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
  17 #include "efx.h"
  18
  19 #include "sfc.h"
  20 #include "sfc_debug.h"
  21 #include "sfc_log.h"
  22 #include "sfc_kvargs.h"
  23 #include "sfc_ev.h"
  24 #include "sfc_rx.h"
  25 #include "sfc_tx.h"
  26 #include "sfc_flow.h"
  27 #include "sfc_dp.h"
  28 #include "sfc_dp_rx.h"
  29
  30 uint32_t sfc_logtype_driver;
  31
  32 static struct sfc_dp_list sfc_dp_head =
  33         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  34
  35 static int
  36 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  37 {
  38         struct sfc_adapter *sa = dev->data->dev_private;
  39         efx_nic_fw_info_t enfi;
  40         int ret;
  41         int rc;
  42
  43         /*
  44          * Return value of the callback is likely supposed to be
  45          * equal to or greater than 0, nevertheless, if an error
  46          * occurs, it will be desirable to pass it to the caller
  47          */
  48         if ((fw_version == NULL) || (fw_size == 0))
  49                 return -EINVAL;
  50
  51         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  52         if (rc != 0)
  53                 return -rc;
  54
  55         ret = snprintf(fw_version, fw_size,
  56                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  57                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  58                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  59         if (ret < 0)
  60                 return ret;
  61
  62         if (enfi.enfi_dpcpu_fw_ids_valid) {
  63                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  64                 int ret_extra;
  65
  66                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  67                                      fw_size - dpcpu_fw_ids_offset,
  68                                      " rx%" PRIx16 " tx%" PRIx16,
  69                                      enfi.enfi_rx_dpcpu_fw_id,
  70                                      enfi.enfi_tx_dpcpu_fw_id);
  71                 if (ret_extra < 0)
  72                         return ret_extra;
  73
  74                 ret += ret_extra;
  75         }
  76
  77         if (fw_size < (size_t)(++ret))
  78                 return ret;
  79         else
  80                 return 0;
  81 }
  82
  83 static void
  84 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  85 {
  86         struct sfc_adapter *sa = dev->data->dev_private;
  87         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
  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         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 142         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 143             !encp->enc_hw_tx_insert_vlan_enabled)
 144                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 145
 146         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 147                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 148
 149         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 150                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 151
 152         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 153                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 154
 155         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 156                 uint64_t rte_hf = 0;
 157                 unsigned int i;
 158
 159                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 160                         rte_hf |= rss->hf_map[i].rte;
 161
 162                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 163                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 164                 dev_info->flow_type_rss_offloads = rte_hf;
 165         }
 166
 167         /* Initialize to hardware limits */
 168         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 169         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 170         /* The RXQ hardware requires that the descriptor count is a power
 171          * of 2, but rx_desc_lim cannot properly describe that constraint.
 172          */
 173         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 174
 175         /* Initialize to hardware limits */
 176         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 177         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 178         /*
 179          * The TXQ hardware requires that the descriptor count is a power
 180          * of 2, but tx_desc_lim cannot properly describe that constraint
 181          */
 182         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 183
 184         if (sa->dp_rx->get_dev_info != NULL)
 185                 sa->dp_rx->get_dev_info(dev_info);
 186         if (sa->dp_tx->get_dev_info != NULL)
 187                 sa->dp_tx->get_dev_info(dev_info);
 188 }
 189
 190 static const uint32_t *
 191 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 192 {
 193         struct sfc_adapter *sa = dev->data->dev_private;
 194         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 195         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 196
 197         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 198 }
 199
 200 static int
 201 sfc_dev_configure(struct rte_eth_dev *dev)
 202 {
 203         struct rte_eth_dev_data *dev_data = dev->data;
 204         struct sfc_adapter *sa = dev_data->dev_private;
 205         int rc;
 206
 207         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 208                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 209
 210         sfc_adapter_lock(sa);
 211         switch (sa->state) {
 212         case SFC_ADAPTER_CONFIGURED:
 213                 /* FALLTHROUGH */
 214         case SFC_ADAPTER_INITIALIZED:
 215                 rc = sfc_configure(sa);
 216                 break;
 217         default:
 218                 sfc_err(sa, "unexpected adapter state %u to configure",
 219                         sa->state);
 220                 rc = EINVAL;
 221                 break;
 222         }
 223         sfc_adapter_unlock(sa);
 224
 225         sfc_log_init(sa, "done %d", rc);
 226         SFC_ASSERT(rc >= 0);
 227         return -rc;
 228 }
 229
 230 static int
 231 sfc_dev_start(struct rte_eth_dev *dev)
 232 {
 233         struct sfc_adapter *sa = dev->data->dev_private;
 234         int rc;
 235
 236         sfc_log_init(sa, "entry");
 237
 238         sfc_adapter_lock(sa);
 239         rc = sfc_start(sa);
 240         sfc_adapter_unlock(sa);
 241
 242         sfc_log_init(sa, "done %d", rc);
 243         SFC_ASSERT(rc >= 0);
 244         return -rc;
 245 }
 246
 247 static int
 248 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 249 {
 250         struct sfc_adapter *sa = dev->data->dev_private;
 251         struct rte_eth_link current_link;
 252         int ret;
 253
 254         sfc_log_init(sa, "entry");
 255
 256         if (sa->state != SFC_ADAPTER_STARTED) {
 257                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 258         } else if (wait_to_complete) {
 259                 efx_link_mode_t link_mode;
 260
 261                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 262                         link_mode = EFX_LINK_UNKNOWN;
 263                 sfc_port_link_mode_to_info(link_mode, &current_link);
 264
 265         } else {
 266                 sfc_ev_mgmt_qpoll(sa);
 267                 rte_eth_linkstatus_get(dev, &current_link);
 268         }
 269
 270         ret = rte_eth_linkstatus_set(dev, &current_link);
 271         if (ret == 0)
 272                 sfc_notice(sa, "Link status is %s",
 273                            current_link.link_status ? "UP" : "DOWN");
 274
 275         return ret;
 276 }
 277
 278 static void
 279 sfc_dev_stop(struct rte_eth_dev *dev)
 280 {
 281         struct sfc_adapter *sa = dev->data->dev_private;
 282
 283         sfc_log_init(sa, "entry");
 284
 285         sfc_adapter_lock(sa);
 286         sfc_stop(sa);
 287         sfc_adapter_unlock(sa);
 288
 289         sfc_log_init(sa, "done");
 290 }
 291
 292 static int
 293 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 294 {
 295         struct sfc_adapter *sa = dev->data->dev_private;
 296         int rc;
 297
 298         sfc_log_init(sa, "entry");
 299
 300         sfc_adapter_lock(sa);
 301         rc = sfc_start(sa);
 302         sfc_adapter_unlock(sa);
 303
 304         SFC_ASSERT(rc >= 0);
 305         return -rc;
 306 }
 307
 308 static int
 309 sfc_dev_set_link_down(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         sfc_stop(sa);
 317         sfc_adapter_unlock(sa);
 318
 319         return 0;
 320 }
 321
 322 static void
 323 sfc_dev_close(struct rte_eth_dev *dev)
 324 {
 325         struct sfc_adapter *sa = dev->data->dev_private;
 326
 327         sfc_log_init(sa, "entry");
 328
 329         sfc_adapter_lock(sa);
 330         switch (sa->state) {
 331         case SFC_ADAPTER_STARTED:
 332                 sfc_stop(sa);
 333                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 334                 /* FALLTHROUGH */
 335         case SFC_ADAPTER_CONFIGURED:
 336                 sfc_close(sa);
 337                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 338                 /* FALLTHROUGH */
 339         case SFC_ADAPTER_INITIALIZED:
 340                 break;
 341         default:
 342                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 343                 break;
 344         }
 345         sfc_adapter_unlock(sa);
 346
 347         sfc_log_init(sa, "done");
 348 }
 349
 350 static void
 351 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 352                    boolean_t enabled)
 353 {
 354         struct sfc_port *port;
 355         boolean_t *toggle;
 356         struct sfc_adapter *sa = dev->data->dev_private;
 357         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 358         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 359
 360         sfc_adapter_lock(sa);
 361
 362         port = &sa->port;
 363         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 364
 365         if (*toggle != enabled) {
 366                 *toggle = enabled;
 367
 368                 if (port->isolated) {
 369                         sfc_warn(sa, "isolated mode is active on the port");
 370                         sfc_warn(sa, "the change is to be applied on the next "
 371                                      "start provided that isolated mode is "
 372                                      "disabled prior the next start");
 373                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 374                            (sfc_set_rx_mode(sa) != 0)) {
 375                         *toggle = !(enabled);
 376                         sfc_warn(sa, "Failed to %s %s mode",
 377                                  ((enabled) ? "enable" : "disable"), desc);
 378                 }
 379         }
 380
 381         sfc_adapter_unlock(sa);
 382 }
 383
 384 static void
 385 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 386 {
 387         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 388 }
 389
 390 static void
 391 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 392 {
 393         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 394 }
 395
 396 static void
 397 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 398 {
 399         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 400 }
 401
 402 static void
 403 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 404 {
 405         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 406 }
 407
 408 static int
 409 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 410                    uint16_t nb_rx_desc, unsigned int socket_id,
 411                    const struct rte_eth_rxconf *rx_conf,
 412                    struct rte_mempool *mb_pool)
 413 {
 414         struct sfc_adapter *sa = dev->data->dev_private;
 415         int rc;
 416
 417         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 418                      rx_queue_id, nb_rx_desc, socket_id);
 419
 420         sfc_adapter_lock(sa);
 421
 422         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 423                           rx_conf, mb_pool);
 424         if (rc != 0)
 425                 goto fail_rx_qinit;
 426
 427         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 428
 429         sfc_adapter_unlock(sa);
 430
 431         return 0;
 432
 433 fail_rx_qinit:
 434         sfc_adapter_unlock(sa);
 435         SFC_ASSERT(rc > 0);
 436         return -rc;
 437 }
 438
 439 static void
 440 sfc_rx_queue_release(void *queue)
 441 {
 442         struct sfc_dp_rxq *dp_rxq = queue;
 443         struct sfc_rxq *rxq;
 444         struct sfc_adapter *sa;
 445         unsigned int sw_index;
 446
 447         if (dp_rxq == NULL)
 448                 return;
 449
 450         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 451         sa = rxq->evq->sa;
 452         sfc_adapter_lock(sa);
 453
 454         sw_index = sfc_rxq_sw_index(rxq);
 455
 456         sfc_log_init(sa, "RxQ=%u", sw_index);
 457
 458         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 459
 460         sfc_rx_qfini(sa, sw_index);
 461
 462         sfc_adapter_unlock(sa);
 463 }
 464
 465 static int
 466 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 467                    uint16_t nb_tx_desc, unsigned int socket_id,
 468                    const struct rte_eth_txconf *tx_conf)
 469 {
 470         struct sfc_adapter *sa = dev->data->dev_private;
 471         int rc;
 472
 473         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 474                      tx_queue_id, nb_tx_desc, socket_id);
 475
 476         sfc_adapter_lock(sa);
 477
 478         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 479         if (rc != 0)
 480                 goto fail_tx_qinit;
 481
 482         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 483
 484         sfc_adapter_unlock(sa);
 485         return 0;
 486
 487 fail_tx_qinit:
 488         sfc_adapter_unlock(sa);
 489         SFC_ASSERT(rc > 0);
 490         return -rc;
 491 }
 492
 493 static void
 494 sfc_tx_queue_release(void *queue)
 495 {
 496         struct sfc_dp_txq *dp_txq = queue;
 497         struct sfc_txq *txq;
 498         unsigned int sw_index;
 499         struct sfc_adapter *sa;
 500
 501         if (dp_txq == NULL)
 502                 return;
 503
 504         txq = sfc_txq_by_dp_txq(dp_txq);
 505         sw_index = sfc_txq_sw_index(txq);
 506
 507         SFC_ASSERT(txq->evq != NULL);
 508         sa = txq->evq->sa;
 509
 510         sfc_log_init(sa, "TxQ = %u", sw_index);
 511
 512         sfc_adapter_lock(sa);
 513
 514         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 515         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 516
 517         sfc_tx_qfini(sa, sw_index);
 518
 519         sfc_adapter_unlock(sa);
 520 }
 521
 522 static int
 523 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 524 {
 525         struct sfc_adapter *sa = dev->data->dev_private;
 526         struct sfc_port *port = &sa->port;
 527         uint64_t *mac_stats;
 528         int ret;
 529
 530         rte_spinlock_lock(&port->mac_stats_lock);
 531
 532         ret = sfc_port_update_mac_stats(sa);
 533         if (ret != 0)
 534                 goto unlock;
 535
 536         mac_stats = port->mac_stats_buf;
 537
 538         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 539                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 540                 stats->ipackets =
 541                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 542                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 543                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 544                 stats->opackets =
 545                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 546                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 547                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 548                 stats->ibytes =
 549                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 550                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 551                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 552                 stats->obytes =
 553                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 554                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 555                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 556                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 557                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 558                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 559         } else {
 560                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 561                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 562                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 563                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 564                 /*
 565                  * Take into account stats which are whenever supported
 566                  * on EF10. If some stat is not supported by current
 567                  * firmware variant or HW revision, it is guaranteed
 568                  * to be zero in mac_stats.
 569                  */
 570                 stats->imissed =
 571                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 572                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 573                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 574                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 575                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 576                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 577                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 578                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 579                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 580                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 581                 stats->ierrors =
 582                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 583                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 584                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 585                 /* no oerrors counters supported on EF10 */
 586         }
 587
 588 unlock:
 589         rte_spinlock_unlock(&port->mac_stats_lock);
 590         SFC_ASSERT(ret >= 0);
 591         return -ret;
 592 }
 593
 594 static void
 595 sfc_stats_reset(struct rte_eth_dev *dev)
 596 {
 597         struct sfc_adapter *sa = dev->data->dev_private;
 598         struct sfc_port *port = &sa->port;
 599         int rc;
 600
 601         if (sa->state != SFC_ADAPTER_STARTED) {
 602                 /*
 603                  * The operation cannot be done if port is not started; it
 604                  * will be scheduled to be done during the next port start
 605                  */
 606                 port->mac_stats_reset_pending = B_TRUE;
 607                 return;
 608         }
 609
 610         rc = sfc_port_reset_mac_stats(sa);
 611         if (rc != 0)
 612                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 613 }
 614
 615 static int
 616 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 617                unsigned int xstats_count)
 618 {
 619         struct sfc_adapter *sa = dev->data->dev_private;
 620         struct sfc_port *port = &sa->port;
 621         uint64_t *mac_stats;
 622         int rc;
 623         unsigned int i;
 624         int nstats = 0;
 625
 626         rte_spinlock_lock(&port->mac_stats_lock);
 627
 628         rc = sfc_port_update_mac_stats(sa);
 629         if (rc != 0) {
 630                 SFC_ASSERT(rc > 0);
 631                 nstats = -rc;
 632                 goto unlock;
 633         }
 634
 635         mac_stats = port->mac_stats_buf;
 636
 637         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 638                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 639                         if (xstats != NULL && nstats < (int)xstats_count) {
 640                                 xstats[nstats].id = nstats;
 641                                 xstats[nstats].value = mac_stats[i];
 642                         }
 643                         nstats++;
 644                 }
 645         }
 646
 647 unlock:
 648         rte_spinlock_unlock(&port->mac_stats_lock);
 649
 650         return nstats;
 651 }
 652
 653 static int
 654 sfc_xstats_get_names(struct rte_eth_dev *dev,
 655                      struct rte_eth_xstat_name *xstats_names,
 656                      unsigned int xstats_count)
 657 {
 658         struct sfc_adapter *sa = dev->data->dev_private;
 659         struct sfc_port *port = &sa->port;
 660         unsigned int i;
 661         unsigned int nstats = 0;
 662
 663         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 664                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 665                         if (xstats_names != NULL && nstats < xstats_count)
 666                                 strncpy(xstats_names[nstats].name,
 667                                         efx_mac_stat_name(sa->nic, i),
 668                                         sizeof(xstats_names[0].name));
 669                         nstats++;
 670                 }
 671         }
 672
 673         return nstats;
 674 }
 675
 676 static int
 677 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 678                      uint64_t *values, unsigned int n)
 679 {
 680         struct sfc_adapter *sa = dev->data->dev_private;
 681         struct sfc_port *port = &sa->port;
 682         uint64_t *mac_stats;
 683         unsigned int nb_supported = 0;
 684         unsigned int nb_written = 0;
 685         unsigned int i;
 686         int ret;
 687         int rc;
 688
 689         if (unlikely(values == NULL) ||
 690             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 691                 return port->mac_stats_nb_supported;
 692
 693         rte_spinlock_lock(&port->mac_stats_lock);
 694
 695         rc = sfc_port_update_mac_stats(sa);
 696         if (rc != 0) {
 697                 SFC_ASSERT(rc > 0);
 698                 ret = -rc;
 699                 goto unlock;
 700         }
 701
 702         mac_stats = port->mac_stats_buf;
 703
 704         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 705                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 706                         continue;
 707
 708                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 709                         values[nb_written++] = mac_stats[i];
 710
 711                 ++nb_supported;
 712         }
 713
 714         ret = nb_written;
 715
 716 unlock:
 717         rte_spinlock_unlock(&port->mac_stats_lock);
 718
 719         return ret;
 720 }
 721
 722 static int
 723 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 724                            struct rte_eth_xstat_name *xstats_names,
 725                            const uint64_t *ids, unsigned int size)
 726 {
 727         struct sfc_adapter *sa = dev->data->dev_private;
 728         struct sfc_port *port = &sa->port;
 729         unsigned int nb_supported = 0;
 730         unsigned int nb_written = 0;
 731         unsigned int i;
 732
 733         if (unlikely(xstats_names == NULL) ||
 734             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 735                 return port->mac_stats_nb_supported;
 736
 737         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 738                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 739                         continue;
 740
 741                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 742                         char *name = xstats_names[nb_written++].name;
 743
 744                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 745                                 sizeof(xstats_names[0].name));
 746                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 747                 }
 748
 749                 ++nb_supported;
 750         }
 751
 752         return nb_written;
 753 }
 754
 755 static int
 756 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 757 {
 758         struct sfc_adapter *sa = dev->data->dev_private;
 759         unsigned int wanted_fc, link_fc;
 760
 761         memset(fc_conf, 0, sizeof(*fc_conf));
 762
 763         sfc_adapter_lock(sa);
 764
 765         if (sa->state == SFC_ADAPTER_STARTED)
 766                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 767         else
 768                 link_fc = sa->port.flow_ctrl;
 769
 770         switch (link_fc) {
 771         case 0:
 772                 fc_conf->mode = RTE_FC_NONE;
 773                 break;
 774         case EFX_FCNTL_RESPOND:
 775                 fc_conf->mode = RTE_FC_RX_PAUSE;
 776                 break;
 777         case EFX_FCNTL_GENERATE:
 778                 fc_conf->mode = RTE_FC_TX_PAUSE;
 779                 break;
 780         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 781                 fc_conf->mode = RTE_FC_FULL;
 782                 break;
 783         default:
 784                 sfc_err(sa, "%s: unexpected flow control value %#x",
 785                         __func__, link_fc);
 786         }
 787
 788         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 789
 790         sfc_adapter_unlock(sa);
 791
 792         return 0;
 793 }
 794
 795 static int
 796 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 797 {
 798         struct sfc_adapter *sa = dev->data->dev_private;
 799         struct sfc_port *port = &sa->port;
 800         unsigned int fcntl;
 801         int rc;
 802
 803         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 804             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 805             fc_conf->mac_ctrl_frame_fwd != 0) {
 806                 sfc_err(sa, "unsupported flow control settings specified");
 807                 rc = EINVAL;
 808                 goto fail_inval;
 809         }
 810
 811         switch (fc_conf->mode) {
 812         case RTE_FC_NONE:
 813                 fcntl = 0;
 814                 break;
 815         case RTE_FC_RX_PAUSE:
 816                 fcntl = EFX_FCNTL_RESPOND;
 817                 break;
 818         case RTE_FC_TX_PAUSE:
 819                 fcntl = EFX_FCNTL_GENERATE;
 820                 break;
 821         case RTE_FC_FULL:
 822                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 823                 break;
 824         default:
 825                 rc = EINVAL;
 826                 goto fail_inval;
 827         }
 828
 829         sfc_adapter_lock(sa);
 830
 831         if (sa->state == SFC_ADAPTER_STARTED) {
 832                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 833                 if (rc != 0)
 834                         goto fail_mac_fcntl_set;
 835         }
 836
 837         port->flow_ctrl = fcntl;
 838         port->flow_ctrl_autoneg = fc_conf->autoneg;
 839
 840         sfc_adapter_unlock(sa);
 841
 842         return 0;
 843
 844 fail_mac_fcntl_set:
 845         sfc_adapter_unlock(sa);
 846 fail_inval:
 847         SFC_ASSERT(rc > 0);
 848         return -rc;
 849 }
 850
 851 static int
 852 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 853 {
 854         struct sfc_adapter *sa = dev->data->dev_private;
 855         size_t pdu = EFX_MAC_PDU(mtu);
 856         size_t old_pdu;
 857         int rc;
 858
 859         sfc_log_init(sa, "mtu=%u", mtu);
 860
 861         rc = EINVAL;
 862         if (pdu < EFX_MAC_PDU_MIN) {
 863                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 864                         (unsigned int)mtu, (unsigned int)pdu,
 865                         EFX_MAC_PDU_MIN);
 866                 goto fail_inval;
 867         }
 868         if (pdu > EFX_MAC_PDU_MAX) {
 869                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 870                         (unsigned int)mtu, (unsigned int)pdu,
 871                         EFX_MAC_PDU_MAX);
 872                 goto fail_inval;
 873         }
 874
 875         sfc_adapter_lock(sa);
 876
 877         if (pdu != sa->port.pdu) {
 878                 if (sa->state == SFC_ADAPTER_STARTED) {
 879                         sfc_stop(sa);
 880
 881                         old_pdu = sa->port.pdu;
 882                         sa->port.pdu = pdu;
 883                         rc = sfc_start(sa);
 884                         if (rc != 0)
 885                                 goto fail_start;
 886                 } else {
 887                         sa->port.pdu = pdu;
 888                 }
 889         }
 890
 891         /*
 892          * The driver does not use it, but other PMDs update jumbo_frame
 893          * flag and max_rx_pkt_len when MTU is set.
 894          */
 895         if (mtu > ETHER_MAX_LEN) {
 896                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 897
 898                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 899                 rxmode->jumbo_frame = 1;
 900         }
 901
 902         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 903
 904         sfc_adapter_unlock(sa);
 905
 906         sfc_log_init(sa, "done");
 907         return 0;
 908
 909 fail_start:
 910         sa->port.pdu = old_pdu;
 911         if (sfc_start(sa) != 0)
 912                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 913                         "PDU max size - port is stopped",
 914                         (unsigned int)pdu, (unsigned int)old_pdu);
 915         sfc_adapter_unlock(sa);
 916
 917 fail_inval:
 918         sfc_log_init(sa, "failed %d", rc);
 919         SFC_ASSERT(rc > 0);
 920         return -rc;
 921 }
 922 static int
 923 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 924 {
 925         struct sfc_adapter *sa = dev->data->dev_private;
 926         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 927         struct sfc_port *port = &sa->port;
 928         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
 929         int rc = 0;
 930
 931         sfc_adapter_lock(sa);
 932
 933         /*
 934          * Copy the address to the device private data so that
 935          * it could be recalled in the case of adapter restart.
 936          */
 937         ether_addr_copy(mac_addr, &port->default_mac_addr);
 938
 939         /*
 940          * Neither of the two following checks can return
 941          * an error. The new MAC address is preserved in
 942          * the device private data and can be activated
 943          * on the next port start if the user prevents
 944          * isolated mode from being enabled.
 945          */
 946         if (port->isolated) {
 947                 sfc_warn(sa, "isolated mode is active on the port");
 948                 sfc_warn(sa, "will not set MAC address");
 949                 goto unlock;
 950         }
 951
 952         if (sa->state != SFC_ADAPTER_STARTED) {
 953                 sfc_notice(sa, "the port is not started");
 954                 sfc_notice(sa, "the new MAC address will be set on port start");
 955
 956                 goto unlock;
 957         }
 958
 959         if (encp->enc_allow_set_mac_with_installed_filters) {
 960                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 961                 if (rc != 0) {
 962                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 963                         goto unlock;
 964                 }
 965
 966                 /*
 967                  * Changing the MAC address by means of MCDI request
 968                  * has no effect on received traffic, therefore
 969                  * we also need to update unicast filters
 970                  */
 971                 rc = sfc_set_rx_mode(sa);
 972                 if (rc != 0) {
 973                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 974                         /* Rollback the old address */
 975                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
 976                         (void)sfc_set_rx_mode(sa);
 977                 }
 978         } else {
 979                 sfc_warn(sa, "cannot set MAC address with filters installed");
 980                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 981                 sfc_warn(sa, "(some traffic may be dropped)");
 982
 983                 /*
 984                  * Since setting MAC address with filters installed is not
 985                  * allowed on the adapter, the new MAC address will be set
 986                  * by means of adapter restart. sfc_start() shall retrieve
 987                  * the new address from the device private data and set it.
 988                  */
 989                 sfc_stop(sa);
 990                 rc = sfc_start(sa);
 991                 if (rc != 0)
 992                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 993         }
 994
 995 unlock:
 996         if (rc != 0)
 997                 ether_addr_copy(old_addr, &port->default_mac_addr);
 998
 999         sfc_adapter_unlock(sa);
1000
1001         SFC_ASSERT(rc >= 0);
1002         return -rc;
1003 }
1004
1005
1006 static int
1007 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1008                      uint32_t nb_mc_addr)
1009 {
1010         struct sfc_adapter *sa = dev->data->dev_private;
1011         struct sfc_port *port = &sa->port;
1012         uint8_t *mc_addrs = port->mcast_addrs;
1013         int rc;
1014         unsigned int i;
1015
1016         if (port->isolated) {
1017                 sfc_err(sa, "isolated mode is active on the port");
1018                 sfc_err(sa, "will not set multicast address list");
1019                 return -ENOTSUP;
1020         }
1021
1022         if (mc_addrs == NULL)
1023                 return -ENOBUFS;
1024
1025         if (nb_mc_addr > port->max_mcast_addrs) {
1026                 sfc_err(sa, "too many multicast addresses: %u > %u",
1027                          nb_mc_addr, port->max_mcast_addrs);
1028                 return -EINVAL;
1029         }
1030
1031         for (i = 0; i < nb_mc_addr; ++i) {
1032                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1033                                  EFX_MAC_ADDR_LEN);
1034                 mc_addrs += EFX_MAC_ADDR_LEN;
1035         }
1036
1037         port->nb_mcast_addrs = nb_mc_addr;
1038
1039         if (sa->state != SFC_ADAPTER_STARTED)
1040                 return 0;
1041
1042         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1043                                         port->nb_mcast_addrs);
1044         if (rc != 0)
1045                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1046
1047         SFC_ASSERT(rc > 0);
1048         return -rc;
1049 }
1050
1051 /*
1052  * The function is used by the secondary process as well. It must not
1053  * use any process-local pointers from the adapter data.
1054  */
1055 static void
1056 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1057                       struct rte_eth_rxq_info *qinfo)
1058 {
1059         struct sfc_adapter *sa = dev->data->dev_private;
1060         struct sfc_rxq_info *rxq_info;
1061         struct sfc_rxq *rxq;
1062
1063         sfc_adapter_lock(sa);
1064
1065         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1066
1067         rxq_info = &sa->rxq_info[rx_queue_id];
1068         rxq = rxq_info->rxq;
1069         SFC_ASSERT(rxq != NULL);
1070
1071         qinfo->mp = rxq->refill_mb_pool;
1072         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1073         qinfo->conf.rx_drop_en = 1;
1074         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1075         qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1076                                DEV_RX_OFFLOAD_UDP_CKSUM |
1077                                DEV_RX_OFFLOAD_TCP_CKSUM;
1078         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1079                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1080                 qinfo->scattered_rx = 1;
1081         }
1082         qinfo->nb_desc = rxq_info->entries;
1083
1084         sfc_adapter_unlock(sa);
1085 }
1086
1087 /*
1088  * The function is used by the secondary process as well. It must not
1089  * use any process-local pointers from the adapter data.
1090  */
1091 static void
1092 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1093                       struct rte_eth_txq_info *qinfo)
1094 {
1095         struct sfc_adapter *sa = dev->data->dev_private;
1096         struct sfc_txq_info *txq_info;
1097
1098         sfc_adapter_lock(sa);
1099
1100         SFC_ASSERT(tx_queue_id < sa->txq_count);
1101
1102         txq_info = &sa->txq_info[tx_queue_id];
1103         SFC_ASSERT(txq_info->txq != NULL);
1104
1105         memset(qinfo, 0, sizeof(*qinfo));
1106
1107         qinfo->conf.txq_flags = txq_info->txq->flags;
1108         qinfo->conf.offloads = txq_info->txq->offloads;
1109         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1110         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1111         qinfo->nb_desc = txq_info->entries;
1112
1113         sfc_adapter_unlock(sa);
1114 }
1115
1116 static uint32_t
1117 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1118 {
1119         struct sfc_adapter *sa = dev->data->dev_private;
1120
1121         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1122
1123         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1124 }
1125
1126 static int
1127 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1128 {
1129         struct sfc_dp_rxq *dp_rxq = queue;
1130
1131         return sfc_rx_qdesc_done(dp_rxq, offset);
1132 }
1133
1134 static int
1135 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1136 {
1137         struct sfc_dp_rxq *dp_rxq = queue;
1138         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1139
1140         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1141 }
1142
1143 static int
1144 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1145 {
1146         struct sfc_dp_txq *dp_txq = queue;
1147         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1148
1149         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1150 }
1151
1152 static int
1153 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1154 {
1155         struct sfc_adapter *sa = dev->data->dev_private;
1156         int rc;
1157
1158         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1159
1160         sfc_adapter_lock(sa);
1161
1162         rc = EINVAL;
1163         if (sa->state != SFC_ADAPTER_STARTED)
1164                 goto fail_not_started;
1165
1166         rc = sfc_rx_qstart(sa, rx_queue_id);
1167         if (rc != 0)
1168                 goto fail_rx_qstart;
1169
1170         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1171
1172         sfc_adapter_unlock(sa);
1173
1174         return 0;
1175
1176 fail_rx_qstart:
1177 fail_not_started:
1178         sfc_adapter_unlock(sa);
1179         SFC_ASSERT(rc > 0);
1180         return -rc;
1181 }
1182
1183 static int
1184 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1185 {
1186         struct sfc_adapter *sa = dev->data->dev_private;
1187
1188         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1189
1190         sfc_adapter_lock(sa);
1191         sfc_rx_qstop(sa, rx_queue_id);
1192
1193         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1194
1195         sfc_adapter_unlock(sa);
1196
1197         return 0;
1198 }
1199
1200 static int
1201 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1202 {
1203         struct sfc_adapter *sa = dev->data->dev_private;
1204         int rc;
1205
1206         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1207
1208         sfc_adapter_lock(sa);
1209
1210         rc = EINVAL;
1211         if (sa->state != SFC_ADAPTER_STARTED)
1212                 goto fail_not_started;
1213
1214         rc = sfc_tx_qstart(sa, tx_queue_id);
1215         if (rc != 0)
1216                 goto fail_tx_qstart;
1217
1218         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1219
1220         sfc_adapter_unlock(sa);
1221         return 0;
1222
1223 fail_tx_qstart:
1224
1225 fail_not_started:
1226         sfc_adapter_unlock(sa);
1227         SFC_ASSERT(rc > 0);
1228         return -rc;
1229 }
1230
1231 static int
1232 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1233 {
1234         struct sfc_adapter *sa = dev->data->dev_private;
1235
1236         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1237
1238         sfc_adapter_lock(sa);
1239
1240         sfc_tx_qstop(sa, tx_queue_id);
1241
1242         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1243
1244         sfc_adapter_unlock(sa);
1245         return 0;
1246 }
1247
1248 static efx_tunnel_protocol_t
1249 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1250 {
1251         switch (rte_type) {
1252         case RTE_TUNNEL_TYPE_VXLAN:
1253                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1254         case RTE_TUNNEL_TYPE_GENEVE:
1255                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1256         default:
1257                 return EFX_TUNNEL_NPROTOS;
1258         }
1259 }
1260
1261 enum sfc_udp_tunnel_op_e {
1262         SFC_UDP_TUNNEL_ADD_PORT,
1263         SFC_UDP_TUNNEL_DEL_PORT,
1264 };
1265
1266 static int
1267 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1268                       struct rte_eth_udp_tunnel *tunnel_udp,
1269                       enum sfc_udp_tunnel_op_e op)
1270 {
1271         struct sfc_adapter *sa = dev->data->dev_private;
1272         efx_tunnel_protocol_t tunnel_proto;
1273         int rc;
1274
1275         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1276                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1277                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1278                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1279
1280         tunnel_proto =
1281                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1282         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1283                 rc = ENOTSUP;
1284                 goto fail_bad_proto;
1285         }
1286
1287         sfc_adapter_lock(sa);
1288
1289         switch (op) {
1290         case SFC_UDP_TUNNEL_ADD_PORT:
1291                 rc = efx_tunnel_config_udp_add(sa->nic,
1292                                                tunnel_udp->udp_port,
1293                                                tunnel_proto);
1294                 break;
1295         case SFC_UDP_TUNNEL_DEL_PORT:
1296                 rc = efx_tunnel_config_udp_remove(sa->nic,
1297                                                   tunnel_udp->udp_port,
1298                                                   tunnel_proto);
1299                 break;
1300         default:
1301                 rc = EINVAL;
1302                 goto fail_bad_op;
1303         }
1304
1305         if (rc != 0)
1306                 goto fail_op;
1307
1308         if (sa->state == SFC_ADAPTER_STARTED) {
1309                 rc = efx_tunnel_reconfigure(sa->nic);
1310                 if (rc == EAGAIN) {
1311                         /*
1312                          * Configuration is accepted by FW and MC reboot
1313                          * is initiated to apply the changes. MC reboot
1314                          * will be handled in a usual way (MC reboot
1315                          * event on management event queue and adapter
1316                          * restart).
1317                          */
1318                         rc = 0;
1319                 } else if (rc != 0) {
1320                         goto fail_reconfigure;
1321                 }
1322         }
1323
1324         sfc_adapter_unlock(sa);
1325         return 0;
1326
1327 fail_reconfigure:
1328         /* Remove/restore entry since the change makes the trouble */
1329         switch (op) {
1330         case SFC_UDP_TUNNEL_ADD_PORT:
1331                 (void)efx_tunnel_config_udp_remove(sa->nic,
1332                                                    tunnel_udp->udp_port,
1333                                                    tunnel_proto);
1334                 break;
1335         case SFC_UDP_TUNNEL_DEL_PORT:
1336                 (void)efx_tunnel_config_udp_add(sa->nic,
1337                                                 tunnel_udp->udp_port,
1338                                                 tunnel_proto);
1339                 break;
1340         }
1341
1342 fail_op:
1343 fail_bad_op:
1344         sfc_adapter_unlock(sa);
1345
1346 fail_bad_proto:
1347         SFC_ASSERT(rc > 0);
1348         return -rc;
1349 }
1350
1351 static int
1352 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1353                             struct rte_eth_udp_tunnel *tunnel_udp)
1354 {
1355         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1356 }
1357
1358 static int
1359 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1360                             struct rte_eth_udp_tunnel *tunnel_udp)
1361 {
1362         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1363 }
1364
1365 static int
1366 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1367                           struct rte_eth_rss_conf *rss_conf)
1368 {
1369         struct sfc_adapter *sa = dev->data->dev_private;
1370         struct sfc_rss *rss = &sa->rss;
1371         struct sfc_port *port = &sa->port;
1372
1373         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1374                 return -ENOTSUP;
1375
1376         if (rss->channels == 0)
1377                 return -EINVAL;
1378
1379         sfc_adapter_lock(sa);
1380
1381         /*
1382          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1383          * hence, conversion is done here to derive a correct set of ETH_RSS
1384          * flags which corresponds to the active EFX configuration stored
1385          * locally in 'sfc_adapter' and kept up-to-date
1386          */
1387         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1388         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1389         if (rss_conf->rss_key != NULL)
1390                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1391
1392         sfc_adapter_unlock(sa);
1393
1394         return 0;
1395 }
1396
1397 static int
1398 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1399                         struct rte_eth_rss_conf *rss_conf)
1400 {
1401         struct sfc_adapter *sa = dev->data->dev_private;
1402         struct sfc_rss *rss = &sa->rss;
1403         struct sfc_port *port = &sa->port;
1404         unsigned int efx_hash_types;
1405         int rc = 0;
1406
1407         if (port->isolated)
1408                 return -ENOTSUP;
1409
1410         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1411                 sfc_err(sa, "RSS is not available");
1412                 return -ENOTSUP;
1413         }
1414
1415         if (rss->channels == 0) {
1416                 sfc_err(sa, "RSS is not configured");
1417                 return -EINVAL;
1418         }
1419
1420         if ((rss_conf->rss_key != NULL) &&
1421             (rss_conf->rss_key_len != sizeof(rss->key))) {
1422                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1423                         sizeof(rss->key));
1424                 return -EINVAL;
1425         }
1426
1427         sfc_adapter_lock(sa);
1428
1429         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1430         if (rc != 0)
1431                 goto fail_rx_hf_rte_to_efx;
1432
1433         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1434                                    rss->hash_alg, efx_hash_types, B_TRUE);
1435         if (rc != 0)
1436                 goto fail_scale_mode_set;
1437
1438         if (rss_conf->rss_key != NULL) {
1439                 if (sa->state == SFC_ADAPTER_STARTED) {
1440                         rc = efx_rx_scale_key_set(sa->nic,
1441                                                   EFX_RSS_CONTEXT_DEFAULT,
1442                                                   rss_conf->rss_key,
1443                                                   sizeof(rss->key));
1444                         if (rc != 0)
1445                                 goto fail_scale_key_set;
1446                 }
1447
1448                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1449         }
1450
1451         rss->hash_types = efx_hash_types;
1452
1453         sfc_adapter_unlock(sa);
1454
1455         return 0;
1456
1457 fail_scale_key_set:
1458         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1459                                   EFX_RX_HASHALG_TOEPLITZ,
1460                                   rss->hash_types, B_TRUE) != 0)
1461                 sfc_err(sa, "failed to restore RSS mode");
1462
1463 fail_scale_mode_set:
1464 fail_rx_hf_rte_to_efx:
1465         sfc_adapter_unlock(sa);
1466         return -rc;
1467 }
1468
1469 static int
1470 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1471                        struct rte_eth_rss_reta_entry64 *reta_conf,
1472                        uint16_t reta_size)
1473 {
1474         struct sfc_adapter *sa = dev->data->dev_private;
1475         struct sfc_rss *rss = &sa->rss;
1476         struct sfc_port *port = &sa->port;
1477         int entry;
1478
1479         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1480                 return -ENOTSUP;
1481
1482         if (rss->channels == 0)
1483                 return -EINVAL;
1484
1485         if (reta_size != EFX_RSS_TBL_SIZE)
1486                 return -EINVAL;
1487
1488         sfc_adapter_lock(sa);
1489
1490         for (entry = 0; entry < reta_size; entry++) {
1491                 int grp = entry / RTE_RETA_GROUP_SIZE;
1492                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1493
1494                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1495                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1496         }
1497
1498         sfc_adapter_unlock(sa);
1499
1500         return 0;
1501 }
1502
1503 static int
1504 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1505                         struct rte_eth_rss_reta_entry64 *reta_conf,
1506                         uint16_t reta_size)
1507 {
1508         struct sfc_adapter *sa = dev->data->dev_private;
1509         struct sfc_rss *rss = &sa->rss;
1510         struct sfc_port *port = &sa->port;
1511         unsigned int *rss_tbl_new;
1512         uint16_t entry;
1513         int rc = 0;
1514
1515
1516         if (port->isolated)
1517                 return -ENOTSUP;
1518
1519         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1520                 sfc_err(sa, "RSS is not available");
1521                 return -ENOTSUP;
1522         }
1523
1524         if (rss->channels == 0) {
1525                 sfc_err(sa, "RSS is not configured");
1526                 return -EINVAL;
1527         }
1528
1529         if (reta_size != EFX_RSS_TBL_SIZE) {
1530                 sfc_err(sa, "RETA size is wrong (should be %u)",
1531                         EFX_RSS_TBL_SIZE);
1532                 return -EINVAL;
1533         }
1534
1535         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1536         if (rss_tbl_new == NULL)
1537                 return -ENOMEM;
1538
1539         sfc_adapter_lock(sa);
1540
1541         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1542
1543         for (entry = 0; entry < reta_size; entry++) {
1544                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1545                 struct rte_eth_rss_reta_entry64 *grp;
1546
1547                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1548
1549                 if (grp->mask & (1ull << grp_idx)) {
1550                         if (grp->reta[grp_idx] >= rss->channels) {
1551                                 rc = EINVAL;
1552                                 goto bad_reta_entry;
1553                         }
1554                         rss_tbl_new[entry] = grp->reta[grp_idx];
1555                 }
1556         }
1557
1558         if (sa->state == SFC_ADAPTER_STARTED) {
1559                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1560                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1561                 if (rc != 0)
1562                         goto fail_scale_tbl_set;
1563         }
1564
1565         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1566
1567 fail_scale_tbl_set:
1568 bad_reta_entry:
1569         sfc_adapter_unlock(sa);
1570
1571         rte_free(rss_tbl_new);
1572
1573         SFC_ASSERT(rc >= 0);
1574         return -rc;
1575 }
1576
1577 static int
1578 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1579                     enum rte_filter_op filter_op,
1580                     void *arg)
1581 {
1582         struct sfc_adapter *sa = dev->data->dev_private;
1583         int rc = ENOTSUP;
1584
1585         sfc_log_init(sa, "entry");
1586
1587         switch (filter_type) {
1588         case RTE_ETH_FILTER_NONE:
1589                 sfc_err(sa, "Global filters configuration not supported");
1590                 break;
1591         case RTE_ETH_FILTER_MACVLAN:
1592                 sfc_err(sa, "MACVLAN filters not supported");
1593                 break;
1594         case RTE_ETH_FILTER_ETHERTYPE:
1595                 sfc_err(sa, "EtherType filters not supported");
1596                 break;
1597         case RTE_ETH_FILTER_FLEXIBLE:
1598                 sfc_err(sa, "Flexible filters not supported");
1599                 break;
1600         case RTE_ETH_FILTER_SYN:
1601                 sfc_err(sa, "SYN filters not supported");
1602                 break;
1603         case RTE_ETH_FILTER_NTUPLE:
1604                 sfc_err(sa, "NTUPLE filters not supported");
1605                 break;
1606         case RTE_ETH_FILTER_TUNNEL:
1607                 sfc_err(sa, "Tunnel filters not supported");
1608                 break;
1609         case RTE_ETH_FILTER_FDIR:
1610                 sfc_err(sa, "Flow Director filters not supported");
1611                 break;
1612         case RTE_ETH_FILTER_HASH:
1613                 sfc_err(sa, "Hash filters not supported");
1614                 break;
1615         case RTE_ETH_FILTER_GENERIC:
1616                 if (filter_op != RTE_ETH_FILTER_GET) {
1617                         rc = EINVAL;
1618                 } else {
1619                         *(const void **)arg = &sfc_flow_ops;
1620                         rc = 0;
1621                 }
1622                 break;
1623         default:
1624                 sfc_err(sa, "Unknown filter type %u", filter_type);
1625                 break;
1626         }
1627
1628         sfc_log_init(sa, "exit: %d", -rc);
1629         SFC_ASSERT(rc >= 0);
1630         return -rc;
1631 }
1632
1633 static int
1634 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1635 {
1636         struct sfc_adapter *sa = dev->data->dev_private;
1637
1638         /*
1639          * If Rx datapath does not provide callback to check mempool,
1640          * all pools are supported.
1641          */
1642         if (sa->dp_rx->pool_ops_supported == NULL)
1643                 return 1;
1644
1645         return sa->dp_rx->pool_ops_supported(pool);
1646 }
1647
1648 static const struct eth_dev_ops sfc_eth_dev_ops = {
1649         .dev_configure                  = sfc_dev_configure,
1650         .dev_start                      = sfc_dev_start,
1651         .dev_stop                       = sfc_dev_stop,
1652         .dev_set_link_up                = sfc_dev_set_link_up,
1653         .dev_set_link_down              = sfc_dev_set_link_down,
1654         .dev_close                      = sfc_dev_close,
1655         .promiscuous_enable             = sfc_dev_promisc_enable,
1656         .promiscuous_disable            = sfc_dev_promisc_disable,
1657         .allmulticast_enable            = sfc_dev_allmulti_enable,
1658         .allmulticast_disable           = sfc_dev_allmulti_disable,
1659         .link_update                    = sfc_dev_link_update,
1660         .stats_get                      = sfc_stats_get,
1661         .stats_reset                    = sfc_stats_reset,
1662         .xstats_get                     = sfc_xstats_get,
1663         .xstats_reset                   = sfc_stats_reset,
1664         .xstats_get_names               = sfc_xstats_get_names,
1665         .dev_infos_get                  = sfc_dev_infos_get,
1666         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1667         .mtu_set                        = sfc_dev_set_mtu,
1668         .rx_queue_start                 = sfc_rx_queue_start,
1669         .rx_queue_stop                  = sfc_rx_queue_stop,
1670         .tx_queue_start                 = sfc_tx_queue_start,
1671         .tx_queue_stop                  = sfc_tx_queue_stop,
1672         .rx_queue_setup                 = sfc_rx_queue_setup,
1673         .rx_queue_release               = sfc_rx_queue_release,
1674         .rx_queue_count                 = sfc_rx_queue_count,
1675         .rx_descriptor_done             = sfc_rx_descriptor_done,
1676         .rx_descriptor_status           = sfc_rx_descriptor_status,
1677         .tx_descriptor_status           = sfc_tx_descriptor_status,
1678         .tx_queue_setup                 = sfc_tx_queue_setup,
1679         .tx_queue_release               = sfc_tx_queue_release,
1680         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1681         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1682         .mac_addr_set                   = sfc_mac_addr_set,
1683         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1684         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1685         .reta_update                    = sfc_dev_rss_reta_update,
1686         .reta_query                     = sfc_dev_rss_reta_query,
1687         .rss_hash_update                = sfc_dev_rss_hash_update,
1688         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1689         .filter_ctrl                    = sfc_dev_filter_ctrl,
1690         .set_mc_addr_list               = sfc_set_mc_addr_list,
1691         .rxq_info_get                   = sfc_rx_queue_info_get,
1692         .txq_info_get                   = sfc_tx_queue_info_get,
1693         .fw_version_get                 = sfc_fw_version_get,
1694         .xstats_get_by_id               = sfc_xstats_get_by_id,
1695         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1696         .pool_ops_supported             = sfc_pool_ops_supported,
1697 };
1698
1699 /**
1700  * Duplicate a string in potentially shared memory required for
1701  * multi-process support.
1702  *
1703  * strdup() allocates from process-local heap/memory.
1704  */
1705 static char *
1706 sfc_strdup(const char *str)
1707 {
1708         size_t size;
1709         char *copy;
1710
1711         if (str == NULL)
1712                 return NULL;
1713
1714         size = strlen(str) + 1;
1715         copy = rte_malloc(__func__, size, 0);
1716         if (copy != NULL)
1717                 rte_memcpy(copy, str, size);
1718
1719         return copy;
1720 }
1721
1722 static int
1723 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1724 {
1725         struct sfc_adapter *sa = dev->data->dev_private;
1726         const efx_nic_cfg_t *encp;
1727         unsigned int avail_caps = 0;
1728         const char *rx_name = NULL;
1729         const char *tx_name = NULL;
1730         int rc;
1731
1732         switch (sa->family) {
1733         case EFX_FAMILY_HUNTINGTON:
1734         case EFX_FAMILY_MEDFORD:
1735         case EFX_FAMILY_MEDFORD2:
1736                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1737                 break;
1738         default:
1739                 break;
1740         }
1741
1742         encp = efx_nic_cfg_get(sa->nic);
1743         if (encp->enc_rx_es_super_buffer_supported)
1744                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1745
1746         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1747                                 sfc_kvarg_string_handler, &rx_name);
1748         if (rc != 0)
1749                 goto fail_kvarg_rx_datapath;
1750
1751         if (rx_name != NULL) {
1752                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1753                 if (sa->dp_rx == NULL) {
1754                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1755                         rc = ENOENT;
1756                         goto fail_dp_rx;
1757                 }
1758                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1759                         sfc_err(sa,
1760                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1761                                 rx_name);
1762                         rc = EINVAL;
1763                         goto fail_dp_rx_caps;
1764                 }
1765         } else {
1766                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1767                 if (sa->dp_rx == NULL) {
1768                         sfc_err(sa, "Rx datapath by caps %#x not found",
1769                                 avail_caps);
1770                         rc = ENOENT;
1771                         goto fail_dp_rx;
1772                 }
1773         }
1774
1775         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1776         if (sa->dp_rx_name == NULL) {
1777                 rc = ENOMEM;
1778                 goto fail_dp_rx_name;
1779         }
1780
1781         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1782
1783         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1784
1785         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1786                                 sfc_kvarg_string_handler, &tx_name);
1787         if (rc != 0)
1788                 goto fail_kvarg_tx_datapath;
1789
1790         if (tx_name != NULL) {
1791                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1792                 if (sa->dp_tx == NULL) {
1793                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1794                         rc = ENOENT;
1795                         goto fail_dp_tx;
1796                 }
1797                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1798                         sfc_err(sa,
1799                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1800                                 tx_name);
1801                         rc = EINVAL;
1802                         goto fail_dp_tx_caps;
1803                 }
1804         } else {
1805                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1806                 if (sa->dp_tx == NULL) {
1807                         sfc_err(sa, "Tx datapath by caps %#x not found",
1808                                 avail_caps);
1809                         rc = ENOENT;
1810                         goto fail_dp_tx;
1811                 }
1812         }
1813
1814         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1815         if (sa->dp_tx_name == NULL) {
1816                 rc = ENOMEM;
1817                 goto fail_dp_tx_name;
1818         }
1819
1820         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1821
1822         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1823
1824         dev->dev_ops = &sfc_eth_dev_ops;
1825
1826         return 0;
1827
1828 fail_dp_tx_name:
1829 fail_dp_tx_caps:
1830         sa->dp_tx = NULL;
1831
1832 fail_dp_tx:
1833 fail_kvarg_tx_datapath:
1834         rte_free(sa->dp_rx_name);
1835         sa->dp_rx_name = NULL;
1836
1837 fail_dp_rx_name:
1838 fail_dp_rx_caps:
1839         sa->dp_rx = NULL;
1840
1841 fail_dp_rx:
1842 fail_kvarg_rx_datapath:
1843         return rc;
1844 }
1845
1846 static void
1847 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1848 {
1849         struct sfc_adapter *sa = dev->data->dev_private;
1850
1851         dev->dev_ops = NULL;
1852         dev->rx_pkt_burst = NULL;
1853         dev->tx_pkt_burst = NULL;
1854
1855         rte_free(sa->dp_tx_name);
1856         sa->dp_tx_name = NULL;
1857         sa->dp_tx = NULL;
1858
1859         rte_free(sa->dp_rx_name);
1860         sa->dp_rx_name = NULL;
1861         sa->dp_rx = NULL;
1862 }
1863
1864 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1865         .rxq_info_get                   = sfc_rx_queue_info_get,
1866         .txq_info_get                   = sfc_tx_queue_info_get,
1867 };
1868
1869 static int
1870 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1871 {
1872         /*
1873          * Device private data has really many process-local pointers.
1874          * Below code should be extremely careful to use data located
1875          * in shared memory only.
1876          */
1877         struct sfc_adapter *sa = dev->data->dev_private;
1878         const struct sfc_dp_rx *dp_rx;
1879         const struct sfc_dp_tx *dp_tx;
1880         int rc;
1881
1882         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1883         if (dp_rx == NULL) {
1884                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1885                 rc = ENOENT;
1886                 goto fail_dp_rx;
1887         }
1888         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1889                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1890                         sa->dp_tx_name);
1891                 rc = EINVAL;
1892                 goto fail_dp_rx_multi_process;
1893         }
1894
1895         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1896         if (dp_tx == NULL) {
1897                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1898                 rc = ENOENT;
1899                 goto fail_dp_tx;
1900         }
1901         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1902                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1903                         sa->dp_tx_name);
1904                 rc = EINVAL;
1905                 goto fail_dp_tx_multi_process;
1906         }
1907
1908         dev->rx_pkt_burst = dp_rx->pkt_burst;
1909         dev->tx_pkt_burst = dp_tx->pkt_burst;
1910         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1911
1912         return 0;
1913
1914 fail_dp_tx_multi_process:
1915 fail_dp_tx:
1916 fail_dp_rx_multi_process:
1917 fail_dp_rx:
1918         return rc;
1919 }
1920
1921 static void
1922 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1923 {
1924         dev->dev_ops = NULL;
1925         dev->tx_pkt_burst = NULL;
1926         dev->rx_pkt_burst = NULL;
1927 }
1928
1929 static void
1930 sfc_register_dp(void)
1931 {
1932         /* Register once */
1933         if (TAILQ_EMPTY(&sfc_dp_head)) {
1934                 /* Prefer EF10 datapath */
1935                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1936                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1937                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1938
1939                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1940                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1941                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1942         }
1943 }
1944
1945 static int
1946 sfc_eth_dev_init(struct rte_eth_dev *dev)
1947 {
1948         struct sfc_adapter *sa = dev->data->dev_private;
1949         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1950         int rc;
1951         const efx_nic_cfg_t *encp;
1952         const struct ether_addr *from;
1953
1954         sfc_register_dp();
1955
1956         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1957                 return -sfc_eth_dev_secondary_set_ops(dev);
1958
1959         /* Required for logging */
1960         sa->pci_addr = pci_dev->addr;
1961         sa->port_id = dev->data->port_id;
1962
1963         sa->eth_dev = dev;
1964
1965         /* Copy PCI device info to the dev->data */
1966         rte_eth_copy_pci_info(dev, pci_dev);
1967
1968         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1969                                                 RTE_LOG_NOTICE);
1970
1971         rc = sfc_kvargs_parse(sa);
1972         if (rc != 0)
1973                 goto fail_kvargs_parse;
1974
1975         sfc_log_init(sa, "entry");
1976
1977         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1978         if (dev->data->mac_addrs == NULL) {
1979                 rc = ENOMEM;
1980                 goto fail_mac_addrs;
1981         }
1982
1983         sfc_adapter_lock_init(sa);
1984         sfc_adapter_lock(sa);
1985
1986         sfc_log_init(sa, "probing");
1987         rc = sfc_probe(sa);
1988         if (rc != 0)
1989                 goto fail_probe;
1990
1991         sfc_log_init(sa, "set device ops");
1992         rc = sfc_eth_dev_set_ops(dev);
1993         if (rc != 0)
1994                 goto fail_set_ops;
1995
1996         sfc_log_init(sa, "attaching");
1997         rc = sfc_attach(sa);
1998         if (rc != 0)
1999                 goto fail_attach;
2000
2001         encp = efx_nic_cfg_get(sa->nic);
2002
2003         /*
2004          * The arguments are really reverse order in comparison to
2005          * Linux kernel. Copy from NIC config to Ethernet device data.
2006          */
2007         from = (const struct ether_addr *)(encp->enc_mac_addr);
2008         ether_addr_copy(from, &dev->data->mac_addrs[0]);
2009
2010         sfc_adapter_unlock(sa);
2011
2012         sfc_log_init(sa, "done");
2013         return 0;
2014
2015 fail_attach:
2016         sfc_eth_dev_clear_ops(dev);
2017
2018 fail_set_ops:
2019         sfc_unprobe(sa);
2020
2021 fail_probe:
2022         sfc_adapter_unlock(sa);
2023         sfc_adapter_lock_fini(sa);
2024         rte_free(dev->data->mac_addrs);
2025         dev->data->mac_addrs = NULL;
2026
2027 fail_mac_addrs:
2028         sfc_kvargs_cleanup(sa);
2029
2030 fail_kvargs_parse:
2031         sfc_log_init(sa, "failed %d", rc);
2032         SFC_ASSERT(rc > 0);
2033         return -rc;
2034 }
2035
2036 static int
2037 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2038 {
2039         struct sfc_adapter *sa;
2040
2041         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2042                 sfc_eth_dev_secondary_clear_ops(dev);
2043                 return 0;
2044         }
2045
2046         sa = dev->data->dev_private;
2047         sfc_log_init(sa, "entry");
2048
2049         sfc_adapter_lock(sa);
2050
2051         sfc_eth_dev_clear_ops(dev);
2052
2053         sfc_detach(sa);
2054         sfc_unprobe(sa);
2055
2056         rte_free(dev->data->mac_addrs);
2057         dev->data->mac_addrs = NULL;
2058
2059         sfc_kvargs_cleanup(sa);
2060
2061         sfc_adapter_unlock(sa);
2062         sfc_adapter_lock_fini(sa);
2063
2064         sfc_log_init(sa, "done");
2065
2066         /* Required for logging, so cleanup last */
2067         sa->eth_dev = NULL;
2068         return 0;
2069 }
2070
2071 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2072         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2073         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2074         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2075         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2076         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2077         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2078         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2079         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2080         { .vendor_id = 0 /* sentinel */ }
2081 };
2082
2083 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2084         struct rte_pci_device *pci_dev)
2085 {
2086         return rte_eth_dev_pci_generic_probe(pci_dev,
2087                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2088 }
2089
2090 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2091 {
2092         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2093 }
2094
2095 static struct rte_pci_driver sfc_efx_pmd = {
2096         .id_table = pci_id_sfc_efx_map,
2097         .drv_flags =
2098                 RTE_PCI_DRV_INTR_LSC |
2099                 RTE_PCI_DRV_NEED_MAPPING,
2100         .probe = sfc_eth_dev_pci_probe,
2101         .remove = sfc_eth_dev_pci_remove,
2102 };
2103
2104 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2105 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2106 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2107 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2108         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2109         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2110         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2111         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2112         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2113
2114 RTE_INIT(sfc_driver_register_logtype);
2115 static void
2116 sfc_driver_register_logtype(void)
2117 {
2118         int ret;
2119
2120         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2121                                                    RTE_LOG_NOTICE);
2122         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2123 }