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                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 883         }
 884
 885         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 886
 887         sfc_adapter_unlock(sa);
 888
 889         sfc_log_init(sa, "done");
 890         return 0;
 891
 892 fail_start:
 893         sa->port.pdu = old_pdu;
 894         if (sfc_start(sa) != 0)
 895                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 896                         "PDU max size - port is stopped",
 897                         (unsigned int)pdu, (unsigned int)old_pdu);
 898         sfc_adapter_unlock(sa);
 899
 900 fail_inval:
 901         sfc_log_init(sa, "failed %d", rc);
 902         SFC_ASSERT(rc > 0);
 903         return -rc;
 904 }
 905 static int
 906 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 907 {
 908         struct sfc_adapter *sa = dev->data->dev_private;
 909         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 910         struct sfc_port *port = &sa->port;
 911         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
 912         int rc = 0;
 913
 914         sfc_adapter_lock(sa);
 915
 916         /*
 917          * Copy the address to the device private data so that
 918          * it could be recalled in the case of adapter restart.
 919          */
 920         ether_addr_copy(mac_addr, &port->default_mac_addr);
 921
 922         /*
 923          * Neither of the two following checks can return
 924          * an error. The new MAC address is preserved in
 925          * the device private data and can be activated
 926          * on the next port start if the user prevents
 927          * isolated mode from being enabled.
 928          */
 929         if (port->isolated) {
 930                 sfc_warn(sa, "isolated mode is active on the port");
 931                 sfc_warn(sa, "will not set MAC address");
 932                 goto unlock;
 933         }
 934
 935         if (sa->state != SFC_ADAPTER_STARTED) {
 936                 sfc_notice(sa, "the port is not started");
 937                 sfc_notice(sa, "the new MAC address will be set on port start");
 938
 939                 goto unlock;
 940         }
 941
 942         if (encp->enc_allow_set_mac_with_installed_filters) {
 943                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 944                 if (rc != 0) {
 945                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 946                         goto unlock;
 947                 }
 948
 949                 /*
 950                  * Changing the MAC address by means of MCDI request
 951                  * has no effect on received traffic, therefore
 952                  * we also need to update unicast filters
 953                  */
 954                 rc = sfc_set_rx_mode(sa);
 955                 if (rc != 0) {
 956                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 957                         /* Rollback the old address */
 958                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
 959                         (void)sfc_set_rx_mode(sa);
 960                 }
 961         } else {
 962                 sfc_warn(sa, "cannot set MAC address with filters installed");
 963                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 964                 sfc_warn(sa, "(some traffic may be dropped)");
 965
 966                 /*
 967                  * Since setting MAC address with filters installed is not
 968                  * allowed on the adapter, the new MAC address will be set
 969                  * by means of adapter restart. sfc_start() shall retrieve
 970                  * the new address from the device private data and set it.
 971                  */
 972                 sfc_stop(sa);
 973                 rc = sfc_start(sa);
 974                 if (rc != 0)
 975                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 976         }
 977
 978 unlock:
 979         if (rc != 0)
 980                 ether_addr_copy(old_addr, &port->default_mac_addr);
 981
 982         sfc_adapter_unlock(sa);
 983
 984         SFC_ASSERT(rc >= 0);
 985         return -rc;
 986 }
 987
 988
 989 static int
 990 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 991                      uint32_t nb_mc_addr)
 992 {
 993         struct sfc_adapter *sa = dev->data->dev_private;
 994         struct sfc_port *port = &sa->port;
 995         uint8_t *mc_addrs = port->mcast_addrs;
 996         int rc;
 997         unsigned int i;
 998
 999         if (port->isolated) {
1000                 sfc_err(sa, "isolated mode is active on the port");
1001                 sfc_err(sa, "will not set multicast address list");
1002                 return -ENOTSUP;
1003         }
1004
1005         if (mc_addrs == NULL)
1006                 return -ENOBUFS;
1007
1008         if (nb_mc_addr > port->max_mcast_addrs) {
1009                 sfc_err(sa, "too many multicast addresses: %u > %u",
1010                          nb_mc_addr, port->max_mcast_addrs);
1011                 return -EINVAL;
1012         }
1013
1014         for (i = 0; i < nb_mc_addr; ++i) {
1015                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1016                                  EFX_MAC_ADDR_LEN);
1017                 mc_addrs += EFX_MAC_ADDR_LEN;
1018         }
1019
1020         port->nb_mcast_addrs = nb_mc_addr;
1021
1022         if (sa->state != SFC_ADAPTER_STARTED)
1023                 return 0;
1024
1025         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1026                                         port->nb_mcast_addrs);
1027         if (rc != 0)
1028                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1029
1030         SFC_ASSERT(rc > 0);
1031         return -rc;
1032 }
1033
1034 /*
1035  * The function is used by the secondary process as well. It must not
1036  * use any process-local pointers from the adapter data.
1037  */
1038 static void
1039 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1040                       struct rte_eth_rxq_info *qinfo)
1041 {
1042         struct sfc_adapter *sa = dev->data->dev_private;
1043         struct sfc_rxq_info *rxq_info;
1044         struct sfc_rxq *rxq;
1045
1046         sfc_adapter_lock(sa);
1047
1048         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1049
1050         rxq_info = &sa->rxq_info[rx_queue_id];
1051         rxq = rxq_info->rxq;
1052         SFC_ASSERT(rxq != NULL);
1053
1054         qinfo->mp = rxq->refill_mb_pool;
1055         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1056         qinfo->conf.rx_drop_en = 1;
1057         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1058         qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1059                                DEV_RX_OFFLOAD_UDP_CKSUM |
1060                                DEV_RX_OFFLOAD_TCP_CKSUM;
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         rc = sfc_rx_qstart(sa, rx_queue_id);
1149         if (rc != 0)
1150                 goto fail_rx_qstart;
1151
1152         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1153
1154         sfc_adapter_unlock(sa);
1155
1156         return 0;
1157
1158 fail_rx_qstart:
1159 fail_not_started:
1160         sfc_adapter_unlock(sa);
1161         SFC_ASSERT(rc > 0);
1162         return -rc;
1163 }
1164
1165 static int
1166 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1167 {
1168         struct sfc_adapter *sa = dev->data->dev_private;
1169
1170         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1171
1172         sfc_adapter_lock(sa);
1173         sfc_rx_qstop(sa, rx_queue_id);
1174
1175         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1176
1177         sfc_adapter_unlock(sa);
1178
1179         return 0;
1180 }
1181
1182 static int
1183 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1184 {
1185         struct sfc_adapter *sa = dev->data->dev_private;
1186         int rc;
1187
1188         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1189
1190         sfc_adapter_lock(sa);
1191
1192         rc = EINVAL;
1193         if (sa->state != SFC_ADAPTER_STARTED)
1194                 goto fail_not_started;
1195
1196         rc = sfc_tx_qstart(sa, tx_queue_id);
1197         if (rc != 0)
1198                 goto fail_tx_qstart;
1199
1200         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1201
1202         sfc_adapter_unlock(sa);
1203         return 0;
1204
1205 fail_tx_qstart:
1206
1207 fail_not_started:
1208         sfc_adapter_unlock(sa);
1209         SFC_ASSERT(rc > 0);
1210         return -rc;
1211 }
1212
1213 static int
1214 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1215 {
1216         struct sfc_adapter *sa = dev->data->dev_private;
1217
1218         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1219
1220         sfc_adapter_lock(sa);
1221
1222         sfc_tx_qstop(sa, tx_queue_id);
1223
1224         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1225
1226         sfc_adapter_unlock(sa);
1227         return 0;
1228 }
1229
1230 static efx_tunnel_protocol_t
1231 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1232 {
1233         switch (rte_type) {
1234         case RTE_TUNNEL_TYPE_VXLAN:
1235                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1236         case RTE_TUNNEL_TYPE_GENEVE:
1237                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1238         default:
1239                 return EFX_TUNNEL_NPROTOS;
1240         }
1241 }
1242
1243 enum sfc_udp_tunnel_op_e {
1244         SFC_UDP_TUNNEL_ADD_PORT,
1245         SFC_UDP_TUNNEL_DEL_PORT,
1246 };
1247
1248 static int
1249 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1250                       struct rte_eth_udp_tunnel *tunnel_udp,
1251                       enum sfc_udp_tunnel_op_e op)
1252 {
1253         struct sfc_adapter *sa = dev->data->dev_private;
1254         efx_tunnel_protocol_t tunnel_proto;
1255         int rc;
1256
1257         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1258                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1259                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1260                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1261
1262         tunnel_proto =
1263                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1264         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1265                 rc = ENOTSUP;
1266                 goto fail_bad_proto;
1267         }
1268
1269         sfc_adapter_lock(sa);
1270
1271         switch (op) {
1272         case SFC_UDP_TUNNEL_ADD_PORT:
1273                 rc = efx_tunnel_config_udp_add(sa->nic,
1274                                                tunnel_udp->udp_port,
1275                                                tunnel_proto);
1276                 break;
1277         case SFC_UDP_TUNNEL_DEL_PORT:
1278                 rc = efx_tunnel_config_udp_remove(sa->nic,
1279                                                   tunnel_udp->udp_port,
1280                                                   tunnel_proto);
1281                 break;
1282         default:
1283                 rc = EINVAL;
1284                 goto fail_bad_op;
1285         }
1286
1287         if (rc != 0)
1288                 goto fail_op;
1289
1290         if (sa->state == SFC_ADAPTER_STARTED) {
1291                 rc = efx_tunnel_reconfigure(sa->nic);
1292                 if (rc == EAGAIN) {
1293                         /*
1294                          * Configuration is accepted by FW and MC reboot
1295                          * is initiated to apply the changes. MC reboot
1296                          * will be handled in a usual way (MC reboot
1297                          * event on management event queue and adapter
1298                          * restart).
1299                          */
1300                         rc = 0;
1301                 } else if (rc != 0) {
1302                         goto fail_reconfigure;
1303                 }
1304         }
1305
1306         sfc_adapter_unlock(sa);
1307         return 0;
1308
1309 fail_reconfigure:
1310         /* Remove/restore entry since the change makes the trouble */
1311         switch (op) {
1312         case SFC_UDP_TUNNEL_ADD_PORT:
1313                 (void)efx_tunnel_config_udp_remove(sa->nic,
1314                                                    tunnel_udp->udp_port,
1315                                                    tunnel_proto);
1316                 break;
1317         case SFC_UDP_TUNNEL_DEL_PORT:
1318                 (void)efx_tunnel_config_udp_add(sa->nic,
1319                                                 tunnel_udp->udp_port,
1320                                                 tunnel_proto);
1321                 break;
1322         }
1323
1324 fail_op:
1325 fail_bad_op:
1326         sfc_adapter_unlock(sa);
1327
1328 fail_bad_proto:
1329         SFC_ASSERT(rc > 0);
1330         return -rc;
1331 }
1332
1333 static int
1334 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1335                             struct rte_eth_udp_tunnel *tunnel_udp)
1336 {
1337         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1338 }
1339
1340 static int
1341 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1342                             struct rte_eth_udp_tunnel *tunnel_udp)
1343 {
1344         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1345 }
1346
1347 static int
1348 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1349                           struct rte_eth_rss_conf *rss_conf)
1350 {
1351         struct sfc_adapter *sa = dev->data->dev_private;
1352         struct sfc_rss *rss = &sa->rss;
1353         struct sfc_port *port = &sa->port;
1354
1355         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1356                 return -ENOTSUP;
1357
1358         if (rss->channels == 0)
1359                 return -EINVAL;
1360
1361         sfc_adapter_lock(sa);
1362
1363         /*
1364          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1365          * hence, conversion is done here to derive a correct set of ETH_RSS
1366          * flags which corresponds to the active EFX configuration stored
1367          * locally in 'sfc_adapter' and kept up-to-date
1368          */
1369         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1370         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1371         if (rss_conf->rss_key != NULL)
1372                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1373
1374         sfc_adapter_unlock(sa);
1375
1376         return 0;
1377 }
1378
1379 static int
1380 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1381                         struct rte_eth_rss_conf *rss_conf)
1382 {
1383         struct sfc_adapter *sa = dev->data->dev_private;
1384         struct sfc_rss *rss = &sa->rss;
1385         struct sfc_port *port = &sa->port;
1386         unsigned int efx_hash_types;
1387         int rc = 0;
1388
1389         if (port->isolated)
1390                 return -ENOTSUP;
1391
1392         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1393                 sfc_err(sa, "RSS is not available");
1394                 return -ENOTSUP;
1395         }
1396
1397         if (rss->channels == 0) {
1398                 sfc_err(sa, "RSS is not configured");
1399                 return -EINVAL;
1400         }
1401
1402         if ((rss_conf->rss_key != NULL) &&
1403             (rss_conf->rss_key_len != sizeof(rss->key))) {
1404                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1405                         sizeof(rss->key));
1406                 return -EINVAL;
1407         }
1408
1409         sfc_adapter_lock(sa);
1410
1411         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1412         if (rc != 0)
1413                 goto fail_rx_hf_rte_to_efx;
1414
1415         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1416                                    rss->hash_alg, efx_hash_types, B_TRUE);
1417         if (rc != 0)
1418                 goto fail_scale_mode_set;
1419
1420         if (rss_conf->rss_key != NULL) {
1421                 if (sa->state == SFC_ADAPTER_STARTED) {
1422                         rc = efx_rx_scale_key_set(sa->nic,
1423                                                   EFX_RSS_CONTEXT_DEFAULT,
1424                                                   rss_conf->rss_key,
1425                                                   sizeof(rss->key));
1426                         if (rc != 0)
1427                                 goto fail_scale_key_set;
1428                 }
1429
1430                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1431         }
1432
1433         rss->hash_types = efx_hash_types;
1434
1435         sfc_adapter_unlock(sa);
1436
1437         return 0;
1438
1439 fail_scale_key_set:
1440         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1441                                   EFX_RX_HASHALG_TOEPLITZ,
1442                                   rss->hash_types, B_TRUE) != 0)
1443                 sfc_err(sa, "failed to restore RSS mode");
1444
1445 fail_scale_mode_set:
1446 fail_rx_hf_rte_to_efx:
1447         sfc_adapter_unlock(sa);
1448         return -rc;
1449 }
1450
1451 static int
1452 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1453                        struct rte_eth_rss_reta_entry64 *reta_conf,
1454                        uint16_t reta_size)
1455 {
1456         struct sfc_adapter *sa = dev->data->dev_private;
1457         struct sfc_rss *rss = &sa->rss;
1458         struct sfc_port *port = &sa->port;
1459         int entry;
1460
1461         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1462                 return -ENOTSUP;
1463
1464         if (rss->channels == 0)
1465                 return -EINVAL;
1466
1467         if (reta_size != EFX_RSS_TBL_SIZE)
1468                 return -EINVAL;
1469
1470         sfc_adapter_lock(sa);
1471
1472         for (entry = 0; entry < reta_size; entry++) {
1473                 int grp = entry / RTE_RETA_GROUP_SIZE;
1474                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1475
1476                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1477                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1478         }
1479
1480         sfc_adapter_unlock(sa);
1481
1482         return 0;
1483 }
1484
1485 static int
1486 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1487                         struct rte_eth_rss_reta_entry64 *reta_conf,
1488                         uint16_t reta_size)
1489 {
1490         struct sfc_adapter *sa = dev->data->dev_private;
1491         struct sfc_rss *rss = &sa->rss;
1492         struct sfc_port *port = &sa->port;
1493         unsigned int *rss_tbl_new;
1494         uint16_t entry;
1495         int rc = 0;
1496
1497
1498         if (port->isolated)
1499                 return -ENOTSUP;
1500
1501         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1502                 sfc_err(sa, "RSS is not available");
1503                 return -ENOTSUP;
1504         }
1505
1506         if (rss->channels == 0) {
1507                 sfc_err(sa, "RSS is not configured");
1508                 return -EINVAL;
1509         }
1510
1511         if (reta_size != EFX_RSS_TBL_SIZE) {
1512                 sfc_err(sa, "RETA size is wrong (should be %u)",
1513                         EFX_RSS_TBL_SIZE);
1514                 return -EINVAL;
1515         }
1516
1517         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1518         if (rss_tbl_new == NULL)
1519                 return -ENOMEM;
1520
1521         sfc_adapter_lock(sa);
1522
1523         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1524
1525         for (entry = 0; entry < reta_size; entry++) {
1526                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1527                 struct rte_eth_rss_reta_entry64 *grp;
1528
1529                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1530
1531                 if (grp->mask & (1ull << grp_idx)) {
1532                         if (grp->reta[grp_idx] >= rss->channels) {
1533                                 rc = EINVAL;
1534                                 goto bad_reta_entry;
1535                         }
1536                         rss_tbl_new[entry] = grp->reta[grp_idx];
1537                 }
1538         }
1539
1540         if (sa->state == SFC_ADAPTER_STARTED) {
1541                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1542                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1543                 if (rc != 0)
1544                         goto fail_scale_tbl_set;
1545         }
1546
1547         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1548
1549 fail_scale_tbl_set:
1550 bad_reta_entry:
1551         sfc_adapter_unlock(sa);
1552
1553         rte_free(rss_tbl_new);
1554
1555         SFC_ASSERT(rc >= 0);
1556         return -rc;
1557 }
1558
1559 static int
1560 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1561                     enum rte_filter_op filter_op,
1562                     void *arg)
1563 {
1564         struct sfc_adapter *sa = dev->data->dev_private;
1565         int rc = ENOTSUP;
1566
1567         sfc_log_init(sa, "entry");
1568
1569         switch (filter_type) {
1570         case RTE_ETH_FILTER_NONE:
1571                 sfc_err(sa, "Global filters configuration not supported");
1572                 break;
1573         case RTE_ETH_FILTER_MACVLAN:
1574                 sfc_err(sa, "MACVLAN filters not supported");
1575                 break;
1576         case RTE_ETH_FILTER_ETHERTYPE:
1577                 sfc_err(sa, "EtherType filters not supported");
1578                 break;
1579         case RTE_ETH_FILTER_FLEXIBLE:
1580                 sfc_err(sa, "Flexible filters not supported");
1581                 break;
1582         case RTE_ETH_FILTER_SYN:
1583                 sfc_err(sa, "SYN filters not supported");
1584                 break;
1585         case RTE_ETH_FILTER_NTUPLE:
1586                 sfc_err(sa, "NTUPLE filters not supported");
1587                 break;
1588         case RTE_ETH_FILTER_TUNNEL:
1589                 sfc_err(sa, "Tunnel filters not supported");
1590                 break;
1591         case RTE_ETH_FILTER_FDIR:
1592                 sfc_err(sa, "Flow Director filters not supported");
1593                 break;
1594         case RTE_ETH_FILTER_HASH:
1595                 sfc_err(sa, "Hash filters not supported");
1596                 break;
1597         case RTE_ETH_FILTER_GENERIC:
1598                 if (filter_op != RTE_ETH_FILTER_GET) {
1599                         rc = EINVAL;
1600                 } else {
1601                         *(const void **)arg = &sfc_flow_ops;
1602                         rc = 0;
1603                 }
1604                 break;
1605         default:
1606                 sfc_err(sa, "Unknown filter type %u", filter_type);
1607                 break;
1608         }
1609
1610         sfc_log_init(sa, "exit: %d", -rc);
1611         SFC_ASSERT(rc >= 0);
1612         return -rc;
1613 }
1614
1615 static int
1616 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1617 {
1618         struct sfc_adapter *sa = dev->data->dev_private;
1619
1620         /*
1621          * If Rx datapath does not provide callback to check mempool,
1622          * all pools are supported.
1623          */
1624         if (sa->dp_rx->pool_ops_supported == NULL)
1625                 return 1;
1626
1627         return sa->dp_rx->pool_ops_supported(pool);
1628 }
1629
1630 static const struct eth_dev_ops sfc_eth_dev_ops = {
1631         .dev_configure                  = sfc_dev_configure,
1632         .dev_start                      = sfc_dev_start,
1633         .dev_stop                       = sfc_dev_stop,
1634         .dev_set_link_up                = sfc_dev_set_link_up,
1635         .dev_set_link_down              = sfc_dev_set_link_down,
1636         .dev_close                      = sfc_dev_close,
1637         .promiscuous_enable             = sfc_dev_promisc_enable,
1638         .promiscuous_disable            = sfc_dev_promisc_disable,
1639         .allmulticast_enable            = sfc_dev_allmulti_enable,
1640         .allmulticast_disable           = sfc_dev_allmulti_disable,
1641         .link_update                    = sfc_dev_link_update,
1642         .stats_get                      = sfc_stats_get,
1643         .stats_reset                    = sfc_stats_reset,
1644         .xstats_get                     = sfc_xstats_get,
1645         .xstats_reset                   = sfc_stats_reset,
1646         .xstats_get_names               = sfc_xstats_get_names,
1647         .dev_infos_get                  = sfc_dev_infos_get,
1648         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1649         .mtu_set                        = sfc_dev_set_mtu,
1650         .rx_queue_start                 = sfc_rx_queue_start,
1651         .rx_queue_stop                  = sfc_rx_queue_stop,
1652         .tx_queue_start                 = sfc_tx_queue_start,
1653         .tx_queue_stop                  = sfc_tx_queue_stop,
1654         .rx_queue_setup                 = sfc_rx_queue_setup,
1655         .rx_queue_release               = sfc_rx_queue_release,
1656         .rx_queue_count                 = sfc_rx_queue_count,
1657         .rx_descriptor_done             = sfc_rx_descriptor_done,
1658         .rx_descriptor_status           = sfc_rx_descriptor_status,
1659         .tx_descriptor_status           = sfc_tx_descriptor_status,
1660         .tx_queue_setup                 = sfc_tx_queue_setup,
1661         .tx_queue_release               = sfc_tx_queue_release,
1662         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1663         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1664         .mac_addr_set                   = sfc_mac_addr_set,
1665         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1666         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1667         .reta_update                    = sfc_dev_rss_reta_update,
1668         .reta_query                     = sfc_dev_rss_reta_query,
1669         .rss_hash_update                = sfc_dev_rss_hash_update,
1670         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1671         .filter_ctrl                    = sfc_dev_filter_ctrl,
1672         .set_mc_addr_list               = sfc_set_mc_addr_list,
1673         .rxq_info_get                   = sfc_rx_queue_info_get,
1674         .txq_info_get                   = sfc_tx_queue_info_get,
1675         .fw_version_get                 = sfc_fw_version_get,
1676         .xstats_get_by_id               = sfc_xstats_get_by_id,
1677         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1678         .pool_ops_supported             = sfc_pool_ops_supported,
1679 };
1680
1681 /**
1682  * Duplicate a string in potentially shared memory required for
1683  * multi-process support.
1684  *
1685  * strdup() allocates from process-local heap/memory.
1686  */
1687 static char *
1688 sfc_strdup(const char *str)
1689 {
1690         size_t size;
1691         char *copy;
1692
1693         if (str == NULL)
1694                 return NULL;
1695
1696         size = strlen(str) + 1;
1697         copy = rte_malloc(__func__, size, 0);
1698         if (copy != NULL)
1699                 rte_memcpy(copy, str, size);
1700
1701         return copy;
1702 }
1703
1704 static int
1705 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1706 {
1707         struct sfc_adapter *sa = dev->data->dev_private;
1708         const efx_nic_cfg_t *encp;
1709         unsigned int avail_caps = 0;
1710         const char *rx_name = NULL;
1711         const char *tx_name = NULL;
1712         int rc;
1713
1714         switch (sa->family) {
1715         case EFX_FAMILY_HUNTINGTON:
1716         case EFX_FAMILY_MEDFORD:
1717         case EFX_FAMILY_MEDFORD2:
1718                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1719                 break;
1720         default:
1721                 break;
1722         }
1723
1724         encp = efx_nic_cfg_get(sa->nic);
1725         if (encp->enc_rx_es_super_buffer_supported)
1726                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1727
1728         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1729                                 sfc_kvarg_string_handler, &rx_name);
1730         if (rc != 0)
1731                 goto fail_kvarg_rx_datapath;
1732
1733         if (rx_name != NULL) {
1734                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1735                 if (sa->dp_rx == NULL) {
1736                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1737                         rc = ENOENT;
1738                         goto fail_dp_rx;
1739                 }
1740                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1741                         sfc_err(sa,
1742                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1743                                 rx_name);
1744                         rc = EINVAL;
1745                         goto fail_dp_rx_caps;
1746                 }
1747         } else {
1748                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1749                 if (sa->dp_rx == NULL) {
1750                         sfc_err(sa, "Rx datapath by caps %#x not found",
1751                                 avail_caps);
1752                         rc = ENOENT;
1753                         goto fail_dp_rx;
1754                 }
1755         }
1756
1757         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1758         if (sa->dp_rx_name == NULL) {
1759                 rc = ENOMEM;
1760                 goto fail_dp_rx_name;
1761         }
1762
1763         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1764
1765         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1766
1767         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1768                                 sfc_kvarg_string_handler, &tx_name);
1769         if (rc != 0)
1770                 goto fail_kvarg_tx_datapath;
1771
1772         if (tx_name != NULL) {
1773                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1774                 if (sa->dp_tx == NULL) {
1775                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1776                         rc = ENOENT;
1777                         goto fail_dp_tx;
1778                 }
1779                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1780                         sfc_err(sa,
1781                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1782                                 tx_name);
1783                         rc = EINVAL;
1784                         goto fail_dp_tx_caps;
1785                 }
1786         } else {
1787                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1788                 if (sa->dp_tx == NULL) {
1789                         sfc_err(sa, "Tx datapath by caps %#x not found",
1790                                 avail_caps);
1791                         rc = ENOENT;
1792                         goto fail_dp_tx;
1793                 }
1794         }
1795
1796         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1797         if (sa->dp_tx_name == NULL) {
1798                 rc = ENOMEM;
1799                 goto fail_dp_tx_name;
1800         }
1801
1802         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1803
1804         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1805
1806         dev->dev_ops = &sfc_eth_dev_ops;
1807
1808         return 0;
1809
1810 fail_dp_tx_name:
1811 fail_dp_tx_caps:
1812         sa->dp_tx = NULL;
1813
1814 fail_dp_tx:
1815 fail_kvarg_tx_datapath:
1816         rte_free(sa->dp_rx_name);
1817         sa->dp_rx_name = NULL;
1818
1819 fail_dp_rx_name:
1820 fail_dp_rx_caps:
1821         sa->dp_rx = NULL;
1822
1823 fail_dp_rx:
1824 fail_kvarg_rx_datapath:
1825         return rc;
1826 }
1827
1828 static void
1829 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1830 {
1831         struct sfc_adapter *sa = dev->data->dev_private;
1832
1833         dev->dev_ops = NULL;
1834         dev->rx_pkt_burst = NULL;
1835         dev->tx_pkt_burst = NULL;
1836
1837         rte_free(sa->dp_tx_name);
1838         sa->dp_tx_name = NULL;
1839         sa->dp_tx = NULL;
1840
1841         rte_free(sa->dp_rx_name);
1842         sa->dp_rx_name = NULL;
1843         sa->dp_rx = NULL;
1844 }
1845
1846 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1847         .rxq_info_get                   = sfc_rx_queue_info_get,
1848         .txq_info_get                   = sfc_tx_queue_info_get,
1849 };
1850
1851 static int
1852 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1853 {
1854         /*
1855          * Device private data has really many process-local pointers.
1856          * Below code should be extremely careful to use data located
1857          * in shared memory only.
1858          */
1859         struct sfc_adapter *sa = dev->data->dev_private;
1860         const struct sfc_dp_rx *dp_rx;
1861         const struct sfc_dp_tx *dp_tx;
1862         int rc;
1863
1864         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1865         if (dp_rx == NULL) {
1866                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1867                 rc = ENOENT;
1868                 goto fail_dp_rx;
1869         }
1870         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1871                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1872                         sa->dp_tx_name);
1873                 rc = EINVAL;
1874                 goto fail_dp_rx_multi_process;
1875         }
1876
1877         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1878         if (dp_tx == NULL) {
1879                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1880                 rc = ENOENT;
1881                 goto fail_dp_tx;
1882         }
1883         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1884                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1885                         sa->dp_tx_name);
1886                 rc = EINVAL;
1887                 goto fail_dp_tx_multi_process;
1888         }
1889
1890         dev->rx_pkt_burst = dp_rx->pkt_burst;
1891         dev->tx_pkt_burst = dp_tx->pkt_burst;
1892         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1893
1894         return 0;
1895
1896 fail_dp_tx_multi_process:
1897 fail_dp_tx:
1898 fail_dp_rx_multi_process:
1899 fail_dp_rx:
1900         return rc;
1901 }
1902
1903 static void
1904 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1905 {
1906         dev->dev_ops = NULL;
1907         dev->tx_pkt_burst = NULL;
1908         dev->rx_pkt_burst = NULL;
1909 }
1910
1911 static void
1912 sfc_register_dp(void)
1913 {
1914         /* Register once */
1915         if (TAILQ_EMPTY(&sfc_dp_head)) {
1916                 /* Prefer EF10 datapath */
1917                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1918                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1919                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1920
1921                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1922                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1923                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1924         }
1925 }
1926
1927 static int
1928 sfc_eth_dev_init(struct rte_eth_dev *dev)
1929 {
1930         struct sfc_adapter *sa = dev->data->dev_private;
1931         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1932         int rc;
1933         const efx_nic_cfg_t *encp;
1934         const struct ether_addr *from;
1935
1936         sfc_register_dp();
1937
1938         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1939                 return -sfc_eth_dev_secondary_set_ops(dev);
1940
1941         /* Required for logging */
1942         sa->pci_addr = pci_dev->addr;
1943         sa->port_id = dev->data->port_id;
1944
1945         sa->eth_dev = dev;
1946
1947         /* Copy PCI device info to the dev->data */
1948         rte_eth_copy_pci_info(dev, pci_dev);
1949
1950         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1951                                                 RTE_LOG_NOTICE);
1952
1953         rc = sfc_kvargs_parse(sa);
1954         if (rc != 0)
1955                 goto fail_kvargs_parse;
1956
1957         sfc_log_init(sa, "entry");
1958
1959         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1960         if (dev->data->mac_addrs == NULL) {
1961                 rc = ENOMEM;
1962                 goto fail_mac_addrs;
1963         }
1964
1965         sfc_adapter_lock_init(sa);
1966         sfc_adapter_lock(sa);
1967
1968         sfc_log_init(sa, "probing");
1969         rc = sfc_probe(sa);
1970         if (rc != 0)
1971                 goto fail_probe;
1972
1973         sfc_log_init(sa, "set device ops");
1974         rc = sfc_eth_dev_set_ops(dev);
1975         if (rc != 0)
1976                 goto fail_set_ops;
1977
1978         sfc_log_init(sa, "attaching");
1979         rc = sfc_attach(sa);
1980         if (rc != 0)
1981                 goto fail_attach;
1982
1983         encp = efx_nic_cfg_get(sa->nic);
1984
1985         /*
1986          * The arguments are really reverse order in comparison to
1987          * Linux kernel. Copy from NIC config to Ethernet device data.
1988          */
1989         from = (const struct ether_addr *)(encp->enc_mac_addr);
1990         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1991
1992         sfc_adapter_unlock(sa);
1993
1994         sfc_log_init(sa, "done");
1995         return 0;
1996
1997 fail_attach:
1998         sfc_eth_dev_clear_ops(dev);
1999
2000 fail_set_ops:
2001         sfc_unprobe(sa);
2002
2003 fail_probe:
2004         sfc_adapter_unlock(sa);
2005         sfc_adapter_lock_fini(sa);
2006         rte_free(dev->data->mac_addrs);
2007         dev->data->mac_addrs = NULL;
2008
2009 fail_mac_addrs:
2010         sfc_kvargs_cleanup(sa);
2011
2012 fail_kvargs_parse:
2013         sfc_log_init(sa, "failed %d", rc);
2014         SFC_ASSERT(rc > 0);
2015         return -rc;
2016 }
2017
2018 static int
2019 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2020 {
2021         struct sfc_adapter *sa;
2022
2023         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2024                 sfc_eth_dev_secondary_clear_ops(dev);
2025                 return 0;
2026         }
2027
2028         sa = dev->data->dev_private;
2029         sfc_log_init(sa, "entry");
2030
2031         sfc_adapter_lock(sa);
2032
2033         sfc_eth_dev_clear_ops(dev);
2034
2035         sfc_detach(sa);
2036         sfc_unprobe(sa);
2037
2038         rte_free(dev->data->mac_addrs);
2039         dev->data->mac_addrs = NULL;
2040
2041         sfc_kvargs_cleanup(sa);
2042
2043         sfc_adapter_unlock(sa);
2044         sfc_adapter_lock_fini(sa);
2045
2046         sfc_log_init(sa, "done");
2047
2048         /* Required for logging, so cleanup last */
2049         sa->eth_dev = NULL;
2050         return 0;
2051 }
2052
2053 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2054         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2055         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2056         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2057         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2058         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2059         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2060         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2061         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2062         { .vendor_id = 0 /* sentinel */ }
2063 };
2064
2065 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2066         struct rte_pci_device *pci_dev)
2067 {
2068         return rte_eth_dev_pci_generic_probe(pci_dev,
2069                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2070 }
2071
2072 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2073 {
2074         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2075 }
2076
2077 static struct rte_pci_driver sfc_efx_pmd = {
2078         .id_table = pci_id_sfc_efx_map,
2079         .drv_flags =
2080                 RTE_PCI_DRV_INTR_LSC |
2081                 RTE_PCI_DRV_NEED_MAPPING,
2082         .probe = sfc_eth_dev_pci_probe,
2083         .remove = sfc_eth_dev_pci_remove,
2084 };
2085
2086 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2087 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2088 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2089 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2090         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2091         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2092         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2093         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2094         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2095         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2096
2097 RTE_INIT(sfc_driver_register_logtype);
2098 static void
2099 sfc_driver_register_logtype(void)
2100 {
2101         int ret;
2102
2103         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2104                                                    RTE_LOG_NOTICE);
2105         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2106 }