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 (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 149                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 150
 151         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 152                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 153
 154 #if EFSYS_OPT_RX_SCALE
 155         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 156                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 157                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 158                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 159         }
 160 #endif
 161
 162         if (sa->tso)
 163                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 164
 165         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 166         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 167         /* The RXQ hardware requires that the descriptor count is a power
 168          * of 2, but rx_desc_lim cannot properly describe that constraint.
 169          */
 170         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 171
 172         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 173         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 174         /*
 175          * The TXQ hardware requires that the descriptor count is a power
 176          * of 2, but tx_desc_lim cannot properly describe that constraint
 177          */
 178         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 179 }
 180
 181 static const uint32_t *
 182 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 183 {
 184         struct sfc_adapter *sa = dev->data->dev_private;
 185
 186         return sa->dp_rx->supported_ptypes_get();
 187 }
 188
 189 static int
 190 sfc_dev_configure(struct rte_eth_dev *dev)
 191 {
 192         struct rte_eth_dev_data *dev_data = dev->data;
 193         struct sfc_adapter *sa = dev_data->dev_private;
 194         int rc;
 195
 196         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 197                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 198
 199         sfc_adapter_lock(sa);
 200         switch (sa->state) {
 201         case SFC_ADAPTER_CONFIGURED:
 202                 /* FALLTHROUGH */
 203         case SFC_ADAPTER_INITIALIZED:
 204                 rc = sfc_configure(sa);
 205                 break;
 206         default:
 207                 sfc_err(sa, "unexpected adapter state %u to configure",
 208                         sa->state);
 209                 rc = EINVAL;
 210                 break;
 211         }
 212         sfc_adapter_unlock(sa);
 213
 214         sfc_log_init(sa, "done %d", rc);
 215         SFC_ASSERT(rc >= 0);
 216         return -rc;
 217 }
 218
 219 static int
 220 sfc_dev_start(struct rte_eth_dev *dev)
 221 {
 222         struct sfc_adapter *sa = dev->data->dev_private;
 223         int rc;
 224
 225         sfc_log_init(sa, "entry");
 226
 227         sfc_adapter_lock(sa);
 228         rc = sfc_start(sa);
 229         sfc_adapter_unlock(sa);
 230
 231         sfc_log_init(sa, "done %d", rc);
 232         SFC_ASSERT(rc >= 0);
 233         return -rc;
 234 }
 235
 236 static int
 237 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 238 {
 239         struct sfc_adapter *sa = dev->data->dev_private;
 240         struct rte_eth_link *dev_link = &dev->data->dev_link;
 241         struct rte_eth_link old_link;
 242         struct rte_eth_link current_link;
 243
 244         sfc_log_init(sa, "entry");
 245
 246 retry:
 247         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 248         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 249
 250         if (sa->state != SFC_ADAPTER_STARTED) {
 251                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 252                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 253                                          *(uint64_t *)&old_link,
 254                                          *(uint64_t *)&current_link))
 255                         goto retry;
 256         } else if (wait_to_complete) {
 257                 efx_link_mode_t link_mode;
 258
 259                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 260                         link_mode = EFX_LINK_UNKNOWN;
 261                 sfc_port_link_mode_to_info(link_mode, &current_link);
 262
 263                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 264                                          *(uint64_t *)&old_link,
 265                                          *(uint64_t *)&current_link))
 266                         goto retry;
 267         } else {
 268                 sfc_ev_mgmt_qpoll(sa);
 269                 *(int64_t *)&current_link =
 270                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 271         }
 272
 273         if (old_link.link_status != current_link.link_status)
 274                 sfc_info(sa, "Link status is %s",
 275                          current_link.link_status ? "UP" : "DOWN");
 276
 277         return old_link.link_status == current_link.link_status ? 0 : -1;
 278 }
 279
 280 static void
 281 sfc_dev_stop(struct rte_eth_dev *dev)
 282 {
 283         struct sfc_adapter *sa = dev->data->dev_private;
 284
 285         sfc_log_init(sa, "entry");
 286
 287         sfc_adapter_lock(sa);
 288         sfc_stop(sa);
 289         sfc_adapter_unlock(sa);
 290
 291         sfc_log_init(sa, "done");
 292 }
 293
 294 static int
 295 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 296 {
 297         struct sfc_adapter *sa = dev->data->dev_private;
 298         int rc;
 299
 300         sfc_log_init(sa, "entry");
 301
 302         sfc_adapter_lock(sa);
 303         rc = sfc_start(sa);
 304         sfc_adapter_unlock(sa);
 305
 306         SFC_ASSERT(rc >= 0);
 307         return -rc;
 308 }
 309
 310 static int
 311 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 312 {
 313         struct sfc_adapter *sa = dev->data->dev_private;
 314
 315         sfc_log_init(sa, "entry");
 316
 317         sfc_adapter_lock(sa);
 318         sfc_stop(sa);
 319         sfc_adapter_unlock(sa);
 320
 321         return 0;
 322 }
 323
 324 static void
 325 sfc_dev_close(struct rte_eth_dev *dev)
 326 {
 327         struct sfc_adapter *sa = dev->data->dev_private;
 328
 329         sfc_log_init(sa, "entry");
 330
 331         sfc_adapter_lock(sa);
 332         switch (sa->state) {
 333         case SFC_ADAPTER_STARTED:
 334                 sfc_stop(sa);
 335                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 336                 /* FALLTHROUGH */
 337         case SFC_ADAPTER_CONFIGURED:
 338                 sfc_close(sa);
 339                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 340                 /* FALLTHROUGH */
 341         case SFC_ADAPTER_INITIALIZED:
 342                 break;
 343         default:
 344                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 345                 break;
 346         }
 347         sfc_adapter_unlock(sa);
 348
 349         sfc_log_init(sa, "done");
 350 }
 351
 352 static void
 353 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 354                    boolean_t enabled)
 355 {
 356         struct sfc_port *port;
 357         boolean_t *toggle;
 358         struct sfc_adapter *sa = dev->data->dev_private;
 359         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 360         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 361
 362         sfc_adapter_lock(sa);
 363
 364         port = &sa->port;
 365         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 366
 367         if (*toggle != enabled) {
 368                 *toggle = enabled;
 369
 370                 if (port->isolated) {
 371                         sfc_warn(sa, "isolated mode is active on the port");
 372                         sfc_warn(sa, "the change is to be applied on the next "
 373                                      "start provided that isolated mode is "
 374                                      "disabled prior the next start");
 375                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 376                            (sfc_set_rx_mode(sa) != 0)) {
 377                         *toggle = !(enabled);
 378                         sfc_warn(sa, "Failed to %s %s mode",
 379                                  ((enabled) ? "enable" : "disable"), desc);
 380                 }
 381         }
 382
 383         sfc_adapter_unlock(sa);
 384 }
 385
 386 static void
 387 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 388 {
 389         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 390 }
 391
 392 static void
 393 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 394 {
 395         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 396 }
 397
 398 static void
 399 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 400 {
 401         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 402 }
 403
 404 static void
 405 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 406 {
 407         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 408 }
 409
 410 static int
 411 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 412                    uint16_t nb_rx_desc, unsigned int socket_id,
 413                    const struct rte_eth_rxconf *rx_conf,
 414                    struct rte_mempool *mb_pool)
 415 {
 416         struct sfc_adapter *sa = dev->data->dev_private;
 417         int rc;
 418
 419         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 420                      rx_queue_id, nb_rx_desc, socket_id);
 421
 422         sfc_adapter_lock(sa);
 423
 424         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 425                           rx_conf, mb_pool);
 426         if (rc != 0)
 427                 goto fail_rx_qinit;
 428
 429         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 430
 431         sfc_adapter_unlock(sa);
 432
 433         return 0;
 434
 435 fail_rx_qinit:
 436         sfc_adapter_unlock(sa);
 437         SFC_ASSERT(rc > 0);
 438         return -rc;
 439 }
 440
 441 static void
 442 sfc_rx_queue_release(void *queue)
 443 {
 444         struct sfc_dp_rxq *dp_rxq = queue;
 445         struct sfc_rxq *rxq;
 446         struct sfc_adapter *sa;
 447         unsigned int sw_index;
 448
 449         if (dp_rxq == NULL)
 450                 return;
 451
 452         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 453         sa = rxq->evq->sa;
 454         sfc_adapter_lock(sa);
 455
 456         sw_index = sfc_rxq_sw_index(rxq);
 457
 458         sfc_log_init(sa, "RxQ=%u", sw_index);
 459
 460         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 461
 462         sfc_rx_qfini(sa, sw_index);
 463
 464         sfc_adapter_unlock(sa);
 465 }
 466
 467 static int
 468 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 469                    uint16_t nb_tx_desc, unsigned int socket_id,
 470                    const struct rte_eth_txconf *tx_conf)
 471 {
 472         struct sfc_adapter *sa = dev->data->dev_private;
 473         int rc;
 474
 475         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 476                      tx_queue_id, nb_tx_desc, socket_id);
 477
 478         sfc_adapter_lock(sa);
 479
 480         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 481         if (rc != 0)
 482                 goto fail_tx_qinit;
 483
 484         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 485
 486         sfc_adapter_unlock(sa);
 487         return 0;
 488
 489 fail_tx_qinit:
 490         sfc_adapter_unlock(sa);
 491         SFC_ASSERT(rc > 0);
 492         return -rc;
 493 }
 494
 495 static void
 496 sfc_tx_queue_release(void *queue)
 497 {
 498         struct sfc_dp_txq *dp_txq = queue;
 499         struct sfc_txq *txq;
 500         unsigned int sw_index;
 501         struct sfc_adapter *sa;
 502
 503         if (dp_txq == NULL)
 504                 return;
 505
 506         txq = sfc_txq_by_dp_txq(dp_txq);
 507         sw_index = sfc_txq_sw_index(txq);
 508
 509         SFC_ASSERT(txq->evq != NULL);
 510         sa = txq->evq->sa;
 511
 512         sfc_log_init(sa, "TxQ = %u", sw_index);
 513
 514         sfc_adapter_lock(sa);
 515
 516         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 517         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 518
 519         sfc_tx_qfini(sa, sw_index);
 520
 521         sfc_adapter_unlock(sa);
 522 }
 523
 524 static void
 525 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 526 {
 527         struct sfc_adapter *sa = dev->data->dev_private;
 528         struct sfc_port *port = &sa->port;
 529         uint64_t *mac_stats;
 530
 531         rte_spinlock_lock(&port->mac_stats_lock);
 532
 533         if (sfc_port_update_mac_stats(sa) != 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 }
 591
 592 static void
 593 sfc_stats_reset(struct rte_eth_dev *dev)
 594 {
 595         struct sfc_adapter *sa = dev->data->dev_private;
 596         struct sfc_port *port = &sa->port;
 597         int rc;
 598
 599         if (sa->state != SFC_ADAPTER_STARTED) {
 600                 /*
 601                  * The operation cannot be done if port is not started; it
 602                  * will be scheduled to be done during the next port start
 603                  */
 604                 port->mac_stats_reset_pending = B_TRUE;
 605                 return;
 606         }
 607
 608         rc = sfc_port_reset_mac_stats(sa);
 609         if (rc != 0)
 610                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 611 }
 612
 613 static int
 614 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 615                unsigned int xstats_count)
 616 {
 617         struct sfc_adapter *sa = dev->data->dev_private;
 618         struct sfc_port *port = &sa->port;
 619         uint64_t *mac_stats;
 620         int rc;
 621         unsigned int i;
 622         int nstats = 0;
 623
 624         rte_spinlock_lock(&port->mac_stats_lock);
 625
 626         rc = sfc_port_update_mac_stats(sa);
 627         if (rc != 0) {
 628                 SFC_ASSERT(rc > 0);
 629                 nstats = -rc;
 630                 goto unlock;
 631         }
 632
 633         mac_stats = port->mac_stats_buf;
 634
 635         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 636                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 637                         if (xstats != NULL && nstats < (int)xstats_count) {
 638                                 xstats[nstats].id = nstats;
 639                                 xstats[nstats].value = mac_stats[i];
 640                         }
 641                         nstats++;
 642                 }
 643         }
 644
 645 unlock:
 646         rte_spinlock_unlock(&port->mac_stats_lock);
 647
 648         return nstats;
 649 }
 650
 651 static int
 652 sfc_xstats_get_names(struct rte_eth_dev *dev,
 653                      struct rte_eth_xstat_name *xstats_names,
 654                      unsigned int xstats_count)
 655 {
 656         struct sfc_adapter *sa = dev->data->dev_private;
 657         struct sfc_port *port = &sa->port;
 658         unsigned int i;
 659         unsigned int nstats = 0;
 660
 661         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 662                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 663                         if (xstats_names != NULL && nstats < xstats_count)
 664                                 strncpy(xstats_names[nstats].name,
 665                                         efx_mac_stat_name(sa->nic, i),
 666                                         sizeof(xstats_names[0].name));
 667                         nstats++;
 668                 }
 669         }
 670
 671         return nstats;
 672 }
 673
 674 static int
 675 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 676                      uint64_t *values, unsigned int n)
 677 {
 678         struct sfc_adapter *sa = dev->data->dev_private;
 679         struct sfc_port *port = &sa->port;
 680         uint64_t *mac_stats;
 681         unsigned int nb_supported = 0;
 682         unsigned int nb_written = 0;
 683         unsigned int i;
 684         int ret;
 685         int rc;
 686
 687         if (unlikely(values == NULL) ||
 688             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 689                 return port->mac_stats_nb_supported;
 690
 691         rte_spinlock_lock(&port->mac_stats_lock);
 692
 693         rc = sfc_port_update_mac_stats(sa);
 694         if (rc != 0) {
 695                 SFC_ASSERT(rc > 0);
 696                 ret = -rc;
 697                 goto unlock;
 698         }
 699
 700         mac_stats = port->mac_stats_buf;
 701
 702         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 703                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 704                         continue;
 705
 706                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 707                         values[nb_written++] = mac_stats[i];
 708
 709                 ++nb_supported;
 710         }
 711
 712         ret = nb_written;
 713
 714 unlock:
 715         rte_spinlock_unlock(&port->mac_stats_lock);
 716
 717         return ret;
 718 }
 719
 720 static int
 721 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 722                            struct rte_eth_xstat_name *xstats_names,
 723                            const uint64_t *ids, unsigned int size)
 724 {
 725         struct sfc_adapter *sa = dev->data->dev_private;
 726         struct sfc_port *port = &sa->port;
 727         unsigned int nb_supported = 0;
 728         unsigned int nb_written = 0;
 729         unsigned int i;
 730
 731         if (unlikely(xstats_names == NULL) ||
 732             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 733                 return port->mac_stats_nb_supported;
 734
 735         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 736                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 737                         continue;
 738
 739                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 740                         char *name = xstats_names[nb_written++].name;
 741
 742                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 743                                 sizeof(xstats_names[0].name));
 744                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 745                 }
 746
 747                 ++nb_supported;
 748         }
 749
 750         return nb_written;
 751 }
 752
 753 static int
 754 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 755 {
 756         struct sfc_adapter *sa = dev->data->dev_private;
 757         unsigned int wanted_fc, link_fc;
 758
 759         memset(fc_conf, 0, sizeof(*fc_conf));
 760
 761         sfc_adapter_lock(sa);
 762
 763         if (sa->state == SFC_ADAPTER_STARTED)
 764                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 765         else
 766                 link_fc = sa->port.flow_ctrl;
 767
 768         switch (link_fc) {
 769         case 0:
 770                 fc_conf->mode = RTE_FC_NONE;
 771                 break;
 772         case EFX_FCNTL_RESPOND:
 773                 fc_conf->mode = RTE_FC_RX_PAUSE;
 774                 break;
 775         case EFX_FCNTL_GENERATE:
 776                 fc_conf->mode = RTE_FC_TX_PAUSE;
 777                 break;
 778         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 779                 fc_conf->mode = RTE_FC_FULL;
 780                 break;
 781         default:
 782                 sfc_err(sa, "%s: unexpected flow control value %#x",
 783                         __func__, link_fc);
 784         }
 785
 786         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 787
 788         sfc_adapter_unlock(sa);
 789
 790         return 0;
 791 }
 792
 793 static int
 794 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 795 {
 796         struct sfc_adapter *sa = dev->data->dev_private;
 797         struct sfc_port *port = &sa->port;
 798         unsigned int fcntl;
 799         int rc;
 800
 801         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 802             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 803             fc_conf->mac_ctrl_frame_fwd != 0) {
 804                 sfc_err(sa, "unsupported flow control settings specified");
 805                 rc = EINVAL;
 806                 goto fail_inval;
 807         }
 808
 809         switch (fc_conf->mode) {
 810         case RTE_FC_NONE:
 811                 fcntl = 0;
 812                 break;
 813         case RTE_FC_RX_PAUSE:
 814                 fcntl = EFX_FCNTL_RESPOND;
 815                 break;
 816         case RTE_FC_TX_PAUSE:
 817                 fcntl = EFX_FCNTL_GENERATE;
 818                 break;
 819         case RTE_FC_FULL:
 820                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 821                 break;
 822         default:
 823                 rc = EINVAL;
 824                 goto fail_inval;
 825         }
 826
 827         sfc_adapter_lock(sa);
 828
 829         if (sa->state == SFC_ADAPTER_STARTED) {
 830                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 831                 if (rc != 0)
 832                         goto fail_mac_fcntl_set;
 833         }
 834
 835         port->flow_ctrl = fcntl;
 836         port->flow_ctrl_autoneg = fc_conf->autoneg;
 837
 838         sfc_adapter_unlock(sa);
 839
 840         return 0;
 841
 842 fail_mac_fcntl_set:
 843         sfc_adapter_unlock(sa);
 844 fail_inval:
 845         SFC_ASSERT(rc > 0);
 846         return -rc;
 847 }
 848
 849 static int
 850 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 851 {
 852         struct sfc_adapter *sa = dev->data->dev_private;
 853         size_t pdu = EFX_MAC_PDU(mtu);
 854         size_t old_pdu;
 855         int rc;
 856
 857         sfc_log_init(sa, "mtu=%u", mtu);
 858
 859         rc = EINVAL;
 860         if (pdu < EFX_MAC_PDU_MIN) {
 861                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 862                         (unsigned int)mtu, (unsigned int)pdu,
 863                         EFX_MAC_PDU_MIN);
 864                 goto fail_inval;
 865         }
 866         if (pdu > EFX_MAC_PDU_MAX) {
 867                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 868                         (unsigned int)mtu, (unsigned int)pdu,
 869                         EFX_MAC_PDU_MAX);
 870                 goto fail_inval;
 871         }
 872
 873         sfc_adapter_lock(sa);
 874
 875         if (pdu != sa->port.pdu) {
 876                 if (sa->state == SFC_ADAPTER_STARTED) {
 877                         sfc_stop(sa);
 878
 879                         old_pdu = sa->port.pdu;
 880                         sa->port.pdu = pdu;
 881                         rc = sfc_start(sa);
 882                         if (rc != 0)
 883                                 goto fail_start;
 884                 } else {
 885                         sa->port.pdu = pdu;
 886                 }
 887         }
 888
 889         /*
 890          * The driver does not use it, but other PMDs update jumbo_frame
 891          * flag and max_rx_pkt_len when MTU is set.
 892          */
 893         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 894         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 895
 896         sfc_adapter_unlock(sa);
 897
 898         sfc_log_init(sa, "done");
 899         return 0;
 900
 901 fail_start:
 902         sa->port.pdu = old_pdu;
 903         if (sfc_start(sa) != 0)
 904                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 905                         "PDU max size - port is stopped",
 906                         (unsigned int)pdu, (unsigned int)old_pdu);
 907         sfc_adapter_unlock(sa);
 908
 909 fail_inval:
 910         sfc_log_init(sa, "failed %d", rc);
 911         SFC_ASSERT(rc > 0);
 912         return -rc;
 913 }
 914 static void
 915 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 916 {
 917         struct sfc_adapter *sa = dev->data->dev_private;
 918         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 919         struct sfc_port *port = &sa->port;
 920         int rc;
 921
 922         sfc_adapter_lock(sa);
 923
 924         if (port->isolated) {
 925                 sfc_err(sa, "isolated mode is active on the port");
 926                 sfc_err(sa, "will not set MAC address");
 927                 goto unlock;
 928         }
 929
 930         if (sa->state != SFC_ADAPTER_STARTED) {
 931                 sfc_info(sa, "the port is not started");
 932                 sfc_info(sa, "the new MAC address will be set on port start");
 933
 934                 goto unlock;
 935         }
 936
 937         if (encp->enc_allow_set_mac_with_installed_filters) {
 938                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 939                 if (rc != 0) {
 940                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 941                         goto unlock;
 942                 }
 943
 944                 /*
 945                  * Changing the MAC address by means of MCDI request
 946                  * has no effect on received traffic, therefore
 947                  * we also need to update unicast filters
 948                  */
 949                 rc = sfc_set_rx_mode(sa);
 950                 if (rc != 0)
 951                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 952         } else {
 953                 sfc_warn(sa, "cannot set MAC address with filters installed");
 954                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 955                 sfc_warn(sa, "(some traffic may be dropped)");
 956
 957                 /*
 958                  * Since setting MAC address with filters installed is not
 959                  * allowed on the adapter, one needs to simply restart adapter
 960                  * so that the new MAC address will be taken from an outer
 961                  * storage and set flawlessly by means of sfc_start() call
 962                  */
 963                 sfc_stop(sa);
 964                 rc = sfc_start(sa);
 965                 if (rc != 0)
 966                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 967         }
 968
 969 unlock:
 970         sfc_adapter_unlock(sa);
 971 }
 972
 973
 974 static int
 975 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 976                      uint32_t nb_mc_addr)
 977 {
 978         struct sfc_adapter *sa = dev->data->dev_private;
 979         struct sfc_port *port = &sa->port;
 980         uint8_t *mc_addrs = port->mcast_addrs;
 981         int rc;
 982         unsigned int i;
 983
 984         if (port->isolated) {
 985                 sfc_err(sa, "isolated mode is active on the port");
 986                 sfc_err(sa, "will not set multicast address list");
 987                 return -ENOTSUP;
 988         }
 989
 990         if (mc_addrs == NULL)
 991                 return -ENOBUFS;
 992
 993         if (nb_mc_addr > port->max_mcast_addrs) {
 994                 sfc_err(sa, "too many multicast addresses: %u > %u",
 995                          nb_mc_addr, port->max_mcast_addrs);
 996                 return -EINVAL;
 997         }
 998
 999         for (i = 0; i < nb_mc_addr; ++i) {
1000                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1001                                  EFX_MAC_ADDR_LEN);
1002                 mc_addrs += EFX_MAC_ADDR_LEN;
1003         }
1004
1005         port->nb_mcast_addrs = nb_mc_addr;
1006
1007         if (sa->state != SFC_ADAPTER_STARTED)
1008                 return 0;
1009
1010         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1011                                         port->nb_mcast_addrs);
1012         if (rc != 0)
1013                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1014
1015         SFC_ASSERT(rc > 0);
1016         return -rc;
1017 }
1018
1019 /*
1020  * The function is used by the secondary process as well. It must not
1021  * use any process-local pointers from the adapter data.
1022  */
1023 static void
1024 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1025                       struct rte_eth_rxq_info *qinfo)
1026 {
1027         struct sfc_adapter *sa = dev->data->dev_private;
1028         struct sfc_rxq_info *rxq_info;
1029         struct sfc_rxq *rxq;
1030
1031         sfc_adapter_lock(sa);
1032
1033         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1034
1035         rxq_info = &sa->rxq_info[rx_queue_id];
1036         rxq = rxq_info->rxq;
1037         SFC_ASSERT(rxq != NULL);
1038
1039         qinfo->mp = rxq->refill_mb_pool;
1040         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1041         qinfo->conf.rx_drop_en = 1;
1042         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1043         qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
1044         qinfo->nb_desc = rxq_info->entries;
1045
1046         sfc_adapter_unlock(sa);
1047 }
1048
1049 /*
1050  * The function is used by the secondary process as well. It must not
1051  * use any process-local pointers from the adapter data.
1052  */
1053 static void
1054 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1055                       struct rte_eth_txq_info *qinfo)
1056 {
1057         struct sfc_adapter *sa = dev->data->dev_private;
1058         struct sfc_txq_info *txq_info;
1059
1060         sfc_adapter_lock(sa);
1061
1062         SFC_ASSERT(tx_queue_id < sa->txq_count);
1063
1064         txq_info = &sa->txq_info[tx_queue_id];
1065         SFC_ASSERT(txq_info->txq != NULL);
1066
1067         memset(qinfo, 0, sizeof(*qinfo));
1068
1069         qinfo->conf.txq_flags = txq_info->txq->flags;
1070         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1071         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1072         qinfo->nb_desc = txq_info->entries;
1073
1074         sfc_adapter_unlock(sa);
1075 }
1076
1077 static uint32_t
1078 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1079 {
1080         struct sfc_adapter *sa = dev->data->dev_private;
1081
1082         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1083
1084         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1085 }
1086
1087 static int
1088 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1089 {
1090         struct sfc_dp_rxq *dp_rxq = queue;
1091
1092         return sfc_rx_qdesc_done(dp_rxq, offset);
1093 }
1094
1095 static int
1096 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1097 {
1098         struct sfc_dp_rxq *dp_rxq = queue;
1099         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1100
1101         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1102 }
1103
1104 static int
1105 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1106 {
1107         struct sfc_dp_txq *dp_txq = queue;
1108         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1109
1110         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1111 }
1112
1113 static int
1114 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1115 {
1116         struct sfc_adapter *sa = dev->data->dev_private;
1117         int rc;
1118
1119         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1120
1121         sfc_adapter_lock(sa);
1122
1123         rc = EINVAL;
1124         if (sa->state != SFC_ADAPTER_STARTED)
1125                 goto fail_not_started;
1126
1127         rc = sfc_rx_qstart(sa, rx_queue_id);
1128         if (rc != 0)
1129                 goto fail_rx_qstart;
1130
1131         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1132
1133         sfc_adapter_unlock(sa);
1134
1135         return 0;
1136
1137 fail_rx_qstart:
1138 fail_not_started:
1139         sfc_adapter_unlock(sa);
1140         SFC_ASSERT(rc > 0);
1141         return -rc;
1142 }
1143
1144 static int
1145 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1146 {
1147         struct sfc_adapter *sa = dev->data->dev_private;
1148
1149         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1150
1151         sfc_adapter_lock(sa);
1152         sfc_rx_qstop(sa, rx_queue_id);
1153
1154         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1155
1156         sfc_adapter_unlock(sa);
1157
1158         return 0;
1159 }
1160
1161 static int
1162 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1163 {
1164         struct sfc_adapter *sa = dev->data->dev_private;
1165         int rc;
1166
1167         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1168
1169         sfc_adapter_lock(sa);
1170
1171         rc = EINVAL;
1172         if (sa->state != SFC_ADAPTER_STARTED)
1173                 goto fail_not_started;
1174
1175         rc = sfc_tx_qstart(sa, tx_queue_id);
1176         if (rc != 0)
1177                 goto fail_tx_qstart;
1178
1179         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1180
1181         sfc_adapter_unlock(sa);
1182         return 0;
1183
1184 fail_tx_qstart:
1185
1186 fail_not_started:
1187         sfc_adapter_unlock(sa);
1188         SFC_ASSERT(rc > 0);
1189         return -rc;
1190 }
1191
1192 static int
1193 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1194 {
1195         struct sfc_adapter *sa = dev->data->dev_private;
1196
1197         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1198
1199         sfc_adapter_lock(sa);
1200
1201         sfc_tx_qstop(sa, tx_queue_id);
1202
1203         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1204
1205         sfc_adapter_unlock(sa);
1206         return 0;
1207 }
1208
1209 #if EFSYS_OPT_RX_SCALE
1210 static int
1211 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1212                           struct rte_eth_rss_conf *rss_conf)
1213 {
1214         struct sfc_adapter *sa = dev->data->dev_private;
1215         struct sfc_port *port = &sa->port;
1216
1217         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1218                 return -ENOTSUP;
1219
1220         if (sa->rss_channels == 0)
1221                 return -EINVAL;
1222
1223         sfc_adapter_lock(sa);
1224
1225         /*
1226          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1227          * hence, conversion is done here to derive a correct set of ETH_RSS
1228          * flags which corresponds to the active EFX configuration stored
1229          * locally in 'sfc_adapter' and kept up-to-date
1230          */
1231         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1232         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1233         if (rss_conf->rss_key != NULL)
1234                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1235
1236         sfc_adapter_unlock(sa);
1237
1238         return 0;
1239 }
1240
1241 static int
1242 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1243                         struct rte_eth_rss_conf *rss_conf)
1244 {
1245         struct sfc_adapter *sa = dev->data->dev_private;
1246         struct sfc_port *port = &sa->port;
1247         unsigned int efx_hash_types;
1248         int rc = 0;
1249
1250         if (port->isolated)
1251                 return -ENOTSUP;
1252
1253         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1254                 sfc_err(sa, "RSS is not available");
1255                 return -ENOTSUP;
1256         }
1257
1258         if (sa->rss_channels == 0) {
1259                 sfc_err(sa, "RSS is not configured");
1260                 return -EINVAL;
1261         }
1262
1263         if ((rss_conf->rss_key != NULL) &&
1264             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1265                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1266                         sizeof(sa->rss_key));
1267                 return -EINVAL;
1268         }
1269
1270         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1271                 sfc_err(sa, "unsupported hash functions requested");
1272                 return -EINVAL;
1273         }
1274
1275         sfc_adapter_lock(sa);
1276
1277         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1278
1279         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1280                                    EFX_RX_HASHALG_TOEPLITZ,
1281                                    efx_hash_types, B_TRUE);
1282         if (rc != 0)
1283                 goto fail_scale_mode_set;
1284
1285         if (rss_conf->rss_key != NULL) {
1286                 if (sa->state == SFC_ADAPTER_STARTED) {
1287                         rc = efx_rx_scale_key_set(sa->nic,
1288                                                   EFX_RSS_CONTEXT_DEFAULT,
1289                                                   rss_conf->rss_key,
1290                                                   sizeof(sa->rss_key));
1291                         if (rc != 0)
1292                                 goto fail_scale_key_set;
1293                 }
1294
1295                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1296         }
1297
1298         sa->rss_hash_types = efx_hash_types;
1299
1300         sfc_adapter_unlock(sa);
1301
1302         return 0;
1303
1304 fail_scale_key_set:
1305         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1306                                   EFX_RX_HASHALG_TOEPLITZ,
1307                                   sa->rss_hash_types, B_TRUE) != 0)
1308                 sfc_err(sa, "failed to restore RSS mode");
1309
1310 fail_scale_mode_set:
1311         sfc_adapter_unlock(sa);
1312         return -rc;
1313 }
1314
1315 static int
1316 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1317                        struct rte_eth_rss_reta_entry64 *reta_conf,
1318                        uint16_t reta_size)
1319 {
1320         struct sfc_adapter *sa = dev->data->dev_private;
1321         struct sfc_port *port = &sa->port;
1322         int entry;
1323
1324         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1325                 return -ENOTSUP;
1326
1327         if (sa->rss_channels == 0)
1328                 return -EINVAL;
1329
1330         if (reta_size != EFX_RSS_TBL_SIZE)
1331                 return -EINVAL;
1332
1333         sfc_adapter_lock(sa);
1334
1335         for (entry = 0; entry < reta_size; entry++) {
1336                 int grp = entry / RTE_RETA_GROUP_SIZE;
1337                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1338
1339                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1340                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1341         }
1342
1343         sfc_adapter_unlock(sa);
1344
1345         return 0;
1346 }
1347
1348 static int
1349 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1350                         struct rte_eth_rss_reta_entry64 *reta_conf,
1351                         uint16_t reta_size)
1352 {
1353         struct sfc_adapter *sa = dev->data->dev_private;
1354         struct sfc_port *port = &sa->port;
1355         unsigned int *rss_tbl_new;
1356         uint16_t entry;
1357         int rc;
1358
1359
1360         if (port->isolated)
1361                 return -ENOTSUP;
1362
1363         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1364                 sfc_err(sa, "RSS is not available");
1365                 return -ENOTSUP;
1366         }
1367
1368         if (sa->rss_channels == 0) {
1369                 sfc_err(sa, "RSS is not configured");
1370                 return -EINVAL;
1371         }
1372
1373         if (reta_size != EFX_RSS_TBL_SIZE) {
1374                 sfc_err(sa, "RETA size is wrong (should be %u)",
1375                         EFX_RSS_TBL_SIZE);
1376                 return -EINVAL;
1377         }
1378
1379         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1380         if (rss_tbl_new == NULL)
1381                 return -ENOMEM;
1382
1383         sfc_adapter_lock(sa);
1384
1385         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1386
1387         for (entry = 0; entry < reta_size; entry++) {
1388                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1389                 struct rte_eth_rss_reta_entry64 *grp;
1390
1391                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1392
1393                 if (grp->mask & (1ull << grp_idx)) {
1394                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1395                                 rc = EINVAL;
1396                                 goto bad_reta_entry;
1397                         }
1398                         rss_tbl_new[entry] = grp->reta[grp_idx];
1399                 }
1400         }
1401
1402         rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1403                                   rss_tbl_new, EFX_RSS_TBL_SIZE);
1404         if (rc == 0)
1405                 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1406
1407 bad_reta_entry:
1408         sfc_adapter_unlock(sa);
1409
1410         rte_free(rss_tbl_new);
1411
1412         SFC_ASSERT(rc >= 0);
1413         return -rc;
1414 }
1415 #endif
1416
1417 static int
1418 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1419                     enum rte_filter_op filter_op,
1420                     void *arg)
1421 {
1422         struct sfc_adapter *sa = dev->data->dev_private;
1423         int rc = ENOTSUP;
1424
1425         sfc_log_init(sa, "entry");
1426
1427         switch (filter_type) {
1428         case RTE_ETH_FILTER_NONE:
1429                 sfc_err(sa, "Global filters configuration not supported");
1430                 break;
1431         case RTE_ETH_FILTER_MACVLAN:
1432                 sfc_err(sa, "MACVLAN filters not supported");
1433                 break;
1434         case RTE_ETH_FILTER_ETHERTYPE:
1435                 sfc_err(sa, "EtherType filters not supported");
1436                 break;
1437         case RTE_ETH_FILTER_FLEXIBLE:
1438                 sfc_err(sa, "Flexible filters not supported");
1439                 break;
1440         case RTE_ETH_FILTER_SYN:
1441                 sfc_err(sa, "SYN filters not supported");
1442                 break;
1443         case RTE_ETH_FILTER_NTUPLE:
1444                 sfc_err(sa, "NTUPLE filters not supported");
1445                 break;
1446         case RTE_ETH_FILTER_TUNNEL:
1447                 sfc_err(sa, "Tunnel filters not supported");
1448                 break;
1449         case RTE_ETH_FILTER_FDIR:
1450                 sfc_err(sa, "Flow Director filters not supported");
1451                 break;
1452         case RTE_ETH_FILTER_HASH:
1453                 sfc_err(sa, "Hash filters not supported");
1454                 break;
1455         case RTE_ETH_FILTER_GENERIC:
1456                 if (filter_op != RTE_ETH_FILTER_GET) {
1457                         rc = EINVAL;
1458                 } else {
1459                         *(const void **)arg = &sfc_flow_ops;
1460                         rc = 0;
1461                 }
1462                 break;
1463         default:
1464                 sfc_err(sa, "Unknown filter type %u", filter_type);
1465                 break;
1466         }
1467
1468         sfc_log_init(sa, "exit: %d", -rc);
1469         SFC_ASSERT(rc >= 0);
1470         return -rc;
1471 }
1472
1473 static const struct eth_dev_ops sfc_eth_dev_ops = {
1474         .dev_configure                  = sfc_dev_configure,
1475         .dev_start                      = sfc_dev_start,
1476         .dev_stop                       = sfc_dev_stop,
1477         .dev_set_link_up                = sfc_dev_set_link_up,
1478         .dev_set_link_down              = sfc_dev_set_link_down,
1479         .dev_close                      = sfc_dev_close,
1480         .promiscuous_enable             = sfc_dev_promisc_enable,
1481         .promiscuous_disable            = sfc_dev_promisc_disable,
1482         .allmulticast_enable            = sfc_dev_allmulti_enable,
1483         .allmulticast_disable           = sfc_dev_allmulti_disable,
1484         .link_update                    = sfc_dev_link_update,
1485         .stats_get                      = sfc_stats_get,
1486         .stats_reset                    = sfc_stats_reset,
1487         .xstats_get                     = sfc_xstats_get,
1488         .xstats_reset                   = sfc_stats_reset,
1489         .xstats_get_names               = sfc_xstats_get_names,
1490         .dev_infos_get                  = sfc_dev_infos_get,
1491         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1492         .mtu_set                        = sfc_dev_set_mtu,
1493         .rx_queue_start                 = sfc_rx_queue_start,
1494         .rx_queue_stop                  = sfc_rx_queue_stop,
1495         .tx_queue_start                 = sfc_tx_queue_start,
1496         .tx_queue_stop                  = sfc_tx_queue_stop,
1497         .rx_queue_setup                 = sfc_rx_queue_setup,
1498         .rx_queue_release               = sfc_rx_queue_release,
1499         .rx_queue_count                 = sfc_rx_queue_count,
1500         .rx_descriptor_done             = sfc_rx_descriptor_done,
1501         .rx_descriptor_status           = sfc_rx_descriptor_status,
1502         .tx_descriptor_status           = sfc_tx_descriptor_status,
1503         .tx_queue_setup                 = sfc_tx_queue_setup,
1504         .tx_queue_release               = sfc_tx_queue_release,
1505         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1506         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1507         .mac_addr_set                   = sfc_mac_addr_set,
1508 #if EFSYS_OPT_RX_SCALE
1509         .reta_update                    = sfc_dev_rss_reta_update,
1510         .reta_query                     = sfc_dev_rss_reta_query,
1511         .rss_hash_update                = sfc_dev_rss_hash_update,
1512         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1513 #endif
1514         .filter_ctrl                    = sfc_dev_filter_ctrl,
1515         .set_mc_addr_list               = sfc_set_mc_addr_list,
1516         .rxq_info_get                   = sfc_rx_queue_info_get,
1517         .txq_info_get                   = sfc_tx_queue_info_get,
1518         .fw_version_get                 = sfc_fw_version_get,
1519         .xstats_get_by_id               = sfc_xstats_get_by_id,
1520         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1521 };
1522
1523 /**
1524  * Duplicate a string in potentially shared memory required for
1525  * multi-process support.
1526  *
1527  * strdup() allocates from process-local heap/memory.
1528  */
1529 static char *
1530 sfc_strdup(const char *str)
1531 {
1532         size_t size;
1533         char *copy;
1534
1535         if (str == NULL)
1536                 return NULL;
1537
1538         size = strlen(str) + 1;
1539         copy = rte_malloc(__func__, size, 0);
1540         if (copy != NULL)
1541                 rte_memcpy(copy, str, size);
1542
1543         return copy;
1544 }
1545
1546 static int
1547 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1548 {
1549         struct sfc_adapter *sa = dev->data->dev_private;
1550         unsigned int avail_caps = 0;
1551         const char *rx_name = NULL;
1552         const char *tx_name = NULL;
1553         int rc;
1554
1555         switch (sa->family) {
1556         case EFX_FAMILY_HUNTINGTON:
1557         case EFX_FAMILY_MEDFORD:
1558                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1559                 break;
1560         default:
1561                 break;
1562         }
1563
1564         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1565                                 sfc_kvarg_string_handler, &rx_name);
1566         if (rc != 0)
1567                 goto fail_kvarg_rx_datapath;
1568
1569         if (rx_name != NULL) {
1570                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1571                 if (sa->dp_rx == NULL) {
1572                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1573                         rc = ENOENT;
1574                         goto fail_dp_rx;
1575                 }
1576                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1577                         sfc_err(sa,
1578                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1579                                 rx_name);
1580                         rc = EINVAL;
1581                         goto fail_dp_rx_caps;
1582                 }
1583         } else {
1584                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1585                 if (sa->dp_rx == NULL) {
1586                         sfc_err(sa, "Rx datapath by caps %#x not found",
1587                                 avail_caps);
1588                         rc = ENOENT;
1589                         goto fail_dp_rx;
1590                 }
1591         }
1592
1593         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1594         if (sa->dp_rx_name == NULL) {
1595                 rc = ENOMEM;
1596                 goto fail_dp_rx_name;
1597         }
1598
1599         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1600
1601         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1602
1603         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1604                                 sfc_kvarg_string_handler, &tx_name);
1605         if (rc != 0)
1606                 goto fail_kvarg_tx_datapath;
1607
1608         if (tx_name != NULL) {
1609                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1610                 if (sa->dp_tx == NULL) {
1611                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1612                         rc = ENOENT;
1613                         goto fail_dp_tx;
1614                 }
1615                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1616                         sfc_err(sa,
1617                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1618                                 tx_name);
1619                         rc = EINVAL;
1620                         goto fail_dp_tx_caps;
1621                 }
1622         } else {
1623                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1624                 if (sa->dp_tx == NULL) {
1625                         sfc_err(sa, "Tx datapath by caps %#x not found",
1626                                 avail_caps);
1627                         rc = ENOENT;
1628                         goto fail_dp_tx;
1629                 }
1630         }
1631
1632         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1633         if (sa->dp_tx_name == NULL) {
1634                 rc = ENOMEM;
1635                 goto fail_dp_tx_name;
1636         }
1637
1638         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1639
1640         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1641
1642         dev->dev_ops = &sfc_eth_dev_ops;
1643
1644         return 0;
1645
1646 fail_dp_tx_name:
1647 fail_dp_tx_caps:
1648         sa->dp_tx = NULL;
1649
1650 fail_dp_tx:
1651 fail_kvarg_tx_datapath:
1652         rte_free(sa->dp_rx_name);
1653         sa->dp_rx_name = NULL;
1654
1655 fail_dp_rx_name:
1656 fail_dp_rx_caps:
1657         sa->dp_rx = NULL;
1658
1659 fail_dp_rx:
1660 fail_kvarg_rx_datapath:
1661         return rc;
1662 }
1663
1664 static void
1665 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1666 {
1667         struct sfc_adapter *sa = dev->data->dev_private;
1668
1669         dev->dev_ops = NULL;
1670         dev->rx_pkt_burst = NULL;
1671         dev->tx_pkt_burst = NULL;
1672
1673         rte_free(sa->dp_tx_name);
1674         sa->dp_tx_name = NULL;
1675         sa->dp_tx = NULL;
1676
1677         rte_free(sa->dp_rx_name);
1678         sa->dp_rx_name = NULL;
1679         sa->dp_rx = NULL;
1680 }
1681
1682 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1683         .rxq_info_get                   = sfc_rx_queue_info_get,
1684         .txq_info_get                   = sfc_tx_queue_info_get,
1685 };
1686
1687 static int
1688 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1689 {
1690         /*
1691          * Device private data has really many process-local pointers.
1692          * Below code should be extremely careful to use data located
1693          * in shared memory only.
1694          */
1695         struct sfc_adapter *sa = dev->data->dev_private;
1696         const struct sfc_dp_rx *dp_rx;
1697         const struct sfc_dp_tx *dp_tx;
1698         int rc;
1699
1700         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1701         if (dp_rx == NULL) {
1702                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1703                 rc = ENOENT;
1704                 goto fail_dp_rx;
1705         }
1706         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1707                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1708                         sa->dp_tx_name);
1709                 rc = EINVAL;
1710                 goto fail_dp_rx_multi_process;
1711         }
1712
1713         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1714         if (dp_tx == NULL) {
1715                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1716                 rc = ENOENT;
1717                 goto fail_dp_tx;
1718         }
1719         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1720                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1721                         sa->dp_tx_name);
1722                 rc = EINVAL;
1723                 goto fail_dp_tx_multi_process;
1724         }
1725
1726         dev->rx_pkt_burst = dp_rx->pkt_burst;
1727         dev->tx_pkt_burst = dp_tx->pkt_burst;
1728         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1729
1730         return 0;
1731
1732 fail_dp_tx_multi_process:
1733 fail_dp_tx:
1734 fail_dp_rx_multi_process:
1735 fail_dp_rx:
1736         return rc;
1737 }
1738
1739 static void
1740 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1741 {
1742         dev->dev_ops = NULL;
1743         dev->tx_pkt_burst = NULL;
1744         dev->rx_pkt_burst = NULL;
1745 }
1746
1747 static void
1748 sfc_register_dp(void)
1749 {
1750         /* Register once */
1751         if (TAILQ_EMPTY(&sfc_dp_head)) {
1752                 /* Prefer EF10 datapath */
1753                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1754                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1755
1756                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1757                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1758                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1759         }
1760 }
1761
1762 static int
1763 sfc_eth_dev_init(struct rte_eth_dev *dev)
1764 {
1765         struct sfc_adapter *sa = dev->data->dev_private;
1766         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1767         int rc;
1768         const efx_nic_cfg_t *encp;
1769         const struct ether_addr *from;
1770
1771         sfc_register_dp();
1772
1773         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1774                 return -sfc_eth_dev_secondary_set_ops(dev);
1775
1776         /* Required for logging */
1777         sa->pci_addr = pci_dev->addr;
1778         sa->port_id = dev->data->port_id;
1779
1780         sa->eth_dev = dev;
1781
1782         /* Copy PCI device info to the dev->data */
1783         rte_eth_copy_pci_info(dev, pci_dev);
1784
1785         dev->data->dev_flags |= RTE_ETH_DEV_DETACHABLE;
1786
1787         rc = sfc_kvargs_parse(sa);
1788         if (rc != 0)
1789                 goto fail_kvargs_parse;
1790
1791         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1792                                 sfc_kvarg_bool_handler, &sa->debug_init);
1793         if (rc != 0)
1794                 goto fail_kvarg_debug_init;
1795
1796         sfc_log_init(sa, "entry");
1797
1798         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1799         if (dev->data->mac_addrs == NULL) {
1800                 rc = ENOMEM;
1801                 goto fail_mac_addrs;
1802         }
1803
1804         sfc_adapter_lock_init(sa);
1805         sfc_adapter_lock(sa);
1806
1807         sfc_log_init(sa, "probing");
1808         rc = sfc_probe(sa);
1809         if (rc != 0)
1810                 goto fail_probe;
1811
1812         sfc_log_init(sa, "set device ops");
1813         rc = sfc_eth_dev_set_ops(dev);
1814         if (rc != 0)
1815                 goto fail_set_ops;
1816
1817         sfc_log_init(sa, "attaching");
1818         rc = sfc_attach(sa);
1819         if (rc != 0)
1820                 goto fail_attach;
1821
1822         encp = efx_nic_cfg_get(sa->nic);
1823
1824         /*
1825          * The arguments are really reverse order in comparison to
1826          * Linux kernel. Copy from NIC config to Ethernet device data.
1827          */
1828         from = (const struct ether_addr *)(encp->enc_mac_addr);
1829         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1830
1831         sfc_adapter_unlock(sa);
1832
1833         sfc_log_init(sa, "done");
1834         return 0;
1835
1836 fail_attach:
1837         sfc_eth_dev_clear_ops(dev);
1838
1839 fail_set_ops:
1840         sfc_unprobe(sa);
1841
1842 fail_probe:
1843         sfc_adapter_unlock(sa);
1844         sfc_adapter_lock_fini(sa);
1845         rte_free(dev->data->mac_addrs);
1846         dev->data->mac_addrs = NULL;
1847
1848 fail_mac_addrs:
1849 fail_kvarg_debug_init:
1850         sfc_kvargs_cleanup(sa);
1851
1852 fail_kvargs_parse:
1853         sfc_log_init(sa, "failed %d", rc);
1854         SFC_ASSERT(rc > 0);
1855         return -rc;
1856 }
1857
1858 static int
1859 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1860 {
1861         struct sfc_adapter *sa;
1862
1863         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1864                 sfc_eth_dev_secondary_clear_ops(dev);
1865                 return 0;
1866         }
1867
1868         sa = dev->data->dev_private;
1869         sfc_log_init(sa, "entry");
1870
1871         sfc_adapter_lock(sa);
1872
1873         sfc_eth_dev_clear_ops(dev);
1874
1875         sfc_detach(sa);
1876         sfc_unprobe(sa);
1877
1878         rte_free(dev->data->mac_addrs);
1879         dev->data->mac_addrs = NULL;
1880
1881         sfc_kvargs_cleanup(sa);
1882
1883         sfc_adapter_unlock(sa);
1884         sfc_adapter_lock_fini(sa);
1885
1886         sfc_log_init(sa, "done");
1887
1888         /* Required for logging, so cleanup last */
1889         sa->eth_dev = NULL;
1890         return 0;
1891 }
1892
1893 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1894         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1895         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1896         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1897         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1898         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1899         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1900         { .vendor_id = 0 /* sentinel */ }
1901 };
1902
1903 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1904         struct rte_pci_device *pci_dev)
1905 {
1906         return rte_eth_dev_pci_generic_probe(pci_dev,
1907                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
1908 }
1909
1910 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
1911 {
1912         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
1913 }
1914
1915 static struct rte_pci_driver sfc_efx_pmd = {
1916         .id_table = pci_id_sfc_efx_map,
1917         .drv_flags =
1918                 RTE_PCI_DRV_INTR_LSC |
1919                 RTE_PCI_DRV_NEED_MAPPING,
1920         .probe = sfc_eth_dev_pci_probe,
1921         .remove = sfc_eth_dev_pci_remove,
1922 };
1923
1924 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
1925 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1926 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
1927 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1928         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1929         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1930         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1931         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1932         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1933         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);