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_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17
  18 #include "efx.h"
  19
  20 #include "sfc.h"
  21 #include "sfc_debug.h"
  22 #include "sfc_log.h"
  23 #include "sfc_kvargs.h"
  24 #include "sfc_ev.h"
  25 #include "sfc_rx.h"
  26 #include "sfc_tx.h"
  27 #include "sfc_flow.h"
  28 #include "sfc_dp.h"
  29 #include "sfc_dp_rx.h"
  30
  31 uint32_t sfc_logtype_driver;
  32
  33 static struct sfc_dp_list sfc_dp_head =
  34         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  35
  36 static int
  37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  38 {
  39         struct sfc_adapter *sa = dev->data->dev_private;
  40         efx_nic_fw_info_t enfi;
  41         int ret;
  42         int rc;
  43
  44         /*
  45          * Return value of the callback is likely supposed to be
  46          * equal to or greater than 0, nevertheless, if an error
  47          * occurs, it will be desirable to pass it to the caller
  48          */
  49         if ((fw_version == NULL) || (fw_size == 0))
  50                 return -EINVAL;
  51
  52         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  53         if (rc != 0)
  54                 return -rc;
  55
  56         ret = snprintf(fw_version, fw_size,
  57                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  58                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  59                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  60         if (ret < 0)
  61                 return ret;
  62
  63         if (enfi.enfi_dpcpu_fw_ids_valid) {
  64                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  65                 int ret_extra;
  66
  67                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  68                                      fw_size - dpcpu_fw_ids_offset,
  69                                      " rx%" PRIx16 " tx%" PRIx16,
  70                                      enfi.enfi_rx_dpcpu_fw_id,
  71                                      enfi.enfi_tx_dpcpu_fw_id);
  72                 if (ret_extra < 0)
  73                         return ret_extra;
  74
  75                 ret += ret_extra;
  76         }
  77
  78         if (fw_size < (size_t)(++ret))
  79                 return ret;
  80         else
  81                 return 0;
  82 }
  83
  84 static void
  85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  86 {
  87         struct sfc_adapter *sa = dev->data->dev_private;
  88         struct sfc_rss *rss = &sa->rss;
  89         uint64_t txq_offloads_def = 0;
  90
  91         sfc_log_init(sa, "entry");
  92
  93         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  94
  95         /* Autonegotiation may be disabled */
  96         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  97         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  98                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  99         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
 100                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 101         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
 102                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 103         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 104                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 105         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
 106                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 107         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
 108                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 109
 110         dev_info->max_rx_queues = sa->rxq_max;
 111         dev_info->max_tx_queues = sa->txq_max;
 112
 113         /* By default packets are dropped if no descriptors are available */
 114         dev_info->default_rxconf.rx_drop_en = 1;
 115
 116         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 117
 118         /*
 119          * rx_offload_capa includes both device and queue offloads since
 120          * the latter may be requested on a per device basis which makes
 121          * sense when some offloads are needed to be set on all queues.
 122          */
 123         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 124                                     dev_info->rx_queue_offload_capa;
 125
 126         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 127
 128         /*
 129          * tx_offload_capa includes both device and queue offloads since
 130          * the latter may be requested on a per device basis which makes
 131          * sense when some offloads are needed to be set on all queues.
 132          */
 133         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 134                                     dev_info->tx_queue_offload_capa;
 135
 136         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 137                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 138
 139         dev_info->default_txconf.offloads |= txq_offloads_def;
 140
 141         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 142                 uint64_t rte_hf = 0;
 143                 unsigned int i;
 144
 145                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 146                         rte_hf |= rss->hf_map[i].rte;
 147
 148                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 149                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 150                 dev_info->flow_type_rss_offloads = rte_hf;
 151         }
 152
 153         /* Initialize to hardware limits */
 154         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 155         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 156         /* The RXQ hardware requires that the descriptor count is a power
 157          * of 2, but rx_desc_lim cannot properly describe that constraint.
 158          */
 159         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 160
 161         /* Initialize to hardware limits */
 162         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 163         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 164         /*
 165          * The TXQ hardware requires that the descriptor count is a power
 166          * of 2, but tx_desc_lim cannot properly describe that constraint
 167          */
 168         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 169
 170         if (sa->dp_rx->get_dev_info != NULL)
 171                 sa->dp_rx->get_dev_info(dev_info);
 172         if (sa->dp_tx->get_dev_info != NULL)
 173                 sa->dp_tx->get_dev_info(dev_info);
 174
 175         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP;
 176 }
 177
 178 static const uint32_t *
 179 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 180 {
 181         struct sfc_adapter *sa = dev->data->dev_private;
 182         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 183         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 184
 185         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 186 }
 187
 188 static int
 189 sfc_dev_configure(struct rte_eth_dev *dev)
 190 {
 191         struct rte_eth_dev_data *dev_data = dev->data;
 192         struct sfc_adapter *sa = dev_data->dev_private;
 193         int rc;
 194
 195         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 196                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 197
 198         sfc_adapter_lock(sa);
 199         switch (sa->state) {
 200         case SFC_ADAPTER_CONFIGURED:
 201                 /* FALLTHROUGH */
 202         case SFC_ADAPTER_INITIALIZED:
 203                 rc = sfc_configure(sa);
 204                 break;
 205         default:
 206                 sfc_err(sa, "unexpected adapter state %u to configure",
 207                         sa->state);
 208                 rc = EINVAL;
 209                 break;
 210         }
 211         sfc_adapter_unlock(sa);
 212
 213         sfc_log_init(sa, "done %d", rc);
 214         SFC_ASSERT(rc >= 0);
 215         return -rc;
 216 }
 217
 218 static int
 219 sfc_dev_start(struct rte_eth_dev *dev)
 220 {
 221         struct sfc_adapter *sa = dev->data->dev_private;
 222         int rc;
 223
 224         sfc_log_init(sa, "entry");
 225
 226         sfc_adapter_lock(sa);
 227         rc = sfc_start(sa);
 228         sfc_adapter_unlock(sa);
 229
 230         sfc_log_init(sa, "done %d", rc);
 231         SFC_ASSERT(rc >= 0);
 232         return -rc;
 233 }
 234
 235 static int
 236 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 237 {
 238         struct sfc_adapter *sa = dev->data->dev_private;
 239         struct rte_eth_link current_link;
 240         int ret;
 241
 242         sfc_log_init(sa, "entry");
 243
 244         if (sa->state != SFC_ADAPTER_STARTED) {
 245                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 246         } else if (wait_to_complete) {
 247                 efx_link_mode_t link_mode;
 248
 249                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 250                         link_mode = EFX_LINK_UNKNOWN;
 251                 sfc_port_link_mode_to_info(link_mode, &current_link);
 252
 253         } else {
 254                 sfc_ev_mgmt_qpoll(sa);
 255                 rte_eth_linkstatus_get(dev, &current_link);
 256         }
 257
 258         ret = rte_eth_linkstatus_set(dev, &current_link);
 259         if (ret == 0)
 260                 sfc_notice(sa, "Link status is %s",
 261                            current_link.link_status ? "UP" : "DOWN");
 262
 263         return ret;
 264 }
 265
 266 static void
 267 sfc_dev_stop(struct rte_eth_dev *dev)
 268 {
 269         struct sfc_adapter *sa = dev->data->dev_private;
 270
 271         sfc_log_init(sa, "entry");
 272
 273         sfc_adapter_lock(sa);
 274         sfc_stop(sa);
 275         sfc_adapter_unlock(sa);
 276
 277         sfc_log_init(sa, "done");
 278 }
 279
 280 static int
 281 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 282 {
 283         struct sfc_adapter *sa = dev->data->dev_private;
 284         int rc;
 285
 286         sfc_log_init(sa, "entry");
 287
 288         sfc_adapter_lock(sa);
 289         rc = sfc_start(sa);
 290         sfc_adapter_unlock(sa);
 291
 292         SFC_ASSERT(rc >= 0);
 293         return -rc;
 294 }
 295
 296 static int
 297 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 298 {
 299         struct sfc_adapter *sa = dev->data->dev_private;
 300
 301         sfc_log_init(sa, "entry");
 302
 303         sfc_adapter_lock(sa);
 304         sfc_stop(sa);
 305         sfc_adapter_unlock(sa);
 306
 307         return 0;
 308 }
 309
 310 static void
 311 sfc_dev_close(struct rte_eth_dev *dev)
 312 {
 313         struct sfc_adapter *sa = dev->data->dev_private;
 314
 315         sfc_log_init(sa, "entry");
 316
 317         sfc_adapter_lock(sa);
 318         switch (sa->state) {
 319         case SFC_ADAPTER_STARTED:
 320                 sfc_stop(sa);
 321                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 322                 /* FALLTHROUGH */
 323         case SFC_ADAPTER_CONFIGURED:
 324                 sfc_close(sa);
 325                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 326                 /* FALLTHROUGH */
 327         case SFC_ADAPTER_INITIALIZED:
 328                 break;
 329         default:
 330                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 331                 break;
 332         }
 333         sfc_adapter_unlock(sa);
 334
 335         sfc_log_init(sa, "done");
 336 }
 337
 338 static void
 339 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 340                    boolean_t enabled)
 341 {
 342         struct sfc_port *port;
 343         boolean_t *toggle;
 344         struct sfc_adapter *sa = dev->data->dev_private;
 345         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 346         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 347
 348         sfc_adapter_lock(sa);
 349
 350         port = &sa->port;
 351         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 352
 353         if (*toggle != enabled) {
 354                 *toggle = enabled;
 355
 356                 if (port->isolated) {
 357                         sfc_warn(sa, "isolated mode is active on the port");
 358                         sfc_warn(sa, "the change is to be applied on the next "
 359                                      "start provided that isolated mode is "
 360                                      "disabled prior the next start");
 361                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 362                            (sfc_set_rx_mode(sa) != 0)) {
 363                         *toggle = !(enabled);
 364                         sfc_warn(sa, "Failed to %s %s mode",
 365                                  ((enabled) ? "enable" : "disable"), desc);
 366                 }
 367         }
 368
 369         sfc_adapter_unlock(sa);
 370 }
 371
 372 static void
 373 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 374 {
 375         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 376 }
 377
 378 static void
 379 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 380 {
 381         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 382 }
 383
 384 static void
 385 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 386 {
 387         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 388 }
 389
 390 static void
 391 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 392 {
 393         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 394 }
 395
 396 static int
 397 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 398                    uint16_t nb_rx_desc, unsigned int socket_id,
 399                    const struct rte_eth_rxconf *rx_conf,
 400                    struct rte_mempool *mb_pool)
 401 {
 402         struct sfc_adapter *sa = dev->data->dev_private;
 403         int rc;
 404
 405         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 406                      rx_queue_id, nb_rx_desc, socket_id);
 407
 408         sfc_adapter_lock(sa);
 409
 410         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 411                           rx_conf, mb_pool);
 412         if (rc != 0)
 413                 goto fail_rx_qinit;
 414
 415         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 416
 417         sfc_adapter_unlock(sa);
 418
 419         return 0;
 420
 421 fail_rx_qinit:
 422         sfc_adapter_unlock(sa);
 423         SFC_ASSERT(rc > 0);
 424         return -rc;
 425 }
 426
 427 static void
 428 sfc_rx_queue_release(void *queue)
 429 {
 430         struct sfc_dp_rxq *dp_rxq = queue;
 431         struct sfc_rxq *rxq;
 432         struct sfc_adapter *sa;
 433         unsigned int sw_index;
 434
 435         if (dp_rxq == NULL)
 436                 return;
 437
 438         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 439         sa = rxq->evq->sa;
 440         sfc_adapter_lock(sa);
 441
 442         sw_index = sfc_rxq_sw_index(rxq);
 443
 444         sfc_log_init(sa, "RxQ=%u", sw_index);
 445
 446         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 447
 448         sfc_rx_qfini(sa, sw_index);
 449
 450         sfc_adapter_unlock(sa);
 451 }
 452
 453 static int
 454 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 455                    uint16_t nb_tx_desc, unsigned int socket_id,
 456                    const struct rte_eth_txconf *tx_conf)
 457 {
 458         struct sfc_adapter *sa = dev->data->dev_private;
 459         int rc;
 460
 461         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 462                      tx_queue_id, nb_tx_desc, socket_id);
 463
 464         sfc_adapter_lock(sa);
 465
 466         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 467         if (rc != 0)
 468                 goto fail_tx_qinit;
 469
 470         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 471
 472         sfc_adapter_unlock(sa);
 473         return 0;
 474
 475 fail_tx_qinit:
 476         sfc_adapter_unlock(sa);
 477         SFC_ASSERT(rc > 0);
 478         return -rc;
 479 }
 480
 481 static void
 482 sfc_tx_queue_release(void *queue)
 483 {
 484         struct sfc_dp_txq *dp_txq = queue;
 485         struct sfc_txq *txq;
 486         unsigned int sw_index;
 487         struct sfc_adapter *sa;
 488
 489         if (dp_txq == NULL)
 490                 return;
 491
 492         txq = sfc_txq_by_dp_txq(dp_txq);
 493         sw_index = sfc_txq_sw_index(txq);
 494
 495         SFC_ASSERT(txq->evq != NULL);
 496         sa = txq->evq->sa;
 497
 498         sfc_log_init(sa, "TxQ = %u", sw_index);
 499
 500         sfc_adapter_lock(sa);
 501
 502         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 503         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 504
 505         sfc_tx_qfini(sa, sw_index);
 506
 507         sfc_adapter_unlock(sa);
 508 }
 509
 510 static int
 511 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 512 {
 513         struct sfc_adapter *sa = dev->data->dev_private;
 514         struct sfc_port *port = &sa->port;
 515         uint64_t *mac_stats;
 516         int ret;
 517
 518         rte_spinlock_lock(&port->mac_stats_lock);
 519
 520         ret = sfc_port_update_mac_stats(sa);
 521         if (ret != 0)
 522                 goto unlock;
 523
 524         mac_stats = port->mac_stats_buf;
 525
 526         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 527                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 528                 stats->ipackets =
 529                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 530                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 531                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 532                 stats->opackets =
 533                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 534                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 535                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 536                 stats->ibytes =
 537                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 538                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 539                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 540                 stats->obytes =
 541                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 542                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 543                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 544                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 545                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 546                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 547         } else {
 548                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 549                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 550                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 551                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 552                 /*
 553                  * Take into account stats which are whenever supported
 554                  * on EF10. If some stat is not supported by current
 555                  * firmware variant or HW revision, it is guaranteed
 556                  * to be zero in mac_stats.
 557                  */
 558                 stats->imissed =
 559                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 560                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 561                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 562                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 563                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 564                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 565                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 566                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 567                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 568                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 569                 stats->ierrors =
 570                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 571                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 572                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 573                 /* no oerrors counters supported on EF10 */
 574         }
 575
 576 unlock:
 577         rte_spinlock_unlock(&port->mac_stats_lock);
 578         SFC_ASSERT(ret >= 0);
 579         return -ret;
 580 }
 581
 582 static void
 583 sfc_stats_reset(struct rte_eth_dev *dev)
 584 {
 585         struct sfc_adapter *sa = dev->data->dev_private;
 586         struct sfc_port *port = &sa->port;
 587         int rc;
 588
 589         if (sa->state != SFC_ADAPTER_STARTED) {
 590                 /*
 591                  * The operation cannot be done if port is not started; it
 592                  * will be scheduled to be done during the next port start
 593                  */
 594                 port->mac_stats_reset_pending = B_TRUE;
 595                 return;
 596         }
 597
 598         rc = sfc_port_reset_mac_stats(sa);
 599         if (rc != 0)
 600                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 601 }
 602
 603 static int
 604 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 605                unsigned int xstats_count)
 606 {
 607         struct sfc_adapter *sa = dev->data->dev_private;
 608         struct sfc_port *port = &sa->port;
 609         uint64_t *mac_stats;
 610         int rc;
 611         unsigned int i;
 612         int nstats = 0;
 613
 614         rte_spinlock_lock(&port->mac_stats_lock);
 615
 616         rc = sfc_port_update_mac_stats(sa);
 617         if (rc != 0) {
 618                 SFC_ASSERT(rc > 0);
 619                 nstats = -rc;
 620                 goto unlock;
 621         }
 622
 623         mac_stats = port->mac_stats_buf;
 624
 625         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 626                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 627                         if (xstats != NULL && nstats < (int)xstats_count) {
 628                                 xstats[nstats].id = nstats;
 629                                 xstats[nstats].value = mac_stats[i];
 630                         }
 631                         nstats++;
 632                 }
 633         }
 634
 635 unlock:
 636         rte_spinlock_unlock(&port->mac_stats_lock);
 637
 638         return nstats;
 639 }
 640
 641 static int
 642 sfc_xstats_get_names(struct rte_eth_dev *dev,
 643                      struct rte_eth_xstat_name *xstats_names,
 644                      unsigned int xstats_count)
 645 {
 646         struct sfc_adapter *sa = dev->data->dev_private;
 647         struct sfc_port *port = &sa->port;
 648         unsigned int i;
 649         unsigned int nstats = 0;
 650
 651         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 652                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 653                         if (xstats_names != NULL && nstats < xstats_count)
 654                                 strlcpy(xstats_names[nstats].name,
 655                                         efx_mac_stat_name(sa->nic, i),
 656                                         sizeof(xstats_names[0].name));
 657                         nstats++;
 658                 }
 659         }
 660
 661         return nstats;
 662 }
 663
 664 static int
 665 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 666                      uint64_t *values, unsigned int n)
 667 {
 668         struct sfc_adapter *sa = dev->data->dev_private;
 669         struct sfc_port *port = &sa->port;
 670         uint64_t *mac_stats;
 671         unsigned int nb_supported = 0;
 672         unsigned int nb_written = 0;
 673         unsigned int i;
 674         int ret;
 675         int rc;
 676
 677         if (unlikely(values == NULL) ||
 678             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 679                 return port->mac_stats_nb_supported;
 680
 681         rte_spinlock_lock(&port->mac_stats_lock);
 682
 683         rc = sfc_port_update_mac_stats(sa);
 684         if (rc != 0) {
 685                 SFC_ASSERT(rc > 0);
 686                 ret = -rc;
 687                 goto unlock;
 688         }
 689
 690         mac_stats = port->mac_stats_buf;
 691
 692         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 693                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 694                         continue;
 695
 696                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 697                         values[nb_written++] = mac_stats[i];
 698
 699                 ++nb_supported;
 700         }
 701
 702         ret = nb_written;
 703
 704 unlock:
 705         rte_spinlock_unlock(&port->mac_stats_lock);
 706
 707         return ret;
 708 }
 709
 710 static int
 711 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 712                            struct rte_eth_xstat_name *xstats_names,
 713                            const uint64_t *ids, unsigned int size)
 714 {
 715         struct sfc_adapter *sa = dev->data->dev_private;
 716         struct sfc_port *port = &sa->port;
 717         unsigned int nb_supported = 0;
 718         unsigned int nb_written = 0;
 719         unsigned int i;
 720
 721         if (unlikely(xstats_names == NULL) ||
 722             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 723                 return port->mac_stats_nb_supported;
 724
 725         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 726                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 727                         continue;
 728
 729                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 730                         char *name = xstats_names[nb_written++].name;
 731
 732                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 733                                 sizeof(xstats_names[0].name));
 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                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 885         }
 886
 887         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 888
 889         sfc_adapter_unlock(sa);
 890
 891         sfc_log_init(sa, "done");
 892         return 0;
 893
 894 fail_start:
 895         sa->port.pdu = old_pdu;
 896         if (sfc_start(sa) != 0)
 897                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 898                         "PDU max size - port is stopped",
 899                         (unsigned int)pdu, (unsigned int)old_pdu);
 900         sfc_adapter_unlock(sa);
 901
 902 fail_inval:
 903         sfc_log_init(sa, "failed %d", rc);
 904         SFC_ASSERT(rc > 0);
 905         return -rc;
 906 }
 907 static int
 908 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 909 {
 910         struct sfc_adapter *sa = dev->data->dev_private;
 911         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 912         struct sfc_port *port = &sa->port;
 913         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
 914         int rc = 0;
 915
 916         sfc_adapter_lock(sa);
 917
 918         /*
 919          * Copy the address to the device private data so that
 920          * it could be recalled in the case of adapter restart.
 921          */
 922         ether_addr_copy(mac_addr, &port->default_mac_addr);
 923
 924         /*
 925          * Neither of the two following checks can return
 926          * an error. The new MAC address is preserved in
 927          * the device private data and can be activated
 928          * on the next port start if the user prevents
 929          * isolated mode from being enabled.
 930          */
 931         if (port->isolated) {
 932                 sfc_warn(sa, "isolated mode is active on the port");
 933                 sfc_warn(sa, "will not set MAC address");
 934                 goto unlock;
 935         }
 936
 937         if (sa->state != SFC_ADAPTER_STARTED) {
 938                 sfc_notice(sa, "the port is not started");
 939                 sfc_notice(sa, "the new MAC address will be set on port start");
 940
 941                 goto unlock;
 942         }
 943
 944         if (encp->enc_allow_set_mac_with_installed_filters) {
 945                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 946                 if (rc != 0) {
 947                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 948                         goto unlock;
 949                 }
 950
 951                 /*
 952                  * Changing the MAC address by means of MCDI request
 953                  * has no effect on received traffic, therefore
 954                  * we also need to update unicast filters
 955                  */
 956                 rc = sfc_set_rx_mode(sa);
 957                 if (rc != 0) {
 958                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 959                         /* Rollback the old address */
 960                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
 961                         (void)sfc_set_rx_mode(sa);
 962                 }
 963         } else {
 964                 sfc_warn(sa, "cannot set MAC address with filters installed");
 965                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 966                 sfc_warn(sa, "(some traffic may be dropped)");
 967
 968                 /*
 969                  * Since setting MAC address with filters installed is not
 970                  * allowed on the adapter, the new MAC address will be set
 971                  * by means of adapter restart. sfc_start() shall retrieve
 972                  * the new address from the device private data and set it.
 973                  */
 974                 sfc_stop(sa);
 975                 rc = sfc_start(sa);
 976                 if (rc != 0)
 977                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 978         }
 979
 980 unlock:
 981         if (rc != 0)
 982                 ether_addr_copy(old_addr, &port->default_mac_addr);
 983
 984         sfc_adapter_unlock(sa);
 985
 986         SFC_ASSERT(rc >= 0);
 987         return -rc;
 988 }
 989
 990
 991 static int
 992 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 993                      uint32_t nb_mc_addr)
 994 {
 995         struct sfc_adapter *sa = dev->data->dev_private;
 996         struct sfc_port *port = &sa->port;
 997         uint8_t *mc_addrs = port->mcast_addrs;
 998         int rc;
 999         unsigned int i;
1000
1001         if (port->isolated) {
1002                 sfc_err(sa, "isolated mode is active on the port");
1003                 sfc_err(sa, "will not set multicast address list");
1004                 return -ENOTSUP;
1005         }
1006
1007         if (mc_addrs == NULL)
1008                 return -ENOBUFS;
1009
1010         if (nb_mc_addr > port->max_mcast_addrs) {
1011                 sfc_err(sa, "too many multicast addresses: %u > %u",
1012                          nb_mc_addr, port->max_mcast_addrs);
1013                 return -EINVAL;
1014         }
1015
1016         for (i = 0; i < nb_mc_addr; ++i) {
1017                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1018                                  EFX_MAC_ADDR_LEN);
1019                 mc_addrs += EFX_MAC_ADDR_LEN;
1020         }
1021
1022         port->nb_mcast_addrs = nb_mc_addr;
1023
1024         if (sa->state != SFC_ADAPTER_STARTED)
1025                 return 0;
1026
1027         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1028                                         port->nb_mcast_addrs);
1029         if (rc != 0)
1030                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1031
1032         SFC_ASSERT(rc >= 0);
1033         return -rc;
1034 }
1035
1036 /*
1037  * The function is used by the secondary process as well. It must not
1038  * use any process-local pointers from the adapter data.
1039  */
1040 static void
1041 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1042                       struct rte_eth_rxq_info *qinfo)
1043 {
1044         struct sfc_adapter *sa = dev->data->dev_private;
1045         struct sfc_rxq_info *rxq_info;
1046         struct sfc_rxq *rxq;
1047
1048         sfc_adapter_lock(sa);
1049
1050         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1051
1052         rxq_info = &sa->rxq_info[rx_queue_id];
1053         rxq = rxq_info->rxq;
1054         SFC_ASSERT(rxq != NULL);
1055
1056         qinfo->mp = rxq->refill_mb_pool;
1057         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1058         qinfo->conf.rx_drop_en = 1;
1059         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1060         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1061         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1062                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1063                 qinfo->scattered_rx = 1;
1064         }
1065         qinfo->nb_desc = rxq_info->entries;
1066
1067         sfc_adapter_unlock(sa);
1068 }
1069
1070 /*
1071  * The function is used by the secondary process as well. It must not
1072  * use any process-local pointers from the adapter data.
1073  */
1074 static void
1075 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1076                       struct rte_eth_txq_info *qinfo)
1077 {
1078         struct sfc_adapter *sa = dev->data->dev_private;
1079         struct sfc_txq_info *txq_info;
1080
1081         sfc_adapter_lock(sa);
1082
1083         SFC_ASSERT(tx_queue_id < sa->txq_count);
1084
1085         txq_info = &sa->txq_info[tx_queue_id];
1086         SFC_ASSERT(txq_info->txq != NULL);
1087
1088         memset(qinfo, 0, sizeof(*qinfo));
1089
1090         qinfo->conf.offloads = txq_info->txq->offloads;
1091         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1092         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1093         qinfo->nb_desc = txq_info->entries;
1094
1095         sfc_adapter_unlock(sa);
1096 }
1097
1098 static uint32_t
1099 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1100 {
1101         struct sfc_adapter *sa = dev->data->dev_private;
1102
1103         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1104
1105         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1106 }
1107
1108 static int
1109 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1110 {
1111         struct sfc_dp_rxq *dp_rxq = queue;
1112
1113         return sfc_rx_qdesc_done(dp_rxq, offset);
1114 }
1115
1116 static int
1117 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1118 {
1119         struct sfc_dp_rxq *dp_rxq = queue;
1120         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1121
1122         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1123 }
1124
1125 static int
1126 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1127 {
1128         struct sfc_dp_txq *dp_txq = queue;
1129         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1130
1131         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1132 }
1133
1134 static int
1135 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1136 {
1137         struct sfc_adapter *sa = dev->data->dev_private;
1138         int rc;
1139
1140         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1141
1142         sfc_adapter_lock(sa);
1143
1144         rc = EINVAL;
1145         if (sa->state != SFC_ADAPTER_STARTED)
1146                 goto fail_not_started;
1147
1148         if (sa->rxq_info[rx_queue_id].rxq == NULL)
1149                 goto fail_not_setup;
1150
1151         rc = sfc_rx_qstart(sa, rx_queue_id);
1152         if (rc != 0)
1153                 goto fail_rx_qstart;
1154
1155         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1156
1157         sfc_adapter_unlock(sa);
1158
1159         return 0;
1160
1161 fail_rx_qstart:
1162 fail_not_setup:
1163 fail_not_started:
1164         sfc_adapter_unlock(sa);
1165         SFC_ASSERT(rc > 0);
1166         return -rc;
1167 }
1168
1169 static int
1170 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1171 {
1172         struct sfc_adapter *sa = dev->data->dev_private;
1173
1174         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1175
1176         sfc_adapter_lock(sa);
1177         sfc_rx_qstop(sa, rx_queue_id);
1178
1179         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1180
1181         sfc_adapter_unlock(sa);
1182
1183         return 0;
1184 }
1185
1186 static int
1187 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1188 {
1189         struct sfc_adapter *sa = dev->data->dev_private;
1190         int rc;
1191
1192         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1193
1194         sfc_adapter_lock(sa);
1195
1196         rc = EINVAL;
1197         if (sa->state != SFC_ADAPTER_STARTED)
1198                 goto fail_not_started;
1199
1200         rc = sfc_tx_qstart(sa, tx_queue_id);
1201         if (rc != 0)
1202                 goto fail_tx_qstart;
1203
1204         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1205
1206         sfc_adapter_unlock(sa);
1207         return 0;
1208
1209 fail_tx_qstart:
1210
1211 fail_not_started:
1212         sfc_adapter_unlock(sa);
1213         SFC_ASSERT(rc > 0);
1214         return -rc;
1215 }
1216
1217 static int
1218 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1219 {
1220         struct sfc_adapter *sa = dev->data->dev_private;
1221
1222         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1223
1224         sfc_adapter_lock(sa);
1225
1226         sfc_tx_qstop(sa, tx_queue_id);
1227
1228         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1229
1230         sfc_adapter_unlock(sa);
1231         return 0;
1232 }
1233
1234 static efx_tunnel_protocol_t
1235 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1236 {
1237         switch (rte_type) {
1238         case RTE_TUNNEL_TYPE_VXLAN:
1239                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1240         case RTE_TUNNEL_TYPE_GENEVE:
1241                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1242         default:
1243                 return EFX_TUNNEL_NPROTOS;
1244         }
1245 }
1246
1247 enum sfc_udp_tunnel_op_e {
1248         SFC_UDP_TUNNEL_ADD_PORT,
1249         SFC_UDP_TUNNEL_DEL_PORT,
1250 };
1251
1252 static int
1253 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1254                       struct rte_eth_udp_tunnel *tunnel_udp,
1255                       enum sfc_udp_tunnel_op_e op)
1256 {
1257         struct sfc_adapter *sa = dev->data->dev_private;
1258         efx_tunnel_protocol_t tunnel_proto;
1259         int rc;
1260
1261         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1262                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1263                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1264                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1265
1266         tunnel_proto =
1267                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1268         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1269                 rc = ENOTSUP;
1270                 goto fail_bad_proto;
1271         }
1272
1273         sfc_adapter_lock(sa);
1274
1275         switch (op) {
1276         case SFC_UDP_TUNNEL_ADD_PORT:
1277                 rc = efx_tunnel_config_udp_add(sa->nic,
1278                                                tunnel_udp->udp_port,
1279                                                tunnel_proto);
1280                 break;
1281         case SFC_UDP_TUNNEL_DEL_PORT:
1282                 rc = efx_tunnel_config_udp_remove(sa->nic,
1283                                                   tunnel_udp->udp_port,
1284                                                   tunnel_proto);
1285                 break;
1286         default:
1287                 rc = EINVAL;
1288                 goto fail_bad_op;
1289         }
1290
1291         if (rc != 0)
1292                 goto fail_op;
1293
1294         if (sa->state == SFC_ADAPTER_STARTED) {
1295                 rc = efx_tunnel_reconfigure(sa->nic);
1296                 if (rc == EAGAIN) {
1297                         /*
1298                          * Configuration is accepted by FW and MC reboot
1299                          * is initiated to apply the changes. MC reboot
1300                          * will be handled in a usual way (MC reboot
1301                          * event on management event queue and adapter
1302                          * restart).
1303                          */
1304                         rc = 0;
1305                 } else if (rc != 0) {
1306                         goto fail_reconfigure;
1307                 }
1308         }
1309
1310         sfc_adapter_unlock(sa);
1311         return 0;
1312
1313 fail_reconfigure:
1314         /* Remove/restore entry since the change makes the trouble */
1315         switch (op) {
1316         case SFC_UDP_TUNNEL_ADD_PORT:
1317                 (void)efx_tunnel_config_udp_remove(sa->nic,
1318                                                    tunnel_udp->udp_port,
1319                                                    tunnel_proto);
1320                 break;
1321         case SFC_UDP_TUNNEL_DEL_PORT:
1322                 (void)efx_tunnel_config_udp_add(sa->nic,
1323                                                 tunnel_udp->udp_port,
1324                                                 tunnel_proto);
1325                 break;
1326         }
1327
1328 fail_op:
1329 fail_bad_op:
1330         sfc_adapter_unlock(sa);
1331
1332 fail_bad_proto:
1333         SFC_ASSERT(rc > 0);
1334         return -rc;
1335 }
1336
1337 static int
1338 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1339                             struct rte_eth_udp_tunnel *tunnel_udp)
1340 {
1341         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1342 }
1343
1344 static int
1345 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1346                             struct rte_eth_udp_tunnel *tunnel_udp)
1347 {
1348         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1349 }
1350
1351 static int
1352 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1353                           struct rte_eth_rss_conf *rss_conf)
1354 {
1355         struct sfc_adapter *sa = dev->data->dev_private;
1356         struct sfc_rss *rss = &sa->rss;
1357         struct sfc_port *port = &sa->port;
1358
1359         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1360                 return -ENOTSUP;
1361
1362         if (rss->channels == 0)
1363                 return -EINVAL;
1364
1365         sfc_adapter_lock(sa);
1366
1367         /*
1368          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1369          * hence, conversion is done here to derive a correct set of ETH_RSS
1370          * flags which corresponds to the active EFX configuration stored
1371          * locally in 'sfc_adapter' and kept up-to-date
1372          */
1373         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1374         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1375         if (rss_conf->rss_key != NULL)
1376                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1377
1378         sfc_adapter_unlock(sa);
1379
1380         return 0;
1381 }
1382
1383 static int
1384 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1385                         struct rte_eth_rss_conf *rss_conf)
1386 {
1387         struct sfc_adapter *sa = dev->data->dev_private;
1388         struct sfc_rss *rss = &sa->rss;
1389         struct sfc_port *port = &sa->port;
1390         unsigned int efx_hash_types;
1391         int rc = 0;
1392
1393         if (port->isolated)
1394                 return -ENOTSUP;
1395
1396         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1397                 sfc_err(sa, "RSS is not available");
1398                 return -ENOTSUP;
1399         }
1400
1401         if (rss->channels == 0) {
1402                 sfc_err(sa, "RSS is not configured");
1403                 return -EINVAL;
1404         }
1405
1406         if ((rss_conf->rss_key != NULL) &&
1407             (rss_conf->rss_key_len != sizeof(rss->key))) {
1408                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1409                         sizeof(rss->key));
1410                 return -EINVAL;
1411         }
1412
1413         sfc_adapter_lock(sa);
1414
1415         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1416         if (rc != 0)
1417                 goto fail_rx_hf_rte_to_efx;
1418
1419         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1420                                    rss->hash_alg, efx_hash_types, B_TRUE);
1421         if (rc != 0)
1422                 goto fail_scale_mode_set;
1423
1424         if (rss_conf->rss_key != NULL) {
1425                 if (sa->state == SFC_ADAPTER_STARTED) {
1426                         rc = efx_rx_scale_key_set(sa->nic,
1427                                                   EFX_RSS_CONTEXT_DEFAULT,
1428                                                   rss_conf->rss_key,
1429                                                   sizeof(rss->key));
1430                         if (rc != 0)
1431                                 goto fail_scale_key_set;
1432                 }
1433
1434                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1435         }
1436
1437         rss->hash_types = efx_hash_types;
1438
1439         sfc_adapter_unlock(sa);
1440
1441         return 0;
1442
1443 fail_scale_key_set:
1444         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1445                                   EFX_RX_HASHALG_TOEPLITZ,
1446                                   rss->hash_types, B_TRUE) != 0)
1447                 sfc_err(sa, "failed to restore RSS mode");
1448
1449 fail_scale_mode_set:
1450 fail_rx_hf_rte_to_efx:
1451         sfc_adapter_unlock(sa);
1452         return -rc;
1453 }
1454
1455 static int
1456 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1457                        struct rte_eth_rss_reta_entry64 *reta_conf,
1458                        uint16_t reta_size)
1459 {
1460         struct sfc_adapter *sa = dev->data->dev_private;
1461         struct sfc_rss *rss = &sa->rss;
1462         struct sfc_port *port = &sa->port;
1463         int entry;
1464
1465         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1466                 return -ENOTSUP;
1467
1468         if (rss->channels == 0)
1469                 return -EINVAL;
1470
1471         if (reta_size != EFX_RSS_TBL_SIZE)
1472                 return -EINVAL;
1473
1474         sfc_adapter_lock(sa);
1475
1476         for (entry = 0; entry < reta_size; entry++) {
1477                 int grp = entry / RTE_RETA_GROUP_SIZE;
1478                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1479
1480                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1481                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1482         }
1483
1484         sfc_adapter_unlock(sa);
1485
1486         return 0;
1487 }
1488
1489 static int
1490 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1491                         struct rte_eth_rss_reta_entry64 *reta_conf,
1492                         uint16_t reta_size)
1493 {
1494         struct sfc_adapter *sa = dev->data->dev_private;
1495         struct sfc_rss *rss = &sa->rss;
1496         struct sfc_port *port = &sa->port;
1497         unsigned int *rss_tbl_new;
1498         uint16_t entry;
1499         int rc = 0;
1500
1501
1502         if (port->isolated)
1503                 return -ENOTSUP;
1504
1505         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1506                 sfc_err(sa, "RSS is not available");
1507                 return -ENOTSUP;
1508         }
1509
1510         if (rss->channels == 0) {
1511                 sfc_err(sa, "RSS is not configured");
1512                 return -EINVAL;
1513         }
1514
1515         if (reta_size != EFX_RSS_TBL_SIZE) {
1516                 sfc_err(sa, "RETA size is wrong (should be %u)",
1517                         EFX_RSS_TBL_SIZE);
1518                 return -EINVAL;
1519         }
1520
1521         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1522         if (rss_tbl_new == NULL)
1523                 return -ENOMEM;
1524
1525         sfc_adapter_lock(sa);
1526
1527         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1528
1529         for (entry = 0; entry < reta_size; entry++) {
1530                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1531                 struct rte_eth_rss_reta_entry64 *grp;
1532
1533                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1534
1535                 if (grp->mask & (1ull << grp_idx)) {
1536                         if (grp->reta[grp_idx] >= rss->channels) {
1537                                 rc = EINVAL;
1538                                 goto bad_reta_entry;
1539                         }
1540                         rss_tbl_new[entry] = grp->reta[grp_idx];
1541                 }
1542         }
1543
1544         if (sa->state == SFC_ADAPTER_STARTED) {
1545                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1546                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1547                 if (rc != 0)
1548                         goto fail_scale_tbl_set;
1549         }
1550
1551         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1552
1553 fail_scale_tbl_set:
1554 bad_reta_entry:
1555         sfc_adapter_unlock(sa);
1556
1557         rte_free(rss_tbl_new);
1558
1559         SFC_ASSERT(rc >= 0);
1560         return -rc;
1561 }
1562
1563 static int
1564 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1565                     enum rte_filter_op filter_op,
1566                     void *arg)
1567 {
1568         struct sfc_adapter *sa = dev->data->dev_private;
1569         int rc = ENOTSUP;
1570
1571         sfc_log_init(sa, "entry");
1572
1573         switch (filter_type) {
1574         case RTE_ETH_FILTER_NONE:
1575                 sfc_err(sa, "Global filters configuration not supported");
1576                 break;
1577         case RTE_ETH_FILTER_MACVLAN:
1578                 sfc_err(sa, "MACVLAN filters not supported");
1579                 break;
1580         case RTE_ETH_FILTER_ETHERTYPE:
1581                 sfc_err(sa, "EtherType filters not supported");
1582                 break;
1583         case RTE_ETH_FILTER_FLEXIBLE:
1584                 sfc_err(sa, "Flexible filters not supported");
1585                 break;
1586         case RTE_ETH_FILTER_SYN:
1587                 sfc_err(sa, "SYN filters not supported");
1588                 break;
1589         case RTE_ETH_FILTER_NTUPLE:
1590                 sfc_err(sa, "NTUPLE filters not supported");
1591                 break;
1592         case RTE_ETH_FILTER_TUNNEL:
1593                 sfc_err(sa, "Tunnel filters not supported");
1594                 break;
1595         case RTE_ETH_FILTER_FDIR:
1596                 sfc_err(sa, "Flow Director filters not supported");
1597                 break;
1598         case RTE_ETH_FILTER_HASH:
1599                 sfc_err(sa, "Hash filters not supported");
1600                 break;
1601         case RTE_ETH_FILTER_GENERIC:
1602                 if (filter_op != RTE_ETH_FILTER_GET) {
1603                         rc = EINVAL;
1604                 } else {
1605                         *(const void **)arg = &sfc_flow_ops;
1606                         rc = 0;
1607                 }
1608                 break;
1609         default:
1610                 sfc_err(sa, "Unknown filter type %u", filter_type);
1611                 break;
1612         }
1613
1614         sfc_log_init(sa, "exit: %d", -rc);
1615         SFC_ASSERT(rc >= 0);
1616         return -rc;
1617 }
1618
1619 static int
1620 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1621 {
1622         struct sfc_adapter *sa = dev->data->dev_private;
1623
1624         /*
1625          * If Rx datapath does not provide callback to check mempool,
1626          * all pools are supported.
1627          */
1628         if (sa->dp_rx->pool_ops_supported == NULL)
1629                 return 1;
1630
1631         return sa->dp_rx->pool_ops_supported(pool);
1632 }
1633
1634 static const struct eth_dev_ops sfc_eth_dev_ops = {
1635         .dev_configure                  = sfc_dev_configure,
1636         .dev_start                      = sfc_dev_start,
1637         .dev_stop                       = sfc_dev_stop,
1638         .dev_set_link_up                = sfc_dev_set_link_up,
1639         .dev_set_link_down              = sfc_dev_set_link_down,
1640         .dev_close                      = sfc_dev_close,
1641         .promiscuous_enable             = sfc_dev_promisc_enable,
1642         .promiscuous_disable            = sfc_dev_promisc_disable,
1643         .allmulticast_enable            = sfc_dev_allmulti_enable,
1644         .allmulticast_disable           = sfc_dev_allmulti_disable,
1645         .link_update                    = sfc_dev_link_update,
1646         .stats_get                      = sfc_stats_get,
1647         .stats_reset                    = sfc_stats_reset,
1648         .xstats_get                     = sfc_xstats_get,
1649         .xstats_reset                   = sfc_stats_reset,
1650         .xstats_get_names               = sfc_xstats_get_names,
1651         .dev_infos_get                  = sfc_dev_infos_get,
1652         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1653         .mtu_set                        = sfc_dev_set_mtu,
1654         .rx_queue_start                 = sfc_rx_queue_start,
1655         .rx_queue_stop                  = sfc_rx_queue_stop,
1656         .tx_queue_start                 = sfc_tx_queue_start,
1657         .tx_queue_stop                  = sfc_tx_queue_stop,
1658         .rx_queue_setup                 = sfc_rx_queue_setup,
1659         .rx_queue_release               = sfc_rx_queue_release,
1660         .rx_queue_count                 = sfc_rx_queue_count,
1661         .rx_descriptor_done             = sfc_rx_descriptor_done,
1662         .rx_descriptor_status           = sfc_rx_descriptor_status,
1663         .tx_descriptor_status           = sfc_tx_descriptor_status,
1664         .tx_queue_setup                 = sfc_tx_queue_setup,
1665         .tx_queue_release               = sfc_tx_queue_release,
1666         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1667         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1668         .mac_addr_set                   = sfc_mac_addr_set,
1669         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1670         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1671         .reta_update                    = sfc_dev_rss_reta_update,
1672         .reta_query                     = sfc_dev_rss_reta_query,
1673         .rss_hash_update                = sfc_dev_rss_hash_update,
1674         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1675         .filter_ctrl                    = sfc_dev_filter_ctrl,
1676         .set_mc_addr_list               = sfc_set_mc_addr_list,
1677         .rxq_info_get                   = sfc_rx_queue_info_get,
1678         .txq_info_get                   = sfc_tx_queue_info_get,
1679         .fw_version_get                 = sfc_fw_version_get,
1680         .xstats_get_by_id               = sfc_xstats_get_by_id,
1681         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1682         .pool_ops_supported             = sfc_pool_ops_supported,
1683 };
1684
1685 /**
1686  * Duplicate a string in potentially shared memory required for
1687  * multi-process support.
1688  *
1689  * strdup() allocates from process-local heap/memory.
1690  */
1691 static char *
1692 sfc_strdup(const char *str)
1693 {
1694         size_t size;
1695         char *copy;
1696
1697         if (str == NULL)
1698                 return NULL;
1699
1700         size = strlen(str) + 1;
1701         copy = rte_malloc(__func__, size, 0);
1702         if (copy != NULL)
1703                 rte_memcpy(copy, str, size);
1704
1705         return copy;
1706 }
1707
1708 static int
1709 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1710 {
1711         struct sfc_adapter *sa = dev->data->dev_private;
1712         const efx_nic_cfg_t *encp;
1713         unsigned int avail_caps = 0;
1714         const char *rx_name = NULL;
1715         const char *tx_name = NULL;
1716         int rc;
1717
1718         switch (sa->family) {
1719         case EFX_FAMILY_HUNTINGTON:
1720         case EFX_FAMILY_MEDFORD:
1721         case EFX_FAMILY_MEDFORD2:
1722                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1723                 break;
1724         default:
1725                 break;
1726         }
1727
1728         encp = efx_nic_cfg_get(sa->nic);
1729         if (encp->enc_rx_es_super_buffer_supported)
1730                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1731
1732         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1733                                 sfc_kvarg_string_handler, &rx_name);
1734         if (rc != 0)
1735                 goto fail_kvarg_rx_datapath;
1736
1737         if (rx_name != NULL) {
1738                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1739                 if (sa->dp_rx == NULL) {
1740                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1741                         rc = ENOENT;
1742                         goto fail_dp_rx;
1743                 }
1744                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1745                         sfc_err(sa,
1746                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1747                                 rx_name);
1748                         rc = EINVAL;
1749                         goto fail_dp_rx_caps;
1750                 }
1751         } else {
1752                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1753                 if (sa->dp_rx == NULL) {
1754                         sfc_err(sa, "Rx datapath by caps %#x not found",
1755                                 avail_caps);
1756                         rc = ENOENT;
1757                         goto fail_dp_rx;
1758                 }
1759         }
1760
1761         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1762         if (sa->dp_rx_name == NULL) {
1763                 rc = ENOMEM;
1764                 goto fail_dp_rx_name;
1765         }
1766
1767         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1768
1769         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1770
1771         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1772                                 sfc_kvarg_string_handler, &tx_name);
1773         if (rc != 0)
1774                 goto fail_kvarg_tx_datapath;
1775
1776         if (tx_name != NULL) {
1777                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1778                 if (sa->dp_tx == NULL) {
1779                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1780                         rc = ENOENT;
1781                         goto fail_dp_tx;
1782                 }
1783                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1784                         sfc_err(sa,
1785                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1786                                 tx_name);
1787                         rc = EINVAL;
1788                         goto fail_dp_tx_caps;
1789                 }
1790         } else {
1791                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1792                 if (sa->dp_tx == NULL) {
1793                         sfc_err(sa, "Tx datapath by caps %#x not found",
1794                                 avail_caps);
1795                         rc = ENOENT;
1796                         goto fail_dp_tx;
1797                 }
1798         }
1799
1800         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1801         if (sa->dp_tx_name == NULL) {
1802                 rc = ENOMEM;
1803                 goto fail_dp_tx_name;
1804         }
1805
1806         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1807
1808         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1809
1810         dev->dev_ops = &sfc_eth_dev_ops;
1811
1812         return 0;
1813
1814 fail_dp_tx_name:
1815 fail_dp_tx_caps:
1816         sa->dp_tx = NULL;
1817
1818 fail_dp_tx:
1819 fail_kvarg_tx_datapath:
1820         rte_free(sa->dp_rx_name);
1821         sa->dp_rx_name = NULL;
1822
1823 fail_dp_rx_name:
1824 fail_dp_rx_caps:
1825         sa->dp_rx = NULL;
1826
1827 fail_dp_rx:
1828 fail_kvarg_rx_datapath:
1829         return rc;
1830 }
1831
1832 static void
1833 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1834 {
1835         struct sfc_adapter *sa = dev->data->dev_private;
1836
1837         dev->dev_ops = NULL;
1838         dev->rx_pkt_burst = NULL;
1839         dev->tx_pkt_burst = NULL;
1840
1841         rte_free(sa->dp_tx_name);
1842         sa->dp_tx_name = NULL;
1843         sa->dp_tx = NULL;
1844
1845         rte_free(sa->dp_rx_name);
1846         sa->dp_rx_name = NULL;
1847         sa->dp_rx = NULL;
1848 }
1849
1850 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1851         .rxq_info_get                   = sfc_rx_queue_info_get,
1852         .txq_info_get                   = sfc_tx_queue_info_get,
1853 };
1854
1855 static int
1856 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1857 {
1858         /*
1859          * Device private data has really many process-local pointers.
1860          * Below code should be extremely careful to use data located
1861          * in shared memory only.
1862          */
1863         struct sfc_adapter *sa = dev->data->dev_private;
1864         const struct sfc_dp_rx *dp_rx;
1865         const struct sfc_dp_tx *dp_tx;
1866         int rc;
1867
1868         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1869         if (dp_rx == NULL) {
1870                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1871                 rc = ENOENT;
1872                 goto fail_dp_rx;
1873         }
1874         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1875                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1876                         sa->dp_tx_name);
1877                 rc = EINVAL;
1878                 goto fail_dp_rx_multi_process;
1879         }
1880
1881         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1882         if (dp_tx == NULL) {
1883                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1884                 rc = ENOENT;
1885                 goto fail_dp_tx;
1886         }
1887         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1888                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1889                         sa->dp_tx_name);
1890                 rc = EINVAL;
1891                 goto fail_dp_tx_multi_process;
1892         }
1893
1894         dev->rx_pkt_burst = dp_rx->pkt_burst;
1895         dev->tx_pkt_burst = dp_tx->pkt_burst;
1896         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1897
1898         return 0;
1899
1900 fail_dp_tx_multi_process:
1901 fail_dp_tx:
1902 fail_dp_rx_multi_process:
1903 fail_dp_rx:
1904         return rc;
1905 }
1906
1907 static void
1908 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1909 {
1910         dev->dev_ops = NULL;
1911         dev->tx_pkt_burst = NULL;
1912         dev->rx_pkt_burst = NULL;
1913 }
1914
1915 static void
1916 sfc_register_dp(void)
1917 {
1918         /* Register once */
1919         if (TAILQ_EMPTY(&sfc_dp_head)) {
1920                 /* Prefer EF10 datapath */
1921                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1922                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1923                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1924
1925                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1926                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1927                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1928         }
1929 }
1930
1931 static int
1932 sfc_eth_dev_init(struct rte_eth_dev *dev)
1933 {
1934         struct sfc_adapter *sa = dev->data->dev_private;
1935         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1936         int rc;
1937         const efx_nic_cfg_t *encp;
1938         const struct ether_addr *from;
1939
1940         sfc_register_dp();
1941
1942         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1943                 return -sfc_eth_dev_secondary_set_ops(dev);
1944
1945         /* Required for logging */
1946         sa->pci_addr = pci_dev->addr;
1947         sa->port_id = dev->data->port_id;
1948
1949         sa->eth_dev = dev;
1950
1951         /* Copy PCI device info to the dev->data */
1952         rte_eth_copy_pci_info(dev, pci_dev);
1953
1954         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1955                                                 RTE_LOG_NOTICE);
1956
1957         rc = sfc_kvargs_parse(sa);
1958         if (rc != 0)
1959                 goto fail_kvargs_parse;
1960
1961         sfc_log_init(sa, "entry");
1962
1963         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1964         if (dev->data->mac_addrs == NULL) {
1965                 rc = ENOMEM;
1966                 goto fail_mac_addrs;
1967         }
1968
1969         sfc_adapter_lock_init(sa);
1970         sfc_adapter_lock(sa);
1971
1972         sfc_log_init(sa, "probing");
1973         rc = sfc_probe(sa);
1974         if (rc != 0)
1975                 goto fail_probe;
1976
1977         sfc_log_init(sa, "set device ops");
1978         rc = sfc_eth_dev_set_ops(dev);
1979         if (rc != 0)
1980                 goto fail_set_ops;
1981
1982         sfc_log_init(sa, "attaching");
1983         rc = sfc_attach(sa);
1984         if (rc != 0)
1985                 goto fail_attach;
1986
1987         encp = efx_nic_cfg_get(sa->nic);
1988
1989         /*
1990          * The arguments are really reverse order in comparison to
1991          * Linux kernel. Copy from NIC config to Ethernet device data.
1992          */
1993         from = (const struct ether_addr *)(encp->enc_mac_addr);
1994         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1995
1996         sfc_adapter_unlock(sa);
1997
1998         sfc_log_init(sa, "done");
1999         return 0;
2000
2001 fail_attach:
2002         sfc_eth_dev_clear_ops(dev);
2003
2004 fail_set_ops:
2005         sfc_unprobe(sa);
2006
2007 fail_probe:
2008         sfc_adapter_unlock(sa);
2009         sfc_adapter_lock_fini(sa);
2010         rte_free(dev->data->mac_addrs);
2011         dev->data->mac_addrs = NULL;
2012
2013 fail_mac_addrs:
2014         sfc_kvargs_cleanup(sa);
2015
2016 fail_kvargs_parse:
2017         sfc_log_init(sa, "failed %d", rc);
2018         SFC_ASSERT(rc > 0);
2019         return -rc;
2020 }
2021
2022 static int
2023 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2024 {
2025         struct sfc_adapter *sa;
2026
2027         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2028                 sfc_eth_dev_secondary_clear_ops(dev);
2029                 return 0;
2030         }
2031
2032         sa = dev->data->dev_private;
2033         sfc_log_init(sa, "entry");
2034
2035         sfc_adapter_lock(sa);
2036
2037         sfc_eth_dev_clear_ops(dev);
2038
2039         sfc_detach(sa);
2040         sfc_unprobe(sa);
2041
2042         rte_free(dev->data->mac_addrs);
2043         dev->data->mac_addrs = NULL;
2044
2045         sfc_kvargs_cleanup(sa);
2046
2047         sfc_adapter_unlock(sa);
2048         sfc_adapter_lock_fini(sa);
2049
2050         sfc_log_init(sa, "done");
2051
2052         /* Required for logging, so cleanup last */
2053         sa->eth_dev = NULL;
2054         return 0;
2055 }
2056
2057 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2058         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2059         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2060         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2061         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2062         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2063         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2064         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2065         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2066         { .vendor_id = 0 /* sentinel */ }
2067 };
2068
2069 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2070         struct rte_pci_device *pci_dev)
2071 {
2072         return rte_eth_dev_pci_generic_probe(pci_dev,
2073                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2074 }
2075
2076 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2077 {
2078         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2079 }
2080
2081 static struct rte_pci_driver sfc_efx_pmd = {
2082         .id_table = pci_id_sfc_efx_map,
2083         .drv_flags =
2084                 RTE_PCI_DRV_INTR_LSC |
2085                 RTE_PCI_DRV_NEED_MAPPING,
2086         .probe = sfc_eth_dev_pci_probe,
2087         .remove = sfc_eth_dev_pci_remove,
2088 };
2089
2090 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2091 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2092 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2093 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2094         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2095         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2096         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2097         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2098         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2099         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2100
2101 RTE_INIT(sfc_driver_register_logtype)
2102 {
2103         int ret;
2104
2105         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2106                                                    RTE_LOG_NOTICE);
2107         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2108 }