drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
   4  * All rights reserved.
   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 static int
  38 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  39 {
  40         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  41         efx_nic_fw_info_t enfi;
  42         int ret;
  43         int rc;
  44
  45         /*
  46          * Return value of the callback is likely supposed to be
  47          * equal to or greater than 0, nevertheless, if an error
  48          * occurs, it will be desirable to pass it to the caller
  49          */
  50         if ((fw_version == NULL) || (fw_size == 0))
  51                 return -EINVAL;
  52
  53         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  54         if (rc != 0)
  55                 return -rc;
  56
  57         ret = snprintf(fw_version, fw_size,
  58                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  59                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  60                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  61         if (ret < 0)
  62                 return ret;
  63
  64         if (enfi.enfi_dpcpu_fw_ids_valid) {
  65                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  66                 int ret_extra;
  67
  68                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  69                                      fw_size - dpcpu_fw_ids_offset,
  70                                      " rx%" PRIx16 " tx%" PRIx16,
  71                                      enfi.enfi_rx_dpcpu_fw_id,
  72                                      enfi.enfi_tx_dpcpu_fw_id);
  73                 if (ret_extra < 0)
  74                         return ret_extra;
  75
  76                 ret += ret_extra;
  77         }
  78
  79         if (fw_size < (size_t)(++ret))
  80                 return ret;
  81         else
  82                 return 0;
  83 }
  84
  85 static void
  86 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  87 {
  88         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  89         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  90         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  91         struct sfc_rss *rss = &sas->rss;
  92         uint64_t txq_offloads_def = 0;
  93
  94         sfc_log_init(sa, "entry");
  95
  96         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
  97         dev_info->max_mtu = EFX_MAC_SDU_MAX;
  98
  99         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 100
 101         /* Autonegotiation may be disabled */
 102         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 103         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 104                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 105         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 106                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 107         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 108                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 110                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 112                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 114                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 115
 116         dev_info->max_rx_queues = sa->rxq_max;
 117         dev_info->max_tx_queues = sa->txq_max;
 118
 119         /* By default packets are dropped if no descriptors are available */
 120         dev_info->default_rxconf.rx_drop_en = 1;
 121
 122         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 123
 124         /*
 125          * rx_offload_capa includes both device and queue offloads since
 126          * the latter may be requested on a per device basis which makes
 127          * sense when some offloads are needed to be set on all queues.
 128          */
 129         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 130                                     dev_info->rx_queue_offload_capa;
 131
 132         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 133
 134         /*
 135          * tx_offload_capa includes both device and queue offloads since
 136          * the latter may be requested on a per device basis which makes
 137          * sense when some offloads are needed to be set on all queues.
 138          */
 139         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 140                                     dev_info->tx_queue_offload_capa;
 141
 142         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 143                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 144
 145         dev_info->default_txconf.offloads |= txq_offloads_def;
 146
 147         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 148                 uint64_t rte_hf = 0;
 149                 unsigned int i;
 150
 151                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 152                         rte_hf |= rss->hf_map[i].rte;
 153
 154                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 155                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 156                 dev_info->flow_type_rss_offloads = rte_hf;
 157         }
 158
 159         /* Initialize to hardware limits */
 160         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 161         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 162         /* The RXQ hardware requires that the descriptor count is a power
 163          * of 2, but rx_desc_lim cannot properly describe that constraint.
 164          */
 165         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 166
 167         /* Initialize to hardware limits */
 168         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 169         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 170         /*
 171          * The TXQ hardware requires that the descriptor count is a power
 172          * of 2, but tx_desc_lim cannot properly describe that constraint
 173          */
 174         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 175
 176         if (sap->dp_rx->get_dev_info != NULL)
 177                 sap->dp_rx->get_dev_info(dev_info);
 178         if (sap->dp_tx->get_dev_info != NULL)
 179                 sap->dp_tx->get_dev_info(dev_info);
 180
 181         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 182                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 183 }
 184
 185 static const uint32_t *
 186 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 187 {
 188         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 189
 190         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 191 }
 192
 193 static int
 194 sfc_dev_configure(struct rte_eth_dev *dev)
 195 {
 196         struct rte_eth_dev_data *dev_data = dev->data;
 197         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 198         int rc;
 199
 200         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 201                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 202
 203         sfc_adapter_lock(sa);
 204         switch (sa->state) {
 205         case SFC_ADAPTER_CONFIGURED:
 206                 /* FALLTHROUGH */
 207         case SFC_ADAPTER_INITIALIZED:
 208                 rc = sfc_configure(sa);
 209                 break;
 210         default:
 211                 sfc_err(sa, "unexpected adapter state %u to configure",
 212                         sa->state);
 213                 rc = EINVAL;
 214                 break;
 215         }
 216         sfc_adapter_unlock(sa);
 217
 218         sfc_log_init(sa, "done %d", rc);
 219         SFC_ASSERT(rc >= 0);
 220         return -rc;
 221 }
 222
 223 static int
 224 sfc_dev_start(struct rte_eth_dev *dev)
 225 {
 226         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 227         int rc;
 228
 229         sfc_log_init(sa, "entry");
 230
 231         sfc_adapter_lock(sa);
 232         rc = sfc_start(sa);
 233         sfc_adapter_unlock(sa);
 234
 235         sfc_log_init(sa, "done %d", rc);
 236         SFC_ASSERT(rc >= 0);
 237         return -rc;
 238 }
 239
 240 static int
 241 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 242 {
 243         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 244         struct rte_eth_link current_link;
 245         int ret;
 246
 247         sfc_log_init(sa, "entry");
 248
 249         if (sa->state != SFC_ADAPTER_STARTED) {
 250                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 251         } else if (wait_to_complete) {
 252                 efx_link_mode_t link_mode;
 253
 254                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 255                         link_mode = EFX_LINK_UNKNOWN;
 256                 sfc_port_link_mode_to_info(link_mode, &current_link);
 257
 258         } else {
 259                 sfc_ev_mgmt_qpoll(sa);
 260                 rte_eth_linkstatus_get(dev, &current_link);
 261         }
 262
 263         ret = rte_eth_linkstatus_set(dev, &current_link);
 264         if (ret == 0)
 265                 sfc_notice(sa, "Link status is %s",
 266                            current_link.link_status ? "UP" : "DOWN");
 267
 268         return ret;
 269 }
 270
 271 static void
 272 sfc_dev_stop(struct rte_eth_dev *dev)
 273 {
 274         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 275
 276         sfc_log_init(sa, "entry");
 277
 278         sfc_adapter_lock(sa);
 279         sfc_stop(sa);
 280         sfc_adapter_unlock(sa);
 281
 282         sfc_log_init(sa, "done");
 283 }
 284
 285 static int
 286 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 287 {
 288         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 289         int rc;
 290
 291         sfc_log_init(sa, "entry");
 292
 293         sfc_adapter_lock(sa);
 294         rc = sfc_start(sa);
 295         sfc_adapter_unlock(sa);
 296
 297         SFC_ASSERT(rc >= 0);
 298         return -rc;
 299 }
 300
 301 static int
 302 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 303 {
 304         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 305
 306         sfc_log_init(sa, "entry");
 307
 308         sfc_adapter_lock(sa);
 309         sfc_stop(sa);
 310         sfc_adapter_unlock(sa);
 311
 312         return 0;
 313 }
 314
 315 static void
 316 sfc_dev_close(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         switch (sa->state) {
 324         case SFC_ADAPTER_STARTED:
 325                 sfc_stop(sa);
 326                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 327                 /* FALLTHROUGH */
 328         case SFC_ADAPTER_CONFIGURED:
 329                 sfc_close(sa);
 330                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 331                 /* FALLTHROUGH */
 332         case SFC_ADAPTER_INITIALIZED:
 333                 break;
 334         default:
 335                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 336                 break;
 337         }
 338         sfc_adapter_unlock(sa);
 339
 340         sfc_log_init(sa, "done");
 341 }
 342
 343 static void
 344 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 345                    boolean_t enabled)
 346 {
 347         struct sfc_port *port;
 348         boolean_t *toggle;
 349         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 350         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 351         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 352
 353         sfc_adapter_lock(sa);
 354
 355         port = &sa->port;
 356         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 357
 358         if (*toggle != enabled) {
 359                 *toggle = enabled;
 360
 361                 if (sfc_sa2shared(sa)->isolated) {
 362                         sfc_warn(sa, "isolated mode is active on the port");
 363                         sfc_warn(sa, "the change is to be applied on the next "
 364                                      "start provided that isolated mode is "
 365                                      "disabled prior the next start");
 366                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 367                            (sfc_set_rx_mode(sa) != 0)) {
 368                         *toggle = !(enabled);
 369                         sfc_warn(sa, "Failed to %s %s mode",
 370                                  ((enabled) ? "enable" : "disable"), desc);
 371                 }
 372         }
 373
 374         sfc_adapter_unlock(sa);
 375 }
 376
 377 static void
 378 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 379 {
 380         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 381 }
 382
 383 static void
 384 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 385 {
 386         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 387 }
 388
 389 static void
 390 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 391 {
 392         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 393 }
 394
 395 static void
 396 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 397 {
 398         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 399 }
 400
 401 static int
 402 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 403                    uint16_t nb_rx_desc, unsigned int socket_id,
 404                    const struct rte_eth_rxconf *rx_conf,
 405                    struct rte_mempool *mb_pool)
 406 {
 407         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 408         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 409         int rc;
 410
 411         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 412                      rx_queue_id, nb_rx_desc, socket_id);
 413
 414         sfc_adapter_lock(sa);
 415
 416         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 417                           rx_conf, mb_pool);
 418         if (rc != 0)
 419                 goto fail_rx_qinit;
 420
 421         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 422
 423         sfc_adapter_unlock(sa);
 424
 425         return 0;
 426
 427 fail_rx_qinit:
 428         sfc_adapter_unlock(sa);
 429         SFC_ASSERT(rc > 0);
 430         return -rc;
 431 }
 432
 433 static void
 434 sfc_rx_queue_release(void *queue)
 435 {
 436         struct sfc_dp_rxq *dp_rxq = queue;
 437         struct sfc_rxq *rxq;
 438         struct sfc_adapter *sa;
 439         unsigned int sw_index;
 440
 441         if (dp_rxq == NULL)
 442                 return;
 443
 444         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 445         sa = rxq->evq->sa;
 446         sfc_adapter_lock(sa);
 447
 448         sw_index = dp_rxq->dpq.queue_id;
 449
 450         sfc_log_init(sa, "RxQ=%u", sw_index);
 451
 452         sfc_rx_qfini(sa, sw_index);
 453
 454         sfc_adapter_unlock(sa);
 455 }
 456
 457 static int
 458 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 459                    uint16_t nb_tx_desc, unsigned int socket_id,
 460                    const struct rte_eth_txconf *tx_conf)
 461 {
 462         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 463         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 464         int rc;
 465
 466         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 467                      tx_queue_id, nb_tx_desc, socket_id);
 468
 469         sfc_adapter_lock(sa);
 470
 471         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 472         if (rc != 0)
 473                 goto fail_tx_qinit;
 474
 475         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
 476
 477         sfc_adapter_unlock(sa);
 478         return 0;
 479
 480 fail_tx_qinit:
 481         sfc_adapter_unlock(sa);
 482         SFC_ASSERT(rc > 0);
 483         return -rc;
 484 }
 485
 486 static void
 487 sfc_tx_queue_release(void *queue)
 488 {
 489         struct sfc_dp_txq *dp_txq = queue;
 490         struct sfc_txq *txq;
 491         unsigned int sw_index;
 492         struct sfc_adapter *sa;
 493
 494         if (dp_txq == NULL)
 495                 return;
 496
 497         txq = sfc_txq_by_dp_txq(dp_txq);
 498         sw_index = dp_txq->dpq.queue_id;
 499
 500         SFC_ASSERT(txq->evq != NULL);
 501         sa = txq->evq->sa;
 502
 503         sfc_log_init(sa, "TxQ = %u", sw_index);
 504
 505         sfc_adapter_lock(sa);
 506
 507         sfc_tx_qfini(sa, sw_index);
 508
 509         sfc_adapter_unlock(sa);
 510 }
 511
 512 /*
 513  * Some statistics are computed as A - B where A and B each increase
 514  * monotonically with some hardware counter(s) and the counters are read
 515  * asynchronously.
 516  *
 517  * If packet X is counted in A, but not counted in B yet, computed value is
 518  * greater than real.
 519  *
 520  * If packet X is not counted in A at the moment of reading the counter,
 521  * but counted in B at the moment of reading the counter, computed value
 522  * is less than real.
 523  *
 524  * However, counter which grows backward is worse evil than slightly wrong
 525  * value. So, let's try to guarantee that it never happens except may be
 526  * the case when the MAC stats are zeroed as a result of a NIC reset.
 527  */
 528 static void
 529 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 530 {
 531         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 532                 *stat = newval;
 533 }
 534
 535 static int
 536 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 537 {
 538         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 539         struct sfc_port *port = &sa->port;
 540         uint64_t *mac_stats;
 541         int ret;
 542
 543         rte_spinlock_lock(&port->mac_stats_lock);
 544
 545         ret = sfc_port_update_mac_stats(sa);
 546         if (ret != 0)
 547                 goto unlock;
 548
 549         mac_stats = port->mac_stats_buf;
 550
 551         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 552                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 553                 stats->ipackets =
 554                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 555                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 556                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 557                 stats->opackets =
 558                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 559                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 560                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 561                 stats->ibytes =
 562                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 563                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 564                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 565                 stats->obytes =
 566                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 567                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 568                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 569                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 570                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 571         } else {
 572                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 573                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 574                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 575                 /*
 576                  * Take into account stats which are whenever supported
 577                  * on EF10. If some stat is not supported by current
 578                  * firmware variant or HW revision, it is guaranteed
 579                  * to be zero in mac_stats.
 580                  */
 581                 stats->imissed =
 582                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 583                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 584                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 585                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 586                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 587                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 588                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 589                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 590                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 591                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 592                 stats->ierrors =
 593                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 594                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 595                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 596                 /* no oerrors counters supported on EF10 */
 597
 598                 /* Exclude missed, errors and pauses from Rx packets */
 599                 sfc_update_diff_stat(&port->ipackets,
 600                         mac_stats[EFX_MAC_RX_PKTS] -
 601                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 602                         stats->imissed - stats->ierrors);
 603                 stats->ipackets = port->ipackets;
 604         }
 605
 606 unlock:
 607         rte_spinlock_unlock(&port->mac_stats_lock);
 608         SFC_ASSERT(ret >= 0);
 609         return -ret;
 610 }
 611
 612 static void
 613 sfc_stats_reset(struct rte_eth_dev *dev)
 614 {
 615         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 616         struct sfc_port *port = &sa->port;
 617         int rc;
 618
 619         if (sa->state != SFC_ADAPTER_STARTED) {
 620                 /*
 621                  * The operation cannot be done if port is not started; it
 622                  * will be scheduled to be done during the next port start
 623                  */
 624                 port->mac_stats_reset_pending = B_TRUE;
 625                 return;
 626         }
 627
 628         rc = sfc_port_reset_mac_stats(sa);
 629         if (rc != 0)
 630                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 631 }
 632
 633 static int
 634 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 635                unsigned int xstats_count)
 636 {
 637         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 638         struct sfc_port *port = &sa->port;
 639         uint64_t *mac_stats;
 640         int rc;
 641         unsigned int i;
 642         int nstats = 0;
 643
 644         rte_spinlock_lock(&port->mac_stats_lock);
 645
 646         rc = sfc_port_update_mac_stats(sa);
 647         if (rc != 0) {
 648                 SFC_ASSERT(rc > 0);
 649                 nstats = -rc;
 650                 goto unlock;
 651         }
 652
 653         mac_stats = port->mac_stats_buf;
 654
 655         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 656                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 657                         if (xstats != NULL && nstats < (int)xstats_count) {
 658                                 xstats[nstats].id = nstats;
 659                                 xstats[nstats].value = mac_stats[i];
 660                         }
 661                         nstats++;
 662                 }
 663         }
 664
 665 unlock:
 666         rte_spinlock_unlock(&port->mac_stats_lock);
 667
 668         return nstats;
 669 }
 670
 671 static int
 672 sfc_xstats_get_names(struct rte_eth_dev *dev,
 673                      struct rte_eth_xstat_name *xstats_names,
 674                      unsigned int xstats_count)
 675 {
 676         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 677         struct sfc_port *port = &sa->port;
 678         unsigned int i;
 679         unsigned int nstats = 0;
 680
 681         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 682                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 683                         if (xstats_names != NULL && nstats < xstats_count)
 684                                 strlcpy(xstats_names[nstats].name,
 685                                         efx_mac_stat_name(sa->nic, i),
 686                                         sizeof(xstats_names[0].name));
 687                         nstats++;
 688                 }
 689         }
 690
 691         return nstats;
 692 }
 693
 694 static int
 695 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 696                      uint64_t *values, unsigned int n)
 697 {
 698         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 699         struct sfc_port *port = &sa->port;
 700         uint64_t *mac_stats;
 701         unsigned int nb_supported = 0;
 702         unsigned int nb_written = 0;
 703         unsigned int i;
 704         int ret;
 705         int rc;
 706
 707         if (unlikely(values == NULL) ||
 708             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 709                 return port->mac_stats_nb_supported;
 710
 711         rte_spinlock_lock(&port->mac_stats_lock);
 712
 713         rc = sfc_port_update_mac_stats(sa);
 714         if (rc != 0) {
 715                 SFC_ASSERT(rc > 0);
 716                 ret = -rc;
 717                 goto unlock;
 718         }
 719
 720         mac_stats = port->mac_stats_buf;
 721
 722         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 723                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 724                         continue;
 725
 726                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 727                         values[nb_written++] = mac_stats[i];
 728
 729                 ++nb_supported;
 730         }
 731
 732         ret = nb_written;
 733
 734 unlock:
 735         rte_spinlock_unlock(&port->mac_stats_lock);
 736
 737         return ret;
 738 }
 739
 740 static int
 741 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 742                            struct rte_eth_xstat_name *xstats_names,
 743                            const uint64_t *ids, unsigned int size)
 744 {
 745         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 746         struct sfc_port *port = &sa->port;
 747         unsigned int nb_supported = 0;
 748         unsigned int nb_written = 0;
 749         unsigned int i;
 750
 751         if (unlikely(xstats_names == NULL) ||
 752             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 753                 return port->mac_stats_nb_supported;
 754
 755         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 756                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 757                         continue;
 758
 759                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 760                         char *name = xstats_names[nb_written++].name;
 761
 762                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 763                                 sizeof(xstats_names[0].name));
 764                 }
 765
 766                 ++nb_supported;
 767         }
 768
 769         return nb_written;
 770 }
 771
 772 static int
 773 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 774 {
 775         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 776         unsigned int wanted_fc, link_fc;
 777
 778         memset(fc_conf, 0, sizeof(*fc_conf));
 779
 780         sfc_adapter_lock(sa);
 781
 782         if (sa->state == SFC_ADAPTER_STARTED)
 783                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 784         else
 785                 link_fc = sa->port.flow_ctrl;
 786
 787         switch (link_fc) {
 788         case 0:
 789                 fc_conf->mode = RTE_FC_NONE;
 790                 break;
 791         case EFX_FCNTL_RESPOND:
 792                 fc_conf->mode = RTE_FC_RX_PAUSE;
 793                 break;
 794         case EFX_FCNTL_GENERATE:
 795                 fc_conf->mode = RTE_FC_TX_PAUSE;
 796                 break;
 797         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 798                 fc_conf->mode = RTE_FC_FULL;
 799                 break;
 800         default:
 801                 sfc_err(sa, "%s: unexpected flow control value %#x",
 802                         __func__, link_fc);
 803         }
 804
 805         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 806
 807         sfc_adapter_unlock(sa);
 808
 809         return 0;
 810 }
 811
 812 static int
 813 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 814 {
 815         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 816         struct sfc_port *port = &sa->port;
 817         unsigned int fcntl;
 818         int rc;
 819
 820         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 821             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 822             fc_conf->mac_ctrl_frame_fwd != 0) {
 823                 sfc_err(sa, "unsupported flow control settings specified");
 824                 rc = EINVAL;
 825                 goto fail_inval;
 826         }
 827
 828         switch (fc_conf->mode) {
 829         case RTE_FC_NONE:
 830                 fcntl = 0;
 831                 break;
 832         case RTE_FC_RX_PAUSE:
 833                 fcntl = EFX_FCNTL_RESPOND;
 834                 break;
 835         case RTE_FC_TX_PAUSE:
 836                 fcntl = EFX_FCNTL_GENERATE;
 837                 break;
 838         case RTE_FC_FULL:
 839                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 840                 break;
 841         default:
 842                 rc = EINVAL;
 843                 goto fail_inval;
 844         }
 845
 846         sfc_adapter_lock(sa);
 847
 848         if (sa->state == SFC_ADAPTER_STARTED) {
 849                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 850                 if (rc != 0)
 851                         goto fail_mac_fcntl_set;
 852         }
 853
 854         port->flow_ctrl = fcntl;
 855         port->flow_ctrl_autoneg = fc_conf->autoneg;
 856
 857         sfc_adapter_unlock(sa);
 858
 859         return 0;
 860
 861 fail_mac_fcntl_set:
 862         sfc_adapter_unlock(sa);
 863 fail_inval:
 864         SFC_ASSERT(rc > 0);
 865         return -rc;
 866 }
 867
 868 static int
 869 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 870 {
 871         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 872         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 873         boolean_t scatter_enabled;
 874         const char *error;
 875         unsigned int i;
 876
 877         for (i = 0; i < sas->rxq_count; i++) {
 878                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 879                         continue;
 880
 881                 scatter_enabled = (sas->rxq_info[i].type_flags &
 882                                    EFX_RXQ_FLAG_SCATTER);
 883
 884                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 885                                           encp->enc_rx_prefix_size,
 886                                           scatter_enabled, &error)) {
 887                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 888                                 error);
 889                         return EINVAL;
 890                 }
 891         }
 892
 893         return 0;
 894 }
 895
 896 static int
 897 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 898 {
 899         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 900         size_t pdu = EFX_MAC_PDU(mtu);
 901         size_t old_pdu;
 902         int rc;
 903
 904         sfc_log_init(sa, "mtu=%u", mtu);
 905
 906         rc = EINVAL;
 907         if (pdu < EFX_MAC_PDU_MIN) {
 908                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 909                         (unsigned int)mtu, (unsigned int)pdu,
 910                         EFX_MAC_PDU_MIN);
 911                 goto fail_inval;
 912         }
 913         if (pdu > EFX_MAC_PDU_MAX) {
 914                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 915                         (unsigned int)mtu, (unsigned int)pdu,
 916                         EFX_MAC_PDU_MAX);
 917                 goto fail_inval;
 918         }
 919
 920         sfc_adapter_lock(sa);
 921
 922         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
 923         if (rc != 0)
 924                 goto fail_check_scatter;
 925
 926         if (pdu != sa->port.pdu) {
 927                 if (sa->state == SFC_ADAPTER_STARTED) {
 928                         sfc_stop(sa);
 929
 930                         old_pdu = sa->port.pdu;
 931                         sa->port.pdu = pdu;
 932                         rc = sfc_start(sa);
 933                         if (rc != 0)
 934                                 goto fail_start;
 935                 } else {
 936                         sa->port.pdu = pdu;
 937                 }
 938         }
 939
 940         /*
 941          * The driver does not use it, but other PMDs update jumbo frame
 942          * flag and max_rx_pkt_len when MTU is set.
 943          */
 944         if (mtu > RTE_ETHER_MAX_LEN) {
 945                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 946                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 947         }
 948
 949         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 950
 951         sfc_adapter_unlock(sa);
 952
 953         sfc_log_init(sa, "done");
 954         return 0;
 955
 956 fail_start:
 957         sa->port.pdu = old_pdu;
 958         if (sfc_start(sa) != 0)
 959                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 960                         "PDU max size - port is stopped",
 961                         (unsigned int)pdu, (unsigned int)old_pdu);
 962
 963 fail_check_scatter:
 964         sfc_adapter_unlock(sa);
 965
 966 fail_inval:
 967         sfc_log_init(sa, "failed %d", rc);
 968         SFC_ASSERT(rc > 0);
 969         return -rc;
 970 }
 971 static int
 972 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
 973 {
 974         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 975         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 976         struct sfc_port *port = &sa->port;
 977         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
 978         int rc = 0;
 979
 980         sfc_adapter_lock(sa);
 981
 982         /*
 983          * Copy the address to the device private data so that
 984          * it could be recalled in the case of adapter restart.
 985          */
 986         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
 987
 988         /*
 989          * Neither of the two following checks can return
 990          * an error. The new MAC address is preserved in
 991          * the device private data and can be activated
 992          * on the next port start if the user prevents
 993          * isolated mode from being enabled.
 994          */
 995         if (sfc_sa2shared(sa)->isolated) {
 996                 sfc_warn(sa, "isolated mode is active on the port");
 997                 sfc_warn(sa, "will not set MAC address");
 998                 goto unlock;
 999         }
1000
1001         if (sa->state != SFC_ADAPTER_STARTED) {
1002                 sfc_notice(sa, "the port is not started");
1003                 sfc_notice(sa, "the new MAC address will be set on port start");
1004
1005                 goto unlock;
1006         }
1007
1008         if (encp->enc_allow_set_mac_with_installed_filters) {
1009                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1010                 if (rc != 0) {
1011                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1012                         goto unlock;
1013                 }
1014
1015                 /*
1016                  * Changing the MAC address by means of MCDI request
1017                  * has no effect on received traffic, therefore
1018                  * we also need to update unicast filters
1019                  */
1020                 rc = sfc_set_rx_mode(sa);
1021                 if (rc != 0) {
1022                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1023                         /* Rollback the old address */
1024                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1025                         (void)sfc_set_rx_mode(sa);
1026                 }
1027         } else {
1028                 sfc_warn(sa, "cannot set MAC address with filters installed");
1029                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1030                 sfc_warn(sa, "(some traffic may be dropped)");
1031
1032                 /*
1033                  * Since setting MAC address with filters installed is not
1034                  * allowed on the adapter, the new MAC address will be set
1035                  * by means of adapter restart. sfc_start() shall retrieve
1036                  * the new address from the device private data and set it.
1037                  */
1038                 sfc_stop(sa);
1039                 rc = sfc_start(sa);
1040                 if (rc != 0)
1041                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1042         }
1043
1044 unlock:
1045         if (rc != 0)
1046                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1047
1048         sfc_adapter_unlock(sa);
1049
1050         SFC_ASSERT(rc >= 0);
1051         return -rc;
1052 }
1053
1054
1055 static int
1056 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1057                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1058 {
1059         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1060         struct sfc_port *port = &sa->port;
1061         uint8_t *mc_addrs = port->mcast_addrs;
1062         int rc;
1063         unsigned int i;
1064
1065         if (sfc_sa2shared(sa)->isolated) {
1066                 sfc_err(sa, "isolated mode is active on the port");
1067                 sfc_err(sa, "will not set multicast address list");
1068                 return -ENOTSUP;
1069         }
1070
1071         if (mc_addrs == NULL)
1072                 return -ENOBUFS;
1073
1074         if (nb_mc_addr > port->max_mcast_addrs) {
1075                 sfc_err(sa, "too many multicast addresses: %u > %u",
1076                          nb_mc_addr, port->max_mcast_addrs);
1077                 return -EINVAL;
1078         }
1079
1080         for (i = 0; i < nb_mc_addr; ++i) {
1081                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1082                                  EFX_MAC_ADDR_LEN);
1083                 mc_addrs += EFX_MAC_ADDR_LEN;
1084         }
1085
1086         port->nb_mcast_addrs = nb_mc_addr;
1087
1088         if (sa->state != SFC_ADAPTER_STARTED)
1089                 return 0;
1090
1091         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1092                                         port->nb_mcast_addrs);
1093         if (rc != 0)
1094                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1095
1096         SFC_ASSERT(rc >= 0);
1097         return -rc;
1098 }
1099
1100 /*
1101  * The function is used by the secondary process as well. It must not
1102  * use any process-local pointers from the adapter data.
1103  */
1104 static void
1105 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1106                       struct rte_eth_rxq_info *qinfo)
1107 {
1108         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1109         struct sfc_rxq_info *rxq_info;
1110
1111         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1112
1113         rxq_info = &sas->rxq_info[rx_queue_id];
1114
1115         qinfo->mp = rxq_info->refill_mb_pool;
1116         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1117         qinfo->conf.rx_drop_en = 1;
1118         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1119         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1120         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1121                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1122                 qinfo->scattered_rx = 1;
1123         }
1124         qinfo->nb_desc = rxq_info->entries;
1125 }
1126
1127 /*
1128  * The function is used by the secondary process as well. It must not
1129  * use any process-local pointers from the adapter data.
1130  */
1131 static void
1132 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1133                       struct rte_eth_txq_info *qinfo)
1134 {
1135         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1136         struct sfc_txq_info *txq_info;
1137
1138         SFC_ASSERT(tx_queue_id < sas->txq_count);
1139
1140         txq_info = &sas->txq_info[tx_queue_id];
1141
1142         memset(qinfo, 0, sizeof(*qinfo));
1143
1144         qinfo->conf.offloads = txq_info->offloads;
1145         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1146         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1147         qinfo->nb_desc = txq_info->entries;
1148 }
1149
1150 /*
1151  * The function is used by the secondary process as well. It must not
1152  * use any process-local pointers from the adapter data.
1153  */
1154 static uint32_t
1155 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1156 {
1157         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1158         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1159         struct sfc_rxq_info *rxq_info;
1160
1161         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1162         rxq_info = &sas->rxq_info[rx_queue_id];
1163
1164         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1165                 return 0;
1166
1167         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1168 }
1169
1170 /*
1171  * The function is used by the secondary process as well. It must not
1172  * use any process-local pointers from the adapter data.
1173  */
1174 static int
1175 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1176 {
1177         struct sfc_dp_rxq *dp_rxq = queue;
1178         const struct sfc_dp_rx *dp_rx;
1179
1180         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1181
1182         return offset < dp_rx->qdesc_npending(dp_rxq);
1183 }
1184
1185 /*
1186  * The function is used by the secondary process as well. It must not
1187  * use any process-local pointers from the adapter data.
1188  */
1189 static int
1190 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1191 {
1192         struct sfc_dp_rxq *dp_rxq = queue;
1193         const struct sfc_dp_rx *dp_rx;
1194
1195         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1196
1197         return dp_rx->qdesc_status(dp_rxq, offset);
1198 }
1199
1200 /*
1201  * The function is used by the secondary process as well. It must not
1202  * use any process-local pointers from the adapter data.
1203  */
1204 static int
1205 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1206 {
1207         struct sfc_dp_txq *dp_txq = queue;
1208         const struct sfc_dp_tx *dp_tx;
1209
1210         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1211
1212         return dp_tx->qdesc_status(dp_txq, offset);
1213 }
1214
1215 static int
1216 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1217 {
1218         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1219         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1220         int rc;
1221
1222         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1223
1224         sfc_adapter_lock(sa);
1225
1226         rc = EINVAL;
1227         if (sa->state != SFC_ADAPTER_STARTED)
1228                 goto fail_not_started;
1229
1230         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1231                 goto fail_not_setup;
1232
1233         rc = sfc_rx_qstart(sa, rx_queue_id);
1234         if (rc != 0)
1235                 goto fail_rx_qstart;
1236
1237         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1238
1239         sfc_adapter_unlock(sa);
1240
1241         return 0;
1242
1243 fail_rx_qstart:
1244 fail_not_setup:
1245 fail_not_started:
1246         sfc_adapter_unlock(sa);
1247         SFC_ASSERT(rc > 0);
1248         return -rc;
1249 }
1250
1251 static int
1252 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1253 {
1254         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1255         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1256
1257         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1258
1259         sfc_adapter_lock(sa);
1260         sfc_rx_qstop(sa, rx_queue_id);
1261
1262         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1263
1264         sfc_adapter_unlock(sa);
1265
1266         return 0;
1267 }
1268
1269 static int
1270 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1271 {
1272         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1273         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1274         int rc;
1275
1276         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1277
1278         sfc_adapter_lock(sa);
1279
1280         rc = EINVAL;
1281         if (sa->state != SFC_ADAPTER_STARTED)
1282                 goto fail_not_started;
1283
1284         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1285                 goto fail_not_setup;
1286
1287         rc = sfc_tx_qstart(sa, tx_queue_id);
1288         if (rc != 0)
1289                 goto fail_tx_qstart;
1290
1291         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1292
1293         sfc_adapter_unlock(sa);
1294         return 0;
1295
1296 fail_tx_qstart:
1297
1298 fail_not_setup:
1299 fail_not_started:
1300         sfc_adapter_unlock(sa);
1301         SFC_ASSERT(rc > 0);
1302         return -rc;
1303 }
1304
1305 static int
1306 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1307 {
1308         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1309         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1310
1311         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1312
1313         sfc_adapter_lock(sa);
1314
1315         sfc_tx_qstop(sa, tx_queue_id);
1316
1317         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1318
1319         sfc_adapter_unlock(sa);
1320         return 0;
1321 }
1322
1323 static efx_tunnel_protocol_t
1324 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1325 {
1326         switch (rte_type) {
1327         case RTE_TUNNEL_TYPE_VXLAN:
1328                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1329         case RTE_TUNNEL_TYPE_GENEVE:
1330                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1331         default:
1332                 return EFX_TUNNEL_NPROTOS;
1333         }
1334 }
1335
1336 enum sfc_udp_tunnel_op_e {
1337         SFC_UDP_TUNNEL_ADD_PORT,
1338         SFC_UDP_TUNNEL_DEL_PORT,
1339 };
1340
1341 static int
1342 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1343                       struct rte_eth_udp_tunnel *tunnel_udp,
1344                       enum sfc_udp_tunnel_op_e op)
1345 {
1346         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1347         efx_tunnel_protocol_t tunnel_proto;
1348         int rc;
1349
1350         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1351                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1352                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1353                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1354
1355         tunnel_proto =
1356                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1357         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1358                 rc = ENOTSUP;
1359                 goto fail_bad_proto;
1360         }
1361
1362         sfc_adapter_lock(sa);
1363
1364         switch (op) {
1365         case SFC_UDP_TUNNEL_ADD_PORT:
1366                 rc = efx_tunnel_config_udp_add(sa->nic,
1367                                                tunnel_udp->udp_port,
1368                                                tunnel_proto);
1369                 break;
1370         case SFC_UDP_TUNNEL_DEL_PORT:
1371                 rc = efx_tunnel_config_udp_remove(sa->nic,
1372                                                   tunnel_udp->udp_port,
1373                                                   tunnel_proto);
1374                 break;
1375         default:
1376                 rc = EINVAL;
1377                 goto fail_bad_op;
1378         }
1379
1380         if (rc != 0)
1381                 goto fail_op;
1382
1383         if (sa->state == SFC_ADAPTER_STARTED) {
1384                 rc = efx_tunnel_reconfigure(sa->nic);
1385                 if (rc == EAGAIN) {
1386                         /*
1387                          * Configuration is accepted by FW and MC reboot
1388                          * is initiated to apply the changes. MC reboot
1389                          * will be handled in a usual way (MC reboot
1390                          * event on management event queue and adapter
1391                          * restart).
1392                          */
1393                         rc = 0;
1394                 } else if (rc != 0) {
1395                         goto fail_reconfigure;
1396                 }
1397         }
1398
1399         sfc_adapter_unlock(sa);
1400         return 0;
1401
1402 fail_reconfigure:
1403         /* Remove/restore entry since the change makes the trouble */
1404         switch (op) {
1405         case SFC_UDP_TUNNEL_ADD_PORT:
1406                 (void)efx_tunnel_config_udp_remove(sa->nic,
1407                                                    tunnel_udp->udp_port,
1408                                                    tunnel_proto);
1409                 break;
1410         case SFC_UDP_TUNNEL_DEL_PORT:
1411                 (void)efx_tunnel_config_udp_add(sa->nic,
1412                                                 tunnel_udp->udp_port,
1413                                                 tunnel_proto);
1414                 break;
1415         }
1416
1417 fail_op:
1418 fail_bad_op:
1419         sfc_adapter_unlock(sa);
1420
1421 fail_bad_proto:
1422         SFC_ASSERT(rc > 0);
1423         return -rc;
1424 }
1425
1426 static int
1427 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1428                             struct rte_eth_udp_tunnel *tunnel_udp)
1429 {
1430         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1431 }
1432
1433 static int
1434 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1435                             struct rte_eth_udp_tunnel *tunnel_udp)
1436 {
1437         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1438 }
1439
1440 /*
1441  * The function is used by the secondary process as well. It must not
1442  * use any process-local pointers from the adapter data.
1443  */
1444 static int
1445 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1446                           struct rte_eth_rss_conf *rss_conf)
1447 {
1448         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1449         struct sfc_rss *rss = &sas->rss;
1450
1451         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1452                 return -ENOTSUP;
1453
1454         /*
1455          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1456          * hence, conversion is done here to derive a correct set of ETH_RSS
1457          * flags which corresponds to the active EFX configuration stored
1458          * locally in 'sfc_adapter' and kept up-to-date
1459          */
1460         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1461         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1462         if (rss_conf->rss_key != NULL)
1463                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1464
1465         return 0;
1466 }
1467
1468 static int
1469 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1470                         struct rte_eth_rss_conf *rss_conf)
1471 {
1472         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1473         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1474         unsigned int efx_hash_types;
1475         int rc = 0;
1476
1477         if (sfc_sa2shared(sa)->isolated)
1478                 return -ENOTSUP;
1479
1480         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1481                 sfc_err(sa, "RSS is not available");
1482                 return -ENOTSUP;
1483         }
1484
1485         if (rss->channels == 0) {
1486                 sfc_err(sa, "RSS is not configured");
1487                 return -EINVAL;
1488         }
1489
1490         if ((rss_conf->rss_key != NULL) &&
1491             (rss_conf->rss_key_len != sizeof(rss->key))) {
1492                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1493                         sizeof(rss->key));
1494                 return -EINVAL;
1495         }
1496
1497         sfc_adapter_lock(sa);
1498
1499         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1500         if (rc != 0)
1501                 goto fail_rx_hf_rte_to_efx;
1502
1503         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1504                                    rss->hash_alg, efx_hash_types, B_TRUE);
1505         if (rc != 0)
1506                 goto fail_scale_mode_set;
1507
1508         if (rss_conf->rss_key != NULL) {
1509                 if (sa->state == SFC_ADAPTER_STARTED) {
1510                         rc = efx_rx_scale_key_set(sa->nic,
1511                                                   EFX_RSS_CONTEXT_DEFAULT,
1512                                                   rss_conf->rss_key,
1513                                                   sizeof(rss->key));
1514                         if (rc != 0)
1515                                 goto fail_scale_key_set;
1516                 }
1517
1518                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1519         }
1520
1521         rss->hash_types = efx_hash_types;
1522
1523         sfc_adapter_unlock(sa);
1524
1525         return 0;
1526
1527 fail_scale_key_set:
1528         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1529                                   EFX_RX_HASHALG_TOEPLITZ,
1530                                   rss->hash_types, B_TRUE) != 0)
1531                 sfc_err(sa, "failed to restore RSS mode");
1532
1533 fail_scale_mode_set:
1534 fail_rx_hf_rte_to_efx:
1535         sfc_adapter_unlock(sa);
1536         return -rc;
1537 }
1538
1539 /*
1540  * The function is used by the secondary process as well. It must not
1541  * use any process-local pointers from the adapter data.
1542  */
1543 static int
1544 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1545                        struct rte_eth_rss_reta_entry64 *reta_conf,
1546                        uint16_t reta_size)
1547 {
1548         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1549         struct sfc_rss *rss = &sas->rss;
1550         int entry;
1551
1552         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1553                 return -ENOTSUP;
1554
1555         if (rss->channels == 0)
1556                 return -EINVAL;
1557
1558         if (reta_size != EFX_RSS_TBL_SIZE)
1559                 return -EINVAL;
1560
1561         for (entry = 0; entry < reta_size; entry++) {
1562                 int grp = entry / RTE_RETA_GROUP_SIZE;
1563                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1564
1565                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1566                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1567         }
1568
1569         return 0;
1570 }
1571
1572 static int
1573 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1574                         struct rte_eth_rss_reta_entry64 *reta_conf,
1575                         uint16_t reta_size)
1576 {
1577         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1578         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1579         unsigned int *rss_tbl_new;
1580         uint16_t entry;
1581         int rc = 0;
1582
1583
1584         if (sfc_sa2shared(sa)->isolated)
1585                 return -ENOTSUP;
1586
1587         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1588                 sfc_err(sa, "RSS is not available");
1589                 return -ENOTSUP;
1590         }
1591
1592         if (rss->channels == 0) {
1593                 sfc_err(sa, "RSS is not configured");
1594                 return -EINVAL;
1595         }
1596
1597         if (reta_size != EFX_RSS_TBL_SIZE) {
1598                 sfc_err(sa, "RETA size is wrong (should be %u)",
1599                         EFX_RSS_TBL_SIZE);
1600                 return -EINVAL;
1601         }
1602
1603         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1604         if (rss_tbl_new == NULL)
1605                 return -ENOMEM;
1606
1607         sfc_adapter_lock(sa);
1608
1609         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1610
1611         for (entry = 0; entry < reta_size; entry++) {
1612                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1613                 struct rte_eth_rss_reta_entry64 *grp;
1614
1615                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1616
1617                 if (grp->mask & (1ull << grp_idx)) {
1618                         if (grp->reta[grp_idx] >= rss->channels) {
1619                                 rc = EINVAL;
1620                                 goto bad_reta_entry;
1621                         }
1622                         rss_tbl_new[entry] = grp->reta[grp_idx];
1623                 }
1624         }
1625
1626         if (sa->state == SFC_ADAPTER_STARTED) {
1627                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1628                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1629                 if (rc != 0)
1630                         goto fail_scale_tbl_set;
1631         }
1632
1633         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1634
1635 fail_scale_tbl_set:
1636 bad_reta_entry:
1637         sfc_adapter_unlock(sa);
1638
1639         rte_free(rss_tbl_new);
1640
1641         SFC_ASSERT(rc >= 0);
1642         return -rc;
1643 }
1644
1645 static int
1646 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1647                     enum rte_filter_op filter_op,
1648                     void *arg)
1649 {
1650         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1651         int rc = ENOTSUP;
1652
1653         sfc_log_init(sa, "entry");
1654
1655         switch (filter_type) {
1656         case RTE_ETH_FILTER_NONE:
1657                 sfc_err(sa, "Global filters configuration not supported");
1658                 break;
1659         case RTE_ETH_FILTER_MACVLAN:
1660                 sfc_err(sa, "MACVLAN filters not supported");
1661                 break;
1662         case RTE_ETH_FILTER_ETHERTYPE:
1663                 sfc_err(sa, "EtherType filters not supported");
1664                 break;
1665         case RTE_ETH_FILTER_FLEXIBLE:
1666                 sfc_err(sa, "Flexible filters not supported");
1667                 break;
1668         case RTE_ETH_FILTER_SYN:
1669                 sfc_err(sa, "SYN filters not supported");
1670                 break;
1671         case RTE_ETH_FILTER_NTUPLE:
1672                 sfc_err(sa, "NTUPLE filters not supported");
1673                 break;
1674         case RTE_ETH_FILTER_TUNNEL:
1675                 sfc_err(sa, "Tunnel filters not supported");
1676                 break;
1677         case RTE_ETH_FILTER_FDIR:
1678                 sfc_err(sa, "Flow Director filters not supported");
1679                 break;
1680         case RTE_ETH_FILTER_HASH:
1681                 sfc_err(sa, "Hash filters not supported");
1682                 break;
1683         case RTE_ETH_FILTER_GENERIC:
1684                 if (filter_op != RTE_ETH_FILTER_GET) {
1685                         rc = EINVAL;
1686                 } else {
1687                         *(const void **)arg = &sfc_flow_ops;
1688                         rc = 0;
1689                 }
1690                 break;
1691         default:
1692                 sfc_err(sa, "Unknown filter type %u", filter_type);
1693                 break;
1694         }
1695
1696         sfc_log_init(sa, "exit: %d", -rc);
1697         SFC_ASSERT(rc >= 0);
1698         return -rc;
1699 }
1700
1701 static int
1702 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1703 {
1704         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1705
1706         /*
1707          * If Rx datapath does not provide callback to check mempool,
1708          * all pools are supported.
1709          */
1710         if (sap->dp_rx->pool_ops_supported == NULL)
1711                 return 1;
1712
1713         return sap->dp_rx->pool_ops_supported(pool);
1714 }
1715
1716 static const struct eth_dev_ops sfc_eth_dev_ops = {
1717         .dev_configure                  = sfc_dev_configure,
1718         .dev_start                      = sfc_dev_start,
1719         .dev_stop                       = sfc_dev_stop,
1720         .dev_set_link_up                = sfc_dev_set_link_up,
1721         .dev_set_link_down              = sfc_dev_set_link_down,
1722         .dev_close                      = sfc_dev_close,
1723         .promiscuous_enable             = sfc_dev_promisc_enable,
1724         .promiscuous_disable            = sfc_dev_promisc_disable,
1725         .allmulticast_enable            = sfc_dev_allmulti_enable,
1726         .allmulticast_disable           = sfc_dev_allmulti_disable,
1727         .link_update                    = sfc_dev_link_update,
1728         .stats_get                      = sfc_stats_get,
1729         .stats_reset                    = sfc_stats_reset,
1730         .xstats_get                     = sfc_xstats_get,
1731         .xstats_reset                   = sfc_stats_reset,
1732         .xstats_get_names               = sfc_xstats_get_names,
1733         .dev_infos_get                  = sfc_dev_infos_get,
1734         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1735         .mtu_set                        = sfc_dev_set_mtu,
1736         .rx_queue_start                 = sfc_rx_queue_start,
1737         .rx_queue_stop                  = sfc_rx_queue_stop,
1738         .tx_queue_start                 = sfc_tx_queue_start,
1739         .tx_queue_stop                  = sfc_tx_queue_stop,
1740         .rx_queue_setup                 = sfc_rx_queue_setup,
1741         .rx_queue_release               = sfc_rx_queue_release,
1742         .rx_queue_count                 = sfc_rx_queue_count,
1743         .rx_descriptor_done             = sfc_rx_descriptor_done,
1744         .rx_descriptor_status           = sfc_rx_descriptor_status,
1745         .tx_descriptor_status           = sfc_tx_descriptor_status,
1746         .tx_queue_setup                 = sfc_tx_queue_setup,
1747         .tx_queue_release               = sfc_tx_queue_release,
1748         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1749         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1750         .mac_addr_set                   = sfc_mac_addr_set,
1751         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1752         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1753         .reta_update                    = sfc_dev_rss_reta_update,
1754         .reta_query                     = sfc_dev_rss_reta_query,
1755         .rss_hash_update                = sfc_dev_rss_hash_update,
1756         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1757         .filter_ctrl                    = sfc_dev_filter_ctrl,
1758         .set_mc_addr_list               = sfc_set_mc_addr_list,
1759         .rxq_info_get                   = sfc_rx_queue_info_get,
1760         .txq_info_get                   = sfc_tx_queue_info_get,
1761         .fw_version_get                 = sfc_fw_version_get,
1762         .xstats_get_by_id               = sfc_xstats_get_by_id,
1763         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1764         .pool_ops_supported             = sfc_pool_ops_supported,
1765 };
1766
1767 /**
1768  * Duplicate a string in potentially shared memory required for
1769  * multi-process support.
1770  *
1771  * strdup() allocates from process-local heap/memory.
1772  */
1773 static char *
1774 sfc_strdup(const char *str)
1775 {
1776         size_t size;
1777         char *copy;
1778
1779         if (str == NULL)
1780                 return NULL;
1781
1782         size = strlen(str) + 1;
1783         copy = rte_malloc(__func__, size, 0);
1784         if (copy != NULL)
1785                 rte_memcpy(copy, str, size);
1786
1787         return copy;
1788 }
1789
1790 static int
1791 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1792 {
1793         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1794         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1795         const struct sfc_dp_rx *dp_rx;
1796         const struct sfc_dp_tx *dp_tx;
1797         const efx_nic_cfg_t *encp;
1798         unsigned int avail_caps = 0;
1799         const char *rx_name = NULL;
1800         const char *tx_name = NULL;
1801         int rc;
1802
1803         switch (sa->family) {
1804         case EFX_FAMILY_HUNTINGTON:
1805         case EFX_FAMILY_MEDFORD:
1806         case EFX_FAMILY_MEDFORD2:
1807                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1808                 break;
1809         default:
1810                 break;
1811         }
1812
1813         encp = efx_nic_cfg_get(sa->nic);
1814         if (encp->enc_rx_es_super_buffer_supported)
1815                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1816
1817         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1818                                 sfc_kvarg_string_handler, &rx_name);
1819         if (rc != 0)
1820                 goto fail_kvarg_rx_datapath;
1821
1822         if (rx_name != NULL) {
1823                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1824                 if (dp_rx == NULL) {
1825                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1826                         rc = ENOENT;
1827                         goto fail_dp_rx;
1828                 }
1829                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1830                         sfc_err(sa,
1831                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1832                                 rx_name);
1833                         rc = EINVAL;
1834                         goto fail_dp_rx_caps;
1835                 }
1836         } else {
1837                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1838                 if (dp_rx == NULL) {
1839                         sfc_err(sa, "Rx datapath by caps %#x not found",
1840                                 avail_caps);
1841                         rc = ENOENT;
1842                         goto fail_dp_rx;
1843                 }
1844         }
1845
1846         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1847         if (sas->dp_rx_name == NULL) {
1848                 rc = ENOMEM;
1849                 goto fail_dp_rx_name;
1850         }
1851
1852         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1853
1854         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1855                                 sfc_kvarg_string_handler, &tx_name);
1856         if (rc != 0)
1857                 goto fail_kvarg_tx_datapath;
1858
1859         if (tx_name != NULL) {
1860                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1861                 if (dp_tx == NULL) {
1862                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1863                         rc = ENOENT;
1864                         goto fail_dp_tx;
1865                 }
1866                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1867                         sfc_err(sa,
1868                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1869                                 tx_name);
1870                         rc = EINVAL;
1871                         goto fail_dp_tx_caps;
1872                 }
1873         } else {
1874                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1875                 if (dp_tx == NULL) {
1876                         sfc_err(sa, "Tx datapath by caps %#x not found",
1877                                 avail_caps);
1878                         rc = ENOENT;
1879                         goto fail_dp_tx;
1880                 }
1881         }
1882
1883         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1884         if (sas->dp_tx_name == NULL) {
1885                 rc = ENOMEM;
1886                 goto fail_dp_tx_name;
1887         }
1888
1889         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1890
1891         sa->priv.dp_rx = dp_rx;
1892         sa->priv.dp_tx = dp_tx;
1893
1894         dev->rx_pkt_burst = dp_rx->pkt_burst;
1895         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
1896         dev->tx_pkt_burst = dp_tx->pkt_burst;
1897
1898         dev->dev_ops = &sfc_eth_dev_ops;
1899
1900         return 0;
1901
1902 fail_dp_tx_name:
1903 fail_dp_tx_caps:
1904 fail_dp_tx:
1905 fail_kvarg_tx_datapath:
1906         rte_free(sas->dp_rx_name);
1907         sas->dp_rx_name = NULL;
1908
1909 fail_dp_rx_name:
1910 fail_dp_rx_caps:
1911 fail_dp_rx:
1912 fail_kvarg_rx_datapath:
1913         return rc;
1914 }
1915
1916 static void
1917 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1918 {
1919         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1920         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1921
1922         dev->dev_ops = NULL;
1923         dev->tx_pkt_prepare = NULL;
1924         dev->rx_pkt_burst = NULL;
1925         dev->tx_pkt_burst = NULL;
1926
1927         rte_free(sas->dp_tx_name);
1928         sas->dp_tx_name = NULL;
1929         sa->priv.dp_tx = NULL;
1930
1931         rte_free(sas->dp_rx_name);
1932         sas->dp_rx_name = NULL;
1933         sa->priv.dp_rx = NULL;
1934 }
1935
1936 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1937         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1938         .rx_queue_count                 = sfc_rx_queue_count,
1939         .rx_descriptor_done             = sfc_rx_descriptor_done,
1940         .rx_descriptor_status           = sfc_rx_descriptor_status,
1941         .tx_descriptor_status           = sfc_tx_descriptor_status,
1942         .reta_query                     = sfc_dev_rss_reta_query,
1943         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1944         .rxq_info_get                   = sfc_rx_queue_info_get,
1945         .txq_info_get                   = sfc_tx_queue_info_get,
1946 };
1947
1948 static int
1949 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1950 {
1951         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1952         struct sfc_adapter_priv *sap;
1953         const struct sfc_dp_rx *dp_rx;
1954         const struct sfc_dp_tx *dp_tx;
1955         int rc;
1956
1957         /*
1958          * Allocate process private data from heap, since it should not
1959          * be located in shared memory allocated using rte_malloc() API.
1960          */
1961         sap = calloc(1, sizeof(*sap));
1962         if (sap == NULL) {
1963                 rc = ENOMEM;
1964                 goto fail_alloc_priv;
1965         }
1966
1967         sap->logtype_main = logtype_main;
1968
1969         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
1970         if (dp_rx == NULL) {
1971                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1972                         "cannot find %s Rx datapath", sas->dp_rx_name);
1973                 rc = ENOENT;
1974                 goto fail_dp_rx;
1975         }
1976         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1977                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1978                         "%s Rx datapath does not support multi-process",
1979                         sas->dp_rx_name);
1980                 rc = EINVAL;
1981                 goto fail_dp_rx_multi_process;
1982         }
1983
1984         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
1985         if (dp_tx == NULL) {
1986                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1987                         "cannot find %s Tx datapath", sas->dp_tx_name);
1988                 rc = ENOENT;
1989                 goto fail_dp_tx;
1990         }
1991         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1992                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1993                         "%s Tx datapath does not support multi-process",
1994                         sas->dp_tx_name);
1995                 rc = EINVAL;
1996                 goto fail_dp_tx_multi_process;
1997         }
1998
1999         sap->dp_rx = dp_rx;
2000         sap->dp_tx = dp_tx;
2001
2002         dev->process_private = sap;
2003         dev->rx_pkt_burst = dp_rx->pkt_burst;
2004         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2005         dev->tx_pkt_burst = dp_tx->pkt_burst;
2006         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2007
2008         return 0;
2009
2010 fail_dp_tx_multi_process:
2011 fail_dp_tx:
2012 fail_dp_rx_multi_process:
2013 fail_dp_rx:
2014         free(sap);
2015
2016 fail_alloc_priv:
2017         return rc;
2018 }
2019
2020 static void
2021 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
2022 {
2023         free(dev->process_private);
2024         dev->process_private = NULL;
2025         dev->dev_ops = NULL;
2026         dev->tx_pkt_prepare = NULL;
2027         dev->tx_pkt_burst = NULL;
2028         dev->rx_pkt_burst = NULL;
2029 }
2030
2031 static void
2032 sfc_register_dp(void)
2033 {
2034         /* Register once */
2035         if (TAILQ_EMPTY(&sfc_dp_head)) {
2036                 /* Prefer EF10 datapath */
2037                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2038                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2039                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2040
2041                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2042                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2043                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2044         }
2045 }
2046
2047 static int
2048 sfc_eth_dev_init(struct rte_eth_dev *dev)
2049 {
2050         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2051         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2052         uint32_t logtype_main;
2053         struct sfc_adapter *sa;
2054         int rc;
2055         const efx_nic_cfg_t *encp;
2056         const struct rte_ether_addr *from;
2057
2058         sfc_register_dp();
2059
2060         logtype_main = sfc_register_logtype(&pci_dev->addr,
2061                                             SFC_LOGTYPE_MAIN_STR,
2062                                             RTE_LOG_NOTICE);
2063
2064         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2065                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2066
2067         /* Required for logging */
2068         sas->pci_addr = pci_dev->addr;
2069         sas->port_id = dev->data->port_id;
2070
2071         /*
2072          * Allocate process private data from heap, since it should not
2073          * be located in shared memory allocated using rte_malloc() API.
2074          */
2075         sa = calloc(1, sizeof(*sa));
2076         if (sa == NULL) {
2077                 rc = ENOMEM;
2078                 goto fail_alloc_sa;
2079         }
2080
2081         dev->process_private = sa;
2082
2083         /* Required for logging */
2084         sa->priv.shared = sas;
2085         sa->priv.logtype_main = logtype_main;
2086
2087         sa->eth_dev = dev;
2088
2089         /* Copy PCI device info to the dev->data */
2090         rte_eth_copy_pci_info(dev, pci_dev);
2091
2092         rc = sfc_kvargs_parse(sa);
2093         if (rc != 0)
2094                 goto fail_kvargs_parse;
2095
2096         sfc_log_init(sa, "entry");
2097
2098         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2099         if (dev->data->mac_addrs == NULL) {
2100                 rc = ENOMEM;
2101                 goto fail_mac_addrs;
2102         }
2103
2104         sfc_adapter_lock_init(sa);
2105         sfc_adapter_lock(sa);
2106
2107         sfc_log_init(sa, "probing");
2108         rc = sfc_probe(sa);
2109         if (rc != 0)
2110                 goto fail_probe;
2111
2112         sfc_log_init(sa, "set device ops");
2113         rc = sfc_eth_dev_set_ops(dev);
2114         if (rc != 0)
2115                 goto fail_set_ops;
2116
2117         sfc_log_init(sa, "attaching");
2118         rc = sfc_attach(sa);
2119         if (rc != 0)
2120                 goto fail_attach;
2121
2122         encp = efx_nic_cfg_get(sa->nic);
2123
2124         /*
2125          * The arguments are really reverse order in comparison to
2126          * Linux kernel. Copy from NIC config to Ethernet device data.
2127          */
2128         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2129         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2130
2131         sfc_adapter_unlock(sa);
2132
2133         sfc_log_init(sa, "done");
2134         return 0;
2135
2136 fail_attach:
2137         sfc_eth_dev_clear_ops(dev);
2138
2139 fail_set_ops:
2140         sfc_unprobe(sa);
2141
2142 fail_probe:
2143         sfc_adapter_unlock(sa);
2144         sfc_adapter_lock_fini(sa);
2145         rte_free(dev->data->mac_addrs);
2146         dev->data->mac_addrs = NULL;
2147
2148 fail_mac_addrs:
2149         sfc_kvargs_cleanup(sa);
2150
2151 fail_kvargs_parse:
2152         sfc_log_init(sa, "failed %d", rc);
2153         dev->process_private = NULL;
2154         free(sa);
2155
2156 fail_alloc_sa:
2157         SFC_ASSERT(rc > 0);
2158         return -rc;
2159 }
2160
2161 static int
2162 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2163 {
2164         struct sfc_adapter *sa;
2165
2166         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2167                 sfc_eth_dev_secondary_clear_ops(dev);
2168                 return 0;
2169         }
2170
2171         sa = sfc_adapter_by_eth_dev(dev);
2172         sfc_log_init(sa, "entry");
2173
2174         sfc_adapter_lock(sa);
2175
2176         sfc_eth_dev_clear_ops(dev);
2177
2178         sfc_detach(sa);
2179         sfc_unprobe(sa);
2180
2181         sfc_kvargs_cleanup(sa);
2182
2183         sfc_adapter_unlock(sa);
2184         sfc_adapter_lock_fini(sa);
2185
2186         sfc_log_init(sa, "done");
2187
2188         /* Required for logging, so cleanup last */
2189         sa->eth_dev = NULL;
2190
2191         dev->process_private = NULL;
2192         free(sa);
2193
2194         return 0;
2195 }
2196
2197 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2198         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2199         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2200         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2201         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2202         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2203         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2204         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2205         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2206         { .vendor_id = 0 /* sentinel */ }
2207 };
2208
2209 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2210         struct rte_pci_device *pci_dev)
2211 {
2212         return rte_eth_dev_pci_generic_probe(pci_dev,
2213                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2214 }
2215
2216 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2217 {
2218         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2219 }
2220
2221 static struct rte_pci_driver sfc_efx_pmd = {
2222         .id_table = pci_id_sfc_efx_map,
2223         .drv_flags =
2224                 RTE_PCI_DRV_INTR_LSC |
2225                 RTE_PCI_DRV_NEED_MAPPING,
2226         .probe = sfc_eth_dev_pci_probe,
2227         .remove = sfc_eth_dev_pci_remove,
2228 };
2229
2230 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2231 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2232 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2233 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2234         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2235         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2236         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2237         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2238         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2239         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2240
2241 RTE_INIT(sfc_driver_register_logtype)
2242 {
2243         int ret;
2244
2245         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2246                                                    RTE_LOG_NOTICE);
2247         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2248 }