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
 176 static const uint32_t *
 177 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 178 {
 179         struct sfc_adapter *sa = dev->data->dev_private;
 180         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 181         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 182
 183         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 184 }
 185
 186 static int
 187 sfc_dev_configure(struct rte_eth_dev *dev)
 188 {
 189         struct rte_eth_dev_data *dev_data = dev->data;
 190         struct sfc_adapter *sa = dev_data->dev_private;
 191         int rc;
 192
 193         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 194                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 195
 196         sfc_adapter_lock(sa);
 197         switch (sa->state) {
 198         case SFC_ADAPTER_CONFIGURED:
 199                 /* FALLTHROUGH */
 200         case SFC_ADAPTER_INITIALIZED:
 201                 rc = sfc_configure(sa);
 202                 break;
 203         default:
 204                 sfc_err(sa, "unexpected adapter state %u to configure",
 205                         sa->state);
 206                 rc = EINVAL;
 207                 break;
 208         }
 209         sfc_adapter_unlock(sa);
 210
 211         sfc_log_init(sa, "done %d", rc);
 212         SFC_ASSERT(rc >= 0);
 213         return -rc;
 214 }
 215
 216 static int
 217 sfc_dev_start(struct rte_eth_dev *dev)
 218 {
 219         struct sfc_adapter *sa = dev->data->dev_private;
 220         int rc;
 221
 222         sfc_log_init(sa, "entry");
 223
 224         sfc_adapter_lock(sa);
 225         rc = sfc_start(sa);
 226         sfc_adapter_unlock(sa);
 227
 228         sfc_log_init(sa, "done %d", rc);
 229         SFC_ASSERT(rc >= 0);
 230         return -rc;
 231 }
 232
 233 static int
 234 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 235 {
 236         struct sfc_adapter *sa = dev->data->dev_private;
 237         struct rte_eth_link current_link;
 238         int ret;
 239
 240         sfc_log_init(sa, "entry");
 241
 242         if (sa->state != SFC_ADAPTER_STARTED) {
 243                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 244         } else if (wait_to_complete) {
 245                 efx_link_mode_t link_mode;
 246
 247                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 248                         link_mode = EFX_LINK_UNKNOWN;
 249                 sfc_port_link_mode_to_info(link_mode, &current_link);
 250
 251         } else {
 252                 sfc_ev_mgmt_qpoll(sa);
 253                 rte_eth_linkstatus_get(dev, &current_link);
 254         }
 255
 256         ret = rte_eth_linkstatus_set(dev, &current_link);
 257         if (ret == 0)
 258                 sfc_notice(sa, "Link status is %s",
 259                            current_link.link_status ? "UP" : "DOWN");
 260
 261         return ret;
 262 }
 263
 264 static void
 265 sfc_dev_stop(struct rte_eth_dev *dev)
 266 {
 267         struct sfc_adapter *sa = dev->data->dev_private;
 268
 269         sfc_log_init(sa, "entry");
 270
 271         sfc_adapter_lock(sa);
 272         sfc_stop(sa);
 273         sfc_adapter_unlock(sa);
 274
 275         sfc_log_init(sa, "done");
 276 }
 277
 278 static int
 279 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 280 {
 281         struct sfc_adapter *sa = dev->data->dev_private;
 282         int rc;
 283
 284         sfc_log_init(sa, "entry");
 285
 286         sfc_adapter_lock(sa);
 287         rc = sfc_start(sa);
 288         sfc_adapter_unlock(sa);
 289
 290         SFC_ASSERT(rc >= 0);
 291         return -rc;
 292 }
 293
 294 static int
 295 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 296 {
 297         struct sfc_adapter *sa = dev->data->dev_private;
 298
 299         sfc_log_init(sa, "entry");
 300
 301         sfc_adapter_lock(sa);
 302         sfc_stop(sa);
 303         sfc_adapter_unlock(sa);
 304
 305         return 0;
 306 }
 307
 308 static void
 309 sfc_dev_close(struct rte_eth_dev *dev)
 310 {
 311         struct sfc_adapter *sa = dev->data->dev_private;
 312
 313         sfc_log_init(sa, "entry");
 314
 315         sfc_adapter_lock(sa);
 316         switch (sa->state) {
 317         case SFC_ADAPTER_STARTED:
 318                 sfc_stop(sa);
 319                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 320                 /* FALLTHROUGH */
 321         case SFC_ADAPTER_CONFIGURED:
 322                 sfc_close(sa);
 323                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 324                 /* FALLTHROUGH */
 325         case SFC_ADAPTER_INITIALIZED:
 326                 break;
 327         default:
 328                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 329                 break;
 330         }
 331         sfc_adapter_unlock(sa);
 332
 333         sfc_log_init(sa, "done");
 334 }
 335
 336 static void
 337 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 338                    boolean_t enabled)
 339 {
 340         struct sfc_port *port;
 341         boolean_t *toggle;
 342         struct sfc_adapter *sa = dev->data->dev_private;
 343         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 344         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 345
 346         sfc_adapter_lock(sa);
 347
 348         port = &sa->port;
 349         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 350
 351         if (*toggle != enabled) {
 352                 *toggle = enabled;
 353
 354                 if (port->isolated) {
 355                         sfc_warn(sa, "isolated mode is active on the port");
 356                         sfc_warn(sa, "the change is to be applied on the next "
 357                                      "start provided that isolated mode is "
 358                                      "disabled prior the next start");
 359                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 360                            (sfc_set_rx_mode(sa) != 0)) {
 361                         *toggle = !(enabled);
 362                         sfc_warn(sa, "Failed to %s %s mode",
 363                                  ((enabled) ? "enable" : "disable"), desc);
 364                 }
 365         }
 366
 367         sfc_adapter_unlock(sa);
 368 }
 369
 370 static void
 371 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 372 {
 373         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 374 }
 375
 376 static void
 377 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 378 {
 379         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 380 }
 381
 382 static void
 383 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 384 {
 385         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 386 }
 387
 388 static void
 389 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 390 {
 391         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 392 }
 393
 394 static int
 395 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 396                    uint16_t nb_rx_desc, unsigned int socket_id,
 397                    const struct rte_eth_rxconf *rx_conf,
 398                    struct rte_mempool *mb_pool)
 399 {
 400         struct sfc_adapter *sa = dev->data->dev_private;
 401         int rc;
 402
 403         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 404                      rx_queue_id, nb_rx_desc, socket_id);
 405
 406         sfc_adapter_lock(sa);
 407
 408         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 409                           rx_conf, mb_pool);
 410         if (rc != 0)
 411                 goto fail_rx_qinit;
 412
 413         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 414
 415         sfc_adapter_unlock(sa);
 416
 417         return 0;
 418
 419 fail_rx_qinit:
 420         sfc_adapter_unlock(sa);
 421         SFC_ASSERT(rc > 0);
 422         return -rc;
 423 }
 424
 425 static void
 426 sfc_rx_queue_release(void *queue)
 427 {
 428         struct sfc_dp_rxq *dp_rxq = queue;
 429         struct sfc_rxq *rxq;
 430         struct sfc_adapter *sa;
 431         unsigned int sw_index;
 432
 433         if (dp_rxq == NULL)
 434                 return;
 435
 436         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 437         sa = rxq->evq->sa;
 438         sfc_adapter_lock(sa);
 439
 440         sw_index = sfc_rxq_sw_index(rxq);
 441
 442         sfc_log_init(sa, "RxQ=%u", sw_index);
 443
 444         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 445
 446         sfc_rx_qfini(sa, sw_index);
 447
 448         sfc_adapter_unlock(sa);
 449 }
 450
 451 static int
 452 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 453                    uint16_t nb_tx_desc, unsigned int socket_id,
 454                    const struct rte_eth_txconf *tx_conf)
 455 {
 456         struct sfc_adapter *sa = dev->data->dev_private;
 457         int rc;
 458
 459         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 460                      tx_queue_id, nb_tx_desc, socket_id);
 461
 462         sfc_adapter_lock(sa);
 463
 464         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 465         if (rc != 0)
 466                 goto fail_tx_qinit;
 467
 468         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 469
 470         sfc_adapter_unlock(sa);
 471         return 0;
 472
 473 fail_tx_qinit:
 474         sfc_adapter_unlock(sa);
 475         SFC_ASSERT(rc > 0);
 476         return -rc;
 477 }
 478
 479 static void
 480 sfc_tx_queue_release(void *queue)
 481 {
 482         struct sfc_dp_txq *dp_txq = queue;
 483         struct sfc_txq *txq;
 484         unsigned int sw_index;
 485         struct sfc_adapter *sa;
 486
 487         if (dp_txq == NULL)
 488                 return;
 489
 490         txq = sfc_txq_by_dp_txq(dp_txq);
 491         sw_index = sfc_txq_sw_index(txq);
 492
 493         SFC_ASSERT(txq->evq != NULL);
 494         sa = txq->evq->sa;
 495
 496         sfc_log_init(sa, "TxQ = %u", sw_index);
 497
 498         sfc_adapter_lock(sa);
 499
 500         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 501         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 502
 503         sfc_tx_qfini(sa, sw_index);
 504
 505         sfc_adapter_unlock(sa);
 506 }
 507
 508 static int
 509 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 510 {
 511         struct sfc_adapter *sa = dev->data->dev_private;
 512         struct sfc_port *port = &sa->port;
 513         uint64_t *mac_stats;
 514         int ret;
 515
 516         rte_spinlock_lock(&port->mac_stats_lock);
 517
 518         ret = sfc_port_update_mac_stats(sa);
 519         if (ret != 0)
 520                 goto unlock;
 521
 522         mac_stats = port->mac_stats_buf;
 523
 524         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 525                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 526                 stats->ipackets =
 527                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 528                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 529                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 530                 stats->opackets =
 531                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 532                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 533                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 534                 stats->ibytes =
 535                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 536                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 537                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 538                 stats->obytes =
 539                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 540                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 541                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 542                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 543                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 544                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 545         } else {
 546                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 547                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 548                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 549                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 550                 /*
 551                  * Take into account stats which are whenever supported
 552                  * on EF10. If some stat is not supported by current
 553                  * firmware variant or HW revision, it is guaranteed
 554                  * to be zero in mac_stats.
 555                  */
 556                 stats->imissed =
 557                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 558                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 559                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 560                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 561                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 562                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 563                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 564                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 565                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 566                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 567                 stats->ierrors =
 568                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 569                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 570                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 571                 /* no oerrors counters supported on EF10 */
 572         }
 573
 574 unlock:
 575         rte_spinlock_unlock(&port->mac_stats_lock);
 576         SFC_ASSERT(ret >= 0);
 577         return -ret;
 578 }
 579
 580 static void
 581 sfc_stats_reset(struct rte_eth_dev *dev)
 582 {
 583         struct sfc_adapter *sa = dev->data->dev_private;
 584         struct sfc_port *port = &sa->port;
 585         int rc;
 586
 587         if (sa->state != SFC_ADAPTER_STARTED) {
 588                 /*
 589                  * The operation cannot be done if port is not started; it
 590                  * will be scheduled to be done during the next port start
 591                  */
 592                 port->mac_stats_reset_pending = B_TRUE;
 593                 return;
 594         }
 595
 596         rc = sfc_port_reset_mac_stats(sa);
 597         if (rc != 0)
 598                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 599 }
 600
 601 static int
 602 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 603                unsigned int xstats_count)
 604 {
 605         struct sfc_adapter *sa = dev->data->dev_private;
 606         struct sfc_port *port = &sa->port;
 607         uint64_t *mac_stats;
 608         int rc;
 609         unsigned int i;
 610         int nstats = 0;
 611
 612         rte_spinlock_lock(&port->mac_stats_lock);
 613
 614         rc = sfc_port_update_mac_stats(sa);
 615         if (rc != 0) {
 616                 SFC_ASSERT(rc > 0);
 617                 nstats = -rc;
 618                 goto unlock;
 619         }
 620
 621         mac_stats = port->mac_stats_buf;
 622
 623         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 624                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 625                         if (xstats != NULL && nstats < (int)xstats_count) {
 626                                 xstats[nstats].id = nstats;
 627                                 xstats[nstats].value = mac_stats[i];
 628                         }
 629                         nstats++;
 630                 }
 631         }
 632
 633 unlock:
 634         rte_spinlock_unlock(&port->mac_stats_lock);
 635
 636         return nstats;
 637 }
 638
 639 static int
 640 sfc_xstats_get_names(struct rte_eth_dev *dev,
 641                      struct rte_eth_xstat_name *xstats_names,
 642                      unsigned int xstats_count)
 643 {
 644         struct sfc_adapter *sa = dev->data->dev_private;
 645         struct sfc_port *port = &sa->port;
 646         unsigned int i;
 647         unsigned int nstats = 0;
 648
 649         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 650                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 651                         if (xstats_names != NULL && nstats < xstats_count)
 652                                 strlcpy(xstats_names[nstats].name,
 653                                         efx_mac_stat_name(sa->nic, i),
 654                                         sizeof(xstats_names[0].name));
 655                         nstats++;
 656                 }
 657         }
 658
 659         return nstats;
 660 }
 661
 662 static int
 663 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 664                      uint64_t *values, unsigned int n)
 665 {
 666         struct sfc_adapter *sa = dev->data->dev_private;
 667         struct sfc_port *port = &sa->port;
 668         uint64_t *mac_stats;
 669         unsigned int nb_supported = 0;
 670         unsigned int nb_written = 0;
 671         unsigned int i;
 672         int ret;
 673         int rc;
 674
 675         if (unlikely(values == NULL) ||
 676             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 677                 return port->mac_stats_nb_supported;
 678
 679         rte_spinlock_lock(&port->mac_stats_lock);
 680
 681         rc = sfc_port_update_mac_stats(sa);
 682         if (rc != 0) {
 683                 SFC_ASSERT(rc > 0);
 684                 ret = -rc;
 685                 goto unlock;
 686         }
 687
 688         mac_stats = port->mac_stats_buf;
 689
 690         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 691                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 692                         continue;
 693
 694                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 695                         values[nb_written++] = mac_stats[i];
 696
 697                 ++nb_supported;
 698         }
 699
 700         ret = nb_written;
 701
 702 unlock:
 703         rte_spinlock_unlock(&port->mac_stats_lock);
 704
 705         return ret;
 706 }
 707
 708 static int
 709 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 710                            struct rte_eth_xstat_name *xstats_names,
 711                            const uint64_t *ids, unsigned int size)
 712 {
 713         struct sfc_adapter *sa = dev->data->dev_private;
 714         struct sfc_port *port = &sa->port;
 715         unsigned int nb_supported = 0;
 716         unsigned int nb_written = 0;
 717         unsigned int i;
 718
 719         if (unlikely(xstats_names == NULL) ||
 720             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 721                 return port->mac_stats_nb_supported;
 722
 723         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 724                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 725                         continue;
 726
 727                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 728                         char *name = xstats_names[nb_written++].name;
 729
 730                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 731                                 sizeof(xstats_names[0].name));
 732                 }
 733
 734                 ++nb_supported;
 735         }
 736
 737         return nb_written;
 738 }
 739
 740 static int
 741 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 742 {
 743         struct sfc_adapter *sa = dev->data->dev_private;
 744         unsigned int wanted_fc, link_fc;
 745
 746         memset(fc_conf, 0, sizeof(*fc_conf));
 747
 748         sfc_adapter_lock(sa);
 749
 750         if (sa->state == SFC_ADAPTER_STARTED)
 751                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 752         else
 753                 link_fc = sa->port.flow_ctrl;
 754
 755         switch (link_fc) {
 756         case 0:
 757                 fc_conf->mode = RTE_FC_NONE;
 758                 break;
 759         case EFX_FCNTL_RESPOND:
 760                 fc_conf->mode = RTE_FC_RX_PAUSE;
 761                 break;
 762         case EFX_FCNTL_GENERATE:
 763                 fc_conf->mode = RTE_FC_TX_PAUSE;
 764                 break;
 765         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 766                 fc_conf->mode = RTE_FC_FULL;
 767                 break;
 768         default:
 769                 sfc_err(sa, "%s: unexpected flow control value %#x",
 770                         __func__, link_fc);
 771         }
 772
 773         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 774
 775         sfc_adapter_unlock(sa);
 776
 777         return 0;
 778 }
 779
 780 static int
 781 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 782 {
 783         struct sfc_adapter *sa = dev->data->dev_private;
 784         struct sfc_port *port = &sa->port;
 785         unsigned int fcntl;
 786         int rc;
 787
 788         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 789             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 790             fc_conf->mac_ctrl_frame_fwd != 0) {
 791                 sfc_err(sa, "unsupported flow control settings specified");
 792                 rc = EINVAL;
 793                 goto fail_inval;
 794         }
 795
 796         switch (fc_conf->mode) {
 797         case RTE_FC_NONE:
 798                 fcntl = 0;
 799                 break;
 800         case RTE_FC_RX_PAUSE:
 801                 fcntl = EFX_FCNTL_RESPOND;
 802                 break;
 803         case RTE_FC_TX_PAUSE:
 804                 fcntl = EFX_FCNTL_GENERATE;
 805                 break;
 806         case RTE_FC_FULL:
 807                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 808                 break;
 809         default:
 810                 rc = EINVAL;
 811                 goto fail_inval;
 812         }
 813
 814         sfc_adapter_lock(sa);
 815
 816         if (sa->state == SFC_ADAPTER_STARTED) {
 817                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 818                 if (rc != 0)
 819                         goto fail_mac_fcntl_set;
 820         }
 821
 822         port->flow_ctrl = fcntl;
 823         port->flow_ctrl_autoneg = fc_conf->autoneg;
 824
 825         sfc_adapter_unlock(sa);
 826
 827         return 0;
 828
 829 fail_mac_fcntl_set:
 830         sfc_adapter_unlock(sa);
 831 fail_inval:
 832         SFC_ASSERT(rc > 0);
 833         return -rc;
 834 }
 835
 836 static int
 837 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 838 {
 839         struct sfc_adapter *sa = dev->data->dev_private;
 840         size_t pdu = EFX_MAC_PDU(mtu);
 841         size_t old_pdu;
 842         int rc;
 843
 844         sfc_log_init(sa, "mtu=%u", mtu);
 845
 846         rc = EINVAL;
 847         if (pdu < EFX_MAC_PDU_MIN) {
 848                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 849                         (unsigned int)mtu, (unsigned int)pdu,
 850                         EFX_MAC_PDU_MIN);
 851                 goto fail_inval;
 852         }
 853         if (pdu > EFX_MAC_PDU_MAX) {
 854                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 855                         (unsigned int)mtu, (unsigned int)pdu,
 856                         EFX_MAC_PDU_MAX);
 857                 goto fail_inval;
 858         }
 859
 860         sfc_adapter_lock(sa);
 861
 862         if (pdu != sa->port.pdu) {
 863                 if (sa->state == SFC_ADAPTER_STARTED) {
 864                         sfc_stop(sa);
 865
 866                         old_pdu = sa->port.pdu;
 867                         sa->port.pdu = pdu;
 868                         rc = sfc_start(sa);
 869                         if (rc != 0)
 870                                 goto fail_start;
 871                 } else {
 872                         sa->port.pdu = pdu;
 873                 }
 874         }
 875
 876         /*
 877          * The driver does not use it, but other PMDs update jumbo_frame
 878          * flag and max_rx_pkt_len when MTU is set.
 879          */
 880         if (mtu > ETHER_MAX_LEN) {
 881                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 882
 883                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 884                 rxmode->jumbo_frame = 1;
 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_RX_OFFLOAD_IPV4_CKSUM |
1061                                DEV_RX_OFFLOAD_UDP_CKSUM |
1062                                DEV_RX_OFFLOAD_TCP_CKSUM;
1063         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1064                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1065                 qinfo->scattered_rx = 1;
1066         }
1067         qinfo->nb_desc = rxq_info->entries;
1068
1069         sfc_adapter_unlock(sa);
1070 }
1071
1072 /*
1073  * The function is used by the secondary process as well. It must not
1074  * use any process-local pointers from the adapter data.
1075  */
1076 static void
1077 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1078                       struct rte_eth_txq_info *qinfo)
1079 {
1080         struct sfc_adapter *sa = dev->data->dev_private;
1081         struct sfc_txq_info *txq_info;
1082
1083         sfc_adapter_lock(sa);
1084
1085         SFC_ASSERT(tx_queue_id < sa->txq_count);
1086
1087         txq_info = &sa->txq_info[tx_queue_id];
1088         SFC_ASSERT(txq_info->txq != NULL);
1089
1090         memset(qinfo, 0, sizeof(*qinfo));
1091
1092         qinfo->conf.txq_flags = txq_info->txq->flags;
1093         qinfo->conf.offloads = txq_info->txq->offloads;
1094         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1095         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1096         qinfo->nb_desc = txq_info->entries;
1097
1098         sfc_adapter_unlock(sa);
1099 }
1100
1101 static uint32_t
1102 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1103 {
1104         struct sfc_adapter *sa = dev->data->dev_private;
1105
1106         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1107
1108         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1109 }
1110
1111 static int
1112 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1113 {
1114         struct sfc_dp_rxq *dp_rxq = queue;
1115
1116         return sfc_rx_qdesc_done(dp_rxq, offset);
1117 }
1118
1119 static int
1120 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1121 {
1122         struct sfc_dp_rxq *dp_rxq = queue;
1123         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1124
1125         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1126 }
1127
1128 static int
1129 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1130 {
1131         struct sfc_dp_txq *dp_txq = queue;
1132         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1133
1134         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1135 }
1136
1137 static int
1138 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1139 {
1140         struct sfc_adapter *sa = dev->data->dev_private;
1141         int rc;
1142
1143         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1144
1145         sfc_adapter_lock(sa);
1146
1147         rc = EINVAL;
1148         if (sa->state != SFC_ADAPTER_STARTED)
1149                 goto fail_not_started;
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_started:
1163         sfc_adapter_unlock(sa);
1164         SFC_ASSERT(rc > 0);
1165         return -rc;
1166 }
1167
1168 static int
1169 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1170 {
1171         struct sfc_adapter *sa = dev->data->dev_private;
1172
1173         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1174
1175         sfc_adapter_lock(sa);
1176         sfc_rx_qstop(sa, rx_queue_id);
1177
1178         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1179
1180         sfc_adapter_unlock(sa);
1181
1182         return 0;
1183 }
1184
1185 static int
1186 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1187 {
1188         struct sfc_adapter *sa = dev->data->dev_private;
1189         int rc;
1190
1191         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1192
1193         sfc_adapter_lock(sa);
1194
1195         rc = EINVAL;
1196         if (sa->state != SFC_ADAPTER_STARTED)
1197                 goto fail_not_started;
1198
1199         rc = sfc_tx_qstart(sa, tx_queue_id);
1200         if (rc != 0)
1201                 goto fail_tx_qstart;
1202
1203         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1204
1205         sfc_adapter_unlock(sa);
1206         return 0;
1207
1208 fail_tx_qstart:
1209
1210 fail_not_started:
1211         sfc_adapter_unlock(sa);
1212         SFC_ASSERT(rc > 0);
1213         return -rc;
1214 }
1215
1216 static int
1217 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1218 {
1219         struct sfc_adapter *sa = dev->data->dev_private;
1220
1221         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1222
1223         sfc_adapter_lock(sa);
1224
1225         sfc_tx_qstop(sa, tx_queue_id);
1226
1227         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1228
1229         sfc_adapter_unlock(sa);
1230         return 0;
1231 }
1232
1233 static efx_tunnel_protocol_t
1234 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1235 {
1236         switch (rte_type) {
1237         case RTE_TUNNEL_TYPE_VXLAN:
1238                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1239         case RTE_TUNNEL_TYPE_GENEVE:
1240                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1241         default:
1242                 return EFX_TUNNEL_NPROTOS;
1243         }
1244 }
1245
1246 enum sfc_udp_tunnel_op_e {
1247         SFC_UDP_TUNNEL_ADD_PORT,
1248         SFC_UDP_TUNNEL_DEL_PORT,
1249 };
1250
1251 static int
1252 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1253                       struct rte_eth_udp_tunnel *tunnel_udp,
1254                       enum sfc_udp_tunnel_op_e op)
1255 {
1256         struct sfc_adapter *sa = dev->data->dev_private;
1257         efx_tunnel_protocol_t tunnel_proto;
1258         int rc;
1259
1260         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1261                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1262                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1263                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1264
1265         tunnel_proto =
1266                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1267         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1268                 rc = ENOTSUP;
1269                 goto fail_bad_proto;
1270         }
1271
1272         sfc_adapter_lock(sa);
1273
1274         switch (op) {
1275         case SFC_UDP_TUNNEL_ADD_PORT:
1276                 rc = efx_tunnel_config_udp_add(sa->nic,
1277                                                tunnel_udp->udp_port,
1278                                                tunnel_proto);
1279                 break;
1280         case SFC_UDP_TUNNEL_DEL_PORT:
1281                 rc = efx_tunnel_config_udp_remove(sa->nic,
1282                                                   tunnel_udp->udp_port,
1283                                                   tunnel_proto);
1284                 break;
1285         default:
1286                 rc = EINVAL;
1287                 goto fail_bad_op;
1288         }
1289
1290         if (rc != 0)
1291                 goto fail_op;
1292
1293         if (sa->state == SFC_ADAPTER_STARTED) {
1294                 rc = efx_tunnel_reconfigure(sa->nic);
1295                 if (rc == EAGAIN) {
1296                         /*
1297                          * Configuration is accepted by FW and MC reboot
1298                          * is initiated to apply the changes. MC reboot
1299                          * will be handled in a usual way (MC reboot
1300                          * event on management event queue and adapter
1301                          * restart).
1302                          */
1303                         rc = 0;
1304                 } else if (rc != 0) {
1305                         goto fail_reconfigure;
1306                 }
1307         }
1308
1309         sfc_adapter_unlock(sa);
1310         return 0;
1311
1312 fail_reconfigure:
1313         /* Remove/restore entry since the change makes the trouble */
1314         switch (op) {
1315         case SFC_UDP_TUNNEL_ADD_PORT:
1316                 (void)efx_tunnel_config_udp_remove(sa->nic,
1317                                                    tunnel_udp->udp_port,
1318                                                    tunnel_proto);
1319                 break;
1320         case SFC_UDP_TUNNEL_DEL_PORT:
1321                 (void)efx_tunnel_config_udp_add(sa->nic,
1322                                                 tunnel_udp->udp_port,
1323                                                 tunnel_proto);
1324                 break;
1325         }
1326
1327 fail_op:
1328 fail_bad_op:
1329         sfc_adapter_unlock(sa);
1330
1331 fail_bad_proto:
1332         SFC_ASSERT(rc > 0);
1333         return -rc;
1334 }
1335
1336 static int
1337 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1338                             struct rte_eth_udp_tunnel *tunnel_udp)
1339 {
1340         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1341 }
1342
1343 static int
1344 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1345                             struct rte_eth_udp_tunnel *tunnel_udp)
1346 {
1347         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1348 }
1349
1350 static int
1351 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1352                           struct rte_eth_rss_conf *rss_conf)
1353 {
1354         struct sfc_adapter *sa = dev->data->dev_private;
1355         struct sfc_rss *rss = &sa->rss;
1356         struct sfc_port *port = &sa->port;
1357
1358         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1359                 return -ENOTSUP;
1360
1361         if (rss->channels == 0)
1362                 return -EINVAL;
1363
1364         sfc_adapter_lock(sa);
1365
1366         /*
1367          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1368          * hence, conversion is done here to derive a correct set of ETH_RSS
1369          * flags which corresponds to the active EFX configuration stored
1370          * locally in 'sfc_adapter' and kept up-to-date
1371          */
1372         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1373         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1374         if (rss_conf->rss_key != NULL)
1375                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1376
1377         sfc_adapter_unlock(sa);
1378
1379         return 0;
1380 }
1381
1382 static int
1383 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1384                         struct rte_eth_rss_conf *rss_conf)
1385 {
1386         struct sfc_adapter *sa = dev->data->dev_private;
1387         struct sfc_rss *rss = &sa->rss;
1388         struct sfc_port *port = &sa->port;
1389         unsigned int efx_hash_types;
1390         int rc = 0;
1391
1392         if (port->isolated)
1393                 return -ENOTSUP;
1394
1395         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1396                 sfc_err(sa, "RSS is not available");
1397                 return -ENOTSUP;
1398         }
1399
1400         if (rss->channels == 0) {
1401                 sfc_err(sa, "RSS is not configured");
1402                 return -EINVAL;
1403         }
1404
1405         if ((rss_conf->rss_key != NULL) &&
1406             (rss_conf->rss_key_len != sizeof(rss->key))) {
1407                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1408                         sizeof(rss->key));
1409                 return -EINVAL;
1410         }
1411
1412         sfc_adapter_lock(sa);
1413
1414         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1415         if (rc != 0)
1416                 goto fail_rx_hf_rte_to_efx;
1417
1418         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1419                                    rss->hash_alg, efx_hash_types, B_TRUE);
1420         if (rc != 0)
1421                 goto fail_scale_mode_set;
1422
1423         if (rss_conf->rss_key != NULL) {
1424                 if (sa->state == SFC_ADAPTER_STARTED) {
1425                         rc = efx_rx_scale_key_set(sa->nic,
1426                                                   EFX_RSS_CONTEXT_DEFAULT,
1427                                                   rss_conf->rss_key,
1428                                                   sizeof(rss->key));
1429                         if (rc != 0)
1430                                 goto fail_scale_key_set;
1431                 }
1432
1433                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1434         }
1435
1436         rss->hash_types = efx_hash_types;
1437
1438         sfc_adapter_unlock(sa);
1439
1440         return 0;
1441
1442 fail_scale_key_set:
1443         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1444                                   EFX_RX_HASHALG_TOEPLITZ,
1445                                   rss->hash_types, B_TRUE) != 0)
1446                 sfc_err(sa, "failed to restore RSS mode");
1447
1448 fail_scale_mode_set:
1449 fail_rx_hf_rte_to_efx:
1450         sfc_adapter_unlock(sa);
1451         return -rc;
1452 }
1453
1454 static int
1455 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1456                        struct rte_eth_rss_reta_entry64 *reta_conf,
1457                        uint16_t reta_size)
1458 {
1459         struct sfc_adapter *sa = dev->data->dev_private;
1460         struct sfc_rss *rss = &sa->rss;
1461         struct sfc_port *port = &sa->port;
1462         int entry;
1463
1464         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1465                 return -ENOTSUP;
1466
1467         if (rss->channels == 0)
1468                 return -EINVAL;
1469
1470         if (reta_size != EFX_RSS_TBL_SIZE)
1471                 return -EINVAL;
1472
1473         sfc_adapter_lock(sa);
1474
1475         for (entry = 0; entry < reta_size; entry++) {
1476                 int grp = entry / RTE_RETA_GROUP_SIZE;
1477                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1478
1479                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1480                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1481         }
1482
1483         sfc_adapter_unlock(sa);
1484
1485         return 0;
1486 }
1487
1488 static int
1489 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1490                         struct rte_eth_rss_reta_entry64 *reta_conf,
1491                         uint16_t reta_size)
1492 {
1493         struct sfc_adapter *sa = dev->data->dev_private;
1494         struct sfc_rss *rss = &sa->rss;
1495         struct sfc_port *port = &sa->port;
1496         unsigned int *rss_tbl_new;
1497         uint16_t entry;
1498         int rc = 0;
1499
1500
1501         if (port->isolated)
1502                 return -ENOTSUP;
1503
1504         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1505                 sfc_err(sa, "RSS is not available");
1506                 return -ENOTSUP;
1507         }
1508
1509         if (rss->channels == 0) {
1510                 sfc_err(sa, "RSS is not configured");
1511                 return -EINVAL;
1512         }
1513
1514         if (reta_size != EFX_RSS_TBL_SIZE) {
1515                 sfc_err(sa, "RETA size is wrong (should be %u)",
1516                         EFX_RSS_TBL_SIZE);
1517                 return -EINVAL;
1518         }
1519
1520         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1521         if (rss_tbl_new == NULL)
1522                 return -ENOMEM;
1523
1524         sfc_adapter_lock(sa);
1525
1526         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1527
1528         for (entry = 0; entry < reta_size; entry++) {
1529                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1530                 struct rte_eth_rss_reta_entry64 *grp;
1531
1532                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1533
1534                 if (grp->mask & (1ull << grp_idx)) {
1535                         if (grp->reta[grp_idx] >= rss->channels) {
1536                                 rc = EINVAL;
1537                                 goto bad_reta_entry;
1538                         }
1539                         rss_tbl_new[entry] = grp->reta[grp_idx];
1540                 }
1541         }
1542
1543         if (sa->state == SFC_ADAPTER_STARTED) {
1544                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1545                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1546                 if (rc != 0)
1547                         goto fail_scale_tbl_set;
1548         }
1549
1550         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1551
1552 fail_scale_tbl_set:
1553 bad_reta_entry:
1554         sfc_adapter_unlock(sa);
1555
1556         rte_free(rss_tbl_new);
1557
1558         SFC_ASSERT(rc >= 0);
1559         return -rc;
1560 }
1561
1562 static int
1563 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1564                     enum rte_filter_op filter_op,
1565                     void *arg)
1566 {
1567         struct sfc_adapter *sa = dev->data->dev_private;
1568         int rc = ENOTSUP;
1569
1570         sfc_log_init(sa, "entry");
1571
1572         switch (filter_type) {
1573         case RTE_ETH_FILTER_NONE:
1574                 sfc_err(sa, "Global filters configuration not supported");
1575                 break;
1576         case RTE_ETH_FILTER_MACVLAN:
1577                 sfc_err(sa, "MACVLAN filters not supported");
1578                 break;
1579         case RTE_ETH_FILTER_ETHERTYPE:
1580                 sfc_err(sa, "EtherType filters not supported");
1581                 break;
1582         case RTE_ETH_FILTER_FLEXIBLE:
1583                 sfc_err(sa, "Flexible filters not supported");
1584                 break;
1585         case RTE_ETH_FILTER_SYN:
1586                 sfc_err(sa, "SYN filters not supported");
1587                 break;
1588         case RTE_ETH_FILTER_NTUPLE:
1589                 sfc_err(sa, "NTUPLE filters not supported");
1590                 break;
1591         case RTE_ETH_FILTER_TUNNEL:
1592                 sfc_err(sa, "Tunnel filters not supported");
1593                 break;
1594         case RTE_ETH_FILTER_FDIR:
1595                 sfc_err(sa, "Flow Director filters not supported");
1596                 break;
1597         case RTE_ETH_FILTER_HASH:
1598                 sfc_err(sa, "Hash filters not supported");
1599                 break;
1600         case RTE_ETH_FILTER_GENERIC:
1601                 if (filter_op != RTE_ETH_FILTER_GET) {
1602                         rc = EINVAL;
1603                 } else {
1604                         *(const void **)arg = &sfc_flow_ops;
1605                         rc = 0;
1606                 }
1607                 break;
1608         default:
1609                 sfc_err(sa, "Unknown filter type %u", filter_type);
1610                 break;
1611         }
1612
1613         sfc_log_init(sa, "exit: %d", -rc);
1614         SFC_ASSERT(rc >= 0);
1615         return -rc;
1616 }
1617
1618 static int
1619 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1620 {
1621         struct sfc_adapter *sa = dev->data->dev_private;
1622
1623         /*
1624          * If Rx datapath does not provide callback to check mempool,
1625          * all pools are supported.
1626          */
1627         if (sa->dp_rx->pool_ops_supported == NULL)
1628                 return 1;
1629
1630         return sa->dp_rx->pool_ops_supported(pool);
1631 }
1632
1633 static const struct eth_dev_ops sfc_eth_dev_ops = {
1634         .dev_configure                  = sfc_dev_configure,
1635         .dev_start                      = sfc_dev_start,
1636         .dev_stop                       = sfc_dev_stop,
1637         .dev_set_link_up                = sfc_dev_set_link_up,
1638         .dev_set_link_down              = sfc_dev_set_link_down,
1639         .dev_close                      = sfc_dev_close,
1640         .promiscuous_enable             = sfc_dev_promisc_enable,
1641         .promiscuous_disable            = sfc_dev_promisc_disable,
1642         .allmulticast_enable            = sfc_dev_allmulti_enable,
1643         .allmulticast_disable           = sfc_dev_allmulti_disable,
1644         .link_update                    = sfc_dev_link_update,
1645         .stats_get                      = sfc_stats_get,
1646         .stats_reset                    = sfc_stats_reset,
1647         .xstats_get                     = sfc_xstats_get,
1648         .xstats_reset                   = sfc_stats_reset,
1649         .xstats_get_names               = sfc_xstats_get_names,
1650         .dev_infos_get                  = sfc_dev_infos_get,
1651         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1652         .mtu_set                        = sfc_dev_set_mtu,
1653         .rx_queue_start                 = sfc_rx_queue_start,
1654         .rx_queue_stop                  = sfc_rx_queue_stop,
1655         .tx_queue_start                 = sfc_tx_queue_start,
1656         .tx_queue_stop                  = sfc_tx_queue_stop,
1657         .rx_queue_setup                 = sfc_rx_queue_setup,
1658         .rx_queue_release               = sfc_rx_queue_release,
1659         .rx_queue_count                 = sfc_rx_queue_count,
1660         .rx_descriptor_done             = sfc_rx_descriptor_done,
1661         .rx_descriptor_status           = sfc_rx_descriptor_status,
1662         .tx_descriptor_status           = sfc_tx_descriptor_status,
1663         .tx_queue_setup                 = sfc_tx_queue_setup,
1664         .tx_queue_release               = sfc_tx_queue_release,
1665         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1666         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1667         .mac_addr_set                   = sfc_mac_addr_set,
1668         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1669         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1670         .reta_update                    = sfc_dev_rss_reta_update,
1671         .reta_query                     = sfc_dev_rss_reta_query,
1672         .rss_hash_update                = sfc_dev_rss_hash_update,
1673         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1674         .filter_ctrl                    = sfc_dev_filter_ctrl,
1675         .set_mc_addr_list               = sfc_set_mc_addr_list,
1676         .rxq_info_get                   = sfc_rx_queue_info_get,
1677         .txq_info_get                   = sfc_tx_queue_info_get,
1678         .fw_version_get                 = sfc_fw_version_get,
1679         .xstats_get_by_id               = sfc_xstats_get_by_id,
1680         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1681         .pool_ops_supported             = sfc_pool_ops_supported,
1682 };
1683
1684 /**
1685  * Duplicate a string in potentially shared memory required for
1686  * multi-process support.
1687  *
1688  * strdup() allocates from process-local heap/memory.
1689  */
1690 static char *
1691 sfc_strdup(const char *str)
1692 {
1693         size_t size;
1694         char *copy;
1695
1696         if (str == NULL)
1697                 return NULL;
1698
1699         size = strlen(str) + 1;
1700         copy = rte_malloc(__func__, size, 0);
1701         if (copy != NULL)
1702                 rte_memcpy(copy, str, size);
1703
1704         return copy;
1705 }
1706
1707 static int
1708 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1709 {
1710         struct sfc_adapter *sa = dev->data->dev_private;
1711         const efx_nic_cfg_t *encp;
1712         unsigned int avail_caps = 0;
1713         const char *rx_name = NULL;
1714         const char *tx_name = NULL;
1715         int rc;
1716
1717         switch (sa->family) {
1718         case EFX_FAMILY_HUNTINGTON:
1719         case EFX_FAMILY_MEDFORD:
1720         case EFX_FAMILY_MEDFORD2:
1721                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1722                 break;
1723         default:
1724                 break;
1725         }
1726
1727         encp = efx_nic_cfg_get(sa->nic);
1728         if (encp->enc_rx_es_super_buffer_supported)
1729                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1730
1731         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1732                                 sfc_kvarg_string_handler, &rx_name);
1733         if (rc != 0)
1734                 goto fail_kvarg_rx_datapath;
1735
1736         if (rx_name != NULL) {
1737                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1738                 if (sa->dp_rx == NULL) {
1739                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1740                         rc = ENOENT;
1741                         goto fail_dp_rx;
1742                 }
1743                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1744                         sfc_err(sa,
1745                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1746                                 rx_name);
1747                         rc = EINVAL;
1748                         goto fail_dp_rx_caps;
1749                 }
1750         } else {
1751                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1752                 if (sa->dp_rx == NULL) {
1753                         sfc_err(sa, "Rx datapath by caps %#x not found",
1754                                 avail_caps);
1755                         rc = ENOENT;
1756                         goto fail_dp_rx;
1757                 }
1758         }
1759
1760         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1761         if (sa->dp_rx_name == NULL) {
1762                 rc = ENOMEM;
1763                 goto fail_dp_rx_name;
1764         }
1765
1766         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1767
1768         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1769
1770         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1771                                 sfc_kvarg_string_handler, &tx_name);
1772         if (rc != 0)
1773                 goto fail_kvarg_tx_datapath;
1774
1775         if (tx_name != NULL) {
1776                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1777                 if (sa->dp_tx == NULL) {
1778                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1779                         rc = ENOENT;
1780                         goto fail_dp_tx;
1781                 }
1782                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1783                         sfc_err(sa,
1784                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1785                                 tx_name);
1786                         rc = EINVAL;
1787                         goto fail_dp_tx_caps;
1788                 }
1789         } else {
1790                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1791                 if (sa->dp_tx == NULL) {
1792                         sfc_err(sa, "Tx datapath by caps %#x not found",
1793                                 avail_caps);
1794                         rc = ENOENT;
1795                         goto fail_dp_tx;
1796                 }
1797         }
1798
1799         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1800         if (sa->dp_tx_name == NULL) {
1801                 rc = ENOMEM;
1802                 goto fail_dp_tx_name;
1803         }
1804
1805         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1806
1807         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1808
1809         dev->dev_ops = &sfc_eth_dev_ops;
1810
1811         return 0;
1812
1813 fail_dp_tx_name:
1814 fail_dp_tx_caps:
1815         sa->dp_tx = NULL;
1816
1817 fail_dp_tx:
1818 fail_kvarg_tx_datapath:
1819         rte_free(sa->dp_rx_name);
1820         sa->dp_rx_name = NULL;
1821
1822 fail_dp_rx_name:
1823 fail_dp_rx_caps:
1824         sa->dp_rx = NULL;
1825
1826 fail_dp_rx:
1827 fail_kvarg_rx_datapath:
1828         return rc;
1829 }
1830
1831 static void
1832 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1833 {
1834         struct sfc_adapter *sa = dev->data->dev_private;
1835
1836         dev->dev_ops = NULL;
1837         dev->rx_pkt_burst = NULL;
1838         dev->tx_pkt_burst = NULL;
1839
1840         rte_free(sa->dp_tx_name);
1841         sa->dp_tx_name = NULL;
1842         sa->dp_tx = NULL;
1843
1844         rte_free(sa->dp_rx_name);
1845         sa->dp_rx_name = NULL;
1846         sa->dp_rx = NULL;
1847 }
1848
1849 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1850         .rxq_info_get                   = sfc_rx_queue_info_get,
1851         .txq_info_get                   = sfc_tx_queue_info_get,
1852 };
1853
1854 static int
1855 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1856 {
1857         /*
1858          * Device private data has really many process-local pointers.
1859          * Below code should be extremely careful to use data located
1860          * in shared memory only.
1861          */
1862         struct sfc_adapter *sa = dev->data->dev_private;
1863         const struct sfc_dp_rx *dp_rx;
1864         const struct sfc_dp_tx *dp_tx;
1865         int rc;
1866
1867         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1868         if (dp_rx == NULL) {
1869                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1870                 rc = ENOENT;
1871                 goto fail_dp_rx;
1872         }
1873         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1874                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1875                         sa->dp_tx_name);
1876                 rc = EINVAL;
1877                 goto fail_dp_rx_multi_process;
1878         }
1879
1880         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1881         if (dp_tx == NULL) {
1882                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1883                 rc = ENOENT;
1884                 goto fail_dp_tx;
1885         }
1886         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1887                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1888                         sa->dp_tx_name);
1889                 rc = EINVAL;
1890                 goto fail_dp_tx_multi_process;
1891         }
1892
1893         dev->rx_pkt_burst = dp_rx->pkt_burst;
1894         dev->tx_pkt_burst = dp_tx->pkt_burst;
1895         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1896
1897         return 0;
1898
1899 fail_dp_tx_multi_process:
1900 fail_dp_tx:
1901 fail_dp_rx_multi_process:
1902 fail_dp_rx:
1903         return rc;
1904 }
1905
1906 static void
1907 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1908 {
1909         dev->dev_ops = NULL;
1910         dev->tx_pkt_burst = NULL;
1911         dev->rx_pkt_burst = NULL;
1912 }
1913
1914 static void
1915 sfc_register_dp(void)
1916 {
1917         /* Register once */
1918         if (TAILQ_EMPTY(&sfc_dp_head)) {
1919                 /* Prefer EF10 datapath */
1920                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1921                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1922                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1923
1924                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1925                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1926                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1927         }
1928 }
1929
1930 static int
1931 sfc_eth_dev_init(struct rte_eth_dev *dev)
1932 {
1933         struct sfc_adapter *sa = dev->data->dev_private;
1934         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1935         int rc;
1936         const efx_nic_cfg_t *encp;
1937         const struct ether_addr *from;
1938
1939         sfc_register_dp();
1940
1941         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1942                 return -sfc_eth_dev_secondary_set_ops(dev);
1943
1944         /* Required for logging */
1945         sa->pci_addr = pci_dev->addr;
1946         sa->port_id = dev->data->port_id;
1947
1948         sa->eth_dev = dev;
1949
1950         /* Copy PCI device info to the dev->data */
1951         rte_eth_copy_pci_info(dev, pci_dev);
1952
1953         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1954                                                 RTE_LOG_NOTICE);
1955
1956         rc = sfc_kvargs_parse(sa);
1957         if (rc != 0)
1958                 goto fail_kvargs_parse;
1959
1960         sfc_log_init(sa, "entry");
1961
1962         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1963         if (dev->data->mac_addrs == NULL) {
1964                 rc = ENOMEM;
1965                 goto fail_mac_addrs;
1966         }
1967
1968         sfc_adapter_lock_init(sa);
1969         sfc_adapter_lock(sa);
1970
1971         sfc_log_init(sa, "probing");
1972         rc = sfc_probe(sa);
1973         if (rc != 0)
1974                 goto fail_probe;
1975
1976         sfc_log_init(sa, "set device ops");
1977         rc = sfc_eth_dev_set_ops(dev);
1978         if (rc != 0)
1979                 goto fail_set_ops;
1980
1981         sfc_log_init(sa, "attaching");
1982         rc = sfc_attach(sa);
1983         if (rc != 0)
1984                 goto fail_attach;
1985
1986         encp = efx_nic_cfg_get(sa->nic);
1987
1988         /*
1989          * The arguments are really reverse order in comparison to
1990          * Linux kernel. Copy from NIC config to Ethernet device data.
1991          */
1992         from = (const struct ether_addr *)(encp->enc_mac_addr);
1993         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1994
1995         sfc_adapter_unlock(sa);
1996
1997         sfc_log_init(sa, "done");
1998         return 0;
1999
2000 fail_attach:
2001         sfc_eth_dev_clear_ops(dev);
2002
2003 fail_set_ops:
2004         sfc_unprobe(sa);
2005
2006 fail_probe:
2007         sfc_adapter_unlock(sa);
2008         sfc_adapter_lock_fini(sa);
2009         rte_free(dev->data->mac_addrs);
2010         dev->data->mac_addrs = NULL;
2011
2012 fail_mac_addrs:
2013         sfc_kvargs_cleanup(sa);
2014
2015 fail_kvargs_parse:
2016         sfc_log_init(sa, "failed %d", rc);
2017         SFC_ASSERT(rc > 0);
2018         return -rc;
2019 }
2020
2021 static int
2022 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2023 {
2024         struct sfc_adapter *sa;
2025
2026         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2027                 sfc_eth_dev_secondary_clear_ops(dev);
2028                 return 0;
2029         }
2030
2031         sa = dev->data->dev_private;
2032         sfc_log_init(sa, "entry");
2033
2034         sfc_adapter_lock(sa);
2035
2036         sfc_eth_dev_clear_ops(dev);
2037
2038         sfc_detach(sa);
2039         sfc_unprobe(sa);
2040
2041         rte_free(dev->data->mac_addrs);
2042         dev->data->mac_addrs = NULL;
2043
2044         sfc_kvargs_cleanup(sa);
2045
2046         sfc_adapter_unlock(sa);
2047         sfc_adapter_lock_fini(sa);
2048
2049         sfc_log_init(sa, "done");
2050
2051         /* Required for logging, so cleanup last */
2052         sa->eth_dev = NULL;
2053         return 0;
2054 }
2055
2056 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2057         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2058         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2059         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2060         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2061         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2062         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2063         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2064         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2065         { .vendor_id = 0 /* sentinel */ }
2066 };
2067
2068 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2069         struct rte_pci_device *pci_dev)
2070 {
2071         return rte_eth_dev_pci_generic_probe(pci_dev,
2072                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2073 }
2074
2075 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2076 {
2077         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2078 }
2079
2080 static struct rte_pci_driver sfc_efx_pmd = {
2081         .id_table = pci_id_sfc_efx_map,
2082         .drv_flags =
2083                 RTE_PCI_DRV_INTR_LSC |
2084                 RTE_PCI_DRV_NEED_MAPPING,
2085         .probe = sfc_eth_dev_pci_probe,
2086         .remove = sfc_eth_dev_pci_remove,
2087 };
2088
2089 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2090 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2091 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2092 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2093         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2094         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2095         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2096         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2097         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2098         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2099
2100 RTE_INIT(sfc_driver_register_logtype);
2101 static void
2102 sfc_driver_register_logtype(void)
2103 {
2104         int ret;
2105
2106         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2107                                                    RTE_LOG_NOTICE);
2108         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2109 }