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 <rte_ethdev_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #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         } else {
 644                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 645                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 646                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 647                 /*
 648                  * Take into account stats which are whenever supported
 649                  * on EF10. If some stat is not supported by current
 650                  * firmware variant or HW revision, it is guaranteed
 651                  * to be zero in mac_stats.
 652                  */
 653                 stats->imissed =
 654                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 655                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 656                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 657                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 658                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 659                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 660                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 661                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 662                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 663                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 664                 stats->ierrors =
 665                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 666                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 667                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 668                 /* no oerrors counters supported on EF10 */
 669
 670                 /* Exclude missed, errors and pauses from Rx packets */
 671                 sfc_update_diff_stat(&port->ipackets,
 672                         mac_stats[EFX_MAC_RX_PKTS] -
 673                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 674                         stats->imissed - stats->ierrors);
 675                 stats->ipackets = port->ipackets;
 676         }
 677
 678 unlock:
 679         rte_spinlock_unlock(&port->mac_stats_lock);
 680         SFC_ASSERT(ret >= 0);
 681         return -ret;
 682 }
 683
 684 static int
 685 sfc_stats_reset(struct rte_eth_dev *dev)
 686 {
 687         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 688         struct sfc_port *port = &sa->port;
 689         int rc;
 690
 691         if (sa->state != SFC_ADAPTER_STARTED) {
 692                 /*
 693                  * The operation cannot be done if port is not started; it
 694                  * will be scheduled to be done during the next port start
 695                  */
 696                 port->mac_stats_reset_pending = B_TRUE;
 697                 return 0;
 698         }
 699
 700         rc = sfc_port_reset_mac_stats(sa);
 701         if (rc != 0)
 702                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 703
 704         SFC_ASSERT(rc >= 0);
 705         return -rc;
 706 }
 707
 708 static int
 709 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 710                unsigned int xstats_count)
 711 {
 712         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 713         struct sfc_port *port = &sa->port;
 714         uint64_t *mac_stats;
 715         int rc;
 716         unsigned int i;
 717         int nstats = 0;
 718
 719         rte_spinlock_lock(&port->mac_stats_lock);
 720
 721         rc = sfc_port_update_mac_stats(sa);
 722         if (rc != 0) {
 723                 SFC_ASSERT(rc > 0);
 724                 nstats = -rc;
 725                 goto unlock;
 726         }
 727
 728         mac_stats = port->mac_stats_buf;
 729
 730         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 731                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 732                         if (xstats != NULL && nstats < (int)xstats_count) {
 733                                 xstats[nstats].id = nstats;
 734                                 xstats[nstats].value = mac_stats[i];
 735                         }
 736                         nstats++;
 737                 }
 738         }
 739
 740 unlock:
 741         rte_spinlock_unlock(&port->mac_stats_lock);
 742
 743         return nstats;
 744 }
 745
 746 static int
 747 sfc_xstats_get_names(struct rte_eth_dev *dev,
 748                      struct rte_eth_xstat_name *xstats_names,
 749                      unsigned int xstats_count)
 750 {
 751         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 752         struct sfc_port *port = &sa->port;
 753         unsigned int i;
 754         unsigned int nstats = 0;
 755
 756         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 757                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 758                         if (xstats_names != NULL && nstats < xstats_count)
 759                                 strlcpy(xstats_names[nstats].name,
 760                                         efx_mac_stat_name(sa->nic, i),
 761                                         sizeof(xstats_names[0].name));
 762                         nstats++;
 763                 }
 764         }
 765
 766         return nstats;
 767 }
 768
 769 static int
 770 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 771                      uint64_t *values, unsigned int n)
 772 {
 773         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 774         struct sfc_port *port = &sa->port;
 775         uint64_t *mac_stats;
 776         unsigned int nb_supported = 0;
 777         unsigned int nb_written = 0;
 778         unsigned int i;
 779         int ret;
 780         int rc;
 781
 782         if (unlikely(values == NULL) ||
 783             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 784                 return port->mac_stats_nb_supported;
 785
 786         rte_spinlock_lock(&port->mac_stats_lock);
 787
 788         rc = sfc_port_update_mac_stats(sa);
 789         if (rc != 0) {
 790                 SFC_ASSERT(rc > 0);
 791                 ret = -rc;
 792                 goto unlock;
 793         }
 794
 795         mac_stats = port->mac_stats_buf;
 796
 797         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 798                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 799                         continue;
 800
 801                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 802                         values[nb_written++] = mac_stats[i];
 803
 804                 ++nb_supported;
 805         }
 806
 807         ret = nb_written;
 808
 809 unlock:
 810         rte_spinlock_unlock(&port->mac_stats_lock);
 811
 812         return ret;
 813 }
 814
 815 static int
 816 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 817                            struct rte_eth_xstat_name *xstats_names,
 818                            const uint64_t *ids, unsigned int size)
 819 {
 820         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 821         struct sfc_port *port = &sa->port;
 822         unsigned int nb_supported = 0;
 823         unsigned int nb_written = 0;
 824         unsigned int i;
 825
 826         if (unlikely(xstats_names == NULL) ||
 827             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 828                 return port->mac_stats_nb_supported;
 829
 830         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 831                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 832                         continue;
 833
 834                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 835                         char *name = xstats_names[nb_written++].name;
 836
 837                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 838                                 sizeof(xstats_names[0].name));
 839                 }
 840
 841                 ++nb_supported;
 842         }
 843
 844         return nb_written;
 845 }
 846
 847 static int
 848 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 849 {
 850         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 851         unsigned int wanted_fc, link_fc;
 852
 853         memset(fc_conf, 0, sizeof(*fc_conf));
 854
 855         sfc_adapter_lock(sa);
 856
 857         if (sa->state == SFC_ADAPTER_STARTED)
 858                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 859         else
 860                 link_fc = sa->port.flow_ctrl;
 861
 862         switch (link_fc) {
 863         case 0:
 864                 fc_conf->mode = RTE_FC_NONE;
 865                 break;
 866         case EFX_FCNTL_RESPOND:
 867                 fc_conf->mode = RTE_FC_RX_PAUSE;
 868                 break;
 869         case EFX_FCNTL_GENERATE:
 870                 fc_conf->mode = RTE_FC_TX_PAUSE;
 871                 break;
 872         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 873                 fc_conf->mode = RTE_FC_FULL;
 874                 break;
 875         default:
 876                 sfc_err(sa, "%s: unexpected flow control value %#x",
 877                         __func__, link_fc);
 878         }
 879
 880         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 881
 882         sfc_adapter_unlock(sa);
 883
 884         return 0;
 885 }
 886
 887 static int
 888 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 889 {
 890         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 891         struct sfc_port *port = &sa->port;
 892         unsigned int fcntl;
 893         int rc;
 894
 895         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 896             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 897             fc_conf->mac_ctrl_frame_fwd != 0) {
 898                 sfc_err(sa, "unsupported flow control settings specified");
 899                 rc = EINVAL;
 900                 goto fail_inval;
 901         }
 902
 903         switch (fc_conf->mode) {
 904         case RTE_FC_NONE:
 905                 fcntl = 0;
 906                 break;
 907         case RTE_FC_RX_PAUSE:
 908                 fcntl = EFX_FCNTL_RESPOND;
 909                 break;
 910         case RTE_FC_TX_PAUSE:
 911                 fcntl = EFX_FCNTL_GENERATE;
 912                 break;
 913         case RTE_FC_FULL:
 914                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 915                 break;
 916         default:
 917                 rc = EINVAL;
 918                 goto fail_inval;
 919         }
 920
 921         sfc_adapter_lock(sa);
 922
 923         if (sa->state == SFC_ADAPTER_STARTED) {
 924                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 925                 if (rc != 0)
 926                         goto fail_mac_fcntl_set;
 927         }
 928
 929         port->flow_ctrl = fcntl;
 930         port->flow_ctrl_autoneg = fc_conf->autoneg;
 931
 932         sfc_adapter_unlock(sa);
 933
 934         return 0;
 935
 936 fail_mac_fcntl_set:
 937         sfc_adapter_unlock(sa);
 938 fail_inval:
 939         SFC_ASSERT(rc > 0);
 940         return -rc;
 941 }
 942
 943 static int
 944 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 945 {
 946         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 947         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 948         boolean_t scatter_enabled;
 949         const char *error;
 950         unsigned int i;
 951
 952         for (i = 0; i < sas->rxq_count; i++) {
 953                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 954                         continue;
 955
 956                 scatter_enabled = (sas->rxq_info[i].type_flags &
 957                                    EFX_RXQ_FLAG_SCATTER);
 958
 959                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 960                                           encp->enc_rx_prefix_size,
 961                                           scatter_enabled,
 962                                           encp->enc_rx_scatter_max, &error)) {
 963                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 964                                 error);
 965                         return EINVAL;
 966                 }
 967         }
 968
 969         return 0;
 970 }
 971
 972 static int
 973 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 974 {
 975         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 976         size_t pdu = EFX_MAC_PDU(mtu);
 977         size_t old_pdu;
 978         int rc;
 979
 980         sfc_log_init(sa, "mtu=%u", mtu);
 981
 982         rc = EINVAL;
 983         if (pdu < EFX_MAC_PDU_MIN) {
 984                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 985                         (unsigned int)mtu, (unsigned int)pdu,
 986                         EFX_MAC_PDU_MIN);
 987                 goto fail_inval;
 988         }
 989         if (pdu > EFX_MAC_PDU_MAX) {
 990                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 991                         (unsigned int)mtu, (unsigned int)pdu,
 992                         (unsigned int)EFX_MAC_PDU_MAX);
 993                 goto fail_inval;
 994         }
 995
 996         sfc_adapter_lock(sa);
 997
 998         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
 999         if (rc != 0)
1000                 goto fail_check_scatter;
1001
1002         if (pdu != sa->port.pdu) {
1003                 if (sa->state == SFC_ADAPTER_STARTED) {
1004                         sfc_stop(sa);
1005
1006                         old_pdu = sa->port.pdu;
1007                         sa->port.pdu = pdu;
1008                         rc = sfc_start(sa);
1009                         if (rc != 0)
1010                                 goto fail_start;
1011                 } else {
1012                         sa->port.pdu = pdu;
1013                 }
1014         }
1015
1016         /*
1017          * The driver does not use it, but other PMDs update jumbo frame
1018          * flag and max_rx_pkt_len when MTU is set.
1019          */
1020         if (mtu > RTE_ETHER_MAX_LEN) {
1021                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1022                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1023         }
1024
1025         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1026
1027         sfc_adapter_unlock(sa);
1028
1029         sfc_log_init(sa, "done");
1030         return 0;
1031
1032 fail_start:
1033         sa->port.pdu = old_pdu;
1034         if (sfc_start(sa) != 0)
1035                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1036                         "PDU max size - port is stopped",
1037                         (unsigned int)pdu, (unsigned int)old_pdu);
1038
1039 fail_check_scatter:
1040         sfc_adapter_unlock(sa);
1041
1042 fail_inval:
1043         sfc_log_init(sa, "failed %d", rc);
1044         SFC_ASSERT(rc > 0);
1045         return -rc;
1046 }
1047 static int
1048 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1049 {
1050         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1051         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1052         struct sfc_port *port = &sa->port;
1053         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1054         int rc = 0;
1055
1056         sfc_adapter_lock(sa);
1057
1058         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1059                 goto unlock;
1060
1061         /*
1062          * Copy the address to the device private data so that
1063          * it could be recalled in the case of adapter restart.
1064          */
1065         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1066
1067         /*
1068          * Neither of the two following checks can return
1069          * an error. The new MAC address is preserved in
1070          * the device private data and can be activated
1071          * on the next port start if the user prevents
1072          * isolated mode from being enabled.
1073          */
1074         if (sfc_sa2shared(sa)->isolated) {
1075                 sfc_warn(sa, "isolated mode is active on the port");
1076                 sfc_warn(sa, "will not set MAC address");
1077                 goto unlock;
1078         }
1079
1080         if (sa->state != SFC_ADAPTER_STARTED) {
1081                 sfc_notice(sa, "the port is not started");
1082                 sfc_notice(sa, "the new MAC address will be set on port start");
1083
1084                 goto unlock;
1085         }
1086
1087         if (encp->enc_allow_set_mac_with_installed_filters) {
1088                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1089                 if (rc != 0) {
1090                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1091                         goto unlock;
1092                 }
1093
1094                 /*
1095                  * Changing the MAC address by means of MCDI request
1096                  * has no effect on received traffic, therefore
1097                  * we also need to update unicast filters
1098                  */
1099                 rc = sfc_set_rx_mode_unchecked(sa);
1100                 if (rc != 0) {
1101                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1102                         /* Rollback the old address */
1103                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1104                         (void)sfc_set_rx_mode_unchecked(sa);
1105                 }
1106         } else {
1107                 sfc_warn(sa, "cannot set MAC address with filters installed");
1108                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1109                 sfc_warn(sa, "(some traffic may be dropped)");
1110
1111                 /*
1112                  * Since setting MAC address with filters installed is not
1113                  * allowed on the adapter, the new MAC address will be set
1114                  * by means of adapter restart. sfc_start() shall retrieve
1115                  * the new address from the device private data and set it.
1116                  */
1117                 sfc_stop(sa);
1118                 rc = sfc_start(sa);
1119                 if (rc != 0)
1120                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1121         }
1122
1123 unlock:
1124         if (rc != 0)
1125                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1126
1127         sfc_adapter_unlock(sa);
1128
1129         SFC_ASSERT(rc >= 0);
1130         return -rc;
1131 }
1132
1133
1134 static int
1135 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1136                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1137 {
1138         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1139         struct sfc_port *port = &sa->port;
1140         uint8_t *mc_addrs = port->mcast_addrs;
1141         int rc;
1142         unsigned int i;
1143
1144         if (sfc_sa2shared(sa)->isolated) {
1145                 sfc_err(sa, "isolated mode is active on the port");
1146                 sfc_err(sa, "will not set multicast address list");
1147                 return -ENOTSUP;
1148         }
1149
1150         if (mc_addrs == NULL)
1151                 return -ENOBUFS;
1152
1153         if (nb_mc_addr > port->max_mcast_addrs) {
1154                 sfc_err(sa, "too many multicast addresses: %u > %u",
1155                          nb_mc_addr, port->max_mcast_addrs);
1156                 return -EINVAL;
1157         }
1158
1159         for (i = 0; i < nb_mc_addr; ++i) {
1160                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1161                                  EFX_MAC_ADDR_LEN);
1162                 mc_addrs += EFX_MAC_ADDR_LEN;
1163         }
1164
1165         port->nb_mcast_addrs = nb_mc_addr;
1166
1167         if (sa->state != SFC_ADAPTER_STARTED)
1168                 return 0;
1169
1170         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1171                                         port->nb_mcast_addrs);
1172         if (rc != 0)
1173                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1174
1175         SFC_ASSERT(rc >= 0);
1176         return -rc;
1177 }
1178
1179 /*
1180  * The function is used by the secondary process as well. It must not
1181  * use any process-local pointers from the adapter data.
1182  */
1183 static void
1184 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1185                       struct rte_eth_rxq_info *qinfo)
1186 {
1187         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1188         struct sfc_rxq_info *rxq_info;
1189
1190         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1191
1192         rxq_info = &sas->rxq_info[rx_queue_id];
1193
1194         qinfo->mp = rxq_info->refill_mb_pool;
1195         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1196         qinfo->conf.rx_drop_en = 1;
1197         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1198         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1199         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1200                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1201                 qinfo->scattered_rx = 1;
1202         }
1203         qinfo->nb_desc = rxq_info->entries;
1204 }
1205
1206 /*
1207  * The function is used by the secondary process as well. It must not
1208  * use any process-local pointers from the adapter data.
1209  */
1210 static void
1211 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1212                       struct rte_eth_txq_info *qinfo)
1213 {
1214         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1215         struct sfc_txq_info *txq_info;
1216
1217         SFC_ASSERT(tx_queue_id < sas->txq_count);
1218
1219         txq_info = &sas->txq_info[tx_queue_id];
1220
1221         memset(qinfo, 0, sizeof(*qinfo));
1222
1223         qinfo->conf.offloads = txq_info->offloads;
1224         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1225         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1226         qinfo->nb_desc = txq_info->entries;
1227 }
1228
1229 /*
1230  * The function is used by the secondary process as well. It must not
1231  * use any process-local pointers from the adapter data.
1232  */
1233 static uint32_t
1234 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1235 {
1236         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1237         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1238         struct sfc_rxq_info *rxq_info;
1239
1240         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1241         rxq_info = &sas->rxq_info[rx_queue_id];
1242
1243         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1244                 return 0;
1245
1246         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1247 }
1248
1249 /*
1250  * The function is used by the secondary process as well. It must not
1251  * use any process-local pointers from the adapter data.
1252  */
1253 static int
1254 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1255 {
1256         struct sfc_dp_rxq *dp_rxq = queue;
1257         const struct sfc_dp_rx *dp_rx;
1258
1259         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1260
1261         return offset < dp_rx->qdesc_npending(dp_rxq);
1262 }
1263
1264 /*
1265  * The function is used by the secondary process as well. It must not
1266  * use any process-local pointers from the adapter data.
1267  */
1268 static int
1269 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1270 {
1271         struct sfc_dp_rxq *dp_rxq = queue;
1272         const struct sfc_dp_rx *dp_rx;
1273
1274         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1275
1276         return dp_rx->qdesc_status(dp_rxq, offset);
1277 }
1278
1279 /*
1280  * The function is used by the secondary process as well. It must not
1281  * use any process-local pointers from the adapter data.
1282  */
1283 static int
1284 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1285 {
1286         struct sfc_dp_txq *dp_txq = queue;
1287         const struct sfc_dp_tx *dp_tx;
1288
1289         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1290
1291         return dp_tx->qdesc_status(dp_txq, offset);
1292 }
1293
1294 static int
1295 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1296 {
1297         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1298         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1299         int rc;
1300
1301         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1302
1303         sfc_adapter_lock(sa);
1304
1305         rc = EINVAL;
1306         if (sa->state != SFC_ADAPTER_STARTED)
1307                 goto fail_not_started;
1308
1309         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1310                 goto fail_not_setup;
1311
1312         rc = sfc_rx_qstart(sa, rx_queue_id);
1313         if (rc != 0)
1314                 goto fail_rx_qstart;
1315
1316         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1317
1318         sfc_adapter_unlock(sa);
1319
1320         return 0;
1321
1322 fail_rx_qstart:
1323 fail_not_setup:
1324 fail_not_started:
1325         sfc_adapter_unlock(sa);
1326         SFC_ASSERT(rc > 0);
1327         return -rc;
1328 }
1329
1330 static int
1331 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1332 {
1333         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1334         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1335
1336         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1337
1338         sfc_adapter_lock(sa);
1339         sfc_rx_qstop(sa, rx_queue_id);
1340
1341         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1342
1343         sfc_adapter_unlock(sa);
1344
1345         return 0;
1346 }
1347
1348 static int
1349 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1350 {
1351         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1352         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1353         int rc;
1354
1355         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1356
1357         sfc_adapter_lock(sa);
1358
1359         rc = EINVAL;
1360         if (sa->state != SFC_ADAPTER_STARTED)
1361                 goto fail_not_started;
1362
1363         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1364                 goto fail_not_setup;
1365
1366         rc = sfc_tx_qstart(sa, tx_queue_id);
1367         if (rc != 0)
1368                 goto fail_tx_qstart;
1369
1370         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1371
1372         sfc_adapter_unlock(sa);
1373         return 0;
1374
1375 fail_tx_qstart:
1376
1377 fail_not_setup:
1378 fail_not_started:
1379         sfc_adapter_unlock(sa);
1380         SFC_ASSERT(rc > 0);
1381         return -rc;
1382 }
1383
1384 static int
1385 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1386 {
1387         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1388         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1389
1390         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1391
1392         sfc_adapter_lock(sa);
1393
1394         sfc_tx_qstop(sa, tx_queue_id);
1395
1396         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1397
1398         sfc_adapter_unlock(sa);
1399         return 0;
1400 }
1401
1402 static efx_tunnel_protocol_t
1403 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1404 {
1405         switch (rte_type) {
1406         case RTE_TUNNEL_TYPE_VXLAN:
1407                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1408         case RTE_TUNNEL_TYPE_GENEVE:
1409                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1410         default:
1411                 return EFX_TUNNEL_NPROTOS;
1412         }
1413 }
1414
1415 enum sfc_udp_tunnel_op_e {
1416         SFC_UDP_TUNNEL_ADD_PORT,
1417         SFC_UDP_TUNNEL_DEL_PORT,
1418 };
1419
1420 static int
1421 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1422                       struct rte_eth_udp_tunnel *tunnel_udp,
1423                       enum sfc_udp_tunnel_op_e op)
1424 {
1425         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1426         efx_tunnel_protocol_t tunnel_proto;
1427         int rc;
1428
1429         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1430                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1431                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1432                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1433
1434         tunnel_proto =
1435                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1436         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1437                 rc = ENOTSUP;
1438                 goto fail_bad_proto;
1439         }
1440
1441         sfc_adapter_lock(sa);
1442
1443         switch (op) {
1444         case SFC_UDP_TUNNEL_ADD_PORT:
1445                 rc = efx_tunnel_config_udp_add(sa->nic,
1446                                                tunnel_udp->udp_port,
1447                                                tunnel_proto);
1448                 break;
1449         case SFC_UDP_TUNNEL_DEL_PORT:
1450                 rc = efx_tunnel_config_udp_remove(sa->nic,
1451                                                   tunnel_udp->udp_port,
1452                                                   tunnel_proto);
1453                 break;
1454         default:
1455                 rc = EINVAL;
1456                 goto fail_bad_op;
1457         }
1458
1459         if (rc != 0)
1460                 goto fail_op;
1461
1462         if (sa->state == SFC_ADAPTER_STARTED) {
1463                 rc = efx_tunnel_reconfigure(sa->nic);
1464                 if (rc == EAGAIN) {
1465                         /*
1466                          * Configuration is accepted by FW and MC reboot
1467                          * is initiated to apply the changes. MC reboot
1468                          * will be handled in a usual way (MC reboot
1469                          * event on management event queue and adapter
1470                          * restart).
1471                          */
1472                         rc = 0;
1473                 } else if (rc != 0) {
1474                         goto fail_reconfigure;
1475                 }
1476         }
1477
1478         sfc_adapter_unlock(sa);
1479         return 0;
1480
1481 fail_reconfigure:
1482         /* Remove/restore entry since the change makes the trouble */
1483         switch (op) {
1484         case SFC_UDP_TUNNEL_ADD_PORT:
1485                 (void)efx_tunnel_config_udp_remove(sa->nic,
1486                                                    tunnel_udp->udp_port,
1487                                                    tunnel_proto);
1488                 break;
1489         case SFC_UDP_TUNNEL_DEL_PORT:
1490                 (void)efx_tunnel_config_udp_add(sa->nic,
1491                                                 tunnel_udp->udp_port,
1492                                                 tunnel_proto);
1493                 break;
1494         }
1495
1496 fail_op:
1497 fail_bad_op:
1498         sfc_adapter_unlock(sa);
1499
1500 fail_bad_proto:
1501         SFC_ASSERT(rc > 0);
1502         return -rc;
1503 }
1504
1505 static int
1506 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1507                             struct rte_eth_udp_tunnel *tunnel_udp)
1508 {
1509         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1510 }
1511
1512 static int
1513 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1514                             struct rte_eth_udp_tunnel *tunnel_udp)
1515 {
1516         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1517 }
1518
1519 /*
1520  * The function is used by the secondary process as well. It must not
1521  * use any process-local pointers from the adapter data.
1522  */
1523 static int
1524 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1525                           struct rte_eth_rss_conf *rss_conf)
1526 {
1527         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1528         struct sfc_rss *rss = &sas->rss;
1529
1530         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1531                 return -ENOTSUP;
1532
1533         /*
1534          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1535          * hence, conversion is done here to derive a correct set of ETH_RSS
1536          * flags which corresponds to the active EFX configuration stored
1537          * locally in 'sfc_adapter' and kept up-to-date
1538          */
1539         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1540         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1541         if (rss_conf->rss_key != NULL)
1542                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1543
1544         return 0;
1545 }
1546
1547 static int
1548 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1549                         struct rte_eth_rss_conf *rss_conf)
1550 {
1551         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1552         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1553         unsigned int efx_hash_types;
1554         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1555         unsigned int n_contexts;
1556         unsigned int mode_i = 0;
1557         unsigned int key_i = 0;
1558         unsigned int i = 0;
1559         int rc = 0;
1560
1561         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1562
1563         if (sfc_sa2shared(sa)->isolated)
1564                 return -ENOTSUP;
1565
1566         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1567                 sfc_err(sa, "RSS is not available");
1568                 return -ENOTSUP;
1569         }
1570
1571         if (rss->channels == 0) {
1572                 sfc_err(sa, "RSS is not configured");
1573                 return -EINVAL;
1574         }
1575
1576         if ((rss_conf->rss_key != NULL) &&
1577             (rss_conf->rss_key_len != sizeof(rss->key))) {
1578                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1579                         sizeof(rss->key));
1580                 return -EINVAL;
1581         }
1582
1583         sfc_adapter_lock(sa);
1584
1585         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1586         if (rc != 0)
1587                 goto fail_rx_hf_rte_to_efx;
1588
1589         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1590                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1591                                            rss->hash_alg, efx_hash_types,
1592                                            B_TRUE);
1593                 if (rc != 0)
1594                         goto fail_scale_mode_set;
1595         }
1596
1597         if (rss_conf->rss_key != NULL) {
1598                 if (sa->state == SFC_ADAPTER_STARTED) {
1599                         for (key_i = 0; key_i < n_contexts; key_i++) {
1600                                 rc = efx_rx_scale_key_set(sa->nic,
1601                                                           contexts[key_i],
1602                                                           rss_conf->rss_key,
1603                                                           sizeof(rss->key));
1604                                 if (rc != 0)
1605                                         goto fail_scale_key_set;
1606                         }
1607                 }
1608
1609                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1610         }
1611
1612         rss->hash_types = efx_hash_types;
1613
1614         sfc_adapter_unlock(sa);
1615
1616         return 0;
1617
1618 fail_scale_key_set:
1619         for (i = 0; i < key_i; i++) {
1620                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1621                                          sizeof(rss->key)) != 0)
1622                         sfc_err(sa, "failed to restore RSS key");
1623         }
1624
1625 fail_scale_mode_set:
1626         for (i = 0; i < mode_i; i++) {
1627                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1628                                           EFX_RX_HASHALG_TOEPLITZ,
1629                                           rss->hash_types, B_TRUE) != 0)
1630                         sfc_err(sa, "failed to restore RSS mode");
1631         }
1632
1633 fail_rx_hf_rte_to_efx:
1634         sfc_adapter_unlock(sa);
1635         return -rc;
1636 }
1637
1638 /*
1639  * The function is used by the secondary process as well. It must not
1640  * use any process-local pointers from the adapter data.
1641  */
1642 static int
1643 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1644                        struct rte_eth_rss_reta_entry64 *reta_conf,
1645                        uint16_t reta_size)
1646 {
1647         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1648         struct sfc_rss *rss = &sas->rss;
1649         int entry;
1650
1651         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1652                 return -ENOTSUP;
1653
1654         if (rss->channels == 0)
1655                 return -EINVAL;
1656
1657         if (reta_size != EFX_RSS_TBL_SIZE)
1658                 return -EINVAL;
1659
1660         for (entry = 0; entry < reta_size; entry++) {
1661                 int grp = entry / RTE_RETA_GROUP_SIZE;
1662                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1663
1664                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1665                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1666         }
1667
1668         return 0;
1669 }
1670
1671 static int
1672 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1673                         struct rte_eth_rss_reta_entry64 *reta_conf,
1674                         uint16_t reta_size)
1675 {
1676         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1677         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1678         unsigned int *rss_tbl_new;
1679         uint16_t entry;
1680         int rc = 0;
1681
1682
1683         if (sfc_sa2shared(sa)->isolated)
1684                 return -ENOTSUP;
1685
1686         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1687                 sfc_err(sa, "RSS is not available");
1688                 return -ENOTSUP;
1689         }
1690
1691         if (rss->channels == 0) {
1692                 sfc_err(sa, "RSS is not configured");
1693                 return -EINVAL;
1694         }
1695
1696         if (reta_size != EFX_RSS_TBL_SIZE) {
1697                 sfc_err(sa, "RETA size is wrong (should be %u)",
1698                         EFX_RSS_TBL_SIZE);
1699                 return -EINVAL;
1700         }
1701
1702         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1703         if (rss_tbl_new == NULL)
1704                 return -ENOMEM;
1705
1706         sfc_adapter_lock(sa);
1707
1708         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1709
1710         for (entry = 0; entry < reta_size; entry++) {
1711                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1712                 struct rte_eth_rss_reta_entry64 *grp;
1713
1714                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1715
1716                 if (grp->mask & (1ull << grp_idx)) {
1717                         if (grp->reta[grp_idx] >= rss->channels) {
1718                                 rc = EINVAL;
1719                                 goto bad_reta_entry;
1720                         }
1721                         rss_tbl_new[entry] = grp->reta[grp_idx];
1722                 }
1723         }
1724
1725         if (sa->state == SFC_ADAPTER_STARTED) {
1726                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1727                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1728                 if (rc != 0)
1729                         goto fail_scale_tbl_set;
1730         }
1731
1732         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1733
1734 fail_scale_tbl_set:
1735 bad_reta_entry:
1736         sfc_adapter_unlock(sa);
1737
1738         rte_free(rss_tbl_new);
1739
1740         SFC_ASSERT(rc >= 0);
1741         return -rc;
1742 }
1743
1744 static int
1745 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1746                     enum rte_filter_op filter_op,
1747                     void *arg)
1748 {
1749         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1750         int rc = ENOTSUP;
1751
1752         sfc_log_init(sa, "entry");
1753
1754         switch (filter_type) {
1755         case RTE_ETH_FILTER_GENERIC:
1756                 if (filter_op != RTE_ETH_FILTER_GET) {
1757                         rc = EINVAL;
1758                 } else {
1759                         *(const void **)arg = &sfc_flow_ops;
1760                         rc = 0;
1761                 }
1762                 break;
1763         default:
1764                 sfc_err(sa, "Unknown filter type %u", filter_type);
1765                 break;
1766         }
1767
1768         sfc_log_init(sa, "exit: %d", -rc);
1769         SFC_ASSERT(rc >= 0);
1770         return -rc;
1771 }
1772
1773 static int
1774 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1775 {
1776         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1777
1778         /*
1779          * If Rx datapath does not provide callback to check mempool,
1780          * all pools are supported.
1781          */
1782         if (sap->dp_rx->pool_ops_supported == NULL)
1783                 return 1;
1784
1785         return sap->dp_rx->pool_ops_supported(pool);
1786 }
1787
1788 static int
1789 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1790 {
1791         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1792         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1793         struct sfc_rxq_info *rxq_info;
1794
1795         SFC_ASSERT(queue_id < sas->rxq_count);
1796         rxq_info = &sas->rxq_info[queue_id];
1797
1798         return sap->dp_rx->intr_enable(rxq_info->dp);
1799 }
1800
1801 static int
1802 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1803 {
1804         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1805         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1806         struct sfc_rxq_info *rxq_info;
1807
1808         SFC_ASSERT(queue_id < sas->rxq_count);
1809         rxq_info = &sas->rxq_info[queue_id];
1810
1811         return sap->dp_rx->intr_disable(rxq_info->dp);
1812 }
1813
1814 static const struct eth_dev_ops sfc_eth_dev_ops = {
1815         .dev_configure                  = sfc_dev_configure,
1816         .dev_start                      = sfc_dev_start,
1817         .dev_stop                       = sfc_dev_stop,
1818         .dev_set_link_up                = sfc_dev_set_link_up,
1819         .dev_set_link_down              = sfc_dev_set_link_down,
1820         .dev_close                      = sfc_dev_close,
1821         .promiscuous_enable             = sfc_dev_promisc_enable,
1822         .promiscuous_disable            = sfc_dev_promisc_disable,
1823         .allmulticast_enable            = sfc_dev_allmulti_enable,
1824         .allmulticast_disable           = sfc_dev_allmulti_disable,
1825         .link_update                    = sfc_dev_link_update,
1826         .stats_get                      = sfc_stats_get,
1827         .stats_reset                    = sfc_stats_reset,
1828         .xstats_get                     = sfc_xstats_get,
1829         .xstats_reset                   = sfc_stats_reset,
1830         .xstats_get_names               = sfc_xstats_get_names,
1831         .dev_infos_get                  = sfc_dev_infos_get,
1832         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1833         .mtu_set                        = sfc_dev_set_mtu,
1834         .rx_queue_start                 = sfc_rx_queue_start,
1835         .rx_queue_stop                  = sfc_rx_queue_stop,
1836         .tx_queue_start                 = sfc_tx_queue_start,
1837         .tx_queue_stop                  = sfc_tx_queue_stop,
1838         .rx_queue_setup                 = sfc_rx_queue_setup,
1839         .rx_queue_release               = sfc_rx_queue_release,
1840         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1841         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1842         .tx_queue_setup                 = sfc_tx_queue_setup,
1843         .tx_queue_release               = sfc_tx_queue_release,
1844         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1845         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1846         .mac_addr_set                   = sfc_mac_addr_set,
1847         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1848         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1849         .reta_update                    = sfc_dev_rss_reta_update,
1850         .reta_query                     = sfc_dev_rss_reta_query,
1851         .rss_hash_update                = sfc_dev_rss_hash_update,
1852         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1853         .filter_ctrl                    = sfc_dev_filter_ctrl,
1854         .set_mc_addr_list               = sfc_set_mc_addr_list,
1855         .rxq_info_get                   = sfc_rx_queue_info_get,
1856         .txq_info_get                   = sfc_tx_queue_info_get,
1857         .fw_version_get                 = sfc_fw_version_get,
1858         .xstats_get_by_id               = sfc_xstats_get_by_id,
1859         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1860         .pool_ops_supported             = sfc_pool_ops_supported,
1861 };
1862
1863 /**
1864  * Duplicate a string in potentially shared memory required for
1865  * multi-process support.
1866  *
1867  * strdup() allocates from process-local heap/memory.
1868  */
1869 static char *
1870 sfc_strdup(const char *str)
1871 {
1872         size_t size;
1873         char *copy;
1874
1875         if (str == NULL)
1876                 return NULL;
1877
1878         size = strlen(str) + 1;
1879         copy = rte_malloc(__func__, size, 0);
1880         if (copy != NULL)
1881                 rte_memcpy(copy, str, size);
1882
1883         return copy;
1884 }
1885
1886 static int
1887 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1888 {
1889         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1890         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1891         const struct sfc_dp_rx *dp_rx;
1892         const struct sfc_dp_tx *dp_tx;
1893         const efx_nic_cfg_t *encp;
1894         unsigned int avail_caps = 0;
1895         const char *rx_name = NULL;
1896         const char *tx_name = NULL;
1897         int rc;
1898
1899         switch (sa->family) {
1900         case EFX_FAMILY_HUNTINGTON:
1901         case EFX_FAMILY_MEDFORD:
1902         case EFX_FAMILY_MEDFORD2:
1903                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1904                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1905                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1906                 break;
1907         case EFX_FAMILY_RIVERHEAD:
1908                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1909                 break;
1910         default:
1911                 break;
1912         }
1913
1914         encp = efx_nic_cfg_get(sa->nic);
1915         if (encp->enc_rx_es_super_buffer_supported)
1916                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1917
1918         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1919                                 sfc_kvarg_string_handler, &rx_name);
1920         if (rc != 0)
1921                 goto fail_kvarg_rx_datapath;
1922
1923         if (rx_name != NULL) {
1924                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1925                 if (dp_rx == NULL) {
1926                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1927                         rc = ENOENT;
1928                         goto fail_dp_rx;
1929                 }
1930                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1931                         sfc_err(sa,
1932                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1933                                 rx_name);
1934                         rc = EINVAL;
1935                         goto fail_dp_rx_caps;
1936                 }
1937         } else {
1938                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1939                 if (dp_rx == NULL) {
1940                         sfc_err(sa, "Rx datapath by caps %#x not found",
1941                                 avail_caps);
1942                         rc = ENOENT;
1943                         goto fail_dp_rx;
1944                 }
1945         }
1946
1947         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1948         if (sas->dp_rx_name == NULL) {
1949                 rc = ENOMEM;
1950                 goto fail_dp_rx_name;
1951         }
1952
1953         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1954
1955         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1956                                 sfc_kvarg_string_handler, &tx_name);
1957         if (rc != 0)
1958                 goto fail_kvarg_tx_datapath;
1959
1960         if (tx_name != NULL) {
1961                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1962                 if (dp_tx == NULL) {
1963                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1964                         rc = ENOENT;
1965                         goto fail_dp_tx;
1966                 }
1967                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1968                         sfc_err(sa,
1969                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1970                                 tx_name);
1971                         rc = EINVAL;
1972                         goto fail_dp_tx_caps;
1973                 }
1974         } else {
1975                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1976                 if (dp_tx == NULL) {
1977                         sfc_err(sa, "Tx datapath by caps %#x not found",
1978                                 avail_caps);
1979                         rc = ENOENT;
1980                         goto fail_dp_tx;
1981                 }
1982         }
1983
1984         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1985         if (sas->dp_tx_name == NULL) {
1986                 rc = ENOMEM;
1987                 goto fail_dp_tx_name;
1988         }
1989
1990         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1991
1992         sa->priv.dp_rx = dp_rx;
1993         sa->priv.dp_tx = dp_tx;
1994
1995         dev->rx_pkt_burst = dp_rx->pkt_burst;
1996         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1997         dev->tx_pkt_burst = dp_tx->pkt_burst;
1998
1999         dev->rx_queue_count = sfc_rx_queue_count;
2000         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2001         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2002         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2003         dev->dev_ops = &sfc_eth_dev_ops;
2004
2005         return 0;
2006
2007 fail_dp_tx_name:
2008 fail_dp_tx_caps:
2009 fail_dp_tx:
2010 fail_kvarg_tx_datapath:
2011         rte_free(sas->dp_rx_name);
2012         sas->dp_rx_name = NULL;
2013
2014 fail_dp_rx_name:
2015 fail_dp_rx_caps:
2016 fail_dp_rx:
2017 fail_kvarg_rx_datapath:
2018         return rc;
2019 }
2020
2021 static void
2022 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2023 {
2024         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2025         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2026
2027         dev->dev_ops = NULL;
2028         dev->tx_pkt_prepare = NULL;
2029         dev->rx_pkt_burst = NULL;
2030         dev->tx_pkt_burst = NULL;
2031
2032         rte_free(sas->dp_tx_name);
2033         sas->dp_tx_name = NULL;
2034         sa->priv.dp_tx = NULL;
2035
2036         rte_free(sas->dp_rx_name);
2037         sas->dp_rx_name = NULL;
2038         sa->priv.dp_rx = NULL;
2039 }
2040
2041 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2042         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2043         .reta_query                     = sfc_dev_rss_reta_query,
2044         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2045         .rxq_info_get                   = sfc_rx_queue_info_get,
2046         .txq_info_get                   = sfc_tx_queue_info_get,
2047 };
2048
2049 static int
2050 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2051 {
2052         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2053         struct sfc_adapter_priv *sap;
2054         const struct sfc_dp_rx *dp_rx;
2055         const struct sfc_dp_tx *dp_tx;
2056         int rc;
2057
2058         /*
2059          * Allocate process private data from heap, since it should not
2060          * be located in shared memory allocated using rte_malloc() API.
2061          */
2062         sap = calloc(1, sizeof(*sap));
2063         if (sap == NULL) {
2064                 rc = ENOMEM;
2065                 goto fail_alloc_priv;
2066         }
2067
2068         sap->logtype_main = logtype_main;
2069
2070         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2071         if (dp_rx == NULL) {
2072                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2073                         "cannot find %s Rx datapath", sas->dp_rx_name);
2074                 rc = ENOENT;
2075                 goto fail_dp_rx;
2076         }
2077         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2078                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2079                         "%s Rx datapath does not support multi-process",
2080                         sas->dp_rx_name);
2081                 rc = EINVAL;
2082                 goto fail_dp_rx_multi_process;
2083         }
2084
2085         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2086         if (dp_tx == NULL) {
2087                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2088                         "cannot find %s Tx datapath", sas->dp_tx_name);
2089                 rc = ENOENT;
2090                 goto fail_dp_tx;
2091         }
2092         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2093                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2094                         "%s Tx datapath does not support multi-process",
2095                         sas->dp_tx_name);
2096                 rc = EINVAL;
2097                 goto fail_dp_tx_multi_process;
2098         }
2099
2100         sap->dp_rx = dp_rx;
2101         sap->dp_tx = dp_tx;
2102
2103         dev->process_private = sap;
2104         dev->rx_pkt_burst = dp_rx->pkt_burst;
2105         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2106         dev->tx_pkt_burst = dp_tx->pkt_burst;
2107         dev->rx_queue_count = sfc_rx_queue_count;
2108         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2109         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2110         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2111         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2112
2113         return 0;
2114
2115 fail_dp_tx_multi_process:
2116 fail_dp_tx:
2117 fail_dp_rx_multi_process:
2118 fail_dp_rx:
2119         free(sap);
2120
2121 fail_alloc_priv:
2122         return rc;
2123 }
2124
2125 static void
2126 sfc_register_dp(void)
2127 {
2128         /* Register once */
2129         if (TAILQ_EMPTY(&sfc_dp_head)) {
2130                 /* Prefer EF10 datapath */
2131                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2132                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2133                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2134                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2135
2136                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2137                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2138                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2139                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2140         }
2141 }
2142
2143 static int
2144 sfc_eth_dev_init(struct rte_eth_dev *dev)
2145 {
2146         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2147         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2148         uint32_t logtype_main;
2149         struct sfc_adapter *sa;
2150         int rc;
2151         const efx_nic_cfg_t *encp;
2152         const struct rte_ether_addr *from;
2153         int ret;
2154
2155         sfc_register_dp();
2156
2157         logtype_main = sfc_register_logtype(&pci_dev->addr,
2158                                             SFC_LOGTYPE_MAIN_STR,
2159                                             RTE_LOG_NOTICE);
2160
2161         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2162                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2163
2164         /* Required for logging */
2165         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2166                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2167                         pci_dev->addr.domain, pci_dev->addr.bus,
2168                         pci_dev->addr.devid, pci_dev->addr.function,
2169                         dev->data->port_id);
2170         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2171                 SFC_GENERIC_LOG(ERR,
2172                         "reserved log prefix is too short for " PCI_PRI_FMT,
2173                         pci_dev->addr.domain, pci_dev->addr.bus,
2174                         pci_dev->addr.devid, pci_dev->addr.function);
2175                 return -EINVAL;
2176         }
2177         sas->pci_addr = pci_dev->addr;
2178         sas->port_id = dev->data->port_id;
2179
2180         /*
2181          * Allocate process private data from heap, since it should not
2182          * be located in shared memory allocated using rte_malloc() API.
2183          */
2184         sa = calloc(1, sizeof(*sa));
2185         if (sa == NULL) {
2186                 rc = ENOMEM;
2187                 goto fail_alloc_sa;
2188         }
2189
2190         dev->process_private = sa;
2191
2192         /* Required for logging */
2193         sa->priv.shared = sas;
2194         sa->priv.logtype_main = logtype_main;
2195
2196         sa->eth_dev = dev;
2197
2198         /* Copy PCI device info to the dev->data */
2199         rte_eth_copy_pci_info(dev, pci_dev);
2200         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2201
2202         rc = sfc_kvargs_parse(sa);
2203         if (rc != 0)
2204                 goto fail_kvargs_parse;
2205
2206         sfc_log_init(sa, "entry");
2207
2208         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2209         if (dev->data->mac_addrs == NULL) {
2210                 rc = ENOMEM;
2211                 goto fail_mac_addrs;
2212         }
2213
2214         sfc_adapter_lock_init(sa);
2215         sfc_adapter_lock(sa);
2216
2217         sfc_log_init(sa, "probing");
2218         rc = sfc_probe(sa);
2219         if (rc != 0)
2220                 goto fail_probe;
2221
2222         sfc_log_init(sa, "set device ops");
2223         rc = sfc_eth_dev_set_ops(dev);
2224         if (rc != 0)
2225                 goto fail_set_ops;
2226
2227         sfc_log_init(sa, "attaching");
2228         rc = sfc_attach(sa);
2229         if (rc != 0)
2230                 goto fail_attach;
2231
2232         encp = efx_nic_cfg_get(sa->nic);
2233
2234         /*
2235          * The arguments are really reverse order in comparison to
2236          * Linux kernel. Copy from NIC config to Ethernet device data.
2237          */
2238         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2239         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2240
2241         sfc_adapter_unlock(sa);
2242
2243         sfc_log_init(sa, "done");
2244         return 0;
2245
2246 fail_attach:
2247         sfc_eth_dev_clear_ops(dev);
2248
2249 fail_set_ops:
2250         sfc_unprobe(sa);
2251
2252 fail_probe:
2253         sfc_adapter_unlock(sa);
2254         sfc_adapter_lock_fini(sa);
2255         rte_free(dev->data->mac_addrs);
2256         dev->data->mac_addrs = NULL;
2257
2258 fail_mac_addrs:
2259         sfc_kvargs_cleanup(sa);
2260
2261 fail_kvargs_parse:
2262         sfc_log_init(sa, "failed %d", rc);
2263         dev->process_private = NULL;
2264         free(sa);
2265
2266 fail_alloc_sa:
2267         SFC_ASSERT(rc > 0);
2268         return -rc;
2269 }
2270
2271 static int
2272 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2273 {
2274         sfc_dev_close(dev);
2275
2276         return 0;
2277 }
2278
2279 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2280         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2281         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2282         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2283         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2284         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2285         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2286         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2287         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2288         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2289         { .vendor_id = 0 /* sentinel */ }
2290 };
2291
2292 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2293         struct rte_pci_device *pci_dev)
2294 {
2295         return rte_eth_dev_pci_generic_probe(pci_dev,
2296                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2297 }
2298
2299 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2300 {
2301         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2302 }
2303
2304 static struct rte_pci_driver sfc_efx_pmd = {
2305         .id_table = pci_id_sfc_efx_map,
2306         .drv_flags =
2307                 RTE_PCI_DRV_INTR_LSC |
2308                 RTE_PCI_DRV_NEED_MAPPING,
2309         .probe = sfc_eth_dev_pci_probe,
2310         .remove = sfc_eth_dev_pci_remove,
2311 };
2312
2313 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2314 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2315 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2316 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2317         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2318         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2319         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2320         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2321         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2322         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2323
2324 RTE_INIT(sfc_driver_register_logtype)
2325 {
2326         int ret;
2327
2328         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2329                                                    RTE_LOG_NOTICE);
2330         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2331 }