drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2021 Xilinx, Inc.
   4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17 #include <rte_ether.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_debug.h"
  23 #include "sfc_log.h"
  24 #include "sfc_kvargs.h"
  25 #include "sfc_ev.h"
  26 #include "sfc_rx.h"
  27 #include "sfc_tx.h"
  28 #include "sfc_flow.h"
  29 #include "sfc_dp.h"
  30 #include "sfc_dp_rx.h"
  31 #include "sfc_repr.h"
  32 #include "sfc_sw_stats.h"
  33
  34 #define SFC_XSTAT_ID_INVALID_VAL  UINT64_MAX
  35 #define SFC_XSTAT_ID_INVALID_NAME '\0'
  36
  37 uint32_t sfc_logtype_driver;
  38
  39 static struct sfc_dp_list sfc_dp_head =
  40         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  41
  42
  43 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  44
  45
  46 static int
  47 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  48 {
  49         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  50         efx_nic_fw_info_t enfi;
  51         int ret;
  52         int rc;
  53
  54         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  55         if (rc != 0)
  56                 return -rc;
  57
  58         ret = snprintf(fw_version, fw_size,
  59                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  60                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  61                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  62         if (ret < 0)
  63                 return ret;
  64
  65         if (enfi.enfi_dpcpu_fw_ids_valid) {
  66                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  67                 int ret_extra;
  68
  69                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  70                                      fw_size - dpcpu_fw_ids_offset,
  71                                      " rx%" PRIx16 " tx%" PRIx16,
  72                                      enfi.enfi_rx_dpcpu_fw_id,
  73                                      enfi.enfi_tx_dpcpu_fw_id);
  74                 if (ret_extra < 0)
  75                         return ret_extra;
  76
  77                 ret += ret_extra;
  78         }
  79
  80         if (fw_size < (size_t)(++ret))
  81                 return ret;
  82         else
  83                 return 0;
  84 }
  85
  86 static int
  87 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  88 {
  89         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  90         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  91         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  92         struct sfc_rss *rss = &sas->rss;
  93         struct sfc_mae *mae = &sa->mae;
  94         uint64_t txq_offloads_def = 0;
  95
  96         sfc_log_init(sa, "entry");
  97
  98         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
  99         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 100
 101         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 102
 103         dev_info->max_vfs = sa->sriov.num_vfs;
 104
 105         /* Autonegotiation may be disabled */
 106         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 107         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 108                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 110                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 112                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 114                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 115         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 116                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 117         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 118                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 119
 120         dev_info->max_rx_queues = sa->rxq_max;
 121         dev_info->max_tx_queues = sa->txq_max;
 122
 123         /* By default packets are dropped if no descriptors are available */
 124         dev_info->default_rxconf.rx_drop_en = 1;
 125
 126         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 127
 128         /*
 129          * rx_offload_capa includes both device and queue offloads since
 130          * the latter may be requested on a per device basis which makes
 131          * sense when some offloads are needed to be set on all queues.
 132          */
 133         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 134                                     dev_info->rx_queue_offload_capa;
 135
 136         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 137
 138         /*
 139          * tx_offload_capa includes both device and queue offloads since
 140          * the latter may be requested on a per device basis which makes
 141          * sense when some offloads are needed to be set on all queues.
 142          */
 143         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 144                                     dev_info->tx_queue_offload_capa;
 145
 146         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 147                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 148
 149         dev_info->default_txconf.offloads |= txq_offloads_def;
 150
 151         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 152                 uint64_t rte_hf = 0;
 153                 unsigned int i;
 154
 155                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 156                         rte_hf |= rss->hf_map[i].rte;
 157
 158                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 159                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 160                 dev_info->flow_type_rss_offloads = rte_hf;
 161         }
 162
 163         /* Initialize to hardware limits */
 164         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 165         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 166         /* The RXQ hardware requires that the descriptor count is a power
 167          * of 2, but rx_desc_lim cannot properly describe that constraint.
 168          */
 169         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 170
 171         /* Initialize to hardware limits */
 172         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 173         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 174         /*
 175          * The TXQ hardware requires that the descriptor count is a power
 176          * of 2, but tx_desc_lim cannot properly describe that constraint
 177          */
 178         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 179
 180         if (sap->dp_rx->get_dev_info != NULL)
 181                 sap->dp_rx->get_dev_info(dev_info);
 182         if (sap->dp_tx->get_dev_info != NULL)
 183                 sap->dp_tx->get_dev_info(dev_info);
 184
 185         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 186                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 187
 188         if (mae->status == SFC_MAE_STATUS_SUPPORTED) {
 189                 dev_info->switch_info.name = dev->device->driver->name;
 190                 dev_info->switch_info.domain_id = mae->switch_domain_id;
 191                 dev_info->switch_info.port_id = mae->switch_port_id;
 192         }
 193
 194         return 0;
 195 }
 196
 197 static const uint32_t *
 198 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 199 {
 200         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 201
 202         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 203 }
 204
 205 static int
 206 sfc_dev_configure(struct rte_eth_dev *dev)
 207 {
 208         struct rte_eth_dev_data *dev_data = dev->data;
 209         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 210         int rc;
 211
 212         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 213                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 214
 215         sfc_adapter_lock(sa);
 216         switch (sa->state) {
 217         case SFC_ETHDEV_CONFIGURED:
 218                 /* FALLTHROUGH */
 219         case SFC_ETHDEV_INITIALIZED:
 220                 rc = sfc_configure(sa);
 221                 break;
 222         default:
 223                 sfc_err(sa, "unexpected adapter state %u to configure",
 224                         sa->state);
 225                 rc = EINVAL;
 226                 break;
 227         }
 228         sfc_adapter_unlock(sa);
 229
 230         sfc_log_init(sa, "done %d", rc);
 231         SFC_ASSERT(rc >= 0);
 232         return -rc;
 233 }
 234
 235 static int
 236 sfc_dev_start(struct rte_eth_dev *dev)
 237 {
 238         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 239         int rc;
 240
 241         sfc_log_init(sa, "entry");
 242
 243         sfc_adapter_lock(sa);
 244         rc = sfc_start(sa);
 245         sfc_adapter_unlock(sa);
 246
 247         sfc_log_init(sa, "done %d", rc);
 248         SFC_ASSERT(rc >= 0);
 249         return -rc;
 250 }
 251
 252 static int
 253 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 254 {
 255         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 256         struct rte_eth_link current_link;
 257         int ret;
 258
 259         sfc_log_init(sa, "entry");
 260
 261         if (sa->state != SFC_ETHDEV_STARTED) {
 262                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 263         } else if (wait_to_complete) {
 264                 efx_link_mode_t link_mode;
 265
 266                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 267                         link_mode = EFX_LINK_UNKNOWN;
 268                 sfc_port_link_mode_to_info(link_mode, &current_link);
 269
 270         } else {
 271                 sfc_ev_mgmt_qpoll(sa);
 272                 rte_eth_linkstatus_get(dev, &current_link);
 273         }
 274
 275         ret = rte_eth_linkstatus_set(dev, &current_link);
 276         if (ret == 0)
 277                 sfc_notice(sa, "Link status is %s",
 278                            current_link.link_status ? "UP" : "DOWN");
 279
 280         return ret;
 281 }
 282
 283 static int
 284 sfc_dev_stop(struct rte_eth_dev *dev)
 285 {
 286         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 287
 288         sfc_log_init(sa, "entry");
 289
 290         sfc_adapter_lock(sa);
 291         sfc_stop(sa);
 292         sfc_adapter_unlock(sa);
 293
 294         sfc_log_init(sa, "done");
 295
 296         return 0;
 297 }
 298
 299 static int
 300 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 301 {
 302         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 303         int rc;
 304
 305         sfc_log_init(sa, "entry");
 306
 307         sfc_adapter_lock(sa);
 308         rc = sfc_start(sa);
 309         sfc_adapter_unlock(sa);
 310
 311         SFC_ASSERT(rc >= 0);
 312         return -rc;
 313 }
 314
 315 static int
 316 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 317 {
 318         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 319
 320         sfc_log_init(sa, "entry");
 321
 322         sfc_adapter_lock(sa);
 323         sfc_stop(sa);
 324         sfc_adapter_unlock(sa);
 325
 326         return 0;
 327 }
 328
 329 static void
 330 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 331 {
 332         free(dev->process_private);
 333         rte_eth_dev_release_port(dev);
 334 }
 335
 336 static int
 337 sfc_dev_close(struct rte_eth_dev *dev)
 338 {
 339         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 340
 341         sfc_log_init(sa, "entry");
 342
 343         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 344                 sfc_eth_dev_secondary_clear_ops(dev);
 345                 return 0;
 346         }
 347
 348         sfc_pre_detach(sa);
 349
 350         sfc_adapter_lock(sa);
 351         switch (sa->state) {
 352         case SFC_ETHDEV_STARTED:
 353                 sfc_stop(sa);
 354                 SFC_ASSERT(sa->state == SFC_ETHDEV_CONFIGURED);
 355                 /* FALLTHROUGH */
 356         case SFC_ETHDEV_CONFIGURED:
 357                 sfc_close(sa);
 358                 SFC_ASSERT(sa->state == SFC_ETHDEV_INITIALIZED);
 359                 /* FALLTHROUGH */
 360         case SFC_ETHDEV_INITIALIZED:
 361                 break;
 362         default:
 363                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 364                 break;
 365         }
 366
 367         /*
 368          * Cleanup all resources.
 369          * Rollback primary process sfc_eth_dev_init() below.
 370          */
 371
 372         sfc_eth_dev_clear_ops(dev);
 373
 374         sfc_detach(sa);
 375         sfc_unprobe(sa);
 376
 377         sfc_kvargs_cleanup(sa);
 378
 379         sfc_adapter_unlock(sa);
 380         sfc_adapter_lock_fini(sa);
 381
 382         sfc_log_init(sa, "done");
 383
 384         /* Required for logging, so cleanup last */
 385         sa->eth_dev = NULL;
 386
 387         free(sa);
 388
 389         return 0;
 390 }
 391
 392 static int
 393 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 394                    boolean_t enabled)
 395 {
 396         struct sfc_port *port;
 397         boolean_t *toggle;
 398         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 399         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 400         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 401         int rc = 0;
 402
 403         sfc_adapter_lock(sa);
 404
 405         port = &sa->port;
 406         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 407
 408         if (*toggle != enabled) {
 409                 *toggle = enabled;
 410
 411                 if (sfc_sa2shared(sa)->isolated) {
 412                         sfc_warn(sa, "isolated mode is active on the port");
 413                         sfc_warn(sa, "the change is to be applied on the next "
 414                                      "start provided that isolated mode is "
 415                                      "disabled prior the next start");
 416                 } else if ((sa->state == SFC_ETHDEV_STARTED) &&
 417                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 418                         *toggle = !(enabled);
 419                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 420                                  ((enabled) ? "enable" : "disable"), desc, rc);
 421
 422                         /*
 423                          * For promiscuous and all-multicast filters a
 424                          * permission failure should be reported as an
 425                          * unsupported filter.
 426                          */
 427                         if (rc == EPERM)
 428                                 rc = ENOTSUP;
 429                 }
 430         }
 431
 432         sfc_adapter_unlock(sa);
 433         return rc;
 434 }
 435
 436 static int
 437 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 438 {
 439         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 440
 441         SFC_ASSERT(rc >= 0);
 442         return -rc;
 443 }
 444
 445 static int
 446 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 447 {
 448         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 449
 450         SFC_ASSERT(rc >= 0);
 451         return -rc;
 452 }
 453
 454 static int
 455 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 456 {
 457         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 458
 459         SFC_ASSERT(rc >= 0);
 460         return -rc;
 461 }
 462
 463 static int
 464 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 465 {
 466         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 467
 468         SFC_ASSERT(rc >= 0);
 469         return -rc;
 470 }
 471
 472 static int
 473 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 474                    uint16_t nb_rx_desc, unsigned int socket_id,
 475                    const struct rte_eth_rxconf *rx_conf,
 476                    struct rte_mempool *mb_pool)
 477 {
 478         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 479         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 480         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
 481         struct sfc_rxq_info *rxq_info;
 482         sfc_sw_index_t sw_index;
 483         int rc;
 484
 485         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 486                      ethdev_qid, nb_rx_desc, socket_id);
 487
 488         sfc_adapter_lock(sa);
 489
 490         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
 491         rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id,
 492                           rx_conf, mb_pool);
 493         if (rc != 0)
 494                 goto fail_rx_qinit;
 495
 496         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
 497         dev->data->rx_queues[ethdev_qid] = rxq_info->dp;
 498
 499         sfc_adapter_unlock(sa);
 500
 501         return 0;
 502
 503 fail_rx_qinit:
 504         sfc_adapter_unlock(sa);
 505         SFC_ASSERT(rc > 0);
 506         return -rc;
 507 }
 508
 509 static void
 510 sfc_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 511 {
 512         struct sfc_dp_rxq *dp_rxq = dev->data->rx_queues[qid];
 513         struct sfc_rxq *rxq;
 514         struct sfc_adapter *sa;
 515         sfc_sw_index_t sw_index;
 516
 517         if (dp_rxq == NULL)
 518                 return;
 519
 520         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 521         sa = rxq->evq->sa;
 522         sfc_adapter_lock(sa);
 523
 524         sw_index = dp_rxq->dpq.queue_id;
 525
 526         sfc_log_init(sa, "RxQ=%u", sw_index);
 527
 528         sfc_rx_qfini(sa, sw_index);
 529
 530         sfc_adapter_unlock(sa);
 531 }
 532
 533 static int
 534 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid,
 535                    uint16_t nb_tx_desc, unsigned int socket_id,
 536                    const struct rte_eth_txconf *tx_conf)
 537 {
 538         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 539         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 540         struct sfc_txq_info *txq_info;
 541         sfc_sw_index_t sw_index;
 542         int rc;
 543
 544         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 545                      ethdev_qid, nb_tx_desc, socket_id);
 546
 547         sfc_adapter_lock(sa);
 548
 549         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
 550         rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf);
 551         if (rc != 0)
 552                 goto fail_tx_qinit;
 553
 554         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
 555         dev->data->tx_queues[ethdev_qid] = txq_info->dp;
 556
 557         sfc_adapter_unlock(sa);
 558         return 0;
 559
 560 fail_tx_qinit:
 561         sfc_adapter_unlock(sa);
 562         SFC_ASSERT(rc > 0);
 563         return -rc;
 564 }
 565
 566 static void
 567 sfc_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid)
 568 {
 569         struct sfc_dp_txq *dp_txq = dev->data->tx_queues[qid];
 570         struct sfc_txq *txq;
 571         sfc_sw_index_t sw_index;
 572         struct sfc_adapter *sa;
 573
 574         if (dp_txq == NULL)
 575                 return;
 576
 577         txq = sfc_txq_by_dp_txq(dp_txq);
 578         sw_index = dp_txq->dpq.queue_id;
 579
 580         SFC_ASSERT(txq->evq != NULL);
 581         sa = txq->evq->sa;
 582
 583         sfc_log_init(sa, "TxQ = %u", sw_index);
 584
 585         sfc_adapter_lock(sa);
 586
 587         sfc_tx_qfini(sa, sw_index);
 588
 589         sfc_adapter_unlock(sa);
 590 }
 591
 592 static void
 593 sfc_stats_get_dp_rx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 594 {
 595         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 596         uint64_t pkts_sum = 0;
 597         uint64_t bytes_sum = 0;
 598         unsigned int i;
 599
 600         for (i = 0; i < sas->ethdev_rxq_count; ++i) {
 601                 struct sfc_rxq_info *rxq_info;
 602
 603                 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, i);
 604                 if (rxq_info->state & SFC_RXQ_INITIALIZED) {
 605                         union sfc_pkts_bytes qstats;
 606
 607                         sfc_pkts_bytes_get(&rxq_info->dp->dpq.stats, &qstats);
 608                         pkts_sum += qstats.pkts -
 609                                         sa->sw_stats.reset_rx_pkts[i];
 610                         bytes_sum += qstats.bytes -
 611                                         sa->sw_stats.reset_rx_bytes[i];
 612                 }
 613         }
 614
 615         *pkts = pkts_sum;
 616         *bytes = bytes_sum;
 617 }
 618
 619 static void
 620 sfc_stats_get_dp_tx(struct sfc_adapter *sa, uint64_t *pkts, uint64_t *bytes)
 621 {
 622         struct sfc_adapter_shared *sas = sfc_sa2shared(sa);
 623         uint64_t pkts_sum = 0;
 624         uint64_t bytes_sum = 0;
 625         unsigned int i;
 626
 627         for (i = 0; i < sas->ethdev_txq_count; ++i) {
 628                 struct sfc_txq_info *txq_info;
 629
 630                 txq_info = sfc_txq_info_by_ethdev_qid(sas, i);
 631                 if (txq_info->state & SFC_TXQ_INITIALIZED) {
 632                         union sfc_pkts_bytes qstats;
 633
 634                         sfc_pkts_bytes_get(&txq_info->dp->dpq.stats, &qstats);
 635                         pkts_sum += qstats.pkts -
 636                                         sa->sw_stats.reset_tx_pkts[i];
 637                         bytes_sum += qstats.bytes -
 638                                         sa->sw_stats.reset_tx_bytes[i];
 639                 }
 640         }
 641
 642         *pkts = pkts_sum;
 643         *bytes = bytes_sum;
 644 }
 645
 646 /*
 647  * Some statistics are computed as A - B where A and B each increase
 648  * monotonically with some hardware counter(s) and the counters are read
 649  * asynchronously.
 650  *
 651  * If packet X is counted in A, but not counted in B yet, computed value is
 652  * greater than real.
 653  *
 654  * If packet X is not counted in A at the moment of reading the counter,
 655  * but counted in B at the moment of reading the counter, computed value
 656  * is less than real.
 657  *
 658  * However, counter which grows backward is worse evil than slightly wrong
 659  * value. So, let's try to guarantee that it never happens except may be
 660  * the case when the MAC stats are zeroed as a result of a NIC reset.
 661  */
 662 static void
 663 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 664 {
 665         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 666                 *stat = newval;
 667 }
 668
 669 static int
 670 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 671 {
 672         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 673         bool have_dp_rx_stats = sap->dp_rx->features & SFC_DP_RX_FEAT_STATS;
 674         bool have_dp_tx_stats = sap->dp_tx->features & SFC_DP_TX_FEAT_STATS;
 675         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 676         struct sfc_port *port = &sa->port;
 677         uint64_t *mac_stats;
 678         int ret;
 679
 680         sfc_adapter_lock(sa);
 681
 682         if (have_dp_rx_stats)
 683                 sfc_stats_get_dp_rx(sa, &stats->ipackets, &stats->ibytes);
 684         if (have_dp_tx_stats)
 685                 sfc_stats_get_dp_tx(sa, &stats->opackets, &stats->obytes);
 686
 687         ret = sfc_port_update_mac_stats(sa, B_FALSE);
 688         if (ret != 0)
 689                 goto unlock;
 690
 691         mac_stats = port->mac_stats_buf;
 692
 693         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 694                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 695                 if (!have_dp_rx_stats) {
 696                         stats->ipackets =
 697                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 698                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 699                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 700                         stats->ibytes =
 701                                 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 702                                 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 703                                 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 704
 705                         /* CRC is included in these stats, but shouldn't be */
 706                         stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
 707                 }
 708                 if (!have_dp_tx_stats) {
 709                         stats->opackets =
 710                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 711                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 712                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 713                         stats->obytes =
 714                                 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 715                                 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 716                                 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 717
 718                         /* CRC is included in these stats, but shouldn't be */
 719                         stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN;
 720                 }
 721                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 722                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 723         } else {
 724                 if (!have_dp_tx_stats) {
 725                         stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 726                         stats->obytes = mac_stats[EFX_MAC_TX_OCTETS] -
 727                                 mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN;
 728                 }
 729
 730                 /*
 731                  * Take into account stats which are whenever supported
 732                  * on EF10. If some stat is not supported by current
 733                  * firmware variant or HW revision, it is guaranteed
 734                  * to be zero in mac_stats.
 735                  */
 736                 stats->imissed =
 737                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 738                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 739                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 740                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 741                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 742                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 743                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 744                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 745                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 746                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 747                 stats->ierrors =
 748                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 749                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 750                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 751                 /* no oerrors counters supported on EF10 */
 752
 753                 if (!have_dp_rx_stats) {
 754                         /* Exclude missed, errors and pauses from Rx packets */
 755                         sfc_update_diff_stat(&port->ipackets,
 756                                 mac_stats[EFX_MAC_RX_PKTS] -
 757                                 mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 758                                 stats->imissed - stats->ierrors);
 759                         stats->ipackets = port->ipackets;
 760                         stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS] -
 761                                 mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN;
 762                 }
 763         }
 764
 765 unlock:
 766         sfc_adapter_unlock(sa);
 767         SFC_ASSERT(ret >= 0);
 768         return -ret;
 769 }
 770
 771 static int
 772 sfc_stats_reset(struct rte_eth_dev *dev)
 773 {
 774         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 775         struct sfc_port *port = &sa->port;
 776         int rc;
 777
 778         sfc_adapter_lock(sa);
 779
 780         if (sa->state != SFC_ETHDEV_STARTED) {
 781                 /*
 782                  * The operation cannot be done if port is not started; it
 783                  * will be scheduled to be done during the next port start
 784                  */
 785                 port->mac_stats_reset_pending = B_TRUE;
 786                 sfc_adapter_unlock(sa);
 787                 return 0;
 788         }
 789
 790         rc = sfc_port_reset_mac_stats(sa);
 791         if (rc != 0)
 792                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 793
 794         sfc_sw_xstats_reset(sa);
 795
 796         sfc_adapter_unlock(sa);
 797
 798         SFC_ASSERT(rc >= 0);
 799         return -rc;
 800 }
 801
 802 static unsigned int
 803 sfc_xstats_get_nb_supported(struct sfc_adapter *sa)
 804 {
 805         struct sfc_port *port = &sa->port;
 806         unsigned int nb_supported;
 807
 808         sfc_adapter_lock(sa);
 809         nb_supported = port->mac_stats_nb_supported +
 810                        sfc_sw_xstats_get_nb_supported(sa);
 811         sfc_adapter_unlock(sa);
 812
 813         return nb_supported;
 814 }
 815
 816 static int
 817 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 818                unsigned int xstats_count)
 819 {
 820         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 821         unsigned int nb_written = 0;
 822         unsigned int nb_supported = 0;
 823         int rc;
 824
 825         if (unlikely(xstats == NULL))
 826                 return sfc_xstats_get_nb_supported(sa);
 827
 828         rc = sfc_port_get_mac_stats(sa, xstats, xstats_count, &nb_written);
 829         if (rc < 0)
 830                 return rc;
 831
 832         nb_supported = rc;
 833         sfc_sw_xstats_get_vals(sa, xstats, xstats_count, &nb_written,
 834                                &nb_supported);
 835
 836         return nb_supported;
 837 }
 838
 839 static int
 840 sfc_xstats_get_names(struct rte_eth_dev *dev,
 841                      struct rte_eth_xstat_name *xstats_names,
 842                      unsigned int xstats_count)
 843 {
 844         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 845         struct sfc_port *port = &sa->port;
 846         unsigned int i;
 847         unsigned int nstats = 0;
 848         unsigned int nb_written = 0;
 849         int ret;
 850
 851         if (unlikely(xstats_names == NULL))
 852                 return sfc_xstats_get_nb_supported(sa);
 853
 854         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 855                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 856                         if (nstats < xstats_count) {
 857                                 strlcpy(xstats_names[nstats].name,
 858                                         efx_mac_stat_name(sa->nic, i),
 859                                         sizeof(xstats_names[0].name));
 860                                 nb_written++;
 861                         }
 862                         nstats++;
 863                 }
 864         }
 865
 866         ret = sfc_sw_xstats_get_names(sa, xstats_names, xstats_count,
 867                                       &nb_written, &nstats);
 868         if (ret != 0) {
 869                 SFC_ASSERT(ret < 0);
 870                 return ret;
 871         }
 872
 873         return nstats;
 874 }
 875
 876 static int
 877 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 878                      uint64_t *values, unsigned int n)
 879 {
 880         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 881         struct sfc_port *port = &sa->port;
 882         unsigned int nb_supported;
 883         unsigned int i;
 884         int rc;
 885
 886         if (unlikely(ids == NULL || values == NULL))
 887                 return -EINVAL;
 888
 889         /*
 890          * Values array could be filled in nonsequential order. Fill values with
 891          * constant indicating invalid ID first.
 892          */
 893         for (i = 0; i < n; i++)
 894                 values[i] = SFC_XSTAT_ID_INVALID_VAL;
 895
 896         rc = sfc_port_get_mac_stats_by_id(sa, ids, values, n);
 897         if (rc != 0)
 898                 return rc;
 899
 900         nb_supported = port->mac_stats_nb_supported;
 901         sfc_sw_xstats_get_vals_by_id(sa, ids, values, n, &nb_supported);
 902
 903         /* Return number of written stats before invalid ID is encountered. */
 904         for (i = 0; i < n; i++) {
 905                 if (values[i] == SFC_XSTAT_ID_INVALID_VAL)
 906                         return i;
 907         }
 908
 909         return n;
 910 }
 911
 912 static int
 913 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 914                            const uint64_t *ids,
 915                            struct rte_eth_xstat_name *xstats_names,
 916                            unsigned int size)
 917 {
 918         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 919         struct sfc_port *port = &sa->port;
 920         unsigned int nb_supported;
 921         unsigned int i;
 922         int ret;
 923
 924         if (unlikely(xstats_names == NULL && ids != NULL) ||
 925             unlikely(xstats_names != NULL && ids == NULL))
 926                 return -EINVAL;
 927
 928         if (unlikely(xstats_names == NULL && ids == NULL))
 929                 return sfc_xstats_get_nb_supported(sa);
 930
 931         /*
 932          * Names array could be filled in nonsequential order. Fill names with
 933          * string indicating invalid ID first.
 934          */
 935         for (i = 0; i < size; i++)
 936                 xstats_names[i].name[0] = SFC_XSTAT_ID_INVALID_NAME;
 937
 938         sfc_adapter_lock(sa);
 939
 940         SFC_ASSERT(port->mac_stats_nb_supported <=
 941                    RTE_DIM(port->mac_stats_by_id));
 942
 943         for (i = 0; i < size; i++) {
 944                 if (ids[i] < port->mac_stats_nb_supported) {
 945                         strlcpy(xstats_names[i].name,
 946                                 efx_mac_stat_name(sa->nic,
 947                                                  port->mac_stats_by_id[ids[i]]),
 948                                 sizeof(xstats_names[0].name));
 949                 }
 950         }
 951
 952         nb_supported = port->mac_stats_nb_supported;
 953
 954         sfc_adapter_unlock(sa);
 955
 956         ret = sfc_sw_xstats_get_names_by_id(sa, ids, xstats_names, size,
 957                                             &nb_supported);
 958         if (ret != 0) {
 959                 SFC_ASSERT(ret < 0);
 960                 return ret;
 961         }
 962
 963         /* Return number of written names before invalid ID is encountered. */
 964         for (i = 0; i < size; i++) {
 965                 if (xstats_names[i].name[0] == SFC_XSTAT_ID_INVALID_NAME)
 966                         return i;
 967         }
 968
 969         return size;
 970 }
 971
 972 static int
 973 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 974 {
 975         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 976         unsigned int wanted_fc, link_fc;
 977
 978         memset(fc_conf, 0, sizeof(*fc_conf));
 979
 980         sfc_adapter_lock(sa);
 981
 982         if (sa->state == SFC_ETHDEV_STARTED)
 983                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 984         else
 985                 link_fc = sa->port.flow_ctrl;
 986
 987         switch (link_fc) {
 988         case 0:
 989                 fc_conf->mode = RTE_FC_NONE;
 990                 break;
 991         case EFX_FCNTL_RESPOND:
 992                 fc_conf->mode = RTE_FC_RX_PAUSE;
 993                 break;
 994         case EFX_FCNTL_GENERATE:
 995                 fc_conf->mode = RTE_FC_TX_PAUSE;
 996                 break;
 997         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 998                 fc_conf->mode = RTE_FC_FULL;
 999                 break;
1000         default:
1001                 sfc_err(sa, "%s: unexpected flow control value %#x",
1002                         __func__, link_fc);
1003         }
1004
1005         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
1006
1007         sfc_adapter_unlock(sa);
1008
1009         return 0;
1010 }
1011
1012 static int
1013 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
1014 {
1015         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1016         struct sfc_port *port = &sa->port;
1017         unsigned int fcntl;
1018         int rc;
1019
1020         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
1021             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
1022             fc_conf->mac_ctrl_frame_fwd != 0) {
1023                 sfc_err(sa, "unsupported flow control settings specified");
1024                 rc = EINVAL;
1025                 goto fail_inval;
1026         }
1027
1028         switch (fc_conf->mode) {
1029         case RTE_FC_NONE:
1030                 fcntl = 0;
1031                 break;
1032         case RTE_FC_RX_PAUSE:
1033                 fcntl = EFX_FCNTL_RESPOND;
1034                 break;
1035         case RTE_FC_TX_PAUSE:
1036                 fcntl = EFX_FCNTL_GENERATE;
1037                 break;
1038         case RTE_FC_FULL:
1039                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
1040                 break;
1041         default:
1042                 rc = EINVAL;
1043                 goto fail_inval;
1044         }
1045
1046         sfc_adapter_lock(sa);
1047
1048         if (sa->state == SFC_ETHDEV_STARTED) {
1049                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
1050                 if (rc != 0)
1051                         goto fail_mac_fcntl_set;
1052         }
1053
1054         port->flow_ctrl = fcntl;
1055         port->flow_ctrl_autoneg = fc_conf->autoneg;
1056
1057         sfc_adapter_unlock(sa);
1058
1059         return 0;
1060
1061 fail_mac_fcntl_set:
1062         sfc_adapter_unlock(sa);
1063 fail_inval:
1064         SFC_ASSERT(rc > 0);
1065         return -rc;
1066 }
1067
1068 static int
1069 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
1070 {
1071         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
1072         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1073         boolean_t scatter_enabled;
1074         const char *error;
1075         unsigned int i;
1076
1077         for (i = 0; i < sas->rxq_count; i++) {
1078                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
1079                         continue;
1080
1081                 scatter_enabled = (sas->rxq_info[i].type_flags &
1082                                    EFX_RXQ_FLAG_SCATTER);
1083
1084                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
1085                                           encp->enc_rx_prefix_size,
1086                                           scatter_enabled,
1087                                           encp->enc_rx_scatter_max, &error)) {
1088                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
1089                                 error);
1090                         return EINVAL;
1091                 }
1092         }
1093
1094         return 0;
1095 }
1096
1097 static int
1098 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
1099 {
1100         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1101         size_t pdu = EFX_MAC_PDU(mtu);
1102         size_t old_pdu;
1103         int rc;
1104
1105         sfc_log_init(sa, "mtu=%u", mtu);
1106
1107         rc = EINVAL;
1108         if (pdu < EFX_MAC_PDU_MIN) {
1109                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
1110                         (unsigned int)mtu, (unsigned int)pdu,
1111                         EFX_MAC_PDU_MIN);
1112                 goto fail_inval;
1113         }
1114         if (pdu > EFX_MAC_PDU_MAX) {
1115                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
1116                         (unsigned int)mtu, (unsigned int)pdu,
1117                         (unsigned int)EFX_MAC_PDU_MAX);
1118                 goto fail_inval;
1119         }
1120
1121         sfc_adapter_lock(sa);
1122
1123         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
1124         if (rc != 0)
1125                 goto fail_check_scatter;
1126
1127         if (pdu != sa->port.pdu) {
1128                 if (sa->state == SFC_ETHDEV_STARTED) {
1129                         sfc_stop(sa);
1130
1131                         old_pdu = sa->port.pdu;
1132                         sa->port.pdu = pdu;
1133                         rc = sfc_start(sa);
1134                         if (rc != 0)
1135                                 goto fail_start;
1136                 } else {
1137                         sa->port.pdu = pdu;
1138                 }
1139         }
1140
1141         /*
1142          * The driver does not use it, but other PMDs update jumbo frame
1143          * flag and max_rx_pkt_len when MTU is set.
1144          */
1145         if (mtu > RTE_ETHER_MTU) {
1146                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1147                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1148         }
1149
1150         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1151
1152         sfc_adapter_unlock(sa);
1153
1154         sfc_log_init(sa, "done");
1155         return 0;
1156
1157 fail_start:
1158         sa->port.pdu = old_pdu;
1159         if (sfc_start(sa) != 0)
1160                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1161                         "PDU max size - port is stopped",
1162                         (unsigned int)pdu, (unsigned int)old_pdu);
1163
1164 fail_check_scatter:
1165         sfc_adapter_unlock(sa);
1166
1167 fail_inval:
1168         sfc_log_init(sa, "failed %d", rc);
1169         SFC_ASSERT(rc > 0);
1170         return -rc;
1171 }
1172 static int
1173 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1174 {
1175         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1176         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1177         struct sfc_port *port = &sa->port;
1178         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1179         int rc = 0;
1180
1181         sfc_adapter_lock(sa);
1182
1183         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1184                 goto unlock;
1185
1186         /*
1187          * Copy the address to the device private data so that
1188          * it could be recalled in the case of adapter restart.
1189          */
1190         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1191
1192         /*
1193          * Neither of the two following checks can return
1194          * an error. The new MAC address is preserved in
1195          * the device private data and can be activated
1196          * on the next port start if the user prevents
1197          * isolated mode from being enabled.
1198          */
1199         if (sfc_sa2shared(sa)->isolated) {
1200                 sfc_warn(sa, "isolated mode is active on the port");
1201                 sfc_warn(sa, "will not set MAC address");
1202                 goto unlock;
1203         }
1204
1205         if (sa->state != SFC_ETHDEV_STARTED) {
1206                 sfc_notice(sa, "the port is not started");
1207                 sfc_notice(sa, "the new MAC address will be set on port start");
1208
1209                 goto unlock;
1210         }
1211
1212         if (encp->enc_allow_set_mac_with_installed_filters) {
1213                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1214                 if (rc != 0) {
1215                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1216                         goto unlock;
1217                 }
1218
1219                 /*
1220                  * Changing the MAC address by means of MCDI request
1221                  * has no effect on received traffic, therefore
1222                  * we also need to update unicast filters
1223                  */
1224                 rc = sfc_set_rx_mode_unchecked(sa);
1225                 if (rc != 0) {
1226                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1227                         /* Rollback the old address */
1228                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1229                         (void)sfc_set_rx_mode_unchecked(sa);
1230                 }
1231         } else {
1232                 sfc_warn(sa, "cannot set MAC address with filters installed");
1233                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1234                 sfc_warn(sa, "(some traffic may be dropped)");
1235
1236                 /*
1237                  * Since setting MAC address with filters installed is not
1238                  * allowed on the adapter, the new MAC address will be set
1239                  * by means of adapter restart. sfc_start() shall retrieve
1240                  * the new address from the device private data and set it.
1241                  */
1242                 sfc_stop(sa);
1243                 rc = sfc_start(sa);
1244                 if (rc != 0)
1245                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1246         }
1247
1248 unlock:
1249         if (rc != 0)
1250                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1251
1252         sfc_adapter_unlock(sa);
1253
1254         SFC_ASSERT(rc >= 0);
1255         return -rc;
1256 }
1257
1258
1259 static int
1260 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1261                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1262 {
1263         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1264         struct sfc_port *port = &sa->port;
1265         uint8_t *mc_addrs = port->mcast_addrs;
1266         int rc;
1267         unsigned int i;
1268
1269         if (sfc_sa2shared(sa)->isolated) {
1270                 sfc_err(sa, "isolated mode is active on the port");
1271                 sfc_err(sa, "will not set multicast address list");
1272                 return -ENOTSUP;
1273         }
1274
1275         if (mc_addrs == NULL)
1276                 return -ENOBUFS;
1277
1278         if (nb_mc_addr > port->max_mcast_addrs) {
1279                 sfc_err(sa, "too many multicast addresses: %u > %u",
1280                          nb_mc_addr, port->max_mcast_addrs);
1281                 return -EINVAL;
1282         }
1283
1284         for (i = 0; i < nb_mc_addr; ++i) {
1285                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1286                                  EFX_MAC_ADDR_LEN);
1287                 mc_addrs += EFX_MAC_ADDR_LEN;
1288         }
1289
1290         port->nb_mcast_addrs = nb_mc_addr;
1291
1292         if (sa->state != SFC_ETHDEV_STARTED)
1293                 return 0;
1294
1295         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1296                                         port->nb_mcast_addrs);
1297         if (rc != 0)
1298                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1299
1300         SFC_ASSERT(rc >= 0);
1301         return -rc;
1302 }
1303
1304 /*
1305  * The function is used by the secondary process as well. It must not
1306  * use any process-local pointers from the adapter data.
1307  */
1308 static void
1309 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1310                       struct rte_eth_rxq_info *qinfo)
1311 {
1312         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1313         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1314         struct sfc_rxq_info *rxq_info;
1315
1316         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1317
1318         qinfo->mp = rxq_info->refill_mb_pool;
1319         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1320         qinfo->conf.rx_drop_en = 1;
1321         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1322         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1323         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1324                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1325                 qinfo->scattered_rx = 1;
1326         }
1327         qinfo->nb_desc = rxq_info->entries;
1328 }
1329
1330 /*
1331  * The function is used by the secondary process as well. It must not
1332  * use any process-local pointers from the adapter data.
1333  */
1334 static void
1335 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid,
1336                       struct rte_eth_txq_info *qinfo)
1337 {
1338         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1339         struct sfc_txq_info *txq_info;
1340
1341         SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count);
1342
1343         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1344
1345         memset(qinfo, 0, sizeof(*qinfo));
1346
1347         qinfo->conf.offloads = txq_info->offloads;
1348         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1349         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1350         qinfo->nb_desc = txq_info->entries;
1351 }
1352
1353 /*
1354  * The function is used by the secondary process as well. It must not
1355  * use any process-local pointers from the adapter data.
1356  */
1357 static uint32_t
1358 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1359 {
1360         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1361         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1362         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1363         struct sfc_rxq_info *rxq_info;
1364
1365         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1366
1367         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1368                 return 0;
1369
1370         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1371 }
1372
1373 /*
1374  * The function is used by the secondary process as well. It must not
1375  * use any process-local pointers from the adapter data.
1376  */
1377 static int
1378 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1379 {
1380         struct sfc_dp_rxq *dp_rxq = queue;
1381         const struct sfc_dp_rx *dp_rx;
1382
1383         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1384
1385         return dp_rx->qdesc_status(dp_rxq, offset);
1386 }
1387
1388 /*
1389  * The function is used by the secondary process as well. It must not
1390  * use any process-local pointers from the adapter data.
1391  */
1392 static int
1393 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1394 {
1395         struct sfc_dp_txq *dp_txq = queue;
1396         const struct sfc_dp_tx *dp_tx;
1397
1398         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1399
1400         return dp_tx->qdesc_status(dp_txq, offset);
1401 }
1402
1403 static int
1404 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1405 {
1406         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1407         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1408         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1409         struct sfc_rxq_info *rxq_info;
1410         sfc_sw_index_t sw_index;
1411         int rc;
1412
1413         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1414
1415         sfc_adapter_lock(sa);
1416
1417         rc = EINVAL;
1418         if (sa->state != SFC_ETHDEV_STARTED)
1419                 goto fail_not_started;
1420
1421         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1422         if (rxq_info->state != SFC_RXQ_INITIALIZED)
1423                 goto fail_not_setup;
1424
1425         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1426         rc = sfc_rx_qstart(sa, sw_index);
1427         if (rc != 0)
1428                 goto fail_rx_qstart;
1429
1430         rxq_info->deferred_started = B_TRUE;
1431
1432         sfc_adapter_unlock(sa);
1433
1434         return 0;
1435
1436 fail_rx_qstart:
1437 fail_not_setup:
1438 fail_not_started:
1439         sfc_adapter_unlock(sa);
1440         SFC_ASSERT(rc > 0);
1441         return -rc;
1442 }
1443
1444 static int
1445 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1446 {
1447         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1448         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1449         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1450         struct sfc_rxq_info *rxq_info;
1451         sfc_sw_index_t sw_index;
1452
1453         sfc_log_init(sa, "RxQ=%u", ethdev_qid);
1454
1455         sfc_adapter_lock(sa);
1456
1457         sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid);
1458         sfc_rx_qstop(sa, sw_index);
1459
1460         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1461         rxq_info->deferred_started = B_FALSE;
1462
1463         sfc_adapter_unlock(sa);
1464
1465         return 0;
1466 }
1467
1468 static int
1469 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1470 {
1471         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1472         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1473         struct sfc_txq_info *txq_info;
1474         sfc_sw_index_t sw_index;
1475         int rc;
1476
1477         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1478
1479         sfc_adapter_lock(sa);
1480
1481         rc = EINVAL;
1482         if (sa->state != SFC_ETHDEV_STARTED)
1483                 goto fail_not_started;
1484
1485         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1486         if (txq_info->state != SFC_TXQ_INITIALIZED)
1487                 goto fail_not_setup;
1488
1489         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1490         rc = sfc_tx_qstart(sa, sw_index);
1491         if (rc != 0)
1492                 goto fail_tx_qstart;
1493
1494         txq_info->deferred_started = B_TRUE;
1495
1496         sfc_adapter_unlock(sa);
1497         return 0;
1498
1499 fail_tx_qstart:
1500
1501 fail_not_setup:
1502 fail_not_started:
1503         sfc_adapter_unlock(sa);
1504         SFC_ASSERT(rc > 0);
1505         return -rc;
1506 }
1507
1508 static int
1509 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1510 {
1511         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1512         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1513         struct sfc_txq_info *txq_info;
1514         sfc_sw_index_t sw_index;
1515
1516         sfc_log_init(sa, "TxQ = %u", ethdev_qid);
1517
1518         sfc_adapter_lock(sa);
1519
1520         sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid);
1521         sfc_tx_qstop(sa, sw_index);
1522
1523         txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid);
1524         txq_info->deferred_started = B_FALSE;
1525
1526         sfc_adapter_unlock(sa);
1527         return 0;
1528 }
1529
1530 static efx_tunnel_protocol_t
1531 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1532 {
1533         switch (rte_type) {
1534         case RTE_TUNNEL_TYPE_VXLAN:
1535                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1536         case RTE_TUNNEL_TYPE_GENEVE:
1537                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1538         default:
1539                 return EFX_TUNNEL_NPROTOS;
1540         }
1541 }
1542
1543 enum sfc_udp_tunnel_op_e {
1544         SFC_UDP_TUNNEL_ADD_PORT,
1545         SFC_UDP_TUNNEL_DEL_PORT,
1546 };
1547
1548 static int
1549 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1550                       struct rte_eth_udp_tunnel *tunnel_udp,
1551                       enum sfc_udp_tunnel_op_e op)
1552 {
1553         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1554         efx_tunnel_protocol_t tunnel_proto;
1555         int rc;
1556
1557         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1558                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1559                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1560                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1561
1562         tunnel_proto =
1563                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1564         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1565                 rc = ENOTSUP;
1566                 goto fail_bad_proto;
1567         }
1568
1569         sfc_adapter_lock(sa);
1570
1571         switch (op) {
1572         case SFC_UDP_TUNNEL_ADD_PORT:
1573                 rc = efx_tunnel_config_udp_add(sa->nic,
1574                                                tunnel_udp->udp_port,
1575                                                tunnel_proto);
1576                 break;
1577         case SFC_UDP_TUNNEL_DEL_PORT:
1578                 rc = efx_tunnel_config_udp_remove(sa->nic,
1579                                                   tunnel_udp->udp_port,
1580                                                   tunnel_proto);
1581                 break;
1582         default:
1583                 rc = EINVAL;
1584                 goto fail_bad_op;
1585         }
1586
1587         if (rc != 0)
1588                 goto fail_op;
1589
1590         if (sa->state == SFC_ETHDEV_STARTED) {
1591                 rc = efx_tunnel_reconfigure(sa->nic);
1592                 if (rc == EAGAIN) {
1593                         /*
1594                          * Configuration is accepted by FW and MC reboot
1595                          * is initiated to apply the changes. MC reboot
1596                          * will be handled in a usual way (MC reboot
1597                          * event on management event queue and adapter
1598                          * restart).
1599                          */
1600                         rc = 0;
1601                 } else if (rc != 0) {
1602                         goto fail_reconfigure;
1603                 }
1604         }
1605
1606         sfc_adapter_unlock(sa);
1607         return 0;
1608
1609 fail_reconfigure:
1610         /* Remove/restore entry since the change makes the trouble */
1611         switch (op) {
1612         case SFC_UDP_TUNNEL_ADD_PORT:
1613                 (void)efx_tunnel_config_udp_remove(sa->nic,
1614                                                    tunnel_udp->udp_port,
1615                                                    tunnel_proto);
1616                 break;
1617         case SFC_UDP_TUNNEL_DEL_PORT:
1618                 (void)efx_tunnel_config_udp_add(sa->nic,
1619                                                 tunnel_udp->udp_port,
1620                                                 tunnel_proto);
1621                 break;
1622         }
1623
1624 fail_op:
1625 fail_bad_op:
1626         sfc_adapter_unlock(sa);
1627
1628 fail_bad_proto:
1629         SFC_ASSERT(rc > 0);
1630         return -rc;
1631 }
1632
1633 static int
1634 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1635                             struct rte_eth_udp_tunnel *tunnel_udp)
1636 {
1637         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1638 }
1639
1640 static int
1641 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1642                             struct rte_eth_udp_tunnel *tunnel_udp)
1643 {
1644         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
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_hash_conf_get(struct rte_eth_dev *dev,
1653                           struct rte_eth_rss_conf *rss_conf)
1654 {
1655         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1656         struct sfc_rss *rss = &sas->rss;
1657
1658         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1659                 return -ENOTSUP;
1660
1661         /*
1662          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1663          * hence, conversion is done here to derive a correct set of ETH_RSS
1664          * flags which corresponds to the active EFX configuration stored
1665          * locally in 'sfc_adapter' and kept up-to-date
1666          */
1667         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1668         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1669         if (rss_conf->rss_key != NULL)
1670                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1671
1672         return 0;
1673 }
1674
1675 static int
1676 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1677                         struct rte_eth_rss_conf *rss_conf)
1678 {
1679         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1680         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1681         unsigned int efx_hash_types;
1682         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1683         unsigned int n_contexts;
1684         unsigned int mode_i = 0;
1685         unsigned int key_i = 0;
1686         unsigned int i = 0;
1687         int rc = 0;
1688
1689         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1690
1691         if (sfc_sa2shared(sa)->isolated)
1692                 return -ENOTSUP;
1693
1694         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1695                 sfc_err(sa, "RSS is not available");
1696                 return -ENOTSUP;
1697         }
1698
1699         if (rss->channels == 0) {
1700                 sfc_err(sa, "RSS is not configured");
1701                 return -EINVAL;
1702         }
1703
1704         if ((rss_conf->rss_key != NULL) &&
1705             (rss_conf->rss_key_len != sizeof(rss->key))) {
1706                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1707                         sizeof(rss->key));
1708                 return -EINVAL;
1709         }
1710
1711         sfc_adapter_lock(sa);
1712
1713         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1714         if (rc != 0)
1715                 goto fail_rx_hf_rte_to_efx;
1716
1717         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1718                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1719                                            rss->hash_alg, efx_hash_types,
1720                                            B_TRUE);
1721                 if (rc != 0)
1722                         goto fail_scale_mode_set;
1723         }
1724
1725         if (rss_conf->rss_key != NULL) {
1726                 if (sa->state == SFC_ETHDEV_STARTED) {
1727                         for (key_i = 0; key_i < n_contexts; key_i++) {
1728                                 rc = efx_rx_scale_key_set(sa->nic,
1729                                                           contexts[key_i],
1730                                                           rss_conf->rss_key,
1731                                                           sizeof(rss->key));
1732                                 if (rc != 0)
1733                                         goto fail_scale_key_set;
1734                         }
1735                 }
1736
1737                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1738         }
1739
1740         rss->hash_types = efx_hash_types;
1741
1742         sfc_adapter_unlock(sa);
1743
1744         return 0;
1745
1746 fail_scale_key_set:
1747         for (i = 0; i < key_i; i++) {
1748                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1749                                          sizeof(rss->key)) != 0)
1750                         sfc_err(sa, "failed to restore RSS key");
1751         }
1752
1753 fail_scale_mode_set:
1754         for (i = 0; i < mode_i; i++) {
1755                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1756                                           EFX_RX_HASHALG_TOEPLITZ,
1757                                           rss->hash_types, B_TRUE) != 0)
1758                         sfc_err(sa, "failed to restore RSS mode");
1759         }
1760
1761 fail_rx_hf_rte_to_efx:
1762         sfc_adapter_unlock(sa);
1763         return -rc;
1764 }
1765
1766 /*
1767  * The function is used by the secondary process as well. It must not
1768  * use any process-local pointers from the adapter data.
1769  */
1770 static int
1771 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1772                        struct rte_eth_rss_reta_entry64 *reta_conf,
1773                        uint16_t reta_size)
1774 {
1775         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1776         struct sfc_rss *rss = &sas->rss;
1777         int entry;
1778
1779         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1780                 return -ENOTSUP;
1781
1782         if (rss->channels == 0)
1783                 return -EINVAL;
1784
1785         if (reta_size != EFX_RSS_TBL_SIZE)
1786                 return -EINVAL;
1787
1788         for (entry = 0; entry < reta_size; entry++) {
1789                 int grp = entry / RTE_RETA_GROUP_SIZE;
1790                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1791
1792                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1793                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1794         }
1795
1796         return 0;
1797 }
1798
1799 static int
1800 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1801                         struct rte_eth_rss_reta_entry64 *reta_conf,
1802                         uint16_t reta_size)
1803 {
1804         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1805         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1806         unsigned int *rss_tbl_new;
1807         uint16_t entry;
1808         int rc = 0;
1809
1810
1811         if (sfc_sa2shared(sa)->isolated)
1812                 return -ENOTSUP;
1813
1814         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1815                 sfc_err(sa, "RSS is not available");
1816                 return -ENOTSUP;
1817         }
1818
1819         if (rss->channels == 0) {
1820                 sfc_err(sa, "RSS is not configured");
1821                 return -EINVAL;
1822         }
1823
1824         if (reta_size != EFX_RSS_TBL_SIZE) {
1825                 sfc_err(sa, "RETA size is wrong (should be %u)",
1826                         EFX_RSS_TBL_SIZE);
1827                 return -EINVAL;
1828         }
1829
1830         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1831         if (rss_tbl_new == NULL)
1832                 return -ENOMEM;
1833
1834         sfc_adapter_lock(sa);
1835
1836         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1837
1838         for (entry = 0; entry < reta_size; entry++) {
1839                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1840                 struct rte_eth_rss_reta_entry64 *grp;
1841
1842                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1843
1844                 if (grp->mask & (1ull << grp_idx)) {
1845                         if (grp->reta[grp_idx] >= rss->channels) {
1846                                 rc = EINVAL;
1847                                 goto bad_reta_entry;
1848                         }
1849                         rss_tbl_new[entry] = grp->reta[grp_idx];
1850                 }
1851         }
1852
1853         if (sa->state == SFC_ETHDEV_STARTED) {
1854                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1855                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1856                 if (rc != 0)
1857                         goto fail_scale_tbl_set;
1858         }
1859
1860         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1861
1862 fail_scale_tbl_set:
1863 bad_reta_entry:
1864         sfc_adapter_unlock(sa);
1865
1866         rte_free(rss_tbl_new);
1867
1868         SFC_ASSERT(rc >= 0);
1869         return -rc;
1870 }
1871
1872 static int
1873 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused,
1874                      const struct rte_flow_ops **ops)
1875 {
1876         *ops = &sfc_flow_ops;
1877         return 0;
1878 }
1879
1880 static int
1881 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1882 {
1883         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1884
1885         /*
1886          * If Rx datapath does not provide callback to check mempool,
1887          * all pools are supported.
1888          */
1889         if (sap->dp_rx->pool_ops_supported == NULL)
1890                 return 1;
1891
1892         return sap->dp_rx->pool_ops_supported(pool);
1893 }
1894
1895 static int
1896 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1897 {
1898         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1899         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1900         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1901         struct sfc_rxq_info *rxq_info;
1902
1903         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1904
1905         return sap->dp_rx->intr_enable(rxq_info->dp);
1906 }
1907
1908 static int
1909 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid)
1910 {
1911         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1912         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1913         sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid;
1914         struct sfc_rxq_info *rxq_info;
1915
1916         rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid);
1917
1918         return sap->dp_rx->intr_disable(rxq_info->dp);
1919 }
1920
1921 static const struct eth_dev_ops sfc_eth_dev_ops = {
1922         .dev_configure                  = sfc_dev_configure,
1923         .dev_start                      = sfc_dev_start,
1924         .dev_stop                       = sfc_dev_stop,
1925         .dev_set_link_up                = sfc_dev_set_link_up,
1926         .dev_set_link_down              = sfc_dev_set_link_down,
1927         .dev_close                      = sfc_dev_close,
1928         .promiscuous_enable             = sfc_dev_promisc_enable,
1929         .promiscuous_disable            = sfc_dev_promisc_disable,
1930         .allmulticast_enable            = sfc_dev_allmulti_enable,
1931         .allmulticast_disable           = sfc_dev_allmulti_disable,
1932         .link_update                    = sfc_dev_link_update,
1933         .stats_get                      = sfc_stats_get,
1934         .stats_reset                    = sfc_stats_reset,
1935         .xstats_get                     = sfc_xstats_get,
1936         .xstats_reset                   = sfc_stats_reset,
1937         .xstats_get_names               = sfc_xstats_get_names,
1938         .dev_infos_get                  = sfc_dev_infos_get,
1939         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1940         .mtu_set                        = sfc_dev_set_mtu,
1941         .rx_queue_start                 = sfc_rx_queue_start,
1942         .rx_queue_stop                  = sfc_rx_queue_stop,
1943         .tx_queue_start                 = sfc_tx_queue_start,
1944         .tx_queue_stop                  = sfc_tx_queue_stop,
1945         .rx_queue_setup                 = sfc_rx_queue_setup,
1946         .rx_queue_release               = sfc_rx_queue_release,
1947         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1948         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1949         .tx_queue_setup                 = sfc_tx_queue_setup,
1950         .tx_queue_release               = sfc_tx_queue_release,
1951         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1952         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1953         .mac_addr_set                   = sfc_mac_addr_set,
1954         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1955         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1956         .reta_update                    = sfc_dev_rss_reta_update,
1957         .reta_query                     = sfc_dev_rss_reta_query,
1958         .rss_hash_update                = sfc_dev_rss_hash_update,
1959         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1960         .flow_ops_get                   = sfc_dev_flow_ops_get,
1961         .set_mc_addr_list               = sfc_set_mc_addr_list,
1962         .rxq_info_get                   = sfc_rx_queue_info_get,
1963         .txq_info_get                   = sfc_tx_queue_info_get,
1964         .fw_version_get                 = sfc_fw_version_get,
1965         .xstats_get_by_id               = sfc_xstats_get_by_id,
1966         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1967         .pool_ops_supported             = sfc_pool_ops_supported,
1968 };
1969
1970 struct sfc_ethdev_init_data {
1971         uint16_t                nb_representors;
1972 };
1973
1974 /**
1975  * Duplicate a string in potentially shared memory required for
1976  * multi-process support.
1977  *
1978  * strdup() allocates from process-local heap/memory.
1979  */
1980 static char *
1981 sfc_strdup(const char *str)
1982 {
1983         size_t size;
1984         char *copy;
1985
1986         if (str == NULL)
1987                 return NULL;
1988
1989         size = strlen(str) + 1;
1990         copy = rte_malloc(__func__, size, 0);
1991         if (copy != NULL)
1992                 rte_memcpy(copy, str, size);
1993
1994         return copy;
1995 }
1996
1997 static int
1998 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1999 {
2000         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2001         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2002         const struct sfc_dp_rx *dp_rx;
2003         const struct sfc_dp_tx *dp_tx;
2004         const efx_nic_cfg_t *encp;
2005         unsigned int avail_caps = 0;
2006         const char *rx_name = NULL;
2007         const char *tx_name = NULL;
2008         int rc;
2009
2010         switch (sa->family) {
2011         case EFX_FAMILY_HUNTINGTON:
2012         case EFX_FAMILY_MEDFORD:
2013         case EFX_FAMILY_MEDFORD2:
2014                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
2015                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
2016                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
2017                 break;
2018         case EFX_FAMILY_RIVERHEAD:
2019                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
2020                 break;
2021         default:
2022                 break;
2023         }
2024
2025         encp = efx_nic_cfg_get(sa->nic);
2026         if (encp->enc_rx_es_super_buffer_supported)
2027                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
2028
2029         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
2030                                 sfc_kvarg_string_handler, &rx_name);
2031         if (rc != 0)
2032                 goto fail_kvarg_rx_datapath;
2033
2034         if (rx_name != NULL) {
2035                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
2036                 if (dp_rx == NULL) {
2037                         sfc_err(sa, "Rx datapath %s not found", rx_name);
2038                         rc = ENOENT;
2039                         goto fail_dp_rx;
2040                 }
2041                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
2042                         sfc_err(sa,
2043                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
2044                                 rx_name);
2045                         rc = EINVAL;
2046                         goto fail_dp_rx_caps;
2047                 }
2048         } else {
2049                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
2050                 if (dp_rx == NULL) {
2051                         sfc_err(sa, "Rx datapath by caps %#x not found",
2052                                 avail_caps);
2053                         rc = ENOENT;
2054                         goto fail_dp_rx;
2055                 }
2056         }
2057
2058         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
2059         if (sas->dp_rx_name == NULL) {
2060                 rc = ENOMEM;
2061                 goto fail_dp_rx_name;
2062         }
2063
2064         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
2065
2066         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
2067                                 sfc_kvarg_string_handler, &tx_name);
2068         if (rc != 0)
2069                 goto fail_kvarg_tx_datapath;
2070
2071         if (tx_name != NULL) {
2072                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
2073                 if (dp_tx == NULL) {
2074                         sfc_err(sa, "Tx datapath %s not found", tx_name);
2075                         rc = ENOENT;
2076                         goto fail_dp_tx;
2077                 }
2078                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
2079                         sfc_err(sa,
2080                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
2081                                 tx_name);
2082                         rc = EINVAL;
2083                         goto fail_dp_tx_caps;
2084                 }
2085         } else {
2086                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
2087                 if (dp_tx == NULL) {
2088                         sfc_err(sa, "Tx datapath by caps %#x not found",
2089                                 avail_caps);
2090                         rc = ENOENT;
2091                         goto fail_dp_tx;
2092                 }
2093         }
2094
2095         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2096         if (sas->dp_tx_name == NULL) {
2097                 rc = ENOMEM;
2098                 goto fail_dp_tx_name;
2099         }
2100
2101         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2102
2103         sa->priv.dp_rx = dp_rx;
2104         sa->priv.dp_tx = dp_tx;
2105
2106         dev->rx_pkt_burst = dp_rx->pkt_burst;
2107         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2108         dev->tx_pkt_burst = dp_tx->pkt_burst;
2109
2110         dev->rx_queue_count = sfc_rx_queue_count;
2111         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2112         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2113         dev->dev_ops = &sfc_eth_dev_ops;
2114
2115         return 0;
2116
2117 fail_dp_tx_name:
2118 fail_dp_tx_caps:
2119 fail_dp_tx:
2120 fail_kvarg_tx_datapath:
2121         rte_free(sas->dp_rx_name);
2122         sas->dp_rx_name = NULL;
2123
2124 fail_dp_rx_name:
2125 fail_dp_rx_caps:
2126 fail_dp_rx:
2127 fail_kvarg_rx_datapath:
2128         return rc;
2129 }
2130
2131 static void
2132 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2133 {
2134         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2135         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2136
2137         dev->dev_ops = NULL;
2138         dev->tx_pkt_prepare = NULL;
2139         dev->rx_pkt_burst = NULL;
2140         dev->tx_pkt_burst = NULL;
2141
2142         rte_free(sas->dp_tx_name);
2143         sas->dp_tx_name = NULL;
2144         sa->priv.dp_tx = NULL;
2145
2146         rte_free(sas->dp_rx_name);
2147         sas->dp_rx_name = NULL;
2148         sa->priv.dp_rx = NULL;
2149 }
2150
2151 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2152         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2153         .reta_query                     = sfc_dev_rss_reta_query,
2154         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2155         .rxq_info_get                   = sfc_rx_queue_info_get,
2156         .txq_info_get                   = sfc_tx_queue_info_get,
2157 };
2158
2159 static int
2160 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2161 {
2162         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2163         struct sfc_adapter_priv *sap;
2164         const struct sfc_dp_rx *dp_rx;
2165         const struct sfc_dp_tx *dp_tx;
2166         int rc;
2167
2168         /*
2169          * Allocate process private data from heap, since it should not
2170          * be located in shared memory allocated using rte_malloc() API.
2171          */
2172         sap = calloc(1, sizeof(*sap));
2173         if (sap == NULL) {
2174                 rc = ENOMEM;
2175                 goto fail_alloc_priv;
2176         }
2177
2178         sap->logtype_main = logtype_main;
2179
2180         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2181         if (dp_rx == NULL) {
2182                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2183                         "cannot find %s Rx datapath", sas->dp_rx_name);
2184                 rc = ENOENT;
2185                 goto fail_dp_rx;
2186         }
2187         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2188                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2189                         "%s Rx datapath does not support multi-process",
2190                         sas->dp_rx_name);
2191                 rc = EINVAL;
2192                 goto fail_dp_rx_multi_process;
2193         }
2194
2195         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2196         if (dp_tx == NULL) {
2197                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2198                         "cannot find %s Tx datapath", sas->dp_tx_name);
2199                 rc = ENOENT;
2200                 goto fail_dp_tx;
2201         }
2202         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2203                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2204                         "%s Tx datapath does not support multi-process",
2205                         sas->dp_tx_name);
2206                 rc = EINVAL;
2207                 goto fail_dp_tx_multi_process;
2208         }
2209
2210         sap->dp_rx = dp_rx;
2211         sap->dp_tx = dp_tx;
2212
2213         dev->process_private = sap;
2214         dev->rx_pkt_burst = dp_rx->pkt_burst;
2215         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2216         dev->tx_pkt_burst = dp_tx->pkt_burst;
2217         dev->rx_queue_count = sfc_rx_queue_count;
2218         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2219         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2220         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2221
2222         return 0;
2223
2224 fail_dp_tx_multi_process:
2225 fail_dp_tx:
2226 fail_dp_rx_multi_process:
2227 fail_dp_rx:
2228         free(sap);
2229
2230 fail_alloc_priv:
2231         return rc;
2232 }
2233
2234 static void
2235 sfc_register_dp(void)
2236 {
2237         /* Register once */
2238         if (TAILQ_EMPTY(&sfc_dp_head)) {
2239                 /* Prefer EF10 datapath */
2240                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2241                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2242                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2243                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2244
2245                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2246                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2247                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2248                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2249         }
2250 }
2251
2252 static int
2253 sfc_parse_switch_mode(struct sfc_adapter *sa, bool has_representors)
2254 {
2255         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2256         const char *switch_mode = NULL;
2257         int rc;
2258
2259         sfc_log_init(sa, "entry");
2260
2261         rc = sfc_kvargs_process(sa, SFC_KVARG_SWITCH_MODE,
2262                                 sfc_kvarg_string_handler, &switch_mode);
2263         if (rc != 0)
2264                 goto fail_kvargs;
2265
2266         if (switch_mode == NULL) {
2267                 sa->switchdev = encp->enc_mae_supported &&
2268                                 (!encp->enc_datapath_cap_evb ||
2269                                  has_representors);
2270         } else if (strcasecmp(switch_mode, SFC_KVARG_SWITCH_MODE_LEGACY) == 0) {
2271                 sa->switchdev = false;
2272         } else if (strcasecmp(switch_mode,
2273                               SFC_KVARG_SWITCH_MODE_SWITCHDEV) == 0) {
2274                 sa->switchdev = true;
2275         } else {
2276                 sfc_err(sa, "invalid switch mode device argument '%s'",
2277                         switch_mode);
2278                 rc = EINVAL;
2279                 goto fail_mode;
2280         }
2281
2282         sfc_log_init(sa, "done");
2283
2284         return 0;
2285
2286 fail_mode:
2287 fail_kvargs:
2288         sfc_log_init(sa, "failed: %s", rte_strerror(rc));
2289
2290         return rc;
2291 }
2292
2293 static int
2294 sfc_eth_dev_init(struct rte_eth_dev *dev, void *init_params)
2295 {
2296         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2297         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2298         struct sfc_ethdev_init_data *init_data = init_params;
2299         uint32_t logtype_main;
2300         struct sfc_adapter *sa;
2301         int rc;
2302         const efx_nic_cfg_t *encp;
2303         const struct rte_ether_addr *from;
2304         int ret;
2305
2306         if (sfc_efx_dev_class_get(pci_dev->device.devargs) !=
2307                         SFC_EFX_DEV_CLASS_NET) {
2308                 SFC_GENERIC_LOG(DEBUG,
2309                         "Incompatible device class: skip probing, should be probed by other sfc driver.");
2310                 return 1;
2311         }
2312
2313         rc = sfc_dp_mport_register();
2314         if (rc != 0)
2315                 return rc;
2316
2317         sfc_register_dp();
2318
2319         logtype_main = sfc_register_logtype(&pci_dev->addr,
2320                                             SFC_LOGTYPE_MAIN_STR,
2321                                             RTE_LOG_NOTICE);
2322
2323         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2324                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2325
2326         /* Required for logging */
2327         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2328                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2329                         pci_dev->addr.domain, pci_dev->addr.bus,
2330                         pci_dev->addr.devid, pci_dev->addr.function,
2331                         dev->data->port_id);
2332         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2333                 SFC_GENERIC_LOG(ERR,
2334                         "reserved log prefix is too short for " PCI_PRI_FMT,
2335                         pci_dev->addr.domain, pci_dev->addr.bus,
2336                         pci_dev->addr.devid, pci_dev->addr.function);
2337                 return -EINVAL;
2338         }
2339         sas->pci_addr = pci_dev->addr;
2340         sas->port_id = dev->data->port_id;
2341
2342         /*
2343          * Allocate process private data from heap, since it should not
2344          * be located in shared memory allocated using rte_malloc() API.
2345          */
2346         sa = calloc(1, sizeof(*sa));
2347         if (sa == NULL) {
2348                 rc = ENOMEM;
2349                 goto fail_alloc_sa;
2350         }
2351
2352         dev->process_private = sa;
2353
2354         /* Required for logging */
2355         sa->priv.shared = sas;
2356         sa->priv.logtype_main = logtype_main;
2357
2358         sa->eth_dev = dev;
2359
2360         /* Copy PCI device info to the dev->data */
2361         rte_eth_copy_pci_info(dev, pci_dev);
2362         dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
2363         dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE;
2364
2365         rc = sfc_kvargs_parse(sa);
2366         if (rc != 0)
2367                 goto fail_kvargs_parse;
2368
2369         sfc_log_init(sa, "entry");
2370
2371         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2372         if (dev->data->mac_addrs == NULL) {
2373                 rc = ENOMEM;
2374                 goto fail_mac_addrs;
2375         }
2376
2377         sfc_adapter_lock_init(sa);
2378         sfc_adapter_lock(sa);
2379
2380         sfc_log_init(sa, "probing");
2381         rc = sfc_probe(sa);
2382         if (rc != 0)
2383                 goto fail_probe;
2384
2385         /*
2386          * Selecting a default switch mode requires the NIC to be probed and
2387          * to have its capabilities filled in.
2388          */
2389         rc = sfc_parse_switch_mode(sa, init_data->nb_representors > 0);
2390         if (rc != 0)
2391                 goto fail_switch_mode;
2392
2393         sfc_log_init(sa, "set device ops");
2394         rc = sfc_eth_dev_set_ops(dev);
2395         if (rc != 0)
2396                 goto fail_set_ops;
2397
2398         sfc_log_init(sa, "attaching");
2399         rc = sfc_attach(sa);
2400         if (rc != 0)
2401                 goto fail_attach;
2402
2403         if (sa->switchdev && sa->mae.status != SFC_MAE_STATUS_SUPPORTED) {
2404                 sfc_err(sa,
2405                         "failed to enable switchdev mode without MAE support");
2406                 rc = ENOTSUP;
2407                 goto fail_switchdev_no_mae;
2408         }
2409
2410         encp = efx_nic_cfg_get(sa->nic);
2411
2412         /*
2413          * The arguments are really reverse order in comparison to
2414          * Linux kernel. Copy from NIC config to Ethernet device data.
2415          */
2416         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2417         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2418
2419         sfc_adapter_unlock(sa);
2420
2421         sfc_log_init(sa, "done");
2422         return 0;
2423
2424 fail_switchdev_no_mae:
2425         sfc_detach(sa);
2426
2427 fail_attach:
2428         sfc_eth_dev_clear_ops(dev);
2429
2430 fail_set_ops:
2431 fail_switch_mode:
2432         sfc_unprobe(sa);
2433
2434 fail_probe:
2435         sfc_adapter_unlock(sa);
2436         sfc_adapter_lock_fini(sa);
2437         rte_free(dev->data->mac_addrs);
2438         dev->data->mac_addrs = NULL;
2439
2440 fail_mac_addrs:
2441         sfc_kvargs_cleanup(sa);
2442
2443 fail_kvargs_parse:
2444         sfc_log_init(sa, "failed %d", rc);
2445         dev->process_private = NULL;
2446         free(sa);
2447
2448 fail_alloc_sa:
2449         SFC_ASSERT(rc > 0);
2450         return -rc;
2451 }
2452
2453 static int
2454 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2455 {
2456         sfc_dev_close(dev);
2457
2458         return 0;
2459 }
2460
2461 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2462         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2463         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2464         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2465         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2466         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2467         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2468         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2469         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2470         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2471         { .vendor_id = 0 /* sentinel */ }
2472 };
2473
2474 static int
2475 sfc_parse_rte_devargs(const char *args, struct rte_eth_devargs *devargs)
2476 {
2477         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
2478         int rc;
2479
2480         if (args != NULL) {
2481                 rc = rte_eth_devargs_parse(args, &eth_da);
2482                 if (rc != 0) {
2483                         SFC_GENERIC_LOG(ERR,
2484                                         "Failed to parse generic devargs '%s'",
2485                                         args);
2486                         return rc;
2487                 }
2488         }
2489
2490         *devargs = eth_da;
2491
2492         return 0;
2493 }
2494
2495 static int
2496 sfc_eth_dev_find_or_create(struct rte_pci_device *pci_dev,
2497                            struct sfc_ethdev_init_data *init_data,
2498                            struct rte_eth_dev **devp,
2499                            bool *dev_created)
2500 {
2501         struct rte_eth_dev *dev;
2502         bool created = false;
2503         int rc;
2504
2505         dev = rte_eth_dev_allocated(pci_dev->device.name);
2506         if (dev == NULL) {
2507                 rc = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
2508                                         sizeof(struct sfc_adapter_shared),
2509                                         eth_dev_pci_specific_init, pci_dev,
2510                                         sfc_eth_dev_init, init_data);
2511                 if (rc != 0) {
2512                         SFC_GENERIC_LOG(ERR, "Failed to create sfc ethdev '%s'",
2513                                         pci_dev->device.name);
2514                         return rc;
2515                 }
2516
2517                 created = true;
2518
2519                 dev = rte_eth_dev_allocated(pci_dev->device.name);
2520                 if (dev == NULL) {
2521                         SFC_GENERIC_LOG(ERR,
2522                                 "Failed to find allocated sfc ethdev '%s'",
2523                                 pci_dev->device.name);
2524                         return -ENODEV;
2525                 }
2526         }
2527
2528         *devp = dev;
2529         *dev_created = created;
2530
2531         return 0;
2532 }
2533
2534 static int
2535 sfc_eth_dev_create_representors(struct rte_eth_dev *dev,
2536                                 const struct rte_eth_devargs *eth_da)
2537 {
2538         struct sfc_adapter *sa;
2539         unsigned int i;
2540         int rc;
2541
2542         if (eth_da->nb_representor_ports == 0)
2543                 return 0;
2544
2545         sa = sfc_adapter_by_eth_dev(dev);
2546
2547         if (!sa->switchdev) {
2548                 sfc_err(sa, "cannot create representors in non-switchdev mode");
2549                 return -EINVAL;
2550         }
2551
2552         if (!sfc_repr_available(sfc_sa2shared(sa))) {
2553                 sfc_err(sa, "cannot create representors: unsupported");
2554
2555                 return -ENOTSUP;
2556         }
2557
2558         for (i = 0; i < eth_da->nb_representor_ports; ++i) {
2559                 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
2560                 efx_mport_sel_t mport_sel;
2561
2562                 rc = efx_mae_mport_by_pcie_function(encp->enc_pf,
2563                                 eth_da->representor_ports[i], &mport_sel);
2564                 if (rc != 0) {
2565                         sfc_err(sa,
2566                                 "failed to get representor %u m-port: %s - ignore",
2567                                 eth_da->representor_ports[i],
2568                                 rte_strerror(-rc));
2569                         continue;
2570                 }
2571
2572                 rc = sfc_repr_create(dev, eth_da->representor_ports[i],
2573                                      sa->mae.switch_domain_id, &mport_sel);
2574                 if (rc != 0) {
2575                         sfc_err(sa, "cannot create representor %u: %s - ignore",
2576                                 eth_da->representor_ports[i],
2577                                 rte_strerror(-rc));
2578                 }
2579         }
2580
2581         return 0;
2582 }
2583
2584 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2585         struct rte_pci_device *pci_dev)
2586 {
2587         struct sfc_ethdev_init_data init_data;
2588         struct rte_eth_devargs eth_da;
2589         struct rte_eth_dev *dev;
2590         bool dev_created;
2591         int rc;
2592
2593         if (pci_dev->device.devargs != NULL) {
2594                 rc = sfc_parse_rte_devargs(pci_dev->device.devargs->args,
2595                                            &eth_da);
2596                 if (rc != 0)
2597                         return rc;
2598         } else {
2599                 memset(&eth_da, 0, sizeof(eth_da));
2600         }
2601
2602         init_data.nb_representors = eth_da.nb_representor_ports;
2603
2604         if (eth_da.nb_representor_ports > 0 &&
2605             rte_eal_process_type() != RTE_PROC_PRIMARY) {
2606                 SFC_GENERIC_LOG(ERR,
2607                         "Create representors from secondary process not supported, dev '%s'",
2608                         pci_dev->device.name);
2609                 return -ENOTSUP;
2610         }
2611
2612         /*
2613          * Driver supports RTE_PCI_DRV_PROBE_AGAIN. Hence create device only
2614          * if it does not already exist. Re-probing an existing device is
2615          * expected to allow additional representors to be configured.
2616          */
2617         rc = sfc_eth_dev_find_or_create(pci_dev, &init_data, &dev,
2618                                         &dev_created);
2619         if (rc != 0)
2620                 return rc;
2621
2622         rc = sfc_eth_dev_create_representors(dev, &eth_da);
2623         if (rc != 0) {
2624                 if (dev_created)
2625                         (void)rte_eth_dev_destroy(dev, sfc_eth_dev_uninit);
2626
2627                 return rc;
2628         }
2629
2630         return 0;
2631 }
2632
2633 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2634 {
2635         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2636 }
2637
2638 static struct rte_pci_driver sfc_efx_pmd = {
2639         .id_table = pci_id_sfc_efx_map,
2640         .drv_flags =
2641                 RTE_PCI_DRV_INTR_LSC |
2642                 RTE_PCI_DRV_NEED_MAPPING |
2643                 RTE_PCI_DRV_PROBE_AGAIN,
2644         .probe = sfc_eth_dev_pci_probe,
2645         .remove = sfc_eth_dev_pci_remove,
2646 };
2647
2648 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2649 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2650 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2651 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2652         SFC_KVARG_SWITCH_MODE "=" SFC_KVARG_VALUES_SWITCH_MODE " "
2653         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2654         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2655         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2656         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2657         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2658         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2659
2660 RTE_INIT(sfc_driver_register_logtype)
2661 {
2662         int ret;
2663
2664         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2665                                                    RTE_LOG_NOTICE);
2666         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2667 }