drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2020 Xilinx, Inc.
   4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <rte_ethdev_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17 #include <rte_ether.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_debug.h"
  23 #include "sfc_log.h"
  24 #include "sfc_kvargs.h"
  25 #include "sfc_ev.h"
  26 #include "sfc_rx.h"
  27 #include "sfc_tx.h"
  28 #include "sfc_flow.h"
  29 #include "sfc_dp.h"
  30 #include "sfc_dp_rx.h"
  31
  32 uint32_t sfc_logtype_driver;
  33
  34 static struct sfc_dp_list sfc_dp_head =
  35         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  36
  37
  38 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  39
  40
  41 static int
  42 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  43 {
  44         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  45         efx_nic_fw_info_t enfi;
  46         int ret;
  47         int rc;
  48
  49         /*
  50          * Return value of the callback is likely supposed to be
  51          * equal to or greater than 0, nevertheless, if an error
  52          * occurs, it will be desirable to pass it to the caller
  53          */
  54         if ((fw_version == NULL) || (fw_size == 0))
  55                 return -EINVAL;
  56
  57         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  58         if (rc != 0)
  59                 return -rc;
  60
  61         ret = snprintf(fw_version, fw_size,
  62                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  63                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  64                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  65         if (ret < 0)
  66                 return ret;
  67
  68         if (enfi.enfi_dpcpu_fw_ids_valid) {
  69                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  70                 int ret_extra;
  71
  72                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  73                                      fw_size - dpcpu_fw_ids_offset,
  74                                      " rx%" PRIx16 " tx%" PRIx16,
  75                                      enfi.enfi_rx_dpcpu_fw_id,
  76                                      enfi.enfi_tx_dpcpu_fw_id);
  77                 if (ret_extra < 0)
  78                         return ret_extra;
  79
  80                 ret += ret_extra;
  81         }
  82
  83         if (fw_size < (size_t)(++ret))
  84                 return ret;
  85         else
  86                 return 0;
  87 }
  88
  89 static int
  90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  91 {
  92         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  93         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  94         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  95         struct sfc_rss *rss = &sas->rss;
  96         uint64_t txq_offloads_def = 0;
  97
  98         sfc_log_init(sa, "entry");
  99
 100         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 101         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 102
 103         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 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         return 0;
 189 }
 190
 191 static const uint32_t *
 192 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 193 {
 194         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 195
 196         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 197 }
 198
 199 static int
 200 sfc_dev_configure(struct rte_eth_dev *dev)
 201 {
 202         struct rte_eth_dev_data *dev_data = dev->data;
 203         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 204         int rc;
 205
 206         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 207                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 208
 209         sfc_adapter_lock(sa);
 210         switch (sa->state) {
 211         case SFC_ADAPTER_CONFIGURED:
 212                 /* FALLTHROUGH */
 213         case SFC_ADAPTER_INITIALIZED:
 214                 rc = sfc_configure(sa);
 215                 break;
 216         default:
 217                 sfc_err(sa, "unexpected adapter state %u to configure",
 218                         sa->state);
 219                 rc = EINVAL;
 220                 break;
 221         }
 222         sfc_adapter_unlock(sa);
 223
 224         sfc_log_init(sa, "done %d", rc);
 225         SFC_ASSERT(rc >= 0);
 226         return -rc;
 227 }
 228
 229 static int
 230 sfc_dev_start(struct rte_eth_dev *dev)
 231 {
 232         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 233         int rc;
 234
 235         sfc_log_init(sa, "entry");
 236
 237         sfc_adapter_lock(sa);
 238         rc = sfc_start(sa);
 239         sfc_adapter_unlock(sa);
 240
 241         sfc_log_init(sa, "done %d", rc);
 242         SFC_ASSERT(rc >= 0);
 243         return -rc;
 244 }
 245
 246 static int
 247 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 248 {
 249         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 250         struct rte_eth_link current_link;
 251         int ret;
 252
 253         sfc_log_init(sa, "entry");
 254
 255         if (sa->state != SFC_ADAPTER_STARTED) {
 256                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 257         } else if (wait_to_complete) {
 258                 efx_link_mode_t link_mode;
 259
 260                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 261                         link_mode = EFX_LINK_UNKNOWN;
 262                 sfc_port_link_mode_to_info(link_mode, &current_link);
 263
 264         } else {
 265                 sfc_ev_mgmt_qpoll(sa);
 266                 rte_eth_linkstatus_get(dev, &current_link);
 267         }
 268
 269         ret = rte_eth_linkstatus_set(dev, &current_link);
 270         if (ret == 0)
 271                 sfc_notice(sa, "Link status is %s",
 272                            current_link.link_status ? "UP" : "DOWN");
 273
 274         return ret;
 275 }
 276
 277 static void
 278 sfc_dev_stop(struct rte_eth_dev *dev)
 279 {
 280         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 281
 282         sfc_log_init(sa, "entry");
 283
 284         sfc_adapter_lock(sa);
 285         sfc_stop(sa);
 286         sfc_adapter_unlock(sa);
 287
 288         sfc_log_init(sa, "done");
 289 }
 290
 291 static int
 292 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 293 {
 294         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 295         int rc;
 296
 297         sfc_log_init(sa, "entry");
 298
 299         sfc_adapter_lock(sa);
 300         rc = sfc_start(sa);
 301         sfc_adapter_unlock(sa);
 302
 303         SFC_ASSERT(rc >= 0);
 304         return -rc;
 305 }
 306
 307 static int
 308 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 309 {
 310         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 311
 312         sfc_log_init(sa, "entry");
 313
 314         sfc_adapter_lock(sa);
 315         sfc_stop(sa);
 316         sfc_adapter_unlock(sa);
 317
 318         return 0;
 319 }
 320
 321 static void
 322 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 323 {
 324         free(dev->process_private);
 325         dev->process_private = NULL;
 326         dev->dev_ops = NULL;
 327         dev->tx_pkt_prepare = NULL;
 328         dev->tx_pkt_burst = NULL;
 329         dev->rx_pkt_burst = NULL;
 330 }
 331
 332 static int
 333 sfc_dev_close(struct rte_eth_dev *dev)
 334 {
 335         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 336
 337         sfc_log_init(sa, "entry");
 338
 339         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 340                 sfc_eth_dev_secondary_clear_ops(dev);
 341                 return 0;
 342         }
 343
 344         sfc_adapter_lock(sa);
 345         switch (sa->state) {
 346         case SFC_ADAPTER_STARTED:
 347                 sfc_stop(sa);
 348                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 349                 /* FALLTHROUGH */
 350         case SFC_ADAPTER_CONFIGURED:
 351                 sfc_close(sa);
 352                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 353                 /* FALLTHROUGH */
 354         case SFC_ADAPTER_INITIALIZED:
 355                 break;
 356         default:
 357                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 358                 break;
 359         }
 360
 361         /*
 362          * Cleanup all resources.
 363          * Rollback primary process sfc_eth_dev_init() below.
 364          */
 365
 366         sfc_eth_dev_clear_ops(dev);
 367
 368         sfc_detach(sa);
 369         sfc_unprobe(sa);
 370
 371         sfc_kvargs_cleanup(sa);
 372
 373         sfc_adapter_unlock(sa);
 374         sfc_adapter_lock_fini(sa);
 375
 376         sfc_log_init(sa, "done");
 377
 378         /* Required for logging, so cleanup last */
 379         sa->eth_dev = NULL;
 380
 381         dev->process_private = NULL;
 382         free(sa);
 383
 384         return 0;
 385 }
 386
 387 static int
 388 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 389                    boolean_t enabled)
 390 {
 391         struct sfc_port *port;
 392         boolean_t *toggle;
 393         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 394         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 395         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 396         int rc = 0;
 397
 398         sfc_adapter_lock(sa);
 399
 400         port = &sa->port;
 401         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 402
 403         if (*toggle != enabled) {
 404                 *toggle = enabled;
 405
 406                 if (sfc_sa2shared(sa)->isolated) {
 407                         sfc_warn(sa, "isolated mode is active on the port");
 408                         sfc_warn(sa, "the change is to be applied on the next "
 409                                      "start provided that isolated mode is "
 410                                      "disabled prior the next start");
 411                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 412                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 413                         *toggle = !(enabled);
 414                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 415                                  ((enabled) ? "enable" : "disable"), desc, rc);
 416
 417                         /*
 418                          * For promiscuous and all-multicast filters a
 419                          * permission failure should be reported as an
 420                          * unsupported filter.
 421                          */
 422                         if (rc == EPERM)
 423                                 rc = ENOTSUP;
 424                 }
 425         }
 426
 427         sfc_adapter_unlock(sa);
 428         return rc;
 429 }
 430
 431 static int
 432 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 433 {
 434         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 435
 436         SFC_ASSERT(rc >= 0);
 437         return -rc;
 438 }
 439
 440 static int
 441 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 442 {
 443         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 444
 445         SFC_ASSERT(rc >= 0);
 446         return -rc;
 447 }
 448
 449 static int
 450 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 451 {
 452         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 453
 454         SFC_ASSERT(rc >= 0);
 455         return -rc;
 456 }
 457
 458 static int
 459 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 460 {
 461         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 462
 463         SFC_ASSERT(rc >= 0);
 464         return -rc;
 465 }
 466
 467 static int
 468 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 469                    uint16_t nb_rx_desc, unsigned int socket_id,
 470                    const struct rte_eth_rxconf *rx_conf,
 471                    struct rte_mempool *mb_pool)
 472 {
 473         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 474         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 475         int rc;
 476
 477         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 478                      rx_queue_id, nb_rx_desc, socket_id);
 479
 480         sfc_adapter_lock(sa);
 481
 482         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 483                           rx_conf, mb_pool);
 484         if (rc != 0)
 485                 goto fail_rx_qinit;
 486
 487         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 488
 489         sfc_adapter_unlock(sa);
 490
 491         return 0;
 492
 493 fail_rx_qinit:
 494         sfc_adapter_unlock(sa);
 495         SFC_ASSERT(rc > 0);
 496         return -rc;
 497 }
 498
 499 static void
 500 sfc_rx_queue_release(void *queue)
 501 {
 502         struct sfc_dp_rxq *dp_rxq = queue;
 503         struct sfc_rxq *rxq;
 504         struct sfc_adapter *sa;
 505         unsigned int sw_index;
 506
 507         if (dp_rxq == NULL)
 508                 return;
 509
 510         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 511         sa = rxq->evq->sa;
 512         sfc_adapter_lock(sa);
 513
 514         sw_index = dp_rxq->dpq.queue_id;
 515
 516         sfc_log_init(sa, "RxQ=%u", sw_index);
 517
 518         sfc_rx_qfini(sa, sw_index);
 519
 520         sfc_adapter_unlock(sa);
 521 }
 522
 523 static int
 524 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 525                    uint16_t nb_tx_desc, unsigned int socket_id,
 526                    const struct rte_eth_txconf *tx_conf)
 527 {
 528         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 529         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 530         int rc;
 531
 532         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 533                      tx_queue_id, nb_tx_desc, socket_id);
 534
 535         sfc_adapter_lock(sa);
 536
 537         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 538         if (rc != 0)
 539                 goto fail_tx_qinit;
 540
 541         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
 542
 543         sfc_adapter_unlock(sa);
 544         return 0;
 545
 546 fail_tx_qinit:
 547         sfc_adapter_unlock(sa);
 548         SFC_ASSERT(rc > 0);
 549         return -rc;
 550 }
 551
 552 static void
 553 sfc_tx_queue_release(void *queue)
 554 {
 555         struct sfc_dp_txq *dp_txq = queue;
 556         struct sfc_txq *txq;
 557         unsigned int sw_index;
 558         struct sfc_adapter *sa;
 559
 560         if (dp_txq == NULL)
 561                 return;
 562
 563         txq = sfc_txq_by_dp_txq(dp_txq);
 564         sw_index = dp_txq->dpq.queue_id;
 565
 566         SFC_ASSERT(txq->evq != NULL);
 567         sa = txq->evq->sa;
 568
 569         sfc_log_init(sa, "TxQ = %u", sw_index);
 570
 571         sfc_adapter_lock(sa);
 572
 573         sfc_tx_qfini(sa, sw_index);
 574
 575         sfc_adapter_unlock(sa);
 576 }
 577
 578 /*
 579  * Some statistics are computed as A - B where A and B each increase
 580  * monotonically with some hardware counter(s) and the counters are read
 581  * asynchronously.
 582  *
 583  * If packet X is counted in A, but not counted in B yet, computed value is
 584  * greater than real.
 585  *
 586  * If packet X is not counted in A at the moment of reading the counter,
 587  * but counted in B at the moment of reading the counter, computed value
 588  * is less than real.
 589  *
 590  * However, counter which grows backward is worse evil than slightly wrong
 591  * value. So, let's try to guarantee that it never happens except may be
 592  * the case when the MAC stats are zeroed as a result of a NIC reset.
 593  */
 594 static void
 595 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 596 {
 597         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 598                 *stat = newval;
 599 }
 600
 601 static int
 602 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 603 {
 604         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 605         struct sfc_port *port = &sa->port;
 606         uint64_t *mac_stats;
 607         int ret;
 608
 609         rte_spinlock_lock(&port->mac_stats_lock);
 610
 611         ret = sfc_port_update_mac_stats(sa);
 612         if (ret != 0)
 613                 goto unlock;
 614
 615         mac_stats = port->mac_stats_buf;
 616
 617         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 618                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 619                 stats->ipackets =
 620                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 621                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 622                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 623                 stats->opackets =
 624                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 625                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 626                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 627                 stats->ibytes =
 628                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 629                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 630                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 631                 stats->obytes =
 632                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 633                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 634                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 635                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 636                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 637         } else {
 638                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 639                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 640                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 641                 /*
 642                  * Take into account stats which are whenever supported
 643                  * on EF10. If some stat is not supported by current
 644                  * firmware variant or HW revision, it is guaranteed
 645                  * to be zero in mac_stats.
 646                  */
 647                 stats->imissed =
 648                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 649                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 650                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 651                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 652                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 653                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 654                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 655                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 656                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 657                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 658                 stats->ierrors =
 659                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 660                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 661                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 662                 /* no oerrors counters supported on EF10 */
 663
 664                 /* Exclude missed, errors and pauses from Rx packets */
 665                 sfc_update_diff_stat(&port->ipackets,
 666                         mac_stats[EFX_MAC_RX_PKTS] -
 667                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 668                         stats->imissed - stats->ierrors);
 669                 stats->ipackets = port->ipackets;
 670         }
 671
 672 unlock:
 673         rte_spinlock_unlock(&port->mac_stats_lock);
 674         SFC_ASSERT(ret >= 0);
 675         return -ret;
 676 }
 677
 678 static int
 679 sfc_stats_reset(struct rte_eth_dev *dev)
 680 {
 681         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 682         struct sfc_port *port = &sa->port;
 683         int rc;
 684
 685         if (sa->state != SFC_ADAPTER_STARTED) {
 686                 /*
 687                  * The operation cannot be done if port is not started; it
 688                  * will be scheduled to be done during the next port start
 689                  */
 690                 port->mac_stats_reset_pending = B_TRUE;
 691                 return 0;
 692         }
 693
 694         rc = sfc_port_reset_mac_stats(sa);
 695         if (rc != 0)
 696                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 697
 698         SFC_ASSERT(rc >= 0);
 699         return -rc;
 700 }
 701
 702 static int
 703 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 704                unsigned int xstats_count)
 705 {
 706         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 707         struct sfc_port *port = &sa->port;
 708         uint64_t *mac_stats;
 709         int rc;
 710         unsigned int i;
 711         int nstats = 0;
 712
 713         rte_spinlock_lock(&port->mac_stats_lock);
 714
 715         rc = sfc_port_update_mac_stats(sa);
 716         if (rc != 0) {
 717                 SFC_ASSERT(rc > 0);
 718                 nstats = -rc;
 719                 goto unlock;
 720         }
 721
 722         mac_stats = port->mac_stats_buf;
 723
 724         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 725                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 726                         if (xstats != NULL && nstats < (int)xstats_count) {
 727                                 xstats[nstats].id = nstats;
 728                                 xstats[nstats].value = mac_stats[i];
 729                         }
 730                         nstats++;
 731                 }
 732         }
 733
 734 unlock:
 735         rte_spinlock_unlock(&port->mac_stats_lock);
 736
 737         return nstats;
 738 }
 739
 740 static int
 741 sfc_xstats_get_names(struct rte_eth_dev *dev,
 742                      struct rte_eth_xstat_name *xstats_names,
 743                      unsigned int xstats_count)
 744 {
 745         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 746         struct sfc_port *port = &sa->port;
 747         unsigned int i;
 748         unsigned int nstats = 0;
 749
 750         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 751                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 752                         if (xstats_names != NULL && nstats < xstats_count)
 753                                 strlcpy(xstats_names[nstats].name,
 754                                         efx_mac_stat_name(sa->nic, i),
 755                                         sizeof(xstats_names[0].name));
 756                         nstats++;
 757                 }
 758         }
 759
 760         return nstats;
 761 }
 762
 763 static int
 764 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 765                      uint64_t *values, unsigned int n)
 766 {
 767         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 768         struct sfc_port *port = &sa->port;
 769         uint64_t *mac_stats;
 770         unsigned int nb_supported = 0;
 771         unsigned int nb_written = 0;
 772         unsigned int i;
 773         int ret;
 774         int rc;
 775
 776         if (unlikely(values == NULL) ||
 777             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 778                 return port->mac_stats_nb_supported;
 779
 780         rte_spinlock_lock(&port->mac_stats_lock);
 781
 782         rc = sfc_port_update_mac_stats(sa);
 783         if (rc != 0) {
 784                 SFC_ASSERT(rc > 0);
 785                 ret = -rc;
 786                 goto unlock;
 787         }
 788
 789         mac_stats = port->mac_stats_buf;
 790
 791         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 792                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 793                         continue;
 794
 795                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 796                         values[nb_written++] = mac_stats[i];
 797
 798                 ++nb_supported;
 799         }
 800
 801         ret = nb_written;
 802
 803 unlock:
 804         rte_spinlock_unlock(&port->mac_stats_lock);
 805
 806         return ret;
 807 }
 808
 809 static int
 810 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 811                            struct rte_eth_xstat_name *xstats_names,
 812                            const uint64_t *ids, unsigned int size)
 813 {
 814         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 815         struct sfc_port *port = &sa->port;
 816         unsigned int nb_supported = 0;
 817         unsigned int nb_written = 0;
 818         unsigned int i;
 819
 820         if (unlikely(xstats_names == NULL) ||
 821             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 822                 return port->mac_stats_nb_supported;
 823
 824         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 825                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 826                         continue;
 827
 828                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 829                         char *name = xstats_names[nb_written++].name;
 830
 831                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 832                                 sizeof(xstats_names[0].name));
 833                 }
 834
 835                 ++nb_supported;
 836         }
 837
 838         return nb_written;
 839 }
 840
 841 static int
 842 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 843 {
 844         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 845         unsigned int wanted_fc, link_fc;
 846
 847         memset(fc_conf, 0, sizeof(*fc_conf));
 848
 849         sfc_adapter_lock(sa);
 850
 851         if (sa->state == SFC_ADAPTER_STARTED)
 852                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 853         else
 854                 link_fc = sa->port.flow_ctrl;
 855
 856         switch (link_fc) {
 857         case 0:
 858                 fc_conf->mode = RTE_FC_NONE;
 859                 break;
 860         case EFX_FCNTL_RESPOND:
 861                 fc_conf->mode = RTE_FC_RX_PAUSE;
 862                 break;
 863         case EFX_FCNTL_GENERATE:
 864                 fc_conf->mode = RTE_FC_TX_PAUSE;
 865                 break;
 866         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 867                 fc_conf->mode = RTE_FC_FULL;
 868                 break;
 869         default:
 870                 sfc_err(sa, "%s: unexpected flow control value %#x",
 871                         __func__, link_fc);
 872         }
 873
 874         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 875
 876         sfc_adapter_unlock(sa);
 877
 878         return 0;
 879 }
 880
 881 static int
 882 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 883 {
 884         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 885         struct sfc_port *port = &sa->port;
 886         unsigned int fcntl;
 887         int rc;
 888
 889         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 890             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 891             fc_conf->mac_ctrl_frame_fwd != 0) {
 892                 sfc_err(sa, "unsupported flow control settings specified");
 893                 rc = EINVAL;
 894                 goto fail_inval;
 895         }
 896
 897         switch (fc_conf->mode) {
 898         case RTE_FC_NONE:
 899                 fcntl = 0;
 900                 break;
 901         case RTE_FC_RX_PAUSE:
 902                 fcntl = EFX_FCNTL_RESPOND;
 903                 break;
 904         case RTE_FC_TX_PAUSE:
 905                 fcntl = EFX_FCNTL_GENERATE;
 906                 break;
 907         case RTE_FC_FULL:
 908                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 909                 break;
 910         default:
 911                 rc = EINVAL;
 912                 goto fail_inval;
 913         }
 914
 915         sfc_adapter_lock(sa);
 916
 917         if (sa->state == SFC_ADAPTER_STARTED) {
 918                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 919                 if (rc != 0)
 920                         goto fail_mac_fcntl_set;
 921         }
 922
 923         port->flow_ctrl = fcntl;
 924         port->flow_ctrl_autoneg = fc_conf->autoneg;
 925
 926         sfc_adapter_unlock(sa);
 927
 928         return 0;
 929
 930 fail_mac_fcntl_set:
 931         sfc_adapter_unlock(sa);
 932 fail_inval:
 933         SFC_ASSERT(rc > 0);
 934         return -rc;
 935 }
 936
 937 static int
 938 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 939 {
 940         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 941         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 942         boolean_t scatter_enabled;
 943         const char *error;
 944         unsigned int i;
 945
 946         for (i = 0; i < sas->rxq_count; i++) {
 947                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 948                         continue;
 949
 950                 scatter_enabled = (sas->rxq_info[i].type_flags &
 951                                    EFX_RXQ_FLAG_SCATTER);
 952
 953                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 954                                           encp->enc_rx_prefix_size,
 955                                           scatter_enabled, &error)) {
 956                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 957                                 error);
 958                         return EINVAL;
 959                 }
 960         }
 961
 962         return 0;
 963 }
 964
 965 static int
 966 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 967 {
 968         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 969         size_t pdu = EFX_MAC_PDU(mtu);
 970         size_t old_pdu;
 971         int rc;
 972
 973         sfc_log_init(sa, "mtu=%u", mtu);
 974
 975         rc = EINVAL;
 976         if (pdu < EFX_MAC_PDU_MIN) {
 977                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 978                         (unsigned int)mtu, (unsigned int)pdu,
 979                         EFX_MAC_PDU_MIN);
 980                 goto fail_inval;
 981         }
 982         if (pdu > EFX_MAC_PDU_MAX) {
 983                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 984                         (unsigned int)mtu, (unsigned int)pdu,
 985                         (unsigned int)EFX_MAC_PDU_MAX);
 986                 goto fail_inval;
 987         }
 988
 989         sfc_adapter_lock(sa);
 990
 991         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
 992         if (rc != 0)
 993                 goto fail_check_scatter;
 994
 995         if (pdu != sa->port.pdu) {
 996                 if (sa->state == SFC_ADAPTER_STARTED) {
 997                         sfc_stop(sa);
 998
 999                         old_pdu = sa->port.pdu;
1000                         sa->port.pdu = pdu;
1001                         rc = sfc_start(sa);
1002                         if (rc != 0)
1003                                 goto fail_start;
1004                 } else {
1005                         sa->port.pdu = pdu;
1006                 }
1007         }
1008
1009         /*
1010          * The driver does not use it, but other PMDs update jumbo frame
1011          * flag and max_rx_pkt_len when MTU is set.
1012          */
1013         if (mtu > RTE_ETHER_MAX_LEN) {
1014                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1015                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1016         }
1017
1018         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1019
1020         sfc_adapter_unlock(sa);
1021
1022         sfc_log_init(sa, "done");
1023         return 0;
1024
1025 fail_start:
1026         sa->port.pdu = old_pdu;
1027         if (sfc_start(sa) != 0)
1028                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1029                         "PDU max size - port is stopped",
1030                         (unsigned int)pdu, (unsigned int)old_pdu);
1031
1032 fail_check_scatter:
1033         sfc_adapter_unlock(sa);
1034
1035 fail_inval:
1036         sfc_log_init(sa, "failed %d", rc);
1037         SFC_ASSERT(rc > 0);
1038         return -rc;
1039 }
1040 static int
1041 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1042 {
1043         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1044         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1045         struct sfc_port *port = &sa->port;
1046         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1047         int rc = 0;
1048
1049         sfc_adapter_lock(sa);
1050
1051         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1052                 goto unlock;
1053
1054         /*
1055          * Copy the address to the device private data so that
1056          * it could be recalled in the case of adapter restart.
1057          */
1058         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1059
1060         /*
1061          * Neither of the two following checks can return
1062          * an error. The new MAC address is preserved in
1063          * the device private data and can be activated
1064          * on the next port start if the user prevents
1065          * isolated mode from being enabled.
1066          */
1067         if (sfc_sa2shared(sa)->isolated) {
1068                 sfc_warn(sa, "isolated mode is active on the port");
1069                 sfc_warn(sa, "will not set MAC address");
1070                 goto unlock;
1071         }
1072
1073         if (sa->state != SFC_ADAPTER_STARTED) {
1074                 sfc_notice(sa, "the port is not started");
1075                 sfc_notice(sa, "the new MAC address will be set on port start");
1076
1077                 goto unlock;
1078         }
1079
1080         if (encp->enc_allow_set_mac_with_installed_filters) {
1081                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1082                 if (rc != 0) {
1083                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1084                         goto unlock;
1085                 }
1086
1087                 /*
1088                  * Changing the MAC address by means of MCDI request
1089                  * has no effect on received traffic, therefore
1090                  * we also need to update unicast filters
1091                  */
1092                 rc = sfc_set_rx_mode_unchecked(sa);
1093                 if (rc != 0) {
1094                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1095                         /* Rollback the old address */
1096                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1097                         (void)sfc_set_rx_mode_unchecked(sa);
1098                 }
1099         } else {
1100                 sfc_warn(sa, "cannot set MAC address with filters installed");
1101                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1102                 sfc_warn(sa, "(some traffic may be dropped)");
1103
1104                 /*
1105                  * Since setting MAC address with filters installed is not
1106                  * allowed on the adapter, the new MAC address will be set
1107                  * by means of adapter restart. sfc_start() shall retrieve
1108                  * the new address from the device private data and set it.
1109                  */
1110                 sfc_stop(sa);
1111                 rc = sfc_start(sa);
1112                 if (rc != 0)
1113                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1114         }
1115
1116 unlock:
1117         if (rc != 0)
1118                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1119
1120         sfc_adapter_unlock(sa);
1121
1122         SFC_ASSERT(rc >= 0);
1123         return -rc;
1124 }
1125
1126
1127 static int
1128 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1129                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1130 {
1131         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1132         struct sfc_port *port = &sa->port;
1133         uint8_t *mc_addrs = port->mcast_addrs;
1134         int rc;
1135         unsigned int i;
1136
1137         if (sfc_sa2shared(sa)->isolated) {
1138                 sfc_err(sa, "isolated mode is active on the port");
1139                 sfc_err(sa, "will not set multicast address list");
1140                 return -ENOTSUP;
1141         }
1142
1143         if (mc_addrs == NULL)
1144                 return -ENOBUFS;
1145
1146         if (nb_mc_addr > port->max_mcast_addrs) {
1147                 sfc_err(sa, "too many multicast addresses: %u > %u",
1148                          nb_mc_addr, port->max_mcast_addrs);
1149                 return -EINVAL;
1150         }
1151
1152         for (i = 0; i < nb_mc_addr; ++i) {
1153                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1154                                  EFX_MAC_ADDR_LEN);
1155                 mc_addrs += EFX_MAC_ADDR_LEN;
1156         }
1157
1158         port->nb_mcast_addrs = nb_mc_addr;
1159
1160         if (sa->state != SFC_ADAPTER_STARTED)
1161                 return 0;
1162
1163         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1164                                         port->nb_mcast_addrs);
1165         if (rc != 0)
1166                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1167
1168         SFC_ASSERT(rc >= 0);
1169         return -rc;
1170 }
1171
1172 /*
1173  * The function is used by the secondary process as well. It must not
1174  * use any process-local pointers from the adapter data.
1175  */
1176 static void
1177 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1178                       struct rte_eth_rxq_info *qinfo)
1179 {
1180         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1181         struct sfc_rxq_info *rxq_info;
1182
1183         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1184
1185         rxq_info = &sas->rxq_info[rx_queue_id];
1186
1187         qinfo->mp = rxq_info->refill_mb_pool;
1188         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1189         qinfo->conf.rx_drop_en = 1;
1190         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1191         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1192         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1193                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1194                 qinfo->scattered_rx = 1;
1195         }
1196         qinfo->nb_desc = rxq_info->entries;
1197 }
1198
1199 /*
1200  * The function is used by the secondary process as well. It must not
1201  * use any process-local pointers from the adapter data.
1202  */
1203 static void
1204 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1205                       struct rte_eth_txq_info *qinfo)
1206 {
1207         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1208         struct sfc_txq_info *txq_info;
1209
1210         SFC_ASSERT(tx_queue_id < sas->txq_count);
1211
1212         txq_info = &sas->txq_info[tx_queue_id];
1213
1214         memset(qinfo, 0, sizeof(*qinfo));
1215
1216         qinfo->conf.offloads = txq_info->offloads;
1217         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1218         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1219         qinfo->nb_desc = txq_info->entries;
1220 }
1221
1222 /*
1223  * The function is used by the secondary process as well. It must not
1224  * use any process-local pointers from the adapter data.
1225  */
1226 static uint32_t
1227 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1228 {
1229         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1230         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1231         struct sfc_rxq_info *rxq_info;
1232
1233         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1234         rxq_info = &sas->rxq_info[rx_queue_id];
1235
1236         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1237                 return 0;
1238
1239         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1240 }
1241
1242 /*
1243  * The function is used by the secondary process as well. It must not
1244  * use any process-local pointers from the adapter data.
1245  */
1246 static int
1247 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1248 {
1249         struct sfc_dp_rxq *dp_rxq = queue;
1250         const struct sfc_dp_rx *dp_rx;
1251
1252         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1253
1254         return offset < dp_rx->qdesc_npending(dp_rxq);
1255 }
1256
1257 /*
1258  * The function is used by the secondary process as well. It must not
1259  * use any process-local pointers from the adapter data.
1260  */
1261 static int
1262 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1263 {
1264         struct sfc_dp_rxq *dp_rxq = queue;
1265         const struct sfc_dp_rx *dp_rx;
1266
1267         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1268
1269         return dp_rx->qdesc_status(dp_rxq, offset);
1270 }
1271
1272 /*
1273  * The function is used by the secondary process as well. It must not
1274  * use any process-local pointers from the adapter data.
1275  */
1276 static int
1277 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1278 {
1279         struct sfc_dp_txq *dp_txq = queue;
1280         const struct sfc_dp_tx *dp_tx;
1281
1282         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1283
1284         return dp_tx->qdesc_status(dp_txq, offset);
1285 }
1286
1287 static int
1288 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1289 {
1290         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1291         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1292         int rc;
1293
1294         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1295
1296         sfc_adapter_lock(sa);
1297
1298         rc = EINVAL;
1299         if (sa->state != SFC_ADAPTER_STARTED)
1300                 goto fail_not_started;
1301
1302         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1303                 goto fail_not_setup;
1304
1305         rc = sfc_rx_qstart(sa, rx_queue_id);
1306         if (rc != 0)
1307                 goto fail_rx_qstart;
1308
1309         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1310
1311         sfc_adapter_unlock(sa);
1312
1313         return 0;
1314
1315 fail_rx_qstart:
1316 fail_not_setup:
1317 fail_not_started:
1318         sfc_adapter_unlock(sa);
1319         SFC_ASSERT(rc > 0);
1320         return -rc;
1321 }
1322
1323 static int
1324 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1325 {
1326         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1327         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1328
1329         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1330
1331         sfc_adapter_lock(sa);
1332         sfc_rx_qstop(sa, rx_queue_id);
1333
1334         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1335
1336         sfc_adapter_unlock(sa);
1337
1338         return 0;
1339 }
1340
1341 static int
1342 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1343 {
1344         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1345         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1346         int rc;
1347
1348         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1349
1350         sfc_adapter_lock(sa);
1351
1352         rc = EINVAL;
1353         if (sa->state != SFC_ADAPTER_STARTED)
1354                 goto fail_not_started;
1355
1356         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1357                 goto fail_not_setup;
1358
1359         rc = sfc_tx_qstart(sa, tx_queue_id);
1360         if (rc != 0)
1361                 goto fail_tx_qstart;
1362
1363         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1364
1365         sfc_adapter_unlock(sa);
1366         return 0;
1367
1368 fail_tx_qstart:
1369
1370 fail_not_setup:
1371 fail_not_started:
1372         sfc_adapter_unlock(sa);
1373         SFC_ASSERT(rc > 0);
1374         return -rc;
1375 }
1376
1377 static int
1378 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1379 {
1380         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1381         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1382
1383         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1384
1385         sfc_adapter_lock(sa);
1386
1387         sfc_tx_qstop(sa, tx_queue_id);
1388
1389         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1390
1391         sfc_adapter_unlock(sa);
1392         return 0;
1393 }
1394
1395 static efx_tunnel_protocol_t
1396 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1397 {
1398         switch (rte_type) {
1399         case RTE_TUNNEL_TYPE_VXLAN:
1400                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1401         case RTE_TUNNEL_TYPE_GENEVE:
1402                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1403         default:
1404                 return EFX_TUNNEL_NPROTOS;
1405         }
1406 }
1407
1408 enum sfc_udp_tunnel_op_e {
1409         SFC_UDP_TUNNEL_ADD_PORT,
1410         SFC_UDP_TUNNEL_DEL_PORT,
1411 };
1412
1413 static int
1414 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1415                       struct rte_eth_udp_tunnel *tunnel_udp,
1416                       enum sfc_udp_tunnel_op_e op)
1417 {
1418         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1419         efx_tunnel_protocol_t tunnel_proto;
1420         int rc;
1421
1422         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1423                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1424                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1425                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1426
1427         tunnel_proto =
1428                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1429         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1430                 rc = ENOTSUP;
1431                 goto fail_bad_proto;
1432         }
1433
1434         sfc_adapter_lock(sa);
1435
1436         switch (op) {
1437         case SFC_UDP_TUNNEL_ADD_PORT:
1438                 rc = efx_tunnel_config_udp_add(sa->nic,
1439                                                tunnel_udp->udp_port,
1440                                                tunnel_proto);
1441                 break;
1442         case SFC_UDP_TUNNEL_DEL_PORT:
1443                 rc = efx_tunnel_config_udp_remove(sa->nic,
1444                                                   tunnel_udp->udp_port,
1445                                                   tunnel_proto);
1446                 break;
1447         default:
1448                 rc = EINVAL;
1449                 goto fail_bad_op;
1450         }
1451
1452         if (rc != 0)
1453                 goto fail_op;
1454
1455         if (sa->state == SFC_ADAPTER_STARTED) {
1456                 rc = efx_tunnel_reconfigure(sa->nic);
1457                 if (rc == EAGAIN) {
1458                         /*
1459                          * Configuration is accepted by FW and MC reboot
1460                          * is initiated to apply the changes. MC reboot
1461                          * will be handled in a usual way (MC reboot
1462                          * event on management event queue and adapter
1463                          * restart).
1464                          */
1465                         rc = 0;
1466                 } else if (rc != 0) {
1467                         goto fail_reconfigure;
1468                 }
1469         }
1470
1471         sfc_adapter_unlock(sa);
1472         return 0;
1473
1474 fail_reconfigure:
1475         /* Remove/restore entry since the change makes the trouble */
1476         switch (op) {
1477         case SFC_UDP_TUNNEL_ADD_PORT:
1478                 (void)efx_tunnel_config_udp_remove(sa->nic,
1479                                                    tunnel_udp->udp_port,
1480                                                    tunnel_proto);
1481                 break;
1482         case SFC_UDP_TUNNEL_DEL_PORT:
1483                 (void)efx_tunnel_config_udp_add(sa->nic,
1484                                                 tunnel_udp->udp_port,
1485                                                 tunnel_proto);
1486                 break;
1487         }
1488
1489 fail_op:
1490 fail_bad_op:
1491         sfc_adapter_unlock(sa);
1492
1493 fail_bad_proto:
1494         SFC_ASSERT(rc > 0);
1495         return -rc;
1496 }
1497
1498 static int
1499 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1500                             struct rte_eth_udp_tunnel *tunnel_udp)
1501 {
1502         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1503 }
1504
1505 static int
1506 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1507                             struct rte_eth_udp_tunnel *tunnel_udp)
1508 {
1509         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1510 }
1511
1512 /*
1513  * The function is used by the secondary process as well. It must not
1514  * use any process-local pointers from the adapter data.
1515  */
1516 static int
1517 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1518                           struct rte_eth_rss_conf *rss_conf)
1519 {
1520         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1521         struct sfc_rss *rss = &sas->rss;
1522
1523         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1524                 return -ENOTSUP;
1525
1526         /*
1527          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1528          * hence, conversion is done here to derive a correct set of ETH_RSS
1529          * flags which corresponds to the active EFX configuration stored
1530          * locally in 'sfc_adapter' and kept up-to-date
1531          */
1532         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1533         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1534         if (rss_conf->rss_key != NULL)
1535                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1536
1537         return 0;
1538 }
1539
1540 static int
1541 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1542                         struct rte_eth_rss_conf *rss_conf)
1543 {
1544         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1545         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1546         unsigned int efx_hash_types;
1547         int rc = 0;
1548
1549         if (sfc_sa2shared(sa)->isolated)
1550                 return -ENOTSUP;
1551
1552         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1553                 sfc_err(sa, "RSS is not available");
1554                 return -ENOTSUP;
1555         }
1556
1557         if (rss->channels == 0) {
1558                 sfc_err(sa, "RSS is not configured");
1559                 return -EINVAL;
1560         }
1561
1562         if ((rss_conf->rss_key != NULL) &&
1563             (rss_conf->rss_key_len != sizeof(rss->key))) {
1564                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1565                         sizeof(rss->key));
1566                 return -EINVAL;
1567         }
1568
1569         sfc_adapter_lock(sa);
1570
1571         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1572         if (rc != 0)
1573                 goto fail_rx_hf_rte_to_efx;
1574
1575         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1576                                    rss->hash_alg, efx_hash_types, B_TRUE);
1577         if (rc != 0)
1578                 goto fail_scale_mode_set;
1579
1580         if (rss_conf->rss_key != NULL) {
1581                 if (sa->state == SFC_ADAPTER_STARTED) {
1582                         rc = efx_rx_scale_key_set(sa->nic,
1583                                                   EFX_RSS_CONTEXT_DEFAULT,
1584                                                   rss_conf->rss_key,
1585                                                   sizeof(rss->key));
1586                         if (rc != 0)
1587                                 goto fail_scale_key_set;
1588                 }
1589
1590                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1591         }
1592
1593         rss->hash_types = efx_hash_types;
1594
1595         sfc_adapter_unlock(sa);
1596
1597         return 0;
1598
1599 fail_scale_key_set:
1600         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1601                                   EFX_RX_HASHALG_TOEPLITZ,
1602                                   rss->hash_types, B_TRUE) != 0)
1603                 sfc_err(sa, "failed to restore RSS mode");
1604
1605 fail_scale_mode_set:
1606 fail_rx_hf_rte_to_efx:
1607         sfc_adapter_unlock(sa);
1608         return -rc;
1609 }
1610
1611 /*
1612  * The function is used by the secondary process as well. It must not
1613  * use any process-local pointers from the adapter data.
1614  */
1615 static int
1616 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1617                        struct rte_eth_rss_reta_entry64 *reta_conf,
1618                        uint16_t reta_size)
1619 {
1620         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1621         struct sfc_rss *rss = &sas->rss;
1622         int entry;
1623
1624         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1625                 return -ENOTSUP;
1626
1627         if (rss->channels == 0)
1628                 return -EINVAL;
1629
1630         if (reta_size != EFX_RSS_TBL_SIZE)
1631                 return -EINVAL;
1632
1633         for (entry = 0; entry < reta_size; entry++) {
1634                 int grp = entry / RTE_RETA_GROUP_SIZE;
1635                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1636
1637                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1638                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1639         }
1640
1641         return 0;
1642 }
1643
1644 static int
1645 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1646                         struct rte_eth_rss_reta_entry64 *reta_conf,
1647                         uint16_t reta_size)
1648 {
1649         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1650         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1651         unsigned int *rss_tbl_new;
1652         uint16_t entry;
1653         int rc = 0;
1654
1655
1656         if (sfc_sa2shared(sa)->isolated)
1657                 return -ENOTSUP;
1658
1659         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1660                 sfc_err(sa, "RSS is not available");
1661                 return -ENOTSUP;
1662         }
1663
1664         if (rss->channels == 0) {
1665                 sfc_err(sa, "RSS is not configured");
1666                 return -EINVAL;
1667         }
1668
1669         if (reta_size != EFX_RSS_TBL_SIZE) {
1670                 sfc_err(sa, "RETA size is wrong (should be %u)",
1671                         EFX_RSS_TBL_SIZE);
1672                 return -EINVAL;
1673         }
1674
1675         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1676         if (rss_tbl_new == NULL)
1677                 return -ENOMEM;
1678
1679         sfc_adapter_lock(sa);
1680
1681         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1682
1683         for (entry = 0; entry < reta_size; entry++) {
1684                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1685                 struct rte_eth_rss_reta_entry64 *grp;
1686
1687                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1688
1689                 if (grp->mask & (1ull << grp_idx)) {
1690                         if (grp->reta[grp_idx] >= rss->channels) {
1691                                 rc = EINVAL;
1692                                 goto bad_reta_entry;
1693                         }
1694                         rss_tbl_new[entry] = grp->reta[grp_idx];
1695                 }
1696         }
1697
1698         if (sa->state == SFC_ADAPTER_STARTED) {
1699                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1700                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1701                 if (rc != 0)
1702                         goto fail_scale_tbl_set;
1703         }
1704
1705         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1706
1707 fail_scale_tbl_set:
1708 bad_reta_entry:
1709         sfc_adapter_unlock(sa);
1710
1711         rte_free(rss_tbl_new);
1712
1713         SFC_ASSERT(rc >= 0);
1714         return -rc;
1715 }
1716
1717 static int
1718 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1719                     enum rte_filter_op filter_op,
1720                     void *arg)
1721 {
1722         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1723         int rc = ENOTSUP;
1724
1725         sfc_log_init(sa, "entry");
1726
1727         switch (filter_type) {
1728         case RTE_ETH_FILTER_NONE:
1729                 sfc_err(sa, "Global filters configuration not supported");
1730                 break;
1731         case RTE_ETH_FILTER_MACVLAN:
1732                 sfc_err(sa, "MACVLAN filters not supported");
1733                 break;
1734         case RTE_ETH_FILTER_ETHERTYPE:
1735                 sfc_err(sa, "EtherType filters not supported");
1736                 break;
1737         case RTE_ETH_FILTER_FLEXIBLE:
1738                 sfc_err(sa, "Flexible filters not supported");
1739                 break;
1740         case RTE_ETH_FILTER_SYN:
1741                 sfc_err(sa, "SYN filters not supported");
1742                 break;
1743         case RTE_ETH_FILTER_NTUPLE:
1744                 sfc_err(sa, "NTUPLE filters not supported");
1745                 break;
1746         case RTE_ETH_FILTER_TUNNEL:
1747                 sfc_err(sa, "Tunnel filters not supported");
1748                 break;
1749         case RTE_ETH_FILTER_FDIR:
1750                 sfc_err(sa, "Flow Director filters not supported");
1751                 break;
1752         case RTE_ETH_FILTER_HASH:
1753                 sfc_err(sa, "Hash filters not supported");
1754                 break;
1755         case RTE_ETH_FILTER_GENERIC:
1756                 if (filter_op != RTE_ETH_FILTER_GET) {
1757                         rc = EINVAL;
1758                 } else {
1759                         *(const void **)arg = &sfc_flow_ops;
1760                         rc = 0;
1761                 }
1762                 break;
1763         default:
1764                 sfc_err(sa, "Unknown filter type %u", filter_type);
1765                 break;
1766         }
1767
1768         sfc_log_init(sa, "exit: %d", -rc);
1769         SFC_ASSERT(rc >= 0);
1770         return -rc;
1771 }
1772
1773 static int
1774 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1775 {
1776         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1777
1778         /*
1779          * If Rx datapath does not provide callback to check mempool,
1780          * all pools are supported.
1781          */
1782         if (sap->dp_rx->pool_ops_supported == NULL)
1783                 return 1;
1784
1785         return sap->dp_rx->pool_ops_supported(pool);
1786 }
1787
1788 static int
1789 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1790 {
1791         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1792         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1793         struct sfc_rxq_info *rxq_info;
1794
1795         SFC_ASSERT(queue_id < sas->rxq_count);
1796         rxq_info = &sas->rxq_info[queue_id];
1797
1798         return sap->dp_rx->intr_enable(rxq_info->dp);
1799 }
1800
1801 static int
1802 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1803 {
1804         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1805         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1806         struct sfc_rxq_info *rxq_info;
1807
1808         SFC_ASSERT(queue_id < sas->rxq_count);
1809         rxq_info = &sas->rxq_info[queue_id];
1810
1811         return sap->dp_rx->intr_disable(rxq_info->dp);
1812 }
1813
1814 static const struct eth_dev_ops sfc_eth_dev_ops = {
1815         .dev_configure                  = sfc_dev_configure,
1816         .dev_start                      = sfc_dev_start,
1817         .dev_stop                       = sfc_dev_stop,
1818         .dev_set_link_up                = sfc_dev_set_link_up,
1819         .dev_set_link_down              = sfc_dev_set_link_down,
1820         .dev_close                      = sfc_dev_close,
1821         .promiscuous_enable             = sfc_dev_promisc_enable,
1822         .promiscuous_disable            = sfc_dev_promisc_disable,
1823         .allmulticast_enable            = sfc_dev_allmulti_enable,
1824         .allmulticast_disable           = sfc_dev_allmulti_disable,
1825         .link_update                    = sfc_dev_link_update,
1826         .stats_get                      = sfc_stats_get,
1827         .stats_reset                    = sfc_stats_reset,
1828         .xstats_get                     = sfc_xstats_get,
1829         .xstats_reset                   = sfc_stats_reset,
1830         .xstats_get_names               = sfc_xstats_get_names,
1831         .dev_infos_get                  = sfc_dev_infos_get,
1832         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1833         .mtu_set                        = sfc_dev_set_mtu,
1834         .rx_queue_start                 = sfc_rx_queue_start,
1835         .rx_queue_stop                  = sfc_rx_queue_stop,
1836         .tx_queue_start                 = sfc_tx_queue_start,
1837         .tx_queue_stop                  = sfc_tx_queue_stop,
1838         .rx_queue_setup                 = sfc_rx_queue_setup,
1839         .rx_queue_release               = sfc_rx_queue_release,
1840         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1841         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1842         .tx_queue_setup                 = sfc_tx_queue_setup,
1843         .tx_queue_release               = sfc_tx_queue_release,
1844         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1845         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1846         .mac_addr_set                   = sfc_mac_addr_set,
1847         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1848         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1849         .reta_update                    = sfc_dev_rss_reta_update,
1850         .reta_query                     = sfc_dev_rss_reta_query,
1851         .rss_hash_update                = sfc_dev_rss_hash_update,
1852         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1853         .filter_ctrl                    = sfc_dev_filter_ctrl,
1854         .set_mc_addr_list               = sfc_set_mc_addr_list,
1855         .rxq_info_get                   = sfc_rx_queue_info_get,
1856         .txq_info_get                   = sfc_tx_queue_info_get,
1857         .fw_version_get                 = sfc_fw_version_get,
1858         .xstats_get_by_id               = sfc_xstats_get_by_id,
1859         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1860         .pool_ops_supported             = sfc_pool_ops_supported,
1861 };
1862
1863 /**
1864  * Duplicate a string in potentially shared memory required for
1865  * multi-process support.
1866  *
1867  * strdup() allocates from process-local heap/memory.
1868  */
1869 static char *
1870 sfc_strdup(const char *str)
1871 {
1872         size_t size;
1873         char *copy;
1874
1875         if (str == NULL)
1876                 return NULL;
1877
1878         size = strlen(str) + 1;
1879         copy = rte_malloc(__func__, size, 0);
1880         if (copy != NULL)
1881                 rte_memcpy(copy, str, size);
1882
1883         return copy;
1884 }
1885
1886 static int
1887 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1888 {
1889         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1890         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1891         const struct sfc_dp_rx *dp_rx;
1892         const struct sfc_dp_tx *dp_tx;
1893         const efx_nic_cfg_t *encp;
1894         unsigned int avail_caps = 0;
1895         const char *rx_name = NULL;
1896         const char *tx_name = NULL;
1897         int rc;
1898
1899         switch (sa->family) {
1900         case EFX_FAMILY_HUNTINGTON:
1901         case EFX_FAMILY_MEDFORD:
1902         case EFX_FAMILY_MEDFORD2:
1903                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1904                 break;
1905         default:
1906                 break;
1907         }
1908
1909         encp = efx_nic_cfg_get(sa->nic);
1910         if (encp->enc_rx_es_super_buffer_supported)
1911                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1912
1913         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1914                                 sfc_kvarg_string_handler, &rx_name);
1915         if (rc != 0)
1916                 goto fail_kvarg_rx_datapath;
1917
1918         if (rx_name != NULL) {
1919                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1920                 if (dp_rx == NULL) {
1921                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1922                         rc = ENOENT;
1923                         goto fail_dp_rx;
1924                 }
1925                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1926                         sfc_err(sa,
1927                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1928                                 rx_name);
1929                         rc = EINVAL;
1930                         goto fail_dp_rx_caps;
1931                 }
1932         } else {
1933                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1934                 if (dp_rx == NULL) {
1935                         sfc_err(sa, "Rx datapath by caps %#x not found",
1936                                 avail_caps);
1937                         rc = ENOENT;
1938                         goto fail_dp_rx;
1939                 }
1940         }
1941
1942         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1943         if (sas->dp_rx_name == NULL) {
1944                 rc = ENOMEM;
1945                 goto fail_dp_rx_name;
1946         }
1947
1948         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1949
1950         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1951                                 sfc_kvarg_string_handler, &tx_name);
1952         if (rc != 0)
1953                 goto fail_kvarg_tx_datapath;
1954
1955         if (tx_name != NULL) {
1956                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1957                 if (dp_tx == NULL) {
1958                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1959                         rc = ENOENT;
1960                         goto fail_dp_tx;
1961                 }
1962                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1963                         sfc_err(sa,
1964                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1965                                 tx_name);
1966                         rc = EINVAL;
1967                         goto fail_dp_tx_caps;
1968                 }
1969         } else {
1970                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1971                 if (dp_tx == NULL) {
1972                         sfc_err(sa, "Tx datapath by caps %#x not found",
1973                                 avail_caps);
1974                         rc = ENOENT;
1975                         goto fail_dp_tx;
1976                 }
1977         }
1978
1979         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1980         if (sas->dp_tx_name == NULL) {
1981                 rc = ENOMEM;
1982                 goto fail_dp_tx_name;
1983         }
1984
1985         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1986
1987         sa->priv.dp_rx = dp_rx;
1988         sa->priv.dp_tx = dp_tx;
1989
1990         dev->rx_pkt_burst = dp_rx->pkt_burst;
1991         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1992         dev->tx_pkt_burst = dp_tx->pkt_burst;
1993
1994         dev->rx_queue_count = sfc_rx_queue_count;
1995         dev->rx_descriptor_done = sfc_rx_descriptor_done;
1996         dev->rx_descriptor_status = sfc_rx_descriptor_status;
1997         dev->tx_descriptor_status = sfc_tx_descriptor_status;
1998         dev->dev_ops = &sfc_eth_dev_ops;
1999
2000         return 0;
2001
2002 fail_dp_tx_name:
2003 fail_dp_tx_caps:
2004 fail_dp_tx:
2005 fail_kvarg_tx_datapath:
2006         rte_free(sas->dp_rx_name);
2007         sas->dp_rx_name = NULL;
2008
2009 fail_dp_rx_name:
2010 fail_dp_rx_caps:
2011 fail_dp_rx:
2012 fail_kvarg_rx_datapath:
2013         return rc;
2014 }
2015
2016 static void
2017 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2018 {
2019         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2020         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2021
2022         dev->dev_ops = NULL;
2023         dev->tx_pkt_prepare = NULL;
2024         dev->rx_pkt_burst = NULL;
2025         dev->tx_pkt_burst = NULL;
2026
2027         rte_free(sas->dp_tx_name);
2028         sas->dp_tx_name = NULL;
2029         sa->priv.dp_tx = NULL;
2030
2031         rte_free(sas->dp_rx_name);
2032         sas->dp_rx_name = NULL;
2033         sa->priv.dp_rx = NULL;
2034 }
2035
2036 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2037         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2038         .reta_query                     = sfc_dev_rss_reta_query,
2039         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2040         .rxq_info_get                   = sfc_rx_queue_info_get,
2041         .txq_info_get                   = sfc_tx_queue_info_get,
2042 };
2043
2044 static int
2045 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2046 {
2047         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2048         struct sfc_adapter_priv *sap;
2049         const struct sfc_dp_rx *dp_rx;
2050         const struct sfc_dp_tx *dp_tx;
2051         int rc;
2052
2053         /*
2054          * Allocate process private data from heap, since it should not
2055          * be located in shared memory allocated using rte_malloc() API.
2056          */
2057         sap = calloc(1, sizeof(*sap));
2058         if (sap == NULL) {
2059                 rc = ENOMEM;
2060                 goto fail_alloc_priv;
2061         }
2062
2063         sap->logtype_main = logtype_main;
2064
2065         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2066         if (dp_rx == NULL) {
2067                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2068                         "cannot find %s Rx datapath", sas->dp_rx_name);
2069                 rc = ENOENT;
2070                 goto fail_dp_rx;
2071         }
2072         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2073                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2074                         "%s Rx datapath does not support multi-process",
2075                         sas->dp_rx_name);
2076                 rc = EINVAL;
2077                 goto fail_dp_rx_multi_process;
2078         }
2079
2080         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2081         if (dp_tx == NULL) {
2082                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2083                         "cannot find %s Tx datapath", sas->dp_tx_name);
2084                 rc = ENOENT;
2085                 goto fail_dp_tx;
2086         }
2087         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2088                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2089                         "%s Tx datapath does not support multi-process",
2090                         sas->dp_tx_name);
2091                 rc = EINVAL;
2092                 goto fail_dp_tx_multi_process;
2093         }
2094
2095         sap->dp_rx = dp_rx;
2096         sap->dp_tx = dp_tx;
2097
2098         dev->process_private = sap;
2099         dev->rx_pkt_burst = dp_rx->pkt_burst;
2100         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2101         dev->tx_pkt_burst = dp_tx->pkt_burst;
2102         dev->rx_queue_count = sfc_rx_queue_count;
2103         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2104         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2105         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2106         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2107
2108         return 0;
2109
2110 fail_dp_tx_multi_process:
2111 fail_dp_tx:
2112 fail_dp_rx_multi_process:
2113 fail_dp_rx:
2114         free(sap);
2115
2116 fail_alloc_priv:
2117         return rc;
2118 }
2119
2120 static void
2121 sfc_register_dp(void)
2122 {
2123         /* Register once */
2124         if (TAILQ_EMPTY(&sfc_dp_head)) {
2125                 /* Prefer EF10 datapath */
2126                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2127                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2128                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2129
2130                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2131                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2132                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2133         }
2134 }
2135
2136 static int
2137 sfc_eth_dev_init(struct rte_eth_dev *dev)
2138 {
2139         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2140         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2141         uint32_t logtype_main;
2142         struct sfc_adapter *sa;
2143         int rc;
2144         const efx_nic_cfg_t *encp;
2145         const struct rte_ether_addr *from;
2146         int ret;
2147
2148         sfc_register_dp();
2149
2150         logtype_main = sfc_register_logtype(&pci_dev->addr,
2151                                             SFC_LOGTYPE_MAIN_STR,
2152                                             RTE_LOG_NOTICE);
2153
2154         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2155                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2156
2157         /* Required for logging */
2158         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2159                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2160                         pci_dev->addr.domain, pci_dev->addr.bus,
2161                         pci_dev->addr.devid, pci_dev->addr.function,
2162                         dev->data->port_id);
2163         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2164                 SFC_GENERIC_LOG(ERR,
2165                         "reserved log prefix is too short for " PCI_PRI_FMT,
2166                         pci_dev->addr.domain, pci_dev->addr.bus,
2167                         pci_dev->addr.devid, pci_dev->addr.function);
2168                 return -EINVAL;
2169         }
2170         sas->pci_addr = pci_dev->addr;
2171         sas->port_id = dev->data->port_id;
2172
2173         /*
2174          * Allocate process private data from heap, since it should not
2175          * be located in shared memory allocated using rte_malloc() API.
2176          */
2177         sa = calloc(1, sizeof(*sa));
2178         if (sa == NULL) {
2179                 rc = ENOMEM;
2180                 goto fail_alloc_sa;
2181         }
2182
2183         dev->process_private = sa;
2184
2185         /* Required for logging */
2186         sa->priv.shared = sas;
2187         sa->priv.logtype_main = logtype_main;
2188
2189         sa->eth_dev = dev;
2190
2191         /* Copy PCI device info to the dev->data */
2192         rte_eth_copy_pci_info(dev, pci_dev);
2193
2194         rc = sfc_kvargs_parse(sa);
2195         if (rc != 0)
2196                 goto fail_kvargs_parse;
2197
2198         sfc_log_init(sa, "entry");
2199
2200         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2201         if (dev->data->mac_addrs == NULL) {
2202                 rc = ENOMEM;
2203                 goto fail_mac_addrs;
2204         }
2205
2206         sfc_adapter_lock_init(sa);
2207         sfc_adapter_lock(sa);
2208
2209         sfc_log_init(sa, "probing");
2210         rc = sfc_probe(sa);
2211         if (rc != 0)
2212                 goto fail_probe;
2213
2214         sfc_log_init(sa, "set device ops");
2215         rc = sfc_eth_dev_set_ops(dev);
2216         if (rc != 0)
2217                 goto fail_set_ops;
2218
2219         sfc_log_init(sa, "attaching");
2220         rc = sfc_attach(sa);
2221         if (rc != 0)
2222                 goto fail_attach;
2223
2224         encp = efx_nic_cfg_get(sa->nic);
2225
2226         /*
2227          * The arguments are really reverse order in comparison to
2228          * Linux kernel. Copy from NIC config to Ethernet device data.
2229          */
2230         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2231         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2232
2233         sfc_adapter_unlock(sa);
2234
2235         sfc_log_init(sa, "done");
2236         return 0;
2237
2238 fail_attach:
2239         sfc_eth_dev_clear_ops(dev);
2240
2241 fail_set_ops:
2242         sfc_unprobe(sa);
2243
2244 fail_probe:
2245         sfc_adapter_unlock(sa);
2246         sfc_adapter_lock_fini(sa);
2247         rte_free(dev->data->mac_addrs);
2248         dev->data->mac_addrs = NULL;
2249
2250 fail_mac_addrs:
2251         sfc_kvargs_cleanup(sa);
2252
2253 fail_kvargs_parse:
2254         sfc_log_init(sa, "failed %d", rc);
2255         dev->process_private = NULL;
2256         free(sa);
2257
2258 fail_alloc_sa:
2259         SFC_ASSERT(rc > 0);
2260         return -rc;
2261 }
2262
2263 static int
2264 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2265 {
2266         sfc_dev_close(dev);
2267
2268         return 0;
2269 }
2270
2271 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2272         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2273         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2274         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2275         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2276         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2277         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2278         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2279         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2280         { .vendor_id = 0 /* sentinel */ }
2281 };
2282
2283 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2284         struct rte_pci_device *pci_dev)
2285 {
2286         return rte_eth_dev_pci_generic_probe(pci_dev,
2287                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2288 }
2289
2290 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2291 {
2292         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2293 }
2294
2295 static struct rte_pci_driver sfc_efx_pmd = {
2296         .id_table = pci_id_sfc_efx_map,
2297         .drv_flags =
2298                 RTE_PCI_DRV_INTR_LSC |
2299                 RTE_PCI_DRV_NEED_MAPPING,
2300         .probe = sfc_eth_dev_pci_probe,
2301         .remove = sfc_eth_dev_pci_remove,
2302 };
2303
2304 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2305 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2306 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2307 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2308         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2309         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2310         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2311         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2312         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2313         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2314
2315 RTE_INIT(sfc_driver_register_logtype)
2316 {
2317         int ret;
2318
2319         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2320                                                    RTE_LOG_NOTICE);
2321         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2322 }