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         sfc_rx_qfini(sa, sw_index);
 448
 449         sfc_adapter_unlock(sa);
 450 }
 451
 452 static int
 453 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 454                    uint16_t nb_tx_desc, unsigned int socket_id,
 455                    const struct rte_eth_txconf *tx_conf)
 456 {
 457         struct sfc_adapter *sa = dev->data->dev_private;
 458         int rc;
 459
 460         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 461                      tx_queue_id, nb_tx_desc, socket_id);
 462
 463         sfc_adapter_lock(sa);
 464
 465         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 466         if (rc != 0)
 467                 goto fail_tx_qinit;
 468
 469         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 470
 471         sfc_adapter_unlock(sa);
 472         return 0;
 473
 474 fail_tx_qinit:
 475         sfc_adapter_unlock(sa);
 476         SFC_ASSERT(rc > 0);
 477         return -rc;
 478 }
 479
 480 static void
 481 sfc_tx_queue_release(void *queue)
 482 {
 483         struct sfc_dp_txq *dp_txq = queue;
 484         struct sfc_txq *txq;
 485         unsigned int sw_index;
 486         struct sfc_adapter *sa;
 487
 488         if (dp_txq == NULL)
 489                 return;
 490
 491         txq = sfc_txq_by_dp_txq(dp_txq);
 492         sw_index = sfc_txq_sw_index(txq);
 493
 494         SFC_ASSERT(txq->evq != NULL);
 495         sa = txq->evq->sa;
 496
 497         sfc_log_init(sa, "TxQ = %u", sw_index);
 498
 499         sfc_adapter_lock(sa);
 500
 501         sfc_tx_qfini(sa, sw_index);
 502
 503         sfc_adapter_unlock(sa);
 504 }
 505
 506 /*
 507  * Some statistics are computed as A - B where A and B each increase
 508  * monotonically with some hardware counter(s) and the counters are read
 509  * asynchronously.
 510  *
 511  * If packet X is counted in A, but not counted in B yet, computed value is
 512  * greater than real.
 513  *
 514  * If packet X is not counted in A at the moment of reading the counter,
 515  * but counted in B at the moment of reading the counter, computed value
 516  * is less than real.
 517  *
 518  * However, counter which grows backward is worse evil than slightly wrong
 519  * value. So, let's try to guarantee that it never happens except may be
 520  * the case when the MAC stats are zeroed as a result of a NIC reset.
 521  */
 522 static void
 523 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 524 {
 525         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 526                 *stat = newval;
 527 }
 528
 529 static int
 530 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 531 {
 532         struct sfc_adapter *sa = dev->data->dev_private;
 533         struct sfc_port *port = &sa->port;
 534         uint64_t *mac_stats;
 535         int ret;
 536
 537         rte_spinlock_lock(&port->mac_stats_lock);
 538
 539         ret = sfc_port_update_mac_stats(sa);
 540         if (ret != 0)
 541                 goto unlock;
 542
 543         mac_stats = port->mac_stats_buf;
 544
 545         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 546                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 547                 stats->ipackets =
 548                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 549                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 550                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 551                 stats->opackets =
 552                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 553                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 554                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 555                 stats->ibytes =
 556                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 557                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 558                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 559                 stats->obytes =
 560                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 561                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 562                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 563                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 564                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 565         } else {
 566                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 567                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 568                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 569                 /*
 570                  * Take into account stats which are whenever supported
 571                  * on EF10. If some stat is not supported by current
 572                  * firmware variant or HW revision, it is guaranteed
 573                  * to be zero in mac_stats.
 574                  */
 575                 stats->imissed =
 576                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 577                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 578                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 579                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 580                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 581                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 582                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 583                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 584                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 585                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 586                 stats->ierrors =
 587                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 588                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 589                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 590                 /* no oerrors counters supported on EF10 */
 591
 592                 /* Exclude missed, errors and pauses from Rx packets */
 593                 sfc_update_diff_stat(&port->ipackets,
 594                         mac_stats[EFX_MAC_RX_PKTS] -
 595                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 596                         stats->imissed - stats->ierrors);
 597                 stats->ipackets = port->ipackets;
 598         }
 599
 600 unlock:
 601         rte_spinlock_unlock(&port->mac_stats_lock);
 602         SFC_ASSERT(ret >= 0);
 603         return -ret;
 604 }
 605
 606 static void
 607 sfc_stats_reset(struct rte_eth_dev *dev)
 608 {
 609         struct sfc_adapter *sa = dev->data->dev_private;
 610         struct sfc_port *port = &sa->port;
 611         int rc;
 612
 613         if (sa->state != SFC_ADAPTER_STARTED) {
 614                 /*
 615                  * The operation cannot be done if port is not started; it
 616                  * will be scheduled to be done during the next port start
 617                  */
 618                 port->mac_stats_reset_pending = B_TRUE;
 619                 return;
 620         }
 621
 622         rc = sfc_port_reset_mac_stats(sa);
 623         if (rc != 0)
 624                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 625 }
 626
 627 static int
 628 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 629                unsigned int xstats_count)
 630 {
 631         struct sfc_adapter *sa = dev->data->dev_private;
 632         struct sfc_port *port = &sa->port;
 633         uint64_t *mac_stats;
 634         int rc;
 635         unsigned int i;
 636         int nstats = 0;
 637
 638         rte_spinlock_lock(&port->mac_stats_lock);
 639
 640         rc = sfc_port_update_mac_stats(sa);
 641         if (rc != 0) {
 642                 SFC_ASSERT(rc > 0);
 643                 nstats = -rc;
 644                 goto unlock;
 645         }
 646
 647         mac_stats = port->mac_stats_buf;
 648
 649         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 650                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 651                         if (xstats != NULL && nstats < (int)xstats_count) {
 652                                 xstats[nstats].id = nstats;
 653                                 xstats[nstats].value = mac_stats[i];
 654                         }
 655                         nstats++;
 656                 }
 657         }
 658
 659 unlock:
 660         rte_spinlock_unlock(&port->mac_stats_lock);
 661
 662         return nstats;
 663 }
 664
 665 static int
 666 sfc_xstats_get_names(struct rte_eth_dev *dev,
 667                      struct rte_eth_xstat_name *xstats_names,
 668                      unsigned int xstats_count)
 669 {
 670         struct sfc_adapter *sa = dev->data->dev_private;
 671         struct sfc_port *port = &sa->port;
 672         unsigned int i;
 673         unsigned int nstats = 0;
 674
 675         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 676                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 677                         if (xstats_names != NULL && nstats < xstats_count)
 678                                 strlcpy(xstats_names[nstats].name,
 679                                         efx_mac_stat_name(sa->nic, i),
 680                                         sizeof(xstats_names[0].name));
 681                         nstats++;
 682                 }
 683         }
 684
 685         return nstats;
 686 }
 687
 688 static int
 689 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 690                      uint64_t *values, unsigned int n)
 691 {
 692         struct sfc_adapter *sa = dev->data->dev_private;
 693         struct sfc_port *port = &sa->port;
 694         uint64_t *mac_stats;
 695         unsigned int nb_supported = 0;
 696         unsigned int nb_written = 0;
 697         unsigned int i;
 698         int ret;
 699         int rc;
 700
 701         if (unlikely(values == NULL) ||
 702             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 703                 return port->mac_stats_nb_supported;
 704
 705         rte_spinlock_lock(&port->mac_stats_lock);
 706
 707         rc = sfc_port_update_mac_stats(sa);
 708         if (rc != 0) {
 709                 SFC_ASSERT(rc > 0);
 710                 ret = -rc;
 711                 goto unlock;
 712         }
 713
 714         mac_stats = port->mac_stats_buf;
 715
 716         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 717                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 718                         continue;
 719
 720                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 721                         values[nb_written++] = mac_stats[i];
 722
 723                 ++nb_supported;
 724         }
 725
 726         ret = nb_written;
 727
 728 unlock:
 729         rte_spinlock_unlock(&port->mac_stats_lock);
 730
 731         return ret;
 732 }
 733
 734 static int
 735 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 736                            struct rte_eth_xstat_name *xstats_names,
 737                            const uint64_t *ids, unsigned int size)
 738 {
 739         struct sfc_adapter *sa = dev->data->dev_private;
 740         struct sfc_port *port = &sa->port;
 741         unsigned int nb_supported = 0;
 742         unsigned int nb_written = 0;
 743         unsigned int i;
 744
 745         if (unlikely(xstats_names == NULL) ||
 746             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 747                 return port->mac_stats_nb_supported;
 748
 749         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 750                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 751                         continue;
 752
 753                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 754                         char *name = xstats_names[nb_written++].name;
 755
 756                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 757                                 sizeof(xstats_names[0].name));
 758                 }
 759
 760                 ++nb_supported;
 761         }
 762
 763         return nb_written;
 764 }
 765
 766 static int
 767 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 768 {
 769         struct sfc_adapter *sa = dev->data->dev_private;
 770         unsigned int wanted_fc, link_fc;
 771
 772         memset(fc_conf, 0, sizeof(*fc_conf));
 773
 774         sfc_adapter_lock(sa);
 775
 776         if (sa->state == SFC_ADAPTER_STARTED)
 777                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 778         else
 779                 link_fc = sa->port.flow_ctrl;
 780
 781         switch (link_fc) {
 782         case 0:
 783                 fc_conf->mode = RTE_FC_NONE;
 784                 break;
 785         case EFX_FCNTL_RESPOND:
 786                 fc_conf->mode = RTE_FC_RX_PAUSE;
 787                 break;
 788         case EFX_FCNTL_GENERATE:
 789                 fc_conf->mode = RTE_FC_TX_PAUSE;
 790                 break;
 791         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 792                 fc_conf->mode = RTE_FC_FULL;
 793                 break;
 794         default:
 795                 sfc_err(sa, "%s: unexpected flow control value %#x",
 796                         __func__, link_fc);
 797         }
 798
 799         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 800
 801         sfc_adapter_unlock(sa);
 802
 803         return 0;
 804 }
 805
 806 static int
 807 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 808 {
 809         struct sfc_adapter *sa = dev->data->dev_private;
 810         struct sfc_port *port = &sa->port;
 811         unsigned int fcntl;
 812         int rc;
 813
 814         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 815             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 816             fc_conf->mac_ctrl_frame_fwd != 0) {
 817                 sfc_err(sa, "unsupported flow control settings specified");
 818                 rc = EINVAL;
 819                 goto fail_inval;
 820         }
 821
 822         switch (fc_conf->mode) {
 823         case RTE_FC_NONE:
 824                 fcntl = 0;
 825                 break;
 826         case RTE_FC_RX_PAUSE:
 827                 fcntl = EFX_FCNTL_RESPOND;
 828                 break;
 829         case RTE_FC_TX_PAUSE:
 830                 fcntl = EFX_FCNTL_GENERATE;
 831                 break;
 832         case RTE_FC_FULL:
 833                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 834                 break;
 835         default:
 836                 rc = EINVAL;
 837                 goto fail_inval;
 838         }
 839
 840         sfc_adapter_lock(sa);
 841
 842         if (sa->state == SFC_ADAPTER_STARTED) {
 843                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 844                 if (rc != 0)
 845                         goto fail_mac_fcntl_set;
 846         }
 847
 848         port->flow_ctrl = fcntl;
 849         port->flow_ctrl_autoneg = fc_conf->autoneg;
 850
 851         sfc_adapter_unlock(sa);
 852
 853         return 0;
 854
 855 fail_mac_fcntl_set:
 856         sfc_adapter_unlock(sa);
 857 fail_inval:
 858         SFC_ASSERT(rc > 0);
 859         return -rc;
 860 }
 861
 862 static int
 863 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 864 {
 865         struct sfc_adapter *sa = dev->data->dev_private;
 866         size_t pdu = EFX_MAC_PDU(mtu);
 867         size_t old_pdu;
 868         int rc;
 869
 870         sfc_log_init(sa, "mtu=%u", mtu);
 871
 872         rc = EINVAL;
 873         if (pdu < EFX_MAC_PDU_MIN) {
 874                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 875                         (unsigned int)mtu, (unsigned int)pdu,
 876                         EFX_MAC_PDU_MIN);
 877                 goto fail_inval;
 878         }
 879         if (pdu > EFX_MAC_PDU_MAX) {
 880                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 881                         (unsigned int)mtu, (unsigned int)pdu,
 882                         EFX_MAC_PDU_MAX);
 883                 goto fail_inval;
 884         }
 885
 886         sfc_adapter_lock(sa);
 887
 888         if (pdu != sa->port.pdu) {
 889                 if (sa->state == SFC_ADAPTER_STARTED) {
 890                         sfc_stop(sa);
 891
 892                         old_pdu = sa->port.pdu;
 893                         sa->port.pdu = pdu;
 894                         rc = sfc_start(sa);
 895                         if (rc != 0)
 896                                 goto fail_start;
 897                 } else {
 898                         sa->port.pdu = pdu;
 899                 }
 900         }
 901
 902         /*
 903          * The driver does not use it, but other PMDs update jumbo frame
 904          * flag and max_rx_pkt_len when MTU is set.
 905          */
 906         if (mtu > ETHER_MAX_LEN) {
 907                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 908                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 909         }
 910
 911         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 912
 913         sfc_adapter_unlock(sa);
 914
 915         sfc_log_init(sa, "done");
 916         return 0;
 917
 918 fail_start:
 919         sa->port.pdu = old_pdu;
 920         if (sfc_start(sa) != 0)
 921                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 922                         "PDU max size - port is stopped",
 923                         (unsigned int)pdu, (unsigned int)old_pdu);
 924         sfc_adapter_unlock(sa);
 925
 926 fail_inval:
 927         sfc_log_init(sa, "failed %d", rc);
 928         SFC_ASSERT(rc > 0);
 929         return -rc;
 930 }
 931 static int
 932 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 933 {
 934         struct sfc_adapter *sa = dev->data->dev_private;
 935         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 936         struct sfc_port *port = &sa->port;
 937         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
 938         int rc = 0;
 939
 940         sfc_adapter_lock(sa);
 941
 942         /*
 943          * Copy the address to the device private data so that
 944          * it could be recalled in the case of adapter restart.
 945          */
 946         ether_addr_copy(mac_addr, &port->default_mac_addr);
 947
 948         /*
 949          * Neither of the two following checks can return
 950          * an error. The new MAC address is preserved in
 951          * the device private data and can be activated
 952          * on the next port start if the user prevents
 953          * isolated mode from being enabled.
 954          */
 955         if (port->isolated) {
 956                 sfc_warn(sa, "isolated mode is active on the port");
 957                 sfc_warn(sa, "will not set MAC address");
 958                 goto unlock;
 959         }
 960
 961         if (sa->state != SFC_ADAPTER_STARTED) {
 962                 sfc_notice(sa, "the port is not started");
 963                 sfc_notice(sa, "the new MAC address will be set on port start");
 964
 965                 goto unlock;
 966         }
 967
 968         if (encp->enc_allow_set_mac_with_installed_filters) {
 969                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 970                 if (rc != 0) {
 971                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 972                         goto unlock;
 973                 }
 974
 975                 /*
 976                  * Changing the MAC address by means of MCDI request
 977                  * has no effect on received traffic, therefore
 978                  * we also need to update unicast filters
 979                  */
 980                 rc = sfc_set_rx_mode(sa);
 981                 if (rc != 0) {
 982                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 983                         /* Rollback the old address */
 984                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
 985                         (void)sfc_set_rx_mode(sa);
 986                 }
 987         } else {
 988                 sfc_warn(sa, "cannot set MAC address with filters installed");
 989                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 990                 sfc_warn(sa, "(some traffic may be dropped)");
 991
 992                 /*
 993                  * Since setting MAC address with filters installed is not
 994                  * allowed on the adapter, the new MAC address will be set
 995                  * by means of adapter restart. sfc_start() shall retrieve
 996                  * the new address from the device private data and set it.
 997                  */
 998                 sfc_stop(sa);
 999                 rc = sfc_start(sa);
1000                 if (rc != 0)
1001                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1002         }
1003
1004 unlock:
1005         if (rc != 0)
1006                 ether_addr_copy(old_addr, &port->default_mac_addr);
1007
1008         sfc_adapter_unlock(sa);
1009
1010         SFC_ASSERT(rc >= 0);
1011         return -rc;
1012 }
1013
1014
1015 static int
1016 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1017                      uint32_t nb_mc_addr)
1018 {
1019         struct sfc_adapter *sa = dev->data->dev_private;
1020         struct sfc_port *port = &sa->port;
1021         uint8_t *mc_addrs = port->mcast_addrs;
1022         int rc;
1023         unsigned int i;
1024
1025         if (port->isolated) {
1026                 sfc_err(sa, "isolated mode is active on the port");
1027                 sfc_err(sa, "will not set multicast address list");
1028                 return -ENOTSUP;
1029         }
1030
1031         if (mc_addrs == NULL)
1032                 return -ENOBUFS;
1033
1034         if (nb_mc_addr > port->max_mcast_addrs) {
1035                 sfc_err(sa, "too many multicast addresses: %u > %u",
1036                          nb_mc_addr, port->max_mcast_addrs);
1037                 return -EINVAL;
1038         }
1039
1040         for (i = 0; i < nb_mc_addr; ++i) {
1041                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1042                                  EFX_MAC_ADDR_LEN);
1043                 mc_addrs += EFX_MAC_ADDR_LEN;
1044         }
1045
1046         port->nb_mcast_addrs = nb_mc_addr;
1047
1048         if (sa->state != SFC_ADAPTER_STARTED)
1049                 return 0;
1050
1051         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1052                                         port->nb_mcast_addrs);
1053         if (rc != 0)
1054                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1055
1056         SFC_ASSERT(rc >= 0);
1057         return -rc;
1058 }
1059
1060 /*
1061  * The function is used by the secondary process as well. It must not
1062  * use any process-local pointers from the adapter data.
1063  */
1064 static void
1065 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1066                       struct rte_eth_rxq_info *qinfo)
1067 {
1068         struct sfc_adapter *sa = dev->data->dev_private;
1069         struct sfc_rxq_info *rxq_info;
1070
1071         sfc_adapter_lock(sa);
1072
1073         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1074
1075         rxq_info = &sa->rxq_info[rx_queue_id];
1076
1077         qinfo->mp = rxq_info->refill_mb_pool;
1078         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1079         qinfo->conf.rx_drop_en = 1;
1080         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1081         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1082         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1083                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1084                 qinfo->scattered_rx = 1;
1085         }
1086         qinfo->nb_desc = rxq_info->entries;
1087
1088         sfc_adapter_unlock(sa);
1089 }
1090
1091 /*
1092  * The function is used by the secondary process as well. It must not
1093  * use any process-local pointers from the adapter data.
1094  */
1095 static void
1096 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1097                       struct rte_eth_txq_info *qinfo)
1098 {
1099         struct sfc_adapter *sa = dev->data->dev_private;
1100         struct sfc_txq_info *txq_info;
1101
1102         sfc_adapter_lock(sa);
1103
1104         SFC_ASSERT(tx_queue_id < sa->txq_count);
1105
1106         txq_info = &sa->txq_info[tx_queue_id];
1107
1108         memset(qinfo, 0, sizeof(*qinfo));
1109
1110         qinfo->conf.offloads = txq_info->offloads;
1111         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1112         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1113         qinfo->nb_desc = txq_info->entries;
1114
1115         sfc_adapter_unlock(sa);
1116 }
1117
1118 static uint32_t
1119 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1120 {
1121         struct sfc_adapter *sa = dev->data->dev_private;
1122         struct sfc_rxq *rxq;
1123
1124         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1125         rxq = sa->rxq_info[rx_queue_id].rxq;
1126
1127         if (rxq == NULL || (rxq->state & SFC_RXQ_STARTED) == 0)
1128                 return 0;
1129
1130         return sa->dp_rx->qdesc_npending(rxq->dp);
1131 }
1132
1133 static int
1134 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1135 {
1136         struct sfc_dp_rxq *dp_rxq = queue;
1137         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1138
1139         return offset < rxq->evq->sa->dp_rx->qdesc_npending(dp_rxq);
1140 }
1141
1142 static int
1143 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1144 {
1145         struct sfc_dp_rxq *dp_rxq = queue;
1146         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1147
1148         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1149 }
1150
1151 static int
1152 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1153 {
1154         struct sfc_dp_txq *dp_txq = queue;
1155         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1156
1157         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1158 }
1159
1160 static int
1161 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1162 {
1163         struct sfc_adapter *sa = dev->data->dev_private;
1164         int rc;
1165
1166         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1167
1168         sfc_adapter_lock(sa);
1169
1170         rc = EINVAL;
1171         if (sa->state != SFC_ADAPTER_STARTED)
1172                 goto fail_not_started;
1173
1174         if (sa->rxq_info[rx_queue_id].rxq == NULL)
1175                 goto fail_not_setup;
1176
1177         rc = sfc_rx_qstart(sa, rx_queue_id);
1178         if (rc != 0)
1179                 goto fail_rx_qstart;
1180
1181         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1182
1183         sfc_adapter_unlock(sa);
1184
1185         return 0;
1186
1187 fail_rx_qstart:
1188 fail_not_setup:
1189 fail_not_started:
1190         sfc_adapter_unlock(sa);
1191         SFC_ASSERT(rc > 0);
1192         return -rc;
1193 }
1194
1195 static int
1196 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1197 {
1198         struct sfc_adapter *sa = dev->data->dev_private;
1199
1200         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1201
1202         sfc_adapter_lock(sa);
1203         sfc_rx_qstop(sa, rx_queue_id);
1204
1205         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1206
1207         sfc_adapter_unlock(sa);
1208
1209         return 0;
1210 }
1211
1212 static int
1213 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1214 {
1215         struct sfc_adapter *sa = dev->data->dev_private;
1216         int rc;
1217
1218         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1219
1220         sfc_adapter_lock(sa);
1221
1222         rc = EINVAL;
1223         if (sa->state != SFC_ADAPTER_STARTED)
1224                 goto fail_not_started;
1225
1226         if (sa->txq_info[tx_queue_id].txq == NULL)
1227                 goto fail_not_setup;
1228
1229         rc = sfc_tx_qstart(sa, tx_queue_id);
1230         if (rc != 0)
1231                 goto fail_tx_qstart;
1232
1233         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1234
1235         sfc_adapter_unlock(sa);
1236         return 0;
1237
1238 fail_tx_qstart:
1239
1240 fail_not_setup:
1241 fail_not_started:
1242         sfc_adapter_unlock(sa);
1243         SFC_ASSERT(rc > 0);
1244         return -rc;
1245 }
1246
1247 static int
1248 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1249 {
1250         struct sfc_adapter *sa = dev->data->dev_private;
1251
1252         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1253
1254         sfc_adapter_lock(sa);
1255
1256         sfc_tx_qstop(sa, tx_queue_id);
1257
1258         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1259
1260         sfc_adapter_unlock(sa);
1261         return 0;
1262 }
1263
1264 static efx_tunnel_protocol_t
1265 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1266 {
1267         switch (rte_type) {
1268         case RTE_TUNNEL_TYPE_VXLAN:
1269                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1270         case RTE_TUNNEL_TYPE_GENEVE:
1271                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1272         default:
1273                 return EFX_TUNNEL_NPROTOS;
1274         }
1275 }
1276
1277 enum sfc_udp_tunnel_op_e {
1278         SFC_UDP_TUNNEL_ADD_PORT,
1279         SFC_UDP_TUNNEL_DEL_PORT,
1280 };
1281
1282 static int
1283 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1284                       struct rte_eth_udp_tunnel *tunnel_udp,
1285                       enum sfc_udp_tunnel_op_e op)
1286 {
1287         struct sfc_adapter *sa = dev->data->dev_private;
1288         efx_tunnel_protocol_t tunnel_proto;
1289         int rc;
1290
1291         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1292                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1293                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1294                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1295
1296         tunnel_proto =
1297                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1298         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1299                 rc = ENOTSUP;
1300                 goto fail_bad_proto;
1301         }
1302
1303         sfc_adapter_lock(sa);
1304
1305         switch (op) {
1306         case SFC_UDP_TUNNEL_ADD_PORT:
1307                 rc = efx_tunnel_config_udp_add(sa->nic,
1308                                                tunnel_udp->udp_port,
1309                                                tunnel_proto);
1310                 break;
1311         case SFC_UDP_TUNNEL_DEL_PORT:
1312                 rc = efx_tunnel_config_udp_remove(sa->nic,
1313                                                   tunnel_udp->udp_port,
1314                                                   tunnel_proto);
1315                 break;
1316         default:
1317                 rc = EINVAL;
1318                 goto fail_bad_op;
1319         }
1320
1321         if (rc != 0)
1322                 goto fail_op;
1323
1324         if (sa->state == SFC_ADAPTER_STARTED) {
1325                 rc = efx_tunnel_reconfigure(sa->nic);
1326                 if (rc == EAGAIN) {
1327                         /*
1328                          * Configuration is accepted by FW and MC reboot
1329                          * is initiated to apply the changes. MC reboot
1330                          * will be handled in a usual way (MC reboot
1331                          * event on management event queue and adapter
1332                          * restart).
1333                          */
1334                         rc = 0;
1335                 } else if (rc != 0) {
1336                         goto fail_reconfigure;
1337                 }
1338         }
1339
1340         sfc_adapter_unlock(sa);
1341         return 0;
1342
1343 fail_reconfigure:
1344         /* Remove/restore entry since the change makes the trouble */
1345         switch (op) {
1346         case SFC_UDP_TUNNEL_ADD_PORT:
1347                 (void)efx_tunnel_config_udp_remove(sa->nic,
1348                                                    tunnel_udp->udp_port,
1349                                                    tunnel_proto);
1350                 break;
1351         case SFC_UDP_TUNNEL_DEL_PORT:
1352                 (void)efx_tunnel_config_udp_add(sa->nic,
1353                                                 tunnel_udp->udp_port,
1354                                                 tunnel_proto);
1355                 break;
1356         }
1357
1358 fail_op:
1359 fail_bad_op:
1360         sfc_adapter_unlock(sa);
1361
1362 fail_bad_proto:
1363         SFC_ASSERT(rc > 0);
1364         return -rc;
1365 }
1366
1367 static int
1368 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1369                             struct rte_eth_udp_tunnel *tunnel_udp)
1370 {
1371         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1372 }
1373
1374 static int
1375 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1376                             struct rte_eth_udp_tunnel *tunnel_udp)
1377 {
1378         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1379 }
1380
1381 static int
1382 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1383                           struct rte_eth_rss_conf *rss_conf)
1384 {
1385         struct sfc_adapter *sa = dev->data->dev_private;
1386         struct sfc_rss *rss = &sa->rss;
1387
1388         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1389                 return -ENOTSUP;
1390
1391         sfc_adapter_lock(sa);
1392
1393         /*
1394          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1395          * hence, conversion is done here to derive a correct set of ETH_RSS
1396          * flags which corresponds to the active EFX configuration stored
1397          * locally in 'sfc_adapter' and kept up-to-date
1398          */
1399         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(sa, rss->hash_types);
1400         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1401         if (rss_conf->rss_key != NULL)
1402                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1403
1404         sfc_adapter_unlock(sa);
1405
1406         return 0;
1407 }
1408
1409 static int
1410 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1411                         struct rte_eth_rss_conf *rss_conf)
1412 {
1413         struct sfc_adapter *sa = dev->data->dev_private;
1414         struct sfc_rss *rss = &sa->rss;
1415         struct sfc_port *port = &sa->port;
1416         unsigned int efx_hash_types;
1417         int rc = 0;
1418
1419         if (port->isolated)
1420                 return -ENOTSUP;
1421
1422         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1423                 sfc_err(sa, "RSS is not available");
1424                 return -ENOTSUP;
1425         }
1426
1427         if (rss->channels == 0) {
1428                 sfc_err(sa, "RSS is not configured");
1429                 return -EINVAL;
1430         }
1431
1432         if ((rss_conf->rss_key != NULL) &&
1433             (rss_conf->rss_key_len != sizeof(rss->key))) {
1434                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1435                         sizeof(rss->key));
1436                 return -EINVAL;
1437         }
1438
1439         sfc_adapter_lock(sa);
1440
1441         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1442         if (rc != 0)
1443                 goto fail_rx_hf_rte_to_efx;
1444
1445         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1446                                    rss->hash_alg, efx_hash_types, B_TRUE);
1447         if (rc != 0)
1448                 goto fail_scale_mode_set;
1449
1450         if (rss_conf->rss_key != NULL) {
1451                 if (sa->state == SFC_ADAPTER_STARTED) {
1452                         rc = efx_rx_scale_key_set(sa->nic,
1453                                                   EFX_RSS_CONTEXT_DEFAULT,
1454                                                   rss_conf->rss_key,
1455                                                   sizeof(rss->key));
1456                         if (rc != 0)
1457                                 goto fail_scale_key_set;
1458                 }
1459
1460                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1461         }
1462
1463         rss->hash_types = efx_hash_types;
1464
1465         sfc_adapter_unlock(sa);
1466
1467         return 0;
1468
1469 fail_scale_key_set:
1470         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1471                                   EFX_RX_HASHALG_TOEPLITZ,
1472                                   rss->hash_types, B_TRUE) != 0)
1473                 sfc_err(sa, "failed to restore RSS mode");
1474
1475 fail_scale_mode_set:
1476 fail_rx_hf_rte_to_efx:
1477         sfc_adapter_unlock(sa);
1478         return -rc;
1479 }
1480
1481 static int
1482 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1483                        struct rte_eth_rss_reta_entry64 *reta_conf,
1484                        uint16_t reta_size)
1485 {
1486         struct sfc_adapter *sa = dev->data->dev_private;
1487         struct sfc_rss *rss = &sa->rss;
1488         struct sfc_port *port = &sa->port;
1489         int entry;
1490
1491         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || port->isolated)
1492                 return -ENOTSUP;
1493
1494         if (rss->channels == 0)
1495                 return -EINVAL;
1496
1497         if (reta_size != EFX_RSS_TBL_SIZE)
1498                 return -EINVAL;
1499
1500         sfc_adapter_lock(sa);
1501
1502         for (entry = 0; entry < reta_size; entry++) {
1503                 int grp = entry / RTE_RETA_GROUP_SIZE;
1504                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1505
1506                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1507                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1508         }
1509
1510         sfc_adapter_unlock(sa);
1511
1512         return 0;
1513 }
1514
1515 static int
1516 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1517                         struct rte_eth_rss_reta_entry64 *reta_conf,
1518                         uint16_t reta_size)
1519 {
1520         struct sfc_adapter *sa = dev->data->dev_private;
1521         struct sfc_rss *rss = &sa->rss;
1522         struct sfc_port *port = &sa->port;
1523         unsigned int *rss_tbl_new;
1524         uint16_t entry;
1525         int rc = 0;
1526
1527
1528         if (port->isolated)
1529                 return -ENOTSUP;
1530
1531         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1532                 sfc_err(sa, "RSS is not available");
1533                 return -ENOTSUP;
1534         }
1535
1536         if (rss->channels == 0) {
1537                 sfc_err(sa, "RSS is not configured");
1538                 return -EINVAL;
1539         }
1540
1541         if (reta_size != EFX_RSS_TBL_SIZE) {
1542                 sfc_err(sa, "RETA size is wrong (should be %u)",
1543                         EFX_RSS_TBL_SIZE);
1544                 return -EINVAL;
1545         }
1546
1547         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1548         if (rss_tbl_new == NULL)
1549                 return -ENOMEM;
1550
1551         sfc_adapter_lock(sa);
1552
1553         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1554
1555         for (entry = 0; entry < reta_size; entry++) {
1556                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1557                 struct rte_eth_rss_reta_entry64 *grp;
1558
1559                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1560
1561                 if (grp->mask & (1ull << grp_idx)) {
1562                         if (grp->reta[grp_idx] >= rss->channels) {
1563                                 rc = EINVAL;
1564                                 goto bad_reta_entry;
1565                         }
1566                         rss_tbl_new[entry] = grp->reta[grp_idx];
1567                 }
1568         }
1569
1570         if (sa->state == SFC_ADAPTER_STARTED) {
1571                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1572                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1573                 if (rc != 0)
1574                         goto fail_scale_tbl_set;
1575         }
1576
1577         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1578
1579 fail_scale_tbl_set:
1580 bad_reta_entry:
1581         sfc_adapter_unlock(sa);
1582
1583         rte_free(rss_tbl_new);
1584
1585         SFC_ASSERT(rc >= 0);
1586         return -rc;
1587 }
1588
1589 static int
1590 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1591                     enum rte_filter_op filter_op,
1592                     void *arg)
1593 {
1594         struct sfc_adapter *sa = dev->data->dev_private;
1595         int rc = ENOTSUP;
1596
1597         sfc_log_init(sa, "entry");
1598
1599         switch (filter_type) {
1600         case RTE_ETH_FILTER_NONE:
1601                 sfc_err(sa, "Global filters configuration not supported");
1602                 break;
1603         case RTE_ETH_FILTER_MACVLAN:
1604                 sfc_err(sa, "MACVLAN filters not supported");
1605                 break;
1606         case RTE_ETH_FILTER_ETHERTYPE:
1607                 sfc_err(sa, "EtherType filters not supported");
1608                 break;
1609         case RTE_ETH_FILTER_FLEXIBLE:
1610                 sfc_err(sa, "Flexible filters not supported");
1611                 break;
1612         case RTE_ETH_FILTER_SYN:
1613                 sfc_err(sa, "SYN filters not supported");
1614                 break;
1615         case RTE_ETH_FILTER_NTUPLE:
1616                 sfc_err(sa, "NTUPLE filters not supported");
1617                 break;
1618         case RTE_ETH_FILTER_TUNNEL:
1619                 sfc_err(sa, "Tunnel filters not supported");
1620                 break;
1621         case RTE_ETH_FILTER_FDIR:
1622                 sfc_err(sa, "Flow Director filters not supported");
1623                 break;
1624         case RTE_ETH_FILTER_HASH:
1625                 sfc_err(sa, "Hash filters not supported");
1626                 break;
1627         case RTE_ETH_FILTER_GENERIC:
1628                 if (filter_op != RTE_ETH_FILTER_GET) {
1629                         rc = EINVAL;
1630                 } else {
1631                         *(const void **)arg = &sfc_flow_ops;
1632                         rc = 0;
1633                 }
1634                 break;
1635         default:
1636                 sfc_err(sa, "Unknown filter type %u", filter_type);
1637                 break;
1638         }
1639
1640         sfc_log_init(sa, "exit: %d", -rc);
1641         SFC_ASSERT(rc >= 0);
1642         return -rc;
1643 }
1644
1645 static int
1646 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1647 {
1648         struct sfc_adapter *sa = dev->data->dev_private;
1649
1650         /*
1651          * If Rx datapath does not provide callback to check mempool,
1652          * all pools are supported.
1653          */
1654         if (sa->dp_rx->pool_ops_supported == NULL)
1655                 return 1;
1656
1657         return sa->dp_rx->pool_ops_supported(pool);
1658 }
1659
1660 static const struct eth_dev_ops sfc_eth_dev_ops = {
1661         .dev_configure                  = sfc_dev_configure,
1662         .dev_start                      = sfc_dev_start,
1663         .dev_stop                       = sfc_dev_stop,
1664         .dev_set_link_up                = sfc_dev_set_link_up,
1665         .dev_set_link_down              = sfc_dev_set_link_down,
1666         .dev_close                      = sfc_dev_close,
1667         .promiscuous_enable             = sfc_dev_promisc_enable,
1668         .promiscuous_disable            = sfc_dev_promisc_disable,
1669         .allmulticast_enable            = sfc_dev_allmulti_enable,
1670         .allmulticast_disable           = sfc_dev_allmulti_disable,
1671         .link_update                    = sfc_dev_link_update,
1672         .stats_get                      = sfc_stats_get,
1673         .stats_reset                    = sfc_stats_reset,
1674         .xstats_get                     = sfc_xstats_get,
1675         .xstats_reset                   = sfc_stats_reset,
1676         .xstats_get_names               = sfc_xstats_get_names,
1677         .dev_infos_get                  = sfc_dev_infos_get,
1678         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1679         .mtu_set                        = sfc_dev_set_mtu,
1680         .rx_queue_start                 = sfc_rx_queue_start,
1681         .rx_queue_stop                  = sfc_rx_queue_stop,
1682         .tx_queue_start                 = sfc_tx_queue_start,
1683         .tx_queue_stop                  = sfc_tx_queue_stop,
1684         .rx_queue_setup                 = sfc_rx_queue_setup,
1685         .rx_queue_release               = sfc_rx_queue_release,
1686         .rx_queue_count                 = sfc_rx_queue_count,
1687         .rx_descriptor_done             = sfc_rx_descriptor_done,
1688         .rx_descriptor_status           = sfc_rx_descriptor_status,
1689         .tx_descriptor_status           = sfc_tx_descriptor_status,
1690         .tx_queue_setup                 = sfc_tx_queue_setup,
1691         .tx_queue_release               = sfc_tx_queue_release,
1692         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1693         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1694         .mac_addr_set                   = sfc_mac_addr_set,
1695         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1696         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1697         .reta_update                    = sfc_dev_rss_reta_update,
1698         .reta_query                     = sfc_dev_rss_reta_query,
1699         .rss_hash_update                = sfc_dev_rss_hash_update,
1700         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1701         .filter_ctrl                    = sfc_dev_filter_ctrl,
1702         .set_mc_addr_list               = sfc_set_mc_addr_list,
1703         .rxq_info_get                   = sfc_rx_queue_info_get,
1704         .txq_info_get                   = sfc_tx_queue_info_get,
1705         .fw_version_get                 = sfc_fw_version_get,
1706         .xstats_get_by_id               = sfc_xstats_get_by_id,
1707         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1708         .pool_ops_supported             = sfc_pool_ops_supported,
1709 };
1710
1711 /**
1712  * Duplicate a string in potentially shared memory required for
1713  * multi-process support.
1714  *
1715  * strdup() allocates from process-local heap/memory.
1716  */
1717 static char *
1718 sfc_strdup(const char *str)
1719 {
1720         size_t size;
1721         char *copy;
1722
1723         if (str == NULL)
1724                 return NULL;
1725
1726         size = strlen(str) + 1;
1727         copy = rte_malloc(__func__, size, 0);
1728         if (copy != NULL)
1729                 rte_memcpy(copy, str, size);
1730
1731         return copy;
1732 }
1733
1734 static int
1735 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1736 {
1737         struct sfc_adapter *sa = dev->data->dev_private;
1738         const efx_nic_cfg_t *encp;
1739         unsigned int avail_caps = 0;
1740         const char *rx_name = NULL;
1741         const char *tx_name = NULL;
1742         int rc;
1743
1744         switch (sa->family) {
1745         case EFX_FAMILY_HUNTINGTON:
1746         case EFX_FAMILY_MEDFORD:
1747         case EFX_FAMILY_MEDFORD2:
1748                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1749                 break;
1750         default:
1751                 break;
1752         }
1753
1754         encp = efx_nic_cfg_get(sa->nic);
1755         if (encp->enc_rx_es_super_buffer_supported)
1756                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1757
1758         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1759                                 sfc_kvarg_string_handler, &rx_name);
1760         if (rc != 0)
1761                 goto fail_kvarg_rx_datapath;
1762
1763         if (rx_name != NULL) {
1764                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1765                 if (sa->dp_rx == NULL) {
1766                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1767                         rc = ENOENT;
1768                         goto fail_dp_rx;
1769                 }
1770                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1771                         sfc_err(sa,
1772                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1773                                 rx_name);
1774                         rc = EINVAL;
1775                         goto fail_dp_rx_caps;
1776                 }
1777         } else {
1778                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1779                 if (sa->dp_rx == NULL) {
1780                         sfc_err(sa, "Rx datapath by caps %#x not found",
1781                                 avail_caps);
1782                         rc = ENOENT;
1783                         goto fail_dp_rx;
1784                 }
1785         }
1786
1787         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1788         if (sa->dp_rx_name == NULL) {
1789                 rc = ENOMEM;
1790                 goto fail_dp_rx_name;
1791         }
1792
1793         sfc_notice(sa, "use %s Rx datapath", sa->dp_rx_name);
1794
1795         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1796
1797         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1798                                 sfc_kvarg_string_handler, &tx_name);
1799         if (rc != 0)
1800                 goto fail_kvarg_tx_datapath;
1801
1802         if (tx_name != NULL) {
1803                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1804                 if (sa->dp_tx == NULL) {
1805                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1806                         rc = ENOENT;
1807                         goto fail_dp_tx;
1808                 }
1809                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1810                         sfc_err(sa,
1811                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1812                                 tx_name);
1813                         rc = EINVAL;
1814                         goto fail_dp_tx_caps;
1815                 }
1816         } else {
1817                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1818                 if (sa->dp_tx == NULL) {
1819                         sfc_err(sa, "Tx datapath by caps %#x not found",
1820                                 avail_caps);
1821                         rc = ENOENT;
1822                         goto fail_dp_tx;
1823                 }
1824         }
1825
1826         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1827         if (sa->dp_tx_name == NULL) {
1828                 rc = ENOMEM;
1829                 goto fail_dp_tx_name;
1830         }
1831
1832         sfc_notice(sa, "use %s Tx datapath", sa->dp_tx_name);
1833
1834         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1835
1836         dev->dev_ops = &sfc_eth_dev_ops;
1837
1838         return 0;
1839
1840 fail_dp_tx_name:
1841 fail_dp_tx_caps:
1842         sa->dp_tx = NULL;
1843
1844 fail_dp_tx:
1845 fail_kvarg_tx_datapath:
1846         rte_free(sa->dp_rx_name);
1847         sa->dp_rx_name = NULL;
1848
1849 fail_dp_rx_name:
1850 fail_dp_rx_caps:
1851         sa->dp_rx = NULL;
1852
1853 fail_dp_rx:
1854 fail_kvarg_rx_datapath:
1855         return rc;
1856 }
1857
1858 static void
1859 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1860 {
1861         struct sfc_adapter *sa = dev->data->dev_private;
1862
1863         dev->dev_ops = NULL;
1864         dev->rx_pkt_burst = NULL;
1865         dev->tx_pkt_burst = NULL;
1866
1867         rte_free(sa->dp_tx_name);
1868         sa->dp_tx_name = NULL;
1869         sa->dp_tx = NULL;
1870
1871         rte_free(sa->dp_rx_name);
1872         sa->dp_rx_name = NULL;
1873         sa->dp_rx = NULL;
1874 }
1875
1876 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1877         .rxq_info_get                   = sfc_rx_queue_info_get,
1878         .txq_info_get                   = sfc_tx_queue_info_get,
1879 };
1880
1881 static int
1882 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev, uint32_t logtype_main)
1883 {
1884         /*
1885          * Device private data has really many process-local pointers.
1886          * Below code should be extremely careful to use data located
1887          * in shared memory only.
1888          */
1889         struct sfc_adapter *sa = dev->data->dev_private;
1890         const struct sfc_dp_rx *dp_rx;
1891         const struct sfc_dp_tx *dp_tx;
1892         int rc;
1893
1894         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1895         if (dp_rx == NULL) {
1896                 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1897                         "cannot find %s Rx datapath", sa->dp_rx_name);
1898                 rc = ENOENT;
1899                 goto fail_dp_rx;
1900         }
1901         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1902                 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1903                         "%s Rx datapath does not support multi-process",
1904                         sa->dp_rx_name);
1905                 rc = EINVAL;
1906                 goto fail_dp_rx_multi_process;
1907         }
1908
1909         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1910         if (dp_tx == NULL) {
1911                 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1912                         "cannot find %s Tx datapath", sa->dp_tx_name);
1913                 rc = ENOENT;
1914                 goto fail_dp_tx;
1915         }
1916         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1917                 SFC_LOG(sa, RTE_LOG_ERR, logtype_main,
1918                         "%s Tx datapath does not support multi-process",
1919                         sa->dp_tx_name);
1920                 rc = EINVAL;
1921                 goto fail_dp_tx_multi_process;
1922         }
1923
1924         dev->rx_pkt_burst = dp_rx->pkt_burst;
1925         dev->tx_pkt_burst = dp_tx->pkt_burst;
1926         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1927
1928         return 0;
1929
1930 fail_dp_tx_multi_process:
1931 fail_dp_tx:
1932 fail_dp_rx_multi_process:
1933 fail_dp_rx:
1934         return rc;
1935 }
1936
1937 static void
1938 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1939 {
1940         dev->dev_ops = NULL;
1941         dev->tx_pkt_burst = NULL;
1942         dev->rx_pkt_burst = NULL;
1943 }
1944
1945 static void
1946 sfc_register_dp(void)
1947 {
1948         /* Register once */
1949         if (TAILQ_EMPTY(&sfc_dp_head)) {
1950                 /* Prefer EF10 datapath */
1951                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1952                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1953                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1954
1955                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1956                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1957                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1958         }
1959 }
1960
1961 static int
1962 sfc_eth_dev_init(struct rte_eth_dev *dev)
1963 {
1964         struct sfc_adapter *sa = dev->data->dev_private;
1965         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1966         uint32_t logtype_main;
1967         int rc;
1968         const efx_nic_cfg_t *encp;
1969         const struct ether_addr *from;
1970
1971         sfc_register_dp();
1972
1973         logtype_main = sfc_register_logtype(&pci_dev->addr,
1974                                             SFC_LOGTYPE_MAIN_STR,
1975                                             RTE_LOG_NOTICE);
1976
1977         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1978                 return -sfc_eth_dev_secondary_set_ops(dev, logtype_main);
1979
1980         /* Required for logging */
1981         sa->pci_addr = pci_dev->addr;
1982         sa->port_id = dev->data->port_id;
1983         sa->logtype_main = logtype_main;
1984
1985         sa->eth_dev = dev;
1986
1987         /* Copy PCI device info to the dev->data */
1988         rte_eth_copy_pci_info(dev, pci_dev);
1989
1990         rc = sfc_kvargs_parse(sa);
1991         if (rc != 0)
1992                 goto fail_kvargs_parse;
1993
1994         sfc_log_init(sa, "entry");
1995
1996         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1997         if (dev->data->mac_addrs == NULL) {
1998                 rc = ENOMEM;
1999                 goto fail_mac_addrs;
2000         }
2001
2002         sfc_adapter_lock_init(sa);
2003         sfc_adapter_lock(sa);
2004
2005         sfc_log_init(sa, "probing");
2006         rc = sfc_probe(sa);
2007         if (rc != 0)
2008                 goto fail_probe;
2009
2010         sfc_log_init(sa, "set device ops");
2011         rc = sfc_eth_dev_set_ops(dev);
2012         if (rc != 0)
2013                 goto fail_set_ops;
2014
2015         sfc_log_init(sa, "attaching");
2016         rc = sfc_attach(sa);
2017         if (rc != 0)
2018                 goto fail_attach;
2019
2020         encp = efx_nic_cfg_get(sa->nic);
2021
2022         /*
2023          * The arguments are really reverse order in comparison to
2024          * Linux kernel. Copy from NIC config to Ethernet device data.
2025          */
2026         from = (const struct ether_addr *)(encp->enc_mac_addr);
2027         ether_addr_copy(from, &dev->data->mac_addrs[0]);
2028
2029         sfc_adapter_unlock(sa);
2030
2031         sfc_log_init(sa, "done");
2032         return 0;
2033
2034 fail_attach:
2035         sfc_eth_dev_clear_ops(dev);
2036
2037 fail_set_ops:
2038         sfc_unprobe(sa);
2039
2040 fail_probe:
2041         sfc_adapter_unlock(sa);
2042         sfc_adapter_lock_fini(sa);
2043         rte_free(dev->data->mac_addrs);
2044         dev->data->mac_addrs = NULL;
2045
2046 fail_mac_addrs:
2047         sfc_kvargs_cleanup(sa);
2048
2049 fail_kvargs_parse:
2050         sfc_log_init(sa, "failed %d", rc);
2051         SFC_ASSERT(rc > 0);
2052         return -rc;
2053 }
2054
2055 static int
2056 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2057 {
2058         struct sfc_adapter *sa;
2059
2060         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2061                 sfc_eth_dev_secondary_clear_ops(dev);
2062                 return 0;
2063         }
2064
2065         sa = dev->data->dev_private;
2066         sfc_log_init(sa, "entry");
2067
2068         sfc_adapter_lock(sa);
2069
2070         sfc_eth_dev_clear_ops(dev);
2071
2072         sfc_detach(sa);
2073         sfc_unprobe(sa);
2074
2075         sfc_kvargs_cleanup(sa);
2076
2077         sfc_adapter_unlock(sa);
2078         sfc_adapter_lock_fini(sa);
2079
2080         sfc_log_init(sa, "done");
2081
2082         /* Required for logging, so cleanup last */
2083         sa->eth_dev = NULL;
2084         return 0;
2085 }
2086
2087 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2088         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2089         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2090         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2091         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2092         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2093         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2094         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2095         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2096         { .vendor_id = 0 /* sentinel */ }
2097 };
2098
2099 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2100         struct rte_pci_device *pci_dev)
2101 {
2102         return rte_eth_dev_pci_generic_probe(pci_dev,
2103                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2104 }
2105
2106 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2107 {
2108         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2109 }
2110
2111 static struct rte_pci_driver sfc_efx_pmd = {
2112         .id_table = pci_id_sfc_efx_map,
2113         .drv_flags =
2114                 RTE_PCI_DRV_INTR_LSC |
2115                 RTE_PCI_DRV_NEED_MAPPING,
2116         .probe = sfc_eth_dev_pci_probe,
2117         .remove = sfc_eth_dev_pci_remove,
2118 };
2119
2120 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2121 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2122 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2123 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2124         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2125         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2126         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2127         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2128         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2129         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2130
2131 RTE_INIT(sfc_driver_register_logtype)
2132 {
2133         int ret;
2134
2135         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2136                                                    RTE_LOG_NOTICE);
2137         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2138 }