drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2021 Xilinx, Inc.
   4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17 #include <rte_ether.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_debug.h"
  23 #include "sfc_log.h"
  24 #include "sfc_kvargs.h"
  25 #include "sfc_ev.h"
  26 #include "sfc_rx.h"
  27 #include "sfc_tx.h"
  28 #include "sfc_flow.h"
  29 #include "sfc_flow_tunnel.h"
  30 #include "sfc_dp.h"
  31 #include "sfc_dp_rx.h"
  32 #include "sfc_repr.h"
  33 #include "sfc_sw_stats.h"
  34 #include "sfc_switch.h"
  35
  36 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
  37 #define SFC_XSTAT_ID_INVALID_NAME '\0'
  38
  39 uint32_t sfc_logtype_driver;
  40
  41 static struct sfc_dp_list sfc_dp_head =
  42         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  43
  44
  45 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  46
  47
  48 static int
  49 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  50 {
  51         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  52         efx_nic_fw_info_t enfi;
  53         int ret;
  54         int rc;
  55
  56         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  57         if (rc != 0)
  58                 return -rc;
  59
  60         ret = snprintf(fw_version, fw_size,
  61                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  62                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  63                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  64         if (ret < 0)
  65                 return ret;
  66
  67         if (enfi.enfi_dpcpu_fw_ids_valid) {
  68                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  69                 int ret_extra;
  70
  71                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  72                                      fw_size - dpcpu_fw_ids_offset,
  73                                      " rx%" PRIx16 " tx%" PRIx16,
  74                                      enfi.enfi_rx_dpcpu_fw_id,
  75                                      enfi.enfi_tx_dpcpu_fw_id);
  76                 if (ret_extra < 0)
  77                         return ret_extra;
  78
  79                 ret += ret_extra;
  80         }
  81
  82         if (fw_size < (size_t)(++ret))
  83                 return ret;
  84         else
  85                 return 0;
  86 }
  87
  88 static int
  89 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  90 {
  91         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  92         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  93         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  94         struct sfc_rss *rss = &sas->rss;
  95         struct sfc_mae *mae = &sa->mae;
  96         uint64_t txq_offloads_def = 0;
  97
  98         sfc_log_init(sa, "entry");
  99
 100         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 101         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 102
 103         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 104
 105         dev_info->max_vfs = sa->sriov.num_vfs;
 106
 107         /* Autonegotiation may be disabled */
 108         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 110                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 112                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 114                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 115         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 116                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 117         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 118                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 119         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 120                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 121
 122         dev_info->max_rx_queues = sa->rxq_max;
 123         dev_info->max_tx_queues = sa->txq_max;
 124
 125         /* By default packets are dropped if no descriptors are available */
 126         dev_info->default_rxconf.rx_drop_en = 1;
 127
 128         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 129
 130         /*
 131          * rx_offload_capa includes both device and queue offloads since
 132          * the latter may be requested on a per device basis which makes
 133          * sense when some offloads are needed to be set on all queues.
 134          */
 135         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 136                                     dev_info->rx_queue_offload_capa;
 137
 138         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 139
 140         /*
 141          * tx_offload_capa includes both device and queue offloads since
 142          * the latter may be requested on a per device basis which makes
 143          * sense when some offloads are needed to be set on all queues.
 144          */
 145         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 146                                     dev_info->tx_queue_offload_capa;
 147
 148         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 149                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 150
 151         dev_info->default_txconf.offloads |= txq_offloads_def;
 152
 153         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 154                 uint64_t rte_hf = 0;
 155                 unsigned int i;
 156
 157                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 158                         rte_hf |= rss->hf_map[i].rte;
 159
 160                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 161                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 162                 dev_info->flow_type_rss_offloads = rte_hf;
 163         }
 164
 165         /* Initialize to hardware limits */
 166         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 167         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 168         /* The RXQ hardware requires that the descriptor count is a power
 169          * of 2, but rx_desc_lim cannot properly describe that constraint.
 170          */
 171         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 172
 173         /* Initialize to hardware limits */
 174         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 175         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 176         /*
 177          * The TXQ hardware requires that the descriptor count is a power
 178          * of 2, but tx_desc_lim cannot properly describe that constraint
 179          */
 180         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 181
 182         if (sap->dp_rx->get_dev_info != NULL)
 183                 sap->dp_rx->get_dev_info(dev_info);
 184         if (sap->dp_tx->get_dev_info != NULL)
 185                 sap->dp_tx->get_dev_info(dev_info);
 186
 187         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 188                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 189
 190         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
 191                 dev_info->switch_info.name = dev->device->driver->name;
 192                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 193                 dev_info->switch_info.port_id = mae->switch_port_id;
 194         }
 195
 196         return 0;
 197 }
 198
 199 static const uint32_t *
 200 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 201 {
 202         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 203
 204         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 205 }
 206
 207 static int
 208 sfc_dev_configure(struct rte_eth_dev *dev)
 209 {
 210         struct rte_eth_dev_data *dev_data = dev->data;
 211         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 212         int rc;
 213
 214         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 215                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 216
 217         sfc_adapter_lock(sa);
 218         switch (sa->state) {
 219         case SFC_ETHDEV_CONFIGURED:
 220                 /* FALLTHROUGH */
 221         case SFC_ETHDEV_INITIALIZED:
 222                 rc = sfc_configure(sa);
 223                 break;
 224         default:
 225                 sfc_err(sa, "unexpected adapter state %u to configure",
 226                         sa->state);
 227                 rc = EINVAL;
 228                 break;
 229         }
 230         sfc_adapter_unlock(sa);
 231
 232         sfc_log_init(sa, "done %d", rc);
 233         SFC_ASSERT(rc >= 0);
 234         return -rc;
 235 }
 236
 237 static int
 238 sfc_dev_start(struct rte_eth_dev *dev)
 239 {
 240         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 241         int rc;
 242
 243         sfc_log_init(sa, "entry");
 244
 245         sfc_adapter_lock(sa);
 246         rc = sfc_start(sa);
 247         sfc_adapter_unlock(sa);
 248
 249         sfc_log_init(sa, "done %d", rc);
 250         SFC_ASSERT(rc >= 0);
 251         return -rc;
 252 }
 253
 254 static int
 255 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 256 {
 257         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 258         struct rte_eth_link current_link;
 259         int ret;
 260
 261         sfc_log_init(sa, "entry");
 262
 263         if (sa->state != SFC_ETHDEV_STARTED) {
 264                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 265         } else if (wait_to_complete) {
 266                 efx_link_mode_t link_mode;
 267
 268                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 269                         link_mode = EFX_LINK_UNKNOWN;
 270                 sfc_port_link_mode_to_info(link_mode, &current_link);
 271
 272         } else {
 273                 sfc_ev_mgmt_qpoll(sa);
 274                 rte_eth_linkstatus_get(dev, &current_link);
 275         }
 276
 277         ret = rte_eth_linkstatus_set(dev, &current_link);
 278         if (ret == 0)
 279                 sfc_notice(sa, "Link status is %s",
 280                            current_link.link_status ? "UP" : "DOWN");
 281
 282         return ret;
 283 }
 284
 285 static int
 286 sfc_dev_stop(struct rte_eth_dev *dev)
 287 {
 288         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 289
 290         sfc_log_init(sa, "entry");
 291
 292         sfc_adapter_lock(sa);
 293         sfc_stop(sa);
 294         sfc_adapter_unlock(sa);
 295
 296         sfc_log_init(sa, "done");
 297
 298         return 0;
 299 }
 300
 301 static int
 302 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 303 {
 304         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 305         int rc;
 306
 307         sfc_log_init(sa, "entry");
 308
 309         sfc_adapter_lock(sa);
 310         rc = sfc_start(sa);
 311         sfc_adapter_unlock(sa);
 312
 313         SFC_ASSERT(rc >= 0);
 314         return -rc;
 315 }
 316
 317 static int
 318 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 319 {
 320         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 321
 322         sfc_log_init(sa, "entry");
 323
 324         sfc_adapter_lock(sa);
 325         sfc_stop(sa);
 326         sfc_adapter_unlock(sa);
 327
 328         return 0;
 329 }
 330
 331 static void
 332 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 333 {
 334         free(dev->process_private);
 335         rte_eth_dev_release_port(dev);
 336 }
 337
 338 static int
 339 sfc_dev_close(struct rte_eth_dev *dev)
 340 {
 341         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 342
 343         sfc_log_init(sa, "entry");
 344
 345         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 346                 sfc_eth_dev_secondary_clear_ops(dev);
 347                 return 0;
 348         }
 349
 350         sfc_pre_detach(sa);
 351
 352         sfc_adapter_lock(sa);
 353         switch (sa->state) {
 354         case SFC_ETHDEV_STARTED:
 355                 sfc_stop(sa);
 356                 SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
 357                 /* FALLTHROUGH */
 358         case SFC_ETHDEV_CONFIGURED:
 359                 sfc_close(sa);
 360                 SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
 361                 /* FALLTHROUGH */
 362         case SFC_ETHDEV_INITIALIZED:
 363                 break;
 364         default:
 365                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 366                 break;
 367         }
 368
 369         /*
 370          * Cleanup all resources.
 371          * Rollback primary process sfc_eth_dev_init() below.
 372          */
 373
 374         sfc_eth_dev_clear_ops(dev);
 375
 376         sfc_detach(sa);
 377         sfc_unprobe(sa);
 378
 379         sfc_kvargs_cleanup(sa);
 380
 381         sfc_adapter_unlock(sa);
 382         sfc_adapter_lock_fini(sa);
 383
 384         sfc_log_init(sa, "done");
 385
 386         /* Required for logging, so cleanup last */
 387         sa->eth_dev = NULL;
 388
 389         free(sa);
 390
 391         return 0;
 392 }
 393
 394 static int
 395 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 396                    boolean_t enabled)
 397 {
 398         struct sfc_port *port;
 399         boolean_t *toggle;
 400         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 401         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 402         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 403         int rc = 0;
 404
 405         sfc_adapter_lock(sa);
 406
 407         port = &sa->port;
 408         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 409
 410         if (*toggle != enabled) {
 411                 *toggle = enabled;
 412
 413                 if (sfc_sa2shared(sa)->isolated) {
 414                         sfc_warn(sa, "isolated mode is active on the port");
 415                         sfc_warn(sa, "the change is to be applied on the next "
 416                                      "start provided that isolated mode is "
 417                                      "disabled prior the next start");
 418                 } else if ((sa->state == SFC_ETHDEV_STARTED) &&
 419                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 420                         *toggle = !(enabled);
 421                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 422                                  ((enabled) ? "enable" : "disable"), desc, rc);
 423
 424                         /*
 425                          * For promiscuous and all-multicast filters a
 426                          * permission failure should be reported as an
 427                          * unsupported filter.
 428                          */
 429                         if (rc == EPERM)
 430                                 rc = ENOTSUP;
 431                 }
 432         }
 433
 434         sfc_adapter_unlock(sa);
 435         return rc;
 436 }
 437
 438 static int
 439 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 440 {
 441         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 442
 443         SFC_ASSERT(rc >= 0);
 444         return -rc;
 445 }
 446
 447 static int
 448 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 449 {
 450         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 451
 452         SFC_ASSERT(rc >= 0);
 453         return -rc;
 454 }
 455
 456 static int
 457 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 458 {
 459         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 460
 461         SFC_ASSERT(rc >= 0);
 462         return -rc;
 463 }
 464
 465 static int
 466 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 467 {
 468         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 469
 470         SFC_ASSERT(rc >= 0);
 471         return -rc;
 472 }
 473
 474 static int
 475 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 476                    uint16_t nb_rx_desc, unsigned int socket_id,
 477                    const struct rte_eth_rxconf *rx_conf,
 478                    struct rte_mempool *mb_pool)
 479 {
 480         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 481         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 482         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
 483         struct sfc_rxq_info *rxq_info;
 484         sfc_sw_index_t sw_index;
 485         int rc;
 486
 487         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 488                      ethdev_qid, nb_rx_desc, socket_id);
 489
 490         sfc_adapter_lock(sa);
 491
 492         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
 493         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
 494                           rx_conf, mb_pool);
 495         if (rc != 0)
 496                 goto fail_rx_qinit;
 497
 498         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
 499         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
 500
 501         sfc_adapter_unlock(sa);
 502
 503         return 0;
 504
 505 fail_rx_qinit:
 506         sfc_adapter_unlock(sa);
 507         SFC_ASSERT(rc > 0);
 508         return -rc;
 509 }
 510
 511 static void
 512 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 513 {
 514         struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
 515         struct sfc_rxq *rxq;
 516         struct sfc_adapter *sa;
 517         sfc_sw_index_t sw_index;
 518
 519         if (dp_rxq == NULL)
 520                 return;
 521
 522         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 523         sa = rxq->evq->sa;
 524         sfc_adapter_lock(sa);
 525
 526         sw_index = dp_rxq->dpq.queue_id;
 527
 528         sfc_log_init(sa, "RxQ=%u", sw_index);
 529
 530         sfc_rx_qfini(sa, sw_index);
 531
 532         sfc_adapter_unlock(sa);
 533 }
 534
 535 static int
 536 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 537                    uint16_t nb_tx_desc, unsigned int socket_id,
 538                    const struct rte_eth_txconf *tx_conf)
 539 {
 540         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 541         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 542         struct sfc_txq_info *txq_info;
 543         sfc_sw_index_t sw_index;
 544         int rc;
 545
 546         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 547                      ethdev_qid, nb_tx_desc, socket_id);
 548
 549         sfc_adapter_lock(sa);
 550
 551         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
 552         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
 553         if (rc != 0)
 554                 goto fail_tx_qinit;
 555
 556         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
 557         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
 558
 559         sfc_adapter_unlock(sa);
 560         return 0;
 561
 562 fail_tx_qinit:
 563         sfc_adapter_unlock(sa);
 564         SFC_ASSERT(rc > 0);
 565         return -rc;
 566 }
 567
 568 static void
 569 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 570 {
 571         struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
 572         struct sfc_txq *txq;
 573         sfc_sw_index_t sw_index;
 574         struct sfc_adapter *sa;
 575
 576         if (dp_txq == NULL)
 577                 return;
 578
 579         txq = sfc_txq_by_dp_txq(dp_txq);
 580         sw_index = dp_txq->dpq.queue_id;
 581
 582         SFC_ASSERT(txq->evq != NULL);
 583         sa = txq->evq->sa;
 584
 585         sfc_log_init(sa, "TxQ = %u", sw_index);
 586
 587         sfc_adapter_lock(sa);
 588
 589         sfc_tx_qfini(sa, sw_index);
 590
 591         sfc_adapter_unlock(sa);
 592 }
 593
 594 static void
 595 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 596 {
 597         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 598         uint64_t pkts_sum = 0;
 599         uint64_t bytes_sum = 0;
 600         unsigned int i;
 601
 602         for (i = 0; i < sas->ethdev_rxq_count; ++i) {
 603                 struct sfc_rxq_info *rxq_info;
 604
 605                 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
 606                 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
 607                         union sfc_pkts_bytes qstats;
 608
 609                         sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
 610                         pkts_sum += qstats.pkts -
 611                                         sa->sw_stats.reset_rx_pkts[i];
 612                         bytes_sum += qstats.bytes -
 613                                         sa->sw_stats.reset_rx_bytes[i];
 614                 }
 615         }
 616
 617         *pkts = pkts_sum;
 618         *bytes = bytes_sum;
 619 }
 620
 621 static void
 622 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 623 {
 624         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 625         uint64_t pkts_sum = 0;
 626         uint64_t bytes_sum = 0;
 627         unsigned int i;
 628
 629         for (i = 0; i < sas->ethdev_txq_count; ++i) {
 630                 struct sfc_txq_info *txq_info;
 631
 632                 txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
 633                 if (txq_info->state & SFC_TXQ_INITIALIZED) {
 634                         union sfc_pkts_bytes qstats;
 635
 636                         sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
 637                         pkts_sum += qstats.pkts -
 638                                         sa->sw_stats.reset_tx_pkts[i];
 639                         bytes_sum += qstats.bytes -
 640                                         sa->sw_stats.reset_tx_bytes[i];
 641                 }
 642         }
 643
 644         *pkts = pkts_sum;
 645         *bytes = bytes_sum;
 646 }
 647
 648 /*
 649  * Some statistics are computed as A - B where A and B each increase
 650  * monotonically with some hardware counter(s) and the counters are read
 651  * asynchronously.
 652  *
 653  * If packet X is counted in A, but not counted in B yet, computed value is
 654  * greater than real.
 655  *
 656  * If packet X is not counted in A at the moment of reading the counter,
 657  * but counted in B at the moment of reading the counter, computed value
 658  * is less than real.
 659  *
 660  * However, counter which grows backward is worse evil than slightly wrong
 661  * value. So, let's try to guarantee that it never happens except may be
 662  * the case when the MAC stats are zeroed as a result of a NIC reset.
 663  */
 664 static void
 665 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 666 {
 667         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 668                 *stat = newval;
 669 }
 670
 671 static int
 672 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 673 {
 674         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 675         bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
 676         bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
 677         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 678         struct sfc_port *port = &sa->port;
 679         uint64_t *mac_stats;
 680         int ret;
 681
 682         sfc_adapter_lock(sa);
 683
 684         if (have_dp_rx_stats)
 685                 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
 686         if (have_dp_tx_stats)
 687                 sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
 688
 689         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 690         if (ret != 0)
 691                 goto unlock;
 692
 693         mac_stats = port->mac_stats_buf;
 694
 695         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 696                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 697                 if (!have_dp_rx_stats) {
 698                         stats->ipackets =
 699                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 700                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 701                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 702                         stats->ibytes =
 703                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 704                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 705                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 706
 707                         /* CRC is included in these stats, but shouldn't be */
 708                         stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 709                 }
 710                 if (!have_dp_tx_stats) {
 711                         stats->opackets =
 712                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 713                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 714                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 715                         stats->obytes =
 716                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 717                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 718                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 719
 720                         /* CRC is included in these stats, but shouldn't be */
 721                         stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 722                 }
 723                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 724                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 725         } else {
 726                 if (!have_dp_tx_stats) {
 727                         stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 728                         stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
 729                                 mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 730                 }
 731
 732                 /*
 733                  * Take into account stats which are whenever supported
 734                  * on EF10. If some stat is not supported by current
 735                  * firmware variant or HW revision, it is guaranteed
 736                  * to be zero in mac_stats.
 737                  */
 738                 stats->imissed =
 739                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 740                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 741                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 742                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 743                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 744                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 745                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 746                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 747                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 748                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 749                 stats->ierrors =
 750                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 751                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 752                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 753                 /* no oerrors counters supported on EF10 */
 754
 755                 if (!have_dp_rx_stats) {
 756                         /* Exclude missed, errors and pauses from Rx packets */
 757                         sfc_update_diff_stat(&port->ipackets,
 758                                 mac_stats[EFX_MAC_RX_PKTS] -
 759                                 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 760                                 stats->imissed - stats->ierrors);
 761                         stats->ipackets = port->ipackets;
 762                         stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
 763                                 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 764                 }
 765         }
 766
 767 unlock:
 768         sfc_adapter_unlock(sa);
 769         SFC_ASSERT(ret >= 0);
 770         return -ret;
 771 }
 772
 773 static int
 774 sfc_stats_reset(struct rte_eth_dev *dev)
 775 {
 776         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 777         struct sfc_port *port = &sa->port;
 778         int rc;
 779
 780         sfc_adapter_lock(sa);
 781
 782         if (sa->state != SFC_ETHDEV_STARTED) {
 783                 /*
 784                  * The operation cannot be done if port is not started; it
 785                  * will be scheduled to be done during the next port start
 786                  */
 787                 port->mac_stats_reset_pending = B_TRUE;
 788                 sfc_adapter_unlock(sa);
 789                 return 0;
 790         }
 791
 792         rc = sfc_port_reset_mac_stats(sa);
 793         if (rc != 0)
 794                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 795
 796         sfc_sw_xstats_reset(sa);
 797
 798         sfc_adapter_unlock(sa);
 799
 800         SFC_ASSERT(rc >= 0);
 801         return -rc;
 802 }
 803
 804 static unsigned int
 805 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 806 {
 807         struct sfc_port *port = &sa->port;
 808         unsigned int nb_supported;
 809
 810         sfc_adapter_lock(sa);
 811         nb_supported = port->mac_stats_nb_supported +
 812                        sfc_sw_xstats_get_nb_supported(sa);
 813         sfc_adapter_unlock(sa);
 814
 815         return nb_supported;
 816 }
 817
 818 static int
 819 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 820                unsigned int xstats_count)
 821 {
 822         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 823         unsigned int nb_written = 0;
 824         unsigned int nb_supported = 0;
 825         int rc;
 826
 827         if (unlikely(xstats == NULL))
 828                 return sfc_xstats_get_nb_supported(sa);
 829
 830         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 831         if (rc < 0)
 832                 return rc;
 833
 834         nb_supported = rc;
 835         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 836                                &nb_supported);
 837
 838         return nb_supported;
 839 }
 840
 841 static int
 842 sfc_xstats_get_names(struct rte_eth_dev *dev,
 843                      struct rte_eth_xstat_name *xstats_names,
 844                      unsigned int xstats_count)
 845 {
 846         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 847         struct sfc_port *port = &sa->port;
 848         unsigned int i;
 849         unsigned int nstats = 0;
 850         unsigned int nb_written = 0;
 851         int ret;
 852
 853         if (unlikely(xstats_names == NULL))
 854                 return sfc_xstats_get_nb_supported(sa);
 855
 856         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 857                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 858                         if (nstats < xstats_count) {
 859                                 strlcpy(xstats_names[nstats].name,
 860                                         efx_mac_stat_name(sa->nic, i),
 861                                         sizeof(xstats_names[0].name));
 862                                 nb_written++;
 863                         }
 864                         nstats++;
 865                 }
 866         }
 867
 868         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 869                                       &nb_written, &nstats);
 870         if (ret != 0) {
 871                 SFC_ASSERT(ret < 0);
 872                 return ret;
 873         }
 874
 875         return nstats;
 876 }
 877
 878 static int
 879 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 880                      uint64_t *values, unsigned int n)
 881 {
 882         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 883         struct sfc_port *port = &sa->port;
 884         unsigned int nb_supported;
 885         unsigned int i;
 886         int rc;
 887
 888         if (unlikely(ids == NULL || values == NULL))
 889                 return -EINVAL;
 890
 891         /*
 892          * Values array could be filled in nonsequential order. Fill values with
 893          * constant indicating invalid ID first.
 894          */
 895         for (i = 0; i < n; i++)
 896                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 897
 898         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 899         if (rc != 0)
 900                 return rc;
 901
 902         nb_supported = port->mac_stats_nb_supported;
 903         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 904
 905         /* Return number of written stats before invalid ID is encountered. */
 906         for (i = 0; i < n; i++) {
 907                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 908                         return i;
 909         }
 910
 911         return n;
 912 }
 913
 914 static int
 915 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 916                            const uint64_t *ids,
 917                            struct rte_eth_xstat_name *xstats_names,
 918                            unsigned int size)
 919 {
 920         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 921         struct sfc_port *port = &sa->port;
 922         unsigned int nb_supported;
 923         unsigned int i;
 924         int ret;
 925
 926         if (unlikely(xstats_names == NULL && ids != NULL) ||
 927             unlikely(xstats_names != NULL && ids == NULL))
 928                 return -EINVAL;
 929
 930         if (unlikely(xstats_names == NULL && ids == NULL))
 931                 return sfc_xstats_get_nb_supported(sa);
 932
 933         /*
 934          * Names array could be filled in nonsequential order. Fill names with
 935          * string indicating invalid ID first.
 936          */
 937         for (i = 0; i < size; i++)
 938                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 939
 940         sfc_adapter_lock(sa);
 941
 942         SFC_ASSERT(port->mac_stats_nb_supported <=
 943                    RTE_DIM(port->mac_stats_by_id));
 944
 945         for (i = 0; i < size; i++) {
 946                 if (ids[i] < port->mac_stats_nb_supported) {
 947                         strlcpy(xstats_names[i].name,
 948                                 efx_mac_stat_name(sa->nic,
 949                                                  port->mac_stats_by_id[ids[i]]),
 950                                 sizeof(xstats_names[0].name));
 951                 }
 952         }
 953
 954         nb_supported = port->mac_stats_nb_supported;
 955
 956         sfc_adapter_unlock(sa);
 957
 958         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 959                                             &nb_supported);
 960         if (ret != 0) {
 961                 SFC_ASSERT(ret < 0);
 962                 return ret;
 963         }
 964
 965         /* Return number of written names before invalid ID is encountered. */
 966         for (i = 0; i < size; i++) {
 967                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 968                         return i;
 969         }
 970
 971         return size;
 972 }
 973
 974 static int
 975 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 976 {
 977         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 978         unsigned int wanted_fc, link_fc;
 979
 980         memset(fc_conf, 0, sizeof(*fc_conf));
 981
 982         sfc_adapter_lock(sa);
 983
 984         if (sa->state == SFC_ETHDEV_STARTED)
 985                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 986         else
 987                 link_fc = sa->port.flow_ctrl;
 988
 989         switch (link_fc) {
 990         case 0:
 991                 fc_conf->mode = RTE_FC_NONE;
 992                 break;
 993         case EFX_FCNTL_RESPOND:
 994                 fc_conf->mode = RTE_FC_RX_PAUSE;
 995                 break;
 996         case EFX_FCNTL_GENERATE:
 997                 fc_conf->mode = RTE_FC_TX_PAUSE;
 998                 break;
 999         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
1000                 fc_conf->mode = RTE_FC_FULL;
1001                 break;
1002         default:
1003                 sfc_err(sa, "%s: unexpected flow control value %#x",
1004                         __func__, link_fc);
1005         }
1006
1007         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1008
1009         sfc_adapter_unlock(sa);
1010
1011         return 0;
1012 }
1013
1014 static int
1015 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1016 {
1017         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1018         struct sfc_port *port = &sa->port;
1019         unsigned int fcntl;
1020         int rc;
1021
1022         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1023             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1024             fc_conf->mac_ctrl_frame_fwd != 0) {
1025                 sfc_err(sa, "unsupported flow control settings specified");
1026                 rc = EINVAL;
1027                 goto fail_inval;
1028         }
1029
1030         switch (fc_conf->mode) {
1031         case RTE_FC_NONE:
1032                 fcntl = 0;
1033                 break;
1034         case RTE_FC_RX_PAUSE:
1035                 fcntl = EFX_FCNTL_RESPOND;
1036                 break;
1037         case RTE_FC_TX_PAUSE:
1038                 fcntl = EFX_FCNTL_GENERATE;
1039                 break;
1040         case RTE_FC_FULL:
1041                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1042                 break;
1043         default:
1044                 rc = EINVAL;
1045                 goto fail_inval;
1046         }
1047
1048         sfc_adapter_lock(sa);
1049
1050         if (sa->state == SFC_ETHDEV_STARTED) {
1051                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1052                 if (rc != 0)
1053                         goto fail_mac_fcntl_set;
1054         }
1055
1056         port->flow_ctrl = fcntl;
1057         port->flow_ctrl_autoneg = fc_conf->autoneg;
1058
1059         sfc_adapter_unlock(sa);
1060
1061         return 0;
1062
1063 fail_mac_fcntl_set:
1064         sfc_adapter_unlock(sa);
1065 fail_inval:
1066         SFC_ASSERT(rc > 0);
1067         return -rc;
1068 }
1069
1070 static int
1071 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1072 {
1073         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1074         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1075         boolean_t scatter_enabled;
1076         const char *error;
1077         unsigned int i;
1078
1079         for (i = 0; i < sas->rxq_count; i++) {
1080                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1081                         continue;
1082
1083                 scatter_enabled = (sas->rxq_info[i].type_flags &
1084                                    EFX_RXQ_FLAG_SCATTER);
1085
1086                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1087                                           encp->enc_rx_prefix_size,
1088                                           scatter_enabled,
1089                                           encp->enc_rx_scatter_max, &error)) {
1090                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1091                                 error);
1092                         return EINVAL;
1093                 }
1094         }
1095
1096         return 0;
1097 }
1098
1099 static int
1100 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1101 {
1102         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1103         size_t pdu = EFX_MAC_PDU(mtu);
1104         size_t old_pdu;
1105         int rc;
1106
1107         sfc_log_init(sa, "mtu=%u", mtu);
1108
1109         rc = EINVAL;
1110         if (pdu < EFX_MAC_PDU_MIN) {
1111                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1112                         (unsigned int)mtu, (unsigned int)pdu,
1113                         EFX_MAC_PDU_MIN);
1114                 goto fail_inval;
1115         }
1116         if (pdu > EFX_MAC_PDU_MAX) {
1117                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1118                         (unsigned int)mtu, (unsigned int)pdu,
1119                         (unsigned int)EFX_MAC_PDU_MAX);
1120                 goto fail_inval;
1121         }
1122
1123         sfc_adapter_lock(sa);
1124
1125         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1126         if (rc != 0)
1127                 goto fail_check_scatter;
1128
1129         if (pdu != sa->port.pdu) {
1130                 if (sa->state == SFC_ETHDEV_STARTED) {
1131                         sfc_stop(sa);
1132
1133                         old_pdu = sa->port.pdu;
1134                         sa->port.pdu = pdu;
1135                         rc = sfc_start(sa);
1136                         if (rc != 0)
1137                                 goto fail_start;
1138                 } else {
1139                         sa->port.pdu = pdu;
1140                 }
1141         }
1142
1143         /*
1144          * The driver does not use it, but other PMDs update jumbo frame
1145          * flag and max_rx_pkt_len when MTU is set.
1146          */
1147         if (mtu > RTE_ETHER_MTU) {
1148                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1149                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1150         }
1151
1152         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1153
1154         sfc_adapter_unlock(sa);
1155
1156         sfc_log_init(sa, "done");
1157         return 0;
1158
1159 fail_start:
1160         sa->port.pdu = old_pdu;
1161         if (sfc_start(sa) != 0)
1162                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1163                         "PDU max size - port is stopped",
1164                         (unsigned int)pdu, (unsigned int)old_pdu);
1165
1166 fail_check_scatter:
1167         sfc_adapter_unlock(sa);
1168
1169 fail_inval:
1170         sfc_log_init(sa, "failed %d", rc);
1171         SFC_ASSERT(rc > 0);
1172         return -rc;
1173 }
1174 static int
1175 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1176 {
1177         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1178         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1179         struct sfc_port *port = &sa->port;
1180         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1181         int rc = 0;
1182
1183         sfc_adapter_lock(sa);
1184
1185         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1186                 goto unlock;
1187
1188         /*
1189          * Copy the address to the device private data so that
1190          * it could be recalled in the case of adapter restart.
1191          */
1192         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1193
1194         /*
1195          * Neither of the two following checks can return
1196          * an error. The new MAC address is preserved in
1197          * the device private data and can be activated
1198          * on the next port start if the user prevents
1199          * isolated mode from being enabled.
1200          */
1201         if (sfc_sa2shared(sa)->isolated) {
1202                 sfc_warn(sa, "isolated mode is active on the port");
1203                 sfc_warn(sa, "will not set MAC address");
1204                 goto unlock;
1205         }
1206
1207         if (sa->state != SFC_ETHDEV_STARTED) {
1208                 sfc_notice(sa, "the port is not started");
1209                 sfc_notice(sa, "the new MAC address will be set on port start");
1210
1211                 goto unlock;
1212         }
1213
1214         if (encp->enc_allow_set_mac_with_installed_filters) {
1215                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1216                 if (rc != 0) {
1217                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1218                         goto unlock;
1219                 }
1220
1221                 /*
1222                  * Changing the MAC address by means of MCDI request
1223                  * has no effect on received traffic, therefore
1224                  * we also need to update unicast filters
1225                  */
1226                 rc = sfc_set_rx_mode_unchecked(sa);
1227                 if (rc != 0) {
1228                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1229                         /* Rollback the old address */
1230                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1231                         (void)sfc_set_rx_mode_unchecked(sa);
1232                 }
1233         } else {
1234                 sfc_warn(sa, "cannot set MAC address with filters installed");
1235                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1236                 sfc_warn(sa, "(some traffic may be dropped)");
1237
1238                 /*
1239                  * Since setting MAC address with filters installed is not
1240                  * allowed on the adapter, the new MAC address will be set
1241                  * by means of adapter restart. sfc_start() shall retrieve
1242                  * the new address from the device private data and set it.
1243                  */
1244                 sfc_stop(sa);
1245                 rc = sfc_start(sa);
1246                 if (rc != 0)
1247                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1248         }
1249
1250 unlock:
1251         if (rc != 0)
1252                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1253
1254         sfc_adapter_unlock(sa);
1255
1256         SFC_ASSERT(rc >= 0);
1257         return -rc;
1258 }
1259
1260
1261 static int
1262 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1263                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1264 {
1265         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1266         struct sfc_port *port = &sa->port;
1267         uint8_t *mc_addrs = port->mcast_addrs;
1268         int rc;
1269         unsigned int i;
1270
1271         if (sfc_sa2shared(sa)->isolated) {
1272                 sfc_err(sa, "isolated mode is active on the port");
1273                 sfc_err(sa, "will not set multicast address list");
1274                 return -ENOTSUP;
1275         }
1276
1277         if (mc_addrs == NULL)
1278                 return -ENOBUFS;
1279
1280         if (nb_mc_addr > port->max_mcast_addrs) {
1281                 sfc_err(sa, "too many multicast addresses: %u > %u",
1282                          nb_mc_addr, port->max_mcast_addrs);
1283                 return -EINVAL;
1284         }
1285
1286         for (i = 0; i < nb_mc_addr; ++i) {
1287                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1288                                  EFX_MAC_ADDR_LEN);
1289                 mc_addrs += EFX_MAC_ADDR_LEN;
1290         }
1291
1292         port->nb_mcast_addrs = nb_mc_addr;
1293
1294         if (sa->state != SFC_ETHDEV_STARTED)
1295                 return 0;
1296
1297         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1298                                         port->nb_mcast_addrs);
1299         if (rc != 0)
1300                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1301
1302         SFC_ASSERT(rc >= 0);
1303         return -rc;
1304 }
1305
1306 /*
1307  * The function is used by the secondary process as well. It must not
1308  * use any process-local pointers from the adapter data.
1309  */
1310 static void
1311 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1312                       struct rte_eth_rxq_info *qinfo)
1313 {
1314         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1315         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1316         struct sfc_rxq_info *rxq_info;
1317
1318         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1319
1320         qinfo->mp = rxq_info->refill_mb_pool;
1321         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1322         qinfo->conf.rx_drop_en = 1;
1323         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1324         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1325         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1326                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1327                 qinfo->scattered_rx = 1;
1328         }
1329         qinfo->nb_desc = rxq_info->entries;
1330 }
1331
1332 /*
1333  * The function is used by the secondary process as well. It must not
1334  * use any process-local pointers from the adapter data.
1335  */
1336 static void
1337 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1338                       struct rte_eth_txq_info *qinfo)
1339 {
1340         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1341         struct sfc_txq_info *txq_info;
1342
1343         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1344
1345         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1346
1347         memset(qinfo, 0, sizeof(*qinfo));
1348
1349         qinfo->conf.offloads = txq_info->offloads;
1350         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1351         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1352         qinfo->nb_desc = txq_info->entries;
1353 }
1354
1355 /*
1356  * The function is used by the secondary process as well. It must not
1357  * use any process-local pointers from the adapter data.
1358  */
1359 static uint32_t
1360 sfc_rx_queue_count(void *rx_queue)
1361 {
1362         struct sfc_dp_rxq *dp_rxq = rx_queue;
1363         const struct sfc_dp_rx *dp_rx;
1364         struct sfc_rxq_info *rxq_info;
1365
1366         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1367         rxq_info = sfc_rxq_info_by_dp_rxq(dp_rxq);
1368
1369         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1370                 return 0;
1371
1372         return dp_rx->qdesc_npending(dp_rxq);
1373 }
1374
1375 /*
1376  * The function is used by the secondary process as well. It must not
1377  * use any process-local pointers from the adapter data.
1378  */
1379 static int
1380 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1381 {
1382         struct sfc_dp_rxq *dp_rxq = queue;
1383         const struct sfc_dp_rx *dp_rx;
1384
1385         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1386
1387         return dp_rx->qdesc_status(dp_rxq, offset);
1388 }
1389
1390 /*
1391  * The function is used by the secondary process as well. It must not
1392  * use any process-local pointers from the adapter data.
1393  */
1394 static int
1395 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1396 {
1397         struct sfc_dp_txq *dp_txq = queue;
1398         const struct sfc_dp_tx *dp_tx;
1399
1400         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1401
1402         return dp_tx->qdesc_status(dp_txq, offset);
1403 }
1404
1405 static int
1406 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1407 {
1408         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1409         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1410         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1411         struct sfc_rxq_info *rxq_info;
1412         sfc_sw_index_t sw_index;
1413         int rc;
1414
1415         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1416
1417         sfc_adapter_lock(sa);
1418
1419         rc = EINVAL;
1420         if (sa->state != SFC_ETHDEV_STARTED)
1421                 goto fail_not_started;
1422
1423         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1424         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1425                 goto fail_not_setup;
1426
1427         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1428         rc = sfc_rx_qstart(sa, sw_index);
1429         if (rc != 0)
1430                 goto fail_rx_qstart;
1431
1432         rxq_info->deferred_started = B_TRUE;
1433
1434         sfc_adapter_unlock(sa);
1435
1436         return 0;
1437
1438 fail_rx_qstart:
1439 fail_not_setup:
1440 fail_not_started:
1441         sfc_adapter_unlock(sa);
1442         SFC_ASSERT(rc > 0);
1443         return -rc;
1444 }
1445
1446 static int
1447 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1448 {
1449         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1450         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1451         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1452         struct sfc_rxq_info *rxq_info;
1453         sfc_sw_index_t sw_index;
1454
1455         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1456
1457         sfc_adapter_lock(sa);
1458
1459         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1460         sfc_rx_qstop(sa, sw_index);
1461
1462         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1463         rxq_info->deferred_started = B_FALSE;
1464
1465         sfc_adapter_unlock(sa);
1466
1467         return 0;
1468 }
1469
1470 static int
1471 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1472 {
1473         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1474         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1475         struct sfc_txq_info *txq_info;
1476         sfc_sw_index_t sw_index;
1477         int rc;
1478
1479         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1480
1481         sfc_adapter_lock(sa);
1482
1483         rc = EINVAL;
1484         if (sa->state != SFC_ETHDEV_STARTED)
1485                 goto fail_not_started;
1486
1487         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1488         if (txq_info->state != SFC_TXQ_INITIALIZED)
1489                 goto fail_not_setup;
1490
1491         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1492         rc = sfc_tx_qstart(sa, sw_index);
1493         if (rc != 0)
1494                 goto fail_tx_qstart;
1495
1496         txq_info->deferred_started = B_TRUE;
1497
1498         sfc_adapter_unlock(sa);
1499         return 0;
1500
1501 fail_tx_qstart:
1502
1503 fail_not_setup:
1504 fail_not_started:
1505         sfc_adapter_unlock(sa);
1506         SFC_ASSERT(rc > 0);
1507         return -rc;
1508 }
1509
1510 static int
1511 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1512 {
1513         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1514         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1515         struct sfc_txq_info *txq_info;
1516         sfc_sw_index_t sw_index;
1517
1518         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1519
1520         sfc_adapter_lock(sa);
1521
1522         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1523         sfc_tx_qstop(sa, sw_index);
1524
1525         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1526         txq_info->deferred_started = B_FALSE;
1527
1528         sfc_adapter_unlock(sa);
1529         return 0;
1530 }
1531
1532 static efx_tunnel_protocol_t
1533 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1534 {
1535         switch (rte_type) {
1536         case RTE_TUNNEL_TYPE_VXLAN:
1537                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1538         case RTE_TUNNEL_TYPE_GENEVE:
1539                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1540         default:
1541                 return EFX_TUNNEL_NPROTOS;
1542         }
1543 }
1544
1545 enum sfc_udp_tunnel_op_e {
1546         SFC_UDP_TUNNEL_ADD_PORT,
1547         SFC_UDP_TUNNEL_DEL_PORT,
1548 };
1549
1550 static int
1551 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1552                       struct rte_eth_udp_tunnel *tunnel_udp,
1553                       enum sfc_udp_tunnel_op_e op)
1554 {
1555         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1556         efx_tunnel_protocol_t tunnel_proto;
1557         int rc;
1558
1559         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1560                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1561                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1562                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1563
1564         tunnel_proto =
1565                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1566         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1567                 rc = ENOTSUP;
1568                 goto fail_bad_proto;
1569         }
1570
1571         sfc_adapter_lock(sa);
1572
1573         switch (op) {
1574         case SFC_UDP_TUNNEL_ADD_PORT:
1575                 rc = efx_tunnel_config_udp_add(sa->nic,
1576                                                tunnel_udp->udp_port,
1577                                                tunnel_proto);
1578                 break;
1579         case SFC_UDP_TUNNEL_DEL_PORT:
1580                 rc = efx_tunnel_config_udp_remove(sa->nic,
1581                                                   tunnel_udp->udp_port,
1582                                                   tunnel_proto);
1583                 break;
1584         default:
1585                 rc = EINVAL;
1586                 goto fail_bad_op;
1587         }
1588
1589         if (rc != 0)
1590                 goto fail_op;
1591
1592         if (sa->state == SFC_ETHDEV_STARTED) {
1593                 rc = efx_tunnel_reconfigure(sa->nic);
1594                 if (rc == EAGAIN) {
1595                         /*
1596                          * Configuration is accepted by FW and MC reboot
1597                          * is initiated to apply the changes. MC reboot
1598                          * will be handled in a usual way (MC reboot
1599                          * event on management event queue and adapter
1600                          * restart).
1601                          */
1602                         rc = 0;
1603                 } else if (rc != 0) {
1604                         goto fail_reconfigure;
1605                 }
1606         }
1607
1608         sfc_adapter_unlock(sa);
1609         return 0;
1610
1611 fail_reconfigure:
1612         /* Remove/restore entry since the change makes the trouble */
1613         switch (op) {
1614         case SFC_UDP_TUNNEL_ADD_PORT:
1615                 (void)efx_tunnel_config_udp_remove(sa->nic,
1616                                                    tunnel_udp->udp_port,
1617                                                    tunnel_proto);
1618                 break;
1619         case SFC_UDP_TUNNEL_DEL_PORT:
1620                 (void)efx_tunnel_config_udp_add(sa->nic,
1621                                                 tunnel_udp->udp_port,
1622                                                 tunnel_proto);
1623                 break;
1624         }
1625
1626 fail_op:
1627 fail_bad_op:
1628         sfc_adapter_unlock(sa);
1629
1630 fail_bad_proto:
1631         SFC_ASSERT(rc > 0);
1632         return -rc;
1633 }
1634
1635 static int
1636 sfc_dev_udp_tunnel_port_add(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_ADD_PORT);
1640 }
1641
1642 static int
1643 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1644                             struct rte_eth_udp_tunnel *tunnel_udp)
1645 {
1646         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1647 }
1648
1649 /*
1650  * The function is used by the secondary process as well. It must not
1651  * use any process-local pointers from the adapter data.
1652  */
1653 static int
1654 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1655                           struct rte_eth_rss_conf *rss_conf)
1656 {
1657         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1658         struct sfc_rss *rss = &sas->rss;
1659
1660         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1661                 return -ENOTSUP;
1662
1663         /*
1664          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1665          * hence, conversion is done here to derive a correct set of ETH_RSS
1666          * flags which corresponds to the active EFX configuration stored
1667          * locally in 'sfc_adapter' and kept up-to-date
1668          */
1669         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1670         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1671         if (rss_conf->rss_key != NULL)
1672                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1673
1674         return 0;
1675 }
1676
1677 static int
1678 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1679                         struct rte_eth_rss_conf *rss_conf)
1680 {
1681         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1682         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1683         unsigned int efx_hash_types;
1684         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1685         unsigned int n_contexts;
1686         unsigned int mode_i = 0;
1687         unsigned int key_i = 0;
1688         unsigned int i = 0;
1689         int rc = 0;
1690
1691         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1692
1693         if (sfc_sa2shared(sa)->isolated)
1694                 return -ENOTSUP;
1695
1696         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1697                 sfc_err(sa, "RSS is not available");
1698                 return -ENOTSUP;
1699         }
1700
1701         if (rss->channels == 0) {
1702                 sfc_err(sa, "RSS is not configured");
1703                 return -EINVAL;
1704         }
1705
1706         if ((rss_conf->rss_key != NULL) &&
1707             (rss_conf->rss_key_len != sizeof(rss->key))) {
1708                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1709                         sizeof(rss->key));
1710                 return -EINVAL;
1711         }
1712
1713         sfc_adapter_lock(sa);
1714
1715         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1716         if (rc != 0)
1717                 goto fail_rx_hf_rte_to_efx;
1718
1719         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1720                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1721                                            rss->hash_alg, efx_hash_types,
1722                                            B_TRUE);
1723                 if (rc != 0)
1724                         goto fail_scale_mode_set;
1725         }
1726
1727         if (rss_conf->rss_key != NULL) {
1728                 if (sa->state == SFC_ETHDEV_STARTED) {
1729                         for (key_i = 0; key_i < n_contexts; key_i++) {
1730                                 rc = efx_rx_scale_key_set(sa->nic,
1731                                                           contexts[key_i],
1732                                                           rss_conf->rss_key,
1733                                                           sizeof(rss->key));
1734                                 if (rc != 0)
1735                                         goto fail_scale_key_set;
1736                         }
1737                 }
1738
1739                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1740         }
1741
1742         rss->hash_types = efx_hash_types;
1743
1744         sfc_adapter_unlock(sa);
1745
1746         return 0;
1747
1748 fail_scale_key_set:
1749         for (i = 0; i < key_i; i++) {
1750                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1751                                          sizeof(rss->key)) != 0)
1752                         sfc_err(sa, "failed to restore RSS key");
1753         }
1754
1755 fail_scale_mode_set:
1756         for (i = 0; i < mode_i; i++) {
1757                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1758                                           EFX_RX_HASHALG_TOEPLITZ,
1759                                           rss->hash_types, B_TRUE) != 0)
1760                         sfc_err(sa, "failed to restore RSS mode");
1761         }
1762
1763 fail_rx_hf_rte_to_efx:
1764         sfc_adapter_unlock(sa);
1765         return -rc;
1766 }
1767
1768 /*
1769  * The function is used by the secondary process as well. It must not
1770  * use any process-local pointers from the adapter data.
1771  */
1772 static int
1773 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1774                        struct rte_eth_rss_reta_entry64 *reta_conf,
1775                        uint16_t reta_size)
1776 {
1777         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1778         struct sfc_rss *rss = &sas->rss;
1779         int entry;
1780
1781         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1782                 return -ENOTSUP;
1783
1784         if (rss->channels == 0)
1785                 return -EINVAL;
1786
1787         if (reta_size != EFX_RSS_TBL_SIZE)
1788                 return -EINVAL;
1789
1790         for (entry = 0; entry < reta_size; entry++) {
1791                 int grp = entry / RTE_RETA_GROUP_SIZE;
1792                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1793
1794                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1795                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1796         }
1797
1798         return 0;
1799 }
1800
1801 static int
1802 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1803                         struct rte_eth_rss_reta_entry64 *reta_conf,
1804                         uint16_t reta_size)
1805 {
1806         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1807         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1808         unsigned int *rss_tbl_new;
1809         uint16_t entry;
1810         int rc = 0;
1811
1812
1813         if (sfc_sa2shared(sa)->isolated)
1814                 return -ENOTSUP;
1815
1816         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1817                 sfc_err(sa, "RSS is not available");
1818                 return -ENOTSUP;
1819         }
1820
1821         if (rss->channels == 0) {
1822                 sfc_err(sa, "RSS is not configured");
1823                 return -EINVAL;
1824         }
1825
1826         if (reta_size != EFX_RSS_TBL_SIZE) {
1827                 sfc_err(sa, "RETA size is wrong (should be %u)",
1828                         EFX_RSS_TBL_SIZE);
1829                 return -EINVAL;
1830         }
1831
1832         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1833         if (rss_tbl_new == NULL)
1834                 return -ENOMEM;
1835
1836         sfc_adapter_lock(sa);
1837
1838         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1839
1840         for (entry = 0; entry < reta_size; entry++) {
1841                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1842                 struct rte_eth_rss_reta_entry64 *grp;
1843
1844                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1845
1846                 if (grp->mask & (1ull << grp_idx)) {
1847                         if (grp->reta[grp_idx] >= rss->channels) {
1848                                 rc = EINVAL;
1849                                 goto bad_reta_entry;
1850                         }
1851                         rss_tbl_new[entry] = grp->reta[grp_idx];
1852                 }
1853         }
1854
1855         if (sa->state == SFC_ETHDEV_STARTED) {
1856                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1857                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1858                 if (rc != 0)
1859                         goto fail_scale_tbl_set;
1860         }
1861
1862         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1863
1864 fail_scale_tbl_set:
1865 bad_reta_entry:
1866         sfc_adapter_unlock(sa);
1867
1868         rte_free(rss_tbl_new);
1869
1870         SFC_ASSERT(rc >= 0);
1871         return -rc;
1872 }
1873
1874 static int
1875 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1876                      const struct rte_flow_ops **ops)
1877 {
1878         *ops = &sfc_flow_ops;
1879         return 0;
1880 }
1881
1882 static int
1883 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1884 {
1885         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1886
1887         /*
1888          * If Rx datapath does not provide callback to check mempool,
1889          * all pools are supported.
1890          */
1891         if (sap->dp_rx->pool_ops_supported == NULL)
1892                 return 1;
1893
1894         return sap->dp_rx->pool_ops_supported(pool);
1895 }
1896
1897 static int
1898 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1899 {
1900         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1901         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1902         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1903         struct sfc_rxq_info *rxq_info;
1904
1905         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1906
1907         return sap->dp_rx->intr_enable(rxq_info->dp);
1908 }
1909
1910 static int
1911 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1912 {
1913         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1914         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1915         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1916         struct sfc_rxq_info *rxq_info;
1917
1918         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1919
1920         return sap->dp_rx->intr_disable(rxq_info->dp);
1921 }
1922
1923 struct sfc_mport_journal_ctx {
1924         struct sfc_adapter              *sa;
1925         uint16_t                        switch_domain_id;
1926         uint32_t                        mcdi_handle;
1927         bool                            controllers_assigned;
1928         efx_pcie_interface_t            *controllers;
1929         size_t                          nb_controllers;
1930 };
1931
1932 static int
1933 sfc_journal_ctx_add_controller(struct sfc_mport_journal_ctx *ctx,
1934                                efx_pcie_interface_t intf)
1935 {
1936         efx_pcie_interface_t *new_controllers;
1937         size_t i, target;
1938         size_t new_size;
1939
1940         if (ctx->controllers == NULL) {
1941                 ctx->controllers = rte_malloc("sfc_controller_mapping",
1942                                               sizeof(ctx->controllers[0]), 0);
1943                 if (ctx->controllers == NULL)
1944                         return ENOMEM;
1945
1946                 ctx->controllers[0] = intf;
1947                 ctx->nb_controllers = 1;
1948
1949                 return 0;
1950         }
1951
1952         for (i = 0; i < ctx->nb_controllers; i++) {
1953                 if (ctx->controllers[i] == intf)
1954                         return 0;
1955                 if (ctx->controllers[i] > intf)
1956                         break;
1957         }
1958         target = i;
1959
1960         ctx->nb_controllers += 1;
1961         new_size = ctx->nb_controllers * sizeof(ctx->controllers[0]);
1962
1963         new_controllers = rte_realloc(ctx->controllers, new_size, 0);
1964         if (new_controllers == NULL) {
1965                 rte_free(ctx->controllers);
1966                 return ENOMEM;
1967         }
1968         ctx->controllers = new_controllers;
1969
1970         for (i = target + 1; i < ctx->nb_controllers; i++)
1971                 ctx->controllers[i] = ctx->controllers[i - 1];
1972
1973         ctx->controllers[target] = intf;
1974
1975         return 0;
1976 }
1977
1978 static efx_rc_t
1979 sfc_process_mport_journal_entry(struct sfc_mport_journal_ctx *ctx,
1980                                 efx_mport_desc_t *mport)
1981 {
1982         struct sfc_mae_switch_port_request req;
1983         efx_mport_sel_t entity_selector;
1984         efx_mport_sel_t ethdev_mport;
1985         uint16_t switch_port_id;
1986         efx_rc_t efx_rc;
1987         int rc;
1988
1989         sfc_dbg(ctx->sa,
1990                 "processing mport id %u (controller %u pf %u vf %u)",
1991                 mport->emd_id.id, mport->emd_vnic.ev_intf,
1992                 mport->emd_vnic.ev_pf, mport->emd_vnic.ev_vf);
1993         efx_mae_mport_invalid(&ethdev_mport);
1994
1995         if (!ctx->controllers_assigned) {
1996                 rc = sfc_journal_ctx_add_controller(ctx,
1997                                                     mport->emd_vnic.ev_intf);
1998                 if (rc != 0)
1999                         return rc;
2000         }
2001
2002         /* Build Mport selector */
2003         efx_rc = efx_mae_mport_by_pcie_mh_function(mport->emd_vnic.ev_intf,
2004                                                 mport->emd_vnic.ev_pf,
2005                                                 mport->emd_vnic.ev_vf,
2006                                                 &entity_selector);
2007         if (efx_rc != 0) {
2008                 sfc_err(ctx->sa, "failed to build entity mport selector for c%upf%uvf%u",
2009                         mport->emd_vnic.ev_intf,
2010                         mport->emd_vnic.ev_pf,
2011                         mport->emd_vnic.ev_vf);
2012                 return efx_rc;
2013         }
2014
2015         rc = sfc_mae_switch_port_id_by_entity(ctx->switch_domain_id,
2016                                               &entity_selector,
2017                                               SFC_MAE_SWITCH_PORT_REPRESENTOR,
2018                                               &switch_port_id);
2019         switch (rc) {
2020         case 0:
2021                 /* Already registered */
2022                 break;
2023         case ENOENT:
2024                 /*
2025                  * No representor has been created for this entity.
2026                  * Create a dummy switch registry entry with an invalid ethdev
2027                  * mport selector. When a corresponding representor is created,
2028                  * this entry will be updated.
2029                  */
2030                 req.type = SFC_MAE_SWITCH_PORT_REPRESENTOR;
2031                 req.entity_mportp = &entity_selector;
2032                 req.ethdev_mportp = &ethdev_mport;
2033                 req.ethdev_port_id = RTE_MAX_ETHPORTS;
2034                 req.port_data.repr.intf = mport->emd_vnic.ev_intf;
2035                 req.port_data.repr.pf = mport->emd_vnic.ev_pf;
2036                 req.port_data.repr.vf = mport->emd_vnic.ev_vf;
2037
2038                 rc = sfc_mae_assign_switch_port(ctx->switch_domain_id,
2039                                                 &req, &switch_port_id);
2040                 if (rc != 0) {
2041                         sfc_err(ctx->sa,
2042                                 "failed to assign MAE switch port for c%upf%uvf%u: %s",
2043                                 mport->emd_vnic.ev_intf,
2044                                 mport->emd_vnic.ev_pf,
2045                                 mport->emd_vnic.ev_vf,
2046                                 rte_strerror(rc));
2047                         return rc;
2048                 }
2049                 break;
2050         default:
2051                 sfc_err(ctx->sa, "failed to find MAE switch port for c%upf%uvf%u: %s",
2052                         mport->emd_vnic.ev_intf,
2053                         mport->emd_vnic.ev_pf,
2054                         mport->emd_vnic.ev_vf,
2055                         rte_strerror(rc));
2056                 return rc;
2057         }
2058
2059         return 0;
2060 }
2061
2062 static efx_rc_t
2063 sfc_process_mport_journal_cb(void *data, efx_mport_desc_t *mport,
2064                              size_t mport_len)
2065 {
2066         struct sfc_mport_journal_ctx *ctx = data;
2067
2068         if (ctx == NULL || ctx->sa == NULL) {
2069                 sfc_err(ctx->sa, "received NULL context or SFC adapter");
2070                 return EINVAL;
2071         }
2072
2073         if (mport_len != sizeof(*mport)) {
2074                 sfc_err(ctx->sa, "actual and expected mport buffer sizes differ");
2075                 return EINVAL;
2076         }
2077
2078         SFC_ASSERT(sfc_adapter_is_locked(ctx->sa));
2079
2080         /*
2081          * If a zombie flag is set, it means the mport has been marked for
2082          * deletion and cannot be used for any new operations. The mport will
2083          * be destroyed completely once all references to it are released.
2084          */
2085         if (mport->emd_zombie) {
2086                 sfc_dbg(ctx->sa, "mport is a zombie, skipping");
2087                 return 0;
2088         }
2089         if (mport->emd_type != EFX_MPORT_TYPE_VNIC) {
2090                 sfc_dbg(ctx->sa, "mport is not a VNIC, skipping");
2091                 return 0;
2092         }
2093         if (mport->emd_vnic.ev_client_type != EFX_MPORT_VNIC_CLIENT_FUNCTION) {
2094                 sfc_dbg(ctx->sa, "mport is not a function, skipping");
2095                 return 0;
2096         }
2097         if (mport->emd_vnic.ev_handle == ctx->mcdi_handle) {
2098                 sfc_dbg(ctx->sa, "mport is this driver instance, skipping");
2099                 return 0;
2100         }
2101
2102         return sfc_process_mport_journal_entry(ctx, mport);
2103 }
2104
2105 static int
2106 sfc_process_mport_journal(struct sfc_adapter *sa)
2107 {
2108         struct sfc_mport_journal_ctx ctx;
2109         const efx_pcie_interface_t *controllers;
2110         size_t nb_controllers;
2111         efx_rc_t efx_rc;
2112         int rc;
2113
2114         memset(&ctx, 0, sizeof(ctx));
2115         ctx.sa = sa;
2116         ctx.switch_domain_id = sa->mae.switch_domain_id;
2117
2118         efx_rc = efx_mcdi_get_own_client_handle(sa->nic, &ctx.mcdi_handle);
2119         if (efx_rc != 0) {
2120                 sfc_err(sa, "failed to get own MCDI handle");
2121                 SFC_ASSERT(efx_rc > 0);
2122                 return efx_rc;
2123         }
2124
2125         rc = sfc_mae_switch_domain_controllers(ctx.switch_domain_id,
2126                                                &controllers, &nb_controllers);
2127         if (rc != 0) {
2128                 sfc_err(sa, "failed to get controller mapping");
2129                 return rc;
2130         }
2131
2132         ctx.controllers_assigned = controllers != NULL;
2133         ctx.controllers = NULL;
2134         ctx.nb_controllers = 0;
2135
2136         efx_rc = efx_mae_read_mport_journal(sa->nic,
2137                                             sfc_process_mport_journal_cb, &ctx);
2138         if (efx_rc != 0) {
2139                 sfc_err(sa, "failed to process MAE mport journal");
2140                 SFC_ASSERT(efx_rc > 0);
2141                 return efx_rc;
2142         }
2143
2144         if (controllers == NULL) {
2145                 rc = sfc_mae_switch_domain_map_controllers(ctx.switch_domain_id,
2146                                                            ctx.controllers,
2147                                                            ctx.nb_controllers);
2148                 if (rc != 0)
2149                         return rc;
2150         }
2151
2152         return 0;
2153 }
2154
2155 static void
2156 sfc_count_representors_cb(enum sfc_mae_switch_port_type type,
2157                           const efx_mport_sel_t *ethdev_mportp __rte_unused,
2158                           uint16_t ethdev_port_id __rte_unused,
2159                           const efx_mport_sel_t *entity_mportp __rte_unused,
2160                           uint16_t switch_port_id __rte_unused,
2161                           union sfc_mae_switch_port_data *port_datap
2162                                 __rte_unused,
2163                           void *user_datap)
2164 {
2165         int *counter = user_datap;
2166
2167         SFC_ASSERT(counter != NULL);
2168
2169         if (type == SFC_MAE_SWITCH_PORT_REPRESENTOR)
2170                 (*counter)++;
2171 }
2172
2173 struct sfc_get_representors_ctx {
2174         struct rte_eth_representor_info *info;
2175         struct sfc_adapter              *sa;
2176         uint16_t                        switch_domain_id;
2177         const efx_pcie_interface_t      *controllers;
2178         size_t                          nb_controllers;
2179 };
2180
2181 static void
2182 sfc_get_representors_cb(enum sfc_mae_switch_port_type type,
2183                         const efx_mport_sel_t *ethdev_mportp __rte_unused,
2184                         uint16_t ethdev_port_id __rte_unused,
2185                         const efx_mport_sel_t *entity_mportp __rte_unused,
2186                         uint16_t switch_port_id,
2187                         union sfc_mae_switch_port_data *port_datap,
2188                         void *user_datap)
2189 {
2190         struct sfc_get_representors_ctx *ctx = user_datap;
2191         struct rte_eth_representor_range *range;
2192         int ret;
2193         int rc;
2194
2195         SFC_ASSERT(ctx != NULL);
2196         SFC_ASSERT(ctx->info != NULL);
2197         SFC_ASSERT(ctx->sa != NULL);
2198
2199         if (type != SFC_MAE_SWITCH_PORT_REPRESENTOR) {
2200                 sfc_dbg(ctx->sa, "not a representor, skipping");
2201                 return;
2202         }
2203         if (ctx->info->nb_ranges >= ctx->info->nb_ranges_alloc) {
2204                 sfc_dbg(ctx->sa, "info structure is full already");
2205                 return;
2206         }
2207
2208         range = &ctx->info->ranges[ctx->info->nb_ranges];
2209         rc = sfc_mae_switch_controller_from_mapping(ctx->controllers,
2210                                                     ctx->nb_controllers,
2211                                                     port_datap->repr.intf,
2212                                                     &range->controller);
2213         if (rc != 0) {
2214                 sfc_err(ctx->sa, "invalid representor controller: %d",
2215                         port_datap->repr.intf);
2216                 range->controller = -1;
2217         }
2218         range->pf = port_datap->repr.pf;
2219         range->id_base = switch_port_id;
2220         range->id_end = switch_port_id;
2221
2222         if (port_datap->repr.vf != EFX_PCI_VF_INVALID) {
2223                 range->type = RTE_ETH_REPRESENTOR_VF;
2224                 range->vf = port_datap->repr.vf;
2225                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2226                                "c%dpf%dvf%d", range->controller, range->pf,
2227                                range->vf);
2228         } else {
2229                 range->type = RTE_ETH_REPRESENTOR_PF;
2230                 ret = snprintf(range->name, RTE_DEV_NAME_MAX_LEN,
2231                          "c%dpf%d", range->controller, range->pf);
2232         }
2233         if (ret >= RTE_DEV_NAME_MAX_LEN) {
2234                 sfc_err(ctx->sa, "representor name has been truncated: %s",
2235                         range->name);
2236         }
2237
2238         ctx->info->nb_ranges++;
2239 }
2240
2241 static int
2242 sfc_representor_info_get(struct rte_eth_dev *dev,
2243                          struct rte_eth_representor_info *info)
2244 {
2245         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2246         struct sfc_get_representors_ctx get_repr_ctx;
2247         const efx_nic_cfg_t *nic_cfg;
2248         uint16_t switch_domain_id;
2249         uint32_t nb_repr;
2250         int controller;
2251         int rc;
2252
2253         sfc_adapter_lock(sa);
2254
2255         if (sa->mae.status != SFC_MAE_STATUS_SUPPORTED) {
2256                 sfc_adapter_unlock(sa);
2257                 return -ENOTSUP;
2258         }
2259
2260         rc = sfc_process_mport_journal(sa);
2261         if (rc != 0) {
2262                 sfc_adapter_unlock(sa);
2263                 SFC_ASSERT(rc > 0);
2264                 return -rc;
2265         }
2266
2267         switch_domain_id = sa->mae.switch_domain_id;
2268
2269         nb_repr = 0;
2270         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2271                                           sfc_count_representors_cb,
2272                                           &nb_repr);
2273         if (rc != 0) {
2274                 sfc_adapter_unlock(sa);
2275                 SFC_ASSERT(rc > 0);
2276                 return -rc;
2277         }
2278
2279         if (info == NULL) {
2280                 sfc_adapter_unlock(sa);
2281                 return nb_repr;
2282         }
2283
2284         rc = sfc_mae_switch_domain_controllers(switch_domain_id,
2285                                                &get_repr_ctx.controllers,
2286                                                &get_repr_ctx.nb_controllers);
2287         if (rc != 0) {
2288                 sfc_adapter_unlock(sa);
2289                 SFC_ASSERT(rc > 0);
2290                 return -rc;
2291         }
2292
2293         nic_cfg = efx_nic_cfg_get(sa->nic);
2294
2295         rc = sfc_mae_switch_domain_get_controller(switch_domain_id,
2296                                                   nic_cfg->enc_intf,
2297                                                   &controller);
2298         if (rc != 0) {
2299                 sfc_err(sa, "invalid controller: %d", nic_cfg->enc_intf);
2300                 controller = -1;
2301         }
2302
2303         info->controller = controller;
2304         info->pf = nic_cfg->enc_pf;
2305
2306         get_repr_ctx.info = info;
2307         get_repr_ctx.sa = sa;
2308         get_repr_ctx.switch_domain_id = switch_domain_id;
2309         rc = sfc_mae_switch_ports_iterate(switch_domain_id,
2310                                           sfc_get_representors_cb,
2311                                           &get_repr_ctx);
2312         if (rc != 0) {
2313                 sfc_adapter_unlock(sa);
2314                 SFC_ASSERT(rc > 0);
2315                 return -rc;
2316         }
2317
2318         sfc_adapter_unlock(sa);
2319         return nb_repr;
2320 }
2321
2322 static int
2323 sfc_rx_metadata_negotiate(struct rte_eth_dev *dev, uint64_t *features)
2324 {
2325         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2326         uint64_t supported = 0;
2327
2328         sfc_adapter_lock(sa);
2329
2330         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_FLAG) != 0)
2331                 supported |= RTE_ETH_RX_METADATA_USER_FLAG;
2332
2333         if ((sa->priv.dp_rx->features & SFC_DP_RX_FEAT_FLOW_MARK) != 0)
2334                 supported |= RTE_ETH_RX_METADATA_USER_MARK;
2335
2336         if (sfc_flow_tunnel_is_supported(sa))
2337                 supported |= RTE_ETH_RX_METADATA_TUNNEL_ID;
2338
2339         sa->negotiated_rx_metadata = supported & *features;
2340         *features = sa->negotiated_rx_metadata;
2341
2342         sfc_adapter_unlock(sa);
2343
2344         return 0;
2345 }
2346
2347 static const struct eth_dev_ops sfc_eth_dev_ops = {
2348         .dev_configure                  = sfc_dev_configure,
2349         .dev_start                      = sfc_dev_start,
2350         .dev_stop                       = sfc_dev_stop,
2351         .dev_set_link_up                = sfc_dev_set_link_up,
2352         .dev_set_link_down              = sfc_dev_set_link_down,
2353         .dev_close                      = sfc_dev_close,
2354         .promiscuous_enable             = sfc_dev_promisc_enable,
2355         .promiscuous_disable            = sfc_dev_promisc_disable,
2356         .allmulticast_enable            = sfc_dev_allmulti_enable,
2357         .allmulticast_disable           = sfc_dev_allmulti_disable,
2358         .link_update                    = sfc_dev_link_update,
2359         .stats_get                      = sfc_stats_get,
2360         .stats_reset                    = sfc_stats_reset,
2361         .xstats_get                     = sfc_xstats_get,
2362         .xstats_reset                   = sfc_stats_reset,
2363         .xstats_get_names               = sfc_xstats_get_names,
2364         .dev_infos_get                  = sfc_dev_infos_get,
2365         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2366         .mtu_set                        = sfc_dev_set_mtu,
2367         .rx_queue_start                 = sfc_rx_queue_start,
2368         .rx_queue_stop                  = sfc_rx_queue_stop,
2369         .tx_queue_start                 = sfc_tx_queue_start,
2370         .tx_queue_stop                  = sfc_tx_queue_stop,
2371         .rx_queue_setup                 = sfc_rx_queue_setup,
2372         .rx_queue_release               = sfc_rx_queue_release,
2373         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
2374         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
2375         .tx_queue_setup                 = sfc_tx_queue_setup,
2376         .tx_queue_release               = sfc_tx_queue_release,
2377         .flow_ctrl_get                  = sfc_flow_ctrl_get,
2378         .flow_ctrl_set                  = sfc_flow_ctrl_set,
2379         .mac_addr_set                   = sfc_mac_addr_set,
2380         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
2381         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
2382         .reta_update                    = sfc_dev_rss_reta_update,
2383         .reta_query                     = sfc_dev_rss_reta_query,
2384         .rss_hash_update                = sfc_dev_rss_hash_update,
2385         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2386         .flow_ops_get                   = sfc_dev_flow_ops_get,
2387         .set_mc_addr_list               = sfc_set_mc_addr_list,
2388         .rxq_info_get                   = sfc_rx_queue_info_get,
2389         .txq_info_get                   = sfc_tx_queue_info_get,
2390         .fw_version_get                 = sfc_fw_version_get,
2391         .xstats_get_by_id               = sfc_xstats_get_by_id,
2392         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
2393         .pool_ops_supported             = sfc_pool_ops_supported,
2394         .representor_info_get           = sfc_representor_info_get,
2395         .rx_metadata_negotiate          = sfc_rx_metadata_negotiate,
2396 };
2397
2398 struct sfc_ethdev_init_data {
2399         uint16_t                nb_representors;
2400 };
2401
2402 /**
2403  * Duplicate a string in potentially shared memory required for
2404  * multi-process support.
2405  *
2406  * strdup() allocates from process-local heap/memory.
2407  */
2408 static char *
2409 sfc_strdup(const char *str)
2410 {
2411         size_t size;
2412         char *copy;
2413
2414         if (str == NULL)
2415                 return NULL;
2416
2417         size = strlen(str) + 1;
2418         copy = rte_malloc(__func__, size, 0);
2419         if (copy != NULL)
2420                 rte_memcpy(copy, str, size);
2421
2422         return copy;
2423 }
2424
2425 static int
2426 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
2427 {
2428         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2429         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2430         const struct sfc_dp_rx *dp_rx;
2431         const struct sfc_dp_tx *dp_tx;
2432         const efx_nic_cfg_t *encp;
2433         unsigned int avail_caps = 0;
2434         const char *rx_name = NULL;
2435         const char *tx_name = NULL;
2436         int rc;
2437
2438         switch (sa->family) {
2439         case EFX_FAMILY_HUNTINGTON:
2440         case EFX_FAMILY_MEDFORD:
2441         case EFX_FAMILY_MEDFORD2:
2442                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2443                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2444                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2445                 break;
2446         case EFX_FAMILY_RIVERHEAD:
2447                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2448                 break;
2449         default:
2450                 break;
2451         }
2452
2453         encp = efx_nic_cfg_get(sa->nic);
2454         if (encp->enc_rx_es_super_buffer_supported)
2455                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2456
2457         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2458                                 sfc_kvarg_string_handler, &rx_name);
2459         if (rc != 0)
2460                 goto fail_kvarg_rx_datapath;
2461
2462         if (rx_name != NULL) {
2463                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2464                 if (dp_rx == NULL) {
2465                         sfc_err(sa, "Rx datapath %s not found", rx_name);
2466                         rc = ENOENT;
2467                         goto fail_dp_rx;
2468                 }
2469                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2470                         sfc_err(sa,
2471                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2472                                 rx_name);
2473                         rc = EINVAL;
2474                         goto fail_dp_rx_caps;
2475                 }
2476         } else {
2477                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2478                 if (dp_rx == NULL) {
2479                         sfc_err(sa, "Rx datapath by caps %#x not found",
2480                                 avail_caps);
2481                         rc = ENOENT;
2482                         goto fail_dp_rx;
2483                 }
2484         }
2485
2486         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2487         if (sas->dp_rx_name == NULL) {
2488                 rc = ENOMEM;
2489                 goto fail_dp_rx_name;
2490         }
2491
2492         if (strcmp(dp_rx->dp.name, SFC_KVARG_DATAPATH_EF10_ESSB) == 0) {
2493                 /* FLAG and MARK are always available from Rx prefix. */
2494                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_FLAG;
2495                 sa->negotiated_rx_metadata |= RTE_ETH_RX_METADATA_USER_MARK;
2496         }
2497
2498         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2499
2500         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2501                                 sfc_kvarg_string_handler, &tx_name);
2502         if (rc != 0)
2503                 goto fail_kvarg_tx_datapath;
2504
2505         if (tx_name != NULL) {
2506                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2507                 if (dp_tx == NULL) {
2508                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2509                         rc = ENOENT;
2510                         goto fail_dp_tx;
2511                 }
2512                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2513                         sfc_err(sa,
2514                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2515                                 tx_name);
2516                         rc = EINVAL;
2517                         goto fail_dp_tx_caps;
2518                 }
2519         } else {
2520                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2521                 if (dp_tx == NULL) {
2522                         sfc_err(sa, "Tx datapath by caps %#x not found",
2523                                 avail_caps);
2524                         rc = ENOENT;
2525                         goto fail_dp_tx;
2526                 }
2527         }
2528
2529         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2530         if (sas->dp_tx_name == NULL) {
2531                 rc = ENOMEM;
2532                 goto fail_dp_tx_name;
2533         }
2534
2535         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2536
2537         sa->priv.dp_rx = dp_rx;
2538         sa->priv.dp_tx = dp_tx;
2539
2540         dev->rx_pkt_burst = dp_rx->pkt_burst;
2541         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2542         dev->tx_pkt_burst = dp_tx->pkt_burst;
2543
2544         dev->rx_queue_count = sfc_rx_queue_count;
2545         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2546         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2547         dev->dev_ops = &sfc_eth_dev_ops;
2548
2549         return 0;
2550
2551 fail_dp_tx_name:
2552 fail_dp_tx_caps:
2553 fail_dp_tx:
2554 fail_kvarg_tx_datapath:
2555         rte_free(sas->dp_rx_name);
2556         sas->dp_rx_name = NULL;
2557
2558 fail_dp_rx_name:
2559 fail_dp_rx_caps:
2560 fail_dp_rx:
2561 fail_kvarg_rx_datapath:
2562         return rc;
2563 }
2564
2565 static void
2566 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2567 {
2568         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2569         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2570
2571         dev->dev_ops = NULL;
2572         dev->tx_pkt_prepare = NULL;
2573         dev->rx_pkt_burst = NULL;
2574         dev->tx_pkt_burst = NULL;
2575
2576         rte_free(sas->dp_tx_name);
2577         sas->dp_tx_name = NULL;
2578         sa->priv.dp_tx = NULL;
2579
2580         rte_free(sas->dp_rx_name);
2581         sas->dp_rx_name = NULL;
2582         sa->priv.dp_rx = NULL;
2583 }
2584
2585 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2586         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2587         .reta_query                     = sfc_dev_rss_reta_query,
2588         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2589         .rxq_info_get                   = sfc_rx_queue_info_get,
2590         .txq_info_get                   = sfc_tx_queue_info_get,
2591 };
2592
2593 static int
2594 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2595 {
2596         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2597         struct sfc_adapter_priv *sap;
2598         const struct sfc_dp_rx *dp_rx;
2599         const struct sfc_dp_tx *dp_tx;
2600         int rc;
2601
2602         /*
2603          * Allocate process private data from heap, since it should not
2604          * be located in shared memory allocated using rte_malloc() API.
2605          */
2606         sap = calloc(1, sizeof(*sap));
2607         if (sap == NULL) {
2608                 rc = ENOMEM;
2609                 goto fail_alloc_priv;
2610         }
2611
2612         sap->logtype_main = logtype_main;
2613
2614         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2615         if (dp_rx == NULL) {
2616                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2617                         "cannot find %s Rx datapath", sas->dp_rx_name);
2618                 rc = ENOENT;
2619                 goto fail_dp_rx;
2620         }
2621         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2622                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2623                         "%s Rx datapath does not support multi-process",
2624                         sas->dp_rx_name);
2625                 rc = EINVAL;
2626                 goto fail_dp_rx_multi_process;
2627         }
2628
2629         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2630         if (dp_tx == NULL) {
2631                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2632                         "cannot find %s Tx datapath", sas->dp_tx_name);
2633                 rc = ENOENT;
2634                 goto fail_dp_tx;
2635         }
2636         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2637                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2638                         "%s Tx datapath does not support multi-process",
2639                         sas->dp_tx_name);
2640                 rc = EINVAL;
2641                 goto fail_dp_tx_multi_process;
2642         }
2643
2644         sap->dp_rx = dp_rx;
2645         sap->dp_tx = dp_tx;
2646
2647         dev->process_private = sap;
2648         dev->rx_pkt_burst = dp_rx->pkt_burst;
2649         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2650         dev->tx_pkt_burst = dp_tx->pkt_burst;
2651         dev->rx_queue_count = sfc_rx_queue_count;
2652         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2653         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2654         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2655
2656         return 0;
2657
2658 fail_dp_tx_multi_process:
2659 fail_dp_tx:
2660 fail_dp_rx_multi_process:
2661 fail_dp_rx:
2662         free(sap);
2663
2664 fail_alloc_priv:
2665         return rc;
2666 }
2667
2668 static void
2669 sfc_register_dp(void)
2670 {
2671         /* Register once */
2672         if (TAILQ_EMPTY(&sfc_dp_head)) {
2673                 /* Prefer EF10 datapath */
2674                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2675                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2676                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2677                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2678
2679                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2680                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2681                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2682                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2683         }
2684 }
2685
2686 static int
2687 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
2688 {
2689         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2690         const char *switch_mode = NULL;
2691         int rc;
2692
2693         sfc_log_init(sa, "entry");
2694
2695         rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2696                                 sfc_kvarg_string_handler, &switch_mode);
2697         if (rc != 0)
2698                 goto fail_kvargs;
2699
2700         if (switch_mode == NULL) {
2701                 sa->switchdev = encp->enc_mae_supported &&
2702                                 (!encp->enc_datapath_cap_evb ||
2703                                  has_representors);
2704         } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2705                 sa->switchdev = false;
2706         } else if (strcasecmp(switch_mode,
2707                               SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2708                 sa->switchdev = true;
2709         } else {
2710                 sfc_err(sa, "invalid switch mode device argument '%s'",
2711                         switch_mode);
2712                 rc = EINVAL;
2713                 goto fail_mode;
2714         }
2715
2716         sfc_log_init(sa, "done");
2717
2718         return 0;
2719
2720 fail_mode:
2721 fail_kvargs:
2722         sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2723
2724         return rc;
2725 }
2726
2727 static int
2728 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
2729 {
2730         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2731         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2732         struct sfc_ethdev_init_data *init_data = init_params;
2733         uint32_t logtype_main;
2734         struct sfc_adapter *sa;
2735         int rc;
2736         const efx_nic_cfg_t *encp;
2737         const struct rte_ether_addr *from;
2738         int ret;
2739
2740         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2741                         SFC_EFX_DEV_CLASS_NET) {
2742                 SFC_GENERIC_LOG(DEBUG,
2743                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2744                 return 1;
2745         }
2746
2747         rc = sfc_dp_mport_register();
2748         if (rc != 0)
2749                 return rc;
2750
2751         sfc_register_dp();
2752
2753         logtype_main = sfc_register_logtype(&pci_dev->addr,
2754                                             SFC_LOGTYPE_MAIN_STR,
2755                                             RTE_LOG_NOTICE);
2756
2757         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2758                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2759
2760         /* Required for logging */
2761         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2762                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2763                         pci_dev->addr.domain, pci_dev->addr.bus,
2764                         pci_dev->addr.devid, pci_dev->addr.function,
2765                         dev->data->port_id);
2766         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2767                 SFC_GENERIC_LOG(ERR,
2768                         "reserved log prefix is too short for " PCI_PRI_FMT,
2769                         pci_dev->addr.domain, pci_dev->addr.bus,
2770                         pci_dev->addr.devid, pci_dev->addr.function);
2771                 return -EINVAL;
2772         }
2773         sas->pci_addr = pci_dev->addr;
2774         sas->port_id = dev->data->port_id;
2775
2776         /*
2777          * Allocate process private data from heap, since it should not
2778          * be located in shared memory allocated using rte_malloc() API.
2779          */
2780         sa = calloc(1, sizeof(*sa));
2781         if (sa == NULL) {
2782                 rc = ENOMEM;
2783                 goto fail_alloc_sa;
2784         }
2785
2786         dev->process_private = sa;
2787
2788         /* Required for logging */
2789         sa->priv.shared = sas;
2790         sa->priv.logtype_main = logtype_main;
2791
2792         sa->eth_dev = dev;
2793
2794         /* Copy PCI device info to the dev->data */
2795         rte_eth_copy_pci_info(dev, pci_dev);
2796         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2797
2798         rc = sfc_kvargs_parse(sa);
2799         if (rc != 0)
2800                 goto fail_kvargs_parse;
2801
2802         sfc_log_init(sa, "entry");
2803
2804         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2805         if (dev->data->mac_addrs == NULL) {
2806                 rc = ENOMEM;
2807                 goto fail_mac_addrs;
2808         }
2809
2810         sfc_adapter_lock_init(sa);
2811         sfc_adapter_lock(sa);
2812
2813         sfc_log_init(sa, "probing");
2814         rc = sfc_probe(sa);
2815         if (rc != 0)
2816                 goto fail_probe;
2817
2818         /*
2819          * Selecting a default switch mode requires the NIC to be probed and
2820          * to have its capabilities filled in.
2821          */
2822         rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
2823         if (rc != 0)
2824                 goto fail_switch_mode;
2825
2826         sfc_log_init(sa, "set device ops");
2827         rc = sfc_eth_dev_set_ops(dev);
2828         if (rc != 0)
2829                 goto fail_set_ops;
2830
2831         sfc_log_init(sa, "attaching");
2832         rc = sfc_attach(sa);
2833         if (rc != 0)
2834                 goto fail_attach;
2835
2836         if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_SUPPORTED) {
2837                 sfc_err(sa,
2838                         "failed to enable switchdev mode without MAE support");
2839                 rc = ENOTSUP;
2840                 goto fail_switchdev_no_mae;
2841         }
2842
2843         encp = efx_nic_cfg_get(sa->nic);
2844
2845         /*
2846          * The arguments are really reverse order in comparison to
2847          * Linux kernel. Copy from NIC config to Ethernet device data.
2848          */
2849         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2850         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2851
2852         sfc_adapter_unlock(sa);
2853
2854         sfc_log_init(sa, "done");
2855         return 0;
2856
2857 fail_switchdev_no_mae:
2858         sfc_detach(sa);
2859
2860 fail_attach:
2861         sfc_eth_dev_clear_ops(dev);
2862
2863 fail_set_ops:
2864 fail_switch_mode:
2865         sfc_unprobe(sa);
2866
2867 fail_probe:
2868         sfc_adapter_unlock(sa);
2869         sfc_adapter_lock_fini(sa);
2870         rte_free(dev->data->mac_addrs);
2871         dev->data->mac_addrs = NULL;
2872
2873 fail_mac_addrs:
2874         sfc_kvargs_cleanup(sa);
2875
2876 fail_kvargs_parse:
2877         sfc_log_init(sa, "failed %d", rc);
2878         dev->process_private = NULL;
2879         free(sa);
2880
2881 fail_alloc_sa:
2882         SFC_ASSERT(rc > 0);
2883         return -rc;
2884 }
2885
2886 static int
2887 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2888 {
2889         sfc_dev_close(dev);
2890
2891         return 0;
2892 }
2893
2894 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2895         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2896         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2897         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2898         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2899         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2900         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2901         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2902         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2903         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2904         { .vendor_id = 0 /* sentinel */ }
2905 };
2906
2907 static int
2908 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
2909 {
2910         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
2911         int rc;
2912
2913         if (args != NULL) {
2914                 rc = rte_eth_devargs_parse(args, &eth_da);
2915                 if (rc != 0) {
2916                         SFC_GENERIC_LOG(ERR,
2917                                         "Failed to parse generic devargs '%s'",
2918                                         args);
2919                         return rc;
2920                 }
2921         }
2922
2923         *devargs = eth_da;
2924
2925         return 0;
2926 }
2927
2928 static int
2929 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
2930                            struct sfc_ethdev_init_data *init_data,
2931                            struct rte_eth_dev **devp,
2932                            bool *dev_created)
2933 {
2934         struct rte_eth_dev *dev;
2935         bool created = false;
2936         int rc;
2937
2938         dev = rte_eth_dev_allocated(pci_dev->device.name);
2939         if (dev == NULL) {
2940                 rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
2941                                         sizeof(struct sfc_adapter_shared),
2942                                         eth_dev_pci_specific_init, pci_dev,
2943                                         sfc_eth_dev_init, init_data);
2944                 if (rc != 0) {
2945                         SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
2946                                         pci_dev->device.name);
2947                         return rc;
2948                 }
2949
2950                 created = true;
2951
2952                 dev = rte_eth_dev_allocated(pci_dev->device.name);
2953                 if (dev == NULL) {
2954                         SFC_GENERIC_LOG(ERR,
2955                                 "Failed to find allocated sfc ethdev '%s'",
2956                                 pci_dev->device.name);
2957                         return -ENODEV;
2958                 }
2959         }
2960
2961         *devp = dev;
2962         *dev_created = created;
2963
2964         return 0;
2965 }
2966
2967 static int
2968 sfc_eth_dev_create_repr(struct sfc_adapter *sa,
2969                         efx_pcie_interface_t controller,
2970                         uint16_t port,
2971                         uint16_t repr_port,
2972                         enum rte_eth_representor_type type)
2973 {
2974         struct sfc_repr_entity_info entity;
2975         efx_mport_sel_t mport_sel;
2976         int rc;
2977
2978         switch (type) {
2979         case RTE_ETH_REPRESENTOR_NONE:
2980                 return 0;
2981         case RTE_ETH_REPRESENTOR_VF:
2982         case RTE_ETH_REPRESENTOR_PF:
2983                 break;
2984         case RTE_ETH_REPRESENTOR_SF:
2985                 sfc_err(sa, "SF representors are not supported");
2986                 return ENOTSUP;
2987         default:
2988                 sfc_err(sa, "unknown representor type: %d", type);
2989                 return ENOTSUP;
2990         }
2991
2992         rc = efx_mae_mport_by_pcie_mh_function(controller,
2993                                                port,
2994                                                repr_port,
2995                                                &mport_sel);
2996         if (rc != 0) {
2997                 sfc_err(sa,
2998                         "failed to get m-port selector for controller %u port %u repr_port %u: %s",
2999                         controller, port, repr_port, rte_strerror(-rc));
3000                 return rc;
3001         }
3002
3003         memset(&entity, 0, sizeof(entity));
3004         entity.type = type;
3005         entity.intf = controller;
3006         entity.pf = port;
3007         entity.vf = repr_port;
3008
3009         rc = sfc_repr_create(sa->eth_dev, &entity, sa->mae.switch_domain_id,
3010                              &mport_sel);
3011         if (rc != 0) {
3012                 sfc_err(sa,
3013                         "failed to create representor for controller %u port %u repr_port %u: %s",
3014                         controller, port, repr_port, rte_strerror(-rc));
3015                 return rc;
3016         }
3017
3018         return 0;
3019 }
3020
3021 static int
3022 sfc_eth_dev_create_repr_port(struct sfc_adapter *sa,
3023                              const struct rte_eth_devargs *eth_da,
3024                              efx_pcie_interface_t controller,
3025                              uint16_t port)
3026 {
3027         int first_error = 0;
3028         uint16_t i;
3029         int rc;
3030
3031         if (eth_da->type == RTE_ETH_REPRESENTOR_PF) {
3032                 return sfc_eth_dev_create_repr(sa, controller, port,
3033                                                EFX_PCI_VF_INVALID,
3034                                                eth_da->type);
3035         }
3036
3037         for (i = 0; i < eth_da->nb_representor_ports; i++) {
3038                 rc = sfc_eth_dev_create_repr(sa, controller, port,
3039                                              eth_da->representor_ports[i],
3040                                              eth_da->type);
3041                 if (rc != 0 && first_error == 0)
3042                         first_error = rc;
3043         }
3044
3045         return first_error;
3046 }
3047
3048 static int
3049 sfc_eth_dev_create_repr_controller(struct sfc_adapter *sa,
3050                                    const struct rte_eth_devargs *eth_da,
3051                                    efx_pcie_interface_t controller)
3052 {
3053         const efx_nic_cfg_t *encp;
3054         int first_error = 0;
3055         uint16_t default_port;
3056         uint16_t i;
3057         int rc;
3058
3059         if (eth_da->nb_ports == 0) {
3060                 encp = efx_nic_cfg_get(sa->nic);
3061                 default_port = encp->enc_intf == controller ? encp->enc_pf : 0;
3062                 return sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3063                                                     default_port);
3064         }
3065
3066         for (i = 0; i < eth_da->nb_ports; i++) {
3067                 rc = sfc_eth_dev_create_repr_port(sa, eth_da, controller,
3068                                                   eth_da->ports[i]);
3069                 if (rc != 0 && first_error == 0)
3070                         first_error = rc;
3071         }
3072
3073         return first_error;
3074 }
3075
3076 static int
3077 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
3078                                 const struct rte_eth_devargs *eth_da)
3079 {
3080         efx_pcie_interface_t intf;
3081         const efx_nic_cfg_t *encp;
3082         struct sfc_adapter *sa;
3083         uint16_t switch_domain_id;
3084         uint16_t i;
3085         int rc;
3086
3087         sa = sfc_adapter_by_eth_dev(dev);
3088         switch_domain_id = sa->mae.switch_domain_id;
3089
3090         switch (eth_da->type) {
3091         case RTE_ETH_REPRESENTOR_NONE:
3092                 return 0;
3093         case RTE_ETH_REPRESENTOR_PF:
3094         case RTE_ETH_REPRESENTOR_VF:
3095                 break;
3096         case RTE_ETH_REPRESENTOR_SF:
3097                 sfc_err(sa, "SF representors are not supported");
3098                 return -ENOTSUP;
3099         default:
3100                 sfc_err(sa, "unknown representor type: %d",
3101                         eth_da->type);
3102                 return -ENOTSUP;
3103         }
3104
3105         if (!sa->switchdev) {
3106                 sfc_err(sa, "cannot create representors in non-switchdev mode");
3107                 return -EINVAL;
3108         }
3109
3110         if (!sfc_repr_available(sfc_sa2shared(sa))) {
3111                 sfc_err(sa, "cannot create representors: unsupported");
3112
3113                 return -ENOTSUP;
3114         }
3115
3116         /*
3117          * This is needed to construct the DPDK controller -> EFX interface
3118          * mapping.
3119          */
3120         sfc_adapter_lock(sa);
3121         rc = sfc_process_mport_journal(sa);
3122         sfc_adapter_unlock(sa);
3123         if (rc != 0) {
3124                 SFC_ASSERT(rc > 0);
3125                 return -rc;
3126         }
3127
3128         if (eth_da->nb_mh_controllers > 0) {
3129                 for (i = 0; i < eth_da->nb_mh_controllers; i++) {
3130                         rc = sfc_mae_switch_domain_get_intf(switch_domain_id,
3131                                                 eth_da->mh_controllers[i],
3132                                                 &intf);
3133                         if (rc != 0) {
3134                                 sfc_err(sa, "failed to get representor");
3135                                 continue;
3136                         }
3137                         sfc_eth_dev_create_repr_controller(sa, eth_da, intf);
3138                 }
3139         } else {
3140                 encp = efx_nic_cfg_get(sa->nic);
3141                 sfc_eth_dev_create_repr_controller(sa, eth_da, encp->enc_intf);
3142         }
3143
3144         return 0;
3145 }
3146
3147 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
3148         struct rte_pci_device *pci_dev)
3149 {
3150         struct sfc_ethdev_init_data init_data;
3151         struct rte_eth_devargs eth_da;
3152         struct rte_eth_dev *dev;
3153         bool dev_created;
3154         int rc;
3155
3156         if (pci_dev->device.devargs != NULL) {
3157                 rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
3158                                            &eth_da);
3159                 if (rc != 0)
3160                         return rc;
3161         } else {
3162                 memset(&eth_da, 0, sizeof(eth_da));
3163         }
3164
3165         /* If no VF representors specified, check for PF ones */
3166         if (eth_da.nb_representor_ports > 0)
3167                 init_data.nb_representors = eth_da.nb_representor_ports;
3168         else
3169                 init_data.nb_representors = eth_da.nb_ports;
3170
3171         if (init_data.nb_representors > 0 &&
3172             rte_eal_process_type() != RTE_PROC_PRIMARY) {
3173                 SFC_GENERIC_LOG(ERR,
3174                         "Create representors from secondary process not supported, dev '%s'",
3175                         pci_dev->device.name);
3176                 return -ENOTSUP;
3177         }
3178
3179         /*
3180          * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
3181          * if it does not already exist. Re-probing an existing device is
3182          * expected to allow additional representors to be configured.
3183          */
3184         rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
3185                                         &dev_created);
3186         if (rc != 0)
3187                 return rc;
3188
3189         rc = sfc_eth_dev_create_representors(dev, &eth_da);
3190         if (rc != 0) {
3191                 if (dev_created)
3192                         (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
3193
3194                 return rc;
3195         }
3196
3197         return 0;
3198 }
3199
3200 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
3201 {
3202         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
3203 }
3204
3205 static struct rte_pci_driver sfc_efx_pmd = {
3206         .id_table = pci_id_sfc_efx_map,
3207         .drv_flags =
3208                 RTE_PCI_DRV_INTR_LSC |
3209                 RTE_PCI_DRV_NEED_MAPPING |
3210                 RTE_PCI_DRV_PROBE_AGAIN,
3211         .probe = sfc_eth_dev_pci_probe,
3212         .remove = sfc_eth_dev_pci_remove,
3213 };
3214
3215 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
3216 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
3217 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
3218 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
3219         SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
3220         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
3221         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
3222         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
3223         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
3224         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
3225         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
3226
3227 RTE_INIT(sfc_driver_register_logtype)
3228 {
3229         int ret;
3230
3231         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
3232                                                    RTE_LOG_NOTICE);
3233         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
3234 }