drivers/net/sfc/sfc_ethdev.c

   1 /*-
   2  * Copyright (c) 2016 Solarflare Communications Inc.
   3  * All rights reserved.
   4  *
   5  * This software was jointly developed between OKTET Labs (under contract
   6  * for Solarflare) and Solarflare Communications, Inc.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright notice,
  12  *    this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright notice,
  14  *    this list of conditions and the following disclaimer in the documentation
  15  *    and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  18  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
  19  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  20  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  21  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  22  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  23  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
  24  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  25  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  26  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
  27  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  28  */
  29
  30 #include <rte_dev.h>
  31 #include <rte_ethdev.h>
  32 #include <rte_pci.h>
  33
  34 #include "efx.h"
  35
  36 #include "sfc.h"
  37 #include "sfc_debug.h"
  38 #include "sfc_log.h"
  39 #include "sfc_kvargs.h"
  40 #include "sfc_ev.h"
  41 #include "sfc_rx.h"
  42 #include "sfc_tx.h"
  43
  44
  45 static void
  46 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  47 {
  48         struct sfc_adapter *sa = dev->data->dev_private;
  49         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
  50
  51         sfc_log_init(sa, "entry");
  52
  53         dev_info->pci_dev = RTE_DEV_TO_PCI(dev->device);
  54         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  55
  56         /* Autonegotiation may be disabled */
  57         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  58         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  59                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  60         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
  61                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
  62         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
  63                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
  64
  65         dev_info->max_rx_queues = sa->rxq_max;
  66         dev_info->max_tx_queues = sa->txq_max;
  67
  68         /* By default packets are dropped if no descriptors are available */
  69         dev_info->default_rxconf.rx_drop_en = 1;
  70
  71         dev_info->rx_offload_capa =
  72                 DEV_RX_OFFLOAD_IPV4_CKSUM |
  73                 DEV_RX_OFFLOAD_UDP_CKSUM |
  74                 DEV_RX_OFFLOAD_TCP_CKSUM;
  75
  76         dev_info->tx_offload_capa =
  77                 DEV_TX_OFFLOAD_IPV4_CKSUM |
  78                 DEV_TX_OFFLOAD_UDP_CKSUM |
  79                 DEV_TX_OFFLOAD_TCP_CKSUM;
  80
  81         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
  82         if (!encp->enc_hw_tx_insert_vlan_enabled)
  83                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
  84         else
  85                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
  86
  87 #if EFSYS_OPT_RX_SCALE
  88         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
  89                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
  90                 dev_info->hash_key_size = SFC_RSS_KEY_SIZE;
  91                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
  92         }
  93 #endif
  94
  95         if (sa->tso)
  96                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
  97
  98         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
  99         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 100         /* The RXQ hardware requires that the descriptor count is a power
 101          * of 2, but rx_desc_lim cannot properly describe that constraint.
 102          */
 103         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 104
 105         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 106         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 107         /*
 108          * The TXQ hardware requires that the descriptor count is a power
 109          * of 2, but tx_desc_lim cannot properly describe that constraint
 110          */
 111         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 112 }
 113
 114 static const uint32_t *
 115 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 116 {
 117         static const uint32_t ptypes[] = {
 118                 RTE_PTYPE_L2_ETHER,
 119                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 120                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 121                 RTE_PTYPE_L4_TCP,
 122                 RTE_PTYPE_L4_UDP,
 123                 RTE_PTYPE_UNKNOWN
 124         };
 125
 126         if (dev->rx_pkt_burst == sfc_recv_pkts)
 127                 return ptypes;
 128
 129         return NULL;
 130 }
 131
 132 static int
 133 sfc_dev_configure(struct rte_eth_dev *dev)
 134 {
 135         struct rte_eth_dev_data *dev_data = dev->data;
 136         struct sfc_adapter *sa = dev_data->dev_private;
 137         int rc;
 138
 139         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 140                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 141
 142         sfc_adapter_lock(sa);
 143         switch (sa->state) {
 144         case SFC_ADAPTER_CONFIGURED:
 145                 sfc_close(sa);
 146                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 147                 /* FALLTHROUGH */
 148         case SFC_ADAPTER_INITIALIZED:
 149                 rc = sfc_configure(sa);
 150                 break;
 151         default:
 152                 sfc_err(sa, "unexpected adapter state %u to configure",
 153                         sa->state);
 154                 rc = EINVAL;
 155                 break;
 156         }
 157         sfc_adapter_unlock(sa);
 158
 159         sfc_log_init(sa, "done %d", rc);
 160         SFC_ASSERT(rc >= 0);
 161         return -rc;
 162 }
 163
 164 static int
 165 sfc_dev_start(struct rte_eth_dev *dev)
 166 {
 167         struct sfc_adapter *sa = dev->data->dev_private;
 168         int rc;
 169
 170         sfc_log_init(sa, "entry");
 171
 172         sfc_adapter_lock(sa);
 173         rc = sfc_start(sa);
 174         sfc_adapter_unlock(sa);
 175
 176         sfc_log_init(sa, "done %d", rc);
 177         SFC_ASSERT(rc >= 0);
 178         return -rc;
 179 }
 180
 181 static int
 182 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 183 {
 184         struct sfc_adapter *sa = dev->data->dev_private;
 185         struct rte_eth_link *dev_link = &dev->data->dev_link;
 186         struct rte_eth_link old_link;
 187         struct rte_eth_link current_link;
 188
 189         sfc_log_init(sa, "entry");
 190
 191 retry:
 192         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 193         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 194
 195         if (sa->state != SFC_ADAPTER_STARTED) {
 196                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 197                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 198                                          *(uint64_t *)&old_link,
 199                                          *(uint64_t *)&current_link))
 200                         goto retry;
 201         } else if (wait_to_complete) {
 202                 efx_link_mode_t link_mode;
 203
 204                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 205                         link_mode = EFX_LINK_UNKNOWN;
 206                 sfc_port_link_mode_to_info(link_mode, &current_link);
 207
 208                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 209                                          *(uint64_t *)&old_link,
 210                                          *(uint64_t *)&current_link))
 211                         goto retry;
 212         } else {
 213                 sfc_ev_mgmt_qpoll(sa);
 214                 *(int64_t *)&current_link =
 215                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 216         }
 217
 218         if (old_link.link_status != current_link.link_status)
 219                 sfc_info(sa, "Link status is %s",
 220                          current_link.link_status ? "UP" : "DOWN");
 221
 222         return old_link.link_status == current_link.link_status ? 0 : -1;
 223 }
 224
 225 static void
 226 sfc_dev_stop(struct rte_eth_dev *dev)
 227 {
 228         struct sfc_adapter *sa = dev->data->dev_private;
 229
 230         sfc_log_init(sa, "entry");
 231
 232         sfc_adapter_lock(sa);
 233         sfc_stop(sa);
 234         sfc_adapter_unlock(sa);
 235
 236         sfc_log_init(sa, "done");
 237 }
 238
 239 static int
 240 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 241 {
 242         struct sfc_adapter *sa = dev->data->dev_private;
 243         int rc;
 244
 245         sfc_log_init(sa, "entry");
 246
 247         sfc_adapter_lock(sa);
 248         rc = sfc_start(sa);
 249         sfc_adapter_unlock(sa);
 250
 251         SFC_ASSERT(rc >= 0);
 252         return -rc;
 253 }
 254
 255 static int
 256 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 257 {
 258         struct sfc_adapter *sa = dev->data->dev_private;
 259
 260         sfc_log_init(sa, "entry");
 261
 262         sfc_adapter_lock(sa);
 263         sfc_stop(sa);
 264         sfc_adapter_unlock(sa);
 265
 266         return 0;
 267 }
 268
 269 static void
 270 sfc_dev_close(struct rte_eth_dev *dev)
 271 {
 272         struct sfc_adapter *sa = dev->data->dev_private;
 273
 274         sfc_log_init(sa, "entry");
 275
 276         sfc_adapter_lock(sa);
 277         switch (sa->state) {
 278         case SFC_ADAPTER_STARTED:
 279                 sfc_stop(sa);
 280                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 281                 /* FALLTHROUGH */
 282         case SFC_ADAPTER_CONFIGURED:
 283                 sfc_close(sa);
 284                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 285                 /* FALLTHROUGH */
 286         case SFC_ADAPTER_INITIALIZED:
 287                 break;
 288         default:
 289                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 290                 break;
 291         }
 292         sfc_adapter_unlock(sa);
 293
 294         sfc_log_init(sa, "done");
 295 }
 296
 297 static void
 298 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 299                    boolean_t enabled)
 300 {
 301         struct sfc_port *port;
 302         boolean_t *toggle;
 303         struct sfc_adapter *sa = dev->data->dev_private;
 304         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 305         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 306
 307         sfc_adapter_lock(sa);
 308
 309         port = &sa->port;
 310         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 311
 312         if (*toggle != enabled) {
 313                 *toggle = enabled;
 314
 315                 if ((sa->state == SFC_ADAPTER_STARTED) &&
 316                     (sfc_set_rx_mode(sa) != 0)) {
 317                         *toggle = !(enabled);
 318                         sfc_warn(sa, "Failed to %s %s mode",
 319                                  ((enabled) ? "enable" : "disable"), desc);
 320                 }
 321         }
 322
 323         sfc_adapter_unlock(sa);
 324 }
 325
 326 static void
 327 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 328 {
 329         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 330 }
 331
 332 static void
 333 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 334 {
 335         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 336 }
 337
 338 static void
 339 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 340 {
 341         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 342 }
 343
 344 static void
 345 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 346 {
 347         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 348 }
 349
 350 static int
 351 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 352                    uint16_t nb_rx_desc, unsigned int socket_id,
 353                    const struct rte_eth_rxconf *rx_conf,
 354                    struct rte_mempool *mb_pool)
 355 {
 356         struct sfc_adapter *sa = dev->data->dev_private;
 357         int rc;
 358
 359         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 360                      rx_queue_id, nb_rx_desc, socket_id);
 361
 362         sfc_adapter_lock(sa);
 363
 364         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 365                           rx_conf, mb_pool);
 366         if (rc != 0)
 367                 goto fail_rx_qinit;
 368
 369         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq;
 370
 371         sfc_adapter_unlock(sa);
 372
 373         return 0;
 374
 375 fail_rx_qinit:
 376         sfc_adapter_unlock(sa);
 377         SFC_ASSERT(rc > 0);
 378         return -rc;
 379 }
 380
 381 static void
 382 sfc_rx_queue_release(void *queue)
 383 {
 384         struct sfc_rxq *rxq = queue;
 385         struct sfc_adapter *sa;
 386         unsigned int sw_index;
 387
 388         if (rxq == NULL)
 389                 return;
 390
 391         sa = rxq->evq->sa;
 392         sfc_adapter_lock(sa);
 393
 394         sw_index = sfc_rxq_sw_index(rxq);
 395
 396         sfc_log_init(sa, "RxQ=%u", sw_index);
 397
 398         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 399
 400         sfc_rx_qfini(sa, sw_index);
 401
 402         sfc_adapter_unlock(sa);
 403 }
 404
 405 static int
 406 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 407                    uint16_t nb_tx_desc, unsigned int socket_id,
 408                    const struct rte_eth_txconf *tx_conf)
 409 {
 410         struct sfc_adapter *sa = dev->data->dev_private;
 411         int rc;
 412
 413         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 414                      tx_queue_id, nb_tx_desc, socket_id);
 415
 416         sfc_adapter_lock(sa);
 417
 418         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 419         if (rc != 0)
 420                 goto fail_tx_qinit;
 421
 422         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq;
 423
 424         sfc_adapter_unlock(sa);
 425         return 0;
 426
 427 fail_tx_qinit:
 428         sfc_adapter_unlock(sa);
 429         SFC_ASSERT(rc > 0);
 430         return -rc;
 431 }
 432
 433 static void
 434 sfc_tx_queue_release(void *queue)
 435 {
 436         struct sfc_txq *txq = queue;
 437         unsigned int sw_index;
 438         struct sfc_adapter *sa;
 439
 440         if (txq == NULL)
 441                 return;
 442
 443         sw_index = sfc_txq_sw_index(txq);
 444
 445         SFC_ASSERT(txq->evq != NULL);
 446         sa = txq->evq->sa;
 447
 448         sfc_log_init(sa, "TxQ = %u", sw_index);
 449
 450         sfc_adapter_lock(sa);
 451
 452         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 453         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 454
 455         sfc_tx_qfini(sa, sw_index);
 456
 457         sfc_adapter_unlock(sa);
 458 }
 459
 460 static void
 461 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 462 {
 463         struct sfc_adapter *sa = dev->data->dev_private;
 464         struct sfc_port *port = &sa->port;
 465         uint64_t *mac_stats;
 466
 467         rte_spinlock_lock(&port->mac_stats_lock);
 468
 469         if (sfc_port_update_mac_stats(sa) != 0)
 470                 goto unlock;
 471
 472         mac_stats = port->mac_stats_buf;
 473
 474         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 475                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 476                 stats->ipackets =
 477                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 478                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 479                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 480                 stats->opackets =
 481                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 482                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 483                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 484                 stats->ibytes =
 485                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 486                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 487                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 488                 stats->obytes =
 489                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 490                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 491                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 492                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 493                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 494                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 495         } else {
 496                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 497                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 498                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 499                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 500                 /*
 501                  * Take into account stats which are whenever supported
 502                  * on EF10. If some stat is not supported by current
 503                  * firmware variant or HW revision, it is guaranteed
 504                  * to be zero in mac_stats.
 505                  */
 506                 stats->imissed =
 507                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 508                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 509                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 510                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 511                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 512                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 513                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 514                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 515                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 516                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 517                 stats->ierrors =
 518                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 519                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 520                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 521                 /* no oerrors counters supported on EF10 */
 522         }
 523
 524 unlock:
 525         rte_spinlock_unlock(&port->mac_stats_lock);
 526 }
 527
 528 static int
 529 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 530                unsigned int xstats_count)
 531 {
 532         struct sfc_adapter *sa = dev->data->dev_private;
 533         struct sfc_port *port = &sa->port;
 534         uint64_t *mac_stats;
 535         int rc;
 536         unsigned int i;
 537         int nstats = 0;
 538
 539         rte_spinlock_lock(&port->mac_stats_lock);
 540
 541         rc = sfc_port_update_mac_stats(sa);
 542         if (rc != 0) {
 543                 SFC_ASSERT(rc > 0);
 544                 nstats = -rc;
 545                 goto unlock;
 546         }
 547
 548         mac_stats = port->mac_stats_buf;
 549
 550         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 551                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 552                         if (xstats != NULL && nstats < (int)xstats_count) {
 553                                 xstats[nstats].id = nstats;
 554                                 xstats[nstats].value = mac_stats[i];
 555                         }
 556                         nstats++;
 557                 }
 558         }
 559
 560 unlock:
 561         rte_spinlock_unlock(&port->mac_stats_lock);
 562
 563         return nstats;
 564 }
 565
 566 static int
 567 sfc_xstats_get_names(struct rte_eth_dev *dev,
 568                      struct rte_eth_xstat_name *xstats_names,
 569                      unsigned int xstats_count)
 570 {
 571         struct sfc_adapter *sa = dev->data->dev_private;
 572         struct sfc_port *port = &sa->port;
 573         unsigned int i;
 574         unsigned int nstats = 0;
 575
 576         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 577                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 578                         if (xstats_names != NULL && nstats < xstats_count)
 579                                 strncpy(xstats_names[nstats].name,
 580                                         efx_mac_stat_name(sa->nic, i),
 581                                         sizeof(xstats_names[0].name));
 582                         nstats++;
 583                 }
 584         }
 585
 586         return nstats;
 587 }
 588
 589 static int
 590 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 591 {
 592         struct sfc_adapter *sa = dev->data->dev_private;
 593         unsigned int wanted_fc, link_fc;
 594
 595         memset(fc_conf, 0, sizeof(*fc_conf));
 596
 597         sfc_adapter_lock(sa);
 598
 599         if (sa->state == SFC_ADAPTER_STARTED)
 600                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 601         else
 602                 link_fc = sa->port.flow_ctrl;
 603
 604         switch (link_fc) {
 605         case 0:
 606                 fc_conf->mode = RTE_FC_NONE;
 607                 break;
 608         case EFX_FCNTL_RESPOND:
 609                 fc_conf->mode = RTE_FC_RX_PAUSE;
 610                 break;
 611         case EFX_FCNTL_GENERATE:
 612                 fc_conf->mode = RTE_FC_TX_PAUSE;
 613                 break;
 614         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 615                 fc_conf->mode = RTE_FC_FULL;
 616                 break;
 617         default:
 618                 sfc_err(sa, "%s: unexpected flow control value %#x",
 619                         __func__, link_fc);
 620         }
 621
 622         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 623
 624         sfc_adapter_unlock(sa);
 625
 626         return 0;
 627 }
 628
 629 static int
 630 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 631 {
 632         struct sfc_adapter *sa = dev->data->dev_private;
 633         struct sfc_port *port = &sa->port;
 634         unsigned int fcntl;
 635         int rc;
 636
 637         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 638             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 639             fc_conf->mac_ctrl_frame_fwd != 0) {
 640                 sfc_err(sa, "unsupported flow control settings specified");
 641                 rc = EINVAL;
 642                 goto fail_inval;
 643         }
 644
 645         switch (fc_conf->mode) {
 646         case RTE_FC_NONE:
 647                 fcntl = 0;
 648                 break;
 649         case RTE_FC_RX_PAUSE:
 650                 fcntl = EFX_FCNTL_RESPOND;
 651                 break;
 652         case RTE_FC_TX_PAUSE:
 653                 fcntl = EFX_FCNTL_GENERATE;
 654                 break;
 655         case RTE_FC_FULL:
 656                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 657                 break;
 658         default:
 659                 rc = EINVAL;
 660                 goto fail_inval;
 661         }
 662
 663         sfc_adapter_lock(sa);
 664
 665         if (sa->state == SFC_ADAPTER_STARTED) {
 666                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 667                 if (rc != 0)
 668                         goto fail_mac_fcntl_set;
 669         }
 670
 671         port->flow_ctrl = fcntl;
 672         port->flow_ctrl_autoneg = fc_conf->autoneg;
 673
 674         sfc_adapter_unlock(sa);
 675
 676         return 0;
 677
 678 fail_mac_fcntl_set:
 679         sfc_adapter_unlock(sa);
 680 fail_inval:
 681         SFC_ASSERT(rc > 0);
 682         return -rc;
 683 }
 684
 685 static int
 686 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 687 {
 688         struct sfc_adapter *sa = dev->data->dev_private;
 689         size_t pdu = EFX_MAC_PDU(mtu);
 690         size_t old_pdu;
 691         int rc;
 692
 693         sfc_log_init(sa, "mtu=%u", mtu);
 694
 695         rc = EINVAL;
 696         if (pdu < EFX_MAC_PDU_MIN) {
 697                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 698                         (unsigned int)mtu, (unsigned int)pdu,
 699                         EFX_MAC_PDU_MIN);
 700                 goto fail_inval;
 701         }
 702         if (pdu > EFX_MAC_PDU_MAX) {
 703                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 704                         (unsigned int)mtu, (unsigned int)pdu,
 705                         EFX_MAC_PDU_MAX);
 706                 goto fail_inval;
 707         }
 708
 709         sfc_adapter_lock(sa);
 710
 711         if (pdu != sa->port.pdu) {
 712                 if (sa->state == SFC_ADAPTER_STARTED) {
 713                         sfc_stop(sa);
 714
 715                         old_pdu = sa->port.pdu;
 716                         sa->port.pdu = pdu;
 717                         rc = sfc_start(sa);
 718                         if (rc != 0)
 719                                 goto fail_start;
 720                 } else {
 721                         sa->port.pdu = pdu;
 722                 }
 723         }
 724
 725         /*
 726          * The driver does not use it, but other PMDs update jumbo_frame
 727          * flag and max_rx_pkt_len when MTU is set.
 728          */
 729         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 730         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 731
 732         sfc_adapter_unlock(sa);
 733
 734         sfc_log_init(sa, "done");
 735         return 0;
 736
 737 fail_start:
 738         sa->port.pdu = old_pdu;
 739         if (sfc_start(sa) != 0)
 740                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 741                         "PDU max size - port is stopped",
 742                         (unsigned int)pdu, (unsigned int)old_pdu);
 743         sfc_adapter_unlock(sa);
 744
 745 fail_inval:
 746         sfc_log_init(sa, "failed %d", rc);
 747         SFC_ASSERT(rc > 0);
 748         return -rc;
 749 }
 750 static void
 751 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 752 {
 753         struct sfc_adapter *sa = dev->data->dev_private;
 754         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 755         int rc;
 756
 757         sfc_adapter_lock(sa);
 758
 759         if (sa->state != SFC_ADAPTER_STARTED) {
 760                 sfc_info(sa, "the port is not started");
 761                 sfc_info(sa, "the new MAC address will be set on port start");
 762
 763                 goto unlock;
 764         }
 765
 766         if (encp->enc_allow_set_mac_with_installed_filters) {
 767                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 768                 if (rc != 0) {
 769                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 770                         goto unlock;
 771                 }
 772
 773                 /*
 774                  * Changing the MAC address by means of MCDI request
 775                  * has no effect on received traffic, therefore
 776                  * we also need to update unicast filters
 777                  */
 778                 rc = sfc_set_rx_mode(sa);
 779                 if (rc != 0)
 780                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 781         } else {
 782                 sfc_warn(sa, "cannot set MAC address with filters installed");
 783                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 784                 sfc_warn(sa, "(some traffic may be dropped)");
 785
 786                 /*
 787                  * Since setting MAC address with filters installed is not
 788                  * allowed on the adapter, one needs to simply restart adapter
 789                  * so that the new MAC address will be taken from an outer
 790                  * storage and set flawlessly by means of sfc_start() call
 791                  */
 792                 sfc_stop(sa);
 793                 rc = sfc_start(sa);
 794                 if (rc != 0)
 795                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 796         }
 797
 798 unlock:
 799         sfc_adapter_unlock(sa);
 800 }
 801
 802
 803 static int
 804 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 805                      uint32_t nb_mc_addr)
 806 {
 807         struct sfc_adapter *sa = dev->data->dev_private;
 808         uint8_t *mc_addrs_p;
 809         uint8_t *mc_addrs;
 810         int rc;
 811         unsigned int i;
 812
 813         if (nb_mc_addr > EFX_MAC_MULTICAST_LIST_MAX) {
 814                 sfc_err(sa, "too many multicast addresses: %u > %u",
 815                          nb_mc_addr, EFX_MAC_MULTICAST_LIST_MAX);
 816                 return -EINVAL;
 817         }
 818
 819         mc_addrs_p = rte_calloc("mc-addrs", nb_mc_addr, EFX_MAC_ADDR_LEN, 0);
 820         if (mc_addrs_p == NULL)
 821                 return -ENOMEM;
 822
 823         mc_addrs = mc_addrs_p;
 824
 825         for (i = 0; i < nb_mc_addr; ++i) {
 826                 (void)rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
 827                                  EFX_MAC_ADDR_LEN);
 828                 mc_addrs += EFX_MAC_ADDR_LEN;
 829         }
 830
 831         rc = efx_mac_multicast_list_set(sa->nic, mc_addrs_p, nb_mc_addr);
 832
 833         rte_free(mc_addrs_p);
 834
 835         if (rc != 0)
 836                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
 837
 838         SFC_ASSERT(rc > 0);
 839         return -rc;
 840 }
 841
 842 static void
 843 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 844                       struct rte_eth_rxq_info *qinfo)
 845 {
 846         struct sfc_adapter *sa = dev->data->dev_private;
 847         struct sfc_rxq_info *rxq_info;
 848         struct sfc_rxq *rxq;
 849
 850         sfc_adapter_lock(sa);
 851
 852         SFC_ASSERT(rx_queue_id < sa->rxq_count);
 853
 854         rxq_info = &sa->rxq_info[rx_queue_id];
 855         rxq = rxq_info->rxq;
 856         SFC_ASSERT(rxq != NULL);
 857
 858         qinfo->mp = rxq->refill_mb_pool;
 859         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
 860         qinfo->conf.rx_drop_en = 1;
 861         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
 862         qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
 863         qinfo->nb_desc = rxq_info->entries;
 864
 865         sfc_adapter_unlock(sa);
 866 }
 867
 868 static void
 869 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 870                       struct rte_eth_txq_info *qinfo)
 871 {
 872         struct sfc_adapter *sa = dev->data->dev_private;
 873         struct sfc_txq_info *txq_info;
 874
 875         sfc_adapter_lock(sa);
 876
 877         SFC_ASSERT(tx_queue_id < sa->txq_count);
 878
 879         txq_info = &sa->txq_info[tx_queue_id];
 880         SFC_ASSERT(txq_info->txq != NULL);
 881
 882         memset(qinfo, 0, sizeof(*qinfo));
 883
 884         qinfo->conf.txq_flags = txq_info->txq->flags;
 885         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
 886         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
 887         qinfo->nb_desc = txq_info->entries;
 888
 889         sfc_adapter_unlock(sa);
 890 }
 891
 892 static uint32_t
 893 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 894 {
 895         struct sfc_adapter *sa = dev->data->dev_private;
 896
 897         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 898
 899         return sfc_rx_qdesc_npending(sa, rx_queue_id);
 900 }
 901
 902 static int
 903 sfc_rx_descriptor_done(void *queue, uint16_t offset)
 904 {
 905         struct sfc_rxq *rxq = queue;
 906
 907         return sfc_rx_qdesc_done(rxq, offset);
 908 }
 909
 910 static int
 911 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 912 {
 913         struct sfc_adapter *sa = dev->data->dev_private;
 914         int rc;
 915
 916         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 917
 918         sfc_adapter_lock(sa);
 919
 920         rc = EINVAL;
 921         if (sa->state != SFC_ADAPTER_STARTED)
 922                 goto fail_not_started;
 923
 924         rc = sfc_rx_qstart(sa, rx_queue_id);
 925         if (rc != 0)
 926                 goto fail_rx_qstart;
 927
 928         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
 929
 930         sfc_adapter_unlock(sa);
 931
 932         return 0;
 933
 934 fail_rx_qstart:
 935 fail_not_started:
 936         sfc_adapter_unlock(sa);
 937         SFC_ASSERT(rc > 0);
 938         return -rc;
 939 }
 940
 941 static int
 942 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 943 {
 944         struct sfc_adapter *sa = dev->data->dev_private;
 945
 946         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 947
 948         sfc_adapter_lock(sa);
 949         sfc_rx_qstop(sa, rx_queue_id);
 950
 951         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
 952
 953         sfc_adapter_unlock(sa);
 954
 955         return 0;
 956 }
 957
 958 static int
 959 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 960 {
 961         struct sfc_adapter *sa = dev->data->dev_private;
 962         int rc;
 963
 964         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
 965
 966         sfc_adapter_lock(sa);
 967
 968         rc = EINVAL;
 969         if (sa->state != SFC_ADAPTER_STARTED)
 970                 goto fail_not_started;
 971
 972         rc = sfc_tx_qstart(sa, tx_queue_id);
 973         if (rc != 0)
 974                 goto fail_tx_qstart;
 975
 976         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
 977
 978         sfc_adapter_unlock(sa);
 979         return 0;
 980
 981 fail_tx_qstart:
 982
 983 fail_not_started:
 984         sfc_adapter_unlock(sa);
 985         SFC_ASSERT(rc > 0);
 986         return -rc;
 987 }
 988
 989 static int
 990 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 991 {
 992         struct sfc_adapter *sa = dev->data->dev_private;
 993
 994         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
 995
 996         sfc_adapter_lock(sa);
 997
 998         sfc_tx_qstop(sa, tx_queue_id);
 999
1000         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1001
1002         sfc_adapter_unlock(sa);
1003         return 0;
1004 }
1005
1006 #if EFSYS_OPT_RX_SCALE
1007 static int
1008 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1009                           struct rte_eth_rss_conf *rss_conf)
1010 {
1011         struct sfc_adapter *sa = dev->data->dev_private;
1012
1013         if ((sa->rss_channels == 1) ||
1014             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1015                 return -ENOTSUP;
1016
1017         sfc_adapter_lock(sa);
1018
1019         /*
1020          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1021          * hence, conversion is done here to derive a correct set of ETH_RSS
1022          * flags which corresponds to the active EFX configuration stored
1023          * locally in 'sfc_adapter' and kept up-to-date
1024          */
1025         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1026         rss_conf->rss_key_len = SFC_RSS_KEY_SIZE;
1027         if (rss_conf->rss_key != NULL)
1028                 rte_memcpy(rss_conf->rss_key, sa->rss_key, SFC_RSS_KEY_SIZE);
1029
1030         sfc_adapter_unlock(sa);
1031
1032         return 0;
1033 }
1034
1035 static int
1036 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1037                         struct rte_eth_rss_conf *rss_conf)
1038 {
1039         struct sfc_adapter *sa = dev->data->dev_private;
1040         unsigned int efx_hash_types;
1041         int rc = 0;
1042
1043         if ((sa->rss_channels == 1) ||
1044             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1045                 sfc_err(sa, "RSS is not available");
1046                 return -ENOTSUP;
1047         }
1048
1049         if ((rss_conf->rss_key != NULL) &&
1050             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1051                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1052                         sizeof(sa->rss_key));
1053                 return -EINVAL;
1054         }
1055
1056         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1057                 sfc_err(sa, "unsupported hash functions requested");
1058                 return -EINVAL;
1059         }
1060
1061         sfc_adapter_lock(sa);
1062
1063         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1064
1065         rc = efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1066                                    efx_hash_types, B_TRUE);
1067         if (rc != 0)
1068                 goto fail_scale_mode_set;
1069
1070         if (rss_conf->rss_key != NULL) {
1071                 if (sa->state == SFC_ADAPTER_STARTED) {
1072                         rc = efx_rx_scale_key_set(sa->nic, rss_conf->rss_key,
1073                                                   sizeof(sa->rss_key));
1074                         if (rc != 0)
1075                                 goto fail_scale_key_set;
1076                 }
1077
1078                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1079         }
1080
1081         sa->rss_hash_types = efx_hash_types;
1082
1083         sfc_adapter_unlock(sa);
1084
1085         return 0;
1086
1087 fail_scale_key_set:
1088         if (efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1089                                   sa->rss_hash_types, B_TRUE) != 0)
1090                 sfc_err(sa, "failed to restore RSS mode");
1091
1092 fail_scale_mode_set:
1093         sfc_adapter_unlock(sa);
1094         return -rc;
1095 }
1096
1097 static int
1098 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1099                        struct rte_eth_rss_reta_entry64 *reta_conf,
1100                        uint16_t reta_size)
1101 {
1102         struct sfc_adapter *sa = dev->data->dev_private;
1103         int entry;
1104
1105         if ((sa->rss_channels == 1) ||
1106             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1107                 return -ENOTSUP;
1108
1109         if (reta_size != EFX_RSS_TBL_SIZE)
1110                 return -EINVAL;
1111
1112         sfc_adapter_lock(sa);
1113
1114         for (entry = 0; entry < reta_size; entry++) {
1115                 int grp = entry / RTE_RETA_GROUP_SIZE;
1116                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1117
1118                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1119                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1120         }
1121
1122         sfc_adapter_unlock(sa);
1123
1124         return 0;
1125 }
1126
1127 static int
1128 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1129                         struct rte_eth_rss_reta_entry64 *reta_conf,
1130                         uint16_t reta_size)
1131 {
1132         struct sfc_adapter *sa = dev->data->dev_private;
1133         unsigned int *rss_tbl_new;
1134         uint16_t entry;
1135         int rc;
1136
1137
1138         if ((sa->rss_channels == 1) ||
1139             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1140                 sfc_err(sa, "RSS is not available");
1141                 return -ENOTSUP;
1142         }
1143
1144         if (reta_size != EFX_RSS_TBL_SIZE) {
1145                 sfc_err(sa, "RETA size is wrong (should be %u)",
1146                         EFX_RSS_TBL_SIZE);
1147                 return -EINVAL;
1148         }
1149
1150         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1151         if (rss_tbl_new == NULL)
1152                 return -ENOMEM;
1153
1154         sfc_adapter_lock(sa);
1155
1156         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1157
1158         for (entry = 0; entry < reta_size; entry++) {
1159                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1160                 struct rte_eth_rss_reta_entry64 *grp;
1161
1162                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1163
1164                 if (grp->mask & (1ull << grp_idx)) {
1165                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1166                                 rc = EINVAL;
1167                                 goto bad_reta_entry;
1168                         }
1169                         rss_tbl_new[entry] = grp->reta[grp_idx];
1170                 }
1171         }
1172
1173         rc = efx_rx_scale_tbl_set(sa->nic, rss_tbl_new, EFX_RSS_TBL_SIZE);
1174         if (rc == 0)
1175                 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1176
1177 bad_reta_entry:
1178         sfc_adapter_unlock(sa);
1179
1180         rte_free(rss_tbl_new);
1181
1182         SFC_ASSERT(rc >= 0);
1183         return -rc;
1184 }
1185 #endif
1186
1187 static const struct eth_dev_ops sfc_eth_dev_ops = {
1188         .dev_configure                  = sfc_dev_configure,
1189         .dev_start                      = sfc_dev_start,
1190         .dev_stop                       = sfc_dev_stop,
1191         .dev_set_link_up                = sfc_dev_set_link_up,
1192         .dev_set_link_down              = sfc_dev_set_link_down,
1193         .dev_close                      = sfc_dev_close,
1194         .promiscuous_enable             = sfc_dev_promisc_enable,
1195         .promiscuous_disable            = sfc_dev_promisc_disable,
1196         .allmulticast_enable            = sfc_dev_allmulti_enable,
1197         .allmulticast_disable           = sfc_dev_allmulti_disable,
1198         .link_update                    = sfc_dev_link_update,
1199         .stats_get                      = sfc_stats_get,
1200         .xstats_get                     = sfc_xstats_get,
1201         .xstats_get_names               = sfc_xstats_get_names,
1202         .dev_infos_get                  = sfc_dev_infos_get,
1203         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1204         .mtu_set                        = sfc_dev_set_mtu,
1205         .rx_queue_start                 = sfc_rx_queue_start,
1206         .rx_queue_stop                  = sfc_rx_queue_stop,
1207         .tx_queue_start                 = sfc_tx_queue_start,
1208         .tx_queue_stop                  = sfc_tx_queue_stop,
1209         .rx_queue_setup                 = sfc_rx_queue_setup,
1210         .rx_queue_release               = sfc_rx_queue_release,
1211         .rx_queue_count                 = sfc_rx_queue_count,
1212         .rx_descriptor_done             = sfc_rx_descriptor_done,
1213         .tx_queue_setup                 = sfc_tx_queue_setup,
1214         .tx_queue_release               = sfc_tx_queue_release,
1215         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1216         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1217         .mac_addr_set                   = sfc_mac_addr_set,
1218 #if EFSYS_OPT_RX_SCALE
1219         .reta_update                    = sfc_dev_rss_reta_update,
1220         .reta_query                     = sfc_dev_rss_reta_query,
1221         .rss_hash_update                = sfc_dev_rss_hash_update,
1222         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1223 #endif
1224         .set_mc_addr_list               = sfc_set_mc_addr_list,
1225         .rxq_info_get                   = sfc_rx_queue_info_get,
1226         .txq_info_get                   = sfc_tx_queue_info_get,
1227 };
1228
1229 static int
1230 sfc_eth_dev_init(struct rte_eth_dev *dev)
1231 {
1232         struct sfc_adapter *sa = dev->data->dev_private;
1233         struct rte_pci_device *pci_dev = SFC_DEV_TO_PCI(dev);
1234         int rc;
1235         const efx_nic_cfg_t *encp;
1236         const struct ether_addr *from;
1237
1238         /* Required for logging */
1239         sa->eth_dev = dev;
1240
1241         /* Copy PCI device info to the dev->data */
1242         rte_eth_copy_pci_info(dev, pci_dev);
1243
1244         rc = sfc_kvargs_parse(sa);
1245         if (rc != 0)
1246                 goto fail_kvargs_parse;
1247
1248         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1249                                 sfc_kvarg_bool_handler, &sa->debug_init);
1250         if (rc != 0)
1251                 goto fail_kvarg_debug_init;
1252
1253         sfc_log_init(sa, "entry");
1254
1255         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1256         if (dev->data->mac_addrs == NULL) {
1257                 rc = ENOMEM;
1258                 goto fail_mac_addrs;
1259         }
1260
1261         sfc_adapter_lock_init(sa);
1262         sfc_adapter_lock(sa);
1263
1264         sfc_log_init(sa, "attaching");
1265         rc = sfc_attach(sa);
1266         if (rc != 0)
1267                 goto fail_attach;
1268
1269         encp = efx_nic_cfg_get(sa->nic);
1270
1271         /*
1272          * The arguments are really reverse order in comparison to
1273          * Linux kernel. Copy from NIC config to Ethernet device data.
1274          */
1275         from = (const struct ether_addr *)(encp->enc_mac_addr);
1276         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1277
1278         dev->dev_ops = &sfc_eth_dev_ops;
1279         dev->rx_pkt_burst = &sfc_recv_pkts;
1280         dev->tx_pkt_burst = &sfc_xmit_pkts;
1281
1282         sfc_adapter_unlock(sa);
1283
1284         sfc_log_init(sa, "done");
1285         return 0;
1286
1287 fail_attach:
1288         sfc_adapter_unlock(sa);
1289         sfc_adapter_lock_fini(sa);
1290         rte_free(dev->data->mac_addrs);
1291         dev->data->mac_addrs = NULL;
1292
1293 fail_mac_addrs:
1294 fail_kvarg_debug_init:
1295         sfc_kvargs_cleanup(sa);
1296
1297 fail_kvargs_parse:
1298         sfc_log_init(sa, "failed %d", rc);
1299         SFC_ASSERT(rc > 0);
1300         return -rc;
1301 }
1302
1303 static int
1304 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1305 {
1306         struct sfc_adapter *sa = dev->data->dev_private;
1307
1308         sfc_log_init(sa, "entry");
1309
1310         sfc_adapter_lock(sa);
1311
1312         sfc_detach(sa);
1313
1314         rte_free(dev->data->mac_addrs);
1315         dev->data->mac_addrs = NULL;
1316
1317         dev->dev_ops = NULL;
1318         dev->rx_pkt_burst = NULL;
1319         dev->tx_pkt_burst = NULL;
1320
1321         sfc_kvargs_cleanup(sa);
1322
1323         sfc_adapter_unlock(sa);
1324         sfc_adapter_lock_fini(sa);
1325
1326         sfc_log_init(sa, "done");
1327
1328         /* Required for logging, so cleanup last */
1329         sa->eth_dev = NULL;
1330         return 0;
1331 }
1332
1333 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1334         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1335         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1336         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1337         { .vendor_id = 0 /* sentinel */ }
1338 };
1339
1340 static struct eth_driver sfc_efx_pmd = {
1341         .pci_drv = {
1342                 .id_table = pci_id_sfc_efx_map,
1343                 .drv_flags =
1344                         RTE_PCI_DRV_INTR_LSC |
1345                         RTE_PCI_DRV_NEED_MAPPING,
1346                 .probe = rte_eth_dev_pci_probe,
1347                 .remove = rte_eth_dev_pci_remove,
1348         },
1349         .eth_dev_init = sfc_eth_dev_init,
1350         .eth_dev_uninit = sfc_eth_dev_uninit,
1351         .dev_private_size = sizeof(struct sfc_adapter),
1352 };
1353
1354 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd.pci_drv);
1355 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1356 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio");
1357 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1358         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1359         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1360         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);