drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
   4  * All rights reserved.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <rte_ethdev.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16
  17 #include "efx.h"
  18
  19 #include "sfc.h"
  20 #include "sfc_debug.h"
  21 #include "sfc_log.h"
  22 #include "sfc_kvargs.h"
  23 #include "sfc_ev.h"
  24 #include "sfc_rx.h"
  25 #include "sfc_tx.h"
  26 #include "sfc_flow.h"
  27 #include "sfc_dp.h"
  28 #include "sfc_dp_rx.h"
  29
  30 static struct sfc_dp_list sfc_dp_head =
  31         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  32
  33 static int
  34 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  35 {
  36         struct sfc_adapter *sa = dev->data->dev_private;
  37         efx_nic_fw_info_t enfi;
  38         int ret;
  39         int rc;
  40
  41         /*
  42          * Return value of the callback is likely supposed to be
  43          * equal to or greater than 0, nevertheless, if an error
  44          * occurs, it will be desirable to pass it to the caller
  45          */
  46         if ((fw_version == NULL) || (fw_size == 0))
  47                 return -EINVAL;
  48
  49         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  50         if (rc != 0)
  51                 return -rc;
  52
  53         ret = snprintf(fw_version, fw_size,
  54                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  55                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  56                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  57         if (ret < 0)
  58                 return ret;
  59
  60         if (enfi.enfi_dpcpu_fw_ids_valid) {
  61                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  62                 int ret_extra;
  63
  64                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  65                                      fw_size - dpcpu_fw_ids_offset,
  66                                      " rx%" PRIx16 " tx%" PRIx16,
  67                                      enfi.enfi_rx_dpcpu_fw_id,
  68                                      enfi.enfi_tx_dpcpu_fw_id);
  69                 if (ret_extra < 0)
  70                         return ret_extra;
  71
  72                 ret += ret_extra;
  73         }
  74
  75         if (fw_size < (size_t)(++ret))
  76                 return ret;
  77         else
  78                 return 0;
  79 }
  80
  81 static void
  82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  83 {
  84         struct sfc_adapter *sa = dev->data->dev_private;
  85         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
  86
  87         sfc_log_init(sa, "entry");
  88
  89         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
  90         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  91
  92         /* Autonegotiation may be disabled */
  93         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  94         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  95                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  96         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
  97                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
  98         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
  99                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 100
 101         dev_info->max_rx_queues = sa->rxq_max;
 102         dev_info->max_tx_queues = sa->txq_max;
 103
 104         /* By default packets are dropped if no descriptors are available */
 105         dev_info->default_rxconf.rx_drop_en = 1;
 106
 107         dev_info->rx_offload_capa =
 108                 DEV_RX_OFFLOAD_IPV4_CKSUM |
 109                 DEV_RX_OFFLOAD_UDP_CKSUM |
 110                 DEV_RX_OFFLOAD_TCP_CKSUM;
 111
 112         if ((encp->enc_tunnel_encapsulations_supported != 0) &&
 113             (sa->dp_rx->features & SFC_DP_RX_FEAT_TUNNELS))
 114                 dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM;
 115
 116         dev_info->tx_offload_capa =
 117                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 118                 DEV_TX_OFFLOAD_UDP_CKSUM |
 119                 DEV_TX_OFFLOAD_TCP_CKSUM;
 120
 121         if (encp->enc_tunnel_encapsulations_supported != 0)
 122                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
 123
 124         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 125         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 126             !encp->enc_hw_tx_insert_vlan_enabled)
 127                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 128         else
 129                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 130
 131         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 132                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 133
 134         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 135                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 136
 137         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 138                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 139
 140 #if EFSYS_OPT_RX_SCALE
 141         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 142                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 143                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 144                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 145         }
 146 #endif
 147
 148         if (sa->tso)
 149                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 150
 151         /* Initialize to hardware limits */
 152         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 153         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 154         /* The RXQ hardware requires that the descriptor count is a power
 155          * of 2, but rx_desc_lim cannot properly describe that constraint.
 156          */
 157         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 158
 159         /* Initialize to hardware limits */
 160         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 161         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 162         /*
 163          * The TXQ hardware requires that the descriptor count is a power
 164          * of 2, but tx_desc_lim cannot properly describe that constraint
 165          */
 166         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 167
 168         if (sa->dp_rx->get_dev_info != NULL)
 169                 sa->dp_rx->get_dev_info(dev_info);
 170         if (sa->dp_tx->get_dev_info != NULL)
 171                 sa->dp_tx->get_dev_info(dev_info);
 172 }
 173
 174 static const uint32_t *
 175 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 176 {
 177         struct sfc_adapter *sa = dev->data->dev_private;
 178         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 179         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 180
 181         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 182 }
 183
 184 static int
 185 sfc_dev_configure(struct rte_eth_dev *dev)
 186 {
 187         struct rte_eth_dev_data *dev_data = dev->data;
 188         struct sfc_adapter *sa = dev_data->dev_private;
 189         int rc;
 190
 191         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 192                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 193
 194         sfc_adapter_lock(sa);
 195         switch (sa->state) {
 196         case SFC_ADAPTER_CONFIGURED:
 197                 /* FALLTHROUGH */
 198         case SFC_ADAPTER_INITIALIZED:
 199                 rc = sfc_configure(sa);
 200                 break;
 201         default:
 202                 sfc_err(sa, "unexpected adapter state %u to configure",
 203                         sa->state);
 204                 rc = EINVAL;
 205                 break;
 206         }
 207         sfc_adapter_unlock(sa);
 208
 209         sfc_log_init(sa, "done %d", rc);
 210         SFC_ASSERT(rc >= 0);
 211         return -rc;
 212 }
 213
 214 static int
 215 sfc_dev_start(struct rte_eth_dev *dev)
 216 {
 217         struct sfc_adapter *sa = dev->data->dev_private;
 218         int rc;
 219
 220         sfc_log_init(sa, "entry");
 221
 222         sfc_adapter_lock(sa);
 223         rc = sfc_start(sa);
 224         sfc_adapter_unlock(sa);
 225
 226         sfc_log_init(sa, "done %d", rc);
 227         SFC_ASSERT(rc >= 0);
 228         return -rc;
 229 }
 230
 231 static int
 232 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 233 {
 234         struct sfc_adapter *sa = dev->data->dev_private;
 235         struct rte_eth_link *dev_link = &dev->data->dev_link;
 236         struct rte_eth_link old_link;
 237         struct rte_eth_link current_link;
 238
 239         sfc_log_init(sa, "entry");
 240
 241 retry:
 242         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 243         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 244
 245         if (sa->state != SFC_ADAPTER_STARTED) {
 246                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 247                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 248                                          *(uint64_t *)&old_link,
 249                                          *(uint64_t *)&current_link))
 250                         goto retry;
 251         } else if (wait_to_complete) {
 252                 efx_link_mode_t link_mode;
 253
 254                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 255                         link_mode = EFX_LINK_UNKNOWN;
 256                 sfc_port_link_mode_to_info(link_mode, &current_link);
 257
 258                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 259                                          *(uint64_t *)&old_link,
 260                                          *(uint64_t *)&current_link))
 261                         goto retry;
 262         } else {
 263                 sfc_ev_mgmt_qpoll(sa);
 264                 *(int64_t *)&current_link =
 265                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 266         }
 267
 268         if (old_link.link_status != current_link.link_status)
 269                 sfc_info(sa, "Link status is %s",
 270                          current_link.link_status ? "UP" : "DOWN");
 271
 272         return old_link.link_status == current_link.link_status ? 0 : -1;
 273 }
 274
 275 static void
 276 sfc_dev_stop(struct rte_eth_dev *dev)
 277 {
 278         struct sfc_adapter *sa = dev->data->dev_private;
 279
 280         sfc_log_init(sa, "entry");
 281
 282         sfc_adapter_lock(sa);
 283         sfc_stop(sa);
 284         sfc_adapter_unlock(sa);
 285
 286         sfc_log_init(sa, "done");
 287 }
 288
 289 static int
 290 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 291 {
 292         struct sfc_adapter *sa = dev->data->dev_private;
 293         int rc;
 294
 295         sfc_log_init(sa, "entry");
 296
 297         sfc_adapter_lock(sa);
 298         rc = sfc_start(sa);
 299         sfc_adapter_unlock(sa);
 300
 301         SFC_ASSERT(rc >= 0);
 302         return -rc;
 303 }
 304
 305 static int
 306 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 307 {
 308         struct sfc_adapter *sa = dev->data->dev_private;
 309
 310         sfc_log_init(sa, "entry");
 311
 312         sfc_adapter_lock(sa);
 313         sfc_stop(sa);
 314         sfc_adapter_unlock(sa);
 315
 316         return 0;
 317 }
 318
 319 static void
 320 sfc_dev_close(struct rte_eth_dev *dev)
 321 {
 322         struct sfc_adapter *sa = dev->data->dev_private;
 323
 324         sfc_log_init(sa, "entry");
 325
 326         sfc_adapter_lock(sa);
 327         switch (sa->state) {
 328         case SFC_ADAPTER_STARTED:
 329                 sfc_stop(sa);
 330                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 331                 /* FALLTHROUGH */
 332         case SFC_ADAPTER_CONFIGURED:
 333                 sfc_close(sa);
 334                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 335                 /* FALLTHROUGH */
 336         case SFC_ADAPTER_INITIALIZED:
 337                 break;
 338         default:
 339                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 340                 break;
 341         }
 342         sfc_adapter_unlock(sa);
 343
 344         sfc_log_init(sa, "done");
 345 }
 346
 347 static void
 348 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 349                    boolean_t enabled)
 350 {
 351         struct sfc_port *port;
 352         boolean_t *toggle;
 353         struct sfc_adapter *sa = dev->data->dev_private;
 354         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 355         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 356
 357         sfc_adapter_lock(sa);
 358
 359         port = &sa->port;
 360         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 361
 362         if (*toggle != enabled) {
 363                 *toggle = enabled;
 364
 365                 if (port->isolated) {
 366                         sfc_warn(sa, "isolated mode is active on the port");
 367                         sfc_warn(sa, "the change is to be applied on the next "
 368                                      "start provided that isolated mode is "
 369                                      "disabled prior the next start");
 370                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 371                            (sfc_set_rx_mode(sa) != 0)) {
 372                         *toggle = !(enabled);
 373                         sfc_warn(sa, "Failed to %s %s mode",
 374                                  ((enabled) ? "enable" : "disable"), desc);
 375                 }
 376         }
 377
 378         sfc_adapter_unlock(sa);
 379 }
 380
 381 static void
 382 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 383 {
 384         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 385 }
 386
 387 static void
 388 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 389 {
 390         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 391 }
 392
 393 static void
 394 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 395 {
 396         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 397 }
 398
 399 static void
 400 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 401 {
 402         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 403 }
 404
 405 static int
 406 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 407                    uint16_t nb_rx_desc, unsigned int socket_id,
 408                    const struct rte_eth_rxconf *rx_conf,
 409                    struct rte_mempool *mb_pool)
 410 {
 411         struct sfc_adapter *sa = dev->data->dev_private;
 412         int rc;
 413
 414         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 415                      rx_queue_id, nb_rx_desc, socket_id);
 416
 417         sfc_adapter_lock(sa);
 418
 419         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 420                           rx_conf, mb_pool);
 421         if (rc != 0)
 422                 goto fail_rx_qinit;
 423
 424         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 425
 426         sfc_adapter_unlock(sa);
 427
 428         return 0;
 429
 430 fail_rx_qinit:
 431         sfc_adapter_unlock(sa);
 432         SFC_ASSERT(rc > 0);
 433         return -rc;
 434 }
 435
 436 static void
 437 sfc_rx_queue_release(void *queue)
 438 {
 439         struct sfc_dp_rxq *dp_rxq = queue;
 440         struct sfc_rxq *rxq;
 441         struct sfc_adapter *sa;
 442         unsigned int sw_index;
 443
 444         if (dp_rxq == NULL)
 445                 return;
 446
 447         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 448         sa = rxq->evq->sa;
 449         sfc_adapter_lock(sa);
 450
 451         sw_index = sfc_rxq_sw_index(rxq);
 452
 453         sfc_log_init(sa, "RxQ=%u", sw_index);
 454
 455         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 456
 457         sfc_rx_qfini(sa, sw_index);
 458
 459         sfc_adapter_unlock(sa);
 460 }
 461
 462 static int
 463 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 464                    uint16_t nb_tx_desc, unsigned int socket_id,
 465                    const struct rte_eth_txconf *tx_conf)
 466 {
 467         struct sfc_adapter *sa = dev->data->dev_private;
 468         int rc;
 469
 470         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 471                      tx_queue_id, nb_tx_desc, socket_id);
 472
 473         sfc_adapter_lock(sa);
 474
 475         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 476         if (rc != 0)
 477                 goto fail_tx_qinit;
 478
 479         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 480
 481         sfc_adapter_unlock(sa);
 482         return 0;
 483
 484 fail_tx_qinit:
 485         sfc_adapter_unlock(sa);
 486         SFC_ASSERT(rc > 0);
 487         return -rc;
 488 }
 489
 490 static void
 491 sfc_tx_queue_release(void *queue)
 492 {
 493         struct sfc_dp_txq *dp_txq = queue;
 494         struct sfc_txq *txq;
 495         unsigned int sw_index;
 496         struct sfc_adapter *sa;
 497
 498         if (dp_txq == NULL)
 499                 return;
 500
 501         txq = sfc_txq_by_dp_txq(dp_txq);
 502         sw_index = sfc_txq_sw_index(txq);
 503
 504         SFC_ASSERT(txq->evq != NULL);
 505         sa = txq->evq->sa;
 506
 507         sfc_log_init(sa, "TxQ = %u", sw_index);
 508
 509         sfc_adapter_lock(sa);
 510
 511         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 512         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 513
 514         sfc_tx_qfini(sa, sw_index);
 515
 516         sfc_adapter_unlock(sa);
 517 }
 518
 519 static int
 520 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 521 {
 522         struct sfc_adapter *sa = dev->data->dev_private;
 523         struct sfc_port *port = &sa->port;
 524         uint64_t *mac_stats;
 525         int ret;
 526
 527         rte_spinlock_lock(&port->mac_stats_lock);
 528
 529         ret = sfc_port_update_mac_stats(sa);
 530         if (ret != 0)
 531                 goto unlock;
 532
 533         mac_stats = port->mac_stats_buf;
 534
 535         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 536                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 537                 stats->ipackets =
 538                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 539                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 540                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 541                 stats->opackets =
 542                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 543                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 544                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 545                 stats->ibytes =
 546                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 547                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 548                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 549                 stats->obytes =
 550                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 551                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 552                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 553                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 554                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 555                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 556         } else {
 557                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 558                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 559                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 560                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 561                 /*
 562                  * Take into account stats which are whenever supported
 563                  * on EF10. If some stat is not supported by current
 564                  * firmware variant or HW revision, it is guaranteed
 565                  * to be zero in mac_stats.
 566                  */
 567                 stats->imissed =
 568                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 569                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 570                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 571                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 572                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 573                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 574                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 575                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 576                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 577                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 578                 stats->ierrors =
 579                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 580                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 581                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 582                 /* no oerrors counters supported on EF10 */
 583         }
 584
 585 unlock:
 586         rte_spinlock_unlock(&port->mac_stats_lock);
 587         SFC_ASSERT(ret >= 0);
 588         return -ret;
 589 }
 590
 591 static void
 592 sfc_stats_reset(struct rte_eth_dev *dev)
 593 {
 594         struct sfc_adapter *sa = dev->data->dev_private;
 595         struct sfc_port *port = &sa->port;
 596         int rc;
 597
 598         if (sa->state != SFC_ADAPTER_STARTED) {
 599                 /*
 600                  * The operation cannot be done if port is not started; it
 601                  * will be scheduled to be done during the next port start
 602                  */
 603                 port->mac_stats_reset_pending = B_TRUE;
 604                 return;
 605         }
 606
 607         rc = sfc_port_reset_mac_stats(sa);
 608         if (rc != 0)
 609                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 610 }
 611
 612 static int
 613 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 614                unsigned int xstats_count)
 615 {
 616         struct sfc_adapter *sa = dev->data->dev_private;
 617         struct sfc_port *port = &sa->port;
 618         uint64_t *mac_stats;
 619         int rc;
 620         unsigned int i;
 621         int nstats = 0;
 622
 623         rte_spinlock_lock(&port->mac_stats_lock);
 624
 625         rc = sfc_port_update_mac_stats(sa);
 626         if (rc != 0) {
 627                 SFC_ASSERT(rc > 0);
 628                 nstats = -rc;
 629                 goto unlock;
 630         }
 631
 632         mac_stats = port->mac_stats_buf;
 633
 634         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 635                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 636                         if (xstats != NULL && nstats < (int)xstats_count) {
 637                                 xstats[nstats].id = nstats;
 638                                 xstats[nstats].value = mac_stats[i];
 639                         }
 640                         nstats++;
 641                 }
 642         }
 643
 644 unlock:
 645         rte_spinlock_unlock(&port->mac_stats_lock);
 646
 647         return nstats;
 648 }
 649
 650 static int
 651 sfc_xstats_get_names(struct rte_eth_dev *dev,
 652                      struct rte_eth_xstat_name *xstats_names,
 653                      unsigned int xstats_count)
 654 {
 655         struct sfc_adapter *sa = dev->data->dev_private;
 656         struct sfc_port *port = &sa->port;
 657         unsigned int i;
 658         unsigned int nstats = 0;
 659
 660         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 661                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 662                         if (xstats_names != NULL && nstats < xstats_count)
 663                                 strncpy(xstats_names[nstats].name,
 664                                         efx_mac_stat_name(sa->nic, i),
 665                                         sizeof(xstats_names[0].name));
 666                         nstats++;
 667                 }
 668         }
 669
 670         return nstats;
 671 }
 672
 673 static int
 674 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 675                      uint64_t *values, unsigned int n)
 676 {
 677         struct sfc_adapter *sa = dev->data->dev_private;
 678         struct sfc_port *port = &sa->port;
 679         uint64_t *mac_stats;
 680         unsigned int nb_supported = 0;
 681         unsigned int nb_written = 0;
 682         unsigned int i;
 683         int ret;
 684         int rc;
 685
 686         if (unlikely(values == NULL) ||
 687             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 688                 return port->mac_stats_nb_supported;
 689
 690         rte_spinlock_lock(&port->mac_stats_lock);
 691
 692         rc = sfc_port_update_mac_stats(sa);
 693         if (rc != 0) {
 694                 SFC_ASSERT(rc > 0);
 695                 ret = -rc;
 696                 goto unlock;
 697         }
 698
 699         mac_stats = port->mac_stats_buf;
 700
 701         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 702                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 703                         continue;
 704
 705                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 706                         values[nb_written++] = mac_stats[i];
 707
 708                 ++nb_supported;
 709         }
 710
 711         ret = nb_written;
 712
 713 unlock:
 714         rte_spinlock_unlock(&port->mac_stats_lock);
 715
 716         return ret;
 717 }
 718
 719 static int
 720 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 721                            struct rte_eth_xstat_name *xstats_names,
 722                            const uint64_t *ids, unsigned int size)
 723 {
 724         struct sfc_adapter *sa = dev->data->dev_private;
 725         struct sfc_port *port = &sa->port;
 726         unsigned int nb_supported = 0;
 727         unsigned int nb_written = 0;
 728         unsigned int i;
 729
 730         if (unlikely(xstats_names == NULL) ||
 731             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 732                 return port->mac_stats_nb_supported;
 733
 734         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 735                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 736                         continue;
 737
 738                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 739                         char *name = xstats_names[nb_written++].name;
 740
 741                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 742                                 sizeof(xstats_names[0].name));
 743                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 744                 }
 745
 746                 ++nb_supported;
 747         }
 748
 749         return nb_written;
 750 }
 751
 752 static int
 753 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 754 {
 755         struct sfc_adapter *sa = dev->data->dev_private;
 756         unsigned int wanted_fc, link_fc;
 757
 758         memset(fc_conf, 0, sizeof(*fc_conf));
 759
 760         sfc_adapter_lock(sa);
 761
 762         if (sa->state == SFC_ADAPTER_STARTED)
 763                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 764         else
 765                 link_fc = sa->port.flow_ctrl;
 766
 767         switch (link_fc) {
 768         case 0:
 769                 fc_conf->mode = RTE_FC_NONE;
 770                 break;
 771         case EFX_FCNTL_RESPOND:
 772                 fc_conf->mode = RTE_FC_RX_PAUSE;
 773                 break;
 774         case EFX_FCNTL_GENERATE:
 775                 fc_conf->mode = RTE_FC_TX_PAUSE;
 776                 break;
 777         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 778                 fc_conf->mode = RTE_FC_FULL;
 779                 break;
 780         default:
 781                 sfc_err(sa, "%s: unexpected flow control value %#x",
 782                         __func__, link_fc);
 783         }
 784
 785         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 786
 787         sfc_adapter_unlock(sa);
 788
 789         return 0;
 790 }
 791
 792 static int
 793 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 794 {
 795         struct sfc_adapter *sa = dev->data->dev_private;
 796         struct sfc_port *port = &sa->port;
 797         unsigned int fcntl;
 798         int rc;
 799
 800         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 801             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 802             fc_conf->mac_ctrl_frame_fwd != 0) {
 803                 sfc_err(sa, "unsupported flow control settings specified");
 804                 rc = EINVAL;
 805                 goto fail_inval;
 806         }
 807
 808         switch (fc_conf->mode) {
 809         case RTE_FC_NONE:
 810                 fcntl = 0;
 811                 break;
 812         case RTE_FC_RX_PAUSE:
 813                 fcntl = EFX_FCNTL_RESPOND;
 814                 break;
 815         case RTE_FC_TX_PAUSE:
 816                 fcntl = EFX_FCNTL_GENERATE;
 817                 break;
 818         case RTE_FC_FULL:
 819                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 820                 break;
 821         default:
 822                 rc = EINVAL;
 823                 goto fail_inval;
 824         }
 825
 826         sfc_adapter_lock(sa);
 827
 828         if (sa->state == SFC_ADAPTER_STARTED) {
 829                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 830                 if (rc != 0)
 831                         goto fail_mac_fcntl_set;
 832         }
 833
 834         port->flow_ctrl = fcntl;
 835         port->flow_ctrl_autoneg = fc_conf->autoneg;
 836
 837         sfc_adapter_unlock(sa);
 838
 839         return 0;
 840
 841 fail_mac_fcntl_set:
 842         sfc_adapter_unlock(sa);
 843 fail_inval:
 844         SFC_ASSERT(rc > 0);
 845         return -rc;
 846 }
 847
 848 static int
 849 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 850 {
 851         struct sfc_adapter *sa = dev->data->dev_private;
 852         size_t pdu = EFX_MAC_PDU(mtu);
 853         size_t old_pdu;
 854         int rc;
 855
 856         sfc_log_init(sa, "mtu=%u", mtu);
 857
 858         rc = EINVAL;
 859         if (pdu < EFX_MAC_PDU_MIN) {
 860                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 861                         (unsigned int)mtu, (unsigned int)pdu,
 862                         EFX_MAC_PDU_MIN);
 863                 goto fail_inval;
 864         }
 865         if (pdu > EFX_MAC_PDU_MAX) {
 866                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 867                         (unsigned int)mtu, (unsigned int)pdu,
 868                         EFX_MAC_PDU_MAX);
 869                 goto fail_inval;
 870         }
 871
 872         sfc_adapter_lock(sa);
 873
 874         if (pdu != sa->port.pdu) {
 875                 if (sa->state == SFC_ADAPTER_STARTED) {
 876                         sfc_stop(sa);
 877
 878                         old_pdu = sa->port.pdu;
 879                         sa->port.pdu = pdu;
 880                         rc = sfc_start(sa);
 881                         if (rc != 0)
 882                                 goto fail_start;
 883                 } else {
 884                         sa->port.pdu = pdu;
 885                 }
 886         }
 887
 888         /*
 889          * The driver does not use it, but other PMDs update jumbo_frame
 890          * flag and max_rx_pkt_len when MTU is set.
 891          */
 892         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 893         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 894
 895         sfc_adapter_unlock(sa);
 896
 897         sfc_log_init(sa, "done");
 898         return 0;
 899
 900 fail_start:
 901         sa->port.pdu = old_pdu;
 902         if (sfc_start(sa) != 0)
 903                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 904                         "PDU max size - port is stopped",
 905                         (unsigned int)pdu, (unsigned int)old_pdu);
 906         sfc_adapter_unlock(sa);
 907
 908 fail_inval:
 909         sfc_log_init(sa, "failed %d", rc);
 910         SFC_ASSERT(rc > 0);
 911         return -rc;
 912 }
 913 static void
 914 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 915 {
 916         struct sfc_adapter *sa = dev->data->dev_private;
 917         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 918         struct sfc_port *port = &sa->port;
 919         int rc;
 920
 921         sfc_adapter_lock(sa);
 922
 923         /*
 924          * Copy the address to the device private data so that
 925          * it could be recalled in the case of adapter restart.
 926          */
 927         ether_addr_copy(mac_addr, &port->default_mac_addr);
 928
 929         if (port->isolated) {
 930                 sfc_err(sa, "isolated mode is active on the port");
 931                 sfc_err(sa, "will not set MAC address");
 932                 goto unlock;
 933         }
 934
 935         if (sa->state != SFC_ADAPTER_STARTED) {
 936                 sfc_info(sa, "the port is not started");
 937                 sfc_info(sa, "the new MAC address will be set on port start");
 938
 939                 goto unlock;
 940         }
 941
 942         if (encp->enc_allow_set_mac_with_installed_filters) {
 943                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 944                 if (rc != 0) {
 945                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 946                         goto unlock;
 947                 }
 948
 949                 /*
 950                  * Changing the MAC address by means of MCDI request
 951                  * has no effect on received traffic, therefore
 952                  * we also need to update unicast filters
 953                  */
 954                 rc = sfc_set_rx_mode(sa);
 955                 if (rc != 0)
 956                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 957         } else {
 958                 sfc_warn(sa, "cannot set MAC address with filters installed");
 959                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 960                 sfc_warn(sa, "(some traffic may be dropped)");
 961
 962                 /*
 963                  * Since setting MAC address with filters installed is not
 964                  * allowed on the adapter, the new MAC address will be set
 965                  * by means of adapter restart. sfc_start() shall retrieve
 966                  * the new address from the device private data and set it.
 967                  */
 968                 sfc_stop(sa);
 969                 rc = sfc_start(sa);
 970                 if (rc != 0)
 971                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 972         }
 973
 974 unlock:
 975         /*
 976          * In the case of failure sa->port->default_mac_addr does not
 977          * need rollback since no error code is returned, and the upper
 978          * API will anyway update the external MAC address storage.
 979          * To be consistent with that new value it is better to keep
 980          * the device private value the same.
 981          */
 982         sfc_adapter_unlock(sa);
 983 }
 984
 985
 986 static int
 987 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 988                      uint32_t nb_mc_addr)
 989 {
 990         struct sfc_adapter *sa = dev->data->dev_private;
 991         struct sfc_port *port = &sa->port;
 992         uint8_t *mc_addrs = port->mcast_addrs;
 993         int rc;
 994         unsigned int i;
 995
 996         if (port->isolated) {
 997                 sfc_err(sa, "isolated mode is active on the port");
 998                 sfc_err(sa, "will not set multicast address list");
 999                 return -ENOTSUP;
1000         }
1001
1002         if (mc_addrs == NULL)
1003                 return -ENOBUFS;
1004
1005         if (nb_mc_addr > port->max_mcast_addrs) {
1006                 sfc_err(sa, "too many multicast addresses: %u > %u",
1007                          nb_mc_addr, port->max_mcast_addrs);
1008                 return -EINVAL;
1009         }
1010
1011         for (i = 0; i < nb_mc_addr; ++i) {
1012                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1013                                  EFX_MAC_ADDR_LEN);
1014                 mc_addrs += EFX_MAC_ADDR_LEN;
1015         }
1016
1017         port->nb_mcast_addrs = nb_mc_addr;
1018
1019         if (sa->state != SFC_ADAPTER_STARTED)
1020                 return 0;
1021
1022         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1023                                         port->nb_mcast_addrs);
1024         if (rc != 0)
1025                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1026
1027         SFC_ASSERT(rc > 0);
1028         return -rc;
1029 }
1030
1031 /*
1032  * The function is used by the secondary process as well. It must not
1033  * use any process-local pointers from the adapter data.
1034  */
1035 static void
1036 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1037                       struct rte_eth_rxq_info *qinfo)
1038 {
1039         struct sfc_adapter *sa = dev->data->dev_private;
1040         struct sfc_rxq_info *rxq_info;
1041         struct sfc_rxq *rxq;
1042
1043         sfc_adapter_lock(sa);
1044
1045         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1046
1047         rxq_info = &sa->rxq_info[rx_queue_id];
1048         rxq = rxq_info->rxq;
1049         SFC_ASSERT(rxq != NULL);
1050
1051         qinfo->mp = rxq->refill_mb_pool;
1052         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1053         qinfo->conf.rx_drop_en = 1;
1054         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1055         qinfo->scattered_rx =
1056                 ((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1057         qinfo->nb_desc = rxq_info->entries;
1058
1059         sfc_adapter_unlock(sa);
1060 }
1061
1062 /*
1063  * The function is used by the secondary process as well. It must not
1064  * use any process-local pointers from the adapter data.
1065  */
1066 static void
1067 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1068                       struct rte_eth_txq_info *qinfo)
1069 {
1070         struct sfc_adapter *sa = dev->data->dev_private;
1071         struct sfc_txq_info *txq_info;
1072
1073         sfc_adapter_lock(sa);
1074
1075         SFC_ASSERT(tx_queue_id < sa->txq_count);
1076
1077         txq_info = &sa->txq_info[tx_queue_id];
1078         SFC_ASSERT(txq_info->txq != NULL);
1079
1080         memset(qinfo, 0, sizeof(*qinfo));
1081
1082         qinfo->conf.txq_flags = txq_info->txq->flags;
1083         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1084         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1085         qinfo->nb_desc = txq_info->entries;
1086
1087         sfc_adapter_unlock(sa);
1088 }
1089
1090 static uint32_t
1091 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1092 {
1093         struct sfc_adapter *sa = dev->data->dev_private;
1094
1095         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1096
1097         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1098 }
1099
1100 static int
1101 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1102 {
1103         struct sfc_dp_rxq *dp_rxq = queue;
1104
1105         return sfc_rx_qdesc_done(dp_rxq, offset);
1106 }
1107
1108 static int
1109 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1110 {
1111         struct sfc_dp_rxq *dp_rxq = queue;
1112         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1113
1114         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1115 }
1116
1117 static int
1118 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1119 {
1120         struct sfc_dp_txq *dp_txq = queue;
1121         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1122
1123         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1124 }
1125
1126 static int
1127 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1128 {
1129         struct sfc_adapter *sa = dev->data->dev_private;
1130         int rc;
1131
1132         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1133
1134         sfc_adapter_lock(sa);
1135
1136         rc = EINVAL;
1137         if (sa->state != SFC_ADAPTER_STARTED)
1138                 goto fail_not_started;
1139
1140         rc = sfc_rx_qstart(sa, rx_queue_id);
1141         if (rc != 0)
1142                 goto fail_rx_qstart;
1143
1144         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1145
1146         sfc_adapter_unlock(sa);
1147
1148         return 0;
1149
1150 fail_rx_qstart:
1151 fail_not_started:
1152         sfc_adapter_unlock(sa);
1153         SFC_ASSERT(rc > 0);
1154         return -rc;
1155 }
1156
1157 static int
1158 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1159 {
1160         struct sfc_adapter *sa = dev->data->dev_private;
1161
1162         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1163
1164         sfc_adapter_lock(sa);
1165         sfc_rx_qstop(sa, rx_queue_id);
1166
1167         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1168
1169         sfc_adapter_unlock(sa);
1170
1171         return 0;
1172 }
1173
1174 static int
1175 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1176 {
1177         struct sfc_adapter *sa = dev->data->dev_private;
1178         int rc;
1179
1180         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1181
1182         sfc_adapter_lock(sa);
1183
1184         rc = EINVAL;
1185         if (sa->state != SFC_ADAPTER_STARTED)
1186                 goto fail_not_started;
1187
1188         rc = sfc_tx_qstart(sa, tx_queue_id);
1189         if (rc != 0)
1190                 goto fail_tx_qstart;
1191
1192         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1193
1194         sfc_adapter_unlock(sa);
1195         return 0;
1196
1197 fail_tx_qstart:
1198
1199 fail_not_started:
1200         sfc_adapter_unlock(sa);
1201         SFC_ASSERT(rc > 0);
1202         return -rc;
1203 }
1204
1205 static int
1206 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1207 {
1208         struct sfc_adapter *sa = dev->data->dev_private;
1209
1210         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1211
1212         sfc_adapter_lock(sa);
1213
1214         sfc_tx_qstop(sa, tx_queue_id);
1215
1216         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1217
1218         sfc_adapter_unlock(sa);
1219         return 0;
1220 }
1221
1222 static efx_tunnel_protocol_t
1223 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1224 {
1225         switch (rte_type) {
1226         case RTE_TUNNEL_TYPE_VXLAN:
1227                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1228         case RTE_TUNNEL_TYPE_GENEVE:
1229                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1230         default:
1231                 return EFX_TUNNEL_NPROTOS;
1232         }
1233 }
1234
1235 enum sfc_udp_tunnel_op_e {
1236         SFC_UDP_TUNNEL_ADD_PORT,
1237         SFC_UDP_TUNNEL_DEL_PORT,
1238 };
1239
1240 static int
1241 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1242                       struct rte_eth_udp_tunnel *tunnel_udp,
1243                       enum sfc_udp_tunnel_op_e op)
1244 {
1245         struct sfc_adapter *sa = dev->data->dev_private;
1246         efx_tunnel_protocol_t tunnel_proto;
1247         int rc;
1248
1249         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1250                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1251                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1252                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1253
1254         tunnel_proto =
1255                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1256         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1257                 rc = ENOTSUP;
1258                 goto fail_bad_proto;
1259         }
1260
1261         sfc_adapter_lock(sa);
1262
1263         switch (op) {
1264         case SFC_UDP_TUNNEL_ADD_PORT:
1265                 rc = efx_tunnel_config_udp_add(sa->nic,
1266                                                tunnel_udp->udp_port,
1267                                                tunnel_proto);
1268                 break;
1269         case SFC_UDP_TUNNEL_DEL_PORT:
1270                 rc = efx_tunnel_config_udp_remove(sa->nic,
1271                                                   tunnel_udp->udp_port,
1272                                                   tunnel_proto);
1273                 break;
1274         default:
1275                 rc = EINVAL;
1276                 goto fail_bad_op;
1277         }
1278
1279         if (rc != 0)
1280                 goto fail_op;
1281
1282         if (sa->state == SFC_ADAPTER_STARTED) {
1283                 rc = efx_tunnel_reconfigure(sa->nic);
1284                 if (rc == EAGAIN) {
1285                         /*
1286                          * Configuration is accepted by FW and MC reboot
1287                          * is initiated to apply the changes. MC reboot
1288                          * will be handled in a usual way (MC reboot
1289                          * event on management event queue and adapter
1290                          * restart).
1291                          */
1292                         rc = 0;
1293                 } else if (rc != 0) {
1294                         goto fail_reconfigure;
1295                 }
1296         }
1297
1298         sfc_adapter_unlock(sa);
1299         return 0;
1300
1301 fail_reconfigure:
1302         /* Remove/restore entry since the change makes the trouble */
1303         switch (op) {
1304         case SFC_UDP_TUNNEL_ADD_PORT:
1305                 (void)efx_tunnel_config_udp_remove(sa->nic,
1306                                                    tunnel_udp->udp_port,
1307                                                    tunnel_proto);
1308                 break;
1309         case SFC_UDP_TUNNEL_DEL_PORT:
1310                 (void)efx_tunnel_config_udp_add(sa->nic,
1311                                                 tunnel_udp->udp_port,
1312                                                 tunnel_proto);
1313                 break;
1314         }
1315
1316 fail_op:
1317 fail_bad_op:
1318         sfc_adapter_unlock(sa);
1319
1320 fail_bad_proto:
1321         SFC_ASSERT(rc > 0);
1322         return -rc;
1323 }
1324
1325 static int
1326 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1327                             struct rte_eth_udp_tunnel *tunnel_udp)
1328 {
1329         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1330 }
1331
1332 static int
1333 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1334                             struct rte_eth_udp_tunnel *tunnel_udp)
1335 {
1336         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1337 }
1338
1339 #if EFSYS_OPT_RX_SCALE
1340 static int
1341 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1342                           struct rte_eth_rss_conf *rss_conf)
1343 {
1344         struct sfc_adapter *sa = dev->data->dev_private;
1345         struct sfc_port *port = &sa->port;
1346
1347         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1348                 return -ENOTSUP;
1349
1350         if (sa->rss_channels == 0)
1351                 return -EINVAL;
1352
1353         sfc_adapter_lock(sa);
1354
1355         /*
1356          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1357          * hence, conversion is done here to derive a correct set of ETH_RSS
1358          * flags which corresponds to the active EFX configuration stored
1359          * locally in 'sfc_adapter' and kept up-to-date
1360          */
1361         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1362         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1363         if (rss_conf->rss_key != NULL)
1364                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1365
1366         sfc_adapter_unlock(sa);
1367
1368         return 0;
1369 }
1370
1371 static int
1372 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1373                         struct rte_eth_rss_conf *rss_conf)
1374 {
1375         struct sfc_adapter *sa = dev->data->dev_private;
1376         struct sfc_port *port = &sa->port;
1377         unsigned int efx_hash_types;
1378         int rc = 0;
1379
1380         if (port->isolated)
1381                 return -ENOTSUP;
1382
1383         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1384                 sfc_err(sa, "RSS is not available");
1385                 return -ENOTSUP;
1386         }
1387
1388         if (sa->rss_channels == 0) {
1389                 sfc_err(sa, "RSS is not configured");
1390                 return -EINVAL;
1391         }
1392
1393         if ((rss_conf->rss_key != NULL) &&
1394             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1395                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1396                         sizeof(sa->rss_key));
1397                 return -EINVAL;
1398         }
1399
1400         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1401                 sfc_err(sa, "unsupported hash functions requested");
1402                 return -EINVAL;
1403         }
1404
1405         sfc_adapter_lock(sa);
1406
1407         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1408
1409         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1410                                    EFX_RX_HASHALG_TOEPLITZ,
1411                                    efx_hash_types, B_TRUE);
1412         if (rc != 0)
1413                 goto fail_scale_mode_set;
1414
1415         if (rss_conf->rss_key != NULL) {
1416                 if (sa->state == SFC_ADAPTER_STARTED) {
1417                         rc = efx_rx_scale_key_set(sa->nic,
1418                                                   EFX_RSS_CONTEXT_DEFAULT,
1419                                                   rss_conf->rss_key,
1420                                                   sizeof(sa->rss_key));
1421                         if (rc != 0)
1422                                 goto fail_scale_key_set;
1423                 }
1424
1425                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1426         }
1427
1428         sa->rss_hash_types = efx_hash_types;
1429
1430         sfc_adapter_unlock(sa);
1431
1432         return 0;
1433
1434 fail_scale_key_set:
1435         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1436                                   EFX_RX_HASHALG_TOEPLITZ,
1437                                   sa->rss_hash_types, B_TRUE) != 0)
1438                 sfc_err(sa, "failed to restore RSS mode");
1439
1440 fail_scale_mode_set:
1441         sfc_adapter_unlock(sa);
1442         return -rc;
1443 }
1444
1445 static int
1446 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1447                        struct rte_eth_rss_reta_entry64 *reta_conf,
1448                        uint16_t reta_size)
1449 {
1450         struct sfc_adapter *sa = dev->data->dev_private;
1451         struct sfc_port *port = &sa->port;
1452         int entry;
1453
1454         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1455                 return -ENOTSUP;
1456
1457         if (sa->rss_channels == 0)
1458                 return -EINVAL;
1459
1460         if (reta_size != EFX_RSS_TBL_SIZE)
1461                 return -EINVAL;
1462
1463         sfc_adapter_lock(sa);
1464
1465         for (entry = 0; entry < reta_size; entry++) {
1466                 int grp = entry / RTE_RETA_GROUP_SIZE;
1467                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1468
1469                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1470                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1471         }
1472
1473         sfc_adapter_unlock(sa);
1474
1475         return 0;
1476 }
1477
1478 static int
1479 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1480                         struct rte_eth_rss_reta_entry64 *reta_conf,
1481                         uint16_t reta_size)
1482 {
1483         struct sfc_adapter *sa = dev->data->dev_private;
1484         struct sfc_port *port = &sa->port;
1485         unsigned int *rss_tbl_new;
1486         uint16_t entry;
1487         int rc = 0;
1488
1489
1490         if (port->isolated)
1491                 return -ENOTSUP;
1492
1493         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1494                 sfc_err(sa, "RSS is not available");
1495                 return -ENOTSUP;
1496         }
1497
1498         if (sa->rss_channels == 0) {
1499                 sfc_err(sa, "RSS is not configured");
1500                 return -EINVAL;
1501         }
1502
1503         if (reta_size != EFX_RSS_TBL_SIZE) {
1504                 sfc_err(sa, "RETA size is wrong (should be %u)",
1505                         EFX_RSS_TBL_SIZE);
1506                 return -EINVAL;
1507         }
1508
1509         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1510         if (rss_tbl_new == NULL)
1511                 return -ENOMEM;
1512
1513         sfc_adapter_lock(sa);
1514
1515         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1516
1517         for (entry = 0; entry < reta_size; entry++) {
1518                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1519                 struct rte_eth_rss_reta_entry64 *grp;
1520
1521                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1522
1523                 if (grp->mask & (1ull << grp_idx)) {
1524                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1525                                 rc = EINVAL;
1526                                 goto bad_reta_entry;
1527                         }
1528                         rss_tbl_new[entry] = grp->reta[grp_idx];
1529                 }
1530         }
1531
1532         if (sa->state == SFC_ADAPTER_STARTED) {
1533                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1534                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1535                 if (rc != 0)
1536                         goto fail_scale_tbl_set;
1537         }
1538
1539         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1540
1541 fail_scale_tbl_set:
1542 bad_reta_entry:
1543         sfc_adapter_unlock(sa);
1544
1545         rte_free(rss_tbl_new);
1546
1547         SFC_ASSERT(rc >= 0);
1548         return -rc;
1549 }
1550 #endif
1551
1552 static int
1553 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1554                     enum rte_filter_op filter_op,
1555                     void *arg)
1556 {
1557         struct sfc_adapter *sa = dev->data->dev_private;
1558         int rc = ENOTSUP;
1559
1560         sfc_log_init(sa, "entry");
1561
1562         switch (filter_type) {
1563         case RTE_ETH_FILTER_NONE:
1564                 sfc_err(sa, "Global filters configuration not supported");
1565                 break;
1566         case RTE_ETH_FILTER_MACVLAN:
1567                 sfc_err(sa, "MACVLAN filters not supported");
1568                 break;
1569         case RTE_ETH_FILTER_ETHERTYPE:
1570                 sfc_err(sa, "EtherType filters not supported");
1571                 break;
1572         case RTE_ETH_FILTER_FLEXIBLE:
1573                 sfc_err(sa, "Flexible filters not supported");
1574                 break;
1575         case RTE_ETH_FILTER_SYN:
1576                 sfc_err(sa, "SYN filters not supported");
1577                 break;
1578         case RTE_ETH_FILTER_NTUPLE:
1579                 sfc_err(sa, "NTUPLE filters not supported");
1580                 break;
1581         case RTE_ETH_FILTER_TUNNEL:
1582                 sfc_err(sa, "Tunnel filters not supported");
1583                 break;
1584         case RTE_ETH_FILTER_FDIR:
1585                 sfc_err(sa, "Flow Director filters not supported");
1586                 break;
1587         case RTE_ETH_FILTER_HASH:
1588                 sfc_err(sa, "Hash filters not supported");
1589                 break;
1590         case RTE_ETH_FILTER_GENERIC:
1591                 if (filter_op != RTE_ETH_FILTER_GET) {
1592                         rc = EINVAL;
1593                 } else {
1594                         *(const void **)arg = &sfc_flow_ops;
1595                         rc = 0;
1596                 }
1597                 break;
1598         default:
1599                 sfc_err(sa, "Unknown filter type %u", filter_type);
1600                 break;
1601         }
1602
1603         sfc_log_init(sa, "exit: %d", -rc);
1604         SFC_ASSERT(rc >= 0);
1605         return -rc;
1606 }
1607
1608 static const struct eth_dev_ops sfc_eth_dev_ops = {
1609         .dev_configure                  = sfc_dev_configure,
1610         .dev_start                      = sfc_dev_start,
1611         .dev_stop                       = sfc_dev_stop,
1612         .dev_set_link_up                = sfc_dev_set_link_up,
1613         .dev_set_link_down              = sfc_dev_set_link_down,
1614         .dev_close                      = sfc_dev_close,
1615         .promiscuous_enable             = sfc_dev_promisc_enable,
1616         .promiscuous_disable            = sfc_dev_promisc_disable,
1617         .allmulticast_enable            = sfc_dev_allmulti_enable,
1618         .allmulticast_disable           = sfc_dev_allmulti_disable,
1619         .link_update                    = sfc_dev_link_update,
1620         .stats_get                      = sfc_stats_get,
1621         .stats_reset                    = sfc_stats_reset,
1622         .xstats_get                     = sfc_xstats_get,
1623         .xstats_reset                   = sfc_stats_reset,
1624         .xstats_get_names               = sfc_xstats_get_names,
1625         .dev_infos_get                  = sfc_dev_infos_get,
1626         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1627         .mtu_set                        = sfc_dev_set_mtu,
1628         .rx_queue_start                 = sfc_rx_queue_start,
1629         .rx_queue_stop                  = sfc_rx_queue_stop,
1630         .tx_queue_start                 = sfc_tx_queue_start,
1631         .tx_queue_stop                  = sfc_tx_queue_stop,
1632         .rx_queue_setup                 = sfc_rx_queue_setup,
1633         .rx_queue_release               = sfc_rx_queue_release,
1634         .rx_queue_count                 = sfc_rx_queue_count,
1635         .rx_descriptor_done             = sfc_rx_descriptor_done,
1636         .rx_descriptor_status           = sfc_rx_descriptor_status,
1637         .tx_descriptor_status           = sfc_tx_descriptor_status,
1638         .tx_queue_setup                 = sfc_tx_queue_setup,
1639         .tx_queue_release               = sfc_tx_queue_release,
1640         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1641         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1642         .mac_addr_set                   = sfc_mac_addr_set,
1643         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1644         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1645 #if EFSYS_OPT_RX_SCALE
1646         .reta_update                    = sfc_dev_rss_reta_update,
1647         .reta_query                     = sfc_dev_rss_reta_query,
1648         .rss_hash_update                = sfc_dev_rss_hash_update,
1649         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1650 #endif
1651         .filter_ctrl                    = sfc_dev_filter_ctrl,
1652         .set_mc_addr_list               = sfc_set_mc_addr_list,
1653         .rxq_info_get                   = sfc_rx_queue_info_get,
1654         .txq_info_get                   = sfc_tx_queue_info_get,
1655         .fw_version_get                 = sfc_fw_version_get,
1656         .xstats_get_by_id               = sfc_xstats_get_by_id,
1657         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1658 };
1659
1660 /**
1661  * Duplicate a string in potentially shared memory required for
1662  * multi-process support.
1663  *
1664  * strdup() allocates from process-local heap/memory.
1665  */
1666 static char *
1667 sfc_strdup(const char *str)
1668 {
1669         size_t size;
1670         char *copy;
1671
1672         if (str == NULL)
1673                 return NULL;
1674
1675         size = strlen(str) + 1;
1676         copy = rte_malloc(__func__, size, 0);
1677         if (copy != NULL)
1678                 rte_memcpy(copy, str, size);
1679
1680         return copy;
1681 }
1682
1683 static int
1684 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1685 {
1686         struct sfc_adapter *sa = dev->data->dev_private;
1687         unsigned int avail_caps = 0;
1688         const char *rx_name = NULL;
1689         const char *tx_name = NULL;
1690         int rc;
1691
1692         switch (sa->family) {
1693         case EFX_FAMILY_HUNTINGTON:
1694         case EFX_FAMILY_MEDFORD:
1695                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1696                 break;
1697         default:
1698                 break;
1699         }
1700
1701         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1702                                 sfc_kvarg_string_handler, &rx_name);
1703         if (rc != 0)
1704                 goto fail_kvarg_rx_datapath;
1705
1706         if (rx_name != NULL) {
1707                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1708                 if (sa->dp_rx == NULL) {
1709                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1710                         rc = ENOENT;
1711                         goto fail_dp_rx;
1712                 }
1713                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1714                         sfc_err(sa,
1715                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1716                                 rx_name);
1717                         rc = EINVAL;
1718                         goto fail_dp_rx_caps;
1719                 }
1720         } else {
1721                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1722                 if (sa->dp_rx == NULL) {
1723                         sfc_err(sa, "Rx datapath by caps %#x not found",
1724                                 avail_caps);
1725                         rc = ENOENT;
1726                         goto fail_dp_rx;
1727                 }
1728         }
1729
1730         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1731         if (sa->dp_rx_name == NULL) {
1732                 rc = ENOMEM;
1733                 goto fail_dp_rx_name;
1734         }
1735
1736         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1737
1738         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1739
1740         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1741                                 sfc_kvarg_string_handler, &tx_name);
1742         if (rc != 0)
1743                 goto fail_kvarg_tx_datapath;
1744
1745         if (tx_name != NULL) {
1746                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1747                 if (sa->dp_tx == NULL) {
1748                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1749                         rc = ENOENT;
1750                         goto fail_dp_tx;
1751                 }
1752                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1753                         sfc_err(sa,
1754                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1755                                 tx_name);
1756                         rc = EINVAL;
1757                         goto fail_dp_tx_caps;
1758                 }
1759         } else {
1760                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1761                 if (sa->dp_tx == NULL) {
1762                         sfc_err(sa, "Tx datapath by caps %#x not found",
1763                                 avail_caps);
1764                         rc = ENOENT;
1765                         goto fail_dp_tx;
1766                 }
1767         }
1768
1769         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1770         if (sa->dp_tx_name == NULL) {
1771                 rc = ENOMEM;
1772                 goto fail_dp_tx_name;
1773         }
1774
1775         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1776
1777         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1778
1779         dev->dev_ops = &sfc_eth_dev_ops;
1780
1781         return 0;
1782
1783 fail_dp_tx_name:
1784 fail_dp_tx_caps:
1785         sa->dp_tx = NULL;
1786
1787 fail_dp_tx:
1788 fail_kvarg_tx_datapath:
1789         rte_free(sa->dp_rx_name);
1790         sa->dp_rx_name = NULL;
1791
1792 fail_dp_rx_name:
1793 fail_dp_rx_caps:
1794         sa->dp_rx = NULL;
1795
1796 fail_dp_rx:
1797 fail_kvarg_rx_datapath:
1798         return rc;
1799 }
1800
1801 static void
1802 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1803 {
1804         struct sfc_adapter *sa = dev->data->dev_private;
1805
1806         dev->dev_ops = NULL;
1807         dev->rx_pkt_burst = NULL;
1808         dev->tx_pkt_burst = NULL;
1809
1810         rte_free(sa->dp_tx_name);
1811         sa->dp_tx_name = NULL;
1812         sa->dp_tx = NULL;
1813
1814         rte_free(sa->dp_rx_name);
1815         sa->dp_rx_name = NULL;
1816         sa->dp_rx = NULL;
1817 }
1818
1819 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1820         .rxq_info_get                   = sfc_rx_queue_info_get,
1821         .txq_info_get                   = sfc_tx_queue_info_get,
1822 };
1823
1824 static int
1825 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1826 {
1827         /*
1828          * Device private data has really many process-local pointers.
1829          * Below code should be extremely careful to use data located
1830          * in shared memory only.
1831          */
1832         struct sfc_adapter *sa = dev->data->dev_private;
1833         const struct sfc_dp_rx *dp_rx;
1834         const struct sfc_dp_tx *dp_tx;
1835         int rc;
1836
1837         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1838         if (dp_rx == NULL) {
1839                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1840                 rc = ENOENT;
1841                 goto fail_dp_rx;
1842         }
1843         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1844                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1845                         sa->dp_tx_name);
1846                 rc = EINVAL;
1847                 goto fail_dp_rx_multi_process;
1848         }
1849
1850         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1851         if (dp_tx == NULL) {
1852                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1853                 rc = ENOENT;
1854                 goto fail_dp_tx;
1855         }
1856         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1857                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1858                         sa->dp_tx_name);
1859                 rc = EINVAL;
1860                 goto fail_dp_tx_multi_process;
1861         }
1862
1863         dev->rx_pkt_burst = dp_rx->pkt_burst;
1864         dev->tx_pkt_burst = dp_tx->pkt_burst;
1865         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1866
1867         return 0;
1868
1869 fail_dp_tx_multi_process:
1870 fail_dp_tx:
1871 fail_dp_rx_multi_process:
1872 fail_dp_rx:
1873         return rc;
1874 }
1875
1876 static void
1877 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1878 {
1879         dev->dev_ops = NULL;
1880         dev->tx_pkt_burst = NULL;
1881         dev->rx_pkt_burst = NULL;
1882 }
1883
1884 static void
1885 sfc_register_dp(void)
1886 {
1887         /* Register once */
1888         if (TAILQ_EMPTY(&sfc_dp_head)) {
1889                 /* Prefer EF10 datapath */
1890                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1891                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1892
1893                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1894                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1895                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1896         }
1897 }
1898
1899 static int
1900 sfc_eth_dev_init(struct rte_eth_dev *dev)
1901 {
1902         struct sfc_adapter *sa = dev->data->dev_private;
1903         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1904         int rc;
1905         const efx_nic_cfg_t *encp;
1906         const struct ether_addr *from;
1907
1908         sfc_register_dp();
1909
1910         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1911                 return -sfc_eth_dev_secondary_set_ops(dev);
1912
1913         /* Required for logging */
1914         sa->pci_addr = pci_dev->addr;
1915         sa->port_id = dev->data->port_id;
1916
1917         sa->eth_dev = dev;
1918
1919         /* Copy PCI device info to the dev->data */
1920         rte_eth_copy_pci_info(dev, pci_dev);
1921
1922         rc = sfc_kvargs_parse(sa);
1923         if (rc != 0)
1924                 goto fail_kvargs_parse;
1925
1926         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1927                                 sfc_kvarg_bool_handler, &sa->debug_init);
1928         if (rc != 0)
1929                 goto fail_kvarg_debug_init;
1930
1931         sfc_log_init(sa, "entry");
1932
1933         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1934         if (dev->data->mac_addrs == NULL) {
1935                 rc = ENOMEM;
1936                 goto fail_mac_addrs;
1937         }
1938
1939         sfc_adapter_lock_init(sa);
1940         sfc_adapter_lock(sa);
1941
1942         sfc_log_init(sa, "probing");
1943         rc = sfc_probe(sa);
1944         if (rc != 0)
1945                 goto fail_probe;
1946
1947         sfc_log_init(sa, "set device ops");
1948         rc = sfc_eth_dev_set_ops(dev);
1949         if (rc != 0)
1950                 goto fail_set_ops;
1951
1952         sfc_log_init(sa, "attaching");
1953         rc = sfc_attach(sa);
1954         if (rc != 0)
1955                 goto fail_attach;
1956
1957         encp = efx_nic_cfg_get(sa->nic);
1958
1959         /*
1960          * The arguments are really reverse order in comparison to
1961          * Linux kernel. Copy from NIC config to Ethernet device data.
1962          */
1963         from = (const struct ether_addr *)(encp->enc_mac_addr);
1964         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1965
1966         sfc_adapter_unlock(sa);
1967
1968         sfc_log_init(sa, "done");
1969         return 0;
1970
1971 fail_attach:
1972         sfc_eth_dev_clear_ops(dev);
1973
1974 fail_set_ops:
1975         sfc_unprobe(sa);
1976
1977 fail_probe:
1978         sfc_adapter_unlock(sa);
1979         sfc_adapter_lock_fini(sa);
1980         rte_free(dev->data->mac_addrs);
1981         dev->data->mac_addrs = NULL;
1982
1983 fail_mac_addrs:
1984 fail_kvarg_debug_init:
1985         sfc_kvargs_cleanup(sa);
1986
1987 fail_kvargs_parse:
1988         sfc_log_init(sa, "failed %d", rc);
1989         SFC_ASSERT(rc > 0);
1990         return -rc;
1991 }
1992
1993 static int
1994 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1995 {
1996         struct sfc_adapter *sa;
1997
1998         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1999                 sfc_eth_dev_secondary_clear_ops(dev);
2000                 return 0;
2001         }
2002
2003         sa = dev->data->dev_private;
2004         sfc_log_init(sa, "entry");
2005
2006         sfc_adapter_lock(sa);
2007
2008         sfc_eth_dev_clear_ops(dev);
2009
2010         sfc_detach(sa);
2011         sfc_unprobe(sa);
2012
2013         rte_free(dev->data->mac_addrs);
2014         dev->data->mac_addrs = NULL;
2015
2016         sfc_kvargs_cleanup(sa);
2017
2018         sfc_adapter_unlock(sa);
2019         sfc_adapter_lock_fini(sa);
2020
2021         sfc_log_init(sa, "done");
2022
2023         /* Required for logging, so cleanup last */
2024         sa->eth_dev = NULL;
2025         return 0;
2026 }
2027
2028 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2029         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2030         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2031         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2032         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2033         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2034         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2035         { .vendor_id = 0 /* sentinel */ }
2036 };
2037
2038 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2039         struct rte_pci_device *pci_dev)
2040 {
2041         return rte_eth_dev_pci_generic_probe(pci_dev,
2042                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2043 }
2044
2045 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2046 {
2047         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2048 }
2049
2050 static struct rte_pci_driver sfc_efx_pmd = {
2051         .id_table = pci_id_sfc_efx_map,
2052         .drv_flags =
2053                 RTE_PCI_DRV_INTR_LSC |
2054                 RTE_PCI_DRV_NEED_MAPPING,
2055         .probe = sfc_eth_dev_pci_probe,
2056         .remove = sfc_eth_dev_pci_remove,
2057 };
2058
2059 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2060 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2061 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2062 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2063         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2064         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2065         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2066         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2067         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2068         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);