drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
   4  * All rights reserved.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <rte_ethdev_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16
  17 #include "efx.h"
  18
  19 #include "sfc.h"
  20 #include "sfc_debug.h"
  21 #include "sfc_log.h"
  22 #include "sfc_kvargs.h"
  23 #include "sfc_ev.h"
  24 #include "sfc_rx.h"
  25 #include "sfc_tx.h"
  26 #include "sfc_flow.h"
  27 #include "sfc_dp.h"
  28 #include "sfc_dp_rx.h"
  29
  30 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         uint64_t txq_offloads_def = 0;
  87
  88         sfc_log_init(sa, "entry");
  89
  90         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
  91         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  92
  93         /* Autonegotiation may be disabled */
  94         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  95         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  96                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  97         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
  98                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
  99         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 100                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 101
 102         dev_info->max_rx_queues = sa->rxq_max;
 103         dev_info->max_tx_queues = sa->txq_max;
 104
 105         /* By default packets are dropped if no descriptors are available */
 106         dev_info->default_rxconf.rx_drop_en = 1;
 107
 108         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 109
 110         /*
 111          * rx_offload_capa includes both device and queue offloads since
 112          * the latter may be requested on a per device basis which makes
 113          * sense when some offloads are needed to be set on all queues.
 114          */
 115         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 116                                     dev_info->rx_queue_offload_capa;
 117
 118         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 119
 120         /*
 121          * tx_offload_capa includes both device and queue offloads since
 122          * the latter may be requested on a per device basis which makes
 123          * sense when some offloads are needed to be set on all queues.
 124          */
 125         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 126                                     dev_info->tx_queue_offload_capa;
 127
 128         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 129                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 130
 131         dev_info->default_txconf.offloads |= txq_offloads_def;
 132
 133         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 134         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 135             !encp->enc_hw_tx_insert_vlan_enabled)
 136                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 137
 138         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 139                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 140
 141         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 142                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 143
 144         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 145                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 146
 147 #if EFSYS_OPT_RX_SCALE
 148         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 149                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 150                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 151                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 152         }
 153 #endif
 154
 155         /* Initialize to hardware limits */
 156         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 157         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 158         /* The RXQ hardware requires that the descriptor count is a power
 159          * of 2, but rx_desc_lim cannot properly describe that constraint.
 160          */
 161         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 162
 163         /* Initialize to hardware limits */
 164         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 165         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 166         /*
 167          * The TXQ hardware requires that the descriptor count is a power
 168          * of 2, but tx_desc_lim cannot properly describe that constraint
 169          */
 170         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 171
 172         if (sa->dp_rx->get_dev_info != NULL)
 173                 sa->dp_rx->get_dev_info(dev_info);
 174         if (sa->dp_tx->get_dev_info != NULL)
 175                 sa->dp_tx->get_dev_info(dev_info);
 176 }
 177
 178 static const uint32_t *
 179 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 180 {
 181         struct sfc_adapter *sa = dev->data->dev_private;
 182         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 183         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 184
 185         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 186 }
 187
 188 static int
 189 sfc_dev_configure(struct rte_eth_dev *dev)
 190 {
 191         struct rte_eth_dev_data *dev_data = dev->data;
 192         struct sfc_adapter *sa = dev_data->dev_private;
 193         int rc;
 194
 195         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 196                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 197
 198         sfc_adapter_lock(sa);
 199         switch (sa->state) {
 200         case SFC_ADAPTER_CONFIGURED:
 201                 /* FALLTHROUGH */
 202         case SFC_ADAPTER_INITIALIZED:
 203                 rc = sfc_configure(sa);
 204                 break;
 205         default:
 206                 sfc_err(sa, "unexpected adapter state %u to configure",
 207                         sa->state);
 208                 rc = EINVAL;
 209                 break;
 210         }
 211         sfc_adapter_unlock(sa);
 212
 213         sfc_log_init(sa, "done %d", rc);
 214         SFC_ASSERT(rc >= 0);
 215         return -rc;
 216 }
 217
 218 static int
 219 sfc_dev_start(struct rte_eth_dev *dev)
 220 {
 221         struct sfc_adapter *sa = dev->data->dev_private;
 222         int rc;
 223
 224         sfc_log_init(sa, "entry");
 225
 226         sfc_adapter_lock(sa);
 227         rc = sfc_start(sa);
 228         sfc_adapter_unlock(sa);
 229
 230         sfc_log_init(sa, "done %d", rc);
 231         SFC_ASSERT(rc >= 0);
 232         return -rc;
 233 }
 234
 235 static int
 236 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 237 {
 238         struct sfc_adapter *sa = dev->data->dev_private;
 239         struct rte_eth_link current_link;
 240         int ret;
 241
 242         sfc_log_init(sa, "entry");
 243
 244         if (sa->state != SFC_ADAPTER_STARTED) {
 245                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 246         } else if (wait_to_complete) {
 247                 efx_link_mode_t link_mode;
 248
 249                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 250                         link_mode = EFX_LINK_UNKNOWN;
 251                 sfc_port_link_mode_to_info(link_mode, &current_link);
 252
 253         } else {
 254                 sfc_ev_mgmt_qpoll(sa);
 255                 rte_eth_linkstatus_get(dev, &current_link);
 256         }
 257
 258         ret = rte_eth_linkstatus_set(dev, &current_link);
 259         if (ret == 0)
 260                 sfc_info(sa, "Link status is %s",
 261                          current_link.link_status ? "UP" : "DOWN");
 262         return ret;
 263 }
 264
 265 static void
 266 sfc_dev_stop(struct rte_eth_dev *dev)
 267 {
 268         struct sfc_adapter *sa = dev->data->dev_private;
 269
 270         sfc_log_init(sa, "entry");
 271
 272         sfc_adapter_lock(sa);
 273         sfc_stop(sa);
 274         sfc_adapter_unlock(sa);
 275
 276         sfc_log_init(sa, "done");
 277 }
 278
 279 static int
 280 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 281 {
 282         struct sfc_adapter *sa = dev->data->dev_private;
 283         int rc;
 284
 285         sfc_log_init(sa, "entry");
 286
 287         sfc_adapter_lock(sa);
 288         rc = sfc_start(sa);
 289         sfc_adapter_unlock(sa);
 290
 291         SFC_ASSERT(rc >= 0);
 292         return -rc;
 293 }
 294
 295 static int
 296 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 297 {
 298         struct sfc_adapter *sa = dev->data->dev_private;
 299
 300         sfc_log_init(sa, "entry");
 301
 302         sfc_adapter_lock(sa);
 303         sfc_stop(sa);
 304         sfc_adapter_unlock(sa);
 305
 306         return 0;
 307 }
 308
 309 static void
 310 sfc_dev_close(struct rte_eth_dev *dev)
 311 {
 312         struct sfc_adapter *sa = dev->data->dev_private;
 313
 314         sfc_log_init(sa, "entry");
 315
 316         sfc_adapter_lock(sa);
 317         switch (sa->state) {
 318         case SFC_ADAPTER_STARTED:
 319                 sfc_stop(sa);
 320                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 321                 /* FALLTHROUGH */
 322         case SFC_ADAPTER_CONFIGURED:
 323                 sfc_close(sa);
 324                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 325                 /* FALLTHROUGH */
 326         case SFC_ADAPTER_INITIALIZED:
 327                 break;
 328         default:
 329                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 330                 break;
 331         }
 332         sfc_adapter_unlock(sa);
 333
 334         sfc_log_init(sa, "done");
 335 }
 336
 337 static void
 338 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 339                    boolean_t enabled)
 340 {
 341         struct sfc_port *port;
 342         boolean_t *toggle;
 343         struct sfc_adapter *sa = dev->data->dev_private;
 344         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 345         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 346
 347         sfc_adapter_lock(sa);
 348
 349         port = &sa->port;
 350         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 351
 352         if (*toggle != enabled) {
 353                 *toggle = enabled;
 354
 355                 if (port->isolated) {
 356                         sfc_warn(sa, "isolated mode is active on the port");
 357                         sfc_warn(sa, "the change is to be applied on the next "
 358                                      "start provided that isolated mode is "
 359                                      "disabled prior the next start");
 360                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 361                            (sfc_set_rx_mode(sa) != 0)) {
 362                         *toggle = !(enabled);
 363                         sfc_warn(sa, "Failed to %s %s mode",
 364                                  ((enabled) ? "enable" : "disable"), desc);
 365                 }
 366         }
 367
 368         sfc_adapter_unlock(sa);
 369 }
 370
 371 static void
 372 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 373 {
 374         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 375 }
 376
 377 static void
 378 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 379 {
 380         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 381 }
 382
 383 static void
 384 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 385 {
 386         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 387 }
 388
 389 static void
 390 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 391 {
 392         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 393 }
 394
 395 static int
 396 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 397                    uint16_t nb_rx_desc, unsigned int socket_id,
 398                    const struct rte_eth_rxconf *rx_conf,
 399                    struct rte_mempool *mb_pool)
 400 {
 401         struct sfc_adapter *sa = dev->data->dev_private;
 402         int rc;
 403
 404         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 405                      rx_queue_id, nb_rx_desc, socket_id);
 406
 407         sfc_adapter_lock(sa);
 408
 409         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 410                           rx_conf, mb_pool);
 411         if (rc != 0)
 412                 goto fail_rx_qinit;
 413
 414         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 415
 416         sfc_adapter_unlock(sa);
 417
 418         return 0;
 419
 420 fail_rx_qinit:
 421         sfc_adapter_unlock(sa);
 422         SFC_ASSERT(rc > 0);
 423         return -rc;
 424 }
 425
 426 static void
 427 sfc_rx_queue_release(void *queue)
 428 {
 429         struct sfc_dp_rxq *dp_rxq = queue;
 430         struct sfc_rxq *rxq;
 431         struct sfc_adapter *sa;
 432         unsigned int sw_index;
 433
 434         if (dp_rxq == NULL)
 435                 return;
 436
 437         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 438         sa = rxq->evq->sa;
 439         sfc_adapter_lock(sa);
 440
 441         sw_index = sfc_rxq_sw_index(rxq);
 442
 443         sfc_log_init(sa, "RxQ=%u", sw_index);
 444
 445         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 446
 447         sfc_rx_qfini(sa, sw_index);
 448
 449         sfc_adapter_unlock(sa);
 450 }
 451
 452 static int
 453 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 454                    uint16_t nb_tx_desc, unsigned int socket_id,
 455                    const struct rte_eth_txconf *tx_conf)
 456 {
 457         struct sfc_adapter *sa = dev->data->dev_private;
 458         int rc;
 459
 460         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 461                      tx_queue_id, nb_tx_desc, socket_id);
 462
 463         sfc_adapter_lock(sa);
 464
 465         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 466         if (rc != 0)
 467                 goto fail_tx_qinit;
 468
 469         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 470
 471         sfc_adapter_unlock(sa);
 472         return 0;
 473
 474 fail_tx_qinit:
 475         sfc_adapter_unlock(sa);
 476         SFC_ASSERT(rc > 0);
 477         return -rc;
 478 }
 479
 480 static void
 481 sfc_tx_queue_release(void *queue)
 482 {
 483         struct sfc_dp_txq *dp_txq = queue;
 484         struct sfc_txq *txq;
 485         unsigned int sw_index;
 486         struct sfc_adapter *sa;
 487
 488         if (dp_txq == NULL)
 489                 return;
 490
 491         txq = sfc_txq_by_dp_txq(dp_txq);
 492         sw_index = sfc_txq_sw_index(txq);
 493
 494         SFC_ASSERT(txq->evq != NULL);
 495         sa = txq->evq->sa;
 496
 497         sfc_log_init(sa, "TxQ = %u", sw_index);
 498
 499         sfc_adapter_lock(sa);
 500
 501         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 502         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 503
 504         sfc_tx_qfini(sa, sw_index);
 505
 506         sfc_adapter_unlock(sa);
 507 }
 508
 509 static int
 510 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 511 {
 512         struct sfc_adapter *sa = dev->data->dev_private;
 513         struct sfc_port *port = &sa->port;
 514         uint64_t *mac_stats;
 515         int ret;
 516
 517         rte_spinlock_lock(&port->mac_stats_lock);
 518
 519         ret = sfc_port_update_mac_stats(sa);
 520         if (ret != 0)
 521                 goto unlock;
 522
 523         mac_stats = port->mac_stats_buf;
 524
 525         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 526                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 527                 stats->ipackets =
 528                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 529                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 530                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 531                 stats->opackets =
 532                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 533                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 534                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 535                 stats->ibytes =
 536                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 537                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 538                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 539                 stats->obytes =
 540                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 541                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 542                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 543                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 544                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 545                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 546         } else {
 547                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 548                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 549                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 550                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 551                 /*
 552                  * Take into account stats which are whenever supported
 553                  * on EF10. If some stat is not supported by current
 554                  * firmware variant or HW revision, it is guaranteed
 555                  * to be zero in mac_stats.
 556                  */
 557                 stats->imissed =
 558                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 559                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 560                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 561                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 562                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 563                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 564                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 565                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 566                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 567                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 568                 stats->ierrors =
 569                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 570                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 571                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 572                 /* no oerrors counters supported on EF10 */
 573         }
 574
 575 unlock:
 576         rte_spinlock_unlock(&port->mac_stats_lock);
 577         SFC_ASSERT(ret >= 0);
 578         return -ret;
 579 }
 580
 581 static void
 582 sfc_stats_reset(struct rte_eth_dev *dev)
 583 {
 584         struct sfc_adapter *sa = dev->data->dev_private;
 585         struct sfc_port *port = &sa->port;
 586         int rc;
 587
 588         if (sa->state != SFC_ADAPTER_STARTED) {
 589                 /*
 590                  * The operation cannot be done if port is not started; it
 591                  * will be scheduled to be done during the next port start
 592                  */
 593                 port->mac_stats_reset_pending = B_TRUE;
 594                 return;
 595         }
 596
 597         rc = sfc_port_reset_mac_stats(sa);
 598         if (rc != 0)
 599                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 600 }
 601
 602 static int
 603 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 604                unsigned int xstats_count)
 605 {
 606         struct sfc_adapter *sa = dev->data->dev_private;
 607         struct sfc_port *port = &sa->port;
 608         uint64_t *mac_stats;
 609         int rc;
 610         unsigned int i;
 611         int nstats = 0;
 612
 613         rte_spinlock_lock(&port->mac_stats_lock);
 614
 615         rc = sfc_port_update_mac_stats(sa);
 616         if (rc != 0) {
 617                 SFC_ASSERT(rc > 0);
 618                 nstats = -rc;
 619                 goto unlock;
 620         }
 621
 622         mac_stats = port->mac_stats_buf;
 623
 624         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 625                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 626                         if (xstats != NULL && nstats < (int)xstats_count) {
 627                                 xstats[nstats].id = nstats;
 628                                 xstats[nstats].value = mac_stats[i];
 629                         }
 630                         nstats++;
 631                 }
 632         }
 633
 634 unlock:
 635         rte_spinlock_unlock(&port->mac_stats_lock);
 636
 637         return nstats;
 638 }
 639
 640 static int
 641 sfc_xstats_get_names(struct rte_eth_dev *dev,
 642                      struct rte_eth_xstat_name *xstats_names,
 643                      unsigned int xstats_count)
 644 {
 645         struct sfc_adapter *sa = dev->data->dev_private;
 646         struct sfc_port *port = &sa->port;
 647         unsigned int i;
 648         unsigned int nstats = 0;
 649
 650         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 651                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 652                         if (xstats_names != NULL && nstats < xstats_count)
 653                                 strncpy(xstats_names[nstats].name,
 654                                         efx_mac_stat_name(sa->nic, i),
 655                                         sizeof(xstats_names[0].name));
 656                         nstats++;
 657                 }
 658         }
 659
 660         return nstats;
 661 }
 662
 663 static int
 664 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 665                      uint64_t *values, unsigned int n)
 666 {
 667         struct sfc_adapter *sa = dev->data->dev_private;
 668         struct sfc_port *port = &sa->port;
 669         uint64_t *mac_stats;
 670         unsigned int nb_supported = 0;
 671         unsigned int nb_written = 0;
 672         unsigned int i;
 673         int ret;
 674         int rc;
 675
 676         if (unlikely(values == NULL) ||
 677             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 678                 return port->mac_stats_nb_supported;
 679
 680         rte_spinlock_lock(&port->mac_stats_lock);
 681
 682         rc = sfc_port_update_mac_stats(sa);
 683         if (rc != 0) {
 684                 SFC_ASSERT(rc > 0);
 685                 ret = -rc;
 686                 goto unlock;
 687         }
 688
 689         mac_stats = port->mac_stats_buf;
 690
 691         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 692                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 693                         continue;
 694
 695                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 696                         values[nb_written++] = mac_stats[i];
 697
 698                 ++nb_supported;
 699         }
 700
 701         ret = nb_written;
 702
 703 unlock:
 704         rte_spinlock_unlock(&port->mac_stats_lock);
 705
 706         return ret;
 707 }
 708
 709 static int
 710 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 711                            struct rte_eth_xstat_name *xstats_names,
 712                            const uint64_t *ids, unsigned int size)
 713 {
 714         struct sfc_adapter *sa = dev->data->dev_private;
 715         struct sfc_port *port = &sa->port;
 716         unsigned int nb_supported = 0;
 717         unsigned int nb_written = 0;
 718         unsigned int i;
 719
 720         if (unlikely(xstats_names == NULL) ||
 721             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 722                 return port->mac_stats_nb_supported;
 723
 724         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 725                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 726                         continue;
 727
 728                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 729                         char *name = xstats_names[nb_written++].name;
 730
 731                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 732                                 sizeof(xstats_names[0].name));
 733                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 734                 }
 735
 736                 ++nb_supported;
 737         }
 738
 739         return nb_written;
 740 }
 741
 742 static int
 743 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 744 {
 745         struct sfc_adapter *sa = dev->data->dev_private;
 746         unsigned int wanted_fc, link_fc;
 747
 748         memset(fc_conf, 0, sizeof(*fc_conf));
 749
 750         sfc_adapter_lock(sa);
 751
 752         if (sa->state == SFC_ADAPTER_STARTED)
 753                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 754         else
 755                 link_fc = sa->port.flow_ctrl;
 756
 757         switch (link_fc) {
 758         case 0:
 759                 fc_conf->mode = RTE_FC_NONE;
 760                 break;
 761         case EFX_FCNTL_RESPOND:
 762                 fc_conf->mode = RTE_FC_RX_PAUSE;
 763                 break;
 764         case EFX_FCNTL_GENERATE:
 765                 fc_conf->mode = RTE_FC_TX_PAUSE;
 766                 break;
 767         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 768                 fc_conf->mode = RTE_FC_FULL;
 769                 break;
 770         default:
 771                 sfc_err(sa, "%s: unexpected flow control value %#x",
 772                         __func__, link_fc);
 773         }
 774
 775         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 776
 777         sfc_adapter_unlock(sa);
 778
 779         return 0;
 780 }
 781
 782 static int
 783 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 784 {
 785         struct sfc_adapter *sa = dev->data->dev_private;
 786         struct sfc_port *port = &sa->port;
 787         unsigned int fcntl;
 788         int rc;
 789
 790         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 791             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 792             fc_conf->mac_ctrl_frame_fwd != 0) {
 793                 sfc_err(sa, "unsupported flow control settings specified");
 794                 rc = EINVAL;
 795                 goto fail_inval;
 796         }
 797
 798         switch (fc_conf->mode) {
 799         case RTE_FC_NONE:
 800                 fcntl = 0;
 801                 break;
 802         case RTE_FC_RX_PAUSE:
 803                 fcntl = EFX_FCNTL_RESPOND;
 804                 break;
 805         case RTE_FC_TX_PAUSE:
 806                 fcntl = EFX_FCNTL_GENERATE;
 807                 break;
 808         case RTE_FC_FULL:
 809                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 810                 break;
 811         default:
 812                 rc = EINVAL;
 813                 goto fail_inval;
 814         }
 815
 816         sfc_adapter_lock(sa);
 817
 818         if (sa->state == SFC_ADAPTER_STARTED) {
 819                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 820                 if (rc != 0)
 821                         goto fail_mac_fcntl_set;
 822         }
 823
 824         port->flow_ctrl = fcntl;
 825         port->flow_ctrl_autoneg = fc_conf->autoneg;
 826
 827         sfc_adapter_unlock(sa);
 828
 829         return 0;
 830
 831 fail_mac_fcntl_set:
 832         sfc_adapter_unlock(sa);
 833 fail_inval:
 834         SFC_ASSERT(rc > 0);
 835         return -rc;
 836 }
 837
 838 static int
 839 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 840 {
 841         struct sfc_adapter *sa = dev->data->dev_private;
 842         size_t pdu = EFX_MAC_PDU(mtu);
 843         size_t old_pdu;
 844         int rc;
 845
 846         sfc_log_init(sa, "mtu=%u", mtu);
 847
 848         rc = EINVAL;
 849         if (pdu < EFX_MAC_PDU_MIN) {
 850                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 851                         (unsigned int)mtu, (unsigned int)pdu,
 852                         EFX_MAC_PDU_MIN);
 853                 goto fail_inval;
 854         }
 855         if (pdu > EFX_MAC_PDU_MAX) {
 856                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 857                         (unsigned int)mtu, (unsigned int)pdu,
 858                         EFX_MAC_PDU_MAX);
 859                 goto fail_inval;
 860         }
 861
 862         sfc_adapter_lock(sa);
 863
 864         if (pdu != sa->port.pdu) {
 865                 if (sa->state == SFC_ADAPTER_STARTED) {
 866                         sfc_stop(sa);
 867
 868                         old_pdu = sa->port.pdu;
 869                         sa->port.pdu = pdu;
 870                         rc = sfc_start(sa);
 871                         if (rc != 0)
 872                                 goto fail_start;
 873                 } else {
 874                         sa->port.pdu = pdu;
 875                 }
 876         }
 877
 878         /*
 879          * The driver does not use it, but other PMDs update jumbo_frame
 880          * flag and max_rx_pkt_len when MTU is set.
 881          */
 882         if (mtu > ETHER_MAX_LEN) {
 883                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 884
 885                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 886                 rxmode->jumbo_frame = 1;
 887         }
 888
 889         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 890
 891         sfc_adapter_unlock(sa);
 892
 893         sfc_log_init(sa, "done");
 894         return 0;
 895
 896 fail_start:
 897         sa->port.pdu = old_pdu;
 898         if (sfc_start(sa) != 0)
 899                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 900                         "PDU max size - port is stopped",
 901                         (unsigned int)pdu, (unsigned int)old_pdu);
 902         sfc_adapter_unlock(sa);
 903
 904 fail_inval:
 905         sfc_log_init(sa, "failed %d", rc);
 906         SFC_ASSERT(rc > 0);
 907         return -rc;
 908 }
 909 static void
 910 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 911 {
 912         struct sfc_adapter *sa = dev->data->dev_private;
 913         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 914         struct sfc_port *port = &sa->port;
 915         int rc;
 916
 917         sfc_adapter_lock(sa);
 918
 919         /*
 920          * Copy the address to the device private data so that
 921          * it could be recalled in the case of adapter restart.
 922          */
 923         ether_addr_copy(mac_addr, &port->default_mac_addr);
 924
 925         if (port->isolated) {
 926                 sfc_err(sa, "isolated mode is active on the port");
 927                 sfc_err(sa, "will not set MAC address");
 928                 goto unlock;
 929         }
 930
 931         if (sa->state != SFC_ADAPTER_STARTED) {
 932                 sfc_info(sa, "the port is not started");
 933                 sfc_info(sa, "the new MAC address will be set on port start");
 934
 935                 goto unlock;
 936         }
 937
 938         if (encp->enc_allow_set_mac_with_installed_filters) {
 939                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 940                 if (rc != 0) {
 941                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 942                         goto unlock;
 943                 }
 944
 945                 /*
 946                  * Changing the MAC address by means of MCDI request
 947                  * has no effect on received traffic, therefore
 948                  * we also need to update unicast filters
 949                  */
 950                 rc = sfc_set_rx_mode(sa);
 951                 if (rc != 0)
 952                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 953         } else {
 954                 sfc_warn(sa, "cannot set MAC address with filters installed");
 955                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 956                 sfc_warn(sa, "(some traffic may be dropped)");
 957
 958                 /*
 959                  * Since setting MAC address with filters installed is not
 960                  * allowed on the adapter, the new MAC address will be set
 961                  * by means of adapter restart. sfc_start() shall retrieve
 962                  * the new address from the device private data and set it.
 963                  */
 964                 sfc_stop(sa);
 965                 rc = sfc_start(sa);
 966                 if (rc != 0)
 967                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 968         }
 969
 970 unlock:
 971         /*
 972          * In the case of failure sa->port->default_mac_addr does not
 973          * need rollback since no error code is returned, and the upper
 974          * API will anyway update the external MAC address storage.
 975          * To be consistent with that new value it is better to keep
 976          * the device private value the same.
 977          */
 978         sfc_adapter_unlock(sa);
 979 }
 980
 981
 982 static int
 983 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 984                      uint32_t nb_mc_addr)
 985 {
 986         struct sfc_adapter *sa = dev->data->dev_private;
 987         struct sfc_port *port = &sa->port;
 988         uint8_t *mc_addrs = port->mcast_addrs;
 989         int rc;
 990         unsigned int i;
 991
 992         if (port->isolated) {
 993                 sfc_err(sa, "isolated mode is active on the port");
 994                 sfc_err(sa, "will not set multicast address list");
 995                 return -ENOTSUP;
 996         }
 997
 998         if (mc_addrs == NULL)
 999                 return -ENOBUFS;
1000
1001         if (nb_mc_addr > port->max_mcast_addrs) {
1002                 sfc_err(sa, "too many multicast addresses: %u > %u",
1003                          nb_mc_addr, port->max_mcast_addrs);
1004                 return -EINVAL;
1005         }
1006
1007         for (i = 0; i < nb_mc_addr; ++i) {
1008                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1009                                  EFX_MAC_ADDR_LEN);
1010                 mc_addrs += EFX_MAC_ADDR_LEN;
1011         }
1012
1013         port->nb_mcast_addrs = nb_mc_addr;
1014
1015         if (sa->state != SFC_ADAPTER_STARTED)
1016                 return 0;
1017
1018         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1019                                         port->nb_mcast_addrs);
1020         if (rc != 0)
1021                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1022
1023         SFC_ASSERT(rc > 0);
1024         return -rc;
1025 }
1026
1027 /*
1028  * The function is used by the secondary process as well. It must not
1029  * use any process-local pointers from the adapter data.
1030  */
1031 static void
1032 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1033                       struct rte_eth_rxq_info *qinfo)
1034 {
1035         struct sfc_adapter *sa = dev->data->dev_private;
1036         struct sfc_rxq_info *rxq_info;
1037         struct sfc_rxq *rxq;
1038
1039         sfc_adapter_lock(sa);
1040
1041         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1042
1043         rxq_info = &sa->rxq_info[rx_queue_id];
1044         rxq = rxq_info->rxq;
1045         SFC_ASSERT(rxq != NULL);
1046
1047         qinfo->mp = rxq->refill_mb_pool;
1048         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1049         qinfo->conf.rx_drop_en = 1;
1050         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1051         qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1052                                DEV_RX_OFFLOAD_UDP_CKSUM |
1053                                DEV_RX_OFFLOAD_TCP_CKSUM;
1054         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1055                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1056                 qinfo->scattered_rx = 1;
1057         }
1058         qinfo->nb_desc = rxq_info->entries;
1059
1060         sfc_adapter_unlock(sa);
1061 }
1062
1063 /*
1064  * The function is used by the secondary process as well. It must not
1065  * use any process-local pointers from the adapter data.
1066  */
1067 static void
1068 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1069                       struct rte_eth_txq_info *qinfo)
1070 {
1071         struct sfc_adapter *sa = dev->data->dev_private;
1072         struct sfc_txq_info *txq_info;
1073
1074         sfc_adapter_lock(sa);
1075
1076         SFC_ASSERT(tx_queue_id < sa->txq_count);
1077
1078         txq_info = &sa->txq_info[tx_queue_id];
1079         SFC_ASSERT(txq_info->txq != NULL);
1080
1081         memset(qinfo, 0, sizeof(*qinfo));
1082
1083         qinfo->conf.txq_flags = txq_info->txq->flags;
1084         qinfo->conf.offloads = txq_info->txq->offloads;
1085         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1086         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1087         qinfo->nb_desc = txq_info->entries;
1088
1089         sfc_adapter_unlock(sa);
1090 }
1091
1092 static uint32_t
1093 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1094 {
1095         struct sfc_adapter *sa = dev->data->dev_private;
1096
1097         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1098
1099         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1100 }
1101
1102 static int
1103 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1104 {
1105         struct sfc_dp_rxq *dp_rxq = queue;
1106
1107         return sfc_rx_qdesc_done(dp_rxq, offset);
1108 }
1109
1110 static int
1111 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1112 {
1113         struct sfc_dp_rxq *dp_rxq = queue;
1114         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1115
1116         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1117 }
1118
1119 static int
1120 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1121 {
1122         struct sfc_dp_txq *dp_txq = queue;
1123         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1124
1125         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1126 }
1127
1128 static int
1129 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1130 {
1131         struct sfc_adapter *sa = dev->data->dev_private;
1132         int rc;
1133
1134         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1135
1136         sfc_adapter_lock(sa);
1137
1138         rc = EINVAL;
1139         if (sa->state != SFC_ADAPTER_STARTED)
1140                 goto fail_not_started;
1141
1142         rc = sfc_rx_qstart(sa, rx_queue_id);
1143         if (rc != 0)
1144                 goto fail_rx_qstart;
1145
1146         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1147
1148         sfc_adapter_unlock(sa);
1149
1150         return 0;
1151
1152 fail_rx_qstart:
1153 fail_not_started:
1154         sfc_adapter_unlock(sa);
1155         SFC_ASSERT(rc > 0);
1156         return -rc;
1157 }
1158
1159 static int
1160 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1161 {
1162         struct sfc_adapter *sa = dev->data->dev_private;
1163
1164         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1165
1166         sfc_adapter_lock(sa);
1167         sfc_rx_qstop(sa, rx_queue_id);
1168
1169         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1170
1171         sfc_adapter_unlock(sa);
1172
1173         return 0;
1174 }
1175
1176 static int
1177 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1178 {
1179         struct sfc_adapter *sa = dev->data->dev_private;
1180         int rc;
1181
1182         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1183
1184         sfc_adapter_lock(sa);
1185
1186         rc = EINVAL;
1187         if (sa->state != SFC_ADAPTER_STARTED)
1188                 goto fail_not_started;
1189
1190         rc = sfc_tx_qstart(sa, tx_queue_id);
1191         if (rc != 0)
1192                 goto fail_tx_qstart;
1193
1194         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1195
1196         sfc_adapter_unlock(sa);
1197         return 0;
1198
1199 fail_tx_qstart:
1200
1201 fail_not_started:
1202         sfc_adapter_unlock(sa);
1203         SFC_ASSERT(rc > 0);
1204         return -rc;
1205 }
1206
1207 static int
1208 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1209 {
1210         struct sfc_adapter *sa = dev->data->dev_private;
1211
1212         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1213
1214         sfc_adapter_lock(sa);
1215
1216         sfc_tx_qstop(sa, tx_queue_id);
1217
1218         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1219
1220         sfc_adapter_unlock(sa);
1221         return 0;
1222 }
1223
1224 static efx_tunnel_protocol_t
1225 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1226 {
1227         switch (rte_type) {
1228         case RTE_TUNNEL_TYPE_VXLAN:
1229                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1230         case RTE_TUNNEL_TYPE_GENEVE:
1231                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1232         default:
1233                 return EFX_TUNNEL_NPROTOS;
1234         }
1235 }
1236
1237 enum sfc_udp_tunnel_op_e {
1238         SFC_UDP_TUNNEL_ADD_PORT,
1239         SFC_UDP_TUNNEL_DEL_PORT,
1240 };
1241
1242 static int
1243 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1244                       struct rte_eth_udp_tunnel *tunnel_udp,
1245                       enum sfc_udp_tunnel_op_e op)
1246 {
1247         struct sfc_adapter *sa = dev->data->dev_private;
1248         efx_tunnel_protocol_t tunnel_proto;
1249         int rc;
1250
1251         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1252                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1253                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1254                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1255
1256         tunnel_proto =
1257                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1258         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1259                 rc = ENOTSUP;
1260                 goto fail_bad_proto;
1261         }
1262
1263         sfc_adapter_lock(sa);
1264
1265         switch (op) {
1266         case SFC_UDP_TUNNEL_ADD_PORT:
1267                 rc = efx_tunnel_config_udp_add(sa->nic,
1268                                                tunnel_udp->udp_port,
1269                                                tunnel_proto);
1270                 break;
1271         case SFC_UDP_TUNNEL_DEL_PORT:
1272                 rc = efx_tunnel_config_udp_remove(sa->nic,
1273                                                   tunnel_udp->udp_port,
1274                                                   tunnel_proto);
1275                 break;
1276         default:
1277                 rc = EINVAL;
1278                 goto fail_bad_op;
1279         }
1280
1281         if (rc != 0)
1282                 goto fail_op;
1283
1284         if (sa->state == SFC_ADAPTER_STARTED) {
1285                 rc = efx_tunnel_reconfigure(sa->nic);
1286                 if (rc == EAGAIN) {
1287                         /*
1288                          * Configuration is accepted by FW and MC reboot
1289                          * is initiated to apply the changes. MC reboot
1290                          * will be handled in a usual way (MC reboot
1291                          * event on management event queue and adapter
1292                          * restart).
1293                          */
1294                         rc = 0;
1295                 } else if (rc != 0) {
1296                         goto fail_reconfigure;
1297                 }
1298         }
1299
1300         sfc_adapter_unlock(sa);
1301         return 0;
1302
1303 fail_reconfigure:
1304         /* Remove/restore entry since the change makes the trouble */
1305         switch (op) {
1306         case SFC_UDP_TUNNEL_ADD_PORT:
1307                 (void)efx_tunnel_config_udp_remove(sa->nic,
1308                                                    tunnel_udp->udp_port,
1309                                                    tunnel_proto);
1310                 break;
1311         case SFC_UDP_TUNNEL_DEL_PORT:
1312                 (void)efx_tunnel_config_udp_add(sa->nic,
1313                                                 tunnel_udp->udp_port,
1314                                                 tunnel_proto);
1315                 break;
1316         }
1317
1318 fail_op:
1319 fail_bad_op:
1320         sfc_adapter_unlock(sa);
1321
1322 fail_bad_proto:
1323         SFC_ASSERT(rc > 0);
1324         return -rc;
1325 }
1326
1327 static int
1328 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1329                             struct rte_eth_udp_tunnel *tunnel_udp)
1330 {
1331         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1332 }
1333
1334 static int
1335 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1336                             struct rte_eth_udp_tunnel *tunnel_udp)
1337 {
1338         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1339 }
1340
1341 #if EFSYS_OPT_RX_SCALE
1342 static int
1343 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1344                           struct rte_eth_rss_conf *rss_conf)
1345 {
1346         struct sfc_adapter *sa = dev->data->dev_private;
1347         struct sfc_port *port = &sa->port;
1348
1349         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1350                 return -ENOTSUP;
1351
1352         if (sa->rss_channels == 0)
1353                 return -EINVAL;
1354
1355         sfc_adapter_lock(sa);
1356
1357         /*
1358          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1359          * hence, conversion is done here to derive a correct set of ETH_RSS
1360          * flags which corresponds to the active EFX configuration stored
1361          * locally in 'sfc_adapter' and kept up-to-date
1362          */
1363         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1364         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1365         if (rss_conf->rss_key != NULL)
1366                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1367
1368         sfc_adapter_unlock(sa);
1369
1370         return 0;
1371 }
1372
1373 static int
1374 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1375                         struct rte_eth_rss_conf *rss_conf)
1376 {
1377         struct sfc_adapter *sa = dev->data->dev_private;
1378         struct sfc_port *port = &sa->port;
1379         unsigned int efx_hash_types;
1380         int rc = 0;
1381
1382         if (port->isolated)
1383                 return -ENOTSUP;
1384
1385         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1386                 sfc_err(sa, "RSS is not available");
1387                 return -ENOTSUP;
1388         }
1389
1390         if (sa->rss_channels == 0) {
1391                 sfc_err(sa, "RSS is not configured");
1392                 return -EINVAL;
1393         }
1394
1395         if ((rss_conf->rss_key != NULL) &&
1396             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1397                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1398                         sizeof(sa->rss_key));
1399                 return -EINVAL;
1400         }
1401
1402         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1403                 sfc_err(sa, "unsupported hash functions requested");
1404                 return -EINVAL;
1405         }
1406
1407         sfc_adapter_lock(sa);
1408
1409         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1410
1411         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1412                                    EFX_RX_HASHALG_TOEPLITZ,
1413                                    efx_hash_types, B_TRUE);
1414         if (rc != 0)
1415                 goto fail_scale_mode_set;
1416
1417         if (rss_conf->rss_key != NULL) {
1418                 if (sa->state == SFC_ADAPTER_STARTED) {
1419                         rc = efx_rx_scale_key_set(sa->nic,
1420                                                   EFX_RSS_CONTEXT_DEFAULT,
1421                                                   rss_conf->rss_key,
1422                                                   sizeof(sa->rss_key));
1423                         if (rc != 0)
1424                                 goto fail_scale_key_set;
1425                 }
1426
1427                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1428         }
1429
1430         sa->rss_hash_types = efx_hash_types;
1431
1432         sfc_adapter_unlock(sa);
1433
1434         return 0;
1435
1436 fail_scale_key_set:
1437         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1438                                   EFX_RX_HASHALG_TOEPLITZ,
1439                                   sa->rss_hash_types, B_TRUE) != 0)
1440                 sfc_err(sa, "failed to restore RSS mode");
1441
1442 fail_scale_mode_set:
1443         sfc_adapter_unlock(sa);
1444         return -rc;
1445 }
1446
1447 static int
1448 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1449                        struct rte_eth_rss_reta_entry64 *reta_conf,
1450                        uint16_t reta_size)
1451 {
1452         struct sfc_adapter *sa = dev->data->dev_private;
1453         struct sfc_port *port = &sa->port;
1454         int entry;
1455
1456         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1457                 return -ENOTSUP;
1458
1459         if (sa->rss_channels == 0)
1460                 return -EINVAL;
1461
1462         if (reta_size != EFX_RSS_TBL_SIZE)
1463                 return -EINVAL;
1464
1465         sfc_adapter_lock(sa);
1466
1467         for (entry = 0; entry < reta_size; entry++) {
1468                 int grp = entry / RTE_RETA_GROUP_SIZE;
1469                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1470
1471                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1472                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1473         }
1474
1475         sfc_adapter_unlock(sa);
1476
1477         return 0;
1478 }
1479
1480 static int
1481 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1482                         struct rte_eth_rss_reta_entry64 *reta_conf,
1483                         uint16_t reta_size)
1484 {
1485         struct sfc_adapter *sa = dev->data->dev_private;
1486         struct sfc_port *port = &sa->port;
1487         unsigned int *rss_tbl_new;
1488         uint16_t entry;
1489         int rc = 0;
1490
1491
1492         if (port->isolated)
1493                 return -ENOTSUP;
1494
1495         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1496                 sfc_err(sa, "RSS is not available");
1497                 return -ENOTSUP;
1498         }
1499
1500         if (sa->rss_channels == 0) {
1501                 sfc_err(sa, "RSS is not configured");
1502                 return -EINVAL;
1503         }
1504
1505         if (reta_size != EFX_RSS_TBL_SIZE) {
1506                 sfc_err(sa, "RETA size is wrong (should be %u)",
1507                         EFX_RSS_TBL_SIZE);
1508                 return -EINVAL;
1509         }
1510
1511         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1512         if (rss_tbl_new == NULL)
1513                 return -ENOMEM;
1514
1515         sfc_adapter_lock(sa);
1516
1517         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1518
1519         for (entry = 0; entry < reta_size; entry++) {
1520                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1521                 struct rte_eth_rss_reta_entry64 *grp;
1522
1523                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1524
1525                 if (grp->mask & (1ull << grp_idx)) {
1526                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1527                                 rc = EINVAL;
1528                                 goto bad_reta_entry;
1529                         }
1530                         rss_tbl_new[entry] = grp->reta[grp_idx];
1531                 }
1532         }
1533
1534         if (sa->state == SFC_ADAPTER_STARTED) {
1535                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1536                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1537                 if (rc != 0)
1538                         goto fail_scale_tbl_set;
1539         }
1540
1541         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1542
1543 fail_scale_tbl_set:
1544 bad_reta_entry:
1545         sfc_adapter_unlock(sa);
1546
1547         rte_free(rss_tbl_new);
1548
1549         SFC_ASSERT(rc >= 0);
1550         return -rc;
1551 }
1552 #endif
1553
1554 static int
1555 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1556                     enum rte_filter_op filter_op,
1557                     void *arg)
1558 {
1559         struct sfc_adapter *sa = dev->data->dev_private;
1560         int rc = ENOTSUP;
1561
1562         sfc_log_init(sa, "entry");
1563
1564         switch (filter_type) {
1565         case RTE_ETH_FILTER_NONE:
1566                 sfc_err(sa, "Global filters configuration not supported");
1567                 break;
1568         case RTE_ETH_FILTER_MACVLAN:
1569                 sfc_err(sa, "MACVLAN filters not supported");
1570                 break;
1571         case RTE_ETH_FILTER_ETHERTYPE:
1572                 sfc_err(sa, "EtherType filters not supported");
1573                 break;
1574         case RTE_ETH_FILTER_FLEXIBLE:
1575                 sfc_err(sa, "Flexible filters not supported");
1576                 break;
1577         case RTE_ETH_FILTER_SYN:
1578                 sfc_err(sa, "SYN filters not supported");
1579                 break;
1580         case RTE_ETH_FILTER_NTUPLE:
1581                 sfc_err(sa, "NTUPLE filters not supported");
1582                 break;
1583         case RTE_ETH_FILTER_TUNNEL:
1584                 sfc_err(sa, "Tunnel filters not supported");
1585                 break;
1586         case RTE_ETH_FILTER_FDIR:
1587                 sfc_err(sa, "Flow Director filters not supported");
1588                 break;
1589         case RTE_ETH_FILTER_HASH:
1590                 sfc_err(sa, "Hash filters not supported");
1591                 break;
1592         case RTE_ETH_FILTER_GENERIC:
1593                 if (filter_op != RTE_ETH_FILTER_GET) {
1594                         rc = EINVAL;
1595                 } else {
1596                         *(const void **)arg = &sfc_flow_ops;
1597                         rc = 0;
1598                 }
1599                 break;
1600         default:
1601                 sfc_err(sa, "Unknown filter type %u", filter_type);
1602                 break;
1603         }
1604
1605         sfc_log_init(sa, "exit: %d", -rc);
1606         SFC_ASSERT(rc >= 0);
1607         return -rc;
1608 }
1609
1610 static const struct eth_dev_ops sfc_eth_dev_ops = {
1611         .dev_configure                  = sfc_dev_configure,
1612         .dev_start                      = sfc_dev_start,
1613         .dev_stop                       = sfc_dev_stop,
1614         .dev_set_link_up                = sfc_dev_set_link_up,
1615         .dev_set_link_down              = sfc_dev_set_link_down,
1616         .dev_close                      = sfc_dev_close,
1617         .promiscuous_enable             = sfc_dev_promisc_enable,
1618         .promiscuous_disable            = sfc_dev_promisc_disable,
1619         .allmulticast_enable            = sfc_dev_allmulti_enable,
1620         .allmulticast_disable           = sfc_dev_allmulti_disable,
1621         .link_update                    = sfc_dev_link_update,
1622         .stats_get                      = sfc_stats_get,
1623         .stats_reset                    = sfc_stats_reset,
1624         .xstats_get                     = sfc_xstats_get,
1625         .xstats_reset                   = sfc_stats_reset,
1626         .xstats_get_names               = sfc_xstats_get_names,
1627         .dev_infos_get                  = sfc_dev_infos_get,
1628         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1629         .mtu_set                        = sfc_dev_set_mtu,
1630         .rx_queue_start                 = sfc_rx_queue_start,
1631         .rx_queue_stop                  = sfc_rx_queue_stop,
1632         .tx_queue_start                 = sfc_tx_queue_start,
1633         .tx_queue_stop                  = sfc_tx_queue_stop,
1634         .rx_queue_setup                 = sfc_rx_queue_setup,
1635         .rx_queue_release               = sfc_rx_queue_release,
1636         .rx_queue_count                 = sfc_rx_queue_count,
1637         .rx_descriptor_done             = sfc_rx_descriptor_done,
1638         .rx_descriptor_status           = sfc_rx_descriptor_status,
1639         .tx_descriptor_status           = sfc_tx_descriptor_status,
1640         .tx_queue_setup                 = sfc_tx_queue_setup,
1641         .tx_queue_release               = sfc_tx_queue_release,
1642         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1643         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1644         .mac_addr_set                   = sfc_mac_addr_set,
1645         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1646         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1647 #if EFSYS_OPT_RX_SCALE
1648         .reta_update                    = sfc_dev_rss_reta_update,
1649         .reta_query                     = sfc_dev_rss_reta_query,
1650         .rss_hash_update                = sfc_dev_rss_hash_update,
1651         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1652 #endif
1653         .filter_ctrl                    = sfc_dev_filter_ctrl,
1654         .set_mc_addr_list               = sfc_set_mc_addr_list,
1655         .rxq_info_get                   = sfc_rx_queue_info_get,
1656         .txq_info_get                   = sfc_tx_queue_info_get,
1657         .fw_version_get                 = sfc_fw_version_get,
1658         .xstats_get_by_id               = sfc_xstats_get_by_id,
1659         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1660 };
1661
1662 /**
1663  * Duplicate a string in potentially shared memory required for
1664  * multi-process support.
1665  *
1666  * strdup() allocates from process-local heap/memory.
1667  */
1668 static char *
1669 sfc_strdup(const char *str)
1670 {
1671         size_t size;
1672         char *copy;
1673
1674         if (str == NULL)
1675                 return NULL;
1676
1677         size = strlen(str) + 1;
1678         copy = rte_malloc(__func__, size, 0);
1679         if (copy != NULL)
1680                 rte_memcpy(copy, str, size);
1681
1682         return copy;
1683 }
1684
1685 static int
1686 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1687 {
1688         struct sfc_adapter *sa = dev->data->dev_private;
1689         unsigned int avail_caps = 0;
1690         const char *rx_name = NULL;
1691         const char *tx_name = NULL;
1692         int rc;
1693
1694         switch (sa->family) {
1695         case EFX_FAMILY_HUNTINGTON:
1696         case EFX_FAMILY_MEDFORD:
1697                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1698                 break;
1699         default:
1700                 break;
1701         }
1702
1703         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1704                                 sfc_kvarg_string_handler, &rx_name);
1705         if (rc != 0)
1706                 goto fail_kvarg_rx_datapath;
1707
1708         if (rx_name != NULL) {
1709                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1710                 if (sa->dp_rx == NULL) {
1711                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1712                         rc = ENOENT;
1713                         goto fail_dp_rx;
1714                 }
1715                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1716                         sfc_err(sa,
1717                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1718                                 rx_name);
1719                         rc = EINVAL;
1720                         goto fail_dp_rx_caps;
1721                 }
1722         } else {
1723                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1724                 if (sa->dp_rx == NULL) {
1725                         sfc_err(sa, "Rx datapath by caps %#x not found",
1726                                 avail_caps);
1727                         rc = ENOENT;
1728                         goto fail_dp_rx;
1729                 }
1730         }
1731
1732         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1733         if (sa->dp_rx_name == NULL) {
1734                 rc = ENOMEM;
1735                 goto fail_dp_rx_name;
1736         }
1737
1738         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1739
1740         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1741
1742         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1743                                 sfc_kvarg_string_handler, &tx_name);
1744         if (rc != 0)
1745                 goto fail_kvarg_tx_datapath;
1746
1747         if (tx_name != NULL) {
1748                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1749                 if (sa->dp_tx == NULL) {
1750                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1751                         rc = ENOENT;
1752                         goto fail_dp_tx;
1753                 }
1754                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1755                         sfc_err(sa,
1756                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1757                                 tx_name);
1758                         rc = EINVAL;
1759                         goto fail_dp_tx_caps;
1760                 }
1761         } else {
1762                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1763                 if (sa->dp_tx == NULL) {
1764                         sfc_err(sa, "Tx datapath by caps %#x not found",
1765                                 avail_caps);
1766                         rc = ENOENT;
1767                         goto fail_dp_tx;
1768                 }
1769         }
1770
1771         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1772         if (sa->dp_tx_name == NULL) {
1773                 rc = ENOMEM;
1774                 goto fail_dp_tx_name;
1775         }
1776
1777         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1778
1779         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1780
1781         dev->dev_ops = &sfc_eth_dev_ops;
1782
1783         return 0;
1784
1785 fail_dp_tx_name:
1786 fail_dp_tx_caps:
1787         sa->dp_tx = NULL;
1788
1789 fail_dp_tx:
1790 fail_kvarg_tx_datapath:
1791         rte_free(sa->dp_rx_name);
1792         sa->dp_rx_name = NULL;
1793
1794 fail_dp_rx_name:
1795 fail_dp_rx_caps:
1796         sa->dp_rx = NULL;
1797
1798 fail_dp_rx:
1799 fail_kvarg_rx_datapath:
1800         return rc;
1801 }
1802
1803 static void
1804 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1805 {
1806         struct sfc_adapter *sa = dev->data->dev_private;
1807
1808         dev->dev_ops = NULL;
1809         dev->rx_pkt_burst = NULL;
1810         dev->tx_pkt_burst = NULL;
1811
1812         rte_free(sa->dp_tx_name);
1813         sa->dp_tx_name = NULL;
1814         sa->dp_tx = NULL;
1815
1816         rte_free(sa->dp_rx_name);
1817         sa->dp_rx_name = NULL;
1818         sa->dp_rx = NULL;
1819 }
1820
1821 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1822         .rxq_info_get                   = sfc_rx_queue_info_get,
1823         .txq_info_get                   = sfc_tx_queue_info_get,
1824 };
1825
1826 static int
1827 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1828 {
1829         /*
1830          * Device private data has really many process-local pointers.
1831          * Below code should be extremely careful to use data located
1832          * in shared memory only.
1833          */
1834         struct sfc_adapter *sa = dev->data->dev_private;
1835         const struct sfc_dp_rx *dp_rx;
1836         const struct sfc_dp_tx *dp_tx;
1837         int rc;
1838
1839         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1840         if (dp_rx == NULL) {
1841                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1842                 rc = ENOENT;
1843                 goto fail_dp_rx;
1844         }
1845         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1846                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1847                         sa->dp_tx_name);
1848                 rc = EINVAL;
1849                 goto fail_dp_rx_multi_process;
1850         }
1851
1852         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1853         if (dp_tx == NULL) {
1854                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1855                 rc = ENOENT;
1856                 goto fail_dp_tx;
1857         }
1858         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1859                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1860                         sa->dp_tx_name);
1861                 rc = EINVAL;
1862                 goto fail_dp_tx_multi_process;
1863         }
1864
1865         dev->rx_pkt_burst = dp_rx->pkt_burst;
1866         dev->tx_pkt_burst = dp_tx->pkt_burst;
1867         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1868
1869         return 0;
1870
1871 fail_dp_tx_multi_process:
1872 fail_dp_tx:
1873 fail_dp_rx_multi_process:
1874 fail_dp_rx:
1875         return rc;
1876 }
1877
1878 static void
1879 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1880 {
1881         dev->dev_ops = NULL;
1882         dev->tx_pkt_burst = NULL;
1883         dev->rx_pkt_burst = NULL;
1884 }
1885
1886 static void
1887 sfc_register_dp(void)
1888 {
1889         /* Register once */
1890         if (TAILQ_EMPTY(&sfc_dp_head)) {
1891                 /* Prefer EF10 datapath */
1892                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1893                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1894
1895                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1896                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1897                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1898         }
1899 }
1900
1901 static int
1902 sfc_eth_dev_init(struct rte_eth_dev *dev)
1903 {
1904         struct sfc_adapter *sa = dev->data->dev_private;
1905         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1906         int rc;
1907         const efx_nic_cfg_t *encp;
1908         const struct ether_addr *from;
1909
1910         sfc_register_dp();
1911
1912         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1913                 return -sfc_eth_dev_secondary_set_ops(dev);
1914
1915         /* Required for logging */
1916         sa->pci_addr = pci_dev->addr;
1917         sa->port_id = dev->data->port_id;
1918
1919         sa->eth_dev = dev;
1920
1921         /* Copy PCI device info to the dev->data */
1922         rte_eth_copy_pci_info(dev, pci_dev);
1923
1924         rc = sfc_kvargs_parse(sa);
1925         if (rc != 0)
1926                 goto fail_kvargs_parse;
1927
1928         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1929                                 sfc_kvarg_bool_handler, &sa->debug_init);
1930         if (rc != 0)
1931                 goto fail_kvarg_debug_init;
1932
1933         sfc_log_init(sa, "entry");
1934
1935         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1936         if (dev->data->mac_addrs == NULL) {
1937                 rc = ENOMEM;
1938                 goto fail_mac_addrs;
1939         }
1940
1941         sfc_adapter_lock_init(sa);
1942         sfc_adapter_lock(sa);
1943
1944         sfc_log_init(sa, "probing");
1945         rc = sfc_probe(sa);
1946         if (rc != 0)
1947                 goto fail_probe;
1948
1949         sfc_log_init(sa, "set device ops");
1950         rc = sfc_eth_dev_set_ops(dev);
1951         if (rc != 0)
1952                 goto fail_set_ops;
1953
1954         sfc_log_init(sa, "attaching");
1955         rc = sfc_attach(sa);
1956         if (rc != 0)
1957                 goto fail_attach;
1958
1959         encp = efx_nic_cfg_get(sa->nic);
1960
1961         /*
1962          * The arguments are really reverse order in comparison to
1963          * Linux kernel. Copy from NIC config to Ethernet device data.
1964          */
1965         from = (const struct ether_addr *)(encp->enc_mac_addr);
1966         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1967
1968         sfc_adapter_unlock(sa);
1969
1970         sfc_log_init(sa, "done");
1971         return 0;
1972
1973 fail_attach:
1974         sfc_eth_dev_clear_ops(dev);
1975
1976 fail_set_ops:
1977         sfc_unprobe(sa);
1978
1979 fail_probe:
1980         sfc_adapter_unlock(sa);
1981         sfc_adapter_lock_fini(sa);
1982         rte_free(dev->data->mac_addrs);
1983         dev->data->mac_addrs = NULL;
1984
1985 fail_mac_addrs:
1986 fail_kvarg_debug_init:
1987         sfc_kvargs_cleanup(sa);
1988
1989 fail_kvargs_parse:
1990         sfc_log_init(sa, "failed %d", rc);
1991         SFC_ASSERT(rc > 0);
1992         return -rc;
1993 }
1994
1995 static int
1996 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1997 {
1998         struct sfc_adapter *sa;
1999
2000         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2001                 sfc_eth_dev_secondary_clear_ops(dev);
2002                 return 0;
2003         }
2004
2005         sa = dev->data->dev_private;
2006         sfc_log_init(sa, "entry");
2007
2008         sfc_adapter_lock(sa);
2009
2010         sfc_eth_dev_clear_ops(dev);
2011
2012         sfc_detach(sa);
2013         sfc_unprobe(sa);
2014
2015         rte_free(dev->data->mac_addrs);
2016         dev->data->mac_addrs = NULL;
2017
2018         sfc_kvargs_cleanup(sa);
2019
2020         sfc_adapter_unlock(sa);
2021         sfc_adapter_lock_fini(sa);
2022
2023         sfc_log_init(sa, "done");
2024
2025         /* Required for logging, so cleanup last */
2026         sa->eth_dev = NULL;
2027         return 0;
2028 }
2029
2030 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2031         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2032         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2033         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2034         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2035         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2036         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2037         { .vendor_id = 0 /* sentinel */ }
2038 };
2039
2040 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2041         struct rte_pci_device *pci_dev)
2042 {
2043         return rte_eth_dev_pci_generic_probe(pci_dev,
2044                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2045 }
2046
2047 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2048 {
2049         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2050 }
2051
2052 static struct rte_pci_driver sfc_efx_pmd = {
2053         .id_table = pci_id_sfc_efx_map,
2054         .drv_flags =
2055                 RTE_PCI_DRV_INTR_LSC |
2056                 RTE_PCI_DRV_NEED_MAPPING,
2057         .probe = sfc_eth_dev_pci_probe,
2058         .remove = sfc_eth_dev_pci_remove,
2059 };
2060
2061 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2062 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2063 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2064 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2065         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2066         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2067         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2068         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2069         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2070         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);