drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2021 Xilinx, Inc.
   4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
   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 <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17 #include <rte_ether.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_debug.h"
  23 #include "sfc_log.h"
  24 #include "sfc_kvargs.h"
  25 #include "sfc_ev.h"
  26 #include "sfc_rx.h"
  27 #include "sfc_tx.h"
  28 #include "sfc_flow.h"
  29 #include "sfc_dp.h"
  30 #include "sfc_dp_rx.h"
  31
  32 uint32_t sfc_logtype_driver;
  33
  34 static struct sfc_dp_list sfc_dp_head =
  35         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  36
  37
  38 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  39
  40
  41 static int
  42 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  43 {
  44         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  45         efx_nic_fw_info_t enfi;
  46         int ret;
  47         int rc;
  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 int
  82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  83 {
  84         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  85         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  86         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  87         struct sfc_rss *rss = &sas->rss;
  88         struct sfc_mae *mae = &sa->mae;
  89         uint64_t txq_offloads_def = 0;
  90
  91         sfc_log_init(sa, "entry");
  92
  93         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
  94         dev_info->max_mtu = EFX_MAC_SDU_MAX;
  95
  96         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  97
  98         dev_info->max_vfs = sa->sriov.num_vfs;
  99
 100         /* Autonegotiation may be disabled */
 101         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 102         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 103                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 104         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 105                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 106         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 107                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 108         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 109                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 110         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 111                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 112         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 113                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 114
 115         dev_info->max_rx_queues = sa->rxq_max;
 116         dev_info->max_tx_queues = sa->txq_max;
 117
 118         /* By default packets are dropped if no descriptors are available */
 119         dev_info->default_rxconf.rx_drop_en = 1;
 120
 121         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 122
 123         /*
 124          * rx_offload_capa includes both device and queue offloads since
 125          * the latter may be requested on a per device basis which makes
 126          * sense when some offloads are needed to be set on all queues.
 127          */
 128         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 129                                     dev_info->rx_queue_offload_capa;
 130
 131         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 132
 133         /*
 134          * tx_offload_capa includes both device and queue offloads since
 135          * the latter may be requested on a per device basis which makes
 136          * sense when some offloads are needed to be set on all queues.
 137          */
 138         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 139                                     dev_info->tx_queue_offload_capa;
 140
 141         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 142                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 143
 144         dev_info->default_txconf.offloads |= txq_offloads_def;
 145
 146         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 147                 uint64_t rte_hf = 0;
 148                 unsigned int i;
 149
 150                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 151                         rte_hf |= rss->hf_map[i].rte;
 152
 153                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 154                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 155                 dev_info->flow_type_rss_offloads = rte_hf;
 156         }
 157
 158         /* Initialize to hardware limits */
 159         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 160         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 161         /* The RXQ hardware requires that the descriptor count is a power
 162          * of 2, but rx_desc_lim cannot properly describe that constraint.
 163          */
 164         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 165
 166         /* Initialize to hardware limits */
 167         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 168         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 169         /*
 170          * The TXQ hardware requires that the descriptor count is a power
 171          * of 2, but tx_desc_lim cannot properly describe that constraint
 172          */
 173         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 174
 175         if (sap->dp_rx->get_dev_info != NULL)
 176                 sap->dp_rx->get_dev_info(dev_info);
 177         if (sap->dp_tx->get_dev_info != NULL)
 178                 sap->dp_tx->get_dev_info(dev_info);
 179
 180         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 181                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 182
 183         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
 184                 dev_info->switch_info.name = dev->device->driver->name;
 185                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 186                 dev_info->switch_info.port_id = mae->switch_port_id;
 187         }
 188
 189         return 0;
 190 }
 191
 192 static const uint32_t *
 193 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 194 {
 195         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 196
 197         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 198 }
 199
 200 static int
 201 sfc_dev_configure(struct rte_eth_dev *dev)
 202 {
 203         struct rte_eth_dev_data *dev_data = dev->data;
 204         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 205         int rc;
 206
 207         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 208                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 209
 210         sfc_adapter_lock(sa);
 211         switch (sa->state) {
 212         case SFC_ADAPTER_CONFIGURED:
 213                 /* FALLTHROUGH */
 214         case SFC_ADAPTER_INITIALIZED:
 215                 rc = sfc_configure(sa);
 216                 break;
 217         default:
 218                 sfc_err(sa, "unexpected adapter state %u to configure",
 219                         sa->state);
 220                 rc = EINVAL;
 221                 break;
 222         }
 223         sfc_adapter_unlock(sa);
 224
 225         sfc_log_init(sa, "done %d", rc);
 226         SFC_ASSERT(rc >= 0);
 227         return -rc;
 228 }
 229
 230 static int
 231 sfc_dev_start(struct rte_eth_dev *dev)
 232 {
 233         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 234         int rc;
 235
 236         sfc_log_init(sa, "entry");
 237
 238         sfc_adapter_lock(sa);
 239         rc = sfc_start(sa);
 240         sfc_adapter_unlock(sa);
 241
 242         sfc_log_init(sa, "done %d", rc);
 243         SFC_ASSERT(rc >= 0);
 244         return -rc;
 245 }
 246
 247 static int
 248 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 249 {
 250         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 251         struct rte_eth_link current_link;
 252         int ret;
 253
 254         sfc_log_init(sa, "entry");
 255
 256         if (sa->state != SFC_ADAPTER_STARTED) {
 257                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 258         } else if (wait_to_complete) {
 259                 efx_link_mode_t link_mode;
 260
 261                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 262                         link_mode = EFX_LINK_UNKNOWN;
 263                 sfc_port_link_mode_to_info(link_mode, &current_link);
 264
 265         } else {
 266                 sfc_ev_mgmt_qpoll(sa);
 267                 rte_eth_linkstatus_get(dev, &current_link);
 268         }
 269
 270         ret = rte_eth_linkstatus_set(dev, &current_link);
 271         if (ret == 0)
 272                 sfc_notice(sa, "Link status is %s",
 273                            current_link.link_status ? "UP" : "DOWN");
 274
 275         return ret;
 276 }
 277
 278 static int
 279 sfc_dev_stop(struct rte_eth_dev *dev)
 280 {
 281         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 282
 283         sfc_log_init(sa, "entry");
 284
 285         sfc_adapter_lock(sa);
 286         sfc_stop(sa);
 287         sfc_adapter_unlock(sa);
 288
 289         sfc_log_init(sa, "done");
 290
 291         return 0;
 292 }
 293
 294 static int
 295 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 296 {
 297         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 298         int rc;
 299
 300         sfc_log_init(sa, "entry");
 301
 302         sfc_adapter_lock(sa);
 303         rc = sfc_start(sa);
 304         sfc_adapter_unlock(sa);
 305
 306         SFC_ASSERT(rc >= 0);
 307         return -rc;
 308 }
 309
 310 static int
 311 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 312 {
 313         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 314
 315         sfc_log_init(sa, "entry");
 316
 317         sfc_adapter_lock(sa);
 318         sfc_stop(sa);
 319         sfc_adapter_unlock(sa);
 320
 321         return 0;
 322 }
 323
 324 static void
 325 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 326 {
 327         free(dev->process_private);
 328         rte_eth_dev_release_port(dev);
 329 }
 330
 331 static int
 332 sfc_dev_close(struct rte_eth_dev *dev)
 333 {
 334         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 335
 336         sfc_log_init(sa, "entry");
 337
 338         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 339                 sfc_eth_dev_secondary_clear_ops(dev);
 340                 return 0;
 341         }
 342
 343         sfc_adapter_lock(sa);
 344         switch (sa->state) {
 345         case SFC_ADAPTER_STARTED:
 346                 sfc_stop(sa);
 347                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 348                 /* FALLTHROUGH */
 349         case SFC_ADAPTER_CONFIGURED:
 350                 sfc_close(sa);
 351                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 352                 /* FALLTHROUGH */
 353         case SFC_ADAPTER_INITIALIZED:
 354                 break;
 355         default:
 356                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 357                 break;
 358         }
 359
 360         /*
 361          * Cleanup all resources.
 362          * Rollback primary process sfc_eth_dev_init() below.
 363          */
 364
 365         sfc_eth_dev_clear_ops(dev);
 366
 367         sfc_detach(sa);
 368         sfc_unprobe(sa);
 369
 370         sfc_kvargs_cleanup(sa);
 371
 372         sfc_adapter_unlock(sa);
 373         sfc_adapter_lock_fini(sa);
 374
 375         sfc_log_init(sa, "done");
 376
 377         /* Required for logging, so cleanup last */
 378         sa->eth_dev = NULL;
 379
 380         free(sa);
 381
 382         return 0;
 383 }
 384
 385 static int
 386 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 387                    boolean_t enabled)
 388 {
 389         struct sfc_port *port;
 390         boolean_t *toggle;
 391         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 392         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 393         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 394         int rc = 0;
 395
 396         sfc_adapter_lock(sa);
 397
 398         port = &sa->port;
 399         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 400
 401         if (*toggle != enabled) {
 402                 *toggle = enabled;
 403
 404                 if (sfc_sa2shared(sa)->isolated) {
 405                         sfc_warn(sa, "isolated mode is active on the port");
 406                         sfc_warn(sa, "the change is to be applied on the next "
 407                                      "start provided that isolated mode is "
 408                                      "disabled prior the next start");
 409                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 410                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 411                         *toggle = !(enabled);
 412                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 413                                  ((enabled) ? "enable" : "disable"), desc, rc);
 414
 415                         /*
 416                          * For promiscuous and all-multicast filters a
 417                          * permission failure should be reported as an
 418                          * unsupported filter.
 419                          */
 420                         if (rc == EPERM)
 421                                 rc = ENOTSUP;
 422                 }
 423         }
 424
 425         sfc_adapter_unlock(sa);
 426         return rc;
 427 }
 428
 429 static int
 430 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 431 {
 432         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 433
 434         SFC_ASSERT(rc >= 0);
 435         return -rc;
 436 }
 437
 438 static int
 439 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 440 {
 441         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 442
 443         SFC_ASSERT(rc >= 0);
 444         return -rc;
 445 }
 446
 447 static int
 448 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 449 {
 450         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 451
 452         SFC_ASSERT(rc >= 0);
 453         return -rc;
 454 }
 455
 456 static int
 457 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 458 {
 459         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 460
 461         SFC_ASSERT(rc >= 0);
 462         return -rc;
 463 }
 464
 465 static int
 466 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 467                    uint16_t nb_rx_desc, unsigned int socket_id,
 468                    const struct rte_eth_rxconf *rx_conf,
 469                    struct rte_mempool *mb_pool)
 470 {
 471         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 472         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 473         int rc;
 474
 475         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 476                      rx_queue_id, nb_rx_desc, socket_id);
 477
 478         sfc_adapter_lock(sa);
 479
 480         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 481                           rx_conf, mb_pool);
 482         if (rc != 0)
 483                 goto fail_rx_qinit;
 484
 485         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 486
 487         sfc_adapter_unlock(sa);
 488
 489         return 0;
 490
 491 fail_rx_qinit:
 492         sfc_adapter_unlock(sa);
 493         SFC_ASSERT(rc > 0);
 494         return -rc;
 495 }
 496
 497 static void
 498 sfc_rx_queue_release(void *queue)
 499 {
 500         struct sfc_dp_rxq *dp_rxq = queue;
 501         struct sfc_rxq *rxq;
 502         struct sfc_adapter *sa;
 503         unsigned int sw_index;
 504
 505         if (dp_rxq == NULL)
 506                 return;
 507
 508         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 509         sa = rxq->evq->sa;
 510         sfc_adapter_lock(sa);
 511
 512         sw_index = dp_rxq->dpq.queue_id;
 513
 514         sfc_log_init(sa, "RxQ=%u", sw_index);
 515
 516         sfc_rx_qfini(sa, sw_index);
 517
 518         sfc_adapter_unlock(sa);
 519 }
 520
 521 static int
 522 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 523                    uint16_t nb_tx_desc, unsigned int socket_id,
 524                    const struct rte_eth_txconf *tx_conf)
 525 {
 526         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 527         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 528         int rc;
 529
 530         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 531                      tx_queue_id, nb_tx_desc, socket_id);
 532
 533         sfc_adapter_lock(sa);
 534
 535         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 536         if (rc != 0)
 537                 goto fail_tx_qinit;
 538
 539         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
 540
 541         sfc_adapter_unlock(sa);
 542         return 0;
 543
 544 fail_tx_qinit:
 545         sfc_adapter_unlock(sa);
 546         SFC_ASSERT(rc > 0);
 547         return -rc;
 548 }
 549
 550 static void
 551 sfc_tx_queue_release(void *queue)
 552 {
 553         struct sfc_dp_txq *dp_txq = queue;
 554         struct sfc_txq *txq;
 555         unsigned int sw_index;
 556         struct sfc_adapter *sa;
 557
 558         if (dp_txq == NULL)
 559                 return;
 560
 561         txq = sfc_txq_by_dp_txq(dp_txq);
 562         sw_index = dp_txq->dpq.queue_id;
 563
 564         SFC_ASSERT(txq->evq != NULL);
 565         sa = txq->evq->sa;
 566
 567         sfc_log_init(sa, "TxQ = %u", sw_index);
 568
 569         sfc_adapter_lock(sa);
 570
 571         sfc_tx_qfini(sa, sw_index);
 572
 573         sfc_adapter_unlock(sa);
 574 }
 575
 576 /*
 577  * Some statistics are computed as A - B where A and B each increase
 578  * monotonically with some hardware counter(s) and the counters are read
 579  * asynchronously.
 580  *
 581  * If packet X is counted in A, but not counted in B yet, computed value is
 582  * greater than real.
 583  *
 584  * If packet X is not counted in A at the moment of reading the counter,
 585  * but counted in B at the moment of reading the counter, computed value
 586  * is less than real.
 587  *
 588  * However, counter which grows backward is worse evil than slightly wrong
 589  * value. So, let's try to guarantee that it never happens except may be
 590  * the case when the MAC stats are zeroed as a result of a NIC reset.
 591  */
 592 static void
 593 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 594 {
 595         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 596                 *stat = newval;
 597 }
 598
 599 static int
 600 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 601 {
 602         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 603         struct sfc_port *port = &sa->port;
 604         uint64_t *mac_stats;
 605         int ret;
 606
 607         rte_spinlock_lock(&port->mac_stats_lock);
 608
 609         ret = sfc_port_update_mac_stats(sa);
 610         if (ret != 0)
 611                 goto unlock;
 612
 613         mac_stats = port->mac_stats_buf;
 614
 615         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 616                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 617                 stats->ipackets =
 618                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 619                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 620                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 621                 stats->opackets =
 622                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 623                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 624                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 625                 stats->ibytes =
 626                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 627                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 628                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 629                 stats->obytes =
 630                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 631                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 632                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 633                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 634                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 635
 636                 /* CRC is included in these stats, but shouldn't be */
 637                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 638                 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 639         } else {
 640                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 641                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 642                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 643
 644                 /* CRC is included in these stats, but shouldn't be */
 645                 stats->ibytes -= mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 646                 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 647
 648                 /*
 649                  * Take into account stats which are whenever supported
 650                  * on EF10. If some stat is not supported by current
 651                  * firmware variant or HW revision, it is guaranteed
 652                  * to be zero in mac_stats.
 653                  */
 654                 stats->imissed =
 655                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 656                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 657                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 658                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 659                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 660                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 661                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 662                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 663                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 664                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 665                 stats->ierrors =
 666                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 667                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 668                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 669                 /* no oerrors counters supported on EF10 */
 670
 671                 /* Exclude missed, errors and pauses from Rx packets */
 672                 sfc_update_diff_stat(&port->ipackets,
 673                         mac_stats[EFX_MAC_RX_PKTS] -
 674                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 675                         stats->imissed - stats->ierrors);
 676                 stats->ipackets = port->ipackets;
 677         }
 678
 679 unlock:
 680         rte_spinlock_unlock(&port->mac_stats_lock);
 681         SFC_ASSERT(ret >= 0);
 682         return -ret;
 683 }
 684
 685 static int
 686 sfc_stats_reset(struct rte_eth_dev *dev)
 687 {
 688         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 689         struct sfc_port *port = &sa->port;
 690         int rc;
 691
 692         if (sa->state != SFC_ADAPTER_STARTED) {
 693                 /*
 694                  * The operation cannot be done if port is not started; it
 695                  * will be scheduled to be done during the next port start
 696                  */
 697                 port->mac_stats_reset_pending = B_TRUE;
 698                 return 0;
 699         }
 700
 701         rc = sfc_port_reset_mac_stats(sa);
 702         if (rc != 0)
 703                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 704
 705         SFC_ASSERT(rc >= 0);
 706         return -rc;
 707 }
 708
 709 static int
 710 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 711                unsigned int xstats_count)
 712 {
 713         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 714         struct sfc_port *port = &sa->port;
 715         uint64_t *mac_stats;
 716         int rc;
 717         unsigned int i;
 718         int nstats = 0;
 719
 720         rte_spinlock_lock(&port->mac_stats_lock);
 721
 722         rc = sfc_port_update_mac_stats(sa);
 723         if (rc != 0) {
 724                 SFC_ASSERT(rc > 0);
 725                 nstats = -rc;
 726                 goto unlock;
 727         }
 728
 729         mac_stats = port->mac_stats_buf;
 730
 731         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 732                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 733                         if (xstats != NULL && nstats < (int)xstats_count) {
 734                                 xstats[nstats].id = nstats;
 735                                 xstats[nstats].value = mac_stats[i];
 736                         }
 737                         nstats++;
 738                 }
 739         }
 740
 741 unlock:
 742         rte_spinlock_unlock(&port->mac_stats_lock);
 743
 744         return nstats;
 745 }
 746
 747 static int
 748 sfc_xstats_get_names(struct rte_eth_dev *dev,
 749                      struct rte_eth_xstat_name *xstats_names,
 750                      unsigned int xstats_count)
 751 {
 752         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 753         struct sfc_port *port = &sa->port;
 754         unsigned int i;
 755         unsigned int nstats = 0;
 756
 757         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 758                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 759                         if (xstats_names != NULL && nstats < xstats_count)
 760                                 strlcpy(xstats_names[nstats].name,
 761                                         efx_mac_stat_name(sa->nic, i),
 762                                         sizeof(xstats_names[0].name));
 763                         nstats++;
 764                 }
 765         }
 766
 767         return nstats;
 768 }
 769
 770 static int
 771 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 772                      uint64_t *values, unsigned int n)
 773 {
 774         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 775         struct sfc_port *port = &sa->port;
 776         uint64_t *mac_stats;
 777         unsigned int nb_supported = 0;
 778         unsigned int nb_written = 0;
 779         unsigned int i;
 780         int ret;
 781         int rc;
 782
 783         if (unlikely(values == NULL) ||
 784             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 785                 return port->mac_stats_nb_supported;
 786
 787         rte_spinlock_lock(&port->mac_stats_lock);
 788
 789         rc = sfc_port_update_mac_stats(sa);
 790         if (rc != 0) {
 791                 SFC_ASSERT(rc > 0);
 792                 ret = -rc;
 793                 goto unlock;
 794         }
 795
 796         mac_stats = port->mac_stats_buf;
 797
 798         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 799                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 800                         continue;
 801
 802                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 803                         values[nb_written++] = mac_stats[i];
 804
 805                 ++nb_supported;
 806         }
 807
 808         ret = nb_written;
 809
 810 unlock:
 811         rte_spinlock_unlock(&port->mac_stats_lock);
 812
 813         return ret;
 814 }
 815
 816 static int
 817 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 818                            struct rte_eth_xstat_name *xstats_names,
 819                            const uint64_t *ids, unsigned int size)
 820 {
 821         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 822         struct sfc_port *port = &sa->port;
 823         unsigned int nb_supported = 0;
 824         unsigned int nb_written = 0;
 825         unsigned int i;
 826
 827         if (unlikely(xstats_names == NULL) ||
 828             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 829                 return port->mac_stats_nb_supported;
 830
 831         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 832                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 833                         continue;
 834
 835                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 836                         char *name = xstats_names[nb_written++].name;
 837
 838                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 839                                 sizeof(xstats_names[0].name));
 840                 }
 841
 842                 ++nb_supported;
 843         }
 844
 845         return nb_written;
 846 }
 847
 848 static int
 849 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 850 {
 851         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 852         unsigned int wanted_fc, link_fc;
 853
 854         memset(fc_conf, 0, sizeof(*fc_conf));
 855
 856         sfc_adapter_lock(sa);
 857
 858         if (sa->state == SFC_ADAPTER_STARTED)
 859                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 860         else
 861                 link_fc = sa->port.flow_ctrl;
 862
 863         switch (link_fc) {
 864         case 0:
 865                 fc_conf->mode = RTE_FC_NONE;
 866                 break;
 867         case EFX_FCNTL_RESPOND:
 868                 fc_conf->mode = RTE_FC_RX_PAUSE;
 869                 break;
 870         case EFX_FCNTL_GENERATE:
 871                 fc_conf->mode = RTE_FC_TX_PAUSE;
 872                 break;
 873         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 874                 fc_conf->mode = RTE_FC_FULL;
 875                 break;
 876         default:
 877                 sfc_err(sa, "%s: unexpected flow control value %#x",
 878                         __func__, link_fc);
 879         }
 880
 881         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 882
 883         sfc_adapter_unlock(sa);
 884
 885         return 0;
 886 }
 887
 888 static int
 889 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 890 {
 891         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 892         struct sfc_port *port = &sa->port;
 893         unsigned int fcntl;
 894         int rc;
 895
 896         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 897             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 898             fc_conf->mac_ctrl_frame_fwd != 0) {
 899                 sfc_err(sa, "unsupported flow control settings specified");
 900                 rc = EINVAL;
 901                 goto fail_inval;
 902         }
 903
 904         switch (fc_conf->mode) {
 905         case RTE_FC_NONE:
 906                 fcntl = 0;
 907                 break;
 908         case RTE_FC_RX_PAUSE:
 909                 fcntl = EFX_FCNTL_RESPOND;
 910                 break;
 911         case RTE_FC_TX_PAUSE:
 912                 fcntl = EFX_FCNTL_GENERATE;
 913                 break;
 914         case RTE_FC_FULL:
 915                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 916                 break;
 917         default:
 918                 rc = EINVAL;
 919                 goto fail_inval;
 920         }
 921
 922         sfc_adapter_lock(sa);
 923
 924         if (sa->state == SFC_ADAPTER_STARTED) {
 925                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 926                 if (rc != 0)
 927                         goto fail_mac_fcntl_set;
 928         }
 929
 930         port->flow_ctrl = fcntl;
 931         port->flow_ctrl_autoneg = fc_conf->autoneg;
 932
 933         sfc_adapter_unlock(sa);
 934
 935         return 0;
 936
 937 fail_mac_fcntl_set:
 938         sfc_adapter_unlock(sa);
 939 fail_inval:
 940         SFC_ASSERT(rc > 0);
 941         return -rc;
 942 }
 943
 944 static int
 945 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 946 {
 947         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 948         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 949         boolean_t scatter_enabled;
 950         const char *error;
 951         unsigned int i;
 952
 953         for (i = 0; i < sas->rxq_count; i++) {
 954                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 955                         continue;
 956
 957                 scatter_enabled = (sas->rxq_info[i].type_flags &
 958                                    EFX_RXQ_FLAG_SCATTER);
 959
 960                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 961                                           encp->enc_rx_prefix_size,
 962                                           scatter_enabled,
 963                                           encp->enc_rx_scatter_max, &error)) {
 964                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 965                                 error);
 966                         return EINVAL;
 967                 }
 968         }
 969
 970         return 0;
 971 }
 972
 973 static int
 974 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 975 {
 976         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 977         size_t pdu = EFX_MAC_PDU(mtu);
 978         size_t old_pdu;
 979         int rc;
 980
 981         sfc_log_init(sa, "mtu=%u", mtu);
 982
 983         rc = EINVAL;
 984         if (pdu < EFX_MAC_PDU_MIN) {
 985                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 986                         (unsigned int)mtu, (unsigned int)pdu,
 987                         EFX_MAC_PDU_MIN);
 988                 goto fail_inval;
 989         }
 990         if (pdu > EFX_MAC_PDU_MAX) {
 991                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 992                         (unsigned int)mtu, (unsigned int)pdu,
 993                         (unsigned int)EFX_MAC_PDU_MAX);
 994                 goto fail_inval;
 995         }
 996
 997         sfc_adapter_lock(sa);
 998
 999         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1000         if (rc != 0)
1001                 goto fail_check_scatter;
1002
1003         if (pdu != sa->port.pdu) {
1004                 if (sa->state == SFC_ADAPTER_STARTED) {
1005                         sfc_stop(sa);
1006
1007                         old_pdu = sa->port.pdu;
1008                         sa->port.pdu = pdu;
1009                         rc = sfc_start(sa);
1010                         if (rc != 0)
1011                                 goto fail_start;
1012                 } else {
1013                         sa->port.pdu = pdu;
1014                 }
1015         }
1016
1017         /*
1018          * The driver does not use it, but other PMDs update jumbo frame
1019          * flag and max_rx_pkt_len when MTU is set.
1020          */
1021         if (mtu > RTE_ETHER_MTU) {
1022                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1023                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1024         }
1025
1026         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1027
1028         sfc_adapter_unlock(sa);
1029
1030         sfc_log_init(sa, "done");
1031         return 0;
1032
1033 fail_start:
1034         sa->port.pdu = old_pdu;
1035         if (sfc_start(sa) != 0)
1036                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1037                         "PDU max size - port is stopped",
1038                         (unsigned int)pdu, (unsigned int)old_pdu);
1039
1040 fail_check_scatter:
1041         sfc_adapter_unlock(sa);
1042
1043 fail_inval:
1044         sfc_log_init(sa, "failed %d", rc);
1045         SFC_ASSERT(rc > 0);
1046         return -rc;
1047 }
1048 static int
1049 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1050 {
1051         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1052         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1053         struct sfc_port *port = &sa->port;
1054         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1055         int rc = 0;
1056
1057         sfc_adapter_lock(sa);
1058
1059         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1060                 goto unlock;
1061
1062         /*
1063          * Copy the address to the device private data so that
1064          * it could be recalled in the case of adapter restart.
1065          */
1066         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1067
1068         /*
1069          * Neither of the two following checks can return
1070          * an error. The new MAC address is preserved in
1071          * the device private data and can be activated
1072          * on the next port start if the user prevents
1073          * isolated mode from being enabled.
1074          */
1075         if (sfc_sa2shared(sa)->isolated) {
1076                 sfc_warn(sa, "isolated mode is active on the port");
1077                 sfc_warn(sa, "will not set MAC address");
1078                 goto unlock;
1079         }
1080
1081         if (sa->state != SFC_ADAPTER_STARTED) {
1082                 sfc_notice(sa, "the port is not started");
1083                 sfc_notice(sa, "the new MAC address will be set on port start");
1084
1085                 goto unlock;
1086         }
1087
1088         if (encp->enc_allow_set_mac_with_installed_filters) {
1089                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1090                 if (rc != 0) {
1091                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1092                         goto unlock;
1093                 }
1094
1095                 /*
1096                  * Changing the MAC address by means of MCDI request
1097                  * has no effect on received traffic, therefore
1098                  * we also need to update unicast filters
1099                  */
1100                 rc = sfc_set_rx_mode_unchecked(sa);
1101                 if (rc != 0) {
1102                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1103                         /* Rollback the old address */
1104                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1105                         (void)sfc_set_rx_mode_unchecked(sa);
1106                 }
1107         } else {
1108                 sfc_warn(sa, "cannot set MAC address with filters installed");
1109                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1110                 sfc_warn(sa, "(some traffic may be dropped)");
1111
1112                 /*
1113                  * Since setting MAC address with filters installed is not
1114                  * allowed on the adapter, the new MAC address will be set
1115                  * by means of adapter restart. sfc_start() shall retrieve
1116                  * the new address from the device private data and set it.
1117                  */
1118                 sfc_stop(sa);
1119                 rc = sfc_start(sa);
1120                 if (rc != 0)
1121                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1122         }
1123
1124 unlock:
1125         if (rc != 0)
1126                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1127
1128         sfc_adapter_unlock(sa);
1129
1130         SFC_ASSERT(rc >= 0);
1131         return -rc;
1132 }
1133
1134
1135 static int
1136 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1137                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1138 {
1139         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1140         struct sfc_port *port = &sa->port;
1141         uint8_t *mc_addrs = port->mcast_addrs;
1142         int rc;
1143         unsigned int i;
1144
1145         if (sfc_sa2shared(sa)->isolated) {
1146                 sfc_err(sa, "isolated mode is active on the port");
1147                 sfc_err(sa, "will not set multicast address list");
1148                 return -ENOTSUP;
1149         }
1150
1151         if (mc_addrs == NULL)
1152                 return -ENOBUFS;
1153
1154         if (nb_mc_addr > port->max_mcast_addrs) {
1155                 sfc_err(sa, "too many multicast addresses: %u > %u",
1156                          nb_mc_addr, port->max_mcast_addrs);
1157                 return -EINVAL;
1158         }
1159
1160         for (i = 0; i < nb_mc_addr; ++i) {
1161                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1162                                  EFX_MAC_ADDR_LEN);
1163                 mc_addrs += EFX_MAC_ADDR_LEN;
1164         }
1165
1166         port->nb_mcast_addrs = nb_mc_addr;
1167
1168         if (sa->state != SFC_ADAPTER_STARTED)
1169                 return 0;
1170
1171         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1172                                         port->nb_mcast_addrs);
1173         if (rc != 0)
1174                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1175
1176         SFC_ASSERT(rc >= 0);
1177         return -rc;
1178 }
1179
1180 /*
1181  * The function is used by the secondary process as well. It must not
1182  * use any process-local pointers from the adapter data.
1183  */
1184 static void
1185 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1186                       struct rte_eth_rxq_info *qinfo)
1187 {
1188         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1189         struct sfc_rxq_info *rxq_info;
1190
1191         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1192
1193         rxq_info = &sas->rxq_info[rx_queue_id];
1194
1195         qinfo->mp = rxq_info->refill_mb_pool;
1196         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1197         qinfo->conf.rx_drop_en = 1;
1198         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1199         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1200         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1201                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1202                 qinfo->scattered_rx = 1;
1203         }
1204         qinfo->nb_desc = rxq_info->entries;
1205 }
1206
1207 /*
1208  * The function is used by the secondary process as well. It must not
1209  * use any process-local pointers from the adapter data.
1210  */
1211 static void
1212 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1213                       struct rte_eth_txq_info *qinfo)
1214 {
1215         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1216         struct sfc_txq_info *txq_info;
1217
1218         SFC_ASSERT(tx_queue_id < sas->txq_count);
1219
1220         txq_info = &sas->txq_info[tx_queue_id];
1221
1222         memset(qinfo, 0, sizeof(*qinfo));
1223
1224         qinfo->conf.offloads = txq_info->offloads;
1225         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1226         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1227         qinfo->nb_desc = txq_info->entries;
1228 }
1229
1230 /*
1231  * The function is used by the secondary process as well. It must not
1232  * use any process-local pointers from the adapter data.
1233  */
1234 static uint32_t
1235 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1236 {
1237         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1238         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1239         struct sfc_rxq_info *rxq_info;
1240
1241         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1242         rxq_info = &sas->rxq_info[rx_queue_id];
1243
1244         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1245                 return 0;
1246
1247         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1248 }
1249
1250 /*
1251  * The function is used by the secondary process as well. It must not
1252  * use any process-local pointers from the adapter data.
1253  */
1254 static int
1255 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1256 {
1257         struct sfc_dp_rxq *dp_rxq = queue;
1258         const struct sfc_dp_rx *dp_rx;
1259
1260         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1261
1262         return offset < dp_rx->qdesc_npending(dp_rxq);
1263 }
1264
1265 /*
1266  * The function is used by the secondary process as well. It must not
1267  * use any process-local pointers from the adapter data.
1268  */
1269 static int
1270 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1271 {
1272         struct sfc_dp_rxq *dp_rxq = queue;
1273         const struct sfc_dp_rx *dp_rx;
1274
1275         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1276
1277         return dp_rx->qdesc_status(dp_rxq, offset);
1278 }
1279
1280 /*
1281  * The function is used by the secondary process as well. It must not
1282  * use any process-local pointers from the adapter data.
1283  */
1284 static int
1285 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1286 {
1287         struct sfc_dp_txq *dp_txq = queue;
1288         const struct sfc_dp_tx *dp_tx;
1289
1290         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1291
1292         return dp_tx->qdesc_status(dp_txq, offset);
1293 }
1294
1295 static int
1296 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1297 {
1298         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1299         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1300         int rc;
1301
1302         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1303
1304         sfc_adapter_lock(sa);
1305
1306         rc = EINVAL;
1307         if (sa->state != SFC_ADAPTER_STARTED)
1308                 goto fail_not_started;
1309
1310         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1311                 goto fail_not_setup;
1312
1313         rc = sfc_rx_qstart(sa, rx_queue_id);
1314         if (rc != 0)
1315                 goto fail_rx_qstart;
1316
1317         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1318
1319         sfc_adapter_unlock(sa);
1320
1321         return 0;
1322
1323 fail_rx_qstart:
1324 fail_not_setup:
1325 fail_not_started:
1326         sfc_adapter_unlock(sa);
1327         SFC_ASSERT(rc > 0);
1328         return -rc;
1329 }
1330
1331 static int
1332 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1333 {
1334         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1335         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1336
1337         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1338
1339         sfc_adapter_lock(sa);
1340         sfc_rx_qstop(sa, rx_queue_id);
1341
1342         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1343
1344         sfc_adapter_unlock(sa);
1345
1346         return 0;
1347 }
1348
1349 static int
1350 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1351 {
1352         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1353         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1354         int rc;
1355
1356         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1357
1358         sfc_adapter_lock(sa);
1359
1360         rc = EINVAL;
1361         if (sa->state != SFC_ADAPTER_STARTED)
1362                 goto fail_not_started;
1363
1364         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1365                 goto fail_not_setup;
1366
1367         rc = sfc_tx_qstart(sa, tx_queue_id);
1368         if (rc != 0)
1369                 goto fail_tx_qstart;
1370
1371         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1372
1373         sfc_adapter_unlock(sa);
1374         return 0;
1375
1376 fail_tx_qstart:
1377
1378 fail_not_setup:
1379 fail_not_started:
1380         sfc_adapter_unlock(sa);
1381         SFC_ASSERT(rc > 0);
1382         return -rc;
1383 }
1384
1385 static int
1386 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1387 {
1388         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1389         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1390
1391         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1392
1393         sfc_adapter_lock(sa);
1394
1395         sfc_tx_qstop(sa, tx_queue_id);
1396
1397         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1398
1399         sfc_adapter_unlock(sa);
1400         return 0;
1401 }
1402
1403 static efx_tunnel_protocol_t
1404 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1405 {
1406         switch (rte_type) {
1407         case RTE_TUNNEL_TYPE_VXLAN:
1408                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1409         case RTE_TUNNEL_TYPE_GENEVE:
1410                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1411         default:
1412                 return EFX_TUNNEL_NPROTOS;
1413         }
1414 }
1415
1416 enum sfc_udp_tunnel_op_e {
1417         SFC_UDP_TUNNEL_ADD_PORT,
1418         SFC_UDP_TUNNEL_DEL_PORT,
1419 };
1420
1421 static int
1422 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1423                       struct rte_eth_udp_tunnel *tunnel_udp,
1424                       enum sfc_udp_tunnel_op_e op)
1425 {
1426         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1427         efx_tunnel_protocol_t tunnel_proto;
1428         int rc;
1429
1430         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1431                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1432                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1433                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1434
1435         tunnel_proto =
1436                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1437         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1438                 rc = ENOTSUP;
1439                 goto fail_bad_proto;
1440         }
1441
1442         sfc_adapter_lock(sa);
1443
1444         switch (op) {
1445         case SFC_UDP_TUNNEL_ADD_PORT:
1446                 rc = efx_tunnel_config_udp_add(sa->nic,
1447                                                tunnel_udp->udp_port,
1448                                                tunnel_proto);
1449                 break;
1450         case SFC_UDP_TUNNEL_DEL_PORT:
1451                 rc = efx_tunnel_config_udp_remove(sa->nic,
1452                                                   tunnel_udp->udp_port,
1453                                                   tunnel_proto);
1454                 break;
1455         default:
1456                 rc = EINVAL;
1457                 goto fail_bad_op;
1458         }
1459
1460         if (rc != 0)
1461                 goto fail_op;
1462
1463         if (sa->state == SFC_ADAPTER_STARTED) {
1464                 rc = efx_tunnel_reconfigure(sa->nic);
1465                 if (rc == EAGAIN) {
1466                         /*
1467                          * Configuration is accepted by FW and MC reboot
1468                          * is initiated to apply the changes. MC reboot
1469                          * will be handled in a usual way (MC reboot
1470                          * event on management event queue and adapter
1471                          * restart).
1472                          */
1473                         rc = 0;
1474                 } else if (rc != 0) {
1475                         goto fail_reconfigure;
1476                 }
1477         }
1478
1479         sfc_adapter_unlock(sa);
1480         return 0;
1481
1482 fail_reconfigure:
1483         /* Remove/restore entry since the change makes the trouble */
1484         switch (op) {
1485         case SFC_UDP_TUNNEL_ADD_PORT:
1486                 (void)efx_tunnel_config_udp_remove(sa->nic,
1487                                                    tunnel_udp->udp_port,
1488                                                    tunnel_proto);
1489                 break;
1490         case SFC_UDP_TUNNEL_DEL_PORT:
1491                 (void)efx_tunnel_config_udp_add(sa->nic,
1492                                                 tunnel_udp->udp_port,
1493                                                 tunnel_proto);
1494                 break;
1495         }
1496
1497 fail_op:
1498 fail_bad_op:
1499         sfc_adapter_unlock(sa);
1500
1501 fail_bad_proto:
1502         SFC_ASSERT(rc > 0);
1503         return -rc;
1504 }
1505
1506 static int
1507 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1508                             struct rte_eth_udp_tunnel *tunnel_udp)
1509 {
1510         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1511 }
1512
1513 static int
1514 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1515                             struct rte_eth_udp_tunnel *tunnel_udp)
1516 {
1517         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1518 }
1519
1520 /*
1521  * The function is used by the secondary process as well. It must not
1522  * use any process-local pointers from the adapter data.
1523  */
1524 static int
1525 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1526                           struct rte_eth_rss_conf *rss_conf)
1527 {
1528         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1529         struct sfc_rss *rss = &sas->rss;
1530
1531         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1532                 return -ENOTSUP;
1533
1534         /*
1535          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1536          * hence, conversion is done here to derive a correct set of ETH_RSS
1537          * flags which corresponds to the active EFX configuration stored
1538          * locally in 'sfc_adapter' and kept up-to-date
1539          */
1540         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1541         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1542         if (rss_conf->rss_key != NULL)
1543                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1544
1545         return 0;
1546 }
1547
1548 static int
1549 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1550                         struct rte_eth_rss_conf *rss_conf)
1551 {
1552         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1553         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1554         unsigned int efx_hash_types;
1555         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1556         unsigned int n_contexts;
1557         unsigned int mode_i = 0;
1558         unsigned int key_i = 0;
1559         unsigned int i = 0;
1560         int rc = 0;
1561
1562         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1563
1564         if (sfc_sa2shared(sa)->isolated)
1565                 return -ENOTSUP;
1566
1567         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1568                 sfc_err(sa, "RSS is not available");
1569                 return -ENOTSUP;
1570         }
1571
1572         if (rss->channels == 0) {
1573                 sfc_err(sa, "RSS is not configured");
1574                 return -EINVAL;
1575         }
1576
1577         if ((rss_conf->rss_key != NULL) &&
1578             (rss_conf->rss_key_len != sizeof(rss->key))) {
1579                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1580                         sizeof(rss->key));
1581                 return -EINVAL;
1582         }
1583
1584         sfc_adapter_lock(sa);
1585
1586         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1587         if (rc != 0)
1588                 goto fail_rx_hf_rte_to_efx;
1589
1590         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1591                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1592                                            rss->hash_alg, efx_hash_types,
1593                                            B_TRUE);
1594                 if (rc != 0)
1595                         goto fail_scale_mode_set;
1596         }
1597
1598         if (rss_conf->rss_key != NULL) {
1599                 if (sa->state == SFC_ADAPTER_STARTED) {
1600                         for (key_i = 0; key_i < n_contexts; key_i++) {
1601                                 rc = efx_rx_scale_key_set(sa->nic,
1602                                                           contexts[key_i],
1603                                                           rss_conf->rss_key,
1604                                                           sizeof(rss->key));
1605                                 if (rc != 0)
1606                                         goto fail_scale_key_set;
1607                         }
1608                 }
1609
1610                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1611         }
1612
1613         rss->hash_types = efx_hash_types;
1614
1615         sfc_adapter_unlock(sa);
1616
1617         return 0;
1618
1619 fail_scale_key_set:
1620         for (i = 0; i < key_i; i++) {
1621                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1622                                          sizeof(rss->key)) != 0)
1623                         sfc_err(sa, "failed to restore RSS key");
1624         }
1625
1626 fail_scale_mode_set:
1627         for (i = 0; i < mode_i; i++) {
1628                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1629                                           EFX_RX_HASHALG_TOEPLITZ,
1630                                           rss->hash_types, B_TRUE) != 0)
1631                         sfc_err(sa, "failed to restore RSS mode");
1632         }
1633
1634 fail_rx_hf_rte_to_efx:
1635         sfc_adapter_unlock(sa);
1636         return -rc;
1637 }
1638
1639 /*
1640  * The function is used by the secondary process as well. It must not
1641  * use any process-local pointers from the adapter data.
1642  */
1643 static int
1644 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1645                        struct rte_eth_rss_reta_entry64 *reta_conf,
1646                        uint16_t reta_size)
1647 {
1648         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1649         struct sfc_rss *rss = &sas->rss;
1650         int entry;
1651
1652         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1653                 return -ENOTSUP;
1654
1655         if (rss->channels == 0)
1656                 return -EINVAL;
1657
1658         if (reta_size != EFX_RSS_TBL_SIZE)
1659                 return -EINVAL;
1660
1661         for (entry = 0; entry < reta_size; entry++) {
1662                 int grp = entry / RTE_RETA_GROUP_SIZE;
1663                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1664
1665                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1666                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1667         }
1668
1669         return 0;
1670 }
1671
1672 static int
1673 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1674                         struct rte_eth_rss_reta_entry64 *reta_conf,
1675                         uint16_t reta_size)
1676 {
1677         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1678         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1679         unsigned int *rss_tbl_new;
1680         uint16_t entry;
1681         int rc = 0;
1682
1683
1684         if (sfc_sa2shared(sa)->isolated)
1685                 return -ENOTSUP;
1686
1687         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1688                 sfc_err(sa, "RSS is not available");
1689                 return -ENOTSUP;
1690         }
1691
1692         if (rss->channels == 0) {
1693                 sfc_err(sa, "RSS is not configured");
1694                 return -EINVAL;
1695         }
1696
1697         if (reta_size != EFX_RSS_TBL_SIZE) {
1698                 sfc_err(sa, "RETA size is wrong (should be %u)",
1699                         EFX_RSS_TBL_SIZE);
1700                 return -EINVAL;
1701         }
1702
1703         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1704         if (rss_tbl_new == NULL)
1705                 return -ENOMEM;
1706
1707         sfc_adapter_lock(sa);
1708
1709         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1710
1711         for (entry = 0; entry < reta_size; entry++) {
1712                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1713                 struct rte_eth_rss_reta_entry64 *grp;
1714
1715                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1716
1717                 if (grp->mask & (1ull << grp_idx)) {
1718                         if (grp->reta[grp_idx] >= rss->channels) {
1719                                 rc = EINVAL;
1720                                 goto bad_reta_entry;
1721                         }
1722                         rss_tbl_new[entry] = grp->reta[grp_idx];
1723                 }
1724         }
1725
1726         if (sa->state == SFC_ADAPTER_STARTED) {
1727                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1728                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1729                 if (rc != 0)
1730                         goto fail_scale_tbl_set;
1731         }
1732
1733         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1734
1735 fail_scale_tbl_set:
1736 bad_reta_entry:
1737         sfc_adapter_unlock(sa);
1738
1739         rte_free(rss_tbl_new);
1740
1741         SFC_ASSERT(rc >= 0);
1742         return -rc;
1743 }
1744
1745 static int
1746 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1747                      const struct rte_flow_ops **ops)
1748 {
1749         *ops = &sfc_flow_ops;
1750         return 0;
1751 }
1752
1753 static int
1754 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1755 {
1756         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1757
1758         /*
1759          * If Rx datapath does not provide callback to check mempool,
1760          * all pools are supported.
1761          */
1762         if (sap->dp_rx->pool_ops_supported == NULL)
1763                 return 1;
1764
1765         return sap->dp_rx->pool_ops_supported(pool);
1766 }
1767
1768 static int
1769 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1770 {
1771         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1772         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1773         struct sfc_rxq_info *rxq_info;
1774
1775         SFC_ASSERT(queue_id < sas->rxq_count);
1776         rxq_info = &sas->rxq_info[queue_id];
1777
1778         return sap->dp_rx->intr_enable(rxq_info->dp);
1779 }
1780
1781 static int
1782 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1783 {
1784         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1785         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1786         struct sfc_rxq_info *rxq_info;
1787
1788         SFC_ASSERT(queue_id < sas->rxq_count);
1789         rxq_info = &sas->rxq_info[queue_id];
1790
1791         return sap->dp_rx->intr_disable(rxq_info->dp);
1792 }
1793
1794 static const struct eth_dev_ops sfc_eth_dev_ops = {
1795         .dev_configure                  = sfc_dev_configure,
1796         .dev_start                      = sfc_dev_start,
1797         .dev_stop                       = sfc_dev_stop,
1798         .dev_set_link_up                = sfc_dev_set_link_up,
1799         .dev_set_link_down              = sfc_dev_set_link_down,
1800         .dev_close                      = sfc_dev_close,
1801         .promiscuous_enable             = sfc_dev_promisc_enable,
1802         .promiscuous_disable            = sfc_dev_promisc_disable,
1803         .allmulticast_enable            = sfc_dev_allmulti_enable,
1804         .allmulticast_disable           = sfc_dev_allmulti_disable,
1805         .link_update                    = sfc_dev_link_update,
1806         .stats_get                      = sfc_stats_get,
1807         .stats_reset                    = sfc_stats_reset,
1808         .xstats_get                     = sfc_xstats_get,
1809         .xstats_reset                   = sfc_stats_reset,
1810         .xstats_get_names               = sfc_xstats_get_names,
1811         .dev_infos_get                  = sfc_dev_infos_get,
1812         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1813         .mtu_set                        = sfc_dev_set_mtu,
1814         .rx_queue_start                 = sfc_rx_queue_start,
1815         .rx_queue_stop                  = sfc_rx_queue_stop,
1816         .tx_queue_start                 = sfc_tx_queue_start,
1817         .tx_queue_stop                  = sfc_tx_queue_stop,
1818         .rx_queue_setup                 = sfc_rx_queue_setup,
1819         .rx_queue_release               = sfc_rx_queue_release,
1820         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1821         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1822         .tx_queue_setup                 = sfc_tx_queue_setup,
1823         .tx_queue_release               = sfc_tx_queue_release,
1824         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1825         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1826         .mac_addr_set                   = sfc_mac_addr_set,
1827         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1828         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1829         .reta_update                    = sfc_dev_rss_reta_update,
1830         .reta_query                     = sfc_dev_rss_reta_query,
1831         .rss_hash_update                = sfc_dev_rss_hash_update,
1832         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1833         .flow_ops_get                   = sfc_dev_flow_ops_get,
1834         .set_mc_addr_list               = sfc_set_mc_addr_list,
1835         .rxq_info_get                   = sfc_rx_queue_info_get,
1836         .txq_info_get                   = sfc_tx_queue_info_get,
1837         .fw_version_get                 = sfc_fw_version_get,
1838         .xstats_get_by_id               = sfc_xstats_get_by_id,
1839         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1840         .pool_ops_supported             = sfc_pool_ops_supported,
1841 };
1842
1843 /**
1844  * Duplicate a string in potentially shared memory required for
1845  * multi-process support.
1846  *
1847  * strdup() allocates from process-local heap/memory.
1848  */
1849 static char *
1850 sfc_strdup(const char *str)
1851 {
1852         size_t size;
1853         char *copy;
1854
1855         if (str == NULL)
1856                 return NULL;
1857
1858         size = strlen(str) + 1;
1859         copy = rte_malloc(__func__, size, 0);
1860         if (copy != NULL)
1861                 rte_memcpy(copy, str, size);
1862
1863         return copy;
1864 }
1865
1866 static int
1867 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1868 {
1869         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1870         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1871         const struct sfc_dp_rx *dp_rx;
1872         const struct sfc_dp_tx *dp_tx;
1873         const efx_nic_cfg_t *encp;
1874         unsigned int avail_caps = 0;
1875         const char *rx_name = NULL;
1876         const char *tx_name = NULL;
1877         int rc;
1878
1879         switch (sa->family) {
1880         case EFX_FAMILY_HUNTINGTON:
1881         case EFX_FAMILY_MEDFORD:
1882         case EFX_FAMILY_MEDFORD2:
1883                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1884                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1885                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1886                 break;
1887         case EFX_FAMILY_RIVERHEAD:
1888                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1889                 break;
1890         default:
1891                 break;
1892         }
1893
1894         encp = efx_nic_cfg_get(sa->nic);
1895         if (encp->enc_rx_es_super_buffer_supported)
1896                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1897
1898         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1899                                 sfc_kvarg_string_handler, &rx_name);
1900         if (rc != 0)
1901                 goto fail_kvarg_rx_datapath;
1902
1903         if (rx_name != NULL) {
1904                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1905                 if (dp_rx == NULL) {
1906                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1907                         rc = ENOENT;
1908                         goto fail_dp_rx;
1909                 }
1910                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1911                         sfc_err(sa,
1912                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1913                                 rx_name);
1914                         rc = EINVAL;
1915                         goto fail_dp_rx_caps;
1916                 }
1917         } else {
1918                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1919                 if (dp_rx == NULL) {
1920                         sfc_err(sa, "Rx datapath by caps %#x not found",
1921                                 avail_caps);
1922                         rc = ENOENT;
1923                         goto fail_dp_rx;
1924                 }
1925         }
1926
1927         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1928         if (sas->dp_rx_name == NULL) {
1929                 rc = ENOMEM;
1930                 goto fail_dp_rx_name;
1931         }
1932
1933         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1934
1935         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1936                                 sfc_kvarg_string_handler, &tx_name);
1937         if (rc != 0)
1938                 goto fail_kvarg_tx_datapath;
1939
1940         if (tx_name != NULL) {
1941                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1942                 if (dp_tx == NULL) {
1943                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1944                         rc = ENOENT;
1945                         goto fail_dp_tx;
1946                 }
1947                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1948                         sfc_err(sa,
1949                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1950                                 tx_name);
1951                         rc = EINVAL;
1952                         goto fail_dp_tx_caps;
1953                 }
1954         } else {
1955                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1956                 if (dp_tx == NULL) {
1957                         sfc_err(sa, "Tx datapath by caps %#x not found",
1958                                 avail_caps);
1959                         rc = ENOENT;
1960                         goto fail_dp_tx;
1961                 }
1962         }
1963
1964         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1965         if (sas->dp_tx_name == NULL) {
1966                 rc = ENOMEM;
1967                 goto fail_dp_tx_name;
1968         }
1969
1970         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1971
1972         sa->priv.dp_rx = dp_rx;
1973         sa->priv.dp_tx = dp_tx;
1974
1975         dev->rx_pkt_burst = dp_rx->pkt_burst;
1976         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1977         dev->tx_pkt_burst = dp_tx->pkt_burst;
1978
1979         dev->rx_queue_count = sfc_rx_queue_count;
1980         dev->rx_descriptor_done = sfc_rx_descriptor_done;
1981         dev->rx_descriptor_status = sfc_rx_descriptor_status;
1982         dev->tx_descriptor_status = sfc_tx_descriptor_status;
1983         dev->dev_ops = &sfc_eth_dev_ops;
1984
1985         return 0;
1986
1987 fail_dp_tx_name:
1988 fail_dp_tx_caps:
1989 fail_dp_tx:
1990 fail_kvarg_tx_datapath:
1991         rte_free(sas->dp_rx_name);
1992         sas->dp_rx_name = NULL;
1993
1994 fail_dp_rx_name:
1995 fail_dp_rx_caps:
1996 fail_dp_rx:
1997 fail_kvarg_rx_datapath:
1998         return rc;
1999 }
2000
2001 static void
2002 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2003 {
2004         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2005         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2006
2007         dev->dev_ops = NULL;
2008         dev->tx_pkt_prepare = NULL;
2009         dev->rx_pkt_burst = NULL;
2010         dev->tx_pkt_burst = NULL;
2011
2012         rte_free(sas->dp_tx_name);
2013         sas->dp_tx_name = NULL;
2014         sa->priv.dp_tx = NULL;
2015
2016         rte_free(sas->dp_rx_name);
2017         sas->dp_rx_name = NULL;
2018         sa->priv.dp_rx = NULL;
2019 }
2020
2021 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2022         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2023         .reta_query                     = sfc_dev_rss_reta_query,
2024         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2025         .rxq_info_get                   = sfc_rx_queue_info_get,
2026         .txq_info_get                   = sfc_tx_queue_info_get,
2027 };
2028
2029 static int
2030 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2031 {
2032         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2033         struct sfc_adapter_priv *sap;
2034         const struct sfc_dp_rx *dp_rx;
2035         const struct sfc_dp_tx *dp_tx;
2036         int rc;
2037
2038         /*
2039          * Allocate process private data from heap, since it should not
2040          * be located in shared memory allocated using rte_malloc() API.
2041          */
2042         sap = calloc(1, sizeof(*sap));
2043         if (sap == NULL) {
2044                 rc = ENOMEM;
2045                 goto fail_alloc_priv;
2046         }
2047
2048         sap->logtype_main = logtype_main;
2049
2050         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2051         if (dp_rx == NULL) {
2052                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2053                         "cannot find %s Rx datapath", sas->dp_rx_name);
2054                 rc = ENOENT;
2055                 goto fail_dp_rx;
2056         }
2057         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2058                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2059                         "%s Rx datapath does not support multi-process",
2060                         sas->dp_rx_name);
2061                 rc = EINVAL;
2062                 goto fail_dp_rx_multi_process;
2063         }
2064
2065         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2066         if (dp_tx == NULL) {
2067                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2068                         "cannot find %s Tx datapath", sas->dp_tx_name);
2069                 rc = ENOENT;
2070                 goto fail_dp_tx;
2071         }
2072         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2073                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2074                         "%s Tx datapath does not support multi-process",
2075                         sas->dp_tx_name);
2076                 rc = EINVAL;
2077                 goto fail_dp_tx_multi_process;
2078         }
2079
2080         sap->dp_rx = dp_rx;
2081         sap->dp_tx = dp_tx;
2082
2083         dev->process_private = sap;
2084         dev->rx_pkt_burst = dp_rx->pkt_burst;
2085         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2086         dev->tx_pkt_burst = dp_tx->pkt_burst;
2087         dev->rx_queue_count = sfc_rx_queue_count;
2088         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2089         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2090         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2091         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2092
2093         return 0;
2094
2095 fail_dp_tx_multi_process:
2096 fail_dp_tx:
2097 fail_dp_rx_multi_process:
2098 fail_dp_rx:
2099         free(sap);
2100
2101 fail_alloc_priv:
2102         return rc;
2103 }
2104
2105 static void
2106 sfc_register_dp(void)
2107 {
2108         /* Register once */
2109         if (TAILQ_EMPTY(&sfc_dp_head)) {
2110                 /* Prefer EF10 datapath */
2111                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2112                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2113                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2114                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2115
2116                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2117                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2118                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2119                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2120         }
2121 }
2122
2123 static int
2124 sfc_eth_dev_init(struct rte_eth_dev *dev)
2125 {
2126         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2127         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2128         uint32_t logtype_main;
2129         struct sfc_adapter *sa;
2130         int rc;
2131         const efx_nic_cfg_t *encp;
2132         const struct rte_ether_addr *from;
2133         int ret;
2134
2135         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2136                         SFC_EFX_DEV_CLASS_NET) {
2137                 SFC_GENERIC_LOG(DEBUG,
2138                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2139                 return 1;
2140         }
2141
2142         sfc_register_dp();
2143
2144         logtype_main = sfc_register_logtype(&pci_dev->addr,
2145                                             SFC_LOGTYPE_MAIN_STR,
2146                                             RTE_LOG_NOTICE);
2147
2148         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2149                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2150
2151         /* Required for logging */
2152         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2153                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2154                         pci_dev->addr.domain, pci_dev->addr.bus,
2155                         pci_dev->addr.devid, pci_dev->addr.function,
2156                         dev->data->port_id);
2157         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2158                 SFC_GENERIC_LOG(ERR,
2159                         "reserved log prefix is too short for " PCI_PRI_FMT,
2160                         pci_dev->addr.domain, pci_dev->addr.bus,
2161                         pci_dev->addr.devid, pci_dev->addr.function);
2162                 return -EINVAL;
2163         }
2164         sas->pci_addr = pci_dev->addr;
2165         sas->port_id = dev->data->port_id;
2166
2167         /*
2168          * Allocate process private data from heap, since it should not
2169          * be located in shared memory allocated using rte_malloc() API.
2170          */
2171         sa = calloc(1, sizeof(*sa));
2172         if (sa == NULL) {
2173                 rc = ENOMEM;
2174                 goto fail_alloc_sa;
2175         }
2176
2177         dev->process_private = sa;
2178
2179         /* Required for logging */
2180         sa->priv.shared = sas;
2181         sa->priv.logtype_main = logtype_main;
2182
2183         sa->eth_dev = dev;
2184
2185         /* Copy PCI device info to the dev->data */
2186         rte_eth_copy_pci_info(dev, pci_dev);
2187         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2188         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2189
2190         rc = sfc_kvargs_parse(sa);
2191         if (rc != 0)
2192                 goto fail_kvargs_parse;
2193
2194         sfc_log_init(sa, "entry");
2195
2196         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2197         if (dev->data->mac_addrs == NULL) {
2198                 rc = ENOMEM;
2199                 goto fail_mac_addrs;
2200         }
2201
2202         sfc_adapter_lock_init(sa);
2203         sfc_adapter_lock(sa);
2204
2205         sfc_log_init(sa, "probing");
2206         rc = sfc_probe(sa);
2207         if (rc != 0)
2208                 goto fail_probe;
2209
2210         sfc_log_init(sa, "set device ops");
2211         rc = sfc_eth_dev_set_ops(dev);
2212         if (rc != 0)
2213                 goto fail_set_ops;
2214
2215         sfc_log_init(sa, "attaching");
2216         rc = sfc_attach(sa);
2217         if (rc != 0)
2218                 goto fail_attach;
2219
2220         encp = efx_nic_cfg_get(sa->nic);
2221
2222         /*
2223          * The arguments are really reverse order in comparison to
2224          * Linux kernel. Copy from NIC config to Ethernet device data.
2225          */
2226         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2227         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2228
2229         sfc_adapter_unlock(sa);
2230
2231         sfc_log_init(sa, "done");
2232         return 0;
2233
2234 fail_attach:
2235         sfc_eth_dev_clear_ops(dev);
2236
2237 fail_set_ops:
2238         sfc_unprobe(sa);
2239
2240 fail_probe:
2241         sfc_adapter_unlock(sa);
2242         sfc_adapter_lock_fini(sa);
2243         rte_free(dev->data->mac_addrs);
2244         dev->data->mac_addrs = NULL;
2245
2246 fail_mac_addrs:
2247         sfc_kvargs_cleanup(sa);
2248
2249 fail_kvargs_parse:
2250         sfc_log_init(sa, "failed %d", rc);
2251         dev->process_private = NULL;
2252         free(sa);
2253
2254 fail_alloc_sa:
2255         SFC_ASSERT(rc > 0);
2256         return -rc;
2257 }
2258
2259 static int
2260 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2261 {
2262         sfc_dev_close(dev);
2263
2264         return 0;
2265 }
2266
2267 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2268         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2269         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2270         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2271         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2272         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2273         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2274         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2275         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2276         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2277         { .vendor_id = 0 /* sentinel */ }
2278 };
2279
2280 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2281         struct rte_pci_device *pci_dev)
2282 {
2283         return rte_eth_dev_pci_generic_probe(pci_dev,
2284                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2285 }
2286
2287 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2288 {
2289         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2290 }
2291
2292 static struct rte_pci_driver sfc_efx_pmd = {
2293         .id_table = pci_id_sfc_efx_map,
2294         .drv_flags =
2295                 RTE_PCI_DRV_INTR_LSC |
2296                 RTE_PCI_DRV_NEED_MAPPING,
2297         .probe = sfc_eth_dev_pci_probe,
2298         .remove = sfc_eth_dev_pci_remove,
2299 };
2300
2301 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2302 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2303 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2304 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2305         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2306         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2307         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2308         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2309         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2310         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2311
2312 RTE_INIT(sfc_driver_register_logtype)
2313 {
2314         int ret;
2315
2316         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2317                                                    RTE_LOG_NOTICE);
2318         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2319 }