return -rc;
1437 }
1438
1439 static int
1440 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1441 {
1442         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1443         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1444         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1445         struct sfc_rxq_info *rxq_info;
1446         sfc_sw_index_t sw_index;
1447
1448         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1449
1450         sfc_adapter_lock(sa);
1451
1452         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1453         sfc_rx_qstop(sa, sw_index);
1454
1455         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1456         rxq_info->deferred_started = B_FALSE;
1457
1458         sfc_adapter_unlock(sa);
1459
1460         return 0;
1461 }
1462
1463 static int
1464 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1465 {
1466         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1467         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1468         struct sfc_txq_info *txq_info;
1469         sfc_sw_index_t sw_index;
1470         int rc;
1471
1472         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1473
1474         sfc_adapter_lock(sa);
1475
1476         rc = EINVAL;
1477         if (sa->state != SFC_ETHDEV_STARTED)
1478                 goto fail_not_started;
1479
1480         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1481         if (txq_info->state != SFC_TXQ_INITIALIZED)
1482                 goto fail_not_setup;
1483
1484         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1485         rc = sfc_tx_qstart(sa, sw_index);
1486         if (rc != 0)
1487                 goto fail_tx_qstart;
1488
1489         txq_info->deferred_started = B_TRUE;
1490
1491         sfc_adapter_unlock(sa);
1492         return 0;
1493
1494 fail_tx_qstart:
1495
1496 fail_not_setup:
1497 fail_not_started:
1498         sfc_adapter_unlock(sa);
1499         SFC_ASSERT(rc > 0);
1500         return -rc;
1501 }
1502
1503 static int
1504 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1505 {
1506         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1507         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1508         struct sfc_txq_info *txq_info;
1509         sfc_sw_index_t sw_index;
1510
1511         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1512
1513         sfc_adapter_lock(sa);
1514
1515         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1516         sfc_tx_qstop(sa, sw_index);
1517
1518         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1519         txq_info->deferred_started = B_FALSE;
1520
1521         sfc_adapter_unlock(sa);
1522         return 0;
1523 }
1524
1525 static efx_tunnel_protocol_t
1526 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1527 {
1528         switch (rte_type) {
1529         case RTE_ETH_TUNNEL_TYPE_VXLAN:
1530                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1531         case RTE_ETH_TUNNEL_TYPE_GENEVE:
1532                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1533         default:
1534                 return EFX_TUNNEL_NPROTOS;
1535         }
1536 }
1537
1538 enum sfc_udp_tunnel_op_e {
1539         SFC_UDP_TUNNEL_ADD_PORT,
1540         SFC_UDP_TUNNEL_DEL_PORT,
1541 };
1542
1543 static int
1544 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1545                       struct rte_eth_udp_tunnel *tunnel_udp,
1546                       enum sfc_udp_tunnel_op_e op)
1547 {
1548         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1549         efx_tunnel_protocol_t tunnel_proto;
1550         int rc;
1551
1552         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1553                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1554                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1555                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1556
1557         tunnel_proto =
1558                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1559         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1560                 rc = ENOTSUP;
1561                 goto fail_bad_proto;
1562         }
1563
1564         sfc_adapter_lock(sa);
1565
1566         switch (op) {
1567         case SFC_UDP_TUNNEL_ADD_PORT:
1568                 rc = efx_tunnel_config_udp_add(sa->nic,
1569                                                tunnel_udp->udp_port,
1570                                                tunnel_proto);
1571                 break;
1572         case SFC_UDP_TUNNEL_DEL_PORT:
1573                 rc = efx_tunnel_config_udp_remove(sa->nic,
1574                                                   tunnel_udp->udp_port,
1575                                                   tunnel_proto);
1576                 break;
1577         default:
1578                 rc = EINVAL;
1579                 goto fail_bad_op;
1580         }
1581
1582         if (rc != 0)
1583                 goto fail_op;
1584
1585         if (sa->state == SFC_ETHDEV_STARTED) {
1586                 rc = efx_tunnel_reconfigure(sa->nic);
1587                 if (rc == EAGAIN) {
1588                         /*
1589                          * Configuration is accepted by FW and MC reboot
1590                          * is initiated to apply the changes. MC reboot
1591                          * will be handled in a usual way (MC reboot
1592                          * event on management event queue and adapter
1593                          * restart).
1594                          */
1595                         rc = 0;
1596                 } else if (rc != 0) {
1597                         goto fail_reconfigure;
1598                 }
1599         }
1600
1601         sfc_adapter_unlock(sa);
1602         return 0;
1603
1604 fail_reconfigure:
1605         /* Remove/restore entry since the change makes the trouble */
1606         switch (op) {
1607         case SFC_UDP_TUNNEL_ADD_PORT:
1608                 (void)efx_tunnel_config_udp_remove(sa->nic,
1609                                                    tunnel_udp->udp_port,
1610                                                    tunnel_proto);
1611                 break;
1612         case SFC_UDP_TUNNEL_DEL_PORT:
1613                 (void)efx_tunnel_config_udp_add(sa->nic,
1614                                                 tunnel_udp->udp_port,
1615                                                 tunnel_proto);
1616                 break;
1617         }
1618
1619 fail_op:
1620 fail_bad_op:
1621         sfc_adapter_unlock(sa);
1622
1623 fail_bad_proto:
1624         SFC_ASSERT(rc > 0);
1625         return -rc;
1626 }
1627
1628 static int
1629 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1630                             struct rte_eth_udp_tunnel *tunnel_udp)
1631 {
1632         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1633 }
1634
1635 static int
1636 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1637                             struct rte_eth_udp_tunnel *tunnel_udp)
1638 {
1639         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1640 }
1641
1642 /*
1643  * The function is used by the secondary process as well. It must not
1644  * use any process-local pointers from the adapter data.
1645  */
1646 static int
1647 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1648                           struct rte_eth_rss_conf *rss_conf)
1649 {
1650         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1651         struct sfc_rss *rss = &sas->rss;
1652
1653         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1654                 return -ENOTSUP;
1655
1656         /*
1657          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1658          * hence, conversion is done here to derive a correct set of RTE_ETH_RSS
1659          * flags which corresponds to the active EFX configuration stored
1660          * locally in 'sfc_adapter' and kept up-to-date
1661          */
1662         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1663         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1664         if (rss_conf->rss_key != NULL)
1665                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1666
1667         return 0;
1668 }
1669
1670 static int
1671 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1672                         struct rte_eth_rss_conf *rss_conf)
1673 {
1674         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1675         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1676         unsigned int efx_hash_types;
1677         unsigned int n_contexts;
1678         unsigned int mode_i = 0;
1679         unsigned int key_i = 0;
1680         uint32_t contexts[2];
1681         unsigned int i = 0;
1682         int rc = 0;
1683
1684         if (sfc_sa2shared(sa)->isolated)
1685                 return -ENOTSUP;
1686
1687         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1688                 sfc_err(sa, "RSS is not available");
1689                 return -ENOTSUP;
1690         }
1691
1692         if (rss->channels == 0) {
1693                 sfc_err(sa, "RSS is not configured");
1694                 return -EINVAL;
1695         }
1696
1697         if ((rss_conf->rss_key != NULL) &&
1698             (rss_conf->rss_key_len != sizeof(rss->key))) {
1699                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1700                         sizeof(rss->key));
1701                 return -EINVAL;
1702         }
1703
1704         sfc_adapter_lock(sa);
1705
1706         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1707         if (rc != 0)
1708                 goto fail_rx_hf_rte_to_efx;
1709
1710         contexts[0] = EFX_RSS_CONTEXT_DEFAULT;
1711         contexts[1] = rss->dummy_ctx.nic_handle;
1712         n_contexts = (rss->dummy_ctx.nic_handle_refcnt == 0) ? 1 : 2;
1713
1714         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1715                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1716                                            rss->hash_alg, efx_hash_types,
1717                                            B_TRUE);
1718                 if (rc != 0)
1719                         goto fail_scale_mode_set;
1720         }
1721
1722         if (rss_conf->rss_key != NULL) {
1723                 if (sa->state == SFC_ETHDEV_STARTED) {
1724                         for (key_i = 0; key_i < n_contexts; key_i++) {
1725                                 rc = efx_rx_scale_key_set(sa->nic,
1726                                                           contexts[key_i],
1727                                                           rss_conf->rss_key,
1728                                                           sizeof(rss->key));
1729                                 if (rc != 0)
1730                                         goto fail_scale_key_set;
1731                         }
1732                 }
1733
1734                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1735         }
1736
1737         rss->hash_types = efx_hash_types;
1738
1739         sfc_adapter_unlock(sa);
1740
1741         return 0;
1742
1743 fail_scale_key_set:
1744         for (i = 0; i < key_i; i++) {
1745                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1746                                          sizeof(rss->key)) != 0)
1747                         sfc_err(sa, "failed to restore RSS key");
1748         }
1749
1750 fail_scale_mode_set:
1751         for (i = 0; i < mode_i; i++) {
1752                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1753                                           EFX_RX_HASHALG_TOEPLITZ,
1754                                           rss->hash_types, B_TRUE) != 0)
1755                         sfc_err(sa, "failed to restore RSS mode");
1756         }
1757
1758 fail_rx_hf_rte_to_efx:
1759         sfc_adapter_unlock(sa);
1760         return -rc;
1761 }
1762
1763 /*
1764  * The function is used by the secondary process as well. It must not
1765  * use any process-local pointers from the adapter data.
1766  */
1767 static int
1768 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1769                        struct rte_eth_rss_reta_entry64 *reta_conf,
1770                        uint16_t reta_size)
1771 {
1772         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1773         struct sfc_rss *rss = &sas->rss;
1774         int entry;
1775
1776         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1777                 return -ENOTSUP;
1778
1779         if (rss->channels == 0)
1780                 return -EINVAL;
1781
1782         if (reta_size != EFX_RSS_TBL_SIZE)
1783                 return -EINVAL;
1784
1785         for (entry = 0; entry < reta_size; entry++) {
1786                 int grp = entry / RTE_ETH_RETA_GROUP_SIZE;
1787                 int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1788
1789                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1790                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1791         }
1792
1793         return 0;
1794 }
1795
1796 static int
1797 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1798                         struct rte_eth_rss_reta_entry64 *reta_conf,
1799                         uint16_t reta_size)
1800 {
1801         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1802         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1803         unsigned int *rss_tbl_new;
1804         uint16_t entry;
1805         int rc = 0;
1806
1807
1808         if (sfc_sa2shared(sa)->isolated)
1809                 return -ENOTSUP;
1810
1811         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1812                 sfc_err(sa, "RSS is not available");
1813                 return -ENOTSUP;
1814         }
1815
1816         if (rss->channels == 0) {
1817                 sfc_err(sa, "RSS is not configured");
1818                 return -EINVAL;
1819         }
1820
1821         if (reta_size != EFX_RSS_TBL_SIZE) {
1822                 sfc_err(sa, "RETA size is wrong (should be %u)",
1823                         EFX_RSS_TBL_SIZE);
1824                 return -EINVAL;
1825         }
1826
1827         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1828         if (rss_tbl_new == NULL)
1829                 return -ENOMEM;
1830
1831         sfc_adapter_lock(sa);
1832
1833         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1834
1835         for (entry = 0; entry < reta_size; entry++) {
1836                 int grp_idx = entry % RTE_ETH_RETA_GROUP_SIZE;
1837                 struct rte_eth_rss_reta_entry64 *grp;
1838
1839                 grp = &reta_conf[entry / RTE_ETH_RETA_GROUP_SIZE];
1840
1841                 if (grp->mask & (1ull << grp_idx)) {
1842                         if (grp->reta[grp_idx] >= rss->channels) {
1843                                 rc = EINVAL;
1844                                 goto bad_reta_entry;
1845                         }
1846                         rss_tbl_new[entry] = grp->reta[grp_idx];
1847                 }
1848         }
1849
1850         if (sa->state == SFC_ETHDEV_STARTED) {
1851                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1852                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1853                 if (rc != 0)
1854                         goto fail_scale_tbl_set;
1855         }
1856
1857         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1858
1859 fail_scale_tbl_set:
1860 bad_reta_entry:
1861         sfc_adapter_unlock(sa);
1862
1863         rte_free(rss_tbl_new);
1864
1865         SFC_ASSERT(rc >= 0);
1866         return -rc;
1867 }
1868
1869 static int
1870 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1871                      const struct rte_flow_ops **ops)
1872 {
1873         *ops = &sfc_flow_ops;
1874         return 0;
1875 }
1876
1877 static int
1878 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1879 {
1880         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1881
1882         /*
1883          * If Rx datapath does not provide callback to check mempool,
1884          * all pools are supported.
1885          */
1886         if (sap->dp_rx->pool_ops_supported == NULL)
1887                 return 1;
1888
1889         return sap->dp_rx->pool_ops_supported(pool);
1890 }
1891
1892 static int
1893 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1894 {
1895         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1896         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1897         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1898         struct sfc_rxq_info *rxq_info;
1899
1900         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1901
1902         return sap->dp_rx->intr_enable(rxq_info->dp);
1903 }
1904
1905 static int
1906 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1907 {
1908         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1909         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1910         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1911         struct sfc_rxq_info *rxq_info;
1912
1913         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1914
1915         return sap->dp_rx->intr_disable(rxq_info->dp);
1916 }
1917
1918 struct sfc_mport_journal_ctx {
1919         struct sfc_adapter              *sa;
1920         uint16_t                        switch_domain_id;
1921         uint32_t                        mcdi_handle;
1922         bool                            controllers_assigned;
1923         efx_pcie_interface_t            *controllers;
1924         size_t                          nb_controllers;
1925 };
1926
1927 static int
1928 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1929                                efx_pcie_interface_t intf)
1930 {
1931         efx_pcie_interface_t *new_controllers;
1932         size_t i, target;
1933         size_t new_size;
1934
1935         if (ctx->controllers == NULL) {
1936                 ctx->controllers = rte_malloc("sfc_controller_mapping",
1937                                               sizeof(ctx->controllers[0]), 0);
1938                 if (ctx->controllers == NULL)
1939                         return ENOMEM;
1940
1941                 ctx->controllers[0] = intf;
1942                 ctx->nb_controllers = 1;
1943
1944                 return 0;
1945         }
1946
1947         for (i = 0; i < ctx->nb_controllers; i++) {
1948                 if (ctx->controllers[i] == intf)
1949                         return 0;
1950                 if (ctx->controllers[i] > intf)
1951                         break;
1952         }
1953         target = i;
1954
1955         ctx->nb_controllers += 1;
1956         new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1957
1958         new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1959         if (new_controllers == NULL) {
1960                 rte_free(ctx->controllers);
1961                 return ENOMEM;
1962         }
1963         ctx->controllers = new_controllers;
1964
1965         for (i = target + 1; i < ctx->nb_controllers; i++)
1966                 ctx->controllers[i] = ctx->controllers[i - 1];
1967
1968         ctx->controllers[target] = intf;
1969
1970         return 0;
1971 }
1972
1973 static efx_rc_t
1974 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1975                                 efx_mport_desc_t *mport)
1976 {
1977         struct sfc_mae_switch_port_request req;
1978         efx_mport_sel_t entity_selector;
1979         efx_mport_sel_t ethdev_mport;
1980         uint16_t switch_port_id;
1981         efx_rc_t efx_rc;
1982         int rc;
1983
1984         sfc_dbg(ctx->sa,
1985                 "processing mport id %u (controller %u pf %u vf %u)",
1986                 mport->emd_id.id, mport->emd_vnic.ev_intf,
1987                 mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1988         efx_mae_mport_invalid(&ethdev_mport);
1989
1990         if (!ctx->controllers_assigned) {
1991                 rc = sfc_journal_ctx_add_controller(ctx,
1992                                                     mport->emd_vnic.ev_intf);
1993                 if (rc != 0)
1994                         return rc;
1995         }
1996
1997         /* Build Mport selector */
1998         efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
1999                                                 mport->emd_vnic.ev_pf,
2000                                                 mport->emd_vnic.ev_vf,
2001                                                 &entity_selector);
2002         if (efx_rc != 0) {
2003                 sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
2004                         mport->emd_vnic.ev_intf,
2005                         mport->emd_vnic.ev_pf,
2006                         mport->emd_vnic.ev_vf);
2007                 return efx_rc;
2008         }
2009
2010         rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2011                                               &entity_selector,
2012                                               SFC_MAE_SWITCH_PORT_REPRESENTOR,
2013                                               &switch_port_id);
2014         switch (rc) {
2015         case 0:
2016                 /* Already registered */
2017                 break;
2018         case ENOENT:
2019                 /*
2020                  * No representor has been created for this entity.
2021                  * Create a dummy switch registry entry with an invalid ethdev
2022                  * mport selector. When a corresponding representor is created,
2023                  * this entry will be updated.
2024                  */
2025                 req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2026                 req.entity_mportp = &entity_selector;
2027                 req.ethdev_mportp = &ethdev_mport;
2028                 req.ethdev_port_id = RTE_MAX_ETHPORTS;
2029                 req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2030                 req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2031                 req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2032
2033                 rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2034                                                 &req, &switch_port_id);
2035                 if (rc != 0) {
2036                         sfc_err(ctx->sa,
2037                                 "failed to assign MAE switch port for c%upf%uvf%u: %s",
2038                                 mport->emd_vnic.ev_intf,
2039                                 mport->emd_vnic.ev_pf,
2040                                 mport->emd_vnic.ev_vf,
2041                                 rte_strerror(rc));
2042                         return rc;
2043                 }
2044                 break;
2045         default:
2046                 sfc_err(ctx->sa, "failed to find MAE switch port for c%upf%uvf%u: %s",
2047                         mport->emd_vnic.ev_intf,
2048                         mport->emd_vnic.ev_pf,
2049                         mport->emd_vnic.ev_vf,
2050                         rte_strerror(rc));
2051                 return rc;
2052         }
2053
2054         return 0;
2055 }
2056
2057 static efx_rc_t
2058 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2059                              size_t mport_len)
2060 {
2061         struct sfc_mport_journal_ctx *ctx = data;
2062
2063         if (ctx == NULL || ctx->sa == NULL) {
2064                 sfc_err(ctx->sa, "received NULL context or SFC adapter");
2065                 return EINVAL;
2066         }
2067
2068         if (mport_len != sizeof(*mport)) {
2069                 sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2070                 return EINVAL;
2071         }
2072
2073         SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2074
2075         /*
2076          * If a zombie flag is set, it means the mport has been marked for
2077          * deletion and cannot be used for any new operations. The mport will
2078          * be destroyed completely once all references to it are released.
2079          */
2080         if (mport->emd_zombie) {
2081                 sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2082                 return 0;
2083         }
2084         if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2085                 sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2086                 return 0;
2087         }
2088         if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2089                 sfc_dbg(ctx->sa, "mport is not a function, skipping");
2090                 return 0;
2091         }
2092         if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2093                 sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2094                 return 0;
2095         }
2096
2097         return sfc_process_mport_journal_entry(ctx, mport);
2098 }
2099
2100 static int
2101 sfc_process_mport_journal(struct sfc_adapter *sa)
2102 {
2103         struct sfc_mport_journal_ctx ctx;
2104         const efx_pcie_interface_t *controllers;
2105         size_t nb_controllers;
2106         efx_rc_t efx_rc;
2107         int rc;
2108
2109         memset(&ctx, 0, sizeof(ctx));
2110         ctx.sa = sa;
2111         ctx.switch_domain_id = sa->mae.switch_domain_id;
2112
2113         efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2114         if (efx_rc != 0) {
2115                 sfc_err(sa, "failed to get own MCDI handle");
2116                 SFC_ASSERT(efx_rc > 0);
2117                 return efx_rc;
2118         }
2119
2120         rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2121                                                &controllers, &nb_controllers);
2122         if (rc != 0) {
2123                 sfc_err(sa, "failed to get controller mapping");
2124                 return rc;
2125         }
2126
2127         ctx.controllers_assigned = controllers != NULL;
2128         ctx.controllers = NULL;
2129         ctx.nb_controllers = 0;
2130
2131         efx_rc = efx_mae_read_mport_journal(sa->nic,
2132                                             sfc_process_mport_journal_cb, &ctx);
2133         if (efx_rc != 0) {
2134                 sfc_err(sa, "failed to process MAE mport journal");
2135                 SFC_ASSERT(efx_rc > 0);
2136                 return efx_rc;
2137         }
2138
2139         if (controllers == NULL) {
2140                 rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2141                                                            ctx.controllers,
2142                                                            ctx.nb_controllers);
2143                 if (rc != 0)
2144                         return rc;
2145         }
2146
2147         return 0;
2148 }
2149
2150 static void
2151 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2152                           const efx_mport_sel_t *ethdev_mportp __rte_unused,
2153                           uint16_t ethdev_port_id __rte_unused,
2154                           const efx_mport_sel_t *entity_mportp __rte_unused,
2155                           uint16_t switch_port_id __rte_unused,
2156                           union sfc_mae_switch_port_data *port_datap
2157                                 __rte_unused,
2158                           void *user_datap)
2159 {
2160         int *counter = user_datap;
2161
2162         SFC_ASSERT(counter != NULL);
2163
2164         if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2165                 (*counter)++;
2166 }
2167
2168 struct sfc_get_representors_ctx {
2169         struct rte_eth_representor_info *info;
2170         struct sfc_adapter              *sa;
2171         uint16_t                        switch_domain_id;
2172         const efx_pcie_interface_t      *controllers;
2173         size_t                          nb_controllers;
2174 };
2175
2176 static void
2177 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2178                         const efx_mport_sel_t *ethdev_mportp __rte_unused,
2179                         uint16_t ethdev_port_id __rte_unused,
2180                         const efx_mport_sel_t *entity_mportp __rte_unused,
2181                         uint16_t switch_port_id,
2182                         union sfc_mae_switch_port_data *port_datap,
2183                         void *user_datap)
2184 {
2185         struct sfc_get_representors_ctx *ctx = user_datap;
2186         struct rte_eth_representor_range *range;
2187         int ret;
2188         int rc;
2189
2190         SFC_ASSERT(ctx != NULL);
2191         SFC_ASSERT(ctx->info != NULL);
2192         SFC_ASSERT(ctx->sa != NULL);
2193
2194         if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2195                 sfc_dbg(ctx->sa, "not a representor, skipping");
2196                 return;
2197         }
2198         if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2199                 sfc_dbg(ctx->sa, "info structure is full already");
2200                 return;
2201         }
2202
2203         range = &ctx->info->ranges[ctx->info->nb_ranges];
2204         rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2205                                                     ctx->nb_controllers,
2206                                                     port_datap->repr.intf,
2207                                                     &range->controller);
2208         if (rc != 0) {
2209                 sfc_err(ctx->sa, "invalid representor controller: %d",
2210                         port_datap->repr.intf);
2211                 range->controller = -1;
2212         }
2213         range->pf = port_datap->repr.pf;
2214         range->id_base = switch_port_id;
2215         range->id_end = switch_port_id;
2216
2217         if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2218                 range->type = RTE_ETH_REPRESENTOR_VF;
2219                 range->vf = port_datap->repr.vf;
2220                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2221                                "c%dpf%dvf%d", range->controller, range->pf,
2222                                range->vf);
2223         } else {
2224                 range->type = RTE_ETH_REPRESENTOR_PF;
2225                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2226                          "c%dpf%d", range->controller, range->pf);
2227         }
2228         if (ret >= RTE_DEV_NAME_MAX_LEN) {
2229                 sfc_err(ctx->sa, "representor name has been truncated: %s",
2230                         range->name);
2231         }
2232
2233         ctx->info->nb_ranges++;
2234 }
2235
2236 static int
2237 sfc_representor_info_get(struct rte_eth_dev *dev,
2238                          struct rte_eth_representor_info *info)
2239 {
2240         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2241         struct sfc_get_representors_ctx get_repr_ctx;
2242         const efx_nic_cfg_t *nic_cfg;
2243         uint16_t switch_domain_id;
2244         uint32_t nb_repr;
2245         int controller;
2246         int rc;
2247
2248         sfc_adapter_lock(sa);
2249
2250         if (sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2251                 sfc_adapter_unlock(sa);
2252                 return -ENOTSUP;
2253         }
2254
2255         rc = sfc_process_mport_journal(sa);
2256         if (rc != 0) {
2257                 sfc_adapter_unlock(sa);
2258                 SFC_ASSERT(rc > 0);
2259                 return -rc;
2260         }
2261
2262         switch_domain_id = sa->mae.switch_domain_id;
2263
2264         nb_repr = 0;
2265         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2266                                           sfc_count_representors_cb,
2267                                           &nb_repr);
2268         if (rc != 0) {
2269                 sfc_adapter_unlock(sa);
2270                 SFC_ASSERT(rc > 0);
2271                 return -rc;
2272         }
2273
2274         if (info == NULL) {
2275                 sfc_adapter_unlock(sa);
2276                 return nb_repr;
2277         }
2278
2279         rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2280                                                &get_repr_ctx.controllers,
2281                                                &get_repr_ctx.nb_controllers);
2282         if (rc != 0) {
2283                 sfc_adapter_unlock(sa);
2284                 SFC_ASSERT(rc > 0);
2285                 return -rc;
2286         }
2287
2288         nic_cfg = efx_nic_cfg_get(sa->nic);
2289
2290         rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2291                                                   nic_cfg->enc_intf,
2292                                                   &controller);
2293         if (rc != 0) {
2294                 sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2295                 controller = -1;
2296         }
2297
2298         info->controller = controller;
2299         info->pf = nic_cfg->enc_pf;
2300
2301         get_repr_ctx.info = info;
2302         get_repr_ctx.sa = sa;
2303         get_repr_ctx.switch_domain_id = switch_domain_id;
2304         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2305                                           sfc_get_representors_cb,
2306                                           &get_repr_ctx);
2307         if (rc != 0) {
2308                 sfc_adapter_unlock(sa);
2309                 SFC_ASSERT(rc > 0);
2310                 return -rc;
2311         }
2312
2313         sfc_adapter_unlock(sa);
2314         return nb_repr;
2315 }
2316
2317 static int
2318 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2319 {
2320         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2321         uint64_t supported = 0;
2322
2323         sfc_adapter_lock(sa);
2324
2325         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2326                 supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2327
2328         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2329                 supported |= RTE_ETH_RX_METADATA_USER_MARK;
2330
2331         if (sfc_flow_tunnel_is_supported(sa))
2332                 supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2333
2334         sa->negotiated_rx_metadata = supported & *features;
2335         *features = sa->negotiated_rx_metadata;
2336
2337         sfc_adapter_unlock(sa);
2338
2339         return 0;
2340 }
2341
2342 static const struct eth_dev_ops sfc_eth_dev_ops = {
2343         .dev_configure                  = sfc_dev_configure,
2344         .dev_start                      = sfc_dev_start,
2345         .dev_stop                       = sfc_dev_stop,
2346         .dev_set_link_up                = sfc_dev_set_link_up,
2347         .dev_set_link_down              = sfc_dev_set_link_down,
2348         .dev_close                      = sfc_dev_close,
2349         .promiscuous_enable             = sfc_dev_promisc_enable,
2350         .promiscuous_disable            = sfc_dev_promisc_disable,
2351         .allmulticast_enable            = sfc_dev_allmulti_enable,
2352         .allmulticast_disable           = sfc_dev_allmulti_disable,
2353         .link_update                    = sfc_dev_link_update,
2354         .stats_get                      = sfc_stats_get,
2355         .stats_reset                    = sfc_stats_reset,
2356         .xstats_get                     = sfc_xstats_get,
2357         .xstats_reset                   = sfc_stats_reset,
2358         .xstats_get_names               = sfc_xstats_get_names,
2359         .dev_infos_get                  = sfc_dev_infos_get,
2360         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2361         .mtu_set                        = sfc_dev_set_mtu,
2362         .rx_queue_start                 = sfc_rx_queue_start,
2363         .rx_queue_stop                  = sfc_rx_queue_stop,
2364         .tx_queue_start                 = sfc_tx_queue_start,
2365         .tx_queue_stop                  = sfc_tx_queue_stop,
2366         .rx_queue_setup                 = sfc_rx_queue_setup,
2367         .rx_queue_release               = sfc_rx_queue_release,
2368         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
2369         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
2370         .tx_queue_setup                 = sfc_tx_queue_setup,
2371         .tx_queue_release               = sfc_tx_queue_release,
2372         .flow_ctrl_get                  = sfc_flow_ctrl_get,
2373         .flow_ctrl_set                  = sfc_flow_ctrl_set,
2374         .mac_addr_set                   = sfc_mac_addr_set,
2375         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
2376         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
2377         .reta_update                    = sfc_dev_rss_reta_update,
2378         .reta_query                     = sfc_dev_rss_reta_query,
2379         .rss_hash_update                = sfc_dev_rss_hash_update,
2380         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2381         .flow_ops_get                   = sfc_dev_flow_ops_get,
2382         .set_mc_addr_list               = sfc_set_mc_addr_list,
2383         .rxq_info_get                   = sfc_rx_queue_info_get,
2384         .txq_info_get                   = sfc_tx_queue_info_get,
2385         .fw_version_get                 = sfc_fw_version_get,
2386         .xstats_get_by_id               = sfc_xstats_get_by_id,
2387         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
2388         .pool_ops_supported             = sfc_pool_ops_supported,
2389         .representor_info_get           = sfc_representor_info_get,
2390         .rx_metadata_negotiate          = sfc_rx_metadata_negotiate,
2391 };
2392
2393 struct sfc_ethdev_init_data {
2394         uint16_t                nb_representors;
2395 };
2396
2397 /**
2398  * Duplicate a string in potentially shared memory required for
2399  * multi-process support.
2400  *
2401  * strdup() allocates from process-local heap/memory.
2402  */
2403 static char *
2404 sfc_strdup(const char *str)
2405 {
2406         size_t size;
2407         char *copy;
2408
2409         if (str == NULL)
2410                 return NULL;
2411
2412         size = strlen(str) + 1;
2413         copy = rte_malloc(__func__, size, 0);
2414         if (copy != NULL)
2415                 rte_memcpy(copy, str, size);
2416
2417         return copy;
2418 }
2419
2420 static int
2421 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2422 {
2423         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2424         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2425         const struct sfc_dp_rx *dp_rx;
2426         const struct sfc_dp_tx *dp_tx;
2427         const efx_nic_cfg_t *encp;
2428         unsigned int avail_caps = 0;
2429         const char *rx_name = NULL;
2430         const char *tx_name = NULL;
2431         int rc;
2432
2433         switch (sa->family) {
2434         case EFX_FAMILY_HUNTINGTON:
2435         case EFX_FAMILY_MEDFORD:
2436         case EFX_FAMILY_MEDFORD2:
2437                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2438                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2439                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2440                 break;
2441         case EFX_FAMILY_RIVERHEAD:
2442                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2443                 break;
2444         default:
2445                 break;
2446         }
2447
2448         encp = efx_nic_cfg_get(sa->nic);
2449         if (encp->enc_rx_es_super_buffer_supported)
2450                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2451
2452         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2453                                 sfc_kvarg_string_handler, &rx_name);
2454         if (rc != 0)
2455                 goto fail_kvarg_rx_datapath;
2456
2457         if (rx_name != NULL) {
2458                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2459                 if (dp_rx == NULL) {
2460                         sfc_err(sa, "Rx datapath %s not found", rx_name);
2461                         rc = ENOENT;
2462                         goto fail_dp_rx;
2463                 }
2464                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2465                         sfc_err(sa,
2466                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2467                                 rx_name);
2468                         rc = EINVAL;
2469                         goto fail_dp_rx_caps;
2470                 }
2471         } else {
2472                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2473                 if (dp_rx == NULL) {
2474                         sfc_err(sa, "Rx datapath by caps %#x not found",
2475                                 avail_caps);
2476                         rc = ENOENT;
2477                         goto fail_dp_rx;
2478                 }
2479         }
2480
2481         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2482         if (sas->dp_rx_name == NULL) {
2483                 rc = ENOMEM;
2484                 goto fail_dp_rx_name;
2485         }
2486
2487         if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2488                 /* FLAG and MARK are always available from Rx prefix. */
2489                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2490                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2491         }
2492
2493         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2494
2495         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2496                                 sfc_kvarg_string_handler, &tx_name);
2497         if (rc != 0)
2498                 goto fail_kvarg_tx_datapath;
2499
2500         if (tx_name != NULL) {
2501                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2502                 if (dp_tx == NULL) {
2503                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2504                         rc = ENOENT;
2505                         goto fail_dp_tx;
2506                 }
2507                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2508                         sfc_err(sa,
2509                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2510                                 tx_name);
2511                         rc = EINVAL;
2512                         goto fail_dp_tx_caps;
2513                 }
2514         } else {
2515                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2516                 if (dp_tx == NULL) {
2517                         sfc_err(sa, "Tx datapath by caps %#x not found",
2518                                 avail_caps);
2519                         rc = ENOENT;
2520                         goto fail_dp_tx;
2521                 }
2522         }
2523
2524         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2525         if (sas->dp_tx_name == NULL) {
2526                 rc = ENOMEM;
2527                 goto fail_dp_tx_name;
2528         }
2529
2530         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2531
2532         sa->priv.dp_rx = dp_rx;
2533         sa->priv.dp_tx = dp_tx;
2534
2535         dev->rx_pkt_burst = dp_rx->pkt_burst;
2536         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2537         dev->tx_pkt_burst = dp_tx->pkt_burst;
2538
2539         dev->rx_queue_count = sfc_rx_queue_count;
2540         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2541         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2542         dev->dev_ops = &sfc_eth_dev_ops;
2543
2544         return 0;
2545
2546 fail_dp_tx_name:
2547 fail_dp_tx_caps:
2548 fail_dp_tx:
2549 fail_kvarg_tx_datapath:
2550         rte_free(sas->dp_rx_name);
2551         sas->dp_rx_name = NULL;
2552
2553 fail_dp_rx_name:
2554 fail_dp_rx_caps:
2555 fail_dp_rx:
2556 fail_kvarg_rx_datapath:
2557         return rc;
2558 }
2559
2560 static void
2561 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2562 {
2563         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2564         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2565
2566         dev->dev_ops = NULL;
2567         dev->tx_pkt_prepare = NULL;
2568         dev->rx_pkt_burst = NULL;
2569         dev->tx_pkt_burst = NULL;
2570
2571         rte_free(sas->dp_tx_name);
2572         sas->dp_tx_name = NULL;
2573         sa->priv.dp_tx = NULL;
2574
2575         rte_free(sas->dp_rx_name);
2576         sas->dp_rx_name = NULL;
2577         sa->priv.dp_rx = NULL;
2578 }
2579
2580 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2581         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2582         .reta_query                     = sfc_dev_rss_reta_query,
2583         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2584         .rxq_info_get                   = sfc_rx_queue_info_get,
2585         .txq_info_get                   = sfc_tx_queue_info_get,
2586 };
2587
2588 static int
2589 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2590 {
2591         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2592         struct sfc_adapter_priv *sap;
2593         const struct sfc_dp_rx *dp_rx;
2594         const struct sfc_dp_tx *dp_tx;
2595         int rc;
2596
2597         /*
2598          * Allocate process private data from heap, since it should not
2599          * be located in shared memory allocated using rte_malloc() API.
2600          */
2601         sap = calloc(1, sizeof(*sap));
2602         if (sap == NULL) {
2603                 rc = ENOMEM;
2604                 goto fail_alloc_priv;
2605         }
2606
2607         sap->logtype_main = logtype_main;
2608
2609         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2610         if (dp_rx == NULL) {
2611                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2612                         "cannot find %s Rx datapath", sas->dp_rx_name);
2613                 rc = ENOENT;
2614                 goto fail_dp_rx;
2615         }
2616         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2617                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2618                         "%s Rx datapath does not support multi-process",
2619                         sas->dp_rx_name);
2620                 rc = EINVAL;
2621                 goto fail_dp_rx_multi_process;
2622         }
2623
2624         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2625         if (dp_tx == NULL) {
2626                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2627                         "cannot find %s Tx datapath", sas->dp_tx_name);
2628                 rc = ENOENT;
2629                 goto fail_dp_tx;
2630         }
2631         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2632                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2633                         "%s Tx datapath does not support multi-process",
2634                         sas->dp_tx_name);
2635                 rc = EINVAL;
2636                 goto fail_dp_tx_multi_process;
2637         }
2638
2639         sap->dp_rx = dp_rx;
2640         sap->dp_tx = dp_tx;
2641
2642         dev->process_private = sap;
2643         dev->rx_pkt_burst = dp_rx->pkt_burst;
2644         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2645         dev->tx_pkt_burst = dp_tx->pkt_burst;
2646         dev->rx_queue_count = sfc_rx_queue_count;
2647         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2648         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2649         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2650
2651         return 0;
2652
2653 fail_dp_tx_multi_process:
2654 fail_dp_tx:
2655 fail_dp_rx_multi_process:
2656 fail_dp_rx:
2657         free(sap);
2658
2659 fail_alloc_priv:
2660         return rc;
2661 }
2662
2663 static void
2664 sfc_register_dp(void)
2665 {
2666         /* Register once */
2667         if (TAILQ_EMPTY(&sfc_dp_head)) {
2668                 /* Prefer EF10 datapath */
2669                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2670                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2671                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2672                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2673
2674                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2675                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2676                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2677                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2678         }
2679 }
2680
2681 static int
2682 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
2683 {
2684         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2685         const char *switch_mode = NULL;
2686         int rc;
2687
2688         sfc_log_init(sa, "entry");
2689
2690         rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2691                                 sfc_kvarg_string_handler, &switch_mode);
2692         if (rc != 0)
2693                 goto fail_kvargs;
2694
2695         if (switch_mode == NULL) {
2696                 sa->switchdev = encp->enc_mae_admin &&
2697                                 (!encp->enc_datapath_cap_evb ||
2698                                  has_representors);
2699         } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2700                 sa->switchdev = false;
2701         } else if (strcasecmp(switch_mode,
2702                               SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2703                 sa->switchdev = true;
2704         } else {
2705                 sfc_err(sa, "invalid switch mode device argument '%s'",
2706                         switch_mode);
2707                 rc = EINVAL;
2708                 goto fail_mode;
2709         }
2710
2711         sfc_log_init(sa, "done");
2712
2713         return 0;
2714
2715 fail_mode:
2716 fail_kvargs:
2717         sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2718
2719         return rc;
2720 }
2721
2722 static int
2723 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
2724 {
2725         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2726         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2727         struct sfc_ethdev_init_data *init_data = init_params;
2728         uint32_t logtype_main;
2729         struct sfc_adapter *sa;
2730         int rc;
2731         const efx_nic_cfg_t *encp;
2732         const struct rte_ether_addr *from;
2733         int ret;
2734
2735         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2736                         SFC_EFX_DEV_CLASS_NET) {
2737                 SFC_GENERIC_LOG(DEBUG,
2738                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2739                 return 1;
2740         }
2741
2742         rc = sfc_dp_mport_register();
2743         if (rc != 0)
2744                 return rc;
2745
2746         sfc_register_dp();
2747
2748         logtype_main = sfc_register_logtype(&pci_dev->addr,
2749                                             SFC_LOGTYPE_MAIN_STR,
2750                                             RTE_LOG_NOTICE);
2751
2752         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2753                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2754
2755         /* Required for logging */
2756         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2757                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2758                         pci_dev->addr.domain, pci_dev->addr.bus,
2759                         pci_dev->addr.devid, pci_dev->addr.function,
2760                         dev->data->port_id);
2761         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2762                 SFC_GENERIC_LOG(ERR,
2763                         "reserved log prefix is too short for " PCI_PRI_FMT,
2764                         pci_dev->addr.domain, pci_dev->addr.bus,
2765                         pci_dev->addr.devid, pci_dev->addr.function);
2766                 return -EINVAL;
2767         }
2768         sas->pci_addr = pci_dev->addr;
2769         sas->port_id = dev->data->port_id;
2770
2771         /*
2772          * Allocate process private data from heap, since it should not
2773          * be located in shared memory allocated using rte_malloc() API.
2774          */
2775         sa = calloc(1, sizeof(*sa));
2776         if (sa == NULL) {
2777                 rc = ENOMEM;
2778                 goto fail_alloc_sa;
2779         }
2780
2781         dev->process_private = sa;
2782
2783         /* Required for logging */
2784         sa->priv.shared = sas;
2785         sa->priv.logtype_main = logtype_main;
2786
2787         sa->eth_dev = dev;
2788
2789         /* Copy PCI device info to the dev->data */
2790         rte_eth_copy_pci_info(dev, pci_dev);
2791         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2792
2793         rc = sfc_kvargs_parse(sa);
2794         if (rc != 0)
2795                 goto fail_kvargs_parse;
2796
2797         sfc_log_init(sa, "entry");
2798
2799         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2800         if (dev->data->mac_addrs == NULL) {
2801                 rc = ENOMEM;
2802                 goto fail_mac_addrs;
2803         }
2804
2805         sfc_adapter_lock_init(sa);
2806         sfc_adapter_lock(sa);
2807
2808         sfc_log_init(sa, "probing");
2809         rc = sfc_probe(sa);
2810         if (rc != 0)
2811                 goto fail_probe;
2812
2813         /*
2814          * Selecting a default switch mode requires the NIC to be probed and
2815          * to have its capabilities filled in.
2816          */
2817         rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
2818         if (rc != 0)
2819                 goto fail_switch_mode;
2820
2821         sfc_log_init(sa, "set device ops");
2822         rc = sfc_eth_dev_set_ops(dev);
2823         if (rc != 0)
2824                 goto fail_set_ops;
2825
2826         sfc_log_init(sa, "attaching");
2827         rc = sfc_attach(sa);
2828         if (rc != 0)
2829                 goto fail_attach;
2830
2831         if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_ADMIN) {
2832                 sfc_err(sa,
2833                         "failed to enable switchdev mode without admin MAE privilege");
2834                 rc = ENOTSUP;
2835                 goto fail_switchdev_no_mae;
2836         }
2837
2838         encp = efx_nic_cfg_get(sa->nic);
2839
2840         /*
2841          * The arguments are really reverse order in comparison to
2842          * Linux kernel. Copy from NIC config to Ethernet device data.
2843          */
2844         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2845         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2846
2847         /*
2848          * Setup the NIC DMA mapping handler. All internal mempools
2849          * MUST be created on attach before this point, and the
2850          * adapter MUST NOT create mempools with the adapter lock
2851          * held after this point.
2852          */
2853         rc = sfc_nic_dma_attach(sa);
2854         if (rc != 0)
2855                 goto fail_nic_dma_attach;
2856
2857         sfc_adapter_unlock(sa);
2858
2859         sfc_log_init(sa, "done");
2860         return 0;
2861
2862 fail_nic_dma_attach:
2863 fail_switchdev_no_mae:
2864         sfc_detach(sa);
2865
2866 fail_attach:
2867         sfc_eth_dev_clear_ops(dev);
2868
2869 fail_set_ops:
2870 fail_switch_mode:
2871         sfc_unprobe(sa);
2872
2873 fail_probe:
2874         sfc_adapter_unlock(sa);
2875         sfc_adapter_lock_fini(sa);
2876         rte_free(dev->data->mac_addrs);
2877         dev->data->mac_addrs = NULL;
2878
2879 fail_mac_addrs:
2880         sfc_kvargs_cleanup(sa);
2881
2882 fail_kvargs_parse:
2883         sfc_log_init(sa, "failed %d", rc);
2884         dev->process_private = NULL;
2885         free(sa);
2886
2887 fail_alloc_sa:
2888         SFC_ASSERT(rc > 0);
2889         return -rc;
2890 }
2891
2892 static int
2893 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2894 {
2895         sfc_dev_close(dev);
2896
2897         return 0;
2898 }
2899
2900 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2901         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2902         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2903         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2904         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2905         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2906         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2907         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2908         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2909         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2910         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD_VF) },
2911         { .vendor_id = 0 /* sentinel */ }
2912 };
2913
2914 static int
2915 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
2916 {
2917         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
2918         int rc;
2919
2920         if (args != NULL) {
2921                 rc = rte_eth_devargs_parse(args, &eth_da);
2922                 if (rc != 0) {
2923                         SFC_GENERIC_LOG(ERR,
2924                                         "Failed to parse generic devargs '%s'",
2925                                         args);
2926                         return rc;
2927                 }
2928         }
2929
2930         *devargs = eth_da;
2931
2932         return 0;
2933 }
2934
2935 static int
2936 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
2937                            struct sfc_ethdev_init_data *init_data,
2938                            struct rte_eth_dev **devp,
2939                            bool *dev_created)
2940 {
2941         struct rte_eth_dev *dev;
2942         bool created = false;
2943         int rc;
2944
2945         dev = rte_eth_dev_allocated(pci_dev->device.name);
2946         if (dev == NULL) {
2947                 rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
2948                                         sizeof(struct sfc_adapter_shared),
2949                                         eth_dev_pci_specific_init, pci_dev,
2950                                         sfc_eth_dev_init, init_data);
2951                 if (rc != 0) {
2952                         SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
2953                                         pci_dev->device.name);
2954                         return rc;
2955                 }
2956
2957                 created = true;
2958
2959                 dev = rte_eth_dev_allocated(pci_dev->device.name);
2960                 if (dev == NULL) {
2961                         SFC_GENERIC_LOG(ERR,
2962                                 "Failed to find allocated sfc ethdev '%s'",
2963                                 pci_dev->device.name);
2964                         return -ENODEV;
2965                 }
2966         }
2967
2968         *devp = dev;
2969         *dev_created = created;
2970
2971         return 0;
2972 }
2973
2974 static int
2975 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
2976                         efx_pcie_interface_t controller,
2977                         uint16_t port,
2978                         uint16_t repr_port,
2979                         enum rte_eth_representor_type type)
2980 {
2981         struct sfc_repr_entity_info entity;
2982         efx_mport_sel_t mport_sel;
2983         int rc;
2984
2985         switch (type) {
2986         case RTE_ETH_REPRESENTOR_NONE:
2987                 return 0;
2988         case RTE_ETH_REPRESENTOR_VF:
2989         case RTE_ETH_REPRESENTOR_PF:
2990                 break;
2991         case RTE_ETH_REPRESENTOR_SF:
2992                 sfc_err(sa, "SF representors are not supported");
2993                 return ENOTSUP;
2994         default:
2995                 sfc_err(sa, "unknown representor type: %d", type);
2996                 return ENOTSUP;
2997         }
2998
2999         rc = efx_mae_mport_by_pcie_mh_function(controller,
3000                                                port,
3001                                                repr_port,
3002                                                &mport_sel);
3003         if (rc != 0) {
3004                 sfc_err(sa,
3005                         "failed to get m-port selector for controller %u port %u repr_port %u: %s",
3006                         controller, port, repr_port, rte_strerror(-rc));
3007                 return rc;
3008         }
3009
3010         memset(&entity, 0, sizeof(entity));
3011         entity.type = type;
3012         entity.intf = controller;
3013         entity.pf = port;
3014         entity.vf = repr_port;
3015
3016         rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3017                              &mport_sel);
3018         if (rc != 0) {
3019                 sfc_err(sa,
3020                         "failed to create representor for controller %u port %u repr_port %u: %s",
3021                         controller, port, repr_port, rte_strerror(-rc));
3022                 return rc;
3023         }
3024
3025         return 0;
3026 }
3027
3028 static int
3029 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3030                              const struct rte_eth_devargs *eth_da,
3031                              efx_pcie_interface_t controller,
3032                              uint16_t port)
3033 {
3034         int first_error = 0;
3035         uint16_t i;
3036         int rc;
3037
3038         if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3039                 return sfc_eth_dev_create_repr(sa, controller, port,
3040                                                EFX_PCI_VF_INVALID,
3041                                                eth_da->type);
3042         }
3043
3044         for (i = 0; i < eth_da->nb_representor_ports; i++) {
3045                 rc = sfc_eth_dev_create_repr(sa, controller, port,
3046                                              eth_da->representor_ports[i],
3047                                              eth_da->type);
3048                 if (rc != 0 && first_error == 0)
3049                         first_error = rc;
3050         }
3051
3052         return first_error;
3053 }
3054
3055 static int
3056 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3057                                    const struct rte_eth_devargs *eth_da,
3058                                    efx_pcie_interface_t controller)
3059 {
3060         const efx_nic_cfg_t *encp;
3061         int first_error = 0;
3062         uint16_t default_port;
3063         uint16_t i;
3064         int rc;
3065
3066         if (eth_da->nb_ports == 0) {
3067                 encp = efx_nic_cfg_get(sa->nic);
3068                 default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3069                 return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3070                                                     default_port);
3071         }
3072
3073         for (i = 0; i < eth_da->nb_ports; i++) {
3074                 rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3075                                                   eth_da->ports[i]);
3076                 if (rc != 0 && first_error == 0)
3077                         first_error = rc;
3078         }
3079
3080         return first_error;
3081 }
3082
3083 static int
3084 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3085                                 const struct rte_eth_devargs *eth_da)
3086 {
3087         efx_pcie_interface_t intf;
3088         const efx_nic_cfg_t *encp;
3089         struct sfc_adapter *sa;
3090         uint16_t switch_domain_id;
3091         uint16_t i;
3092         int rc;
3093
3094         sa = sfc_adapter_by_eth_dev(dev);
3095         switch_domain_id = sa->mae.switch_domain_id;
3096
3097         switch (eth_da->type) {
3098         case RTE_ETH_REPRESENTOR_NONE:
3099                 return 0;
3100         case RTE_ETH_REPRESENTOR_PF:
3101         case RTE_ETH_REPRESENTOR_VF:
3102                 break;
3103         case RTE_ETH_REPRESENTOR_SF:
3104                 sfc_err(sa, "SF representors are not supported");
3105                 return -ENOTSUP;
3106         default:
3107                 sfc_err(sa, "unknown representor type: %d",
3108                         eth_da->type);
3109                 return -ENOTSUP;
3110         }
3111
3112         if (!sa->switchdev) {
3113                 sfc_err(sa, "cannot create representors in non-switchdev mode");
3114                 return -EINVAL;
3115         }
3116
3117         if (!sfc_repr_available(sfc_sa2shared(sa))) {
3118                 sfc_err(sa, "cannot create representors: unsupported");
3119
3120                 return -ENOTSUP;
3121         }
3122
3123         /*
3124          * This is needed to construct the DPDK controller -> EFX interface
3125          * mapping.
3126          */
3127         sfc_adapter_lock(sa);
3128         rc = sfc_process_mport_journal(sa);
3129         sfc_adapter_unlock(sa);
3130         if (rc != 0) {
3131                 SFC_ASSERT(rc > 0);
3132                 return -rc;
3133         }
3134
3135         if (eth_da->nb_mh_controllers > 0) {
3136                 for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3137                         rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3138                                                 eth_da->mh_controllers[i],
3139                                                 &intf);
3140                         if (rc != 0) {
3141                                 sfc_err(sa, "failed to get representor");
3142                                 continue;
3143                         }
3144                         sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3145                 }
3146         } else {
3147                 encp = efx_nic_cfg_get(sa->nic);
3148                 sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3149         }
3150
3151         return 0;
3152 }
3153
3154 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3155         struct rte_pci_device *pci_dev)
3156 {
3157         struct sfc_ethdev_init_data init_data;
3158         struct rte_eth_devargs eth_da;
3159         struct rte_eth_dev *dev;
3160         bool dev_created;
3161         int rc;
3162
3163         if (pci_dev->device.devargs != NULL) {
3164                 rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3165                                            &eth_da);
3166                 if (rc != 0)
3167                         return rc;
3168         } else {
3169                 memset(&eth_da, 0, sizeof(eth_da));
3170         }
3171
3172         /* If no VF representors specified, check for PF ones */
3173         if (eth_da.nb_representor_ports > 0)
3174                 init_data.nb_representors = eth_da.nb_representor_ports;
3175         else
3176                 init_data.nb_representors = eth_da.nb_ports;
3177
3178         if (init_data.nb_representors > 0 &&
3179             rte_eal_process_type() != RTE_PROC_PRIMARY) {
3180                 SFC_GENERIC_LOG(ERR,
3181                         "Create representors from secondary process not supported, dev '%s'",
3182                         pci_dev->device.name);
3183                 return -ENOTSUP;
3184         }
3185
3186         /*
3187          * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3188          * if it does not already exist. Re-probing an existing device is
3189          * expected to allow additional representors to be configured.
3190          */
3191         rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3192                                         &dev_created);
3193         if (rc != 0)
3194                 return rc;
3195
3196         rc = sfc_eth_dev_create_representors(dev, &eth_da);
3197         if (rc != 0) {
3198                 if (dev_created)
3199                         (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3200
3201                 return rc;
3202         }
3203
3204         return 0;
3205 }
3206
3207 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3208 {
3209         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3210 }
3211
3212 static struct rte_pci_driver sfc_efx_pmd = {
3213         .id_table = pci_id_sfc_efx_map,
3214         .drv_flags =
3215                 RTE_PCI_DRV_INTR_LSC |
3216                 RTE_PCI_DRV_NEED_MAPPING |
3217                 RTE_PCI_DRV_PROBE_AGAIN,
3218         .probe = sfc_eth_dev_pci_probe,
3219         .remove = sfc_eth_dev_pci_remove,
3220 };
3221
3222 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3223 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3224 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3225 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3226         SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3227         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3228         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3229         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3230         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3231         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3232         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3233
3234 RTE_INIT(sfc_driver_register_logtype)
3235 {
3236         int ret;
3237
3238         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
3239                                                    RTE_LOG_NOTICE);
3240         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
3241 }