drivers/net/sfc/sfc_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
   5  * All rights reserved.
   6  *
   7  * This software was jointly developed between OKTET Labs (under contract
   8  * for Solarflare) and Solarflare Communications, Inc.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright notice,
  14  *    this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright notice,
  16  *    this list of conditions and the following disclaimer in the documentation
  17  *    and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
  21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
  26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
  29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 #include <rte_dev.h>
  33 #include <rte_ethdev.h>
  34 #include <rte_ethdev_pci.h>
  35 #include <rte_pci.h>
  36 #include <rte_errno.h>
  37
  38 #include "efx.h"
  39
  40 #include "sfc.h"
  41 #include "sfc_debug.h"
  42 #include "sfc_log.h"
  43 #include "sfc_kvargs.h"
  44 #include "sfc_ev.h"
  45 #include "sfc_rx.h"
  46 #include "sfc_tx.h"
  47 #include "sfc_flow.h"
  48 #include "sfc_dp.h"
  49 #include "sfc_dp_rx.h"
  50
  51 static struct sfc_dp_list sfc_dp_head =
  52         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  53
  54 static int
  55 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  56 {
  57         struct sfc_adapter *sa = dev->data->dev_private;
  58         efx_nic_fw_info_t enfi;
  59         int ret;
  60         int rc;
  61
  62         /*
  63          * Return value of the callback is likely supposed to be
  64          * equal to or greater than 0, nevertheless, if an error
  65          * occurs, it will be desirable to pass it to the caller
  66          */
  67         if ((fw_version == NULL) || (fw_size == 0))
  68                 return -EINVAL;
  69
  70         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  71         if (rc != 0)
  72                 return -rc;
  73
  74         ret = snprintf(fw_version, fw_size,
  75                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  76                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  77                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  78         if (ret < 0)
  79                 return ret;
  80
  81         if (enfi.enfi_dpcpu_fw_ids_valid) {
  82                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  83                 int ret_extra;
  84
  85                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  86                                      fw_size - dpcpu_fw_ids_offset,
  87                                      " rx%" PRIx16 " tx%" PRIx16,
  88                                      enfi.enfi_rx_dpcpu_fw_id,
  89                                      enfi.enfi_tx_dpcpu_fw_id);
  90                 if (ret_extra < 0)
  91                         return ret_extra;
  92
  93                 ret += ret_extra;
  94         }
  95
  96         if (fw_size < (size_t)(++ret))
  97                 return ret;
  98         else
  99                 return 0;
 100 }
 101
 102 static void
 103 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 104 {
 105         struct sfc_adapter *sa = dev->data->dev_private;
 106         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 107
 108         sfc_log_init(sa, "entry");
 109
 110         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 111         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 112
 113         /* Autonegotiation may be disabled */
 114         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 115         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
 116                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 117         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
 118                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 119         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 120                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 121
 122         dev_info->max_rx_queues = sa->rxq_max;
 123         dev_info->max_tx_queues = sa->txq_max;
 124
 125         /* By default packets are dropped if no descriptors are available */
 126         dev_info->default_rxconf.rx_drop_en = 1;
 127
 128         dev_info->rx_offload_capa =
 129                 DEV_RX_OFFLOAD_IPV4_CKSUM |
 130                 DEV_RX_OFFLOAD_UDP_CKSUM |
 131                 DEV_RX_OFFLOAD_TCP_CKSUM;
 132
 133         dev_info->tx_offload_capa =
 134                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 135                 DEV_TX_OFFLOAD_UDP_CKSUM |
 136                 DEV_TX_OFFLOAD_TCP_CKSUM;
 137
 138         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 139         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 140             !encp->enc_hw_tx_insert_vlan_enabled)
 141                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 142         else
 143                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 144
 145         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 146                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 147
 148 #if EFSYS_OPT_RX_SCALE
 149         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 150                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 151                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 152                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 153         }
 154 #endif
 155
 156         if (sa->tso)
 157                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 158
 159         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 160         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 161         /* The RXQ hardware requires that the descriptor count is a power
 162          * of 2, but rx_desc_lim cannot properly describe that constraint.
 163          */
 164         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 165
 166         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 167         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 168         /*
 169          * The TXQ hardware requires that the descriptor count is a power
 170          * of 2, but tx_desc_lim cannot properly describe that constraint
 171          */
 172         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 173 }
 174
 175 static const uint32_t *
 176 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 177 {
 178         struct sfc_adapter *sa = dev->data->dev_private;
 179
 180         return sa->dp_rx->supported_ptypes_get();
 181 }
 182
 183 static int
 184 sfc_dev_configure(struct rte_eth_dev *dev)
 185 {
 186         struct rte_eth_dev_data *dev_data = dev->data;
 187         struct sfc_adapter *sa = dev_data->dev_private;
 188         int rc;
 189
 190         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 191                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 192
 193         sfc_adapter_lock(sa);
 194         switch (sa->state) {
 195         case SFC_ADAPTER_CONFIGURED:
 196                 /* FALLTHROUGH */
 197         case SFC_ADAPTER_INITIALIZED:
 198                 rc = sfc_configure(sa);
 199                 break;
 200         default:
 201                 sfc_err(sa, "unexpected adapter state %u to configure",
 202                         sa->state);
 203                 rc = EINVAL;
 204                 break;
 205         }
 206         sfc_adapter_unlock(sa);
 207
 208         sfc_log_init(sa, "done %d", rc);
 209         SFC_ASSERT(rc >= 0);
 210         return -rc;
 211 }
 212
 213 static int
 214 sfc_dev_start(struct rte_eth_dev *dev)
 215 {
 216         struct sfc_adapter *sa = dev->data->dev_private;
 217         int rc;
 218
 219         sfc_log_init(sa, "entry");
 220
 221         sfc_adapter_lock(sa);
 222         rc = sfc_start(sa);
 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_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 232 {
 233         struct sfc_adapter *sa = dev->data->dev_private;
 234         struct rte_eth_link *dev_link = &dev->data->dev_link;
 235         struct rte_eth_link old_link;
 236         struct rte_eth_link current_link;
 237
 238         sfc_log_init(sa, "entry");
 239
 240 retry:
 241         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 242         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 243
 244         if (sa->state != SFC_ADAPTER_STARTED) {
 245                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 246                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 247                                          *(uint64_t *)&old_link,
 248                                          *(uint64_t *)&current_link))
 249                         goto retry;
 250         } else if (wait_to_complete) {
 251                 efx_link_mode_t link_mode;
 252
 253                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 254                         link_mode = EFX_LINK_UNKNOWN;
 255                 sfc_port_link_mode_to_info(link_mode, &current_link);
 256
 257                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 258                                          *(uint64_t *)&old_link,
 259                                          *(uint64_t *)&current_link))
 260                         goto retry;
 261         } else {
 262                 sfc_ev_mgmt_qpoll(sa);
 263                 *(int64_t *)&current_link =
 264                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 265         }
 266
 267         if (old_link.link_status != current_link.link_status)
 268                 sfc_info(sa, "Link status is %s",
 269                          current_link.link_status ? "UP" : "DOWN");
 270
 271         return old_link.link_status == current_link.link_status ? 0 : -1;
 272 }
 273
 274 static void
 275 sfc_dev_stop(struct rte_eth_dev *dev)
 276 {
 277         struct sfc_adapter *sa = dev->data->dev_private;
 278
 279         sfc_log_init(sa, "entry");
 280
 281         sfc_adapter_lock(sa);
 282         sfc_stop(sa);
 283         sfc_adapter_unlock(sa);
 284
 285         sfc_log_init(sa, "done");
 286 }
 287
 288 static int
 289 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 290 {
 291         struct sfc_adapter *sa = dev->data->dev_private;
 292         int rc;
 293
 294         sfc_log_init(sa, "entry");
 295
 296         sfc_adapter_lock(sa);
 297         rc = sfc_start(sa);
 298         sfc_adapter_unlock(sa);
 299
 300         SFC_ASSERT(rc >= 0);
 301         return -rc;
 302 }
 303
 304 static int
 305 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 306 {
 307         struct sfc_adapter *sa = dev->data->dev_private;
 308
 309         sfc_log_init(sa, "entry");
 310
 311         sfc_adapter_lock(sa);
 312         sfc_stop(sa);
 313         sfc_adapter_unlock(sa);
 314
 315         return 0;
 316 }
 317
 318 static void
 319 sfc_dev_close(struct rte_eth_dev *dev)
 320 {
 321         struct sfc_adapter *sa = dev->data->dev_private;
 322
 323         sfc_log_init(sa, "entry");
 324
 325         sfc_adapter_lock(sa);
 326         switch (sa->state) {
 327         case SFC_ADAPTER_STARTED:
 328                 sfc_stop(sa);
 329                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 330                 /* FALLTHROUGH */
 331         case SFC_ADAPTER_CONFIGURED:
 332                 sfc_close(sa);
 333                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 334                 /* FALLTHROUGH */
 335         case SFC_ADAPTER_INITIALIZED:
 336                 break;
 337         default:
 338                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 339                 break;
 340         }
 341         sfc_adapter_unlock(sa);
 342
 343         sfc_log_init(sa, "done");
 344 }
 345
 346 static void
 347 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 348                    boolean_t enabled)
 349 {
 350         struct sfc_port *port;
 351         boolean_t *toggle;
 352         struct sfc_adapter *sa = dev->data->dev_private;
 353         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 354         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 355
 356         sfc_adapter_lock(sa);
 357
 358         port = &sa->port;
 359         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 360
 361         if (*toggle != enabled) {
 362                 *toggle = enabled;
 363
 364                 if (port->isolated) {
 365                         sfc_warn(sa, "isolated mode is active on the port");
 366                         sfc_warn(sa, "the change is to be applied on the next "
 367                                      "start provided that isolated mode is "
 368                                      "disabled prior the next start");
 369                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 370                            (sfc_set_rx_mode(sa) != 0)) {
 371                         *toggle = !(enabled);
 372                         sfc_warn(sa, "Failed to %s %s mode",
 373                                  ((enabled) ? "enable" : "disable"), desc);
 374                 }
 375         }
 376
 377         sfc_adapter_unlock(sa);
 378 }
 379
 380 static void
 381 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 382 {
 383         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 384 }
 385
 386 static void
 387 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 388 {
 389         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 390 }
 391
 392 static void
 393 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 394 {
 395         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 396 }
 397
 398 static void
 399 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 400 {
 401         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 402 }
 403
 404 static int
 405 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 406                    uint16_t nb_rx_desc, unsigned int socket_id,
 407                    const struct rte_eth_rxconf *rx_conf,
 408                    struct rte_mempool *mb_pool)
 409 {
 410         struct sfc_adapter *sa = dev->data->dev_private;
 411         int rc;
 412
 413         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 414                      rx_queue_id, nb_rx_desc, socket_id);
 415
 416         sfc_adapter_lock(sa);
 417
 418         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 419                           rx_conf, mb_pool);
 420         if (rc != 0)
 421                 goto fail_rx_qinit;
 422
 423         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 424
 425         sfc_adapter_unlock(sa);
 426
 427         return 0;
 428
 429 fail_rx_qinit:
 430         sfc_adapter_unlock(sa);
 431         SFC_ASSERT(rc > 0);
 432         return -rc;
 433 }
 434
 435 static void
 436 sfc_rx_queue_release(void *queue)
 437 {
 438         struct sfc_dp_rxq *dp_rxq = queue;
 439         struct sfc_rxq *rxq;
 440         struct sfc_adapter *sa;
 441         unsigned int sw_index;
 442
 443         if (dp_rxq == NULL)
 444                 return;
 445
 446         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 447         sa = rxq->evq->sa;
 448         sfc_adapter_lock(sa);
 449
 450         sw_index = sfc_rxq_sw_index(rxq);
 451
 452         sfc_log_init(sa, "RxQ=%u", sw_index);
 453
 454         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 455
 456         sfc_rx_qfini(sa, sw_index);
 457
 458         sfc_adapter_unlock(sa);
 459 }
 460
 461 static int
 462 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 463                    uint16_t nb_tx_desc, unsigned int socket_id,
 464                    const struct rte_eth_txconf *tx_conf)
 465 {
 466         struct sfc_adapter *sa = dev->data->dev_private;
 467         int rc;
 468
 469         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 470                      tx_queue_id, nb_tx_desc, socket_id);
 471
 472         sfc_adapter_lock(sa);
 473
 474         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 475         if (rc != 0)
 476                 goto fail_tx_qinit;
 477
 478         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 479
 480         sfc_adapter_unlock(sa);
 481         return 0;
 482
 483 fail_tx_qinit:
 484         sfc_adapter_unlock(sa);
 485         SFC_ASSERT(rc > 0);
 486         return -rc;
 487 }
 488
 489 static void
 490 sfc_tx_queue_release(void *queue)
 491 {
 492         struct sfc_dp_txq *dp_txq = queue;
 493         struct sfc_txq *txq;
 494         unsigned int sw_index;
 495         struct sfc_adapter *sa;
 496
 497         if (dp_txq == NULL)
 498                 return;
 499
 500         txq = sfc_txq_by_dp_txq(dp_txq);
 501         sw_index = sfc_txq_sw_index(txq);
 502
 503         SFC_ASSERT(txq->evq != NULL);
 504         sa = txq->evq->sa;
 505
 506         sfc_log_init(sa, "TxQ = %u", sw_index);
 507
 508         sfc_adapter_lock(sa);
 509
 510         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 511         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 512
 513         sfc_tx_qfini(sa, sw_index);
 514
 515         sfc_adapter_unlock(sa);
 516 }
 517
 518 static void
 519 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 520 {
 521         struct sfc_adapter *sa = dev->data->dev_private;
 522         struct sfc_port *port = &sa->port;
 523         uint64_t *mac_stats;
 524
 525         rte_spinlock_lock(&port->mac_stats_lock);
 526
 527         if (sfc_port_update_mac_stats(sa) != 0)
 528                 goto unlock;
 529
 530         mac_stats = port->mac_stats_buf;
 531
 532         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 533                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 534                 stats->ipackets =
 535                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 536                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 537                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 538                 stats->opackets =
 539                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 540                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 541                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 542                 stats->ibytes =
 543                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 544                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 545                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 546                 stats->obytes =
 547                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 548                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 549                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 550                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 551                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 552                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 553         } else {
 554                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 555                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 556                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 557                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 558                 /*
 559                  * Take into account stats which are whenever supported
 560                  * on EF10. If some stat is not supported by current
 561                  * firmware variant or HW revision, it is guaranteed
 562                  * to be zero in mac_stats.
 563                  */
 564                 stats->imissed =
 565                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 566                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 567                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 568                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 569                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 570                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 571                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 572                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 573                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 574                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 575                 stats->ierrors =
 576                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 577                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 578                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 579                 /* no oerrors counters supported on EF10 */
 580         }
 581
 582 unlock:
 583         rte_spinlock_unlock(&port->mac_stats_lock);
 584 }
 585
 586 static void
 587 sfc_stats_reset(struct rte_eth_dev *dev)
 588 {
 589         struct sfc_adapter *sa = dev->data->dev_private;
 590         struct sfc_port *port = &sa->port;
 591         int rc;
 592
 593         if (sa->state != SFC_ADAPTER_STARTED) {
 594                 /*
 595                  * The operation cannot be done if port is not started; it
 596                  * will be scheduled to be done during the next port start
 597                  */
 598                 port->mac_stats_reset_pending = B_TRUE;
 599                 return;
 600         }
 601
 602         rc = sfc_port_reset_mac_stats(sa);
 603         if (rc != 0)
 604                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 605 }
 606
 607 static int
 608 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 609                unsigned int xstats_count)
 610 {
 611         struct sfc_adapter *sa = dev->data->dev_private;
 612         struct sfc_port *port = &sa->port;
 613         uint64_t *mac_stats;
 614         int rc;
 615         unsigned int i;
 616         int nstats = 0;
 617
 618         rte_spinlock_lock(&port->mac_stats_lock);
 619
 620         rc = sfc_port_update_mac_stats(sa);
 621         if (rc != 0) {
 622                 SFC_ASSERT(rc > 0);
 623                 nstats = -rc;
 624                 goto unlock;
 625         }
 626
 627         mac_stats = port->mac_stats_buf;
 628
 629         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 630                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 631                         if (xstats != NULL && nstats < (int)xstats_count) {
 632                                 xstats[nstats].id = nstats;
 633                                 xstats[nstats].value = mac_stats[i];
 634                         }
 635                         nstats++;
 636                 }
 637         }
 638
 639 unlock:
 640         rte_spinlock_unlock(&port->mac_stats_lock);
 641
 642         return nstats;
 643 }
 644
 645 static int
 646 sfc_xstats_get_names(struct rte_eth_dev *dev,
 647                      struct rte_eth_xstat_name *xstats_names,
 648                      unsigned int xstats_count)
 649 {
 650         struct sfc_adapter *sa = dev->data->dev_private;
 651         struct sfc_port *port = &sa->port;
 652         unsigned int i;
 653         unsigned int nstats = 0;
 654
 655         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 656                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 657                         if (xstats_names != NULL && nstats < xstats_count)
 658                                 strncpy(xstats_names[nstats].name,
 659                                         efx_mac_stat_name(sa->nic, i),
 660                                         sizeof(xstats_names[0].name));
 661                         nstats++;
 662                 }
 663         }
 664
 665         return nstats;
 666 }
 667
 668 static int
 669 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 670                      uint64_t *values, unsigned int n)
 671 {
 672         struct sfc_adapter *sa = dev->data->dev_private;
 673         struct sfc_port *port = &sa->port;
 674         uint64_t *mac_stats;
 675         unsigned int nb_supported = 0;
 676         unsigned int nb_written = 0;
 677         unsigned int i;
 678         int ret;
 679         int rc;
 680
 681         if (unlikely(values == NULL) ||
 682             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 683                 return port->mac_stats_nb_supported;
 684
 685         rte_spinlock_lock(&port->mac_stats_lock);
 686
 687         rc = sfc_port_update_mac_stats(sa);
 688         if (rc != 0) {
 689                 SFC_ASSERT(rc > 0);
 690                 ret = -rc;
 691                 goto unlock;
 692         }
 693
 694         mac_stats = port->mac_stats_buf;
 695
 696         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 697                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 698                         continue;
 699
 700                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 701                         values[nb_written++] = mac_stats[i];
 702
 703                 ++nb_supported;
 704         }
 705
 706         ret = nb_written;
 707
 708 unlock:
 709         rte_spinlock_unlock(&port->mac_stats_lock);
 710
 711         return ret;
 712 }
 713
 714 static int
 715 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 716                            struct rte_eth_xstat_name *xstats_names,
 717                            const uint64_t *ids, unsigned int size)
 718 {
 719         struct sfc_adapter *sa = dev->data->dev_private;
 720         struct sfc_port *port = &sa->port;
 721         unsigned int nb_supported = 0;
 722         unsigned int nb_written = 0;
 723         unsigned int i;
 724
 725         if (unlikely(xstats_names == NULL) ||
 726             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 727                 return port->mac_stats_nb_supported;
 728
 729         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 730                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 731                         continue;
 732
 733                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 734                         char *name = xstats_names[nb_written++].name;
 735
 736                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 737                                 sizeof(xstats_names[0].name));
 738                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 739                 }
 740
 741                 ++nb_supported;
 742         }
 743
 744         return nb_written;
 745 }
 746
 747 static int
 748 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 749 {
 750         struct sfc_adapter *sa = dev->data->dev_private;
 751         unsigned int wanted_fc, link_fc;
 752
 753         memset(fc_conf, 0, sizeof(*fc_conf));
 754
 755         sfc_adapter_lock(sa);
 756
 757         if (sa->state == SFC_ADAPTER_STARTED)
 758                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 759         else
 760                 link_fc = sa->port.flow_ctrl;
 761
 762         switch (link_fc) {
 763         case 0:
 764                 fc_conf->mode = RTE_FC_NONE;
 765                 break;
 766         case EFX_FCNTL_RESPOND:
 767                 fc_conf->mode = RTE_FC_RX_PAUSE;
 768                 break;
 769         case EFX_FCNTL_GENERATE:
 770                 fc_conf->mode = RTE_FC_TX_PAUSE;
 771                 break;
 772         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 773                 fc_conf->mode = RTE_FC_FULL;
 774                 break;
 775         default:
 776                 sfc_err(sa, "%s: unexpected flow control value %#x",
 777                         __func__, link_fc);
 778         }
 779
 780         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 781
 782         sfc_adapter_unlock(sa);
 783
 784         return 0;
 785 }
 786
 787 static int
 788 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 789 {
 790         struct sfc_adapter *sa = dev->data->dev_private;
 791         struct sfc_port *port = &sa->port;
 792         unsigned int fcntl;
 793         int rc;
 794
 795         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 796             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 797             fc_conf->mac_ctrl_frame_fwd != 0) {
 798                 sfc_err(sa, "unsupported flow control settings specified");
 799                 rc = EINVAL;
 800                 goto fail_inval;
 801         }
 802
 803         switch (fc_conf->mode) {
 804         case RTE_FC_NONE:
 805                 fcntl = 0;
 806                 break;
 807         case RTE_FC_RX_PAUSE:
 808                 fcntl = EFX_FCNTL_RESPOND;
 809                 break;
 810         case RTE_FC_TX_PAUSE:
 811                 fcntl = EFX_FCNTL_GENERATE;
 812                 break;
 813         case RTE_FC_FULL:
 814                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 815                 break;
 816         default:
 817                 rc = EINVAL;
 818                 goto fail_inval;
 819         }
 820
 821         sfc_adapter_lock(sa);
 822
 823         if (sa->state == SFC_ADAPTER_STARTED) {
 824                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 825                 if (rc != 0)
 826                         goto fail_mac_fcntl_set;
 827         }
 828
 829         port->flow_ctrl = fcntl;
 830         port->flow_ctrl_autoneg = fc_conf->autoneg;
 831
 832         sfc_adapter_unlock(sa);
 833
 834         return 0;
 835
 836 fail_mac_fcntl_set:
 837         sfc_adapter_unlock(sa);
 838 fail_inval:
 839         SFC_ASSERT(rc > 0);
 840         return -rc;
 841 }
 842
 843 static int
 844 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 845 {
 846         struct sfc_adapter *sa = dev->data->dev_private;
 847         size_t pdu = EFX_MAC_PDU(mtu);
 848         size_t old_pdu;
 849         int rc;
 850
 851         sfc_log_init(sa, "mtu=%u", mtu);
 852
 853         rc = EINVAL;
 854         if (pdu < EFX_MAC_PDU_MIN) {
 855                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 856                         (unsigned int)mtu, (unsigned int)pdu,
 857                         EFX_MAC_PDU_MIN);
 858                 goto fail_inval;
 859         }
 860         if (pdu > EFX_MAC_PDU_MAX) {
 861                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 862                         (unsigned int)mtu, (unsigned int)pdu,
 863                         EFX_MAC_PDU_MAX);
 864                 goto fail_inval;
 865         }
 866
 867         sfc_adapter_lock(sa);
 868
 869         if (pdu != sa->port.pdu) {
 870                 if (sa->state == SFC_ADAPTER_STARTED) {
 871                         sfc_stop(sa);
 872
 873                         old_pdu = sa->port.pdu;
 874                         sa->port.pdu = pdu;
 875                         rc = sfc_start(sa);
 876                         if (rc != 0)
 877                                 goto fail_start;
 878                 } else {
 879                         sa->port.pdu = pdu;
 880                 }
 881         }
 882
 883         /*
 884          * The driver does not use it, but other PMDs update jumbo_frame
 885          * flag and max_rx_pkt_len when MTU is set.
 886          */
 887         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 888         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 889
 890         sfc_adapter_unlock(sa);
 891
 892         sfc_log_init(sa, "done");
 893         return 0;
 894
 895 fail_start:
 896         sa->port.pdu = old_pdu;
 897         if (sfc_start(sa) != 0)
 898                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 899                         "PDU max size - port is stopped",
 900                         (unsigned int)pdu, (unsigned int)old_pdu);
 901         sfc_adapter_unlock(sa);
 902
 903 fail_inval:
 904         sfc_log_init(sa, "failed %d", rc);
 905         SFC_ASSERT(rc > 0);
 906         return -rc;
 907 }
 908 static void
 909 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 910 {
 911         struct sfc_adapter *sa = dev->data->dev_private;
 912         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 913         struct sfc_port *port = &sa->port;
 914         int rc;
 915
 916         sfc_adapter_lock(sa);
 917
 918         if (port->isolated) {
 919                 sfc_err(sa, "isolated mode is active on the port");
 920                 sfc_err(sa, "will not set MAC address");
 921                 goto unlock;
 922         }
 923
 924         if (sa->state != SFC_ADAPTER_STARTED) {
 925                 sfc_info(sa, "the port is not started");
 926                 sfc_info(sa, "the new MAC address will be set on port start");
 927
 928                 goto unlock;
 929         }
 930
 931         if (encp->enc_allow_set_mac_with_installed_filters) {
 932                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 933                 if (rc != 0) {
 934                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 935                         goto unlock;
 936                 }
 937
 938                 /*
 939                  * Changing the MAC address by means of MCDI request
 940                  * has no effect on received traffic, therefore
 941                  * we also need to update unicast filters
 942                  */
 943                 rc = sfc_set_rx_mode(sa);
 944                 if (rc != 0)
 945                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 946         } else {
 947                 sfc_warn(sa, "cannot set MAC address with filters installed");
 948                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 949                 sfc_warn(sa, "(some traffic may be dropped)");
 950
 951                 /*
 952                  * Since setting MAC address with filters installed is not
 953                  * allowed on the adapter, one needs to simply restart adapter
 954                  * so that the new MAC address will be taken from an outer
 955                  * storage and set flawlessly by means of sfc_start() call
 956                  */
 957                 sfc_stop(sa);
 958                 rc = sfc_start(sa);
 959                 if (rc != 0)
 960                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 961         }
 962
 963 unlock:
 964         sfc_adapter_unlock(sa);
 965 }
 966
 967
 968 static int
 969 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 970                      uint32_t nb_mc_addr)
 971 {
 972         struct sfc_adapter *sa = dev->data->dev_private;
 973         struct sfc_port *port = &sa->port;
 974         uint8_t *mc_addrs = port->mcast_addrs;
 975         int rc;
 976         unsigned int i;
 977
 978         if (port->isolated) {
 979                 sfc_err(sa, "isolated mode is active on the port");
 980                 sfc_err(sa, "will not set multicast address list");
 981                 return -ENOTSUP;
 982         }
 983
 984         if (mc_addrs == NULL)
 985                 return -ENOBUFS;
 986
 987         if (nb_mc_addr > port->max_mcast_addrs) {
 988                 sfc_err(sa, "too many multicast addresses: %u > %u",
 989                          nb_mc_addr, port->max_mcast_addrs);
 990                 return -EINVAL;
 991         }
 992
 993         for (i = 0; i < nb_mc_addr; ++i) {
 994                 (void)rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
 995                                  EFX_MAC_ADDR_LEN);
 996                 mc_addrs += EFX_MAC_ADDR_LEN;
 997         }
 998
 999         port->nb_mcast_addrs = nb_mc_addr;
1000
1001         if (sa->state != SFC_ADAPTER_STARTED)
1002                 return 0;
1003
1004         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1005                                         port->nb_mcast_addrs);
1006         if (rc != 0)
1007                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1008
1009         SFC_ASSERT(rc > 0);
1010         return -rc;
1011 }
1012
1013 /*
1014  * The function is used by the secondary process as well. It must not
1015  * use any process-local pointers from the adapter data.
1016  */
1017 static void
1018 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1019                       struct rte_eth_rxq_info *qinfo)
1020 {
1021         struct sfc_adapter *sa = dev->data->dev_private;
1022         struct sfc_rxq_info *rxq_info;
1023         struct sfc_rxq *rxq;
1024
1025         sfc_adapter_lock(sa);
1026
1027         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1028
1029         rxq_info = &sa->rxq_info[rx_queue_id];
1030         rxq = rxq_info->rxq;
1031         SFC_ASSERT(rxq != NULL);
1032
1033         qinfo->mp = rxq->refill_mb_pool;
1034         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1035         qinfo->conf.rx_drop_en = 1;
1036         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1037         qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
1038         qinfo->nb_desc = rxq_info->entries;
1039
1040         sfc_adapter_unlock(sa);
1041 }
1042
1043 /*
1044  * The function is used by the secondary process as well. It must not
1045  * use any process-local pointers from the adapter data.
1046  */
1047 static void
1048 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1049                       struct rte_eth_txq_info *qinfo)
1050 {
1051         struct sfc_adapter *sa = dev->data->dev_private;
1052         struct sfc_txq_info *txq_info;
1053
1054         sfc_adapter_lock(sa);
1055
1056         SFC_ASSERT(tx_queue_id < sa->txq_count);
1057
1058         txq_info = &sa->txq_info[tx_queue_id];
1059         SFC_ASSERT(txq_info->txq != NULL);
1060
1061         memset(qinfo, 0, sizeof(*qinfo));
1062
1063         qinfo->conf.txq_flags = txq_info->txq->flags;
1064         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1065         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1066         qinfo->nb_desc = txq_info->entries;
1067
1068         sfc_adapter_unlock(sa);
1069 }
1070
1071 static uint32_t
1072 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1073 {
1074         struct sfc_adapter *sa = dev->data->dev_private;
1075
1076         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1077
1078         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1079 }
1080
1081 static int
1082 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1083 {
1084         struct sfc_dp_rxq *dp_rxq = queue;
1085
1086         return sfc_rx_qdesc_done(dp_rxq, offset);
1087 }
1088
1089 static int
1090 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1091 {
1092         struct sfc_dp_rxq *dp_rxq = queue;
1093         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1094
1095         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1096 }
1097
1098 static int
1099 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1100 {
1101         struct sfc_dp_txq *dp_txq = queue;
1102         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1103
1104         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1105 }
1106
1107 static int
1108 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1109 {
1110         struct sfc_adapter *sa = dev->data->dev_private;
1111         int rc;
1112
1113         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1114
1115         sfc_adapter_lock(sa);
1116
1117         rc = EINVAL;
1118         if (sa->state != SFC_ADAPTER_STARTED)
1119                 goto fail_not_started;
1120
1121         rc = sfc_rx_qstart(sa, rx_queue_id);
1122         if (rc != 0)
1123                 goto fail_rx_qstart;
1124
1125         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1126
1127         sfc_adapter_unlock(sa);
1128
1129         return 0;
1130
1131 fail_rx_qstart:
1132 fail_not_started:
1133         sfc_adapter_unlock(sa);
1134         SFC_ASSERT(rc > 0);
1135         return -rc;
1136 }
1137
1138 static int
1139 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1140 {
1141         struct sfc_adapter *sa = dev->data->dev_private;
1142
1143         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1144
1145         sfc_adapter_lock(sa);
1146         sfc_rx_qstop(sa, rx_queue_id);
1147
1148         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1149
1150         sfc_adapter_unlock(sa);
1151
1152         return 0;
1153 }
1154
1155 static int
1156 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1157 {
1158         struct sfc_adapter *sa = dev->data->dev_private;
1159         int rc;
1160
1161         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1162
1163         sfc_adapter_lock(sa);
1164
1165         rc = EINVAL;
1166         if (sa->state != SFC_ADAPTER_STARTED)
1167                 goto fail_not_started;
1168
1169         rc = sfc_tx_qstart(sa, tx_queue_id);
1170         if (rc != 0)
1171                 goto fail_tx_qstart;
1172
1173         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1174
1175         sfc_adapter_unlock(sa);
1176         return 0;
1177
1178 fail_tx_qstart:
1179
1180 fail_not_started:
1181         sfc_adapter_unlock(sa);
1182         SFC_ASSERT(rc > 0);
1183         return -rc;
1184 }
1185
1186 static int
1187 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1188 {
1189         struct sfc_adapter *sa = dev->data->dev_private;
1190
1191         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1192
1193         sfc_adapter_lock(sa);
1194
1195         sfc_tx_qstop(sa, tx_queue_id);
1196
1197         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1198
1199         sfc_adapter_unlock(sa);
1200         return 0;
1201 }
1202
1203 #if EFSYS_OPT_RX_SCALE
1204 static int
1205 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1206                           struct rte_eth_rss_conf *rss_conf)
1207 {
1208         struct sfc_adapter *sa = dev->data->dev_private;
1209         struct sfc_port *port = &sa->port;
1210
1211         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1212                 return -ENOTSUP;
1213
1214         if (sa->rss_channels == 0)
1215                 return -EINVAL;
1216
1217         sfc_adapter_lock(sa);
1218
1219         /*
1220          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1221          * hence, conversion is done here to derive a correct set of ETH_RSS
1222          * flags which corresponds to the active EFX configuration stored
1223          * locally in 'sfc_adapter' and kept up-to-date
1224          */
1225         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1226         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1227         if (rss_conf->rss_key != NULL)
1228                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1229
1230         sfc_adapter_unlock(sa);
1231
1232         return 0;
1233 }
1234
1235 static int
1236 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1237                         struct rte_eth_rss_conf *rss_conf)
1238 {
1239         struct sfc_adapter *sa = dev->data->dev_private;
1240         struct sfc_port *port = &sa->port;
1241         unsigned int efx_hash_types;
1242         int rc = 0;
1243
1244         if (port->isolated)
1245                 return -ENOTSUP;
1246
1247         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1248                 sfc_err(sa, "RSS is not available");
1249                 return -ENOTSUP;
1250         }
1251
1252         if (sa->rss_channels == 0) {
1253                 sfc_err(sa, "RSS is not configured");
1254                 return -EINVAL;
1255         }
1256
1257         if ((rss_conf->rss_key != NULL) &&
1258             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1259                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1260                         sizeof(sa->rss_key));
1261                 return -EINVAL;
1262         }
1263
1264         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1265                 sfc_err(sa, "unsupported hash functions requested");
1266                 return -EINVAL;
1267         }
1268
1269         sfc_adapter_lock(sa);
1270
1271         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1272
1273         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1274                                    EFX_RX_HASHALG_TOEPLITZ,
1275                                    efx_hash_types, B_TRUE);
1276         if (rc != 0)
1277                 goto fail_scale_mode_set;
1278
1279         if (rss_conf->rss_key != NULL) {
1280                 if (sa->state == SFC_ADAPTER_STARTED) {
1281                         rc = efx_rx_scale_key_set(sa->nic,
1282                                                   EFX_RSS_CONTEXT_DEFAULT,
1283                                                   rss_conf->rss_key,
1284                                                   sizeof(sa->rss_key));
1285                         if (rc != 0)
1286                                 goto fail_scale_key_set;
1287                 }
1288
1289                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1290         }
1291
1292         sa->rss_hash_types = efx_hash_types;
1293
1294         sfc_adapter_unlock(sa);
1295
1296         return 0;
1297
1298 fail_scale_key_set:
1299         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1300                                   EFX_RX_HASHALG_TOEPLITZ,
1301                                   sa->rss_hash_types, B_TRUE) != 0)
1302                 sfc_err(sa, "failed to restore RSS mode");
1303
1304 fail_scale_mode_set:
1305         sfc_adapter_unlock(sa);
1306         return -rc;
1307 }
1308
1309 static int
1310 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1311                        struct rte_eth_rss_reta_entry64 *reta_conf,
1312                        uint16_t reta_size)
1313 {
1314         struct sfc_adapter *sa = dev->data->dev_private;
1315         struct sfc_port *port = &sa->port;
1316         int entry;
1317
1318         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1319                 return -ENOTSUP;
1320
1321         if (sa->rss_channels == 0)
1322                 return -EINVAL;
1323
1324         if (reta_size != EFX_RSS_TBL_SIZE)
1325                 return -EINVAL;
1326
1327         sfc_adapter_lock(sa);
1328
1329         for (entry = 0; entry < reta_size; entry++) {
1330                 int grp = entry / RTE_RETA_GROUP_SIZE;
1331                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1332
1333                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1334                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1335         }
1336
1337         sfc_adapter_unlock(sa);
1338
1339         return 0;
1340 }
1341
1342 static int
1343 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1344                         struct rte_eth_rss_reta_entry64 *reta_conf,
1345                         uint16_t reta_size)
1346 {
1347         struct sfc_adapter *sa = dev->data->dev_private;
1348         struct sfc_port *port = &sa->port;
1349         unsigned int *rss_tbl_new;
1350         uint16_t entry;
1351         int rc;
1352
1353
1354         if (port->isolated)
1355                 return -ENOTSUP;
1356
1357         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1358                 sfc_err(sa, "RSS is not available");
1359                 return -ENOTSUP;
1360         }
1361
1362         if (sa->rss_channels == 0) {
1363                 sfc_err(sa, "RSS is not configured");
1364                 return -EINVAL;
1365         }
1366
1367         if (reta_size != EFX_RSS_TBL_SIZE) {
1368                 sfc_err(sa, "RETA size is wrong (should be %u)",
1369                         EFX_RSS_TBL_SIZE);
1370                 return -EINVAL;
1371         }
1372
1373         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1374         if (rss_tbl_new == NULL)
1375                 return -ENOMEM;
1376
1377         sfc_adapter_lock(sa);
1378
1379         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1380
1381         for (entry = 0; entry < reta_size; entry++) {
1382                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1383                 struct rte_eth_rss_reta_entry64 *grp;
1384
1385                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1386
1387                 if (grp->mask & (1ull << grp_idx)) {
1388                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1389                                 rc = EINVAL;
1390                                 goto bad_reta_entry;
1391                         }
1392                         rss_tbl_new[entry] = grp->reta[grp_idx];
1393                 }
1394         }
1395
1396         rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1397                                   rss_tbl_new, EFX_RSS_TBL_SIZE);
1398         if (rc == 0)
1399                 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1400
1401 bad_reta_entry:
1402         sfc_adapter_unlock(sa);
1403
1404         rte_free(rss_tbl_new);
1405
1406         SFC_ASSERT(rc >= 0);
1407         return -rc;
1408 }
1409 #endif
1410
1411 static int
1412 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1413                     enum rte_filter_op filter_op,
1414                     void *arg)
1415 {
1416         struct sfc_adapter *sa = dev->data->dev_private;
1417         int rc = ENOTSUP;
1418
1419         sfc_log_init(sa, "entry");
1420
1421         switch (filter_type) {
1422         case RTE_ETH_FILTER_NONE:
1423                 sfc_err(sa, "Global filters configuration not supported");
1424                 break;
1425         case RTE_ETH_FILTER_MACVLAN:
1426                 sfc_err(sa, "MACVLAN filters not supported");
1427                 break;
1428         case RTE_ETH_FILTER_ETHERTYPE:
1429                 sfc_err(sa, "EtherType filters not supported");
1430                 break;
1431         case RTE_ETH_FILTER_FLEXIBLE:
1432                 sfc_err(sa, "Flexible filters not supported");
1433                 break;
1434         case RTE_ETH_FILTER_SYN:
1435                 sfc_err(sa, "SYN filters not supported");
1436                 break;
1437         case RTE_ETH_FILTER_NTUPLE:
1438                 sfc_err(sa, "NTUPLE filters not supported");
1439                 break;
1440         case RTE_ETH_FILTER_TUNNEL:
1441                 sfc_err(sa, "Tunnel filters not supported");
1442                 break;
1443         case RTE_ETH_FILTER_FDIR:
1444                 sfc_err(sa, "Flow Director filters not supported");
1445                 break;
1446         case RTE_ETH_FILTER_HASH:
1447                 sfc_err(sa, "Hash filters not supported");
1448                 break;
1449         case RTE_ETH_FILTER_GENERIC:
1450                 if (filter_op != RTE_ETH_FILTER_GET) {
1451                         rc = EINVAL;
1452                 } else {
1453                         *(const void **)arg = &sfc_flow_ops;
1454                         rc = 0;
1455                 }
1456                 break;
1457         default:
1458                 sfc_err(sa, "Unknown filter type %u", filter_type);
1459                 break;
1460         }
1461
1462         sfc_log_init(sa, "exit: %d", -rc);
1463         SFC_ASSERT(rc >= 0);
1464         return -rc;
1465 }
1466
1467 static const struct eth_dev_ops sfc_eth_dev_ops = {
1468         .dev_configure                  = sfc_dev_configure,
1469         .dev_start                      = sfc_dev_start,
1470         .dev_stop                       = sfc_dev_stop,
1471         .dev_set_link_up                = sfc_dev_set_link_up,
1472         .dev_set_link_down              = sfc_dev_set_link_down,
1473         .dev_close                      = sfc_dev_close,
1474         .promiscuous_enable             = sfc_dev_promisc_enable,
1475         .promiscuous_disable            = sfc_dev_promisc_disable,
1476         .allmulticast_enable            = sfc_dev_allmulti_enable,
1477         .allmulticast_disable           = sfc_dev_allmulti_disable,
1478         .link_update                    = sfc_dev_link_update,
1479         .stats_get                      = sfc_stats_get,
1480         .stats_reset                    = sfc_stats_reset,
1481         .xstats_get                     = sfc_xstats_get,
1482         .xstats_reset                   = sfc_stats_reset,
1483         .xstats_get_names               = sfc_xstats_get_names,
1484         .dev_infos_get                  = sfc_dev_infos_get,
1485         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1486         .mtu_set                        = sfc_dev_set_mtu,
1487         .rx_queue_start                 = sfc_rx_queue_start,
1488         .rx_queue_stop                  = sfc_rx_queue_stop,
1489         .tx_queue_start                 = sfc_tx_queue_start,
1490         .tx_queue_stop                  = sfc_tx_queue_stop,
1491         .rx_queue_setup                 = sfc_rx_queue_setup,
1492         .rx_queue_release               = sfc_rx_queue_release,
1493         .rx_queue_count                 = sfc_rx_queue_count,
1494         .rx_descriptor_done             = sfc_rx_descriptor_done,
1495         .rx_descriptor_status           = sfc_rx_descriptor_status,
1496         .tx_descriptor_status           = sfc_tx_descriptor_status,
1497         .tx_queue_setup                 = sfc_tx_queue_setup,
1498         .tx_queue_release               = sfc_tx_queue_release,
1499         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1500         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1501         .mac_addr_set                   = sfc_mac_addr_set,
1502 #if EFSYS_OPT_RX_SCALE
1503         .reta_update                    = sfc_dev_rss_reta_update,
1504         .reta_query                     = sfc_dev_rss_reta_query,
1505         .rss_hash_update                = sfc_dev_rss_hash_update,
1506         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1507 #endif
1508         .filter_ctrl                    = sfc_dev_filter_ctrl,
1509         .set_mc_addr_list               = sfc_set_mc_addr_list,
1510         .rxq_info_get                   = sfc_rx_queue_info_get,
1511         .txq_info_get                   = sfc_tx_queue_info_get,
1512         .fw_version_get                 = sfc_fw_version_get,
1513         .xstats_get_by_id               = sfc_xstats_get_by_id,
1514         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1515 };
1516
1517 /**
1518  * Duplicate a string in potentially shared memory required for
1519  * multi-process support.
1520  *
1521  * strdup() allocates from process-local heap/memory.
1522  */
1523 static char *
1524 sfc_strdup(const char *str)
1525 {
1526         size_t size;
1527         char *copy;
1528
1529         if (str == NULL)
1530                 return NULL;
1531
1532         size = strlen(str) + 1;
1533         copy = rte_malloc(__func__, size, 0);
1534         if (copy != NULL)
1535                 rte_memcpy(copy, str, size);
1536
1537         return copy;
1538 }
1539
1540 static int
1541 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1542 {
1543         struct sfc_adapter *sa = dev->data->dev_private;
1544         unsigned int avail_caps = 0;
1545         const char *rx_name = NULL;
1546         const char *tx_name = NULL;
1547         int rc;
1548
1549         switch (sa->family) {
1550         case EFX_FAMILY_HUNTINGTON:
1551         case EFX_FAMILY_MEDFORD:
1552                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1553                 break;
1554         default:
1555                 break;
1556         }
1557
1558         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1559                                 sfc_kvarg_string_handler, &rx_name);
1560         if (rc != 0)
1561                 goto fail_kvarg_rx_datapath;
1562
1563         if (rx_name != NULL) {
1564                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1565                 if (sa->dp_rx == NULL) {
1566                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1567                         rc = ENOENT;
1568                         goto fail_dp_rx;
1569                 }
1570                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1571                         sfc_err(sa,
1572                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1573                                 rx_name);
1574                         rc = EINVAL;
1575                         goto fail_dp_rx_caps;
1576                 }
1577         } else {
1578                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1579                 if (sa->dp_rx == NULL) {
1580                         sfc_err(sa, "Rx datapath by caps %#x not found",
1581                                 avail_caps);
1582                         rc = ENOENT;
1583                         goto fail_dp_rx;
1584                 }
1585         }
1586
1587         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1588         if (sa->dp_rx_name == NULL) {
1589                 rc = ENOMEM;
1590                 goto fail_dp_rx_name;
1591         }
1592
1593         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1594
1595         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1596
1597         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1598                                 sfc_kvarg_string_handler, &tx_name);
1599         if (rc != 0)
1600                 goto fail_kvarg_tx_datapath;
1601
1602         if (tx_name != NULL) {
1603                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1604                 if (sa->dp_tx == NULL) {
1605                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1606                         rc = ENOENT;
1607                         goto fail_dp_tx;
1608                 }
1609                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1610                         sfc_err(sa,
1611                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1612                                 tx_name);
1613                         rc = EINVAL;
1614                         goto fail_dp_tx_caps;
1615                 }
1616         } else {
1617                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1618                 if (sa->dp_tx == NULL) {
1619                         sfc_err(sa, "Tx datapath by caps %#x not found",
1620                                 avail_caps);
1621                         rc = ENOENT;
1622                         goto fail_dp_tx;
1623                 }
1624         }
1625
1626         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1627         if (sa->dp_tx_name == NULL) {
1628                 rc = ENOMEM;
1629                 goto fail_dp_tx_name;
1630         }
1631
1632         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1633
1634         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1635
1636         dev->dev_ops = &sfc_eth_dev_ops;
1637
1638         return 0;
1639
1640 fail_dp_tx_name:
1641 fail_dp_tx_caps:
1642         sa->dp_tx = NULL;
1643
1644 fail_dp_tx:
1645 fail_kvarg_tx_datapath:
1646         rte_free(sa->dp_rx_name);
1647         sa->dp_rx_name = NULL;
1648
1649 fail_dp_rx_name:
1650 fail_dp_rx_caps:
1651         sa->dp_rx = NULL;
1652
1653 fail_dp_rx:
1654 fail_kvarg_rx_datapath:
1655         return rc;
1656 }
1657
1658 static void
1659 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1660 {
1661         struct sfc_adapter *sa = dev->data->dev_private;
1662
1663         dev->dev_ops = NULL;
1664         dev->rx_pkt_burst = NULL;
1665         dev->tx_pkt_burst = NULL;
1666
1667         rte_free(sa->dp_tx_name);
1668         sa->dp_tx_name = NULL;
1669         sa->dp_tx = NULL;
1670
1671         rte_free(sa->dp_rx_name);
1672         sa->dp_rx_name = NULL;
1673         sa->dp_rx = NULL;
1674 }
1675
1676 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1677         .rxq_info_get                   = sfc_rx_queue_info_get,
1678         .txq_info_get                   = sfc_tx_queue_info_get,
1679 };
1680
1681 static int
1682 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1683 {
1684         /*
1685          * Device private data has really many process-local pointers.
1686          * Below code should be extremely careful to use data located
1687          * in shared memory only.
1688          */
1689         struct sfc_adapter *sa = dev->data->dev_private;
1690         const struct sfc_dp_rx *dp_rx;
1691         const struct sfc_dp_tx *dp_tx;
1692         int rc;
1693
1694         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1695         if (dp_rx == NULL) {
1696                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1697                 rc = ENOENT;
1698                 goto fail_dp_rx;
1699         }
1700         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1701                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1702                         sa->dp_tx_name);
1703                 rc = EINVAL;
1704                 goto fail_dp_rx_multi_process;
1705         }
1706
1707         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1708         if (dp_tx == NULL) {
1709                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1710                 rc = ENOENT;
1711                 goto fail_dp_tx;
1712         }
1713         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1714                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1715                         sa->dp_tx_name);
1716                 rc = EINVAL;
1717                 goto fail_dp_tx_multi_process;
1718         }
1719
1720         dev->rx_pkt_burst = dp_rx->pkt_burst;
1721         dev->tx_pkt_burst = dp_tx->pkt_burst;
1722         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1723
1724         return 0;
1725
1726 fail_dp_tx_multi_process:
1727 fail_dp_tx:
1728 fail_dp_rx_multi_process:
1729 fail_dp_rx:
1730         return rc;
1731 }
1732
1733 static void
1734 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1735 {
1736         dev->dev_ops = NULL;
1737         dev->tx_pkt_burst = NULL;
1738         dev->rx_pkt_burst = NULL;
1739 }
1740
1741 static void
1742 sfc_register_dp(void)
1743 {
1744         /* Register once */
1745         if (TAILQ_EMPTY(&sfc_dp_head)) {
1746                 /* Prefer EF10 datapath */
1747                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1748                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1749
1750                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1751                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1752                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1753         }
1754 }
1755
1756 static int
1757 sfc_eth_dev_init(struct rte_eth_dev *dev)
1758 {
1759         struct sfc_adapter *sa = dev->data->dev_private;
1760         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1761         int rc;
1762         const efx_nic_cfg_t *encp;
1763         const struct ether_addr *from;
1764
1765         sfc_register_dp();
1766
1767         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1768                 return -sfc_eth_dev_secondary_set_ops(dev);
1769
1770         /* Required for logging */
1771         sa->pci_addr = pci_dev->addr;
1772         sa->port_id = dev->data->port_id;
1773
1774         sa->eth_dev = dev;
1775
1776         /* Copy PCI device info to the dev->data */
1777         rte_eth_copy_pci_info(dev, pci_dev);
1778
1779         dev->data->dev_flags |= RTE_ETH_DEV_DETACHABLE;
1780
1781         rc = sfc_kvargs_parse(sa);
1782         if (rc != 0)
1783                 goto fail_kvargs_parse;
1784
1785         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1786                                 sfc_kvarg_bool_handler, &sa->debug_init);
1787         if (rc != 0)
1788                 goto fail_kvarg_debug_init;
1789
1790         sfc_log_init(sa, "entry");
1791
1792         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1793         if (dev->data->mac_addrs == NULL) {
1794                 rc = ENOMEM;
1795                 goto fail_mac_addrs;
1796         }
1797
1798         sfc_adapter_lock_init(sa);
1799         sfc_adapter_lock(sa);
1800
1801         sfc_log_init(sa, "probing");
1802         rc = sfc_probe(sa);
1803         if (rc != 0)
1804                 goto fail_probe;
1805
1806         sfc_log_init(sa, "set device ops");
1807         rc = sfc_eth_dev_set_ops(dev);
1808         if (rc != 0)
1809                 goto fail_set_ops;
1810
1811         sfc_log_init(sa, "attaching");
1812         rc = sfc_attach(sa);
1813         if (rc != 0)
1814                 goto fail_attach;
1815
1816         encp = efx_nic_cfg_get(sa->nic);
1817
1818         /*
1819          * The arguments are really reverse order in comparison to
1820          * Linux kernel. Copy from NIC config to Ethernet device data.
1821          */
1822         from = (const struct ether_addr *)(encp->enc_mac_addr);
1823         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1824
1825         sfc_adapter_unlock(sa);
1826
1827         sfc_log_init(sa, "done");
1828         return 0;
1829
1830 fail_attach:
1831         sfc_eth_dev_clear_ops(dev);
1832
1833 fail_set_ops:
1834         sfc_unprobe(sa);
1835
1836 fail_probe:
1837         sfc_adapter_unlock(sa);
1838         sfc_adapter_lock_fini(sa);
1839         rte_free(dev->data->mac_addrs);
1840         dev->data->mac_addrs = NULL;
1841
1842 fail_mac_addrs:
1843 fail_kvarg_debug_init:
1844         sfc_kvargs_cleanup(sa);
1845
1846 fail_kvargs_parse:
1847         sfc_log_init(sa, "failed %d", rc);
1848         SFC_ASSERT(rc > 0);
1849         return -rc;
1850 }
1851
1852 static int
1853 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1854 {
1855         struct sfc_adapter *sa;
1856
1857         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1858                 sfc_eth_dev_secondary_clear_ops(dev);
1859                 return 0;
1860         }
1861
1862         sa = dev->data->dev_private;
1863         sfc_log_init(sa, "entry");
1864
1865         sfc_adapter_lock(sa);
1866
1867         sfc_eth_dev_clear_ops(dev);
1868
1869         sfc_detach(sa);
1870         sfc_unprobe(sa);
1871
1872         rte_free(dev->data->mac_addrs);
1873         dev->data->mac_addrs = NULL;
1874
1875         sfc_kvargs_cleanup(sa);
1876
1877         sfc_adapter_unlock(sa);
1878         sfc_adapter_lock_fini(sa);
1879
1880         sfc_log_init(sa, "done");
1881
1882         /* Required for logging, so cleanup last */
1883         sa->eth_dev = NULL;
1884         return 0;
1885 }
1886
1887 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1888         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1889         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1890         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1891         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1892         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1893         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1894         { .vendor_id = 0 /* sentinel */ }
1895 };
1896
1897 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1898         struct rte_pci_device *pci_dev)
1899 {
1900         return rte_eth_dev_pci_generic_probe(pci_dev,
1901                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
1902 }
1903
1904 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
1905 {
1906         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
1907 }
1908
1909 static struct rte_pci_driver sfc_efx_pmd = {
1910         .id_table = pci_id_sfc_efx_map,
1911         .drv_flags =
1912                 RTE_PCI_DRV_INTR_LSC |
1913                 RTE_PCI_DRV_NEED_MAPPING,
1914         .probe = sfc_eth_dev_pci_probe,
1915         .remove = sfc_eth_dev_pci_remove,
1916 };
1917
1918 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
1919 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1920 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
1921 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1922         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1923         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1924         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1925         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1926         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1927         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);