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