drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
   4  * All rights reserved.
   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 <rte_ethdev.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16
  17 #include "efx.h"
  18
  19 #include "sfc.h"
  20 #include "sfc_debug.h"
  21 #include "sfc_log.h"
  22 #include "sfc_kvargs.h"
  23 #include "sfc_ev.h"
  24 #include "sfc_rx.h"
  25 #include "sfc_tx.h"
  26 #include "sfc_flow.h"
  27 #include "sfc_dp.h"
  28 #include "sfc_dp_rx.h"
  29
  30 static struct sfc_dp_list sfc_dp_head =
  31         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  32
  33 static int
  34 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  35 {
  36         struct sfc_adapter *sa = dev->data->dev_private;
  37         efx_nic_fw_info_t enfi;
  38         int ret;
  39         int rc;
  40
  41         /*
  42          * Return value of the callback is likely supposed to be
  43          * equal to or greater than 0, nevertheless, if an error
  44          * occurs, it will be desirable to pass it to the caller
  45          */
  46         if ((fw_version == NULL) || (fw_size == 0))
  47                 return -EINVAL;
  48
  49         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  50         if (rc != 0)
  51                 return -rc;
  52
  53         ret = snprintf(fw_version, fw_size,
  54                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  55                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  56                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  57         if (ret < 0)
  58                 return ret;
  59
  60         if (enfi.enfi_dpcpu_fw_ids_valid) {
  61                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  62                 int ret_extra;
  63
  64                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  65                                      fw_size - dpcpu_fw_ids_offset,
  66                                      " rx%" PRIx16 " tx%" PRIx16,
  67                                      enfi.enfi_rx_dpcpu_fw_id,
  68                                      enfi.enfi_tx_dpcpu_fw_id);
  69                 if (ret_extra < 0)
  70                         return ret_extra;
  71
  72                 ret += ret_extra;
  73         }
  74
  75         if (fw_size < (size_t)(++ret))
  76                 return ret;
  77         else
  78                 return 0;
  79 }
  80
  81 static void
  82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  83 {
  84         struct sfc_adapter *sa = dev->data->dev_private;
  85         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
  86
  87         sfc_log_init(sa, "entry");
  88
  89         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
  90         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  91
  92         /* Autonegotiation may be disabled */
  93         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  94         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  95                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  96         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
  97                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
  98         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
  99                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 100
 101         dev_info->max_rx_queues = sa->rxq_max;
 102         dev_info->max_tx_queues = sa->txq_max;
 103
 104         /* By default packets are dropped if no descriptors are available */
 105         dev_info->default_rxconf.rx_drop_en = 1;
 106
 107         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa);
 108
 109         dev_info->tx_offload_capa =
 110                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 111                 DEV_TX_OFFLOAD_UDP_CKSUM |
 112                 DEV_TX_OFFLOAD_TCP_CKSUM;
 113
 114         if (encp->enc_tunnel_encapsulations_supported != 0)
 115                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
 116
 117         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 118         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 119             !encp->enc_hw_tx_insert_vlan_enabled)
 120                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 121         else
 122                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 123
 124         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 125                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 126
 127         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 128                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 129
 130         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 131                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 132
 133 #if EFSYS_OPT_RX_SCALE
 134         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 135                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 136                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 137                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 138         }
 139 #endif
 140
 141         if (sa->tso)
 142                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 143
 144         /* Initialize to hardware limits */
 145         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 146         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 147         /* The RXQ hardware requires that the descriptor count is a power
 148          * of 2, but rx_desc_lim cannot properly describe that constraint.
 149          */
 150         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 151
 152         /* Initialize to hardware limits */
 153         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 154         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 155         /*
 156          * The TXQ hardware requires that the descriptor count is a power
 157          * of 2, but tx_desc_lim cannot properly describe that constraint
 158          */
 159         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 160
 161         if (sa->dp_rx->get_dev_info != NULL)
 162                 sa->dp_rx->get_dev_info(dev_info);
 163         if (sa->dp_tx->get_dev_info != NULL)
 164                 sa->dp_tx->get_dev_info(dev_info);
 165 }
 166
 167 static const uint32_t *
 168 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 169 {
 170         struct sfc_adapter *sa = dev->data->dev_private;
 171         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 172         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 173
 174         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 175 }
 176
 177 static int
 178 sfc_dev_configure(struct rte_eth_dev *dev)
 179 {
 180         struct rte_eth_dev_data *dev_data = dev->data;
 181         struct sfc_adapter *sa = dev_data->dev_private;
 182         int rc;
 183
 184         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 185                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 186
 187         sfc_adapter_lock(sa);
 188         switch (sa->state) {
 189         case SFC_ADAPTER_CONFIGURED:
 190                 /* FALLTHROUGH */
 191         case SFC_ADAPTER_INITIALIZED:
 192                 rc = sfc_configure(sa);
 193                 break;
 194         default:
 195                 sfc_err(sa, "unexpected adapter state %u to configure",
 196                         sa->state);
 197                 rc = EINVAL;
 198                 break;
 199         }
 200         sfc_adapter_unlock(sa);
 201
 202         sfc_log_init(sa, "done %d", rc);
 203         SFC_ASSERT(rc >= 0);
 204         return -rc;
 205 }
 206
 207 static int
 208 sfc_dev_start(struct rte_eth_dev *dev)
 209 {
 210         struct sfc_adapter *sa = dev->data->dev_private;
 211         int rc;
 212
 213         sfc_log_init(sa, "entry");
 214
 215         sfc_adapter_lock(sa);
 216         rc = sfc_start(sa);
 217         sfc_adapter_unlock(sa);
 218
 219         sfc_log_init(sa, "done %d", rc);
 220         SFC_ASSERT(rc >= 0);
 221         return -rc;
 222 }
 223
 224 static int
 225 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 226 {
 227         struct sfc_adapter *sa = dev->data->dev_private;
 228         struct rte_eth_link *dev_link = &dev->data->dev_link;
 229         struct rte_eth_link old_link;
 230         struct rte_eth_link current_link;
 231
 232         sfc_log_init(sa, "entry");
 233
 234 retry:
 235         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 236         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 237
 238         if (sa->state != SFC_ADAPTER_STARTED) {
 239                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 240                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 241                                          *(uint64_t *)&old_link,
 242                                          *(uint64_t *)&current_link))
 243                         goto retry;
 244         } else if (wait_to_complete) {
 245                 efx_link_mode_t link_mode;
 246
 247                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 248                         link_mode = EFX_LINK_UNKNOWN;
 249                 sfc_port_link_mode_to_info(link_mode, &current_link);
 250
 251                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 252                                          *(uint64_t *)&old_link,
 253                                          *(uint64_t *)&current_link))
 254                         goto retry;
 255         } else {
 256                 sfc_ev_mgmt_qpoll(sa);
 257                 *(int64_t *)&current_link =
 258                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 259         }
 260
 261         if (old_link.link_status != current_link.link_status)
 262                 sfc_info(sa, "Link status is %s",
 263                          current_link.link_status ? "UP" : "DOWN");
 264
 265         return old_link.link_status == current_link.link_status ? 0 : -1;
 266 }
 267
 268 static void
 269 sfc_dev_stop(struct rte_eth_dev *dev)
 270 {
 271         struct sfc_adapter *sa = dev->data->dev_private;
 272
 273         sfc_log_init(sa, "entry");
 274
 275         sfc_adapter_lock(sa);
 276         sfc_stop(sa);
 277         sfc_adapter_unlock(sa);
 278
 279         sfc_log_init(sa, "done");
 280 }
 281
 282 static int
 283 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 284 {
 285         struct sfc_adapter *sa = dev->data->dev_private;
 286         int rc;
 287
 288         sfc_log_init(sa, "entry");
 289
 290         sfc_adapter_lock(sa);
 291         rc = sfc_start(sa);
 292         sfc_adapter_unlock(sa);
 293
 294         SFC_ASSERT(rc >= 0);
 295         return -rc;
 296 }
 297
 298 static int
 299 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 300 {
 301         struct sfc_adapter *sa = dev->data->dev_private;
 302
 303         sfc_log_init(sa, "entry");
 304
 305         sfc_adapter_lock(sa);
 306         sfc_stop(sa);
 307         sfc_adapter_unlock(sa);
 308
 309         return 0;
 310 }
 311
 312 static void
 313 sfc_dev_close(struct rte_eth_dev *dev)
 314 {
 315         struct sfc_adapter *sa = dev->data->dev_private;
 316
 317         sfc_log_init(sa, "entry");
 318
 319         sfc_adapter_lock(sa);
 320         switch (sa->state) {
 321         case SFC_ADAPTER_STARTED:
 322                 sfc_stop(sa);
 323                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 324                 /* FALLTHROUGH */
 325         case SFC_ADAPTER_CONFIGURED:
 326                 sfc_close(sa);
 327                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 328                 /* FALLTHROUGH */
 329         case SFC_ADAPTER_INITIALIZED:
 330                 break;
 331         default:
 332                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 333                 break;
 334         }
 335         sfc_adapter_unlock(sa);
 336
 337         sfc_log_init(sa, "done");
 338 }
 339
 340 static void
 341 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 342                    boolean_t enabled)
 343 {
 344         struct sfc_port *port;
 345         boolean_t *toggle;
 346         struct sfc_adapter *sa = dev->data->dev_private;
 347         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 348         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 349
 350         sfc_adapter_lock(sa);
 351
 352         port = &sa->port;
 353         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 354
 355         if (*toggle != enabled) {
 356                 *toggle = enabled;
 357
 358                 if (port->isolated) {
 359                         sfc_warn(sa, "isolated mode is active on the port");
 360                         sfc_warn(sa, "the change is to be applied on the next "
 361                                      "start provided that isolated mode is "
 362                                      "disabled prior the next start");
 363                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 364                            (sfc_set_rx_mode(sa) != 0)) {
 365                         *toggle = !(enabled);
 366                         sfc_warn(sa, "Failed to %s %s mode",
 367                                  ((enabled) ? "enable" : "disable"), desc);
 368                 }
 369         }
 370
 371         sfc_adapter_unlock(sa);
 372 }
 373
 374 static void
 375 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 376 {
 377         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 378 }
 379
 380 static void
 381 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 382 {
 383         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 384 }
 385
 386 static void
 387 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 388 {
 389         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 390 }
 391
 392 static void
 393 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 394 {
 395         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 396 }
 397
 398 static int
 399 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 400                    uint16_t nb_rx_desc, unsigned int socket_id,
 401                    const struct rte_eth_rxconf *rx_conf,
 402                    struct rte_mempool *mb_pool)
 403 {
 404         struct sfc_adapter *sa = dev->data->dev_private;
 405         int rc;
 406
 407         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 408                      rx_queue_id, nb_rx_desc, socket_id);
 409
 410         sfc_adapter_lock(sa);
 411
 412         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 413                           rx_conf, mb_pool);
 414         if (rc != 0)
 415                 goto fail_rx_qinit;
 416
 417         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 418
 419         sfc_adapter_unlock(sa);
 420
 421         return 0;
 422
 423 fail_rx_qinit:
 424         sfc_adapter_unlock(sa);
 425         SFC_ASSERT(rc > 0);
 426         return -rc;
 427 }
 428
 429 static void
 430 sfc_rx_queue_release(void *queue)
 431 {
 432         struct sfc_dp_rxq *dp_rxq = queue;
 433         struct sfc_rxq *rxq;
 434         struct sfc_adapter *sa;
 435         unsigned int sw_index;
 436
 437         if (dp_rxq == NULL)
 438                 return;
 439
 440         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 441         sa = rxq->evq->sa;
 442         sfc_adapter_lock(sa);
 443
 444         sw_index = sfc_rxq_sw_index(rxq);
 445
 446         sfc_log_init(sa, "RxQ=%u", sw_index);
 447
 448         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 449
 450         sfc_rx_qfini(sa, sw_index);
 451
 452         sfc_adapter_unlock(sa);
 453 }
 454
 455 static int
 456 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 457                    uint16_t nb_tx_desc, unsigned int socket_id,
 458                    const struct rte_eth_txconf *tx_conf)
 459 {
 460         struct sfc_adapter *sa = dev->data->dev_private;
 461         int rc;
 462
 463         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 464                      tx_queue_id, nb_tx_desc, socket_id);
 465
 466         sfc_adapter_lock(sa);
 467
 468         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 469         if (rc != 0)
 470                 goto fail_tx_qinit;
 471
 472         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 473
 474         sfc_adapter_unlock(sa);
 475         return 0;
 476
 477 fail_tx_qinit:
 478         sfc_adapter_unlock(sa);
 479         SFC_ASSERT(rc > 0);
 480         return -rc;
 481 }
 482
 483 static void
 484 sfc_tx_queue_release(void *queue)
 485 {
 486         struct sfc_dp_txq *dp_txq = queue;
 487         struct sfc_txq *txq;
 488         unsigned int sw_index;
 489         struct sfc_adapter *sa;
 490
 491         if (dp_txq == NULL)
 492                 return;
 493
 494         txq = sfc_txq_by_dp_txq(dp_txq);
 495         sw_index = sfc_txq_sw_index(txq);
 496
 497         SFC_ASSERT(txq->evq != NULL);
 498         sa = txq->evq->sa;
 499
 500         sfc_log_init(sa, "TxQ = %u", sw_index);
 501
 502         sfc_adapter_lock(sa);
 503
 504         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 505         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 506
 507         sfc_tx_qfini(sa, sw_index);
 508
 509         sfc_adapter_unlock(sa);
 510 }
 511
 512 static int
 513 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 514 {
 515         struct sfc_adapter *sa = dev->data->dev_private;
 516         struct sfc_port *port = &sa->port;
 517         uint64_t *mac_stats;
 518         int ret;
 519
 520         rte_spinlock_lock(&port->mac_stats_lock);
 521
 522         ret = sfc_port_update_mac_stats(sa);
 523         if (ret != 0)
 524                 goto unlock;
 525
 526         mac_stats = port->mac_stats_buf;
 527
 528         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 529                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 530                 stats->ipackets =
 531                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 532                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 533                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 534                 stats->opackets =
 535                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 536                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 537                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 538                 stats->ibytes =
 539                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 540                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 541                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 542                 stats->obytes =
 543                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 544                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 545                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 546                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 547                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 548                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 549         } else {
 550                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 551                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 552                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 553                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 554                 /*
 555                  * Take into account stats which are whenever supported
 556                  * on EF10. If some stat is not supported by current
 557                  * firmware variant or HW revision, it is guaranteed
 558                  * to be zero in mac_stats.
 559                  */
 560                 stats->imissed =
 561                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 562                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 563                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 564                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 565                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 566                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 567                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 568                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 569                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 570                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 571                 stats->ierrors =
 572                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 573                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 574                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 575                 /* no oerrors counters supported on EF10 */
 576         }
 577
 578 unlock:
 579         rte_spinlock_unlock(&port->mac_stats_lock);
 580         SFC_ASSERT(ret >= 0);
 581         return -ret;
 582 }
 583
 584 static void
 585 sfc_stats_reset(struct rte_eth_dev *dev)
 586 {
 587         struct sfc_adapter *sa = dev->data->dev_private;
 588         struct sfc_port *port = &sa->port;
 589         int rc;
 590
 591         if (sa->state != SFC_ADAPTER_STARTED) {
 592                 /*
 593                  * The operation cannot be done if port is not started; it
 594                  * will be scheduled to be done during the next port start
 595                  */
 596                 port->mac_stats_reset_pending = B_TRUE;
 597                 return;
 598         }
 599
 600         rc = sfc_port_reset_mac_stats(sa);
 601         if (rc != 0)
 602                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 603 }
 604
 605 static int
 606 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 607                unsigned int xstats_count)
 608 {
 609         struct sfc_adapter *sa = dev->data->dev_private;
 610         struct sfc_port *port = &sa->port;
 611         uint64_t *mac_stats;
 612         int rc;
 613         unsigned int i;
 614         int nstats = 0;
 615
 616         rte_spinlock_lock(&port->mac_stats_lock);
 617
 618         rc = sfc_port_update_mac_stats(sa);
 619         if (rc != 0) {
 620                 SFC_ASSERT(rc > 0);
 621                 nstats = -rc;
 622                 goto unlock;
 623         }
 624
 625         mac_stats = port->mac_stats_buf;
 626
 627         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 628                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 629                         if (xstats != NULL && nstats < (int)xstats_count) {
 630                                 xstats[nstats].id = nstats;
 631                                 xstats[nstats].value = mac_stats[i];
 632                         }
 633                         nstats++;
 634                 }
 635         }
 636
 637 unlock:
 638         rte_spinlock_unlock(&port->mac_stats_lock);
 639
 640         return nstats;
 641 }
 642
 643 static int
 644 sfc_xstats_get_names(struct rte_eth_dev *dev,
 645                      struct rte_eth_xstat_name *xstats_names,
 646                      unsigned int xstats_count)
 647 {
 648         struct sfc_adapter *sa = dev->data->dev_private;
 649         struct sfc_port *port = &sa->port;
 650         unsigned int i;
 651         unsigned int nstats = 0;
 652
 653         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 654                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 655                         if (xstats_names != NULL && nstats < xstats_count)
 656                                 strncpy(xstats_names[nstats].name,
 657                                         efx_mac_stat_name(sa->nic, i),
 658                                         sizeof(xstats_names[0].name));
 659                         nstats++;
 660                 }
 661         }
 662
 663         return nstats;
 664 }
 665
 666 static int
 667 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 668                      uint64_t *values, unsigned int n)
 669 {
 670         struct sfc_adapter *sa = dev->data->dev_private;
 671         struct sfc_port *port = &sa->port;
 672         uint64_t *mac_stats;
 673         unsigned int nb_supported = 0;
 674         unsigned int nb_written = 0;
 675         unsigned int i;
 676         int ret;
 677         int rc;
 678
 679         if (unlikely(values == NULL) ||
 680             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 681                 return port->mac_stats_nb_supported;
 682
 683         rte_spinlock_lock(&port->mac_stats_lock);
 684
 685         rc = sfc_port_update_mac_stats(sa);
 686         if (rc != 0) {
 687                 SFC_ASSERT(rc > 0);
 688                 ret = -rc;
 689                 goto unlock;
 690         }
 691
 692         mac_stats = port->mac_stats_buf;
 693
 694         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 695                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 696                         continue;
 697
 698                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 699                         values[nb_written++] = mac_stats[i];
 700
 701                 ++nb_supported;
 702         }
 703
 704         ret = nb_written;
 705
 706 unlock:
 707         rte_spinlock_unlock(&port->mac_stats_lock);
 708
 709         return ret;
 710 }
 711
 712 static int
 713 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 714                            struct rte_eth_xstat_name *xstats_names,
 715                            const uint64_t *ids, unsigned int size)
 716 {
 717         struct sfc_adapter *sa = dev->data->dev_private;
 718         struct sfc_port *port = &sa->port;
 719         unsigned int nb_supported = 0;
 720         unsigned int nb_written = 0;
 721         unsigned int i;
 722
 723         if (unlikely(xstats_names == NULL) ||
 724             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 725                 return port->mac_stats_nb_supported;
 726
 727         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 728                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 729                         continue;
 730
 731                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 732                         char *name = xstats_names[nb_written++].name;
 733
 734                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 735                                 sizeof(xstats_names[0].name));
 736                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 737                 }
 738
 739                 ++nb_supported;
 740         }
 741
 742         return nb_written;
 743 }
 744
 745 static int
 746 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 747 {
 748         struct sfc_adapter *sa = dev->data->dev_private;
 749         unsigned int wanted_fc, link_fc;
 750
 751         memset(fc_conf, 0, sizeof(*fc_conf));
 752
 753         sfc_adapter_lock(sa);
 754
 755         if (sa->state == SFC_ADAPTER_STARTED)
 756                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 757         else
 758                 link_fc = sa->port.flow_ctrl;
 759
 760         switch (link_fc) {
 761         case 0:
 762                 fc_conf->mode = RTE_FC_NONE;
 763                 break;
 764         case EFX_FCNTL_RESPOND:
 765                 fc_conf->mode = RTE_FC_RX_PAUSE;
 766                 break;
 767         case EFX_FCNTL_GENERATE:
 768                 fc_conf->mode = RTE_FC_TX_PAUSE;
 769                 break;
 770         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 771                 fc_conf->mode = RTE_FC_FULL;
 772                 break;
 773         default:
 774                 sfc_err(sa, "%s: unexpected flow control value %#x",
 775                         __func__, link_fc);
 776         }
 777
 778         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 779
 780         sfc_adapter_unlock(sa);
 781
 782         return 0;
 783 }
 784
 785 static int
 786 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 787 {
 788         struct sfc_adapter *sa = dev->data->dev_private;
 789         struct sfc_port *port = &sa->port;
 790         unsigned int fcntl;
 791         int rc;
 792
 793         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 794             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 795             fc_conf->mac_ctrl_frame_fwd != 0) {
 796                 sfc_err(sa, "unsupported flow control settings specified");
 797                 rc = EINVAL;
 798                 goto fail_inval;
 799         }
 800
 801         switch (fc_conf->mode) {
 802         case RTE_FC_NONE:
 803                 fcntl = 0;
 804                 break;
 805         case RTE_FC_RX_PAUSE:
 806                 fcntl = EFX_FCNTL_RESPOND;
 807                 break;
 808         case RTE_FC_TX_PAUSE:
 809                 fcntl = EFX_FCNTL_GENERATE;
 810                 break;
 811         case RTE_FC_FULL:
 812                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 813                 break;
 814         default:
 815                 rc = EINVAL;
 816                 goto fail_inval;
 817         }
 818
 819         sfc_adapter_lock(sa);
 820
 821         if (sa->state == SFC_ADAPTER_STARTED) {
 822                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 823                 if (rc != 0)
 824                         goto fail_mac_fcntl_set;
 825         }
 826
 827         port->flow_ctrl = fcntl;
 828         port->flow_ctrl_autoneg = fc_conf->autoneg;
 829
 830         sfc_adapter_unlock(sa);
 831
 832         return 0;
 833
 834 fail_mac_fcntl_set:
 835         sfc_adapter_unlock(sa);
 836 fail_inval:
 837         SFC_ASSERT(rc > 0);
 838         return -rc;
 839 }
 840
 841 static int
 842 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 843 {
 844         struct sfc_adapter *sa = dev->data->dev_private;
 845         size_t pdu = EFX_MAC_PDU(mtu);
 846         size_t old_pdu;
 847         int rc;
 848
 849         sfc_log_init(sa, "mtu=%u", mtu);
 850
 851         rc = EINVAL;
 852         if (pdu < EFX_MAC_PDU_MIN) {
 853                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 854                         (unsigned int)mtu, (unsigned int)pdu,
 855                         EFX_MAC_PDU_MIN);
 856                 goto fail_inval;
 857         }
 858         if (pdu > EFX_MAC_PDU_MAX) {
 859                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 860                         (unsigned int)mtu, (unsigned int)pdu,
 861                         EFX_MAC_PDU_MAX);
 862                 goto fail_inval;
 863         }
 864
 865         sfc_adapter_lock(sa);
 866
 867         if (pdu != sa->port.pdu) {
 868                 if (sa->state == SFC_ADAPTER_STARTED) {
 869                         sfc_stop(sa);
 870
 871                         old_pdu = sa->port.pdu;
 872                         sa->port.pdu = pdu;
 873                         rc = sfc_start(sa);
 874                         if (rc != 0)
 875                                 goto fail_start;
 876                 } else {
 877                         sa->port.pdu = pdu;
 878                 }
 879         }
 880
 881         /*
 882          * The driver does not use it, but other PMDs update jumbo_frame
 883          * flag and max_rx_pkt_len when MTU is set.
 884          */
 885         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 886         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 887
 888         sfc_adapter_unlock(sa);
 889
 890         sfc_log_init(sa, "done");
 891         return 0;
 892
 893 fail_start:
 894         sa->port.pdu = old_pdu;
 895         if (sfc_start(sa) != 0)
 896                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 897                         "PDU max size - port is stopped",
 898                         (unsigned int)pdu, (unsigned int)old_pdu);
 899         sfc_adapter_unlock(sa);
 900
 901 fail_inval:
 902         sfc_log_init(sa, "failed %d", rc);
 903         SFC_ASSERT(rc > 0);
 904         return -rc;
 905 }
 906 static void
 907 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 908 {
 909         struct sfc_adapter *sa = dev->data->dev_private;
 910         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 911         struct sfc_port *port = &sa->port;
 912         int rc;
 913
 914         sfc_adapter_lock(sa);
 915
 916         /*
 917          * Copy the address to the device private data so that
 918          * it could be recalled in the case of adapter restart.
 919          */
 920         ether_addr_copy(mac_addr, &port->default_mac_addr);
 921
 922         if (port->isolated) {
 923                 sfc_err(sa, "isolated mode is active on the port");
 924                 sfc_err(sa, "will not set MAC address");
 925                 goto unlock;
 926         }
 927
 928         if (sa->state != SFC_ADAPTER_STARTED) {
 929                 sfc_info(sa, "the port is not started");
 930                 sfc_info(sa, "the new MAC address will be set on port start");
 931
 932                 goto unlock;
 933         }
 934
 935         if (encp->enc_allow_set_mac_with_installed_filters) {
 936                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 937                 if (rc != 0) {
 938                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 939                         goto unlock;
 940                 }
 941
 942                 /*
 943                  * Changing the MAC address by means of MCDI request
 944                  * has no effect on received traffic, therefore
 945                  * we also need to update unicast filters
 946                  */
 947                 rc = sfc_set_rx_mode(sa);
 948                 if (rc != 0)
 949                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 950         } else {
 951                 sfc_warn(sa, "cannot set MAC address with filters installed");
 952                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 953                 sfc_warn(sa, "(some traffic may be dropped)");
 954
 955                 /*
 956                  * Since setting MAC address with filters installed is not
 957                  * allowed on the adapter, the new MAC address will be set
 958                  * by means of adapter restart. sfc_start() shall retrieve
 959                  * the new address from the device private data and set it.
 960                  */
 961                 sfc_stop(sa);
 962                 rc = sfc_start(sa);
 963                 if (rc != 0)
 964                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 965         }
 966
 967 unlock:
 968         /*
 969          * In the case of failure sa->port->default_mac_addr does not
 970          * need rollback since no error code is returned, and the upper
 971          * API will anyway update the external MAC address storage.
 972          * To be consistent with that new value it is better to keep
 973          * the device private value the same.
 974          */
 975         sfc_adapter_unlock(sa);
 976 }
 977
 978
 979 static int
 980 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 981                      uint32_t nb_mc_addr)
 982 {
 983         struct sfc_adapter *sa = dev->data->dev_private;
 984         struct sfc_port *port = &sa->port;
 985         uint8_t *mc_addrs = port->mcast_addrs;
 986         int rc;
 987         unsigned int i;
 988
 989         if (port->isolated) {
 990                 sfc_err(sa, "isolated mode is active on the port");
 991                 sfc_err(sa, "will not set multicast address list");
 992                 return -ENOTSUP;
 993         }
 994
 995         if (mc_addrs == NULL)
 996                 return -ENOBUFS;
 997
 998         if (nb_mc_addr > port->max_mcast_addrs) {
 999                 sfc_err(sa, "too many multicast addresses: %u > %u",
1000                          nb_mc_addr, port->max_mcast_addrs);
1001                 return -EINVAL;
1002         }
1003
1004         for (i = 0; i < nb_mc_addr; ++i) {
1005                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1006                                  EFX_MAC_ADDR_LEN);
1007                 mc_addrs += EFX_MAC_ADDR_LEN;
1008         }
1009
1010         port->nb_mcast_addrs = nb_mc_addr;
1011
1012         if (sa->state != SFC_ADAPTER_STARTED)
1013                 return 0;
1014
1015         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1016                                         port->nb_mcast_addrs);
1017         if (rc != 0)
1018                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1019
1020         SFC_ASSERT(rc > 0);
1021         return -rc;
1022 }
1023
1024 /*
1025  * The function is used by the secondary process as well. It must not
1026  * use any process-local pointers from the adapter data.
1027  */
1028 static void
1029 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1030                       struct rte_eth_rxq_info *qinfo)
1031 {
1032         struct sfc_adapter *sa = dev->data->dev_private;
1033         struct sfc_rxq_info *rxq_info;
1034         struct sfc_rxq *rxq;
1035
1036         sfc_adapter_lock(sa);
1037
1038         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1039
1040         rxq_info = &sa->rxq_info[rx_queue_id];
1041         rxq = rxq_info->rxq;
1042         SFC_ASSERT(rxq != NULL);
1043
1044         qinfo->mp = rxq->refill_mb_pool;
1045         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1046         qinfo->conf.rx_drop_en = 1;
1047         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1048         qinfo->scattered_rx =
1049                 ((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1050         qinfo->nb_desc = rxq_info->entries;
1051
1052         sfc_adapter_unlock(sa);
1053 }
1054
1055 /*
1056  * The function is used by the secondary process as well. It must not
1057  * use any process-local pointers from the adapter data.
1058  */
1059 static void
1060 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1061                       struct rte_eth_txq_info *qinfo)
1062 {
1063         struct sfc_adapter *sa = dev->data->dev_private;
1064         struct sfc_txq_info *txq_info;
1065
1066         sfc_adapter_lock(sa);
1067
1068         SFC_ASSERT(tx_queue_id < sa->txq_count);
1069
1070         txq_info = &sa->txq_info[tx_queue_id];
1071         SFC_ASSERT(txq_info->txq != NULL);
1072
1073         memset(qinfo, 0, sizeof(*qinfo));
1074
1075         qinfo->conf.txq_flags = txq_info->txq->flags;
1076         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1077         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1078         qinfo->nb_desc = txq_info->entries;
1079
1080         sfc_adapter_unlock(sa);
1081 }
1082
1083 static uint32_t
1084 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1085 {
1086         struct sfc_adapter *sa = dev->data->dev_private;
1087
1088         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1089
1090         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1091 }
1092
1093 static int
1094 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1095 {
1096         struct sfc_dp_rxq *dp_rxq = queue;
1097
1098         return sfc_rx_qdesc_done(dp_rxq, offset);
1099 }
1100
1101 static int
1102 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1103 {
1104         struct sfc_dp_rxq *dp_rxq = queue;
1105         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1106
1107         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1108 }
1109
1110 static int
1111 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1112 {
1113         struct sfc_dp_txq *dp_txq = queue;
1114         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1115
1116         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1117 }
1118
1119 static int
1120 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1121 {
1122         struct sfc_adapter *sa = dev->data->dev_private;
1123         int rc;
1124
1125         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1126
1127         sfc_adapter_lock(sa);
1128
1129         rc = EINVAL;
1130         if (sa->state != SFC_ADAPTER_STARTED)
1131                 goto fail_not_started;
1132
1133         rc = sfc_rx_qstart(sa, rx_queue_id);
1134         if (rc != 0)
1135                 goto fail_rx_qstart;
1136
1137         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1138
1139         sfc_adapter_unlock(sa);
1140
1141         return 0;
1142
1143 fail_rx_qstart:
1144 fail_not_started:
1145         sfc_adapter_unlock(sa);
1146         SFC_ASSERT(rc > 0);
1147         return -rc;
1148 }
1149
1150 static int
1151 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1152 {
1153         struct sfc_adapter *sa = dev->data->dev_private;
1154
1155         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1156
1157         sfc_adapter_lock(sa);
1158         sfc_rx_qstop(sa, rx_queue_id);
1159
1160         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1161
1162         sfc_adapter_unlock(sa);
1163
1164         return 0;
1165 }
1166
1167 static int
1168 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1169 {
1170         struct sfc_adapter *sa = dev->data->dev_private;
1171         int rc;
1172
1173         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1174
1175         sfc_adapter_lock(sa);
1176
1177         rc = EINVAL;
1178         if (sa->state != SFC_ADAPTER_STARTED)
1179                 goto fail_not_started;
1180
1181         rc = sfc_tx_qstart(sa, tx_queue_id);
1182         if (rc != 0)
1183                 goto fail_tx_qstart;
1184
1185         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1186
1187         sfc_adapter_unlock(sa);
1188         return 0;
1189
1190 fail_tx_qstart:
1191
1192 fail_not_started:
1193         sfc_adapter_unlock(sa);
1194         SFC_ASSERT(rc > 0);
1195         return -rc;
1196 }
1197
1198 static int
1199 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1200 {
1201         struct sfc_adapter *sa = dev->data->dev_private;
1202
1203         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1204
1205         sfc_adapter_lock(sa);
1206
1207         sfc_tx_qstop(sa, tx_queue_id);
1208
1209         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1210
1211         sfc_adapter_unlock(sa);
1212         return 0;
1213 }
1214
1215 static efx_tunnel_protocol_t
1216 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1217 {
1218         switch (rte_type) {
1219         case RTE_TUNNEL_TYPE_VXLAN:
1220                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1221         case RTE_TUNNEL_TYPE_GENEVE:
1222                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1223         default:
1224                 return EFX_TUNNEL_NPROTOS;
1225         }
1226 }
1227
1228 enum sfc_udp_tunnel_op_e {
1229         SFC_UDP_TUNNEL_ADD_PORT,
1230         SFC_UDP_TUNNEL_DEL_PORT,
1231 };
1232
1233 static int
1234 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1235                       struct rte_eth_udp_tunnel *tunnel_udp,
1236                       enum sfc_udp_tunnel_op_e op)
1237 {
1238         struct sfc_adapter *sa = dev->data->dev_private;
1239         efx_tunnel_protocol_t tunnel_proto;
1240         int rc;
1241
1242         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1243                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1244                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1245                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1246
1247         tunnel_proto =
1248                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1249         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1250                 rc = ENOTSUP;
1251                 goto fail_bad_proto;
1252         }
1253
1254         sfc_adapter_lock(sa);
1255
1256         switch (op) {
1257         case SFC_UDP_TUNNEL_ADD_PORT:
1258                 rc = efx_tunnel_config_udp_add(sa->nic,
1259                                                tunnel_udp->udp_port,
1260                                                tunnel_proto);
1261                 break;
1262         case SFC_UDP_TUNNEL_DEL_PORT:
1263                 rc = efx_tunnel_config_udp_remove(sa->nic,
1264                                                   tunnel_udp->udp_port,
1265                                                   tunnel_proto);
1266                 break;
1267         default:
1268                 rc = EINVAL;
1269                 goto fail_bad_op;
1270         }
1271
1272         if (rc != 0)
1273                 goto fail_op;
1274
1275         if (sa->state == SFC_ADAPTER_STARTED) {
1276                 rc = efx_tunnel_reconfigure(sa->nic);
1277                 if (rc == EAGAIN) {
1278                         /*
1279                          * Configuration is accepted by FW and MC reboot
1280                          * is initiated to apply the changes. MC reboot
1281                          * will be handled in a usual way (MC reboot
1282                          * event on management event queue and adapter
1283                          * restart).
1284                          */
1285                         rc = 0;
1286                 } else if (rc != 0) {
1287                         goto fail_reconfigure;
1288                 }
1289         }
1290
1291         sfc_adapter_unlock(sa);
1292         return 0;
1293
1294 fail_reconfigure:
1295         /* Remove/restore entry since the change makes the trouble */
1296         switch (op) {
1297         case SFC_UDP_TUNNEL_ADD_PORT:
1298                 (void)efx_tunnel_config_udp_remove(sa->nic,
1299                                                    tunnel_udp->udp_port,
1300                                                    tunnel_proto);
1301                 break;
1302         case SFC_UDP_TUNNEL_DEL_PORT:
1303                 (void)efx_tunnel_config_udp_add(sa->nic,
1304                                                 tunnel_udp->udp_port,
1305                                                 tunnel_proto);
1306                 break;
1307         }
1308
1309 fail_op:
1310 fail_bad_op:
1311         sfc_adapter_unlock(sa);
1312
1313 fail_bad_proto:
1314         SFC_ASSERT(rc > 0);
1315         return -rc;
1316 }
1317
1318 static int
1319 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1320                             struct rte_eth_udp_tunnel *tunnel_udp)
1321 {
1322         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1323 }
1324
1325 static int
1326 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1327                             struct rte_eth_udp_tunnel *tunnel_udp)
1328 {
1329         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1330 }
1331
1332 #if EFSYS_OPT_RX_SCALE
1333 static int
1334 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1335                           struct rte_eth_rss_conf *rss_conf)
1336 {
1337         struct sfc_adapter *sa = dev->data->dev_private;
1338         struct sfc_port *port = &sa->port;
1339
1340         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1341                 return -ENOTSUP;
1342
1343         if (sa->rss_channels == 0)
1344                 return -EINVAL;
1345
1346         sfc_adapter_lock(sa);
1347
1348         /*
1349          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1350          * hence, conversion is done here to derive a correct set of ETH_RSS
1351          * flags which corresponds to the active EFX configuration stored
1352          * locally in 'sfc_adapter' and kept up-to-date
1353          */
1354         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1355         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1356         if (rss_conf->rss_key != NULL)
1357                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1358
1359         sfc_adapter_unlock(sa);
1360
1361         return 0;
1362 }
1363
1364 static int
1365 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1366                         struct rte_eth_rss_conf *rss_conf)
1367 {
1368         struct sfc_adapter *sa = dev->data->dev_private;
1369         struct sfc_port *port = &sa->port;
1370         unsigned int efx_hash_types;
1371         int rc = 0;
1372
1373         if (port->isolated)
1374                 return -ENOTSUP;
1375
1376         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1377                 sfc_err(sa, "RSS is not available");
1378                 return -ENOTSUP;
1379         }
1380
1381         if (sa->rss_channels == 0) {
1382                 sfc_err(sa, "RSS is not configured");
1383                 return -EINVAL;
1384         }
1385
1386         if ((rss_conf->rss_key != NULL) &&
1387             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1388                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1389                         sizeof(sa->rss_key));
1390                 return -EINVAL;
1391         }
1392
1393         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1394                 sfc_err(sa, "unsupported hash functions requested");
1395                 return -EINVAL;
1396         }
1397
1398         sfc_adapter_lock(sa);
1399
1400         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1401
1402         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1403                                    EFX_RX_HASHALG_TOEPLITZ,
1404                                    efx_hash_types, B_TRUE);
1405         if (rc != 0)
1406                 goto fail_scale_mode_set;
1407
1408         if (rss_conf->rss_key != NULL) {
1409                 if (sa->state == SFC_ADAPTER_STARTED) {
1410                         rc = efx_rx_scale_key_set(sa->nic,
1411                                                   EFX_RSS_CONTEXT_DEFAULT,
1412                                                   rss_conf->rss_key,
1413                                                   sizeof(sa->rss_key));
1414                         if (rc != 0)
1415                                 goto fail_scale_key_set;
1416                 }
1417
1418                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1419         }
1420
1421         sa->rss_hash_types = efx_hash_types;
1422
1423         sfc_adapter_unlock(sa);
1424
1425         return 0;
1426
1427 fail_scale_key_set:
1428         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1429                                   EFX_RX_HASHALG_TOEPLITZ,
1430                                   sa->rss_hash_types, B_TRUE) != 0)
1431                 sfc_err(sa, "failed to restore RSS mode");
1432
1433 fail_scale_mode_set:
1434         sfc_adapter_unlock(sa);
1435         return -rc;
1436 }
1437
1438 static int
1439 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1440                        struct rte_eth_rss_reta_entry64 *reta_conf,
1441                        uint16_t reta_size)
1442 {
1443         struct sfc_adapter *sa = dev->data->dev_private;
1444         struct sfc_port *port = &sa->port;
1445         int entry;
1446
1447         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1448                 return -ENOTSUP;
1449
1450         if (sa->rss_channels == 0)
1451                 return -EINVAL;
1452
1453         if (reta_size != EFX_RSS_TBL_SIZE)
1454                 return -EINVAL;
1455
1456         sfc_adapter_lock(sa);
1457
1458         for (entry = 0; entry < reta_size; entry++) {
1459                 int grp = entry / RTE_RETA_GROUP_SIZE;
1460                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1461
1462                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1463                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1464         }
1465
1466         sfc_adapter_unlock(sa);
1467
1468         return 0;
1469 }
1470
1471 static int
1472 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1473                         struct rte_eth_rss_reta_entry64 *reta_conf,
1474                         uint16_t reta_size)
1475 {
1476         struct sfc_adapter *sa = dev->data->dev_private;
1477         struct sfc_port *port = &sa->port;
1478         unsigned int *rss_tbl_new;
1479         uint16_t entry;
1480         int rc = 0;
1481
1482
1483         if (port->isolated)
1484                 return -ENOTSUP;
1485
1486         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1487                 sfc_err(sa, "RSS is not available");
1488                 return -ENOTSUP;
1489         }
1490
1491         if (sa->rss_channels == 0) {
1492                 sfc_err(sa, "RSS is not configured");
1493                 return -EINVAL;
1494         }
1495
1496         if (reta_size != EFX_RSS_TBL_SIZE) {
1497                 sfc_err(sa, "RETA size is wrong (should be %u)",
1498                         EFX_RSS_TBL_SIZE);
1499                 return -EINVAL;
1500         }
1501
1502         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1503         if (rss_tbl_new == NULL)
1504                 return -ENOMEM;
1505
1506         sfc_adapter_lock(sa);
1507
1508         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1509
1510         for (entry = 0; entry < reta_size; entry++) {
1511                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1512                 struct rte_eth_rss_reta_entry64 *grp;
1513
1514                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1515
1516                 if (grp->mask & (1ull << grp_idx)) {
1517                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1518                                 rc = EINVAL;
1519                                 goto bad_reta_entry;
1520                         }
1521                         rss_tbl_new[entry] = grp->reta[grp_idx];
1522                 }
1523         }
1524
1525         if (sa->state == SFC_ADAPTER_STARTED) {
1526                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1527                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1528                 if (rc != 0)
1529                         goto fail_scale_tbl_set;
1530         }
1531
1532         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1533
1534 fail_scale_tbl_set:
1535 bad_reta_entry:
1536         sfc_adapter_unlock(sa);
1537
1538         rte_free(rss_tbl_new);
1539
1540         SFC_ASSERT(rc >= 0);
1541         return -rc;
1542 }
1543 #endif
1544
1545 static int
1546 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1547                     enum rte_filter_op filter_op,
1548                     void *arg)
1549 {
1550         struct sfc_adapter *sa = dev->data->dev_private;
1551         int rc = ENOTSUP;
1552
1553         sfc_log_init(sa, "entry");
1554
1555         switch (filter_type) {
1556         case RTE_ETH_FILTER_NONE:
1557                 sfc_err(sa, "Global filters configuration not supported");
1558                 break;
1559         case RTE_ETH_FILTER_MACVLAN:
1560                 sfc_err(sa, "MACVLAN filters not supported");
1561                 break;
1562         case RTE_ETH_FILTER_ETHERTYPE:
1563                 sfc_err(sa, "EtherType filters not supported");
1564                 break;
1565         case RTE_ETH_FILTER_FLEXIBLE:
1566                 sfc_err(sa, "Flexible filters not supported");
1567                 break;
1568         case RTE_ETH_FILTER_SYN:
1569                 sfc_err(sa, "SYN filters not supported");
1570                 break;
1571         case RTE_ETH_FILTER_NTUPLE:
1572                 sfc_err(sa, "NTUPLE filters not supported");
1573                 break;
1574         case RTE_ETH_FILTER_TUNNEL:
1575                 sfc_err(sa, "Tunnel filters not supported");
1576                 break;
1577         case RTE_ETH_FILTER_FDIR:
1578                 sfc_err(sa, "Flow Director filters not supported");
1579                 break;
1580         case RTE_ETH_FILTER_HASH:
1581                 sfc_err(sa, "Hash filters not supported");
1582                 break;
1583         case RTE_ETH_FILTER_GENERIC:
1584                 if (filter_op != RTE_ETH_FILTER_GET) {
1585                         rc = EINVAL;
1586                 } else {
1587                         *(const void **)arg = &sfc_flow_ops;
1588                         rc = 0;
1589                 }
1590                 break;
1591         default:
1592                 sfc_err(sa, "Unknown filter type %u", filter_type);
1593                 break;
1594         }
1595
1596         sfc_log_init(sa, "exit: %d", -rc);
1597         SFC_ASSERT(rc >= 0);
1598         return -rc;
1599 }
1600
1601 static const struct eth_dev_ops sfc_eth_dev_ops = {
1602         .dev_configure                  = sfc_dev_configure,
1603         .dev_start                      = sfc_dev_start,
1604         .dev_stop                       = sfc_dev_stop,
1605         .dev_set_link_up                = sfc_dev_set_link_up,
1606         .dev_set_link_down              = sfc_dev_set_link_down,
1607         .dev_close                      = sfc_dev_close,
1608         .promiscuous_enable             = sfc_dev_promisc_enable,
1609         .promiscuous_disable            = sfc_dev_promisc_disable,
1610         .allmulticast_enable            = sfc_dev_allmulti_enable,
1611         .allmulticast_disable           = sfc_dev_allmulti_disable,
1612         .link_update                    = sfc_dev_link_update,
1613         .stats_get                      = sfc_stats_get,
1614         .stats_reset                    = sfc_stats_reset,
1615         .xstats_get                     = sfc_xstats_get,
1616         .xstats_reset                   = sfc_stats_reset,
1617         .xstats_get_names               = sfc_xstats_get_names,
1618         .dev_infos_get                  = sfc_dev_infos_get,
1619         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1620         .mtu_set                        = sfc_dev_set_mtu,
1621         .rx_queue_start                 = sfc_rx_queue_start,
1622         .rx_queue_stop                  = sfc_rx_queue_stop,
1623         .tx_queue_start                 = sfc_tx_queue_start,
1624         .tx_queue_stop                  = sfc_tx_queue_stop,
1625         .rx_queue_setup                 = sfc_rx_queue_setup,
1626         .rx_queue_release               = sfc_rx_queue_release,
1627         .rx_queue_count                 = sfc_rx_queue_count,
1628         .rx_descriptor_done             = sfc_rx_descriptor_done,
1629         .rx_descriptor_status           = sfc_rx_descriptor_status,
1630         .tx_descriptor_status           = sfc_tx_descriptor_status,
1631         .tx_queue_setup                 = sfc_tx_queue_setup,
1632         .tx_queue_release               = sfc_tx_queue_release,
1633         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1634         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1635         .mac_addr_set                   = sfc_mac_addr_set,
1636         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1637         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1638 #if EFSYS_OPT_RX_SCALE
1639         .reta_update                    = sfc_dev_rss_reta_update,
1640         .reta_query                     = sfc_dev_rss_reta_query,
1641         .rss_hash_update                = sfc_dev_rss_hash_update,
1642         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1643 #endif
1644         .filter_ctrl                    = sfc_dev_filter_ctrl,
1645         .set_mc_addr_list               = sfc_set_mc_addr_list,
1646         .rxq_info_get                   = sfc_rx_queue_info_get,
1647         .txq_info_get                   = sfc_tx_queue_info_get,
1648         .fw_version_get                 = sfc_fw_version_get,
1649         .xstats_get_by_id               = sfc_xstats_get_by_id,
1650         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1651 };
1652
1653 /**
1654  * Duplicate a string in potentially shared memory required for
1655  * multi-process support.
1656  *
1657  * strdup() allocates from process-local heap/memory.
1658  */
1659 static char *
1660 sfc_strdup(const char *str)
1661 {
1662         size_t size;
1663         char *copy;
1664
1665         if (str == NULL)
1666                 return NULL;
1667
1668         size = strlen(str) + 1;
1669         copy = rte_malloc(__func__, size, 0);
1670         if (copy != NULL)
1671                 rte_memcpy(copy, str, size);
1672
1673         return copy;
1674 }
1675
1676 static int
1677 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1678 {
1679         struct sfc_adapter *sa = dev->data->dev_private;
1680         unsigned int avail_caps = 0;
1681         const char *rx_name = NULL;
1682         const char *tx_name = NULL;
1683         int rc;
1684
1685         switch (sa->family) {
1686         case EFX_FAMILY_HUNTINGTON:
1687         case EFX_FAMILY_MEDFORD:
1688                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1689                 break;
1690         default:
1691                 break;
1692         }
1693
1694         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1695                                 sfc_kvarg_string_handler, &rx_name);
1696         if (rc != 0)
1697                 goto fail_kvarg_rx_datapath;
1698
1699         if (rx_name != NULL) {
1700                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1701                 if (sa->dp_rx == NULL) {
1702                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1703                         rc = ENOENT;
1704                         goto fail_dp_rx;
1705                 }
1706                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1707                         sfc_err(sa,
1708                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1709                                 rx_name);
1710                         rc = EINVAL;
1711                         goto fail_dp_rx_caps;
1712                 }
1713         } else {
1714                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1715                 if (sa->dp_rx == NULL) {
1716                         sfc_err(sa, "Rx datapath by caps %#x not found",
1717                                 avail_caps);
1718                         rc = ENOENT;
1719                         goto fail_dp_rx;
1720                 }
1721         }
1722
1723         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1724         if (sa->dp_rx_name == NULL) {
1725                 rc = ENOMEM;
1726                 goto fail_dp_rx_name;
1727         }
1728
1729         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1730
1731         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1732
1733         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1734                                 sfc_kvarg_string_handler, &tx_name);
1735         if (rc != 0)
1736                 goto fail_kvarg_tx_datapath;
1737
1738         if (tx_name != NULL) {
1739                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1740                 if (sa->dp_tx == NULL) {
1741                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1742                         rc = ENOENT;
1743                         goto fail_dp_tx;
1744                 }
1745                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1746                         sfc_err(sa,
1747                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1748                                 tx_name);
1749                         rc = EINVAL;
1750                         goto fail_dp_tx_caps;
1751                 }
1752         } else {
1753                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1754                 if (sa->dp_tx == NULL) {
1755                         sfc_err(sa, "Tx datapath by caps %#x not found",
1756                                 avail_caps);
1757                         rc = ENOENT;
1758                         goto fail_dp_tx;
1759                 }
1760         }
1761
1762         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1763         if (sa->dp_tx_name == NULL) {
1764                 rc = ENOMEM;
1765                 goto fail_dp_tx_name;
1766         }
1767
1768         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1769
1770         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1771
1772         dev->dev_ops = &sfc_eth_dev_ops;
1773
1774         return 0;
1775
1776 fail_dp_tx_name:
1777 fail_dp_tx_caps:
1778         sa->dp_tx = NULL;
1779
1780 fail_dp_tx:
1781 fail_kvarg_tx_datapath:
1782         rte_free(sa->dp_rx_name);
1783         sa->dp_rx_name = NULL;
1784
1785 fail_dp_rx_name:
1786 fail_dp_rx_caps:
1787         sa->dp_rx = NULL;
1788
1789 fail_dp_rx:
1790 fail_kvarg_rx_datapath:
1791         return rc;
1792 }
1793
1794 static void
1795 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1796 {
1797         struct sfc_adapter *sa = dev->data->dev_private;
1798
1799         dev->dev_ops = NULL;
1800         dev->rx_pkt_burst = NULL;
1801         dev->tx_pkt_burst = NULL;
1802
1803         rte_free(sa->dp_tx_name);
1804         sa->dp_tx_name = NULL;
1805         sa->dp_tx = NULL;
1806
1807         rte_free(sa->dp_rx_name);
1808         sa->dp_rx_name = NULL;
1809         sa->dp_rx = NULL;
1810 }
1811
1812 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1813         .rxq_info_get                   = sfc_rx_queue_info_get,
1814         .txq_info_get                   = sfc_tx_queue_info_get,
1815 };
1816
1817 static int
1818 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1819 {
1820         /*
1821          * Device private data has really many process-local pointers.
1822          * Below code should be extremely careful to use data located
1823          * in shared memory only.
1824          */
1825         struct sfc_adapter *sa = dev->data->dev_private;
1826         const struct sfc_dp_rx *dp_rx;
1827         const struct sfc_dp_tx *dp_tx;
1828         int rc;
1829
1830         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1831         if (dp_rx == NULL) {
1832                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1833                 rc = ENOENT;
1834                 goto fail_dp_rx;
1835         }
1836         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1837                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1838                         sa->dp_tx_name);
1839                 rc = EINVAL;
1840                 goto fail_dp_rx_multi_process;
1841         }
1842
1843         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1844         if (dp_tx == NULL) {
1845                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1846                 rc = ENOENT;
1847                 goto fail_dp_tx;
1848         }
1849         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1850                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1851                         sa->dp_tx_name);
1852                 rc = EINVAL;
1853                 goto fail_dp_tx_multi_process;
1854         }
1855
1856         dev->rx_pkt_burst = dp_rx->pkt_burst;
1857         dev->tx_pkt_burst = dp_tx->pkt_burst;
1858         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1859
1860         return 0;
1861
1862 fail_dp_tx_multi_process:
1863 fail_dp_tx:
1864 fail_dp_rx_multi_process:
1865 fail_dp_rx:
1866         return rc;
1867 }
1868
1869 static void
1870 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1871 {
1872         dev->dev_ops = NULL;
1873         dev->tx_pkt_burst = NULL;
1874         dev->rx_pkt_burst = NULL;
1875 }
1876
1877 static void
1878 sfc_register_dp(void)
1879 {
1880         /* Register once */
1881         if (TAILQ_EMPTY(&sfc_dp_head)) {
1882                 /* Prefer EF10 datapath */
1883                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1884                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1885
1886                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1887                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1888                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1889         }
1890 }
1891
1892 static int
1893 sfc_eth_dev_init(struct rte_eth_dev *dev)
1894 {
1895         struct sfc_adapter *sa = dev->data->dev_private;
1896         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1897         int rc;
1898         const efx_nic_cfg_t *encp;
1899         const struct ether_addr *from;
1900
1901         sfc_register_dp();
1902
1903         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1904                 return -sfc_eth_dev_secondary_set_ops(dev);
1905
1906         /* Required for logging */
1907         sa->pci_addr = pci_dev->addr;
1908         sa->port_id = dev->data->port_id;
1909
1910         sa->eth_dev = dev;
1911
1912         /* Copy PCI device info to the dev->data */
1913         rte_eth_copy_pci_info(dev, pci_dev);
1914
1915         rc = sfc_kvargs_parse(sa);
1916         if (rc != 0)
1917                 goto fail_kvargs_parse;
1918
1919         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1920                                 sfc_kvarg_bool_handler, &sa->debug_init);
1921         if (rc != 0)
1922                 goto fail_kvarg_debug_init;
1923
1924         sfc_log_init(sa, "entry");
1925
1926         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1927         if (dev->data->mac_addrs == NULL) {
1928                 rc = ENOMEM;
1929                 goto fail_mac_addrs;
1930         }
1931
1932         sfc_adapter_lock_init(sa);
1933         sfc_adapter_lock(sa);
1934
1935         sfc_log_init(sa, "probing");
1936         rc = sfc_probe(sa);
1937         if (rc != 0)
1938                 goto fail_probe;
1939
1940         sfc_log_init(sa, "set device ops");
1941         rc = sfc_eth_dev_set_ops(dev);
1942         if (rc != 0)
1943                 goto fail_set_ops;
1944
1945         sfc_log_init(sa, "attaching");
1946         rc = sfc_attach(sa);
1947         if (rc != 0)
1948                 goto fail_attach;
1949
1950         encp = efx_nic_cfg_get(sa->nic);
1951
1952         /*
1953          * The arguments are really reverse order in comparison to
1954          * Linux kernel. Copy from NIC config to Ethernet device data.
1955          */
1956         from = (const struct ether_addr *)(encp->enc_mac_addr);
1957         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1958
1959         sfc_adapter_unlock(sa);
1960
1961         sfc_log_init(sa, "done");
1962         return 0;
1963
1964 fail_attach:
1965         sfc_eth_dev_clear_ops(dev);
1966
1967 fail_set_ops:
1968         sfc_unprobe(sa);
1969
1970 fail_probe:
1971         sfc_adapter_unlock(sa);
1972         sfc_adapter_lock_fini(sa);
1973         rte_free(dev->data->mac_addrs);
1974         dev->data->mac_addrs = NULL;
1975
1976 fail_mac_addrs:
1977 fail_kvarg_debug_init:
1978         sfc_kvargs_cleanup(sa);
1979
1980 fail_kvargs_parse:
1981         sfc_log_init(sa, "failed %d", rc);
1982         SFC_ASSERT(rc > 0);
1983         return -rc;
1984 }
1985
1986 static int
1987 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1988 {
1989         struct sfc_adapter *sa;
1990
1991         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1992                 sfc_eth_dev_secondary_clear_ops(dev);
1993                 return 0;
1994         }
1995
1996         sa = dev->data->dev_private;
1997         sfc_log_init(sa, "entry");
1998
1999         sfc_adapter_lock(sa);
2000
2001         sfc_eth_dev_clear_ops(dev);
2002
2003         sfc_detach(sa);
2004         sfc_unprobe(sa);
2005
2006         rte_free(dev->data->mac_addrs);
2007         dev->data->mac_addrs = NULL;
2008
2009         sfc_kvargs_cleanup(sa);
2010
2011         sfc_adapter_unlock(sa);
2012         sfc_adapter_lock_fini(sa);
2013
2014         sfc_log_init(sa, "done");
2015
2016         /* Required for logging, so cleanup last */
2017         sa->eth_dev = NULL;
2018         return 0;
2019 }
2020
2021 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2022         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2023         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2024         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2025         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2026         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2027         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2028         { .vendor_id = 0 /* sentinel */ }
2029 };
2030
2031 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2032         struct rte_pci_device *pci_dev)
2033 {
2034         return rte_eth_dev_pci_generic_probe(pci_dev,
2035                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2036 }
2037
2038 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2039 {
2040         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2041 }
2042
2043 static struct rte_pci_driver sfc_efx_pmd = {
2044         .id_table = pci_id_sfc_efx_map,
2045         .drv_flags =
2046                 RTE_PCI_DRV_INTR_LSC |
2047                 RTE_PCI_DRV_NEED_MAPPING,
2048         .probe = sfc_eth_dev_pci_probe,
2049         .remove = sfc_eth_dev_pci_remove,
2050 };
2051
2052 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2053 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2054 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2055 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2056         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2057         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2058         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2059         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2060         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2061         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);