drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2020 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         /*
  50          * Return value of the callback is likely supposed to be
  51          * equal to or greater than 0, nevertheless, if an error
  52          * occurs, it will be desirable to pass it to the caller
  53          */
  54         if ((fw_version == NULL) || (fw_size == 0))
  55                 return -EINVAL;
  56
  57         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  58         if (rc != 0)
  59                 return -rc;
  60
  61         ret = snprintf(fw_version, fw_size,
  62                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  63                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  64                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  65         if (ret < 0)
  66                 return ret;
  67
  68         if (enfi.enfi_dpcpu_fw_ids_valid) {
  69                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  70                 int ret_extra;
  71
  72                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  73                                      fw_size - dpcpu_fw_ids_offset,
  74                                      " rx%" PRIx16 " tx%" PRIx16,
  75                                      enfi.enfi_rx_dpcpu_fw_id,
  76                                      enfi.enfi_tx_dpcpu_fw_id);
  77                 if (ret_extra < 0)
  78                         return ret_extra;
  79
  80                 ret += ret_extra;
  81         }
  82
  83         if (fw_size < (size_t)(++ret))
  84                 return ret;
  85         else
  86                 return 0;
  87 }
  88
  89 static int
  90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  91 {
  92         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  93         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  94         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  95         struct sfc_rss *rss = &sas->rss;
  96         struct sfc_mae *mae = &sa->mae;
  97         uint64_t txq_offloads_def = 0;
  98
  99         sfc_log_init(sa, "entry");
 100
 101         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 102         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 103
 104         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 105
 106         dev_info->max_vfs = sa->sriov.num_vfs;
 107
 108         /* Autonegotiation may be disabled */
 109         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 110         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 111                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 112         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 113                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 114         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 115                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 116         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 117                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 118         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 119                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 120         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 121                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 122
 123         dev_info->max_rx_queues = sa->rxq_max;
 124         dev_info->max_tx_queues = sa->txq_max;
 125
 126         /* By default packets are dropped if no descriptors are available */
 127         dev_info->default_rxconf.rx_drop_en = 1;
 128
 129         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 130
 131         /*
 132          * rx_offload_capa includes both device and queue offloads since
 133          * the latter may be requested on a per device basis which makes
 134          * sense when some offloads are needed to be set on all queues.
 135          */
 136         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 137                                     dev_info->rx_queue_offload_capa;
 138
 139         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 140
 141         /*
 142          * tx_offload_capa includes both device and queue offloads since
 143          * the latter may be requested on a per device basis which makes
 144          * sense when some offloads are needed to be set on all queues.
 145          */
 146         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 147                                     dev_info->tx_queue_offload_capa;
 148
 149         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 150                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 151
 152         dev_info->default_txconf.offloads |= txq_offloads_def;
 153
 154         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 155                 uint64_t rte_hf = 0;
 156                 unsigned int i;
 157
 158                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 159                         rte_hf |= rss->hf_map[i].rte;
 160
 161                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 162                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 163                 dev_info->flow_type_rss_offloads = rte_hf;
 164         }
 165
 166         /* Initialize to hardware limits */
 167         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 168         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 169         /* The RXQ hardware requires that the descriptor count is a power
 170          * of 2, but rx_desc_lim cannot properly describe that constraint.
 171          */
 172         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 173
 174         /* Initialize to hardware limits */
 175         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 176         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 177         /*
 178          * The TXQ hardware requires that the descriptor count is a power
 179          * of 2, but tx_desc_lim cannot properly describe that constraint
 180          */
 181         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 182
 183         if (sap->dp_rx->get_dev_info != NULL)
 184                 sap->dp_rx->get_dev_info(dev_info);
 185         if (sap->dp_tx->get_dev_info != NULL)
 186                 sap->dp_tx->get_dev_info(dev_info);
 187
 188         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 189                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 190
 191         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
 192                 dev_info->switch_info.name = dev->device->driver->name;
 193                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 194                 dev_info->switch_info.port_id = mae->switch_port_id;
 195         }
 196
 197         return 0;
 198 }
 199
 200 static const uint32_t *
 201 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 202 {
 203         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 204
 205         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 206 }
 207
 208 static int
 209 sfc_dev_configure(struct rte_eth_dev *dev)
 210 {
 211         struct rte_eth_dev_data *dev_data = dev->data;
 212         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 213         int rc;
 214
 215         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 216                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 217
 218         sfc_adapter_lock(sa);
 219         switch (sa->state) {
 220         case SFC_ADAPTER_CONFIGURED:
 221                 /* FALLTHROUGH */
 222         case SFC_ADAPTER_INITIALIZED:
 223                 rc = sfc_configure(sa);
 224                 break;
 225         default:
 226                 sfc_err(sa, "unexpected adapter state %u to configure",
 227                         sa->state);
 228                 rc = EINVAL;
 229                 break;
 230         }
 231         sfc_adapter_unlock(sa);
 232
 233         sfc_log_init(sa, "done %d", rc);
 234         SFC_ASSERT(rc >= 0);
 235         return -rc;
 236 }
 237
 238 static int
 239 sfc_dev_start(struct rte_eth_dev *dev)
 240 {
 241         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 242         int rc;
 243
 244         sfc_log_init(sa, "entry");
 245
 246         sfc_adapter_lock(sa);
 247         rc = sfc_start(sa);
 248         sfc_adapter_unlock(sa);
 249
 250         sfc_log_init(sa, "done %d", rc);
 251         SFC_ASSERT(rc >= 0);
 252         return -rc;
 253 }
 254
 255 static int
 256 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 257 {
 258         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 259         struct rte_eth_link current_link;
 260         int ret;
 261
 262         sfc_log_init(sa, "entry");
 263
 264         if (sa->state != SFC_ADAPTER_STARTED) {
 265                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 266         } else if (wait_to_complete) {
 267                 efx_link_mode_t link_mode;
 268
 269                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 270                         link_mode = EFX_LINK_UNKNOWN;
 271                 sfc_port_link_mode_to_info(link_mode, &current_link);
 272
 273         } else {
 274                 sfc_ev_mgmt_qpoll(sa);
 275                 rte_eth_linkstatus_get(dev, &current_link);
 276         }
 277
 278         ret = rte_eth_linkstatus_set(dev, &current_link);
 279         if (ret == 0)
 280                 sfc_notice(sa, "Link status is %s",
 281                            current_link.link_status ? "UP" : "DOWN");
 282
 283         return ret;
 284 }
 285
 286 static int
 287 sfc_dev_stop(struct rte_eth_dev *dev)
 288 {
 289         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 290
 291         sfc_log_init(sa, "entry");
 292
 293         sfc_adapter_lock(sa);
 294         sfc_stop(sa);
 295         sfc_adapter_unlock(sa);
 296
 297         sfc_log_init(sa, "done");
 298
 299         return 0;
 300 }
 301
 302 static int
 303 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 304 {
 305         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 306         int rc;
 307
 308         sfc_log_init(sa, "entry");
 309
 310         sfc_adapter_lock(sa);
 311         rc = sfc_start(sa);
 312         sfc_adapter_unlock(sa);
 313
 314         SFC_ASSERT(rc >= 0);
 315         return -rc;
 316 }
 317
 318 static int
 319 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 320 {
 321         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 322
 323         sfc_log_init(sa, "entry");
 324
 325         sfc_adapter_lock(sa);
 326         sfc_stop(sa);
 327         sfc_adapter_unlock(sa);
 328
 329         return 0;
 330 }
 331
 332 static void
 333 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 334 {
 335         free(dev->process_private);
 336         rte_eth_dev_release_port(dev);
 337 }
 338
 339 static int
 340 sfc_dev_close(struct rte_eth_dev *dev)
 341 {
 342         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 343
 344         sfc_log_init(sa, "entry");
 345
 346         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 347                 sfc_eth_dev_secondary_clear_ops(dev);
 348                 return 0;
 349         }
 350
 351         sfc_adapter_lock(sa);
 352         switch (sa->state) {
 353         case SFC_ADAPTER_STARTED:
 354                 sfc_stop(sa);
 355                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 356                 /* FALLTHROUGH */
 357         case SFC_ADAPTER_CONFIGURED:
 358                 sfc_close(sa);
 359                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 360                 /* FALLTHROUGH */
 361         case SFC_ADAPTER_INITIALIZED:
 362                 break;
 363         default:
 364                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 365                 break;
 366         }
 367
 368         /*
 369          * Cleanup all resources.
 370          * Rollback primary process sfc_eth_dev_init() below.
 371          */
 372
 373         sfc_eth_dev_clear_ops(dev);
 374
 375         sfc_detach(sa);
 376         sfc_unprobe(sa);
 377
 378         sfc_kvargs_cleanup(sa);
 379
 380         sfc_adapter_unlock(sa);
 381         sfc_adapter_lock_fini(sa);
 382
 383         sfc_log_init(sa, "done");
 384
 385         /* Required for logging, so cleanup last */
 386         sa->eth_dev = NULL;
 387
 388         free(sa);
 389
 390         return 0;
 391 }
 392
 393 static int
 394 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 395                    boolean_t enabled)
 396 {
 397         struct sfc_port *port;
 398         boolean_t *toggle;
 399         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 400         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 401         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 402         int rc = 0;
 403
 404         sfc_adapter_lock(sa);
 405
 406         port = &sa->port;
 407         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 408
 409         if (*toggle != enabled) {
 410                 *toggle = enabled;
 411
 412                 if (sfc_sa2shared(sa)->isolated) {
 413                         sfc_warn(sa, "isolated mode is active on the port");
 414                         sfc_warn(sa, "the change is to be applied on the next "
 415                                      "start provided that isolated mode is "
 416                                      "disabled prior the next start");
 417                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 418                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 419                         *toggle = !(enabled);
 420                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 421                                  ((enabled) ? "enable" : "disable"), desc, rc);
 422
 423                         /*
 424                          * For promiscuous and all-multicast filters a
 425                          * permission failure should be reported as an
 426                          * unsupported filter.
 427                          */
 428                         if (rc == EPERM)
 429                                 rc = ENOTSUP;
 430                 }
 431         }
 432
 433         sfc_adapter_unlock(sa);
 434         return rc;
 435 }
 436
 437 static int
 438 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 439 {
 440         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 441
 442         SFC_ASSERT(rc >= 0);
 443         return -rc;
 444 }
 445
 446 static int
 447 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 448 {
 449         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 450
 451         SFC_ASSERT(rc >= 0);
 452         return -rc;
 453 }
 454
 455 static int
 456 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 457 {
 458         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 459
 460         SFC_ASSERT(rc >= 0);
 461         return -rc;
 462 }
 463
 464 static int
 465 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 466 {
 467         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 468
 469         SFC_ASSERT(rc >= 0);
 470         return -rc;
 471 }
 472
 473 static int
 474 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 475                    uint16_t nb_rx_desc, unsigned int socket_id,
 476                    const struct rte_eth_rxconf *rx_conf,
 477                    struct rte_mempool *mb_pool)
 478 {
 479         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 480         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 481         int rc;
 482
 483         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 484                      rx_queue_id, nb_rx_desc, socket_id);
 485
 486         sfc_adapter_lock(sa);
 487
 488         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 489                           rx_conf, mb_pool);
 490         if (rc != 0)
 491                 goto fail_rx_qinit;
 492
 493         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 494
 495         sfc_adapter_unlock(sa);
 496
 497         return 0;
 498
 499 fail_rx_qinit:
 500         sfc_adapter_unlock(sa);
 501         SFC_ASSERT(rc > 0);
 502         return -rc;
 503 }
 504
 505 static void
 506 sfc_rx_queue_release(void *queue)
 507 {
 508         struct sfc_dp_rxq *dp_rxq = queue;
 509         struct sfc_rxq *rxq;
 510         struct sfc_adapter *sa;
 511         unsigned int sw_index;
 512
 513         if (dp_rxq == NULL)
 514                 return;
 515
 516         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 517         sa = rxq->evq->sa;
 518         sfc_adapter_lock(sa);
 519
 520         sw_index = dp_rxq->dpq.queue_id;
 521
 522         sfc_log_init(sa, "RxQ=%u", sw_index);
 523
 524         sfc_rx_qfini(sa, sw_index);
 525
 526         sfc_adapter_unlock(sa);
 527 }
 528
 529 static int
 530 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 531                    uint16_t nb_tx_desc, unsigned int socket_id,
 532                    const struct rte_eth_txconf *tx_conf)
 533 {
 534         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 535         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 536         int rc;
 537
 538         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 539                      tx_queue_id, nb_tx_desc, socket_id);
 540
 541         sfc_adapter_lock(sa);
 542
 543         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 544         if (rc != 0)
 545                 goto fail_tx_qinit;
 546
 547         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
 548
 549         sfc_adapter_unlock(sa);
 550         return 0;
 551
 552 fail_tx_qinit:
 553         sfc_adapter_unlock(sa);
 554         SFC_ASSERT(rc > 0);
 555         return -rc;
 556 }
 557
 558 static void
 559 sfc_tx_queue_release(void *queue)
 560 {
 561         struct sfc_dp_txq *dp_txq = queue;
 562         struct sfc_txq *txq;
 563         unsigned int sw_index;
 564         struct sfc_adapter *sa;
 565
 566         if (dp_txq == NULL)
 567                 return;
 568
 569         txq = sfc_txq_by_dp_txq(dp_txq);
 570         sw_index = dp_txq->dpq.queue_id;
 571
 572         SFC_ASSERT(txq->evq != NULL);
 573         sa = txq->evq->sa;
 574
 575         sfc_log_init(sa, "TxQ = %u", sw_index);
 576
 577         sfc_adapter_lock(sa);
 578
 579         sfc_tx_qfini(sa, sw_index);
 580
 581         sfc_adapter_unlock(sa);
 582 }
 583
 584 /*
 585  * Some statistics are computed as A - B where A and B each increase
 586  * monotonically with some hardware counter(s) and the counters are read
 587  * asynchronously.
 588  *
 589  * If packet X is counted in A, but not counted in B yet, computed value is
 590  * greater than real.
 591  *
 592  * If packet X is not counted in A at the moment of reading the counter,
 593  * but counted in B at the moment of reading the counter, computed value
 594  * is less than real.
 595  *
 596  * However, counter which grows backward is worse evil than slightly wrong
 597  * value. So, let's try to guarantee that it never happens except may be
 598  * the case when the MAC stats are zeroed as a result of a NIC reset.
 599  */
 600 static void
 601 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 602 {
 603         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 604                 *stat = newval;
 605 }
 606
 607 static int
 608 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 609 {
 610         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 611         struct sfc_port *port = &sa->port;
 612         uint64_t *mac_stats;
 613         int ret;
 614
 615         rte_spinlock_lock(&port->mac_stats_lock);
 616
 617         ret = sfc_port_update_mac_stats(sa);
 618         if (ret != 0)
 619                 goto unlock;
 620
 621         mac_stats = port->mac_stats_buf;
 622
 623         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 624                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 625                 stats->ipackets =
 626                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 627                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 628                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 629                 stats->opackets =
 630                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 631                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 632                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 633                 stats->ibytes =
 634                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 635                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 636                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 637                 stats->obytes =
 638                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 639                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 640                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 641                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 642                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 643
 644                 /* CRC is included in these stats, but shouldn't be */
 645                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 646                 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 647         } else {
 648                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 649                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 650                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 651
 652                 /* CRC is included in these stats, but shouldn't be */
 653                 stats->ibytes -= mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 654                 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 655
 656                 /*
 657                  * Take into account stats which are whenever supported
 658                  * on EF10. If some stat is not supported by current
 659                  * firmware variant or HW revision, it is guaranteed
 660                  * to be zero in mac_stats.
 661                  */
 662                 stats->imissed =
 663                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 664                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 665                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 666                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 667                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 668                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 669                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 670                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 671                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 672                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 673                 stats->ierrors =
 674                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 675                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 676                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 677                 /* no oerrors counters supported on EF10 */
 678
 679                 /* Exclude missed, errors and pauses from Rx packets */
 680                 sfc_update_diff_stat(&port->ipackets,
 681                         mac_stats[EFX_MAC_RX_PKTS] -
 682                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 683                         stats->imissed - stats->ierrors);
 684                 stats->ipackets = port->ipackets;
 685         }
 686
 687 unlock:
 688         rte_spinlock_unlock(&port->mac_stats_lock);
 689         SFC_ASSERT(ret >= 0);
 690         return -ret;
 691 }
 692
 693 static int
 694 sfc_stats_reset(struct rte_eth_dev *dev)
 695 {
 696         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 697         struct sfc_port *port = &sa->port;
 698         int rc;
 699
 700         if (sa->state != SFC_ADAPTER_STARTED) {
 701                 /*
 702                  * The operation cannot be done if port is not started; it
 703                  * will be scheduled to be done during the next port start
 704                  */
 705                 port->mac_stats_reset_pending = B_TRUE;
 706                 return 0;
 707         }
 708
 709         rc = sfc_port_reset_mac_stats(sa);
 710         if (rc != 0)
 711                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 712
 713         SFC_ASSERT(rc >= 0);
 714         return -rc;
 715 }
 716
 717 static int
 718 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 719                unsigned int xstats_count)
 720 {
 721         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 722         struct sfc_port *port = &sa->port;
 723         uint64_t *mac_stats;
 724         int rc;
 725         unsigned int i;
 726         int nstats = 0;
 727
 728         rte_spinlock_lock(&port->mac_stats_lock);
 729
 730         rc = sfc_port_update_mac_stats(sa);
 731         if (rc != 0) {
 732                 SFC_ASSERT(rc > 0);
 733                 nstats = -rc;
 734                 goto unlock;
 735         }
 736
 737         mac_stats = port->mac_stats_buf;
 738
 739         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 740                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 741                         if (xstats != NULL && nstats < (int)xstats_count) {
 742                                 xstats[nstats].id = nstats;
 743                                 xstats[nstats].value = mac_stats[i];
 744                         }
 745                         nstats++;
 746                 }
 747         }
 748
 749 unlock:
 750         rte_spinlock_unlock(&port->mac_stats_lock);
 751
 752         return nstats;
 753 }
 754
 755 static int
 756 sfc_xstats_get_names(struct rte_eth_dev *dev,
 757                      struct rte_eth_xstat_name *xstats_names,
 758                      unsigned int xstats_count)
 759 {
 760         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 761         struct sfc_port *port = &sa->port;
 762         unsigned int i;
 763         unsigned int nstats = 0;
 764
 765         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 766                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 767                         if (xstats_names != NULL && nstats < xstats_count)
 768                                 strlcpy(xstats_names[nstats].name,
 769                                         efx_mac_stat_name(sa->nic, i),
 770                                         sizeof(xstats_names[0].name));
 771                         nstats++;
 772                 }
 773         }
 774
 775         return nstats;
 776 }
 777
 778 static int
 779 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 780                      uint64_t *values, unsigned int n)
 781 {
 782         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 783         struct sfc_port *port = &sa->port;
 784         uint64_t *mac_stats;
 785         unsigned int nb_supported = 0;
 786         unsigned int nb_written = 0;
 787         unsigned int i;
 788         int ret;
 789         int rc;
 790
 791         if (unlikely(values == NULL) ||
 792             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 793                 return port->mac_stats_nb_supported;
 794
 795         rte_spinlock_lock(&port->mac_stats_lock);
 796
 797         rc = sfc_port_update_mac_stats(sa);
 798         if (rc != 0) {
 799                 SFC_ASSERT(rc > 0);
 800                 ret = -rc;
 801                 goto unlock;
 802         }
 803
 804         mac_stats = port->mac_stats_buf;
 805
 806         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 807                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 808                         continue;
 809
 810                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 811                         values[nb_written++] = mac_stats[i];
 812
 813                 ++nb_supported;
 814         }
 815
 816         ret = nb_written;
 817
 818 unlock:
 819         rte_spinlock_unlock(&port->mac_stats_lock);
 820
 821         return ret;
 822 }
 823
 824 static int
 825 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 826                            struct rte_eth_xstat_name *xstats_names,
 827                            const uint64_t *ids, unsigned int size)
 828 {
 829         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 830         struct sfc_port *port = &sa->port;
 831         unsigned int nb_supported = 0;
 832         unsigned int nb_written = 0;
 833         unsigned int i;
 834
 835         if (unlikely(xstats_names == NULL) ||
 836             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 837                 return port->mac_stats_nb_supported;
 838
 839         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 840                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 841                         continue;
 842
 843                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 844                         char *name = xstats_names[nb_written++].name;
 845
 846                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 847                                 sizeof(xstats_names[0].name));
 848                 }
 849
 850                 ++nb_supported;
 851         }
 852
 853         return nb_written;
 854 }
 855
 856 static int
 857 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 858 {
 859         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 860         unsigned int wanted_fc, link_fc;
 861
 862         memset(fc_conf, 0, sizeof(*fc_conf));
 863
 864         sfc_adapter_lock(sa);
 865
 866         if (sa->state == SFC_ADAPTER_STARTED)
 867                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 868         else
 869                 link_fc = sa->port.flow_ctrl;
 870
 871         switch (link_fc) {
 872         case 0:
 873                 fc_conf->mode = RTE_FC_NONE;
 874                 break;
 875         case EFX_FCNTL_RESPOND:
 876                 fc_conf->mode = RTE_FC_RX_PAUSE;
 877                 break;
 878         case EFX_FCNTL_GENERATE:
 879                 fc_conf->mode = RTE_FC_TX_PAUSE;
 880                 break;
 881         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 882                 fc_conf->mode = RTE_FC_FULL;
 883                 break;
 884         default:
 885                 sfc_err(sa, "%s: unexpected flow control value %#x",
 886                         __func__, link_fc);
 887         }
 888
 889         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 890
 891         sfc_adapter_unlock(sa);
 892
 893         return 0;
 894 }
 895
 896 static int
 897 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 898 {
 899         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 900         struct sfc_port *port = &sa->port;
 901         unsigned int fcntl;
 902         int rc;
 903
 904         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 905             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 906             fc_conf->mac_ctrl_frame_fwd != 0) {
 907                 sfc_err(sa, "unsupported flow control settings specified");
 908                 rc = EINVAL;
 909                 goto fail_inval;
 910         }
 911
 912         switch (fc_conf->mode) {
 913         case RTE_FC_NONE:
 914                 fcntl = 0;
 915                 break;
 916         case RTE_FC_RX_PAUSE:
 917                 fcntl = EFX_FCNTL_RESPOND;
 918                 break;
 919         case RTE_FC_TX_PAUSE:
 920                 fcntl = EFX_FCNTL_GENERATE;
 921                 break;
 922         case RTE_FC_FULL:
 923                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 924                 break;
 925         default:
 926                 rc = EINVAL;
 927                 goto fail_inval;
 928         }
 929
 930         sfc_adapter_lock(sa);
 931
 932         if (sa->state == SFC_ADAPTER_STARTED) {
 933                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 934                 if (rc != 0)
 935                         goto fail_mac_fcntl_set;
 936         }
 937
 938         port->flow_ctrl = fcntl;
 939         port->flow_ctrl_autoneg = fc_conf->autoneg;
 940
 941         sfc_adapter_unlock(sa);
 942
 943         return 0;
 944
 945 fail_mac_fcntl_set:
 946         sfc_adapter_unlock(sa);
 947 fail_inval:
 948         SFC_ASSERT(rc > 0);
 949         return -rc;
 950 }
 951
 952 static int
 953 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 954 {
 955         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 956         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 957         boolean_t scatter_enabled;
 958         const char *error;
 959         unsigned int i;
 960
 961         for (i = 0; i < sas->rxq_count; i++) {
 962                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 963                         continue;
 964
 965                 scatter_enabled = (sas->rxq_info[i].type_flags &
 966                                    EFX_RXQ_FLAG_SCATTER);
 967
 968                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 969                                           encp->enc_rx_prefix_size,
 970                                           scatter_enabled,
 971                                           encp->enc_rx_scatter_max, &error)) {
 972                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 973                                 error);
 974                         return EINVAL;
 975                 }
 976         }
 977
 978         return 0;
 979 }
 980
 981 static int
 982 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 983 {
 984         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 985         size_t pdu = EFX_MAC_PDU(mtu);
 986         size_t old_pdu;
 987         int rc;
 988
 989         sfc_log_init(sa, "mtu=%u", mtu);
 990
 991         rc = EINVAL;
 992         if (pdu < EFX_MAC_PDU_MIN) {
 993                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 994                         (unsigned int)mtu, (unsigned int)pdu,
 995                         EFX_MAC_PDU_MIN);
 996                 goto fail_inval;
 997         }
 998         if (pdu > EFX_MAC_PDU_MAX) {
 999                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1000                         (unsigned int)mtu, (unsigned int)pdu,
1001                         (unsigned int)EFX_MAC_PDU_MAX);
1002                 goto fail_inval;
1003         }
1004
1005         sfc_adapter_lock(sa);
1006
1007         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1008         if (rc != 0)
1009                 goto fail_check_scatter;
1010
1011         if (pdu != sa->port.pdu) {
1012                 if (sa->state == SFC_ADAPTER_STARTED) {
1013                         sfc_stop(sa);
1014
1015                         old_pdu = sa->port.pdu;
1016                         sa->port.pdu = pdu;
1017                         rc = sfc_start(sa);
1018                         if (rc != 0)
1019                                 goto fail_start;
1020                 } else {
1021                         sa->port.pdu = pdu;
1022                 }
1023         }
1024
1025         /*
1026          * The driver does not use it, but other PMDs update jumbo frame
1027          * flag and max_rx_pkt_len when MTU is set.
1028          */
1029         if (mtu > RTE_ETHER_MTU) {
1030                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1031                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1032         }
1033
1034         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1035
1036         sfc_adapter_unlock(sa);
1037
1038         sfc_log_init(sa, "done");
1039         return 0;
1040
1041 fail_start:
1042         sa->port.pdu = old_pdu;
1043         if (sfc_start(sa) != 0)
1044                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1045                         "PDU max size - port is stopped",
1046                         (unsigned int)pdu, (unsigned int)old_pdu);
1047
1048 fail_check_scatter:
1049         sfc_adapter_unlock(sa);
1050
1051 fail_inval:
1052         sfc_log_init(sa, "failed %d", rc);
1053         SFC_ASSERT(rc > 0);
1054         return -rc;
1055 }
1056 static int
1057 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1058 {
1059         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1060         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1061         struct sfc_port *port = &sa->port;
1062         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1063         int rc = 0;
1064
1065         sfc_adapter_lock(sa);
1066
1067         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1068                 goto unlock;
1069
1070         /*
1071          * Copy the address to the device private data so that
1072          * it could be recalled in the case of adapter restart.
1073          */
1074         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1075
1076         /*
1077          * Neither of the two following checks can return
1078          * an error. The new MAC address is preserved in
1079          * the device private data and can be activated
1080          * on the next port start if the user prevents
1081          * isolated mode from being enabled.
1082          */
1083         if (sfc_sa2shared(sa)->isolated) {
1084                 sfc_warn(sa, "isolated mode is active on the port");
1085                 sfc_warn(sa, "will not set MAC address");
1086                 goto unlock;
1087         }
1088
1089         if (sa->state != SFC_ADAPTER_STARTED) {
1090                 sfc_notice(sa, "the port is not started");
1091                 sfc_notice(sa, "the new MAC address will be set on port start");
1092
1093                 goto unlock;
1094         }
1095
1096         if (encp->enc_allow_set_mac_with_installed_filters) {
1097                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1098                 if (rc != 0) {
1099                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1100                         goto unlock;
1101                 }
1102
1103                 /*
1104                  * Changing the MAC address by means of MCDI request
1105                  * has no effect on received traffic, therefore
1106                  * we also need to update unicast filters
1107                  */
1108                 rc = sfc_set_rx_mode_unchecked(sa);
1109                 if (rc != 0) {
1110                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1111                         /* Rollback the old address */
1112                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1113                         (void)sfc_set_rx_mode_unchecked(sa);
1114                 }
1115         } else {
1116                 sfc_warn(sa, "cannot set MAC address with filters installed");
1117                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1118                 sfc_warn(sa, "(some traffic may be dropped)");
1119
1120                 /*
1121                  * Since setting MAC address with filters installed is not
1122                  * allowed on the adapter, the new MAC address will be set
1123                  * by means of adapter restart. sfc_start() shall retrieve
1124                  * the new address from the device private data and set it.
1125                  */
1126                 sfc_stop(sa);
1127                 rc = sfc_start(sa);
1128                 if (rc != 0)
1129                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1130         }
1131
1132 unlock:
1133         if (rc != 0)
1134                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1135
1136         sfc_adapter_unlock(sa);
1137
1138         SFC_ASSERT(rc >= 0);
1139         return -rc;
1140 }
1141
1142
1143 static int
1144 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1145                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1146 {
1147         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1148         struct sfc_port *port = &sa->port;
1149         uint8_t *mc_addrs = port->mcast_addrs;
1150         int rc;
1151         unsigned int i;
1152
1153         if (sfc_sa2shared(sa)->isolated) {
1154                 sfc_err(sa, "isolated mode is active on the port");
1155                 sfc_err(sa, "will not set multicast address list");
1156                 return -ENOTSUP;
1157         }
1158
1159         if (mc_addrs == NULL)
1160                 return -ENOBUFS;
1161
1162         if (nb_mc_addr > port->max_mcast_addrs) {
1163                 sfc_err(sa, "too many multicast addresses: %u > %u",
1164                          nb_mc_addr, port->max_mcast_addrs);
1165                 return -EINVAL;
1166         }
1167
1168         for (i = 0; i < nb_mc_addr; ++i) {
1169                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1170                                  EFX_MAC_ADDR_LEN);
1171                 mc_addrs += EFX_MAC_ADDR_LEN;
1172         }
1173
1174         port->nb_mcast_addrs = nb_mc_addr;
1175
1176         if (sa->state != SFC_ADAPTER_STARTED)
1177                 return 0;
1178
1179         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1180                                         port->nb_mcast_addrs);
1181         if (rc != 0)
1182                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1183
1184         SFC_ASSERT(rc >= 0);
1185         return -rc;
1186 }
1187
1188 /*
1189  * The function is used by the secondary process as well. It must not
1190  * use any process-local pointers from the adapter data.
1191  */
1192 static void
1193 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1194                       struct rte_eth_rxq_info *qinfo)
1195 {
1196         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1197         struct sfc_rxq_info *rxq_info;
1198
1199         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1200
1201         rxq_info = &sas->rxq_info[rx_queue_id];
1202
1203         qinfo->mp = rxq_info->refill_mb_pool;
1204         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1205         qinfo->conf.rx_drop_en = 1;
1206         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1207         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1208         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1209                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1210                 qinfo->scattered_rx = 1;
1211         }
1212         qinfo->nb_desc = rxq_info->entries;
1213 }
1214
1215 /*
1216  * The function is used by the secondary process as well. It must not
1217  * use any process-local pointers from the adapter data.
1218  */
1219 static void
1220 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1221                       struct rte_eth_txq_info *qinfo)
1222 {
1223         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1224         struct sfc_txq_info *txq_info;
1225
1226         SFC_ASSERT(tx_queue_id < sas->txq_count);
1227
1228         txq_info = &sas->txq_info[tx_queue_id];
1229
1230         memset(qinfo, 0, sizeof(*qinfo));
1231
1232         qinfo->conf.offloads = txq_info->offloads;
1233         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1234         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1235         qinfo->nb_desc = txq_info->entries;
1236 }
1237
1238 /*
1239  * The function is used by the secondary process as well. It must not
1240  * use any process-local pointers from the adapter data.
1241  */
1242 static uint32_t
1243 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1244 {
1245         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1246         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1247         struct sfc_rxq_info *rxq_info;
1248
1249         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1250         rxq_info = &sas->rxq_info[rx_queue_id];
1251
1252         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1253                 return 0;
1254
1255         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1256 }
1257
1258 /*
1259  * The function is used by the secondary process as well. It must not
1260  * use any process-local pointers from the adapter data.
1261  */
1262 static int
1263 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1264 {
1265         struct sfc_dp_rxq *dp_rxq = queue;
1266         const struct sfc_dp_rx *dp_rx;
1267
1268         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1269
1270         return offset < dp_rx->qdesc_npending(dp_rxq);
1271 }
1272
1273 /*
1274  * The function is used by the secondary process as well. It must not
1275  * use any process-local pointers from the adapter data.
1276  */
1277 static int
1278 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1279 {
1280         struct sfc_dp_rxq *dp_rxq = queue;
1281         const struct sfc_dp_rx *dp_rx;
1282
1283         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1284
1285         return dp_rx->qdesc_status(dp_rxq, offset);
1286 }
1287
1288 /*
1289  * The function is used by the secondary process as well. It must not
1290  * use any process-local pointers from the adapter data.
1291  */
1292 static int
1293 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1294 {
1295         struct sfc_dp_txq *dp_txq = queue;
1296         const struct sfc_dp_tx *dp_tx;
1297
1298         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1299
1300         return dp_tx->qdesc_status(dp_txq, offset);
1301 }
1302
1303 static int
1304 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1305 {
1306         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1307         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1308         int rc;
1309
1310         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1311
1312         sfc_adapter_lock(sa);
1313
1314         rc = EINVAL;
1315         if (sa->state != SFC_ADAPTER_STARTED)
1316                 goto fail_not_started;
1317
1318         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1319                 goto fail_not_setup;
1320
1321         rc = sfc_rx_qstart(sa, rx_queue_id);
1322         if (rc != 0)
1323                 goto fail_rx_qstart;
1324
1325         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1326
1327         sfc_adapter_unlock(sa);
1328
1329         return 0;
1330
1331 fail_rx_qstart:
1332 fail_not_setup:
1333 fail_not_started:
1334         sfc_adapter_unlock(sa);
1335         SFC_ASSERT(rc > 0);
1336         return -rc;
1337 }
1338
1339 static int
1340 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1341 {
1342         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1343         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1344
1345         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1346
1347         sfc_adapter_lock(sa);
1348         sfc_rx_qstop(sa, rx_queue_id);
1349
1350         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1351
1352         sfc_adapter_unlock(sa);
1353
1354         return 0;
1355 }
1356
1357 static int
1358 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1359 {
1360         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1361         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1362         int rc;
1363
1364         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1365
1366         sfc_adapter_lock(sa);
1367
1368         rc = EINVAL;
1369         if (sa->state != SFC_ADAPTER_STARTED)
1370                 goto fail_not_started;
1371
1372         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1373                 goto fail_not_setup;
1374
1375         rc = sfc_tx_qstart(sa, tx_queue_id);
1376         if (rc != 0)
1377                 goto fail_tx_qstart;
1378
1379         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1380
1381         sfc_adapter_unlock(sa);
1382         return 0;
1383
1384 fail_tx_qstart:
1385
1386 fail_not_setup:
1387 fail_not_started:
1388         sfc_adapter_unlock(sa);
1389         SFC_ASSERT(rc > 0);
1390         return -rc;
1391 }
1392
1393 static int
1394 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1395 {
1396         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1397         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1398
1399         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1400
1401         sfc_adapter_lock(sa);
1402
1403         sfc_tx_qstop(sa, tx_queue_id);
1404
1405         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1406
1407         sfc_adapter_unlock(sa);
1408         return 0;
1409 }
1410
1411 static efx_tunnel_protocol_t
1412 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1413 {
1414         switch (rte_type) {
1415         case RTE_TUNNEL_TYPE_VXLAN:
1416                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1417         case RTE_TUNNEL_TYPE_GENEVE:
1418                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1419         default:
1420                 return EFX_TUNNEL_NPROTOS;
1421         }
1422 }
1423
1424 enum sfc_udp_tunnel_op_e {
1425         SFC_UDP_TUNNEL_ADD_PORT,
1426         SFC_UDP_TUNNEL_DEL_PORT,
1427 };
1428
1429 static int
1430 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1431                       struct rte_eth_udp_tunnel *tunnel_udp,
1432                       enum sfc_udp_tunnel_op_e op)
1433 {
1434         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1435         efx_tunnel_protocol_t tunnel_proto;
1436         int rc;
1437
1438         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1439                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1440                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1441                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1442
1443         tunnel_proto =
1444                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1445         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1446                 rc = ENOTSUP;
1447                 goto fail_bad_proto;
1448         }
1449
1450         sfc_adapter_lock(sa);
1451
1452         switch (op) {
1453         case SFC_UDP_TUNNEL_ADD_PORT:
1454                 rc = efx_tunnel_config_udp_add(sa->nic,
1455                                                tunnel_udp->udp_port,
1456                                                tunnel_proto);
1457                 break;
1458         case SFC_UDP_TUNNEL_DEL_PORT:
1459                 rc = efx_tunnel_config_udp_remove(sa->nic,
1460                                                   tunnel_udp->udp_port,
1461                                                   tunnel_proto);
1462                 break;
1463         default:
1464                 rc = EINVAL;
1465                 goto fail_bad_op;
1466         }
1467
1468         if (rc != 0)
1469                 goto fail_op;
1470
1471         if (sa->state == SFC_ADAPTER_STARTED) {
1472                 rc = efx_tunnel_reconfigure(sa->nic);
1473                 if (rc == EAGAIN) {
1474                         /*
1475                          * Configuration is accepted by FW and MC reboot
1476                          * is initiated to apply the changes. MC reboot
1477                          * will be handled in a usual way (MC reboot
1478                          * event on management event queue and adapter
1479                          * restart).
1480                          */
1481                         rc = 0;
1482                 } else if (rc != 0) {
1483                         goto fail_reconfigure;
1484                 }
1485         }
1486
1487         sfc_adapter_unlock(sa);
1488         return 0;
1489
1490 fail_reconfigure:
1491         /* Remove/restore entry since the change makes the trouble */
1492         switch (op) {
1493         case SFC_UDP_TUNNEL_ADD_PORT:
1494                 (void)efx_tunnel_config_udp_remove(sa->nic,
1495                                                    tunnel_udp->udp_port,
1496                                                    tunnel_proto);
1497                 break;
1498         case SFC_UDP_TUNNEL_DEL_PORT:
1499                 (void)efx_tunnel_config_udp_add(sa->nic,
1500                                                 tunnel_udp->udp_port,
1501                                                 tunnel_proto);
1502                 break;
1503         }
1504
1505 fail_op:
1506 fail_bad_op:
1507         sfc_adapter_unlock(sa);
1508
1509 fail_bad_proto:
1510         SFC_ASSERT(rc > 0);
1511         return -rc;
1512 }
1513
1514 static int
1515 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1516                             struct rte_eth_udp_tunnel *tunnel_udp)
1517 {
1518         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1519 }
1520
1521 static int
1522 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1523                             struct rte_eth_udp_tunnel *tunnel_udp)
1524 {
1525         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1526 }
1527
1528 /*
1529  * The function is used by the secondary process as well. It must not
1530  * use any process-local pointers from the adapter data.
1531  */
1532 static int
1533 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1534                           struct rte_eth_rss_conf *rss_conf)
1535 {
1536         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1537         struct sfc_rss *rss = &sas->rss;
1538
1539         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1540                 return -ENOTSUP;
1541
1542         /*
1543          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1544          * hence, conversion is done here to derive a correct set of ETH_RSS
1545          * flags which corresponds to the active EFX configuration stored
1546          * locally in 'sfc_adapter' and kept up-to-date
1547          */
1548         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1549         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1550         if (rss_conf->rss_key != NULL)
1551                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1552
1553         return 0;
1554 }
1555
1556 static int
1557 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1558                         struct rte_eth_rss_conf *rss_conf)
1559 {
1560         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1561         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1562         unsigned int efx_hash_types;
1563         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1564         unsigned int n_contexts;
1565         unsigned int mode_i = 0;
1566         unsigned int key_i = 0;
1567         unsigned int i = 0;
1568         int rc = 0;
1569
1570         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1571
1572         if (sfc_sa2shared(sa)->isolated)
1573                 return -ENOTSUP;
1574
1575         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1576                 sfc_err(sa, "RSS is not available");
1577                 return -ENOTSUP;
1578         }
1579
1580         if (rss->channels == 0) {
1581                 sfc_err(sa, "RSS is not configured");
1582                 return -EINVAL;
1583         }
1584
1585         if ((rss_conf->rss_key != NULL) &&
1586             (rss_conf->rss_key_len != sizeof(rss->key))) {
1587                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1588                         sizeof(rss->key));
1589                 return -EINVAL;
1590         }
1591
1592         sfc_adapter_lock(sa);
1593
1594         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1595         if (rc != 0)
1596                 goto fail_rx_hf_rte_to_efx;
1597
1598         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1599                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1600                                            rss->hash_alg, efx_hash_types,
1601                                            B_TRUE);
1602                 if (rc != 0)
1603                         goto fail_scale_mode_set;
1604         }
1605
1606         if (rss_conf->rss_key != NULL) {
1607                 if (sa->state == SFC_ADAPTER_STARTED) {
1608                         for (key_i = 0; key_i < n_contexts; key_i++) {
1609                                 rc = efx_rx_scale_key_set(sa->nic,
1610                                                           contexts[key_i],
1611                                                           rss_conf->rss_key,
1612                                                           sizeof(rss->key));
1613                                 if (rc != 0)
1614                                         goto fail_scale_key_set;
1615                         }
1616                 }
1617
1618                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1619         }
1620
1621         rss->hash_types = efx_hash_types;
1622
1623         sfc_adapter_unlock(sa);
1624
1625         return 0;
1626
1627 fail_scale_key_set:
1628         for (i = 0; i < key_i; i++) {
1629                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1630                                          sizeof(rss->key)) != 0)
1631                         sfc_err(sa, "failed to restore RSS key");
1632         }
1633
1634 fail_scale_mode_set:
1635         for (i = 0; i < mode_i; i++) {
1636                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1637                                           EFX_RX_HASHALG_TOEPLITZ,
1638                                           rss->hash_types, B_TRUE) != 0)
1639                         sfc_err(sa, "failed to restore RSS mode");
1640         }
1641
1642 fail_rx_hf_rte_to_efx:
1643         sfc_adapter_unlock(sa);
1644         return -rc;
1645 }
1646
1647 /*
1648  * The function is used by the secondary process as well. It must not
1649  * use any process-local pointers from the adapter data.
1650  */
1651 static int
1652 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1653                        struct rte_eth_rss_reta_entry64 *reta_conf,
1654                        uint16_t reta_size)
1655 {
1656         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1657         struct sfc_rss *rss = &sas->rss;
1658         int entry;
1659
1660         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1661                 return -ENOTSUP;
1662
1663         if (rss->channels == 0)
1664                 return -EINVAL;
1665
1666         if (reta_size != EFX_RSS_TBL_SIZE)
1667                 return -EINVAL;
1668
1669         for (entry = 0; entry < reta_size; entry++) {
1670                 int grp = entry / RTE_RETA_GROUP_SIZE;
1671                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1672
1673                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1674                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1675         }
1676
1677         return 0;
1678 }
1679
1680 static int
1681 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1682                         struct rte_eth_rss_reta_entry64 *reta_conf,
1683                         uint16_t reta_size)
1684 {
1685         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1686         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1687         unsigned int *rss_tbl_new;
1688         uint16_t entry;
1689         int rc = 0;
1690
1691
1692         if (sfc_sa2shared(sa)->isolated)
1693                 return -ENOTSUP;
1694
1695         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1696                 sfc_err(sa, "RSS is not available");
1697                 return -ENOTSUP;
1698         }
1699
1700         if (rss->channels == 0) {
1701                 sfc_err(sa, "RSS is not configured");
1702                 return -EINVAL;
1703         }
1704
1705         if (reta_size != EFX_RSS_TBL_SIZE) {
1706                 sfc_err(sa, "RETA size is wrong (should be %u)",
1707                         EFX_RSS_TBL_SIZE);
1708                 return -EINVAL;
1709         }
1710
1711         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1712         if (rss_tbl_new == NULL)
1713                 return -ENOMEM;
1714
1715         sfc_adapter_lock(sa);
1716
1717         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1718
1719         for (entry = 0; entry < reta_size; entry++) {
1720                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1721                 struct rte_eth_rss_reta_entry64 *grp;
1722
1723                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1724
1725                 if (grp->mask & (1ull << grp_idx)) {
1726                         if (grp->reta[grp_idx] >= rss->channels) {
1727                                 rc = EINVAL;
1728                                 goto bad_reta_entry;
1729                         }
1730                         rss_tbl_new[entry] = grp->reta[grp_idx];
1731                 }
1732         }
1733
1734         if (sa->state == SFC_ADAPTER_STARTED) {
1735                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1736                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1737                 if (rc != 0)
1738                         goto fail_scale_tbl_set;
1739         }
1740
1741         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1742
1743 fail_scale_tbl_set:
1744 bad_reta_entry:
1745         sfc_adapter_unlock(sa);
1746
1747         rte_free(rss_tbl_new);
1748
1749         SFC_ASSERT(rc >= 0);
1750         return -rc;
1751 }
1752
1753 static int
1754 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1755                     enum rte_filter_op filter_op,
1756                     void *arg)
1757 {
1758         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1759         int rc = ENOTSUP;
1760
1761         sfc_log_init(sa, "entry");
1762
1763         switch (filter_type) {
1764         case RTE_ETH_FILTER_GENERIC:
1765                 if (filter_op != RTE_ETH_FILTER_GET) {
1766                         rc = EINVAL;
1767                 } else {
1768                         *(const void **)arg = &sfc_flow_ops;
1769                         rc = 0;
1770                 }
1771                 break;
1772         default:
1773                 sfc_err(sa, "Unknown filter type %u", filter_type);
1774                 break;
1775         }
1776
1777         sfc_log_init(sa, "exit: %d", -rc);
1778         SFC_ASSERT(rc >= 0);
1779         return -rc;
1780 }
1781
1782 static int
1783 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1784 {
1785         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1786
1787         /*
1788          * If Rx datapath does not provide callback to check mempool,
1789          * all pools are supported.
1790          */
1791         if (sap->dp_rx->pool_ops_supported == NULL)
1792                 return 1;
1793
1794         return sap->dp_rx->pool_ops_supported(pool);
1795 }
1796
1797 static int
1798 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1799 {
1800         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1801         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1802         struct sfc_rxq_info *rxq_info;
1803
1804         SFC_ASSERT(queue_id < sas->rxq_count);
1805         rxq_info = &sas->rxq_info[queue_id];
1806
1807         return sap->dp_rx->intr_enable(rxq_info->dp);
1808 }
1809
1810 static int
1811 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1812 {
1813         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1814         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1815         struct sfc_rxq_info *rxq_info;
1816
1817         SFC_ASSERT(queue_id < sas->rxq_count);
1818         rxq_info = &sas->rxq_info[queue_id];
1819
1820         return sap->dp_rx->intr_disable(rxq_info->dp);
1821 }
1822
1823 static const struct eth_dev_ops sfc_eth_dev_ops = {
1824         .dev_configure                  = sfc_dev_configure,
1825         .dev_start                      = sfc_dev_start,
1826         .dev_stop                       = sfc_dev_stop,
1827         .dev_set_link_up                = sfc_dev_set_link_up,
1828         .dev_set_link_down              = sfc_dev_set_link_down,
1829         .dev_close                      = sfc_dev_close,
1830         .promiscuous_enable             = sfc_dev_promisc_enable,
1831         .promiscuous_disable            = sfc_dev_promisc_disable,
1832         .allmulticast_enable            = sfc_dev_allmulti_enable,
1833         .allmulticast_disable           = sfc_dev_allmulti_disable,
1834         .link_update                    = sfc_dev_link_update,
1835         .stats_get                      = sfc_stats_get,
1836         .stats_reset                    = sfc_stats_reset,
1837         .xstats_get                     = sfc_xstats_get,
1838         .xstats_reset                   = sfc_stats_reset,
1839         .xstats_get_names               = sfc_xstats_get_names,
1840         .dev_infos_get                  = sfc_dev_infos_get,
1841         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1842         .mtu_set                        = sfc_dev_set_mtu,
1843         .rx_queue_start                 = sfc_rx_queue_start,
1844         .rx_queue_stop                  = sfc_rx_queue_stop,
1845         .tx_queue_start                 = sfc_tx_queue_start,
1846         .tx_queue_stop                  = sfc_tx_queue_stop,
1847         .rx_queue_setup                 = sfc_rx_queue_setup,
1848         .rx_queue_release               = sfc_rx_queue_release,
1849         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1850         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1851         .tx_queue_setup                 = sfc_tx_queue_setup,
1852         .tx_queue_release               = sfc_tx_queue_release,
1853         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1854         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1855         .mac_addr_set                   = sfc_mac_addr_set,
1856         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1857         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1858         .reta_update                    = sfc_dev_rss_reta_update,
1859         .reta_query                     = sfc_dev_rss_reta_query,
1860         .rss_hash_update                = sfc_dev_rss_hash_update,
1861         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1862         .filter_ctrl                    = sfc_dev_filter_ctrl,
1863         .set_mc_addr_list               = sfc_set_mc_addr_list,
1864         .rxq_info_get                   = sfc_rx_queue_info_get,
1865         .txq_info_get                   = sfc_tx_queue_info_get,
1866         .fw_version_get                 = sfc_fw_version_get,
1867         .xstats_get_by_id               = sfc_xstats_get_by_id,
1868         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1869         .pool_ops_supported             = sfc_pool_ops_supported,
1870 };
1871
1872 /**
1873  * Duplicate a string in potentially shared memory required for
1874  * multi-process support.
1875  *
1876  * strdup() allocates from process-local heap/memory.
1877  */
1878 static char *
1879 sfc_strdup(const char *str)
1880 {
1881         size_t size;
1882         char *copy;
1883
1884         if (str == NULL)
1885                 return NULL;
1886
1887         size = strlen(str) + 1;
1888         copy = rte_malloc(__func__, size, 0);
1889         if (copy != NULL)
1890                 rte_memcpy(copy, str, size);
1891
1892         return copy;
1893 }
1894
1895 static int
1896 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1897 {
1898         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1899         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1900         const struct sfc_dp_rx *dp_rx;
1901         const struct sfc_dp_tx *dp_tx;
1902         const efx_nic_cfg_t *encp;
1903         unsigned int avail_caps = 0;
1904         const char *rx_name = NULL;
1905         const char *tx_name = NULL;
1906         int rc;
1907
1908         switch (sa->family) {
1909         case EFX_FAMILY_HUNTINGTON:
1910         case EFX_FAMILY_MEDFORD:
1911         case EFX_FAMILY_MEDFORD2:
1912                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1913                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1914                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1915                 break;
1916         case EFX_FAMILY_RIVERHEAD:
1917                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1918                 break;
1919         default:
1920                 break;
1921         }
1922
1923         encp = efx_nic_cfg_get(sa->nic);
1924         if (encp->enc_rx_es_super_buffer_supported)
1925                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1926
1927         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1928                                 sfc_kvarg_string_handler, &rx_name);
1929         if (rc != 0)
1930                 goto fail_kvarg_rx_datapath;
1931
1932         if (rx_name != NULL) {
1933                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1934                 if (dp_rx == NULL) {
1935                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1936                         rc = ENOENT;
1937                         goto fail_dp_rx;
1938                 }
1939                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1940                         sfc_err(sa,
1941                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1942                                 rx_name);
1943                         rc = EINVAL;
1944                         goto fail_dp_rx_caps;
1945                 }
1946         } else {
1947                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1948                 if (dp_rx == NULL) {
1949                         sfc_err(sa, "Rx datapath by caps %#x not found",
1950                                 avail_caps);
1951                         rc = ENOENT;
1952                         goto fail_dp_rx;
1953                 }
1954         }
1955
1956         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1957         if (sas->dp_rx_name == NULL) {
1958                 rc = ENOMEM;
1959                 goto fail_dp_rx_name;
1960         }
1961
1962         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1963
1964         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1965                                 sfc_kvarg_string_handler, &tx_name);
1966         if (rc != 0)
1967                 goto fail_kvarg_tx_datapath;
1968
1969         if (tx_name != NULL) {
1970                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1971                 if (dp_tx == NULL) {
1972                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1973                         rc = ENOENT;
1974                         goto fail_dp_tx;
1975                 }
1976                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1977                         sfc_err(sa,
1978                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1979                                 tx_name);
1980                         rc = EINVAL;
1981                         goto fail_dp_tx_caps;
1982                 }
1983         } else {
1984                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1985                 if (dp_tx == NULL) {
1986                         sfc_err(sa, "Tx datapath by caps %#x not found",
1987                                 avail_caps);
1988                         rc = ENOENT;
1989                         goto fail_dp_tx;
1990                 }
1991         }
1992
1993         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1994         if (sas->dp_tx_name == NULL) {
1995                 rc = ENOMEM;
1996                 goto fail_dp_tx_name;
1997         }
1998
1999         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2000
2001         sa->priv.dp_rx = dp_rx;
2002         sa->priv.dp_tx = dp_tx;
2003
2004         dev->rx_pkt_burst = dp_rx->pkt_burst;
2005         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2006         dev->tx_pkt_burst = dp_tx->pkt_burst;
2007
2008         dev->rx_queue_count = sfc_rx_queue_count;
2009         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2010         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2011         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2012         dev->dev_ops = &sfc_eth_dev_ops;
2013
2014         return 0;
2015
2016 fail_dp_tx_name:
2017 fail_dp_tx_caps:
2018 fail_dp_tx:
2019 fail_kvarg_tx_datapath:
2020         rte_free(sas->dp_rx_name);
2021         sas->dp_rx_name = NULL;
2022
2023 fail_dp_rx_name:
2024 fail_dp_rx_caps:
2025 fail_dp_rx:
2026 fail_kvarg_rx_datapath:
2027         return rc;
2028 }
2029
2030 static void
2031 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2032 {
2033         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2034         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2035
2036         dev->dev_ops = NULL;
2037         dev->tx_pkt_prepare = NULL;
2038         dev->rx_pkt_burst = NULL;
2039         dev->tx_pkt_burst = NULL;
2040
2041         rte_free(sas->dp_tx_name);
2042         sas->dp_tx_name = NULL;
2043         sa->priv.dp_tx = NULL;
2044
2045         rte_free(sas->dp_rx_name);
2046         sas->dp_rx_name = NULL;
2047         sa->priv.dp_rx = NULL;
2048 }
2049
2050 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2051         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2052         .reta_query                     = sfc_dev_rss_reta_query,
2053         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2054         .rxq_info_get                   = sfc_rx_queue_info_get,
2055         .txq_info_get                   = sfc_tx_queue_info_get,
2056 };
2057
2058 static int
2059 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2060 {
2061         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2062         struct sfc_adapter_priv *sap;
2063         const struct sfc_dp_rx *dp_rx;
2064         const struct sfc_dp_tx *dp_tx;
2065         int rc;
2066
2067         /*
2068          * Allocate process private data from heap, since it should not
2069          * be located in shared memory allocated using rte_malloc() API.
2070          */
2071         sap = calloc(1, sizeof(*sap));
2072         if (sap == NULL) {
2073                 rc = ENOMEM;
2074                 goto fail_alloc_priv;
2075         }
2076
2077         sap->logtype_main = logtype_main;
2078
2079         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2080         if (dp_rx == NULL) {
2081                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2082                         "cannot find %s Rx datapath", sas->dp_rx_name);
2083                 rc = ENOENT;
2084                 goto fail_dp_rx;
2085         }
2086         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2087                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2088                         "%s Rx datapath does not support multi-process",
2089                         sas->dp_rx_name);
2090                 rc = EINVAL;
2091                 goto fail_dp_rx_multi_process;
2092         }
2093
2094         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2095         if (dp_tx == NULL) {
2096                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2097                         "cannot find %s Tx datapath", sas->dp_tx_name);
2098                 rc = ENOENT;
2099                 goto fail_dp_tx;
2100         }
2101         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2102                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2103                         "%s Tx datapath does not support multi-process",
2104                         sas->dp_tx_name);
2105                 rc = EINVAL;
2106                 goto fail_dp_tx_multi_process;
2107         }
2108
2109         sap->dp_rx = dp_rx;
2110         sap->dp_tx = dp_tx;
2111
2112         dev->process_private = sap;
2113         dev->rx_pkt_burst = dp_rx->pkt_burst;
2114         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2115         dev->tx_pkt_burst = dp_tx->pkt_burst;
2116         dev->rx_queue_count = sfc_rx_queue_count;
2117         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2118         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2119         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2120         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2121
2122         return 0;
2123
2124 fail_dp_tx_multi_process:
2125 fail_dp_tx:
2126 fail_dp_rx_multi_process:
2127 fail_dp_rx:
2128         free(sap);
2129
2130 fail_alloc_priv:
2131         return rc;
2132 }
2133
2134 static void
2135 sfc_register_dp(void)
2136 {
2137         /* Register once */
2138         if (TAILQ_EMPTY(&sfc_dp_head)) {
2139                 /* Prefer EF10 datapath */
2140                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2141                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2142                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2143                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2144
2145                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2146                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2147                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2148                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2149         }
2150 }
2151
2152 static int
2153 sfc_eth_dev_init(struct rte_eth_dev *dev)
2154 {
2155         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2156         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2157         uint32_t logtype_main;
2158         struct sfc_adapter *sa;
2159         int rc;
2160         const efx_nic_cfg_t *encp;
2161         const struct rte_ether_addr *from;
2162         int ret;
2163
2164         sfc_register_dp();
2165
2166         logtype_main = sfc_register_logtype(&pci_dev->addr,
2167                                             SFC_LOGTYPE_MAIN_STR,
2168                                             RTE_LOG_NOTICE);
2169
2170         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2171                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2172
2173         /* Required for logging */
2174         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2175                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2176                         pci_dev->addr.domain, pci_dev->addr.bus,
2177                         pci_dev->addr.devid, pci_dev->addr.function,
2178                         dev->data->port_id);
2179         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2180                 SFC_GENERIC_LOG(ERR,
2181                         "reserved log prefix is too short for " PCI_PRI_FMT,
2182                         pci_dev->addr.domain, pci_dev->addr.bus,
2183                         pci_dev->addr.devid, pci_dev->addr.function);
2184                 return -EINVAL;
2185         }
2186         sas->pci_addr = pci_dev->addr;
2187         sas->port_id = dev->data->port_id;
2188
2189         /*
2190          * Allocate process private data from heap, since it should not
2191          * be located in shared memory allocated using rte_malloc() API.
2192          */
2193         sa = calloc(1, sizeof(*sa));
2194         if (sa == NULL) {
2195                 rc = ENOMEM;
2196                 goto fail_alloc_sa;
2197         }
2198
2199         dev->process_private = sa;
2200
2201         /* Required for logging */
2202         sa->priv.shared = sas;
2203         sa->priv.logtype_main = logtype_main;
2204
2205         sa->eth_dev = dev;
2206
2207         /* Copy PCI device info to the dev->data */
2208         rte_eth_copy_pci_info(dev, pci_dev);
2209         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2210         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2211
2212         rc = sfc_kvargs_parse(sa);
2213         if (rc != 0)
2214                 goto fail_kvargs_parse;
2215
2216         sfc_log_init(sa, "entry");
2217
2218         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2219         if (dev->data->mac_addrs == NULL) {
2220                 rc = ENOMEM;
2221                 goto fail_mac_addrs;
2222         }
2223
2224         sfc_adapter_lock_init(sa);
2225         sfc_adapter_lock(sa);
2226
2227         sfc_log_init(sa, "probing");
2228         rc = sfc_probe(sa);
2229         if (rc != 0)
2230                 goto fail_probe;
2231
2232         sfc_log_init(sa, "set device ops");
2233         rc = sfc_eth_dev_set_ops(dev);
2234         if (rc != 0)
2235                 goto fail_set_ops;
2236
2237         sfc_log_init(sa, "attaching");
2238         rc = sfc_attach(sa);
2239         if (rc != 0)
2240                 goto fail_attach;
2241
2242         encp = efx_nic_cfg_get(sa->nic);
2243
2244         /*
2245          * The arguments are really reverse order in comparison to
2246          * Linux kernel. Copy from NIC config to Ethernet device data.
2247          */
2248         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2249         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2250
2251         sfc_adapter_unlock(sa);
2252
2253         sfc_log_init(sa, "done");
2254         return 0;
2255
2256 fail_attach:
2257         sfc_eth_dev_clear_ops(dev);
2258
2259 fail_set_ops:
2260         sfc_unprobe(sa);
2261
2262 fail_probe:
2263         sfc_adapter_unlock(sa);
2264         sfc_adapter_lock_fini(sa);
2265         rte_free(dev->data->mac_addrs);
2266         dev->data->mac_addrs = NULL;
2267
2268 fail_mac_addrs:
2269         sfc_kvargs_cleanup(sa);
2270
2271 fail_kvargs_parse:
2272         sfc_log_init(sa, "failed %d", rc);
2273         dev->process_private = NULL;
2274         free(sa);
2275
2276 fail_alloc_sa:
2277         SFC_ASSERT(rc > 0);
2278         return -rc;
2279 }
2280
2281 static int
2282 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2283 {
2284         sfc_dev_close(dev);
2285
2286         return 0;
2287 }
2288
2289 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2290         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2291         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2292         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2293         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2294         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2295         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2296         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2297         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2298         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2299         { .vendor_id = 0 /* sentinel */ }
2300 };
2301
2302 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2303         struct rte_pci_device *pci_dev)
2304 {
2305         return rte_eth_dev_pci_generic_probe(pci_dev,
2306                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2307 }
2308
2309 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2310 {
2311         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2312 }
2313
2314 static struct rte_pci_driver sfc_efx_pmd = {
2315         .id_table = pci_id_sfc_efx_map,
2316         .drv_flags =
2317                 RTE_PCI_DRV_INTR_LSC |
2318                 RTE_PCI_DRV_NEED_MAPPING,
2319         .probe = sfc_eth_dev_pci_probe,
2320         .remove = sfc_eth_dev_pci_remove,
2321 };
2322
2323 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2324 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2325 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2326 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2327         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2328         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2329         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2330         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2331         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2332         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2333
2334 RTE_INIT(sfc_driver_register_logtype)
2335 {
2336         int ret;
2337
2338         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2339                                                    RTE_LOG_NOTICE);
2340         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2341 }