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_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 108
 109         /*
 110          * rx_offload_capa includes both device and queue offloads since
 111          * the latter may be requested on a per device basis which makes
 112          * sense when some offloads are needed to be set on all queues.
 113          */
 114         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 115                                     dev_info->rx_queue_offload_capa;
 116
 117         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa);
 118
 119         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 120         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 121             !encp->enc_hw_tx_insert_vlan_enabled)
 122                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 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         /* Initialize to hardware limits */
 142         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 143         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 144         /* The RXQ hardware requires that the descriptor count is a power
 145          * of 2, but rx_desc_lim cannot properly describe that constraint.
 146          */
 147         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 148
 149         /* Initialize to hardware limits */
 150         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 151         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 152         /*
 153          * The TXQ hardware requires that the descriptor count is a power
 154          * of 2, but tx_desc_lim cannot properly describe that constraint
 155          */
 156         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 157
 158         if (sa->dp_rx->get_dev_info != NULL)
 159                 sa->dp_rx->get_dev_info(dev_info);
 160         if (sa->dp_tx->get_dev_info != NULL)
 161                 sa->dp_tx->get_dev_info(dev_info);
 162 }
 163
 164 static const uint32_t *
 165 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 166 {
 167         struct sfc_adapter *sa = dev->data->dev_private;
 168         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 169         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 170
 171         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 172 }
 173
 174 static int
 175 sfc_dev_configure(struct rte_eth_dev *dev)
 176 {
 177         struct rte_eth_dev_data *dev_data = dev->data;
 178         struct sfc_adapter *sa = dev_data->dev_private;
 179         int rc;
 180
 181         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 182                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 183
 184         sfc_adapter_lock(sa);
 185         switch (sa->state) {
 186         case SFC_ADAPTER_CONFIGURED:
 187                 /* FALLTHROUGH */
 188         case SFC_ADAPTER_INITIALIZED:
 189                 rc = sfc_configure(sa);
 190                 break;
 191         default:
 192                 sfc_err(sa, "unexpected adapter state %u to configure",
 193                         sa->state);
 194                 rc = EINVAL;
 195                 break;
 196         }
 197         sfc_adapter_unlock(sa);
 198
 199         sfc_log_init(sa, "done %d", rc);
 200         SFC_ASSERT(rc >= 0);
 201         return -rc;
 202 }
 203
 204 static int
 205 sfc_dev_start(struct rte_eth_dev *dev)
 206 {
 207         struct sfc_adapter *sa = dev->data->dev_private;
 208         int rc;
 209
 210         sfc_log_init(sa, "entry");
 211
 212         sfc_adapter_lock(sa);
 213         rc = sfc_start(sa);
 214         sfc_adapter_unlock(sa);
 215
 216         sfc_log_init(sa, "done %d", rc);
 217         SFC_ASSERT(rc >= 0);
 218         return -rc;
 219 }
 220
 221 static int
 222 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 223 {
 224         struct sfc_adapter *sa = dev->data->dev_private;
 225         struct rte_eth_link *dev_link = &dev->data->dev_link;
 226         struct rte_eth_link old_link;
 227         struct rte_eth_link current_link;
 228
 229         sfc_log_init(sa, "entry");
 230
 231 retry:
 232         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 233         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 234
 235         if (sa->state != SFC_ADAPTER_STARTED) {
 236                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 237                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 238                                          *(uint64_t *)&old_link,
 239                                          *(uint64_t *)&current_link))
 240                         goto retry;
 241         } else if (wait_to_complete) {
 242                 efx_link_mode_t link_mode;
 243
 244                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 245                         link_mode = EFX_LINK_UNKNOWN;
 246                 sfc_port_link_mode_to_info(link_mode, &current_link);
 247
 248                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 249                                          *(uint64_t *)&old_link,
 250                                          *(uint64_t *)&current_link))
 251                         goto retry;
 252         } else {
 253                 sfc_ev_mgmt_qpoll(sa);
 254                 *(int64_t *)&current_link =
 255                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 256         }
 257
 258         if (old_link.link_status != current_link.link_status)
 259                 sfc_info(sa, "Link status is %s",
 260                          current_link.link_status ? "UP" : "DOWN");
 261
 262         return old_link.link_status == current_link.link_status ? 0 : -1;
 263 }
 264
 265 static void
 266 sfc_dev_stop(struct rte_eth_dev *dev)
 267 {
 268         struct sfc_adapter *sa = dev->data->dev_private;
 269
 270         sfc_log_init(sa, "entry");
 271
 272         sfc_adapter_lock(sa);
 273         sfc_stop(sa);
 274         sfc_adapter_unlock(sa);
 275
 276         sfc_log_init(sa, "done");
 277 }
 278
 279 static int
 280 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 281 {
 282         struct sfc_adapter *sa = dev->data->dev_private;
 283         int rc;
 284
 285         sfc_log_init(sa, "entry");
 286
 287         sfc_adapter_lock(sa);
 288         rc = sfc_start(sa);
 289         sfc_adapter_unlock(sa);
 290
 291         SFC_ASSERT(rc >= 0);
 292         return -rc;
 293 }
 294
 295 static int
 296 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 297 {
 298         struct sfc_adapter *sa = dev->data->dev_private;
 299
 300         sfc_log_init(sa, "entry");
 301
 302         sfc_adapter_lock(sa);
 303         sfc_stop(sa);
 304         sfc_adapter_unlock(sa);
 305
 306         return 0;
 307 }
 308
 309 static void
 310 sfc_dev_close(struct rte_eth_dev *dev)
 311 {
 312         struct sfc_adapter *sa = dev->data->dev_private;
 313
 314         sfc_log_init(sa, "entry");
 315
 316         sfc_adapter_lock(sa);
 317         switch (sa->state) {
 318         case SFC_ADAPTER_STARTED:
 319                 sfc_stop(sa);
 320                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 321                 /* FALLTHROUGH */
 322         case SFC_ADAPTER_CONFIGURED:
 323                 sfc_close(sa);
 324                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 325                 /* FALLTHROUGH */
 326         case SFC_ADAPTER_INITIALIZED:
 327                 break;
 328         default:
 329                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 330                 break;
 331         }
 332         sfc_adapter_unlock(sa);
 333
 334         sfc_log_init(sa, "done");
 335 }
 336
 337 static void
 338 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 339                    boolean_t enabled)
 340 {
 341         struct sfc_port *port;
 342         boolean_t *toggle;
 343         struct sfc_adapter *sa = dev->data->dev_private;
 344         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 345         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 346
 347         sfc_adapter_lock(sa);
 348
 349         port = &sa->port;
 350         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 351
 352         if (*toggle != enabled) {
 353                 *toggle = enabled;
 354
 355                 if (port->isolated) {
 356                         sfc_warn(sa, "isolated mode is active on the port");
 357                         sfc_warn(sa, "the change is to be applied on the next "
 358                                      "start provided that isolated mode is "
 359                                      "disabled prior the next start");
 360                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 361                            (sfc_set_rx_mode(sa) != 0)) {
 362                         *toggle = !(enabled);
 363                         sfc_warn(sa, "Failed to %s %s mode",
 364                                  ((enabled) ? "enable" : "disable"), desc);
 365                 }
 366         }
 367
 368         sfc_adapter_unlock(sa);
 369 }
 370
 371 static void
 372 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 373 {
 374         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 375 }
 376
 377 static void
 378 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 379 {
 380         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 381 }
 382
 383 static void
 384 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 385 {
 386         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 387 }
 388
 389 static void
 390 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 391 {
 392         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 393 }
 394
 395 static int
 396 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 397                    uint16_t nb_rx_desc, unsigned int socket_id,
 398                    const struct rte_eth_rxconf *rx_conf,
 399                    struct rte_mempool *mb_pool)
 400 {
 401         struct sfc_adapter *sa = dev->data->dev_private;
 402         int rc;
 403
 404         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 405                      rx_queue_id, nb_rx_desc, socket_id);
 406
 407         sfc_adapter_lock(sa);
 408
 409         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 410                           rx_conf, mb_pool);
 411         if (rc != 0)
 412                 goto fail_rx_qinit;
 413
 414         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 415
 416         sfc_adapter_unlock(sa);
 417
 418         return 0;
 419
 420 fail_rx_qinit:
 421         sfc_adapter_unlock(sa);
 422         SFC_ASSERT(rc > 0);
 423         return -rc;
 424 }
 425
 426 static void
 427 sfc_rx_queue_release(void *queue)
 428 {
 429         struct sfc_dp_rxq *dp_rxq = queue;
 430         struct sfc_rxq *rxq;
 431         struct sfc_adapter *sa;
 432         unsigned int sw_index;
 433
 434         if (dp_rxq == NULL)
 435                 return;
 436
 437         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 438         sa = rxq->evq->sa;
 439         sfc_adapter_lock(sa);
 440
 441         sw_index = sfc_rxq_sw_index(rxq);
 442
 443         sfc_log_init(sa, "RxQ=%u", sw_index);
 444
 445         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 446
 447         sfc_rx_qfini(sa, sw_index);
 448
 449         sfc_adapter_unlock(sa);
 450 }
 451
 452 static int
 453 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 454                    uint16_t nb_tx_desc, unsigned int socket_id,
 455                    const struct rte_eth_txconf *tx_conf)
 456 {
 457         struct sfc_adapter *sa = dev->data->dev_private;
 458         int rc;
 459
 460         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 461                      tx_queue_id, nb_tx_desc, socket_id);
 462
 463         sfc_adapter_lock(sa);
 464
 465         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 466         if (rc != 0)
 467                 goto fail_tx_qinit;
 468
 469         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 470
 471         sfc_adapter_unlock(sa);
 472         return 0;
 473
 474 fail_tx_qinit:
 475         sfc_adapter_unlock(sa);
 476         SFC_ASSERT(rc > 0);
 477         return -rc;
 478 }
 479
 480 static void
 481 sfc_tx_queue_release(void *queue)
 482 {
 483         struct sfc_dp_txq *dp_txq = queue;
 484         struct sfc_txq *txq;
 485         unsigned int sw_index;
 486         struct sfc_adapter *sa;
 487
 488         if (dp_txq == NULL)
 489                 return;
 490
 491         txq = sfc_txq_by_dp_txq(dp_txq);
 492         sw_index = sfc_txq_sw_index(txq);
 493
 494         SFC_ASSERT(txq->evq != NULL);
 495         sa = txq->evq->sa;
 496
 497         sfc_log_init(sa, "TxQ = %u", sw_index);
 498
 499         sfc_adapter_lock(sa);
 500
 501         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 502         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 503
 504         sfc_tx_qfini(sa, sw_index);
 505
 506         sfc_adapter_unlock(sa);
 507 }
 508
 509 static int
 510 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 511 {
 512         struct sfc_adapter *sa = dev->data->dev_private;
 513         struct sfc_port *port = &sa->port;
 514         uint64_t *mac_stats;
 515         int ret;
 516
 517         rte_spinlock_lock(&port->mac_stats_lock);
 518
 519         ret = sfc_port_update_mac_stats(sa);
 520         if (ret != 0)
 521                 goto unlock;
 522
 523         mac_stats = port->mac_stats_buf;
 524
 525         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 526                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 527                 stats->ipackets =
 528                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 529                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 530                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 531                 stats->opackets =
 532                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 533                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 534                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 535                 stats->ibytes =
 536                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 537                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 538                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 539                 stats->obytes =
 540                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 541                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 542                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 543                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 544                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 545                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 546         } else {
 547                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 548                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 549                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 550                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 551                 /*
 552                  * Take into account stats which are whenever supported
 553                  * on EF10. If some stat is not supported by current
 554                  * firmware variant or HW revision, it is guaranteed
 555                  * to be zero in mac_stats.
 556                  */
 557                 stats->imissed =
 558                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 559                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 560                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 561                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 562                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 563                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 564                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 565                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 566                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 567                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 568                 stats->ierrors =
 569                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 570                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 571                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 572                 /* no oerrors counters supported on EF10 */
 573         }
 574
 575 unlock:
 576         rte_spinlock_unlock(&port->mac_stats_lock);
 577         SFC_ASSERT(ret >= 0);
 578         return -ret;
 579 }
 580
 581 static void
 582 sfc_stats_reset(struct rte_eth_dev *dev)
 583 {
 584         struct sfc_adapter *sa = dev->data->dev_private;
 585         struct sfc_port *port = &sa->port;
 586         int rc;
 587
 588         if (sa->state != SFC_ADAPTER_STARTED) {
 589                 /*
 590                  * The operation cannot be done if port is not started; it
 591                  * will be scheduled to be done during the next port start
 592                  */
 593                 port->mac_stats_reset_pending = B_TRUE;
 594                 return;
 595         }
 596
 597         rc = sfc_port_reset_mac_stats(sa);
 598         if (rc != 0)
 599                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 600 }
 601
 602 static int
 603 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 604                unsigned int xstats_count)
 605 {
 606         struct sfc_adapter *sa = dev->data->dev_private;
 607         struct sfc_port *port = &sa->port;
 608         uint64_t *mac_stats;
 609         int rc;
 610         unsigned int i;
 611         int nstats = 0;
 612
 613         rte_spinlock_lock(&port->mac_stats_lock);
 614
 615         rc = sfc_port_update_mac_stats(sa);
 616         if (rc != 0) {
 617                 SFC_ASSERT(rc > 0);
 618                 nstats = -rc;
 619                 goto unlock;
 620         }
 621
 622         mac_stats = port->mac_stats_buf;
 623
 624         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 625                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 626                         if (xstats != NULL && nstats < (int)xstats_count) {
 627                                 xstats[nstats].id = nstats;
 628                                 xstats[nstats].value = mac_stats[i];
 629                         }
 630                         nstats++;
 631                 }
 632         }
 633
 634 unlock:
 635         rte_spinlock_unlock(&port->mac_stats_lock);
 636
 637         return nstats;
 638 }
 639
 640 static int
 641 sfc_xstats_get_names(struct rte_eth_dev *dev,
 642                      struct rte_eth_xstat_name *xstats_names,
 643                      unsigned int xstats_count)
 644 {
 645         struct sfc_adapter *sa = dev->data->dev_private;
 646         struct sfc_port *port = &sa->port;
 647         unsigned int i;
 648         unsigned int nstats = 0;
 649
 650         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 651                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 652                         if (xstats_names != NULL && nstats < xstats_count)
 653                                 strncpy(xstats_names[nstats].name,
 654                                         efx_mac_stat_name(sa->nic, i),
 655                                         sizeof(xstats_names[0].name));
 656                         nstats++;
 657                 }
 658         }
 659
 660         return nstats;
 661 }
 662
 663 static int
 664 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 665                      uint64_t *values, unsigned int n)
 666 {
 667         struct sfc_adapter *sa = dev->data->dev_private;
 668         struct sfc_port *port = &sa->port;
 669         uint64_t *mac_stats;
 670         unsigned int nb_supported = 0;
 671         unsigned int nb_written = 0;
 672         unsigned int i;
 673         int ret;
 674         int rc;
 675
 676         if (unlikely(values == NULL) ||
 677             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 678                 return port->mac_stats_nb_supported;
 679
 680         rte_spinlock_lock(&port->mac_stats_lock);
 681
 682         rc = sfc_port_update_mac_stats(sa);
 683         if (rc != 0) {
 684                 SFC_ASSERT(rc > 0);
 685                 ret = -rc;
 686                 goto unlock;
 687         }
 688
 689         mac_stats = port->mac_stats_buf;
 690
 691         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 692                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 693                         continue;
 694
 695                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 696                         values[nb_written++] = mac_stats[i];
 697
 698                 ++nb_supported;
 699         }
 700
 701         ret = nb_written;
 702
 703 unlock:
 704         rte_spinlock_unlock(&port->mac_stats_lock);
 705
 706         return ret;
 707 }
 708
 709 static int
 710 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 711                            struct rte_eth_xstat_name *xstats_names,
 712                            const uint64_t *ids, unsigned int size)
 713 {
 714         struct sfc_adapter *sa = dev->data->dev_private;
 715         struct sfc_port *port = &sa->port;
 716         unsigned int nb_supported = 0;
 717         unsigned int nb_written = 0;
 718         unsigned int i;
 719
 720         if (unlikely(xstats_names == NULL) ||
 721             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 722                 return port->mac_stats_nb_supported;
 723
 724         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 725                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 726                         continue;
 727
 728                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 729                         char *name = xstats_names[nb_written++].name;
 730
 731                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 732                                 sizeof(xstats_names[0].name));
 733                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 734                 }
 735
 736                 ++nb_supported;
 737         }
 738
 739         return nb_written;
 740 }
 741
 742 static int
 743 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 744 {
 745         struct sfc_adapter *sa = dev->data->dev_private;
 746         unsigned int wanted_fc, link_fc;
 747
 748         memset(fc_conf, 0, sizeof(*fc_conf));
 749
 750         sfc_adapter_lock(sa);
 751
 752         if (sa->state == SFC_ADAPTER_STARTED)
 753                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 754         else
 755                 link_fc = sa->port.flow_ctrl;
 756
 757         switch (link_fc) {
 758         case 0:
 759                 fc_conf->mode = RTE_FC_NONE;
 760                 break;
 761         case EFX_FCNTL_RESPOND:
 762                 fc_conf->mode = RTE_FC_RX_PAUSE;
 763                 break;
 764         case EFX_FCNTL_GENERATE:
 765                 fc_conf->mode = RTE_FC_TX_PAUSE;
 766                 break;
 767         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 768                 fc_conf->mode = RTE_FC_FULL;
 769                 break;
 770         default:
 771                 sfc_err(sa, "%s: unexpected flow control value %#x",
 772                         __func__, link_fc);
 773         }
 774
 775         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 776
 777         sfc_adapter_unlock(sa);
 778
 779         return 0;
 780 }
 781
 782 static int
 783 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 784 {
 785         struct sfc_adapter *sa = dev->data->dev_private;
 786         struct sfc_port *port = &sa->port;
 787         unsigned int fcntl;
 788         int rc;
 789
 790         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 791             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 792             fc_conf->mac_ctrl_frame_fwd != 0) {
 793                 sfc_err(sa, "unsupported flow control settings specified");
 794                 rc = EINVAL;
 795                 goto fail_inval;
 796         }
 797
 798         switch (fc_conf->mode) {
 799         case RTE_FC_NONE:
 800                 fcntl = 0;
 801                 break;
 802         case RTE_FC_RX_PAUSE:
 803                 fcntl = EFX_FCNTL_RESPOND;
 804                 break;
 805         case RTE_FC_TX_PAUSE:
 806                 fcntl = EFX_FCNTL_GENERATE;
 807                 break;
 808         case RTE_FC_FULL:
 809                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 810                 break;
 811         default:
 812                 rc = EINVAL;
 813                 goto fail_inval;
 814         }
 815
 816         sfc_adapter_lock(sa);
 817
 818         if (sa->state == SFC_ADAPTER_STARTED) {
 819                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 820                 if (rc != 0)
 821                         goto fail_mac_fcntl_set;
 822         }
 823
 824         port->flow_ctrl = fcntl;
 825         port->flow_ctrl_autoneg = fc_conf->autoneg;
 826
 827         sfc_adapter_unlock(sa);
 828
 829         return 0;
 830
 831 fail_mac_fcntl_set:
 832         sfc_adapter_unlock(sa);
 833 fail_inval:
 834         SFC_ASSERT(rc > 0);
 835         return -rc;
 836 }
 837
 838 static int
 839 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 840 {
 841         struct sfc_adapter *sa = dev->data->dev_private;
 842         size_t pdu = EFX_MAC_PDU(mtu);
 843         size_t old_pdu;
 844         int rc;
 845
 846         sfc_log_init(sa, "mtu=%u", mtu);
 847
 848         rc = EINVAL;
 849         if (pdu < EFX_MAC_PDU_MIN) {
 850                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 851                         (unsigned int)mtu, (unsigned int)pdu,
 852                         EFX_MAC_PDU_MIN);
 853                 goto fail_inval;
 854         }
 855         if (pdu > EFX_MAC_PDU_MAX) {
 856                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 857                         (unsigned int)mtu, (unsigned int)pdu,
 858                         EFX_MAC_PDU_MAX);
 859                 goto fail_inval;
 860         }
 861
 862         sfc_adapter_lock(sa);
 863
 864         if (pdu != sa->port.pdu) {
 865                 if (sa->state == SFC_ADAPTER_STARTED) {
 866                         sfc_stop(sa);
 867
 868                         old_pdu = sa->port.pdu;
 869                         sa->port.pdu = pdu;
 870                         rc = sfc_start(sa);
 871                         if (rc != 0)
 872                                 goto fail_start;
 873                 } else {
 874                         sa->port.pdu = pdu;
 875                 }
 876         }
 877
 878         /*
 879          * The driver does not use it, but other PMDs update jumbo_frame
 880          * flag and max_rx_pkt_len when MTU is set.
 881          */
 882         if (mtu > ETHER_MAX_LEN) {
 883                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 884
 885                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 886                 rxmode->jumbo_frame = 1;
 887         }
 888
 889         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 890
 891         sfc_adapter_unlock(sa);
 892
 893         sfc_log_init(sa, "done");
 894         return 0;
 895
 896 fail_start:
 897         sa->port.pdu = old_pdu;
 898         if (sfc_start(sa) != 0)
 899                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 900                         "PDU max size - port is stopped",
 901                         (unsigned int)pdu, (unsigned int)old_pdu);
 902         sfc_adapter_unlock(sa);
 903
 904 fail_inval:
 905         sfc_log_init(sa, "failed %d", rc);
 906         SFC_ASSERT(rc > 0);
 907         return -rc;
 908 }
 909 static void
 910 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 911 {
 912         struct sfc_adapter *sa = dev->data->dev_private;
 913         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 914         struct sfc_port *port = &sa->port;
 915         int rc;
 916
 917         sfc_adapter_lock(sa);
 918
 919         /*
 920          * Copy the address to the device private data so that
 921          * it could be recalled in the case of adapter restart.
 922          */
 923         ether_addr_copy(mac_addr, &port->default_mac_addr);
 924
 925         if (port->isolated) {
 926                 sfc_err(sa, "isolated mode is active on the port");
 927                 sfc_err(sa, "will not set MAC address");
 928                 goto unlock;
 929         }
 930
 931         if (sa->state != SFC_ADAPTER_STARTED) {
 932                 sfc_info(sa, "the port is not started");
 933                 sfc_info(sa, "the new MAC address will be set on port start");
 934
 935                 goto unlock;
 936         }
 937
 938         if (encp->enc_allow_set_mac_with_installed_filters) {
 939                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 940                 if (rc != 0) {
 941                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 942                         goto unlock;
 943                 }
 944
 945                 /*
 946                  * Changing the MAC address by means of MCDI request
 947                  * has no effect on received traffic, therefore
 948                  * we also need to update unicast filters
 949                  */
 950                 rc = sfc_set_rx_mode(sa);
 951                 if (rc != 0)
 952                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 953         } else {
 954                 sfc_warn(sa, "cannot set MAC address with filters installed");
 955                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 956                 sfc_warn(sa, "(some traffic may be dropped)");
 957
 958                 /*
 959                  * Since setting MAC address with filters installed is not
 960                  * allowed on the adapter, the new MAC address will be set
 961                  * by means of adapter restart. sfc_start() shall retrieve
 962                  * the new address from the device private data and set it.
 963                  */
 964                 sfc_stop(sa);
 965                 rc = sfc_start(sa);
 966                 if (rc != 0)
 967                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 968         }
 969
 970 unlock:
 971         /*
 972          * In the case of failure sa->port->default_mac_addr does not
 973          * need rollback since no error code is returned, and the upper
 974          * API will anyway update the external MAC address storage.
 975          * To be consistent with that new value it is better to keep
 976          * the device private value the same.
 977          */
 978         sfc_adapter_unlock(sa);
 979 }
 980
 981
 982 static int
 983 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 984                      uint32_t nb_mc_addr)
 985 {
 986         struct sfc_adapter *sa = dev->data->dev_private;
 987         struct sfc_port *port = &sa->port;
 988         uint8_t *mc_addrs = port->mcast_addrs;
 989         int rc;
 990         unsigned int i;
 991
 992         if (port->isolated) {
 993                 sfc_err(sa, "isolated mode is active on the port");
 994                 sfc_err(sa, "will not set multicast address list");
 995                 return -ENOTSUP;
 996         }
 997
 998         if (mc_addrs == NULL)
 999                 return -ENOBUFS;
1000
1001         if (nb_mc_addr > port->max_mcast_addrs) {
1002                 sfc_err(sa, "too many multicast addresses: %u > %u",
1003                          nb_mc_addr, port->max_mcast_addrs);
1004                 return -EINVAL;
1005         }
1006
1007         for (i = 0; i < nb_mc_addr; ++i) {
1008                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1009                                  EFX_MAC_ADDR_LEN);
1010                 mc_addrs += EFX_MAC_ADDR_LEN;
1011         }
1012
1013         port->nb_mcast_addrs = nb_mc_addr;
1014
1015         if (sa->state != SFC_ADAPTER_STARTED)
1016                 return 0;
1017
1018         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1019                                         port->nb_mcast_addrs);
1020         if (rc != 0)
1021                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1022
1023         SFC_ASSERT(rc > 0);
1024         return -rc;
1025 }
1026
1027 /*
1028  * The function is used by the secondary process as well. It must not
1029  * use any process-local pointers from the adapter data.
1030  */
1031 static void
1032 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1033                       struct rte_eth_rxq_info *qinfo)
1034 {
1035         struct sfc_adapter *sa = dev->data->dev_private;
1036         struct sfc_rxq_info *rxq_info;
1037         struct sfc_rxq *rxq;
1038
1039         sfc_adapter_lock(sa);
1040
1041         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1042
1043         rxq_info = &sa->rxq_info[rx_queue_id];
1044         rxq = rxq_info->rxq;
1045         SFC_ASSERT(rxq != NULL);
1046
1047         qinfo->mp = rxq->refill_mb_pool;
1048         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1049         qinfo->conf.rx_drop_en = 1;
1050         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1051         qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1052                                DEV_RX_OFFLOAD_UDP_CKSUM |
1053                                DEV_RX_OFFLOAD_TCP_CKSUM;
1054         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1055                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1056                 qinfo->scattered_rx = 1;
1057         }
1058         qinfo->nb_desc = rxq_info->entries;
1059
1060         sfc_adapter_unlock(sa);
1061 }
1062
1063 /*
1064  * The function is used by the secondary process as well. It must not
1065  * use any process-local pointers from the adapter data.
1066  */
1067 static void
1068 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1069                       struct rte_eth_txq_info *qinfo)
1070 {
1071         struct sfc_adapter *sa = dev->data->dev_private;
1072         struct sfc_txq_info *txq_info;
1073
1074         sfc_adapter_lock(sa);
1075
1076         SFC_ASSERT(tx_queue_id < sa->txq_count);
1077
1078         txq_info = &sa->txq_info[tx_queue_id];
1079         SFC_ASSERT(txq_info->txq != NULL);
1080
1081         memset(qinfo, 0, sizeof(*qinfo));
1082
1083         qinfo->conf.txq_flags = txq_info->txq->flags;
1084         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1085         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1086         qinfo->nb_desc = txq_info->entries;
1087
1088         sfc_adapter_unlock(sa);
1089 }
1090
1091 static uint32_t
1092 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1093 {
1094         struct sfc_adapter *sa = dev->data->dev_private;
1095
1096         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1097
1098         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1099 }
1100
1101 static int
1102 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1103 {
1104         struct sfc_dp_rxq *dp_rxq = queue;
1105
1106         return sfc_rx_qdesc_done(dp_rxq, offset);
1107 }
1108
1109 static int
1110 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1111 {
1112         struct sfc_dp_rxq *dp_rxq = queue;
1113         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1114
1115         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1116 }
1117
1118 static int
1119 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1120 {
1121         struct sfc_dp_txq *dp_txq = queue;
1122         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1123
1124         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1125 }
1126
1127 static int
1128 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1129 {
1130         struct sfc_adapter *sa = dev->data->dev_private;
1131         int rc;
1132
1133         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1134
1135         sfc_adapter_lock(sa);
1136
1137         rc = EINVAL;
1138         if (sa->state != SFC_ADAPTER_STARTED)
1139                 goto fail_not_started;
1140
1141         rc = sfc_rx_qstart(sa, rx_queue_id);
1142         if (rc != 0)
1143                 goto fail_rx_qstart;
1144
1145         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1146
1147         sfc_adapter_unlock(sa);
1148
1149         return 0;
1150
1151 fail_rx_qstart:
1152 fail_not_started:
1153         sfc_adapter_unlock(sa);
1154         SFC_ASSERT(rc > 0);
1155         return -rc;
1156 }
1157
1158 static int
1159 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1160 {
1161         struct sfc_adapter *sa = dev->data->dev_private;
1162
1163         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1164
1165         sfc_adapter_lock(sa);
1166         sfc_rx_qstop(sa, rx_queue_id);
1167
1168         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1169
1170         sfc_adapter_unlock(sa);
1171
1172         return 0;
1173 }
1174
1175 static int
1176 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1177 {
1178         struct sfc_adapter *sa = dev->data->dev_private;
1179         int rc;
1180
1181         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1182
1183         sfc_adapter_lock(sa);
1184
1185         rc = EINVAL;
1186         if (sa->state != SFC_ADAPTER_STARTED)
1187                 goto fail_not_started;
1188
1189         rc = sfc_tx_qstart(sa, tx_queue_id);
1190         if (rc != 0)
1191                 goto fail_tx_qstart;
1192
1193         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1194
1195         sfc_adapter_unlock(sa);
1196         return 0;
1197
1198 fail_tx_qstart:
1199
1200 fail_not_started:
1201         sfc_adapter_unlock(sa);
1202         SFC_ASSERT(rc > 0);
1203         return -rc;
1204 }
1205
1206 static int
1207 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1208 {
1209         struct sfc_adapter *sa = dev->data->dev_private;
1210
1211         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1212
1213         sfc_adapter_lock(sa);
1214
1215         sfc_tx_qstop(sa, tx_queue_id);
1216
1217         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1218
1219         sfc_adapter_unlock(sa);
1220         return 0;
1221 }
1222
1223 static efx_tunnel_protocol_t
1224 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1225 {
1226         switch (rte_type) {
1227         case RTE_TUNNEL_TYPE_VXLAN:
1228                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1229         case RTE_TUNNEL_TYPE_GENEVE:
1230                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1231         default:
1232                 return EFX_TUNNEL_NPROTOS;
1233         }
1234 }
1235
1236 enum sfc_udp_tunnel_op_e {
1237         SFC_UDP_TUNNEL_ADD_PORT,
1238         SFC_UDP_TUNNEL_DEL_PORT,
1239 };
1240
1241 static int
1242 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1243                       struct rte_eth_udp_tunnel *tunnel_udp,
1244                       enum sfc_udp_tunnel_op_e op)
1245 {
1246         struct sfc_adapter *sa = dev->data->dev_private;
1247         efx_tunnel_protocol_t tunnel_proto;
1248         int rc;
1249
1250         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1251                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1252                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1253                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1254
1255         tunnel_proto =
1256                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1257         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1258                 rc = ENOTSUP;
1259                 goto fail_bad_proto;
1260         }
1261
1262         sfc_adapter_lock(sa);
1263
1264         switch (op) {
1265         case SFC_UDP_TUNNEL_ADD_PORT:
1266                 rc = efx_tunnel_config_udp_add(sa->nic,
1267                                                tunnel_udp->udp_port,
1268                                                tunnel_proto);
1269                 break;
1270         case SFC_UDP_TUNNEL_DEL_PORT:
1271                 rc = efx_tunnel_config_udp_remove(sa->nic,
1272                                                   tunnel_udp->udp_port,
1273                                                   tunnel_proto);
1274                 break;
1275         default:
1276                 rc = EINVAL;
1277                 goto fail_bad_op;
1278         }
1279
1280         if (rc != 0)
1281                 goto fail_op;
1282
1283         if (sa->state == SFC_ADAPTER_STARTED) {
1284                 rc = efx_tunnel_reconfigure(sa->nic);
1285                 if (rc == EAGAIN) {
1286                         /*
1287                          * Configuration is accepted by FW and MC reboot
1288                          * is initiated to apply the changes. MC reboot
1289                          * will be handled in a usual way (MC reboot
1290                          * event on management event queue and adapter
1291                          * restart).
1292                          */
1293                         rc = 0;
1294                 } else if (rc != 0) {
1295                         goto fail_reconfigure;
1296                 }
1297         }
1298
1299         sfc_adapter_unlock(sa);
1300         return 0;
1301
1302 fail_reconfigure:
1303         /* Remove/restore entry since the change makes the trouble */
1304         switch (op) {
1305         case SFC_UDP_TUNNEL_ADD_PORT:
1306                 (void)efx_tunnel_config_udp_remove(sa->nic,
1307                                                    tunnel_udp->udp_port,
1308                                                    tunnel_proto);
1309                 break;
1310         case SFC_UDP_TUNNEL_DEL_PORT:
1311                 (void)efx_tunnel_config_udp_add(sa->nic,
1312                                                 tunnel_udp->udp_port,
1313                                                 tunnel_proto);
1314                 break;
1315         }
1316
1317 fail_op:
1318 fail_bad_op:
1319         sfc_adapter_unlock(sa);
1320
1321 fail_bad_proto:
1322         SFC_ASSERT(rc > 0);
1323         return -rc;
1324 }
1325
1326 static int
1327 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1328                             struct rte_eth_udp_tunnel *tunnel_udp)
1329 {
1330         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1331 }
1332
1333 static int
1334 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1335                             struct rte_eth_udp_tunnel *tunnel_udp)
1336 {
1337         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1338 }
1339
1340 #if EFSYS_OPT_RX_SCALE
1341 static int
1342 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1343                           struct rte_eth_rss_conf *rss_conf)
1344 {
1345         struct sfc_adapter *sa = dev->data->dev_private;
1346         struct sfc_port *port = &sa->port;
1347
1348         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1349                 return -ENOTSUP;
1350
1351         if (sa->rss_channels == 0)
1352                 return -EINVAL;
1353
1354         sfc_adapter_lock(sa);
1355
1356         /*
1357          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1358          * hence, conversion is done here to derive a correct set of ETH_RSS
1359          * flags which corresponds to the active EFX configuration stored
1360          * locally in 'sfc_adapter' and kept up-to-date
1361          */
1362         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1363         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1364         if (rss_conf->rss_key != NULL)
1365                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1366
1367         sfc_adapter_unlock(sa);
1368
1369         return 0;
1370 }
1371
1372 static int
1373 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1374                         struct rte_eth_rss_conf *rss_conf)
1375 {
1376         struct sfc_adapter *sa = dev->data->dev_private;
1377         struct sfc_port *port = &sa->port;
1378         unsigned int efx_hash_types;
1379         int rc = 0;
1380
1381         if (port->isolated)
1382                 return -ENOTSUP;
1383
1384         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1385                 sfc_err(sa, "RSS is not available");
1386                 return -ENOTSUP;
1387         }
1388
1389         if (sa->rss_channels == 0) {
1390                 sfc_err(sa, "RSS is not configured");
1391                 return -EINVAL;
1392         }
1393
1394         if ((rss_conf->rss_key != NULL) &&
1395             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1396                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1397                         sizeof(sa->rss_key));
1398                 return -EINVAL;
1399         }
1400
1401         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1402                 sfc_err(sa, "unsupported hash functions requested");
1403                 return -EINVAL;
1404         }
1405
1406         sfc_adapter_lock(sa);
1407
1408         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1409
1410         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1411                                    EFX_RX_HASHALG_TOEPLITZ,
1412                                    efx_hash_types, B_TRUE);
1413         if (rc != 0)
1414                 goto fail_scale_mode_set;
1415
1416         if (rss_conf->rss_key != NULL) {
1417                 if (sa->state == SFC_ADAPTER_STARTED) {
1418                         rc = efx_rx_scale_key_set(sa->nic,
1419                                                   EFX_RSS_CONTEXT_DEFAULT,
1420                                                   rss_conf->rss_key,
1421                                                   sizeof(sa->rss_key));
1422                         if (rc != 0)
1423                                 goto fail_scale_key_set;
1424                 }
1425
1426                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1427         }
1428
1429         sa->rss_hash_types = efx_hash_types;
1430
1431         sfc_adapter_unlock(sa);
1432
1433         return 0;
1434
1435 fail_scale_key_set:
1436         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1437                                   EFX_RX_HASHALG_TOEPLITZ,
1438                                   sa->rss_hash_types, B_TRUE) != 0)
1439                 sfc_err(sa, "failed to restore RSS mode");
1440
1441 fail_scale_mode_set:
1442         sfc_adapter_unlock(sa);
1443         return -rc;
1444 }
1445
1446 static int
1447 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1448                        struct rte_eth_rss_reta_entry64 *reta_conf,
1449                        uint16_t reta_size)
1450 {
1451         struct sfc_adapter *sa = dev->data->dev_private;
1452         struct sfc_port *port = &sa->port;
1453         int entry;
1454
1455         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1456                 return -ENOTSUP;
1457
1458         if (sa->rss_channels == 0)
1459                 return -EINVAL;
1460
1461         if (reta_size != EFX_RSS_TBL_SIZE)
1462                 return -EINVAL;
1463
1464         sfc_adapter_lock(sa);
1465
1466         for (entry = 0; entry < reta_size; entry++) {
1467                 int grp = entry / RTE_RETA_GROUP_SIZE;
1468                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1469
1470                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1471                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1472         }
1473
1474         sfc_adapter_unlock(sa);
1475
1476         return 0;
1477 }
1478
1479 static int
1480 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1481                         struct rte_eth_rss_reta_entry64 *reta_conf,
1482                         uint16_t reta_size)
1483 {
1484         struct sfc_adapter *sa = dev->data->dev_private;
1485         struct sfc_port *port = &sa->port;
1486         unsigned int *rss_tbl_new;
1487         uint16_t entry;
1488         int rc = 0;
1489
1490
1491         if (port->isolated)
1492                 return -ENOTSUP;
1493
1494         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1495                 sfc_err(sa, "RSS is not available");
1496                 return -ENOTSUP;
1497         }
1498
1499         if (sa->rss_channels == 0) {
1500                 sfc_err(sa, "RSS is not configured");
1501                 return -EINVAL;
1502         }
1503
1504         if (reta_size != EFX_RSS_TBL_SIZE) {
1505                 sfc_err(sa, "RETA size is wrong (should be %u)",
1506                         EFX_RSS_TBL_SIZE);
1507                 return -EINVAL;
1508         }
1509
1510         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1511         if (rss_tbl_new == NULL)
1512                 return -ENOMEM;
1513
1514         sfc_adapter_lock(sa);
1515
1516         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1517
1518         for (entry = 0; entry < reta_size; entry++) {
1519                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1520                 struct rte_eth_rss_reta_entry64 *grp;
1521
1522                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1523
1524                 if (grp->mask & (1ull << grp_idx)) {
1525                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1526                                 rc = EINVAL;
1527                                 goto bad_reta_entry;
1528                         }
1529                         rss_tbl_new[entry] = grp->reta[grp_idx];
1530                 }
1531         }
1532
1533         if (sa->state == SFC_ADAPTER_STARTED) {
1534                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1535                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1536                 if (rc != 0)
1537                         goto fail_scale_tbl_set;
1538         }
1539
1540         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1541
1542 fail_scale_tbl_set:
1543 bad_reta_entry:
1544         sfc_adapter_unlock(sa);
1545
1546         rte_free(rss_tbl_new);
1547
1548         SFC_ASSERT(rc >= 0);
1549         return -rc;
1550 }
1551 #endif
1552
1553 static int
1554 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1555                     enum rte_filter_op filter_op,
1556                     void *arg)
1557 {
1558         struct sfc_adapter *sa = dev->data->dev_private;
1559         int rc = ENOTSUP;
1560
1561         sfc_log_init(sa, "entry");
1562
1563         switch (filter_type) {
1564         case RTE_ETH_FILTER_NONE:
1565                 sfc_err(sa, "Global filters configuration not supported");
1566                 break;
1567         case RTE_ETH_FILTER_MACVLAN:
1568                 sfc_err(sa, "MACVLAN filters not supported");
1569                 break;
1570         case RTE_ETH_FILTER_ETHERTYPE:
1571                 sfc_err(sa, "EtherType filters not supported");
1572                 break;
1573         case RTE_ETH_FILTER_FLEXIBLE:
1574                 sfc_err(sa, "Flexible filters not supported");
1575                 break;
1576         case RTE_ETH_FILTER_SYN:
1577                 sfc_err(sa, "SYN filters not supported");
1578                 break;
1579         case RTE_ETH_FILTER_NTUPLE:
1580                 sfc_err(sa, "NTUPLE filters not supported");
1581                 break;
1582         case RTE_ETH_FILTER_TUNNEL:
1583                 sfc_err(sa, "Tunnel filters not supported");
1584                 break;
1585         case RTE_ETH_FILTER_FDIR:
1586                 sfc_err(sa, "Flow Director filters not supported");
1587                 break;
1588         case RTE_ETH_FILTER_HASH:
1589                 sfc_err(sa, "Hash filters not supported");
1590                 break;
1591         case RTE_ETH_FILTER_GENERIC:
1592                 if (filter_op != RTE_ETH_FILTER_GET) {
1593                         rc = EINVAL;
1594                 } else {
1595                         *(const void **)arg = &sfc_flow_ops;
1596                         rc = 0;
1597                 }
1598                 break;
1599         default:
1600                 sfc_err(sa, "Unknown filter type %u", filter_type);
1601                 break;
1602         }
1603
1604         sfc_log_init(sa, "exit: %d", -rc);
1605         SFC_ASSERT(rc >= 0);
1606         return -rc;
1607 }
1608
1609 static const struct eth_dev_ops sfc_eth_dev_ops = {
1610         .dev_configure                  = sfc_dev_configure,
1611         .dev_start                      = sfc_dev_start,
1612         .dev_stop                       = sfc_dev_stop,
1613         .dev_set_link_up                = sfc_dev_set_link_up,
1614         .dev_set_link_down              = sfc_dev_set_link_down,
1615         .dev_close                      = sfc_dev_close,
1616         .promiscuous_enable             = sfc_dev_promisc_enable,
1617         .promiscuous_disable            = sfc_dev_promisc_disable,
1618         .allmulticast_enable            = sfc_dev_allmulti_enable,
1619         .allmulticast_disable           = sfc_dev_allmulti_disable,
1620         .link_update                    = sfc_dev_link_update,
1621         .stats_get                      = sfc_stats_get,
1622         .stats_reset                    = sfc_stats_reset,
1623         .xstats_get                     = sfc_xstats_get,
1624         .xstats_reset                   = sfc_stats_reset,
1625         .xstats_get_names               = sfc_xstats_get_names,
1626         .dev_infos_get                  = sfc_dev_infos_get,
1627         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1628         .mtu_set                        = sfc_dev_set_mtu,
1629         .rx_queue_start                 = sfc_rx_queue_start,
1630         .rx_queue_stop                  = sfc_rx_queue_stop,
1631         .tx_queue_start                 = sfc_tx_queue_start,
1632         .tx_queue_stop                  = sfc_tx_queue_stop,
1633         .rx_queue_setup                 = sfc_rx_queue_setup,
1634         .rx_queue_release               = sfc_rx_queue_release,
1635         .rx_queue_count                 = sfc_rx_queue_count,
1636         .rx_descriptor_done             = sfc_rx_descriptor_done,
1637         .rx_descriptor_status           = sfc_rx_descriptor_status,
1638         .tx_descriptor_status           = sfc_tx_descriptor_status,
1639         .tx_queue_setup                 = sfc_tx_queue_setup,
1640         .tx_queue_release               = sfc_tx_queue_release,
1641         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1642         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1643         .mac_addr_set                   = sfc_mac_addr_set,
1644         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1645         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1646 #if EFSYS_OPT_RX_SCALE
1647         .reta_update                    = sfc_dev_rss_reta_update,
1648         .reta_query                     = sfc_dev_rss_reta_query,
1649         .rss_hash_update                = sfc_dev_rss_hash_update,
1650         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1651 #endif
1652         .filter_ctrl                    = sfc_dev_filter_ctrl,
1653         .set_mc_addr_list               = sfc_set_mc_addr_list,
1654         .rxq_info_get                   = sfc_rx_queue_info_get,
1655         .txq_info_get                   = sfc_tx_queue_info_get,
1656         .fw_version_get                 = sfc_fw_version_get,
1657         .xstats_get_by_id               = sfc_xstats_get_by_id,
1658         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1659 };
1660
1661 /**
1662  * Duplicate a string in potentially shared memory required for
1663  * multi-process support.
1664  *
1665  * strdup() allocates from process-local heap/memory.
1666  */
1667 static char *
1668 sfc_strdup(const char *str)
1669 {
1670         size_t size;
1671         char *copy;
1672
1673         if (str == NULL)
1674                 return NULL;
1675
1676         size = strlen(str) + 1;
1677         copy = rte_malloc(__func__, size, 0);
1678         if (copy != NULL)
1679                 rte_memcpy(copy, str, size);
1680
1681         return copy;
1682 }
1683
1684 static int
1685 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1686 {
1687         struct sfc_adapter *sa = dev->data->dev_private;
1688         unsigned int avail_caps = 0;
1689         const char *rx_name = NULL;
1690         const char *tx_name = NULL;
1691         int rc;
1692
1693         switch (sa->family) {
1694         case EFX_FAMILY_HUNTINGTON:
1695         case EFX_FAMILY_MEDFORD:
1696                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1697                 break;
1698         default:
1699                 break;
1700         }
1701
1702         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1703                                 sfc_kvarg_string_handler, &rx_name);
1704         if (rc != 0)
1705                 goto fail_kvarg_rx_datapath;
1706
1707         if (rx_name != NULL) {
1708                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1709                 if (sa->dp_rx == NULL) {
1710                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1711                         rc = ENOENT;
1712                         goto fail_dp_rx;
1713                 }
1714                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1715                         sfc_err(sa,
1716                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1717                                 rx_name);
1718                         rc = EINVAL;
1719                         goto fail_dp_rx_caps;
1720                 }
1721         } else {
1722                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1723                 if (sa->dp_rx == NULL) {
1724                         sfc_err(sa, "Rx datapath by caps %#x not found",
1725                                 avail_caps);
1726                         rc = ENOENT;
1727                         goto fail_dp_rx;
1728                 }
1729         }
1730
1731         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1732         if (sa->dp_rx_name == NULL) {
1733                 rc = ENOMEM;
1734                 goto fail_dp_rx_name;
1735         }
1736
1737         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1738
1739         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1740
1741         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1742                                 sfc_kvarg_string_handler, &tx_name);
1743         if (rc != 0)
1744                 goto fail_kvarg_tx_datapath;
1745
1746         if (tx_name != NULL) {
1747                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1748                 if (sa->dp_tx == NULL) {
1749                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1750                         rc = ENOENT;
1751                         goto fail_dp_tx;
1752                 }
1753                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1754                         sfc_err(sa,
1755                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1756                                 tx_name);
1757                         rc = EINVAL;
1758                         goto fail_dp_tx_caps;
1759                 }
1760         } else {
1761                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1762                 if (sa->dp_tx == NULL) {
1763                         sfc_err(sa, "Tx datapath by caps %#x not found",
1764                                 avail_caps);
1765                         rc = ENOENT;
1766                         goto fail_dp_tx;
1767                 }
1768         }
1769
1770         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1771         if (sa->dp_tx_name == NULL) {
1772                 rc = ENOMEM;
1773                 goto fail_dp_tx_name;
1774         }
1775
1776         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1777
1778         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1779
1780         dev->dev_ops = &sfc_eth_dev_ops;
1781
1782         return 0;
1783
1784 fail_dp_tx_name:
1785 fail_dp_tx_caps:
1786         sa->dp_tx = NULL;
1787
1788 fail_dp_tx:
1789 fail_kvarg_tx_datapath:
1790         rte_free(sa->dp_rx_name);
1791         sa->dp_rx_name = NULL;
1792
1793 fail_dp_rx_name:
1794 fail_dp_rx_caps:
1795         sa->dp_rx = NULL;
1796
1797 fail_dp_rx:
1798 fail_kvarg_rx_datapath:
1799         return rc;
1800 }
1801
1802 static void
1803 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1804 {
1805         struct sfc_adapter *sa = dev->data->dev_private;
1806
1807         dev->dev_ops = NULL;
1808         dev->rx_pkt_burst = NULL;
1809         dev->tx_pkt_burst = NULL;
1810
1811         rte_free(sa->dp_tx_name);
1812         sa->dp_tx_name = NULL;
1813         sa->dp_tx = NULL;
1814
1815         rte_free(sa->dp_rx_name);
1816         sa->dp_rx_name = NULL;
1817         sa->dp_rx = NULL;
1818 }
1819
1820 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1821         .rxq_info_get                   = sfc_rx_queue_info_get,
1822         .txq_info_get                   = sfc_tx_queue_info_get,
1823 };
1824
1825 static int
1826 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1827 {
1828         /*
1829          * Device private data has really many process-local pointers.
1830          * Below code should be extremely careful to use data located
1831          * in shared memory only.
1832          */
1833         struct sfc_adapter *sa = dev->data->dev_private;
1834         const struct sfc_dp_rx *dp_rx;
1835         const struct sfc_dp_tx *dp_tx;
1836         int rc;
1837
1838         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1839         if (dp_rx == NULL) {
1840                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1841                 rc = ENOENT;
1842                 goto fail_dp_rx;
1843         }
1844         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1845                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1846                         sa->dp_tx_name);
1847                 rc = EINVAL;
1848                 goto fail_dp_rx_multi_process;
1849         }
1850
1851         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1852         if (dp_tx == NULL) {
1853                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1854                 rc = ENOENT;
1855                 goto fail_dp_tx;
1856         }
1857         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1858                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1859                         sa->dp_tx_name);
1860                 rc = EINVAL;
1861                 goto fail_dp_tx_multi_process;
1862         }
1863
1864         dev->rx_pkt_burst = dp_rx->pkt_burst;
1865         dev->tx_pkt_burst = dp_tx->pkt_burst;
1866         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1867
1868         return 0;
1869
1870 fail_dp_tx_multi_process:
1871 fail_dp_tx:
1872 fail_dp_rx_multi_process:
1873 fail_dp_rx:
1874         return rc;
1875 }
1876
1877 static void
1878 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1879 {
1880         dev->dev_ops = NULL;
1881         dev->tx_pkt_burst = NULL;
1882         dev->rx_pkt_burst = NULL;
1883 }
1884
1885 static void
1886 sfc_register_dp(void)
1887 {
1888         /* Register once */
1889         if (TAILQ_EMPTY(&sfc_dp_head)) {
1890                 /* Prefer EF10 datapath */
1891                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1892                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1893
1894                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1895                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1896                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1897         }
1898 }
1899
1900 static int
1901 sfc_eth_dev_init(struct rte_eth_dev *dev)
1902 {
1903         struct sfc_adapter *sa = dev->data->dev_private;
1904         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1905         int rc;
1906         const efx_nic_cfg_t *encp;
1907         const struct ether_addr *from;
1908
1909         sfc_register_dp();
1910
1911         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1912                 return -sfc_eth_dev_secondary_set_ops(dev);
1913
1914         /* Required for logging */
1915         sa->pci_addr = pci_dev->addr;
1916         sa->port_id = dev->data->port_id;
1917
1918         sa->eth_dev = dev;
1919
1920         /* Copy PCI device info to the dev->data */
1921         rte_eth_copy_pci_info(dev, pci_dev);
1922
1923         rc = sfc_kvargs_parse(sa);
1924         if (rc != 0)
1925                 goto fail_kvargs_parse;
1926
1927         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1928                                 sfc_kvarg_bool_handler, &sa->debug_init);
1929         if (rc != 0)
1930                 goto fail_kvarg_debug_init;
1931
1932         sfc_log_init(sa, "entry");
1933
1934         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1935         if (dev->data->mac_addrs == NULL) {
1936                 rc = ENOMEM;
1937                 goto fail_mac_addrs;
1938         }
1939
1940         sfc_adapter_lock_init(sa);
1941         sfc_adapter_lock(sa);
1942
1943         sfc_log_init(sa, "probing");
1944         rc = sfc_probe(sa);
1945         if (rc != 0)
1946                 goto fail_probe;
1947
1948         sfc_log_init(sa, "set device ops");
1949         rc = sfc_eth_dev_set_ops(dev);
1950         if (rc != 0)
1951                 goto fail_set_ops;
1952
1953         sfc_log_init(sa, "attaching");
1954         rc = sfc_attach(sa);
1955         if (rc != 0)
1956                 goto fail_attach;
1957
1958         encp = efx_nic_cfg_get(sa->nic);
1959
1960         /*
1961          * The arguments are really reverse order in comparison to
1962          * Linux kernel. Copy from NIC config to Ethernet device data.
1963          */
1964         from = (const struct ether_addr *)(encp->enc_mac_addr);
1965         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1966
1967         sfc_adapter_unlock(sa);
1968
1969         sfc_log_init(sa, "done");
1970         return 0;
1971
1972 fail_attach:
1973         sfc_eth_dev_clear_ops(dev);
1974
1975 fail_set_ops:
1976         sfc_unprobe(sa);
1977
1978 fail_probe:
1979         sfc_adapter_unlock(sa);
1980         sfc_adapter_lock_fini(sa);
1981         rte_free(dev->data->mac_addrs);
1982         dev->data->mac_addrs = NULL;
1983
1984 fail_mac_addrs:
1985 fail_kvarg_debug_init:
1986         sfc_kvargs_cleanup(sa);
1987
1988 fail_kvargs_parse:
1989         sfc_log_init(sa, "failed %d", rc);
1990         SFC_ASSERT(rc > 0);
1991         return -rc;
1992 }
1993
1994 static int
1995 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1996 {
1997         struct sfc_adapter *sa;
1998
1999         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2000                 sfc_eth_dev_secondary_clear_ops(dev);
2001                 return 0;
2002         }
2003
2004         sa = dev->data->dev_private;
2005         sfc_log_init(sa, "entry");
2006
2007         sfc_adapter_lock(sa);
2008
2009         sfc_eth_dev_clear_ops(dev);
2010
2011         sfc_detach(sa);
2012         sfc_unprobe(sa);
2013
2014         rte_free(dev->data->mac_addrs);
2015         dev->data->mac_addrs = NULL;
2016
2017         sfc_kvargs_cleanup(sa);
2018
2019         sfc_adapter_unlock(sa);
2020         sfc_adapter_lock_fini(sa);
2021
2022         sfc_log_init(sa, "done");
2023
2024         /* Required for logging, so cleanup last */
2025         sa->eth_dev = NULL;
2026         return 0;
2027 }
2028
2029 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2030         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2031         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2032         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2033         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2034         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2035         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2036         { .vendor_id = 0 /* sentinel */ }
2037 };
2038
2039 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2040         struct rte_pci_device *pci_dev)
2041 {
2042         return rte_eth_dev_pci_generic_probe(pci_dev,
2043                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2044 }
2045
2046 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2047 {
2048         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2049 }
2050
2051 static struct rte_pci_driver sfc_efx_pmd = {
2052         .id_table = pci_id_sfc_efx_map,
2053         .drv_flags =
2054                 RTE_PCI_DRV_INTR_LSC |
2055                 RTE_PCI_DRV_NEED_MAPPING,
2056         .probe = sfc_eth_dev_pci_probe,
2057         .remove = sfc_eth_dev_pci_remove,
2058 };
2059
2060 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2061 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2062 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2063 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2064         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2065         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2066         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2067         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2068         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2069         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);