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
  18 #include "efx.h"
  19
  20 #include "sfc.h"
  21 #include "sfc_debug.h"
  22 #include "sfc_log.h"
  23 #include "sfc_kvargs.h"
  24 #include "sfc_ev.h"
  25 #include "sfc_rx.h"
  26 #include "sfc_tx.h"
  27 #include "sfc_flow.h"
  28 #include "sfc_dp.h"
  29 #include "sfc_dp_rx.h"
  30
  31 uint32_t sfc_logtype_driver;
  32
  33 static struct sfc_dp_list sfc_dp_head =
  34         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  35
  36 static int
  37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  38 {
  39         struct sfc_adapter *sa = dev->data->dev_private;
  40         efx_nic_fw_info_t enfi;
  41         int ret;
  42         int rc;
  43
  44         /*
  45          * Return value of the callback is likely supposed to be
  46          * equal to or greater than 0, nevertheless, if an error
  47          * occurs, it will be desirable to pass it to the caller
  48          */
  49         if ((fw_version == NULL) || (fw_size == 0))
  50                 return -EINVAL;
  51
  52         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  53         if (rc != 0)
  54                 return -rc;
  55
  56         ret = snprintf(fw_version, fw_size,
  57                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  58                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  59                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  60         if (ret < 0)
  61                 return ret;
  62
  63         if (enfi.enfi_dpcpu_fw_ids_valid) {
  64                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  65                 int ret_extra;
  66
  67                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  68                                      fw_size - dpcpu_fw_ids_offset,
  69                                      " rx%" PRIx16 " tx%" PRIx16,
  70                                      enfi.enfi_rx_dpcpu_fw_id,
  71                                      enfi.enfi_tx_dpcpu_fw_id);
  72                 if (ret_extra < 0)
  73                         return ret_extra;
  74
  75                 ret += ret_extra;
  76         }
  77
  78         if (fw_size < (size_t)(++ret))
  79                 return ret;
  80         else
  81                 return 0;
  82 }
  83
  84 static void
  85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  86 {
  87         struct sfc_adapter *sa = dev->data->dev_private;
  88         struct sfc_rss *rss = &sa->rss;
  89         uint64_t txq_offloads_def = 0;
  90
  91         sfc_log_init(sa, "entry");
  92
  93         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  94
  95         /* Autonegotiation may be disabled */
  96         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  97         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  98                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  99         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
 100                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 101         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
 102                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 103         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 104                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 105         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
 106                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 107         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
 108                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 109
 110         dev_info->max_rx_queues = sa->rxq_max;
 111         dev_info->max_tx_queues = sa->txq_max;
 112
 113         /* By default packets are dropped if no descriptors are available */
 114         dev_info->default_rxconf.rx_drop_en = 1;
 115
 116         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 117
 118         /*
 119          * rx_offload_capa includes both device and queue offloads since
 120          * the latter may be requested on a per device basis which makes
 121          * sense when some offloads are needed to be set on all queues.
 122          */
 123         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 124                                     dev_info->rx_queue_offload_capa;
 125
 126         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 127
 128         /*
 129          * tx_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->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 134                                     dev_info->tx_queue_offload_capa;
 135
 136         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 137                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 138
 139         dev_info->default_txconf.offloads |= txq_offloads_def;
 140
 141         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 142                 uint64_t rte_hf = 0;
 143                 unsigned int i;
 144
 145                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 146                         rte_hf |= rss->hf_map[i].rte;
 147
 148                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 149                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 150                 dev_info->flow_type_rss_offloads = rte_hf;
 151         }
 152
 153         /* Initialize to hardware limits */
 154         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 155         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 156         /* The RXQ hardware requires that the descriptor count is a power
 157          * of 2, but rx_desc_lim cannot properly describe that constraint.
 158          */
 159         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 160
 161         /* Initialize to hardware limits */
 162         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 163         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 164         /*
 165          * The TXQ hardware requires that the descriptor count is a power
 166          * of 2, but tx_desc_lim cannot properly describe that constraint
 167          */
 168         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 169
 170         if (sa->dp_rx->get_dev_info != NULL)
 171                 sa->dp_rx->get_dev_info(dev_info);
 172         if (sa->dp_tx->get_dev_info != NULL)
 173                 sa->dp_tx->get_dev_info(dev_info);
 174
 175         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 176                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 177 }
 178
 179 static const uint32_t *
 180 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 181 {
 182         struct sfc_adapter *sa = dev->data->dev_private;
 183         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 184         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 185
 186         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 187 }
 188
 189 static int
 190 sfc_dev_configure(struct rte_eth_dev *dev)
 191 {
 192         struct rte_eth_dev_data *dev_data = dev->data;
 193         struct sfc_adapter *sa = dev_data->dev_private;
 194         int rc;
 195
 196         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 197                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 198
 199         sfc_adapter_lock(sa);
 200         switch (sa->state) {
 201         case SFC_ADAPTER_CONFIGURED:
 202                 /* FALLTHROUGH */
 203         case SFC_ADAPTER_INITIALIZED:
 204                 rc = sfc_configure(sa);
 205                 break;
 206         default:
 207                 sfc_err(sa, "unexpected adapter state %u to configure",
 208                         sa->state);
 209                 rc = EINVAL;
 210                 break;
 211         }
 212         sfc_adapter_unlock(sa);
 213
 214         sfc_log_init(sa, "done %d", rc);
 215         SFC_ASSERT(rc >= 0);
 216         return -rc;
 217 }
 218
 219 static int
 220 sfc_dev_start(struct rte_eth_dev *dev)
 221 {
 222         struct sfc_adapter *sa = dev->data->dev_private;
 223         int rc;
 224
 225         sfc_log_init(sa, "entry");
 226
 227         sfc_adapter_lock(sa);
 228         rc = sfc_start(sa);
 229         sfc_adapter_unlock(sa);
 230
 231         sfc_log_init(sa, "done %d", rc);
 232         SFC_ASSERT(rc >= 0);
 233         return -rc;
 234 }
 235
 236 static int
 237 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 238 {
 239         struct sfc_adapter *sa = dev->data->dev_private;
 240         struct rte_eth_link current_link;
 241         int ret;
 242
 243         sfc_log_init(sa, "entry");
 244
 245         if (sa->state != SFC_ADAPTER_STARTED) {
 246                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 247         } else if (wait_to_complete) {
 248                 efx_link_mode_t link_mode;
 249
 250                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 251                         link_mode = EFX_LINK_UNKNOWN;
 252                 sfc_port_link_mode_to_info(link_mode, &current_link);
 253
 254         } else {
 255                 sfc_ev_mgmt_qpoll(sa);
 256                 rte_eth_linkstatus_get(dev, &current_link);
 257         }
 258
 259         ret = rte_eth_linkstatus_set(dev, &current_link);
 260         if (ret == 0)
 261                 sfc_notice(sa, "Link status is %s",
 262                            current_link.link_status ? "UP" : "DOWN");
 263
 264         return ret;
 265 }
 266
 267 static void
 268 sfc_dev_stop(struct rte_eth_dev *dev)
 269 {
 270         struct sfc_adapter *sa = dev->data->dev_private;
 271
 272         sfc_log_init(sa, "entry");
 273
 274         sfc_adapter_lock(sa);
 275         sfc_stop(sa);
 276         sfc_adapter_unlock(sa);
 277
 278         sfc_log_init(sa, "done");
 279 }
 280
 281 static int
 282 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 283 {
 284         struct sfc_adapter *sa = dev->data->dev_private;
 285         int rc;
 286
 287         sfc_log_init(sa, "entry");
 288
 289         sfc_adapter_lock(sa);
 290         rc = sfc_start(sa);
 291         sfc_adapter_unlock(sa);
 292
 293         SFC_ASSERT(rc >= 0);
 294         return -rc;
 295 }
 296
 297 static int
 298 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 299 {
 300         struct sfc_adapter *sa = dev->data->dev_private;
 301
 302         sfc_log_init(sa, "entry");
 303
 304         sfc_adapter_lock(sa);
 305         sfc_stop(sa);
 306         sfc_adapter_unlock(sa);
 307
 308         return 0;
 309 }
 310
 311 static void
 312 sfc_dev_close(struct rte_eth_dev *dev)
 313 {
 314         struct sfc_adapter *sa = dev->data->dev_private;
 315
 316         sfc_log_init(sa, "entry");
 317
 318         sfc_adapter_lock(sa);
 319         switch (sa->state) {
 320         case SFC_ADAPTER_STARTED:
 321                 sfc_stop(sa);
 322                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 323                 /* FALLTHROUGH */
 324         case SFC_ADAPTER_CONFIGURED:
 325                 sfc_close(sa);
 326                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 327                 /* FALLTHROUGH */
 328         case SFC_ADAPTER_INITIALIZED:
 329                 break;
 330         default:
 331                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 332                 break;
 333         }
 334         sfc_adapter_unlock(sa);
 335
 336         sfc_log_init(sa, "done");
 337 }
 338
 339 static void
 340 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 341                    boolean_t enabled)
 342 {
 343         struct sfc_port *port;
 344         boolean_t *toggle;
 345         struct sfc_adapter *sa = dev->data->dev_private;
 346         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 347         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 348
 349         sfc_adapter_lock(sa);
 350
 351         port = &sa->port;
 352         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 353
 354         if (*toggle != enabled) {
 355                 *toggle = enabled;
 356
 357                 if (port->isolated) {
 358                         sfc_warn(sa, "isolated mode is active on the port");
 359                         sfc_warn(sa, "the change is to be applied on the next "
 360                                      "start provided that isolated mode is "
 361                                      "disabled prior the next start");
 362                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 363                            (sfc_set_rx_mode(sa) != 0)) {
 364                         *toggle = !(enabled);
 365                         sfc_warn(sa, "Failed to %s %s mode",
 366                                  ((enabled) ? "enable" : "disable"), desc);
 367                 }
 368         }
 369
 370         sfc_adapter_unlock(sa);
 371 }
 372
 373 static void
 374 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 375 {
 376         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 377 }
 378
 379 static void
 380 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 381 {
 382         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 383 }
 384
 385 static void
 386 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 387 {
 388         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 389 }
 390
 391 static void
 392 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 393 {
 394         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 395 }
 396
 397 static int
 398 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 399                    uint16_t nb_rx_desc, unsigned int socket_id,
 400                    const struct rte_eth_rxconf *rx_conf,
 401                    struct rte_mempool *mb_pool)
 402 {
 403         struct sfc_adapter *sa = dev->data->dev_private;
 404         int rc;
 405
 406         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 407                      rx_queue_id, nb_rx_desc, socket_id);
 408
 409         sfc_adapter_lock(sa);
 410
 411         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 412                           rx_conf, mb_pool);
 413         if (rc != 0)
 414                 goto fail_rx_qinit;
 415
 416         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 417
 418         sfc_adapter_unlock(sa);
 419
 420         return 0;
 421
 422 fail_rx_qinit:
 423         sfc_adapter_unlock(sa);
 424         SFC_ASSERT(rc > 0);
 425         return -rc;
 426 }
 427
 428 static void
 429 sfc_rx_queue_release(void *queue)
 430 {
 431         struct sfc_dp_rxq *dp_rxq = queue;
 432         struct sfc_rxq *rxq;
 433         struct sfc_adapter *sa;
 434         unsigned int sw_index;
 435
 436         if (dp_rxq == NULL)
 437                 return;
 438
 439         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 440         sa = rxq->evq->sa;
 441         sfc_adapter_lock(sa);
 442
 443         sw_index = sfc_rxq_sw_index(rxq);
 444
 445         sfc_log_init(sa, "RxQ=%u", sw_index);
 446
 447         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 448
 449         sfc_rx_qfini(sa, sw_index);
 450
 451         sfc_adapter_unlock(sa);
 452 }
 453
 454 static int
 455 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 456                    uint16_t nb_tx_desc, unsigned int socket_id,
 457                    const struct rte_eth_txconf *tx_conf)
 458 {
 459         struct sfc_adapter *sa = dev->data->dev_private;
 460         int rc;
 461
 462         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 463                      tx_queue_id, nb_tx_desc, socket_id);
 464
 465         sfc_adapter_lock(sa);
 466
 467         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 468         if (rc != 0)
 469                 goto fail_tx_qinit;
 470
 471         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 472
 473         sfc_adapter_unlock(sa);
 474         return 0;
 475
 476 fail_tx_qinit:
 477         sfc_adapter_unlock(sa);
 478         SFC_ASSERT(rc > 0);
 479         return -rc;
 480 }
 481
 482 static void
 483 sfc_tx_queue_release(void *queue)
 484 {
 485         struct sfc_dp_txq *dp_txq = queue;
 486         struct sfc_txq *txq;
 487         unsigned int sw_index;
 488         struct sfc_adapter *sa;
 489
 490         if (dp_txq == NULL)
 491                 return;
 492
 493         txq = sfc_txq_by_dp_txq(dp_txq);
 494         sw_index = sfc_txq_sw_index(txq);
 495
 496         SFC_ASSERT(txq->evq != NULL);
 497         sa = txq->evq->sa;
 498
 499         sfc_log_init(sa, "TxQ = %u", sw_index);
 500
 501         sfc_adapter_lock(sa);
 502
 503         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 504         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 505
 506         sfc_tx_qfini(sa, sw_index);
 507
 508         sfc_adapter_unlock(sa);
 509 }
 510
 511 static int
 512 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 513 {
 514         struct sfc_adapter *sa = dev->data->dev_private;
 515         struct sfc_port *port = &sa->port;
 516         uint64_t *mac_stats;
 517         int ret;
 518
 519         rte_spinlock_lock(&port->mac_stats_lock);
 520
 521         ret = sfc_port_update_mac_stats(sa);
 522         if (ret != 0)
 523                 goto unlock;
 524
 525         mac_stats = port->mac_stats_buf;
 526
 527         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 528                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 529                 stats->ipackets =
 530                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 531                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 532                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 533                 stats->opackets =
 534                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 535                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 536                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 537                 stats->ibytes =
 538                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 539                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 540                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 541                 stats->obytes =
 542                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 543                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 544                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 545                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 546                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 547                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 548         } else {
 549                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 550                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 551                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 552                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 553                 /*
 554                  * Take into account stats which are whenever supported
 555                  * on EF10. If some stat is not supported by current
 556                  * firmware variant or HW revision, it is guaranteed
 557                  * to be zero in mac_stats.
 558                  */
 559                 stats->imissed =
 560                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 561                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 562                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 563                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 564                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 565                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 566                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 567                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 568                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 569                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 570                 stats->ierrors =
 571                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 572                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 573                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 574                 /* no oerrors counters supported on EF10 */
 575         }
 576
 577 unlock:
 578         rte_spinlock_unlock(&port->mac_stats_lock);
 579         SFC_ASSERT(ret >= 0);
 580         return -ret;
 581 }
 582
 583 static void
 584 sfc_stats_reset(struct rte_eth_dev *dev)
 585 {
 586         struct sfc_adapter *sa = dev->data->dev_private;
 587         struct sfc_port *port = &sa->port;
 588         int rc;
 589
 590         if (sa->state != SFC_ADAPTER_STARTED) {
 591                 /*
 592                  * The operation cannot be done if port is not started; it
 593                  * will be scheduled to be done during the next port start
 594                  */
 595                 port->mac_stats_reset_pending = B_TRUE;
 596                 return;
 597         }
 598
 599         rc = sfc_port_reset_mac_stats(sa);
 600         if (rc != 0)
 601                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 602 }
 603
 604 static int
 605 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 606                unsigned int xstats_count)
 607 {
 608         struct sfc_adapter *sa = dev->data->dev_private;
 609         struct sfc_port *port = &sa->port;
 610         uint64_t *mac_stats;
 611         int rc;
 612         unsigned int i;
 613         int nstats = 0;
 614
 615         rte_spinlock_lock(&port->mac_stats_lock);
 616
 617         rc = sfc_port_update_mac_stats(sa);
 618         if (rc != 0) {
 619                 SFC_ASSERT(rc > 0);
 620                 nstats = -rc;
 621                 goto unlock;
 622         }
 623
 624         mac_stats = port->mac_stats_buf;
 625
 626         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 627                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 628                         if (xstats != NULL && nstats < (int)xstats_count) {
 629                                 xstats[nstats].id = nstats;
 630                                 xstats[nstats].value = mac_stats[i];
 631                         }
 632                         nstats++;
 633                 }
 634         }
 635
 636 unlock:
 637         rte_spinlock_unlock(&port->mac_stats_lock);
 638
 639         return nstats;
 640 }
 641
 642 static int
 643 sfc_xstats_get_names(struct rte_eth_dev *dev,
 644                      struct rte_eth_xstat_name *xstats_names,
 645                      unsigned int xstats_count)
 646 {
 647         struct sfc_adapter *sa = dev->data->dev_private;
 648         struct sfc_port *port = &sa->port;
 649         unsigned int i;
 650         unsigned int nstats = 0;
 651
 652         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 653                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 654                         if (xstats_names != NULL && nstats < xstats_count)
 655                                 strlcpy(xstats_names[nstats].name,
 656                                         efx_mac_stat_name(sa->nic, i),
 657                                         sizeof(xstats_names[0].name));
 658                         nstats++;
 659                 }
 660         }
 661
 662         return nstats;
 663 }
 664
 665 static int
 666 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 667                      uint64_t *values, unsigned int n)
 668 {
 669         struct sfc_adapter *sa = dev->data->dev_private;
 670         struct sfc_port *port = &sa->port;
 671         uint64_t *mac_stats;
 672         unsigned int nb_supported = 0;
 673         unsigned int nb_written = 0;
 674         unsigned int i;
 675         int ret;
 676         int rc;
 677
 678         if (unlikely(values == NULL) ||
 679             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 680                 return port->mac_stats_nb_supported;
 681
 682         rte_spinlock_lock(&port->mac_stats_lock);
 683
 684         rc = sfc_port_update_mac_stats(sa);
 685         if (rc != 0) {
 686                 SFC_ASSERT(rc > 0);
 687                 ret = -rc;
 688                 goto unlock;
 689         }
 690
 691         mac_stats = port->mac_stats_buf;
 692
 693         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 694                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 695                         continue;
 696
 697                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 698                         values[nb_written++] = mac_stats[i];
 699
 700                 ++nb_supported;
 701         }
 702
 703         ret = nb_written;
 704
 705 unlock:
 706         rte_spinlock_unlock(&port->mac_stats_lock);
 707
 708         return ret;
 709 }
 710
 711 static int
 712 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 713                            struct rte_eth_xstat_name *xstats_names,
 714                            const uint64_t *ids, unsigned int size)
 715 {
 716         struct sfc_adapter *sa = dev->data->dev_private;
 717         struct sfc_port *port = &sa->port;
 718         unsigned int nb_supported = 0;
 719         unsigned int nb_written = 0;
 720         unsigned int i;
 721
 722         if (unlikely(xstats_names == NULL) ||
 723             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 724                 return port->mac_stats_nb_supported;
 725
 726         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 727                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 728                         continue;
 729
 730                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 731                         char *name = xstats_names[nb_written++].name;
 732
 733                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 734                                 sizeof(xstats_names[0].name));
 735                 }
 736
 737                 ++nb_supported;
 738         }
 739
 740         return nb_written;
 741 }
 742
 743 static int
 744 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 745 {
 746         struct sfc_adapter *sa = dev->data->dev_private;
 747         unsigned int wanted_fc, link_fc;
 748
 749         memset(fc_conf, 0, sizeof(*fc_conf));
 750
 751         sfc_adapter_lock(sa);
 752
 753         if (sa->state == SFC_ADAPTER_STARTED)
 754                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 755         else
 756                 link_fc = sa->port.flow_ctrl;
 757
 758         switch (link_fc) {
 759         case 0:
 760                 fc_conf->mode = RTE_FC_NONE;
 761                 break;
 762         case EFX_FCNTL_RESPOND:
 763                 fc_conf->mode = RTE_FC_RX_PAUSE;
 764                 break;
 765         case EFX_FCNTL_GENERATE:
 766                 fc_conf->mode = RTE_FC_TX_PAUSE;
 767                 break;
 768         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 769                 fc_conf->mode = RTE_FC_FULL;
 770                 break;
 771         default:
 772                 sfc_err(sa, "%s: unexpected flow control value %#x",
 773                         __func__, link_fc);
 774         }
 775
 776         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 777
 778         sfc_adapter_unlock(sa);
 779
 780         return 0;
 781 }
 782
 783 static int
 784 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 785 {
 786         struct sfc_adapter *sa = dev->data->dev_private;
 787         struct sfc_port *port = &sa->port;
 788         unsigned int fcntl;
 789         int rc;
 790
 791         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 792             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 793             fc_conf->mac_ctrl_frame_fwd != 0) {
 794                 sfc_err(sa, "unsupported flow control settings specified");
 795                 rc = EINVAL;
 796                 goto fail_inval;
 797         }
 798
 799         switch (fc_conf->mode) {
 800         case RTE_FC_NONE:
 801                 fcntl = 0;
 802                 break;
 803         case RTE_FC_RX_PAUSE:
 804                 fcntl = EFX_FCNTL_RESPOND;
 805                 break;
 806         case RTE_FC_TX_PAUSE:
 807                 fcntl = EFX_FCNTL_GENERATE;
 808                 break;
 809         case RTE_FC_FULL:
 810                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 811                 break;
 812         default:
 813                 rc = EINVAL;
 814                 goto fail_inval;
 815         }
 816
 817         sfc_adapter_lock(sa);
 818
 819         if (sa->state == SFC_ADAPTER_STARTED) {
 820                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 821                 if (rc != 0)
 822                         goto fail_mac_fcntl_set;
 823         }
 824
 825         port->flow_ctrl = fcntl;
 826         port->flow_ctrl_autoneg = fc_conf->autoneg;
 827
 828         sfc_adapter_unlock(sa);
 829
 830         return 0;
 831
 832 fail_mac_fcntl_set:
 833         sfc_adapter_unlock(sa);
 834 fail_inval:
 835         SFC_ASSERT(rc > 0);
 836         return -rc;
 837 }
 838
 839 static int
 840 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 841 {
 842         struct sfc_adapter *sa = dev->data->dev_private;
 843         size_t pdu = EFX_MAC_PDU(mtu);
 844         size_t old_pdu;
 845         int rc;
 846
 847         sfc_log_init(sa, "mtu=%u", mtu);
 848
 849         rc = EINVAL;
 850         if (pdu < EFX_MAC_PDU_MIN) {
 851                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 852                         (unsigned int)mtu, (unsigned int)pdu,
 853                         EFX_MAC_PDU_MIN);
 854                 goto fail_inval;
 855         }
 856         if (pdu > EFX_MAC_PDU_MAX) {
 857                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 858                         (unsigned int)mtu, (unsigned int)pdu,
 859                         EFX_MAC_PDU_MAX);
 860                 goto fail_inval;
 861         }
 862
 863         sfc_adapter_lock(sa);
 864
 865         if (pdu != sa->port.pdu) {
 866                 if (sa->state == SFC_ADAPTER_STARTED) {
 867                         sfc_stop(sa);
 868
 869                         old_pdu = sa->port.pdu;
 870                         sa->port.pdu = pdu;
 871                         rc = sfc_start(sa);
 872                         if (rc != 0)
 873                                 goto fail_start;
 874                 } else {
 875                         sa->port.pdu = pdu;
 876                 }
 877         }
 878
 879         /*
 880          * The driver does not use it, but other PMDs update jumbo frame
 881          * flag and max_rx_pkt_len when MTU is set.
 882          */
 883         if (mtu > ETHER_MAX_LEN) {
 884                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 885                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 886         }
 887
 888         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 889
 890         sfc_adapter_unlock(sa);
 891
 892         sfc_log_init(sa, "done");
 893         return 0;
 894
 895 fail_start:
 896         sa->port.pdu = old_pdu;
 897         if (sfc_start(sa) != 0)
 898                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 899                         "PDU max size - port is stopped",
 900                         (unsigned int)pdu, (unsigned int)old_pdu);
 901         sfc_adapter_unlock(sa);
 902
 903 fail_inval:
 904         sfc_log_init(sa, "failed %d", rc);
 905         SFC_ASSERT(rc > 0);
 906         return -rc;
 907 }
 908 static int
 909 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 910 {
 911         struct sfc_adapter *sa = dev->data->dev_private;
 912         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 913         struct sfc_port *port = &sa->port;
 914         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
 915         int rc = 0;
 916
 917         sfc_adapter_lock(sa);
 918
 919         /*
 920          * Copy the address to the device private data so that
 921          * it could be recalled in the case of adapter restart.
 922          */
 923         ether_addr_copy(mac_addr, &port->default_mac_addr);
 924
 925         /*
 926          * Neither of the two following checks can return
 927          * an error. The new MAC address is preserved in
 928          * the device private data and can be activated
 929          * on the next port start if the user prevents
 930          * isolated mode from being enabled.
 931          */
 932         if (port->isolated) {
 933                 sfc_warn(sa, "isolated mode is active on the port");
 934                 sfc_warn(sa, "will not set MAC address");
 935                 goto unlock;
 936         }
 937
 938         if (sa->state != SFC_ADAPTER_STARTED) {
 939                 sfc_notice(sa, "the port is not started");
 940                 sfc_notice(sa, "the new MAC address will be set on port start");
 941
 942                 goto unlock;
 943         }
 944
 945         if (encp->enc_allow_set_mac_with_installed_filters) {
 946                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 947                 if (rc != 0) {
 948                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 949                         goto unlock;
 950                 }
 951
 952                 /*
 953                  * Changing the MAC address by means of MCDI request
 954                  * has no effect on received traffic, therefore
 955                  * we also need to update unicast filters
 956                  */
 957                 rc = sfc_set_rx_mode(sa);
 958                 if (rc != 0) {
 959                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 960                         /* Rollback the old address */
 961                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
 962                         (void)sfc_set_rx_mode(sa);
 963                 }
 964         } else {
 965                 sfc_warn(sa, "cannot set MAC address with filters installed");
 966                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 967                 sfc_warn(sa, "(some traffic may be dropped)");
 968
 969                 /*
 970                  * Since setting MAC address with filters installed is not
 971                  * allowed on the adapter, the new MAC address will be set
 972                  * by means of adapter restart. sfc_start() shall retrieve
 973                  * the new address from the device private data and set it.
 974                  */
 975                 sfc_stop(sa);
 976                 rc = sfc_start(sa);
 977                 if (rc != 0)
 978                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 979         }
 980
 981 unlock:
 982         if (rc != 0)
 983                 ether_addr_copy(old_addr, &port->default_mac_addr);
 984
 985         sfc_adapter_unlock(sa);
 986
 987         SFC_ASSERT(rc >= 0);
 988         return -rc;
 989 }
 990
 991
 992 static int
 993 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 994                      uint32_t nb_mc_addr)
 995 {
 996         struct sfc_adapter *sa = dev->data->dev_private;
 997         struct sfc_port *port = &sa->port;
 998         uint8_t *mc_addrs = port->mcast_addrs;
 999         int rc;
1000         unsigned int i;
1001
1002         if (port->isolated) {
1003                 sfc_err(sa, "isolated mode is active on the port");
1004                 sfc_err(sa, "will not set multicast address list");
1005                 return -ENOTSUP;
1006         }
1007
1008         if (mc_addrs == NULL)
1009                 return -ENOBUFS;
1010
1011         if (nb_mc_addr > port->max_mcast_addrs) {
1012                 sfc_err(sa, "too many multicast addresses: %u > %u",
1013                          nb_mc_addr, port->max_mcast_addrs);
1014                 return -EINVAL;
1015         }
1016
1017         for (i = 0; i < nb_mc_addr; ++i) {
1018                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1019                                  EFX_MAC_ADDR_LEN);
1020                 mc_addrs += EFX_MAC_ADDR_LEN;
1021         }
1022
1023         port->nb_mcast_addrs = nb_mc_addr;
1024
1025         if (sa->state != SFC_ADAPTER_STARTED)
1026                 return 0;
1027
1028         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1029                                         port->nb_mcast_addrs);
1030         if (rc != 0)
1031                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1032
1033         SFC_ASSERT(rc >= 0);
1034         return -rc;
1035 }
1036
1037 /*
1038  * The function is used by the secondary process as well. It must not
1039  * use any process-local pointers from the adapter data.
1040  */
1041 static void
1042 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1043                       struct rte_eth_rxq_info *qinfo)
1044 {
1045         struct sfc_adapter *sa = dev->data->dev_private;
1046         struct sfc_rxq_info *rxq_info;
1047         struct sfc_rxq *rxq;
1048
1049         sfc_adapter_lock(sa);
1050
1051         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1052
1053         rxq_info = &sa->rxq_info[rx_queue_id];
1054         rxq = rxq_info->rxq;
1055         SFC_ASSERT(rxq != NULL);
1056
1057         qinfo->mp = rxq->refill_mb_pool;
1058         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1059         qinfo->conf.rx_drop_en = 1;
1060         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1061         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1062         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1063                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1064                 qinfo->scattered_rx = 1;
1065         }
1066         qinfo->nb_desc = rxq_info->entries;
1067
1068         sfc_adapter_unlock(sa);
1069 }
1070
1071 /*
1072  * The function is used by the secondary process as well. It must not
1073  * use any process-local pointers from the adapter data.
1074  */
1075 static void
1076 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1077                       struct rte_eth_txq_info *qinfo)
1078 {
1079         struct sfc_adapter *sa = dev->data->dev_private;
1080         struct sfc_txq_info *txq_info;
1081
1082         sfc_adapter_lock(sa);
1083
1084         SFC_ASSERT(tx_queue_id < sa->txq_count);
1085
1086         txq_info = &sa->txq_info[tx_queue_id];
1087         SFC_ASSERT(txq_info->txq != NULL);
1088
1089         memset(qinfo, 0, sizeof(*qinfo));
1090
1091         qinfo->conf.offloads = txq_info->txq->offloads;
1092         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1093         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1094         qinfo->nb_desc = txq_info->entries;
1095
1096         sfc_adapter_unlock(sa);
1097 }
1098
1099 static uint32_t
1100 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1101 {
1102         struct sfc_adapter *sa = dev->data->dev_private;
1103
1104         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1105
1106         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1107 }
1108
1109 static int
1110 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1111 {
1112         struct sfc_dp_rxq *dp_rxq = queue;
1113
1114         return sfc_rx_qdesc_done(dp_rxq, offset);
1115 }
1116
1117 static int
1118 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1119 {
1120         struct sfc_dp_rxq *dp_rxq = queue;
1121         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1122
1123         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1124 }
1125
1126 static int
1127 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1128 {
1129         struct sfc_dp_txq *dp_txq = queue;
1130         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1131
1132         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1133 }
1134
1135 static int
1136 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1137 {
1138         struct sfc_adapter *sa = dev->data->dev_private;
1139         int rc;
1140
1141         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1142
1143         sfc_adapter_lock(sa);
1144
1145         rc = EINVAL;
1146         if (sa->state != SFC_ADAPTER_STARTED)
1147                 goto fail_not_started;
1148
1149         if (sa->rxq_info[rx_queue_id].rxq == NULL)
1150                 goto fail_not_setup;
1151
1152         rc = sfc_rx_qstart(sa, rx_queue_id);
1153         if (rc != 0)
1154                 goto fail_rx_qstart;
1155
1156         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1157
1158         sfc_adapter_unlock(sa);
1159
1160         return 0;
1161
1162 fail_rx_qstart:
1163 fail_not_setup:
1164 fail_not_started:
1165         sfc_adapter_unlock(sa);
1166         SFC_ASSERT(rc > 0);
1167         return -rc;
1168 }
1169
1170 static int
1171 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1172 {
1173         struct sfc_adapter *sa = dev->data->dev_private;
1174
1175         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1176
1177         sfc_adapter_lock(sa);
1178         sfc_rx_qstop(sa, rx_queue_id);
1179
1180         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1181
1182         sfc_adapter_unlock(sa);
1183
1184         return 0;
1185 }
1186
1187 static int
1188 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1189 {
1190         struct sfc_adapter *sa = dev->data->dev_private;
1191         int rc;
1192
1193         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1194
1195         sfc_adapter_lock(sa);
1196
1197         rc = EINVAL;
1198         if (sa->state != SFC_ADAPTER_STARTED)
1199                 goto fail_not_started;
1200
1201         if (sa->txq_info[tx_queue_id].txq == NULL)
1202                 goto fail_not_setup;
1203
1204         rc = sfc_tx_qstart(sa, tx_queue_id);
1205         if (rc != 0)
1206                 goto fail_tx_qstart;
1207
1208         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1209
1210         sfc_adapter_unlock(sa);
1211         return 0;
1212
1213 fail_tx_qstart:
1214
1215 fail_not_setup:
1216 fail_not_started:
1217         sfc_adapter_unlock(sa);
1218         SFC_ASSERT(rc > 0);
1219         return -rc;
1220 }
1221
1222 static int
1223 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1224 {
1225         struct sfc_adapter *sa = dev->data->dev_private;
1226
1227         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1228
1229         sfc_adapter_lock(sa);
1230
1231         sfc_tx_qstop(sa, tx_queue_id);
1232
1233         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1234
1235         sfc_adapter_unlock(sa);
1236         return 0;
1237 }
1238
1239 static efx_tunnel_protocol_t
1240 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1241 {
1242         switch (rte_type) {
1243         case RTE_TUNNEL_TYPE_VXLAN:
1244                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1245         case RTE_TUNNEL_TYPE_GENEVE:
1246                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1247         default:
1248                 return EFX_TUNNEL_NPROTOS;
1249         }
1250 }
1251
1252 enum sfc_udp_tunnel_op_e {
1253         SFC_UDP_TUNNEL_ADD_PORT,
1254         SFC_UDP_TUNNEL_DEL_PORT,
1255 };
1256
1257 static int
1258 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1259                       struct rte_eth_udp_tunnel *tunnel_udp,
1260                       enum sfc_udp_tunnel_op_e op)
1261 {
1262         struct sfc_adapter *sa = dev->data->dev_private;
1263         efx_tunnel_protocol_t tunnel_proto;
1264         int rc;
1265
1266         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1267                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1268                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1269                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1270
1271         tunnel_proto =
1272                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1273         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1274                 rc = ENOTSUP;
1275                 goto fail_bad_proto;
1276         }
1277
1278         sfc_adapter_lock(sa);
1279
1280         switch (op) {
1281         case SFC_UDP_TUNNEL_ADD_PORT:
1282                 rc = efx_tunnel_config_udp_add(sa->nic,
1283                                                tunnel_udp->udp_port,
1284                                                tunnel_proto);
1285                 break;
1286         case SFC_UDP_TUNNEL_DEL_PORT:
1287                 rc = efx_tunnel_config_udp_remove(sa->nic,
1288                                                   tunnel_udp->udp_port,
1289                                                   tunnel_proto);
1290                 break;
1291         default:
1292                 rc = EINVAL;
1293                 goto fail_bad_op;
1294         }
1295
1296         if (rc != 0)
1297                 goto fail_op;
1298
1299         if (sa->state == SFC_ADAPTER_STARTED) {
1300                 rc = efx_tunnel_reconfigure(sa->nic);
1301                 if (rc == EAGAIN) {
1302                         /*
1303                          * Configuration is accepted by FW and MC reboot
1304                          * is initiated to apply the changes. MC reboot
1305                          * will be handled in a usual way (MC reboot
1306                          * event on management event queue and adapter
1307                          * restart).
1308                          */
1309                         rc = 0;
1310                 } else if (rc != 0) {
1311                         goto fail_reconfigure;
1312                 }
1313         }
1314
1315         sfc_adapter_unlock(sa);
1316         return 0;
1317
1318 fail_reconfigure:
1319         /* Remove/restore entry since the change makes the trouble */
1320         switch (op) {
1321         case SFC_UDP_TUNNEL_ADD_PORT:
1322                 (void)efx_tunnel_config_udp_remove(sa->nic,
1323                                                    tunnel_udp->udp_port,
1324                                                    tunnel_proto);
1325                 break;
1326         case SFC_UDP_TUNNEL_DEL_PORT:
1327                 (void)efx_tunnel_config_udp_add(sa->nic,
1328                                                 tunnel_udp->udp_port,
1329                                                 tunnel_proto);
1330                 break;
1331         }
1332
1333 fail_op:
1334 fail_bad_op:
1335         sfc_adapter_unlock(sa);
1336
1337 fail_bad_proto:
1338         SFC_ASSERT(rc > 0);
1339         return -rc;
1340 }
1341
1342 static int
1343 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1344                             struct rte_eth_udp_tunnel *tunnel_udp)
1345 {
1346         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1347 }
1348
1349 static int
1350 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1351                             struct rte_eth_udp_tunnel *tunnel_udp)
1352 {
1353         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1354 }
1355
1356 static int
1357 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1358                           struct rte_eth_rss_conf *rss_conf)
1359 {
1360         struct sfc_adapter *sa = dev->data->dev_private;
1361         struct sfc_rss *rss = &sa->rss;
1362         struct sfc_port *port = &sa->port;
1363
1364         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1365                 return -ENOTSUP;
1366
1367         if (rss->channels == 0)
1368                 return -EINVAL;
1369
1370         sfc_adapter_lock(sa);
1371
1372         /*
1373          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1374          * hence, conversion is done here to derive a correct set of ETH_RSS
1375          * flags which corresponds to the active EFX configuration stored
1376          * locally in 'sfc_adapter' and kept up-to-date
1377          */
1378         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1379         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1380         if (rss_conf->rss_key != NULL)
1381                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1382
1383         sfc_adapter_unlock(sa);
1384
1385         return 0;
1386 }
1387
1388 static int
1389 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1390                         struct rte_eth_rss_conf *rss_conf)
1391 {
1392         struct sfc_adapter *sa = dev->data->dev_private;
1393         struct sfc_rss *rss = &sa->rss;
1394         struct sfc_port *port = &sa->port;
1395         unsigned int efx_hash_types;
1396         int rc = 0;
1397
1398         if (port->isolated)
1399                 return -ENOTSUP;
1400
1401         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1402                 sfc_err(sa, "RSS is not available");
1403                 return -ENOTSUP;
1404         }
1405
1406         if (rss->channels == 0) {
1407                 sfc_err(sa, "RSS is not configured");
1408                 return -EINVAL;
1409         }
1410
1411         if ((rss_conf->rss_key != NULL) &&
1412             (rss_conf->rss_key_len != sizeof(rss->key))) {
1413                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1414                         sizeof(rss->key));
1415                 return -EINVAL;
1416         }
1417
1418         sfc_adapter_lock(sa);
1419
1420         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1421         if (rc != 0)
1422                 goto fail_rx_hf_rte_to_efx;
1423
1424         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1425                                    rss->hash_alg, efx_hash_types, B_TRUE);
1426         if (rc != 0)
1427                 goto fail_scale_mode_set;
1428
1429         if (rss_conf->rss_key != NULL) {
1430                 if (sa->state == SFC_ADAPTER_STARTED) {
1431                         rc = efx_rx_scale_key_set(sa->nic,
1432                                                   EFX_RSS_CONTEXT_DEFAULT,
1433                                                   rss_conf->rss_key,
1434                                                   sizeof(rss->key));
1435                         if (rc != 0)
1436                                 goto fail_scale_key_set;
1437                 }
1438
1439                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1440         }
1441
1442         rss->hash_types = efx_hash_types;
1443
1444         sfc_adapter_unlock(sa);
1445
1446         return 0;
1447
1448 fail_scale_key_set:
1449         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1450                                   EFX_RX_HASHALG_TOEPLITZ,
1451                                   rss->hash_types, B_TRUE) != 0)
1452                 sfc_err(sa, "failed to restore RSS mode");
1453
1454 fail_scale_mode_set:
1455 fail_rx_hf_rte_to_efx:
1456         sfc_adapter_unlock(sa);
1457         return -rc;
1458 }
1459
1460 static int
1461 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1462                        struct rte_eth_rss_reta_entry64 *reta_conf,
1463                        uint16_t reta_size)
1464 {
1465         struct sfc_adapter *sa = dev->data->dev_private;
1466         struct sfc_rss *rss = &sa->rss;
1467         struct sfc_port *port = &sa->port;
1468         int entry;
1469
1470         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1471                 return -ENOTSUP;
1472
1473         if (rss->channels == 0)
1474                 return -EINVAL;
1475
1476         if (reta_size != EFX_RSS_TBL_SIZE)
1477                 return -EINVAL;
1478
1479         sfc_adapter_lock(sa);
1480
1481         for (entry = 0; entry < reta_size; entry++) {
1482                 int grp = entry / RTE_RETA_GROUP_SIZE;
1483                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1484
1485                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1486                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1487         }
1488
1489         sfc_adapter_unlock(sa);
1490
1491         return 0;
1492 }
1493
1494 static int
1495 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1496                         struct rte_eth_rss_reta_entry64 *reta_conf,
1497                         uint16_t reta_size)
1498 {
1499         struct sfc_adapter *sa = dev->data->dev_private;
1500         struct sfc_rss *rss = &sa->rss;
1501         struct sfc_port *port = &sa->port;
1502         unsigned int *rss_tbl_new;
1503         uint16_t entry;
1504         int rc = 0;
1505
1506
1507         if (port->isolated)
1508                 return -ENOTSUP;
1509
1510         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1511                 sfc_err(sa, "RSS is not available");
1512                 return -ENOTSUP;
1513         }
1514
1515         if (rss->channels == 0) {
1516                 sfc_err(sa, "RSS is not configured");
1517                 return -EINVAL;
1518         }
1519
1520         if (reta_size != EFX_RSS_TBL_SIZE) {
1521                 sfc_err(sa, "RETA size is wrong (should be %u)",
1522                         EFX_RSS_TBL_SIZE);
1523                 return -EINVAL;
1524         }
1525
1526         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1527         if (rss_tbl_new == NULL)
1528                 return -ENOMEM;
1529
1530         sfc_adapter_lock(sa);
1531
1532         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1533
1534         for (entry = 0; entry < reta_size; entry++) {
1535                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1536                 struct rte_eth_rss_reta_entry64 *grp;
1537
1538                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1539
1540                 if (grp->mask & (1ull << grp_idx)) {
1541                         if (grp->reta[grp_idx] >= rss->channels) {
1542                                 rc = EINVAL;
1543                                 goto bad_reta_entry;
1544                         }
1545                         rss_tbl_new[entry] = grp->reta[grp_idx];
1546                 }
1547         }
1548
1549         if (sa->state == SFC_ADAPTER_STARTED) {
1550                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1551                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1552                 if (rc != 0)
1553                         goto fail_scale_tbl_set;
1554         }
1555
1556         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1557
1558 fail_scale_tbl_set:
1559 bad_reta_entry:
1560         sfc_adapter_unlock(sa);
1561
1562         rte_free(rss_tbl_new);
1563
1564         SFC_ASSERT(rc >= 0);
1565         return -rc;
1566 }
1567
1568 static int
1569 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1570                     enum rte_filter_op filter_op,
1571                     void *arg)
1572 {
1573         struct sfc_adapter *sa = dev->data->dev_private;
1574         int rc = ENOTSUP;
1575
1576         sfc_log_init(sa, "entry");
1577
1578         switch (filter_type) {
1579         case RTE_ETH_FILTER_NONE:
1580                 sfc_err(sa, "Global filters configuration not supported");
1581                 break;
1582         case RTE_ETH_FILTER_MACVLAN:
1583                 sfc_err(sa, "MACVLAN filters not supported");
1584                 break;
1585         case RTE_ETH_FILTER_ETHERTYPE:
1586                 sfc_err(sa, "EtherType filters not supported");
1587                 break;
1588         case RTE_ETH_FILTER_FLEXIBLE:
1589                 sfc_err(sa, "Flexible filters not supported");
1590                 break;
1591         case RTE_ETH_FILTER_SYN:
1592                 sfc_err(sa, "SYN filters not supported");
1593                 break;
1594         case RTE_ETH_FILTER_NTUPLE:
1595                 sfc_err(sa, "NTUPLE filters not supported");
1596                 break;
1597         case RTE_ETH_FILTER_TUNNEL:
1598                 sfc_err(sa, "Tunnel filters not supported");
1599                 break;
1600         case RTE_ETH_FILTER_FDIR:
1601                 sfc_err(sa, "Flow Director filters not supported");
1602                 break;
1603         case RTE_ETH_FILTER_HASH:
1604                 sfc_err(sa, "Hash filters not supported");
1605                 break;
1606         case RTE_ETH_FILTER_GENERIC:
1607                 if (filter_op != RTE_ETH_FILTER_GET) {
1608                         rc = EINVAL;
1609                 } else {
1610                         *(const void **)arg = &sfc_flow_ops;
1611                         rc = 0;
1612                 }
1613                 break;
1614         default:
1615                 sfc_err(sa, "Unknown filter type %u", filter_type);
1616                 break;
1617         }
1618
1619         sfc_log_init(sa, "exit: %d", -rc);
1620         SFC_ASSERT(rc >= 0);
1621         return -rc;
1622 }
1623
1624 static int
1625 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1626 {
1627         struct sfc_adapter *sa = dev->data->dev_private;
1628
1629         /*
1630          * If Rx datapath does not provide callback to check mempool,
1631          * all pools are supported.
1632          */
1633         if (sa->dp_rx->pool_ops_supported == NULL)
1634                 return 1;
1635
1636         return sa->dp_rx->pool_ops_supported(pool);
1637 }
1638
1639 static const struct eth_dev_ops sfc_eth_dev_ops = {
1640         .dev_configure                  = sfc_dev_configure,
1641         .dev_start                      = sfc_dev_start,
1642         .dev_stop                       = sfc_dev_stop,
1643         .dev_set_link_up                = sfc_dev_set_link_up,
1644         .dev_set_link_down              = sfc_dev_set_link_down,
1645         .dev_close                      = sfc_dev_close,
1646         .promiscuous_enable             = sfc_dev_promisc_enable,
1647         .promiscuous_disable            = sfc_dev_promisc_disable,
1648         .allmulticast_enable            = sfc_dev_allmulti_enable,
1649         .allmulticast_disable           = sfc_dev_allmulti_disable,
1650         .link_update                    = sfc_dev_link_update,
1651         .stats_get                      = sfc_stats_get,
1652         .stats_reset                    = sfc_stats_reset,
1653         .xstats_get                     = sfc_xstats_get,
1654         .xstats_reset                   = sfc_stats_reset,
1655         .xstats_get_names               = sfc_xstats_get_names,
1656         .dev_infos_get                  = sfc_dev_infos_get,
1657         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1658         .mtu_set                        = sfc_dev_set_mtu,
1659         .rx_queue_start                 = sfc_rx_queue_start,
1660         .rx_queue_stop                  = sfc_rx_queue_stop,
1661         .tx_queue_start                 = sfc_tx_queue_start,
1662         .tx_queue_stop                  = sfc_tx_queue_stop,
1663         .rx_queue_setup                 = sfc_rx_queue_setup,
1664         .rx_queue_release               = sfc_rx_queue_release,
1665         .rx_queue_count                 = sfc_rx_queue_count,
1666         .rx_descriptor_done             = sfc_rx_descriptor_done,
1667         .rx_descriptor_status           = sfc_rx_descriptor_status,
1668         .tx_descriptor_status           = sfc_tx_descriptor_status,
1669         .tx_queue_setup                 = sfc_tx_queue_setup,
1670         .tx_queue_release               = sfc_tx_queue_release,
1671         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1672         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1673         .mac_addr_set                   = sfc_mac_addr_set,
1674         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1675         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1676         .reta_update                    = sfc_dev_rss_reta_update,
1677         .reta_query                     = sfc_dev_rss_reta_query,
1678         .rss_hash_update                = sfc_dev_rss_hash_update,
1679         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1680         .filter_ctrl                    = sfc_dev_filter_ctrl,
1681         .set_mc_addr_list               = sfc_set_mc_addr_list,
1682         .rxq_info_get                   = sfc_rx_queue_info_get,
1683         .txq_info_get                   = sfc_tx_queue_info_get,
1684         .fw_version_get                 = sfc_fw_version_get,
1685         .xstats_get_by_id               = sfc_xstats_get_by_id,
1686         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1687         .pool_ops_supported             = sfc_pool_ops_supported,
1688 };
1689
1690 /**
1691  * Duplicate a string in potentially shared memory required for
1692  * multi-process support.
1693  *
1694  * strdup() allocates from process-local heap/memory.
1695  */
1696 static char *
1697 sfc_strdup(const char *str)
1698 {
1699         size_t size;
1700         char *copy;
1701
1702         if (str == NULL)
1703                 return NULL;
1704
1705         size = strlen(str) + 1;
1706         copy = rte_malloc(__func__, size, 0);
1707         if (copy != NULL)
1708                 rte_memcpy(copy, str, size);
1709
1710         return copy;
1711 }
1712
1713 static int
1714 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1715 {
1716         struct sfc_adapter *sa = dev->data->dev_private;
1717         const efx_nic_cfg_t *encp;
1718         unsigned int avail_caps = 0;
1719         const char *rx_name = NULL;
1720         const char *tx_name = NULL;
1721         int rc;
1722
1723         switch (sa->family) {
1724         case EFX_FAMILY_HUNTINGTON:
1725         case EFX_FAMILY_MEDFORD:
1726         case EFX_FAMILY_MEDFORD2:
1727                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1728                 break;
1729         default:
1730                 break;
1731         }
1732
1733         encp = efx_nic_cfg_get(sa->nic);
1734         if (encp->enc_rx_es_super_buffer_supported)
1735                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1736
1737         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1738                                 sfc_kvarg_string_handler, &rx_name);
1739         if (rc != 0)
1740                 goto fail_kvarg_rx_datapath;
1741
1742         if (rx_name != NULL) {
1743                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1744                 if (sa->dp_rx == NULL) {
1745                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1746                         rc = ENOENT;
1747                         goto fail_dp_rx;
1748                 }
1749                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1750                         sfc_err(sa,
1751                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1752                                 rx_name);
1753                         rc = EINVAL;
1754                         goto fail_dp_rx_caps;
1755                 }
1756         } else {
1757                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1758                 if (sa->dp_rx == NULL) {
1759                         sfc_err(sa, "Rx datapath by caps %#x not found",
1760                                 avail_caps);
1761                         rc = ENOENT;
1762                         goto fail_dp_rx;
1763                 }
1764         }
1765
1766         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1767         if (sa->dp_rx_name == NULL) {
1768                 rc = ENOMEM;
1769                 goto fail_dp_rx_name;
1770         }
1771
1772         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1773
1774         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1775
1776         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1777                                 sfc_kvarg_string_handler, &tx_name);
1778         if (rc != 0)
1779                 goto fail_kvarg_tx_datapath;
1780
1781         if (tx_name != NULL) {
1782                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1783                 if (sa->dp_tx == NULL) {
1784                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1785                         rc = ENOENT;
1786                         goto fail_dp_tx;
1787                 }
1788                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1789                         sfc_err(sa,
1790                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1791                                 tx_name);
1792                         rc = EINVAL;
1793                         goto fail_dp_tx_caps;
1794                 }
1795         } else {
1796                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1797                 if (sa->dp_tx == NULL) {
1798                         sfc_err(sa, "Tx datapath by caps %#x not found",
1799                                 avail_caps);
1800                         rc = ENOENT;
1801                         goto fail_dp_tx;
1802                 }
1803         }
1804
1805         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1806         if (sa->dp_tx_name == NULL) {
1807                 rc = ENOMEM;
1808                 goto fail_dp_tx_name;
1809         }
1810
1811         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1812
1813         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1814
1815         dev->dev_ops = &sfc_eth_dev_ops;
1816
1817         return 0;
1818
1819 fail_dp_tx_name:
1820 fail_dp_tx_caps:
1821         sa->dp_tx = NULL;
1822
1823 fail_dp_tx:
1824 fail_kvarg_tx_datapath:
1825         rte_free(sa->dp_rx_name);
1826         sa->dp_rx_name = NULL;
1827
1828 fail_dp_rx_name:
1829 fail_dp_rx_caps:
1830         sa->dp_rx = NULL;
1831
1832 fail_dp_rx:
1833 fail_kvarg_rx_datapath:
1834         return rc;
1835 }
1836
1837 static void
1838 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1839 {
1840         struct sfc_adapter *sa = dev->data->dev_private;
1841
1842         dev->dev_ops = NULL;
1843         dev->rx_pkt_burst = NULL;
1844         dev->tx_pkt_burst = NULL;
1845
1846         rte_free(sa->dp_tx_name);
1847         sa->dp_tx_name = NULL;
1848         sa->dp_tx = NULL;
1849
1850         rte_free(sa->dp_rx_name);
1851         sa->dp_rx_name = NULL;
1852         sa->dp_rx = NULL;
1853 }
1854
1855 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1856         .rxq_info_get                   = sfc_rx_queue_info_get,
1857         .txq_info_get                   = sfc_tx_queue_info_get,
1858 };
1859
1860 static int
1861 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1862 {
1863         /*
1864          * Device private data has really many process-local pointers.
1865          * Below code should be extremely careful to use data located
1866          * in shared memory only.
1867          */
1868         struct sfc_adapter *sa = dev->data->dev_private;
1869         const struct sfc_dp_rx *dp_rx;
1870         const struct sfc_dp_tx *dp_tx;
1871         int rc;
1872
1873         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1874         if (dp_rx == NULL) {
1875                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1876                 rc = ENOENT;
1877                 goto fail_dp_rx;
1878         }
1879         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1880                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1881                         sa->dp_tx_name);
1882                 rc = EINVAL;
1883                 goto fail_dp_rx_multi_process;
1884         }
1885
1886         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1887         if (dp_tx == NULL) {
1888                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1889                 rc = ENOENT;
1890                 goto fail_dp_tx;
1891         }
1892         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1893                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1894                         sa->dp_tx_name);
1895                 rc = EINVAL;
1896                 goto fail_dp_tx_multi_process;
1897         }
1898
1899         dev->rx_pkt_burst = dp_rx->pkt_burst;
1900         dev->tx_pkt_burst = dp_tx->pkt_burst;
1901         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1902
1903         return 0;
1904
1905 fail_dp_tx_multi_process:
1906 fail_dp_tx:
1907 fail_dp_rx_multi_process:
1908 fail_dp_rx:
1909         return rc;
1910 }
1911
1912 static void
1913 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1914 {
1915         dev->dev_ops = NULL;
1916         dev->tx_pkt_burst = NULL;
1917         dev->rx_pkt_burst = NULL;
1918 }
1919
1920 static void
1921 sfc_register_dp(void)
1922 {
1923         /* Register once */
1924         if (TAILQ_EMPTY(&sfc_dp_head)) {
1925                 /* Prefer EF10 datapath */
1926                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1927                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1928                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1929
1930                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1931                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1932                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1933         }
1934 }
1935
1936 static int
1937 sfc_eth_dev_init(struct rte_eth_dev *dev)
1938 {
1939         struct sfc_adapter *sa = dev->data->dev_private;
1940         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1941         int rc;
1942         const efx_nic_cfg_t *encp;
1943         const struct ether_addr *from;
1944
1945         sfc_register_dp();
1946
1947         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1948                 return -sfc_eth_dev_secondary_set_ops(dev);
1949
1950         /* Required for logging */
1951         sa->pci_addr = pci_dev->addr;
1952         sa->port_id = dev->data->port_id;
1953
1954         sa->eth_dev = dev;
1955
1956         /* Copy PCI device info to the dev->data */
1957         rte_eth_copy_pci_info(dev, pci_dev);
1958
1959         sa->logtype_main = sfc_register_logtype(sa, SFC_LOGTYPE_MAIN_STR,
1960                                                 RTE_LOG_NOTICE);
1961
1962         rc = sfc_kvargs_parse(sa);
1963         if (rc != 0)
1964                 goto fail_kvargs_parse;
1965
1966         sfc_log_init(sa, "entry");
1967
1968         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1969         if (dev->data->mac_addrs == NULL) {
1970                 rc = ENOMEM;
1971                 goto fail_mac_addrs;
1972         }
1973
1974         sfc_adapter_lock_init(sa);
1975         sfc_adapter_lock(sa);
1976
1977         sfc_log_init(sa, "probing");
1978         rc = sfc_probe(sa);
1979         if (rc != 0)
1980                 goto fail_probe;
1981
1982         sfc_log_init(sa, "set device ops");
1983         rc = sfc_eth_dev_set_ops(dev);
1984         if (rc != 0)
1985                 goto fail_set_ops;
1986
1987         sfc_log_init(sa, "attaching");
1988         rc = sfc_attach(sa);
1989         if (rc != 0)
1990                 goto fail_attach;
1991
1992         encp = efx_nic_cfg_get(sa->nic);
1993
1994         /*
1995          * The arguments are really reverse order in comparison to
1996          * Linux kernel. Copy from NIC config to Ethernet device data.
1997          */
1998         from = (const struct ether_addr *)(encp->enc_mac_addr);
1999         ether_addr_copy(from, &dev->data->mac_addrs[0]);
2000
2001         sfc_adapter_unlock(sa);
2002
2003         sfc_log_init(sa, "done");
2004         return 0;
2005
2006 fail_attach:
2007         sfc_eth_dev_clear_ops(dev);
2008
2009 fail_set_ops:
2010         sfc_unprobe(sa);
2011
2012 fail_probe:
2013         sfc_adapter_unlock(sa);
2014         sfc_adapter_lock_fini(sa);
2015         rte_free(dev->data->mac_addrs);
2016         dev->data->mac_addrs = NULL;
2017
2018 fail_mac_addrs:
2019         sfc_kvargs_cleanup(sa);
2020
2021 fail_kvargs_parse:
2022         sfc_log_init(sa, "failed %d", rc);
2023         SFC_ASSERT(rc > 0);
2024         return -rc;
2025 }
2026
2027 static int
2028 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2029 {
2030         struct sfc_adapter *sa;
2031
2032         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2033                 sfc_eth_dev_secondary_clear_ops(dev);
2034                 return 0;
2035         }
2036
2037         sa = dev->data->dev_private;
2038         sfc_log_init(sa, "entry");
2039
2040         sfc_adapter_lock(sa);
2041
2042         sfc_eth_dev_clear_ops(dev);
2043
2044         sfc_detach(sa);
2045         sfc_unprobe(sa);
2046
2047         rte_free(dev->data->mac_addrs);
2048         dev->data->mac_addrs = NULL;
2049
2050         sfc_kvargs_cleanup(sa);
2051
2052         sfc_adapter_unlock(sa);
2053         sfc_adapter_lock_fini(sa);
2054
2055         sfc_log_init(sa, "done");
2056
2057         /* Required for logging, so cleanup last */
2058         sa->eth_dev = NULL;
2059         return 0;
2060 }
2061
2062 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2063         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2064         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2065         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2066         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2067         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2068         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2069         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2070         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2071         { .vendor_id = 0 /* sentinel */ }
2072 };
2073
2074 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2075         struct rte_pci_device *pci_dev)
2076 {
2077         return rte_eth_dev_pci_generic_probe(pci_dev,
2078                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2079 }
2080
2081 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2082 {
2083         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2084 }
2085
2086 static struct rte_pci_driver sfc_efx_pmd = {
2087         .id_table = pci_id_sfc_efx_map,
2088         .drv_flags =
2089                 RTE_PCI_DRV_INTR_LSC |
2090                 RTE_PCI_DRV_NEED_MAPPING,
2091         .probe = sfc_eth_dev_pci_probe,
2092         .remove = sfc_eth_dev_pci_remove,
2093 };
2094
2095 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2096 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2097 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2098 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2099         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2100         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2101         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2102         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2103         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2104         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2105
2106 RTE_INIT(sfc_driver_register_logtype)
2107 {
2108         int ret;
2109
2110         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2111                                                    RTE_LOG_NOTICE);
2112         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2113 }