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.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16
  17 #include "efx.h"
  18
  19 #include "sfc.h"
  20 #include "sfc_debug.h"
  21 #include "sfc_log.h"
  22 #include "sfc_kvargs.h"
  23 #include "sfc_ev.h"
  24 #include "sfc_rx.h"
  25 #include "sfc_tx.h"
  26 #include "sfc_flow.h"
  27 #include "sfc_dp.h"
  28 #include "sfc_dp_rx.h"
  29
  30 static struct sfc_dp_list sfc_dp_head =
  31         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  32
  33 static int
  34 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  35 {
  36         struct sfc_adapter *sa = dev->data->dev_private;
  37         efx_nic_fw_info_t enfi;
  38         int ret;
  39         int rc;
  40
  41         /*
  42          * Return value of the callback is likely supposed to be
  43          * equal to or greater than 0, nevertheless, if an error
  44          * occurs, it will be desirable to pass it to the caller
  45          */
  46         if ((fw_version == NULL) || (fw_size == 0))
  47                 return -EINVAL;
  48
  49         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  50         if (rc != 0)
  51                 return -rc;
  52
  53         ret = snprintf(fw_version, fw_size,
  54                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  55                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  56                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  57         if (ret < 0)
  58                 return ret;
  59
  60         if (enfi.enfi_dpcpu_fw_ids_valid) {
  61                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  62                 int ret_extra;
  63
  64                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  65                                      fw_size - dpcpu_fw_ids_offset,
  66                                      " rx%" PRIx16 " tx%" PRIx16,
  67                                      enfi.enfi_rx_dpcpu_fw_id,
  68                                      enfi.enfi_tx_dpcpu_fw_id);
  69                 if (ret_extra < 0)
  70                         return ret_extra;
  71
  72                 ret += ret_extra;
  73         }
  74
  75         if (fw_size < (size_t)(++ret))
  76                 return ret;
  77         else
  78                 return 0;
  79 }
  80
  81 static void
  82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  83 {
  84         struct sfc_adapter *sa = dev->data->dev_private;
  85         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
  86
  87         sfc_log_init(sa, "entry");
  88
  89         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
  90         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  91
  92         /* Autonegotiation may be disabled */
  93         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  94         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
  95                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
  96         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
  97                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
  98         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
  99                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 100
 101         dev_info->max_rx_queues = sa->rxq_max;
 102         dev_info->max_tx_queues = sa->txq_max;
 103
 104         /* By default packets are dropped if no descriptors are available */
 105         dev_info->default_rxconf.rx_drop_en = 1;
 106
 107         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 108
 109         /*
 110          * rx_offload_capa includes both device and queue offloads since
 111          * the latter may be requested on a per device basis which makes
 112          * sense when some offloads are needed to be set on all queues.
 113          */
 114         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 115                                     dev_info->rx_queue_offload_capa;
 116
 117         dev_info->tx_offload_capa =
 118                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 119                 DEV_TX_OFFLOAD_UDP_CKSUM |
 120                 DEV_TX_OFFLOAD_TCP_CKSUM;
 121
 122         if (encp->enc_tunnel_encapsulations_supported != 0)
 123                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM;
 124
 125         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 126         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 127             !encp->enc_hw_tx_insert_vlan_enabled)
 128                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 129         else
 130                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 131
 132         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 133                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 134
 135         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 136                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 137
 138         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 139                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 140
 141 #if EFSYS_OPT_RX_SCALE
 142         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 143                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 144                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 145                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 146         }
 147 #endif
 148
 149         if (sa->tso)
 150                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 151
 152         /* Initialize to hardware limits */
 153         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 154         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 155         /* The RXQ hardware requires that the descriptor count is a power
 156          * of 2, but rx_desc_lim cannot properly describe that constraint.
 157          */
 158         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 159
 160         /* Initialize to hardware limits */
 161         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 162         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 163         /*
 164          * The TXQ hardware requires that the descriptor count is a power
 165          * of 2, but tx_desc_lim cannot properly describe that constraint
 166          */
 167         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 168
 169         if (sa->dp_rx->get_dev_info != NULL)
 170                 sa->dp_rx->get_dev_info(dev_info);
 171         if (sa->dp_tx->get_dev_info != NULL)
 172                 sa->dp_tx->get_dev_info(dev_info);
 173 }
 174
 175 static const uint32_t *
 176 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 177 {
 178         struct sfc_adapter *sa = dev->data->dev_private;
 179         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 180         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 181
 182         return sa->dp_rx->supported_ptypes_get(tunnel_encaps);
 183 }
 184
 185 static int
 186 sfc_dev_configure(struct rte_eth_dev *dev)
 187 {
 188         struct rte_eth_dev_data *dev_data = dev->data;
 189         struct sfc_adapter *sa = dev_data->dev_private;
 190         int rc;
 191
 192         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 193                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 194
 195         sfc_adapter_lock(sa);
 196         switch (sa->state) {
 197         case SFC_ADAPTER_CONFIGURED:
 198                 /* FALLTHROUGH */
 199         case SFC_ADAPTER_INITIALIZED:
 200                 rc = sfc_configure(sa);
 201                 break;
 202         default:
 203                 sfc_err(sa, "unexpected adapter state %u to configure",
 204                         sa->state);
 205                 rc = EINVAL;
 206                 break;
 207         }
 208         sfc_adapter_unlock(sa);
 209
 210         sfc_log_init(sa, "done %d", rc);
 211         SFC_ASSERT(rc >= 0);
 212         return -rc;
 213 }
 214
 215 static int
 216 sfc_dev_start(struct rte_eth_dev *dev)
 217 {
 218         struct sfc_adapter *sa = dev->data->dev_private;
 219         int rc;
 220
 221         sfc_log_init(sa, "entry");
 222
 223         sfc_adapter_lock(sa);
 224         rc = sfc_start(sa);
 225         sfc_adapter_unlock(sa);
 226
 227         sfc_log_init(sa, "done %d", rc);
 228         SFC_ASSERT(rc >= 0);
 229         return -rc;
 230 }
 231
 232 static int
 233 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 234 {
 235         struct sfc_adapter *sa = dev->data->dev_private;
 236         struct rte_eth_link *dev_link = &dev->data->dev_link;
 237         struct rte_eth_link old_link;
 238         struct rte_eth_link current_link;
 239
 240         sfc_log_init(sa, "entry");
 241
 242 retry:
 243         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 244         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 245
 246         if (sa->state != SFC_ADAPTER_STARTED) {
 247                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 248                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 249                                          *(uint64_t *)&old_link,
 250                                          *(uint64_t *)&current_link))
 251                         goto retry;
 252         } else if (wait_to_complete) {
 253                 efx_link_mode_t link_mode;
 254
 255                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 256                         link_mode = EFX_LINK_UNKNOWN;
 257                 sfc_port_link_mode_to_info(link_mode, &current_link);
 258
 259                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 260                                          *(uint64_t *)&old_link,
 261                                          *(uint64_t *)&current_link))
 262                         goto retry;
 263         } else {
 264                 sfc_ev_mgmt_qpoll(sa);
 265                 *(int64_t *)&current_link =
 266                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 267         }
 268
 269         if (old_link.link_status != current_link.link_status)
 270                 sfc_info(sa, "Link status is %s",
 271                          current_link.link_status ? "UP" : "DOWN");
 272
 273         return old_link.link_status == current_link.link_status ? 0 : -1;
 274 }
 275
 276 static void
 277 sfc_dev_stop(struct rte_eth_dev *dev)
 278 {
 279         struct sfc_adapter *sa = dev->data->dev_private;
 280
 281         sfc_log_init(sa, "entry");
 282
 283         sfc_adapter_lock(sa);
 284         sfc_stop(sa);
 285         sfc_adapter_unlock(sa);
 286
 287         sfc_log_init(sa, "done");
 288 }
 289
 290 static int
 291 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 292 {
 293         struct sfc_adapter *sa = dev->data->dev_private;
 294         int rc;
 295
 296         sfc_log_init(sa, "entry");
 297
 298         sfc_adapter_lock(sa);
 299         rc = sfc_start(sa);
 300         sfc_adapter_unlock(sa);
 301
 302         SFC_ASSERT(rc >= 0);
 303         return -rc;
 304 }
 305
 306 static int
 307 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 308 {
 309         struct sfc_adapter *sa = dev->data->dev_private;
 310
 311         sfc_log_init(sa, "entry");
 312
 313         sfc_adapter_lock(sa);
 314         sfc_stop(sa);
 315         sfc_adapter_unlock(sa);
 316
 317         return 0;
 318 }
 319
 320 static void
 321 sfc_dev_close(struct rte_eth_dev *dev)
 322 {
 323         struct sfc_adapter *sa = dev->data->dev_private;
 324
 325         sfc_log_init(sa, "entry");
 326
 327         sfc_adapter_lock(sa);
 328         switch (sa->state) {
 329         case SFC_ADAPTER_STARTED:
 330                 sfc_stop(sa);
 331                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 332                 /* FALLTHROUGH */
 333         case SFC_ADAPTER_CONFIGURED:
 334                 sfc_close(sa);
 335                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 336                 /* FALLTHROUGH */
 337         case SFC_ADAPTER_INITIALIZED:
 338                 break;
 339         default:
 340                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 341                 break;
 342         }
 343         sfc_adapter_unlock(sa);
 344
 345         sfc_log_init(sa, "done");
 346 }
 347
 348 static void
 349 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 350                    boolean_t enabled)
 351 {
 352         struct sfc_port *port;
 353         boolean_t *toggle;
 354         struct sfc_adapter *sa = dev->data->dev_private;
 355         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 356         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 357
 358         sfc_adapter_lock(sa);
 359
 360         port = &sa->port;
 361         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 362
 363         if (*toggle != enabled) {
 364                 *toggle = enabled;
 365
 366                 if (port->isolated) {
 367                         sfc_warn(sa, "isolated mode is active on the port");
 368                         sfc_warn(sa, "the change is to be applied on the next "
 369                                      "start provided that isolated mode is "
 370                                      "disabled prior the next start");
 371                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 372                            (sfc_set_rx_mode(sa) != 0)) {
 373                         *toggle = !(enabled);
 374                         sfc_warn(sa, "Failed to %s %s mode",
 375                                  ((enabled) ? "enable" : "disable"), desc);
 376                 }
 377         }
 378
 379         sfc_adapter_unlock(sa);
 380 }
 381
 382 static void
 383 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 384 {
 385         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 386 }
 387
 388 static void
 389 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 390 {
 391         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 392 }
 393
 394 static void
 395 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 396 {
 397         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 398 }
 399
 400 static void
 401 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 402 {
 403         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 404 }
 405
 406 static int
 407 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 408                    uint16_t nb_rx_desc, unsigned int socket_id,
 409                    const struct rte_eth_rxconf *rx_conf,
 410                    struct rte_mempool *mb_pool)
 411 {
 412         struct sfc_adapter *sa = dev->data->dev_private;
 413         int rc;
 414
 415         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 416                      rx_queue_id, nb_rx_desc, socket_id);
 417
 418         sfc_adapter_lock(sa);
 419
 420         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 421                           rx_conf, mb_pool);
 422         if (rc != 0)
 423                 goto fail_rx_qinit;
 424
 425         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 426
 427         sfc_adapter_unlock(sa);
 428
 429         return 0;
 430
 431 fail_rx_qinit:
 432         sfc_adapter_unlock(sa);
 433         SFC_ASSERT(rc > 0);
 434         return -rc;
 435 }
 436
 437 static void
 438 sfc_rx_queue_release(void *queue)
 439 {
 440         struct sfc_dp_rxq *dp_rxq = queue;
 441         struct sfc_rxq *rxq;
 442         struct sfc_adapter *sa;
 443         unsigned int sw_index;
 444
 445         if (dp_rxq == NULL)
 446                 return;
 447
 448         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 449         sa = rxq->evq->sa;
 450         sfc_adapter_lock(sa);
 451
 452         sw_index = sfc_rxq_sw_index(rxq);
 453
 454         sfc_log_init(sa, "RxQ=%u", sw_index);
 455
 456         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 457
 458         sfc_rx_qfini(sa, sw_index);
 459
 460         sfc_adapter_unlock(sa);
 461 }
 462
 463 static int
 464 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 465                    uint16_t nb_tx_desc, unsigned int socket_id,
 466                    const struct rte_eth_txconf *tx_conf)
 467 {
 468         struct sfc_adapter *sa = dev->data->dev_private;
 469         int rc;
 470
 471         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 472                      tx_queue_id, nb_tx_desc, socket_id);
 473
 474         sfc_adapter_lock(sa);
 475
 476         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 477         if (rc != 0)
 478                 goto fail_tx_qinit;
 479
 480         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 481
 482         sfc_adapter_unlock(sa);
 483         return 0;
 484
 485 fail_tx_qinit:
 486         sfc_adapter_unlock(sa);
 487         SFC_ASSERT(rc > 0);
 488         return -rc;
 489 }
 490
 491 static void
 492 sfc_tx_queue_release(void *queue)
 493 {
 494         struct sfc_dp_txq *dp_txq = queue;
 495         struct sfc_txq *txq;
 496         unsigned int sw_index;
 497         struct sfc_adapter *sa;
 498
 499         if (dp_txq == NULL)
 500                 return;
 501
 502         txq = sfc_txq_by_dp_txq(dp_txq);
 503         sw_index = sfc_txq_sw_index(txq);
 504
 505         SFC_ASSERT(txq->evq != NULL);
 506         sa = txq->evq->sa;
 507
 508         sfc_log_init(sa, "TxQ = %u", sw_index);
 509
 510         sfc_adapter_lock(sa);
 511
 512         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 513         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 514
 515         sfc_tx_qfini(sa, sw_index);
 516
 517         sfc_adapter_unlock(sa);
 518 }
 519
 520 static int
 521 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 522 {
 523         struct sfc_adapter *sa = dev->data->dev_private;
 524         struct sfc_port *port = &sa->port;
 525         uint64_t *mac_stats;
 526         int ret;
 527
 528         rte_spinlock_lock(&port->mac_stats_lock);
 529
 530         ret = sfc_port_update_mac_stats(sa);
 531         if (ret != 0)
 532                 goto unlock;
 533
 534         mac_stats = port->mac_stats_buf;
 535
 536         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 537                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 538                 stats->ipackets =
 539                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 540                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 541                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 542                 stats->opackets =
 543                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 544                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 545                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 546                 stats->ibytes =
 547                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 548                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 549                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 550                 stats->obytes =
 551                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 552                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 553                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 554                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 555                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 556                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 557         } else {
 558                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 559                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 560                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 561                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 562                 /*
 563                  * Take into account stats which are whenever supported
 564                  * on EF10. If some stat is not supported by current
 565                  * firmware variant or HW revision, it is guaranteed
 566                  * to be zero in mac_stats.
 567                  */
 568                 stats->imissed =
 569                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 570                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 571                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 572                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 573                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 574                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 575                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 576                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 577                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 578                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 579                 stats->ierrors =
 580                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 581                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 582                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 583                 /* no oerrors counters supported on EF10 */
 584         }
 585
 586 unlock:
 587         rte_spinlock_unlock(&port->mac_stats_lock);
 588         SFC_ASSERT(ret >= 0);
 589         return -ret;
 590 }
 591
 592 static void
 593 sfc_stats_reset(struct rte_eth_dev *dev)
 594 {
 595         struct sfc_adapter *sa = dev->data->dev_private;
 596         struct sfc_port *port = &sa->port;
 597         int rc;
 598
 599         if (sa->state != SFC_ADAPTER_STARTED) {
 600                 /*
 601                  * The operation cannot be done if port is not started; it
 602                  * will be scheduled to be done during the next port start
 603                  */
 604                 port->mac_stats_reset_pending = B_TRUE;
 605                 return;
 606         }
 607
 608         rc = sfc_port_reset_mac_stats(sa);
 609         if (rc != 0)
 610                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 611 }
 612
 613 static int
 614 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 615                unsigned int xstats_count)
 616 {
 617         struct sfc_adapter *sa = dev->data->dev_private;
 618         struct sfc_port *port = &sa->port;
 619         uint64_t *mac_stats;
 620         int rc;
 621         unsigned int i;
 622         int nstats = 0;
 623
 624         rte_spinlock_lock(&port->mac_stats_lock);
 625
 626         rc = sfc_port_update_mac_stats(sa);
 627         if (rc != 0) {
 628                 SFC_ASSERT(rc > 0);
 629                 nstats = -rc;
 630                 goto unlock;
 631         }
 632
 633         mac_stats = port->mac_stats_buf;
 634
 635         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 636                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 637                         if (xstats != NULL && nstats < (int)xstats_count) {
 638                                 xstats[nstats].id = nstats;
 639                                 xstats[nstats].value = mac_stats[i];
 640                         }
 641                         nstats++;
 642                 }
 643         }
 644
 645 unlock:
 646         rte_spinlock_unlock(&port->mac_stats_lock);
 647
 648         return nstats;
 649 }
 650
 651 static int
 652 sfc_xstats_get_names(struct rte_eth_dev *dev,
 653                      struct rte_eth_xstat_name *xstats_names,
 654                      unsigned int xstats_count)
 655 {
 656         struct sfc_adapter *sa = dev->data->dev_private;
 657         struct sfc_port *port = &sa->port;
 658         unsigned int i;
 659         unsigned int nstats = 0;
 660
 661         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 662                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 663                         if (xstats_names != NULL && nstats < xstats_count)
 664                                 strncpy(xstats_names[nstats].name,
 665                                         efx_mac_stat_name(sa->nic, i),
 666                                         sizeof(xstats_names[0].name));
 667                         nstats++;
 668                 }
 669         }
 670
 671         return nstats;
 672 }
 673
 674 static int
 675 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 676                      uint64_t *values, unsigned int n)
 677 {
 678         struct sfc_adapter *sa = dev->data->dev_private;
 679         struct sfc_port *port = &sa->port;
 680         uint64_t *mac_stats;
 681         unsigned int nb_supported = 0;
 682         unsigned int nb_written = 0;
 683         unsigned int i;
 684         int ret;
 685         int rc;
 686
 687         if (unlikely(values == NULL) ||
 688             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 689                 return port->mac_stats_nb_supported;
 690
 691         rte_spinlock_lock(&port->mac_stats_lock);
 692
 693         rc = sfc_port_update_mac_stats(sa);
 694         if (rc != 0) {
 695                 SFC_ASSERT(rc > 0);
 696                 ret = -rc;
 697                 goto unlock;
 698         }
 699
 700         mac_stats = port->mac_stats_buf;
 701
 702         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 703                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 704                         continue;
 705
 706                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 707                         values[nb_written++] = mac_stats[i];
 708
 709                 ++nb_supported;
 710         }
 711
 712         ret = nb_written;
 713
 714 unlock:
 715         rte_spinlock_unlock(&port->mac_stats_lock);
 716
 717         return ret;
 718 }
 719
 720 static int
 721 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 722                            struct rte_eth_xstat_name *xstats_names,
 723                            const uint64_t *ids, unsigned int size)
 724 {
 725         struct sfc_adapter *sa = dev->data->dev_private;
 726         struct sfc_port *port = &sa->port;
 727         unsigned int nb_supported = 0;
 728         unsigned int nb_written = 0;
 729         unsigned int i;
 730
 731         if (unlikely(xstats_names == NULL) ||
 732             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 733                 return port->mac_stats_nb_supported;
 734
 735         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 736                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 737                         continue;
 738
 739                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 740                         char *name = xstats_names[nb_written++].name;
 741
 742                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 743                                 sizeof(xstats_names[0].name));
 744                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 745                 }
 746
 747                 ++nb_supported;
 748         }
 749
 750         return nb_written;
 751 }
 752
 753 static int
 754 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 755 {
 756         struct sfc_adapter *sa = dev->data->dev_private;
 757         unsigned int wanted_fc, link_fc;
 758
 759         memset(fc_conf, 0, sizeof(*fc_conf));
 760
 761         sfc_adapter_lock(sa);
 762
 763         if (sa->state == SFC_ADAPTER_STARTED)
 764                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 765         else
 766                 link_fc = sa->port.flow_ctrl;
 767
 768         switch (link_fc) {
 769         case 0:
 770                 fc_conf->mode = RTE_FC_NONE;
 771                 break;
 772         case EFX_FCNTL_RESPOND:
 773                 fc_conf->mode = RTE_FC_RX_PAUSE;
 774                 break;
 775         case EFX_FCNTL_GENERATE:
 776                 fc_conf->mode = RTE_FC_TX_PAUSE;
 777                 break;
 778         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 779                 fc_conf->mode = RTE_FC_FULL;
 780                 break;
 781         default:
 782                 sfc_err(sa, "%s: unexpected flow control value %#x",
 783                         __func__, link_fc);
 784         }
 785
 786         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 787
 788         sfc_adapter_unlock(sa);
 789
 790         return 0;
 791 }
 792
 793 static int
 794 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 795 {
 796         struct sfc_adapter *sa = dev->data->dev_private;
 797         struct sfc_port *port = &sa->port;
 798         unsigned int fcntl;
 799         int rc;
 800
 801         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 802             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 803             fc_conf->mac_ctrl_frame_fwd != 0) {
 804                 sfc_err(sa, "unsupported flow control settings specified");
 805                 rc = EINVAL;
 806                 goto fail_inval;
 807         }
 808
 809         switch (fc_conf->mode) {
 810         case RTE_FC_NONE:
 811                 fcntl = 0;
 812                 break;
 813         case RTE_FC_RX_PAUSE:
 814                 fcntl = EFX_FCNTL_RESPOND;
 815                 break;
 816         case RTE_FC_TX_PAUSE:
 817                 fcntl = EFX_FCNTL_GENERATE;
 818                 break;
 819         case RTE_FC_FULL:
 820                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 821                 break;
 822         default:
 823                 rc = EINVAL;
 824                 goto fail_inval;
 825         }
 826
 827         sfc_adapter_lock(sa);
 828
 829         if (sa->state == SFC_ADAPTER_STARTED) {
 830                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 831                 if (rc != 0)
 832                         goto fail_mac_fcntl_set;
 833         }
 834
 835         port->flow_ctrl = fcntl;
 836         port->flow_ctrl_autoneg = fc_conf->autoneg;
 837
 838         sfc_adapter_unlock(sa);
 839
 840         return 0;
 841
 842 fail_mac_fcntl_set:
 843         sfc_adapter_unlock(sa);
 844 fail_inval:
 845         SFC_ASSERT(rc > 0);
 846         return -rc;
 847 }
 848
 849 static int
 850 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 851 {
 852         struct sfc_adapter *sa = dev->data->dev_private;
 853         size_t pdu = EFX_MAC_PDU(mtu);
 854         size_t old_pdu;
 855         int rc;
 856
 857         sfc_log_init(sa, "mtu=%u", mtu);
 858
 859         rc = EINVAL;
 860         if (pdu < EFX_MAC_PDU_MIN) {
 861                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 862                         (unsigned int)mtu, (unsigned int)pdu,
 863                         EFX_MAC_PDU_MIN);
 864                 goto fail_inval;
 865         }
 866         if (pdu > EFX_MAC_PDU_MAX) {
 867                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 868                         (unsigned int)mtu, (unsigned int)pdu,
 869                         EFX_MAC_PDU_MAX);
 870                 goto fail_inval;
 871         }
 872
 873         sfc_adapter_lock(sa);
 874
 875         if (pdu != sa->port.pdu) {
 876                 if (sa->state == SFC_ADAPTER_STARTED) {
 877                         sfc_stop(sa);
 878
 879                         old_pdu = sa->port.pdu;
 880                         sa->port.pdu = pdu;
 881                         rc = sfc_start(sa);
 882                         if (rc != 0)
 883                                 goto fail_start;
 884                 } else {
 885                         sa->port.pdu = pdu;
 886                 }
 887         }
 888
 889         /*
 890          * The driver does not use it, but other PMDs update jumbo_frame
 891          * flag and max_rx_pkt_len when MTU is set.
 892          */
 893         if (mtu > ETHER_MAX_LEN) {
 894                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 895
 896                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 897                 rxmode->jumbo_frame = 1;
 898         }
 899
 900         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 901
 902         sfc_adapter_unlock(sa);
 903
 904         sfc_log_init(sa, "done");
 905         return 0;
 906
 907 fail_start:
 908         sa->port.pdu = old_pdu;
 909         if (sfc_start(sa) != 0)
 910                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 911                         "PDU max size - port is stopped",
 912                         (unsigned int)pdu, (unsigned int)old_pdu);
 913         sfc_adapter_unlock(sa);
 914
 915 fail_inval:
 916         sfc_log_init(sa, "failed %d", rc);
 917         SFC_ASSERT(rc > 0);
 918         return -rc;
 919 }
 920 static void
 921 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 922 {
 923         struct sfc_adapter *sa = dev->data->dev_private;
 924         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 925         struct sfc_port *port = &sa->port;
 926         int rc;
 927
 928         sfc_adapter_lock(sa);
 929
 930         /*
 931          * Copy the address to the device private data so that
 932          * it could be recalled in the case of adapter restart.
 933          */
 934         ether_addr_copy(mac_addr, &port->default_mac_addr);
 935
 936         if (port->isolated) {
 937                 sfc_err(sa, "isolated mode is active on the port");
 938                 sfc_err(sa, "will not set MAC address");
 939                 goto unlock;
 940         }
 941
 942         if (sa->state != SFC_ADAPTER_STARTED) {
 943                 sfc_info(sa, "the port is not started");
 944                 sfc_info(sa, "the new MAC address will be set on port start");
 945
 946                 goto unlock;
 947         }
 948
 949         if (encp->enc_allow_set_mac_with_installed_filters) {
 950                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 951                 if (rc != 0) {
 952                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 953                         goto unlock;
 954                 }
 955
 956                 /*
 957                  * Changing the MAC address by means of MCDI request
 958                  * has no effect on received traffic, therefore
 959                  * we also need to update unicast filters
 960                  */
 961                 rc = sfc_set_rx_mode(sa);
 962                 if (rc != 0)
 963                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 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         /*
 983          * In the case of failure sa->port->default_mac_addr does not
 984          * need rollback since no error code is returned, and the upper
 985          * API will anyway update the external MAC address storage.
 986          * To be consistent with that new value it is better to keep
 987          * the device private value the same.
 988          */
 989         sfc_adapter_unlock(sa);
 990 }
 991
 992
 993 static int
 994 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 995                      uint32_t nb_mc_addr)
 996 {
 997         struct sfc_adapter *sa = dev->data->dev_private;
 998         struct sfc_port *port = &sa->port;
 999         uint8_t *mc_addrs = port->mcast_addrs;
1000         int rc;
1001         unsigned int i;
1002
1003         if (port->isolated) {
1004                 sfc_err(sa, "isolated mode is active on the port");
1005                 sfc_err(sa, "will not set multicast address list");
1006                 return -ENOTSUP;
1007         }
1008
1009         if (mc_addrs == NULL)
1010                 return -ENOBUFS;
1011
1012         if (nb_mc_addr > port->max_mcast_addrs) {
1013                 sfc_err(sa, "too many multicast addresses: %u > %u",
1014                          nb_mc_addr, port->max_mcast_addrs);
1015                 return -EINVAL;
1016         }
1017
1018         for (i = 0; i < nb_mc_addr; ++i) {
1019                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1020                                  EFX_MAC_ADDR_LEN);
1021                 mc_addrs += EFX_MAC_ADDR_LEN;
1022         }
1023
1024         port->nb_mcast_addrs = nb_mc_addr;
1025
1026         if (sa->state != SFC_ADAPTER_STARTED)
1027                 return 0;
1028
1029         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1030                                         port->nb_mcast_addrs);
1031         if (rc != 0)
1032                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1033
1034         SFC_ASSERT(rc > 0);
1035         return -rc;
1036 }
1037
1038 /*
1039  * The function is used by the secondary process as well. It must not
1040  * use any process-local pointers from the adapter data.
1041  */
1042 static void
1043 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1044                       struct rte_eth_rxq_info *qinfo)
1045 {
1046         struct sfc_adapter *sa = dev->data->dev_private;
1047         struct sfc_rxq_info *rxq_info;
1048         struct sfc_rxq *rxq;
1049
1050         sfc_adapter_lock(sa);
1051
1052         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1053
1054         rxq_info = &sa->rxq_info[rx_queue_id];
1055         rxq = rxq_info->rxq;
1056         SFC_ASSERT(rxq != NULL);
1057
1058         qinfo->mp = rxq->refill_mb_pool;
1059         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1060         qinfo->conf.rx_drop_en = 1;
1061         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1062         qinfo->conf.offloads = DEV_RX_OFFLOAD_IPV4_CKSUM |
1063                                DEV_RX_OFFLOAD_UDP_CKSUM |
1064                                DEV_RX_OFFLOAD_TCP_CKSUM;
1065         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1066                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1067                 qinfo->scattered_rx = 1;
1068         }
1069         qinfo->nb_desc = rxq_info->entries;
1070
1071         sfc_adapter_unlock(sa);
1072 }
1073
1074 /*
1075  * The function is used by the secondary process as well. It must not
1076  * use any process-local pointers from the adapter data.
1077  */
1078 static void
1079 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1080                       struct rte_eth_txq_info *qinfo)
1081 {
1082         struct sfc_adapter *sa = dev->data->dev_private;
1083         struct sfc_txq_info *txq_info;
1084
1085         sfc_adapter_lock(sa);
1086
1087         SFC_ASSERT(tx_queue_id < sa->txq_count);
1088
1089         txq_info = &sa->txq_info[tx_queue_id];
1090         SFC_ASSERT(txq_info->txq != NULL);
1091
1092         memset(qinfo, 0, sizeof(*qinfo));
1093
1094         qinfo->conf.txq_flags = txq_info->txq->flags;
1095         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1096         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1097         qinfo->nb_desc = txq_info->entries;
1098
1099         sfc_adapter_unlock(sa);
1100 }
1101
1102 static uint32_t
1103 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1104 {
1105         struct sfc_adapter *sa = dev->data->dev_private;
1106
1107         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1108
1109         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1110 }
1111
1112 static int
1113 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1114 {
1115         struct sfc_dp_rxq *dp_rxq = queue;
1116
1117         return sfc_rx_qdesc_done(dp_rxq, offset);
1118 }
1119
1120 static int
1121 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1122 {
1123         struct sfc_dp_rxq *dp_rxq = queue;
1124         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1125
1126         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1127 }
1128
1129 static int
1130 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1131 {
1132         struct sfc_dp_txq *dp_txq = queue;
1133         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1134
1135         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1136 }
1137
1138 static int
1139 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1140 {
1141         struct sfc_adapter *sa = dev->data->dev_private;
1142         int rc;
1143
1144         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1145
1146         sfc_adapter_lock(sa);
1147
1148         rc = EINVAL;
1149         if (sa->state != SFC_ADAPTER_STARTED)
1150                 goto fail_not_started;
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_started:
1164         sfc_adapter_unlock(sa);
1165         SFC_ASSERT(rc > 0);
1166         return -rc;
1167 }
1168
1169 static int
1170 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1171 {
1172         struct sfc_adapter *sa = dev->data->dev_private;
1173
1174         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1175
1176         sfc_adapter_lock(sa);
1177         sfc_rx_qstop(sa, rx_queue_id);
1178
1179         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1180
1181         sfc_adapter_unlock(sa);
1182
1183         return 0;
1184 }
1185
1186 static int
1187 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1188 {
1189         struct sfc_adapter *sa = dev->data->dev_private;
1190         int rc;
1191
1192         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1193
1194         sfc_adapter_lock(sa);
1195
1196         rc = EINVAL;
1197         if (sa->state != SFC_ADAPTER_STARTED)
1198                 goto fail_not_started;
1199
1200         rc = sfc_tx_qstart(sa, tx_queue_id);
1201         if (rc != 0)
1202                 goto fail_tx_qstart;
1203
1204         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1205
1206         sfc_adapter_unlock(sa);
1207         return 0;
1208
1209 fail_tx_qstart:
1210
1211 fail_not_started:
1212         sfc_adapter_unlock(sa);
1213         SFC_ASSERT(rc > 0);
1214         return -rc;
1215 }
1216
1217 static int
1218 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1219 {
1220         struct sfc_adapter *sa = dev->data->dev_private;
1221
1222         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1223
1224         sfc_adapter_lock(sa);
1225
1226         sfc_tx_qstop(sa, tx_queue_id);
1227
1228         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1229
1230         sfc_adapter_unlock(sa);
1231         return 0;
1232 }
1233
1234 static efx_tunnel_protocol_t
1235 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1236 {
1237         switch (rte_type) {
1238         case RTE_TUNNEL_TYPE_VXLAN:
1239                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1240         case RTE_TUNNEL_TYPE_GENEVE:
1241                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1242         default:
1243                 return EFX_TUNNEL_NPROTOS;
1244         }
1245 }
1246
1247 enum sfc_udp_tunnel_op_e {
1248         SFC_UDP_TUNNEL_ADD_PORT,
1249         SFC_UDP_TUNNEL_DEL_PORT,
1250 };
1251
1252 static int
1253 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1254                       struct rte_eth_udp_tunnel *tunnel_udp,
1255                       enum sfc_udp_tunnel_op_e op)
1256 {
1257         struct sfc_adapter *sa = dev->data->dev_private;
1258         efx_tunnel_protocol_t tunnel_proto;
1259         int rc;
1260
1261         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1262                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1263                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1264                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1265
1266         tunnel_proto =
1267                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1268         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1269                 rc = ENOTSUP;
1270                 goto fail_bad_proto;
1271         }
1272
1273         sfc_adapter_lock(sa);
1274
1275         switch (op) {
1276         case SFC_UDP_TUNNEL_ADD_PORT:
1277                 rc = efx_tunnel_config_udp_add(sa->nic,
1278                                                tunnel_udp->udp_port,
1279                                                tunnel_proto);
1280                 break;
1281         case SFC_UDP_TUNNEL_DEL_PORT:
1282                 rc = efx_tunnel_config_udp_remove(sa->nic,
1283                                                   tunnel_udp->udp_port,
1284                                                   tunnel_proto);
1285                 break;
1286         default:
1287                 rc = EINVAL;
1288                 goto fail_bad_op;
1289         }
1290
1291         if (rc != 0)
1292                 goto fail_op;
1293
1294         if (sa->state == SFC_ADAPTER_STARTED) {
1295                 rc = efx_tunnel_reconfigure(sa->nic);
1296                 if (rc == EAGAIN) {
1297                         /*
1298                          * Configuration is accepted by FW and MC reboot
1299                          * is initiated to apply the changes. MC reboot
1300                          * will be handled in a usual way (MC reboot
1301                          * event on management event queue and adapter
1302                          * restart).
1303                          */
1304                         rc = 0;
1305                 } else if (rc != 0) {
1306                         goto fail_reconfigure;
1307                 }
1308         }
1309
1310         sfc_adapter_unlock(sa);
1311         return 0;
1312
1313 fail_reconfigure:
1314         /* Remove/restore entry since the change makes the trouble */
1315         switch (op) {
1316         case SFC_UDP_TUNNEL_ADD_PORT:
1317                 (void)efx_tunnel_config_udp_remove(sa->nic,
1318                                                    tunnel_udp->udp_port,
1319                                                    tunnel_proto);
1320                 break;
1321         case SFC_UDP_TUNNEL_DEL_PORT:
1322                 (void)efx_tunnel_config_udp_add(sa->nic,
1323                                                 tunnel_udp->udp_port,
1324                                                 tunnel_proto);
1325                 break;
1326         }
1327
1328 fail_op:
1329 fail_bad_op:
1330         sfc_adapter_unlock(sa);
1331
1332 fail_bad_proto:
1333         SFC_ASSERT(rc > 0);
1334         return -rc;
1335 }
1336
1337 static int
1338 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1339                             struct rte_eth_udp_tunnel *tunnel_udp)
1340 {
1341         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1342 }
1343
1344 static int
1345 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1346                             struct rte_eth_udp_tunnel *tunnel_udp)
1347 {
1348         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1349 }
1350
1351 #if EFSYS_OPT_RX_SCALE
1352 static int
1353 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1354                           struct rte_eth_rss_conf *rss_conf)
1355 {
1356         struct sfc_adapter *sa = dev->data->dev_private;
1357         struct sfc_port *port = &sa->port;
1358
1359         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1360                 return -ENOTSUP;
1361
1362         if (sa->rss_channels == 0)
1363                 return -EINVAL;
1364
1365         sfc_adapter_lock(sa);
1366
1367         /*
1368          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1369          * hence, conversion is done here to derive a correct set of ETH_RSS
1370          * flags which corresponds to the active EFX configuration stored
1371          * locally in 'sfc_adapter' and kept up-to-date
1372          */
1373         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1374         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1375         if (rss_conf->rss_key != NULL)
1376                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1377
1378         sfc_adapter_unlock(sa);
1379
1380         return 0;
1381 }
1382
1383 static int
1384 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1385                         struct rte_eth_rss_conf *rss_conf)
1386 {
1387         struct sfc_adapter *sa = dev->data->dev_private;
1388         struct sfc_port *port = &sa->port;
1389         unsigned int efx_hash_types;
1390         int rc = 0;
1391
1392         if (port->isolated)
1393                 return -ENOTSUP;
1394
1395         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1396                 sfc_err(sa, "RSS is not available");
1397                 return -ENOTSUP;
1398         }
1399
1400         if (sa->rss_channels == 0) {
1401                 sfc_err(sa, "RSS is not configured");
1402                 return -EINVAL;
1403         }
1404
1405         if ((rss_conf->rss_key != NULL) &&
1406             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1407                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1408                         sizeof(sa->rss_key));
1409                 return -EINVAL;
1410         }
1411
1412         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1413                 sfc_err(sa, "unsupported hash functions requested");
1414                 return -EINVAL;
1415         }
1416
1417         sfc_adapter_lock(sa);
1418
1419         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1420
1421         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1422                                    EFX_RX_HASHALG_TOEPLITZ,
1423                                    efx_hash_types, B_TRUE);
1424         if (rc != 0)
1425                 goto fail_scale_mode_set;
1426
1427         if (rss_conf->rss_key != NULL) {
1428                 if (sa->state == SFC_ADAPTER_STARTED) {
1429                         rc = efx_rx_scale_key_set(sa->nic,
1430                                                   EFX_RSS_CONTEXT_DEFAULT,
1431                                                   rss_conf->rss_key,
1432                                                   sizeof(sa->rss_key));
1433                         if (rc != 0)
1434                                 goto fail_scale_key_set;
1435                 }
1436
1437                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1438         }
1439
1440         sa->rss_hash_types = efx_hash_types;
1441
1442         sfc_adapter_unlock(sa);
1443
1444         return 0;
1445
1446 fail_scale_key_set:
1447         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1448                                   EFX_RX_HASHALG_TOEPLITZ,
1449                                   sa->rss_hash_types, B_TRUE) != 0)
1450                 sfc_err(sa, "failed to restore RSS mode");
1451
1452 fail_scale_mode_set:
1453         sfc_adapter_unlock(sa);
1454         return -rc;
1455 }
1456
1457 static int
1458 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1459                        struct rte_eth_rss_reta_entry64 *reta_conf,
1460                        uint16_t reta_size)
1461 {
1462         struct sfc_adapter *sa = dev->data->dev_private;
1463         struct sfc_port *port = &sa->port;
1464         int entry;
1465
1466         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1467                 return -ENOTSUP;
1468
1469         if (sa->rss_channels == 0)
1470                 return -EINVAL;
1471
1472         if (reta_size != EFX_RSS_TBL_SIZE)
1473                 return -EINVAL;
1474
1475         sfc_adapter_lock(sa);
1476
1477         for (entry = 0; entry < reta_size; entry++) {
1478                 int grp = entry / RTE_RETA_GROUP_SIZE;
1479                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1480
1481                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1482                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1483         }
1484
1485         sfc_adapter_unlock(sa);
1486
1487         return 0;
1488 }
1489
1490 static int
1491 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1492                         struct rte_eth_rss_reta_entry64 *reta_conf,
1493                         uint16_t reta_size)
1494 {
1495         struct sfc_adapter *sa = dev->data->dev_private;
1496         struct sfc_port *port = &sa->port;
1497         unsigned int *rss_tbl_new;
1498         uint16_t entry;
1499         int rc = 0;
1500
1501
1502         if (port->isolated)
1503                 return -ENOTSUP;
1504
1505         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1506                 sfc_err(sa, "RSS is not available");
1507                 return -ENOTSUP;
1508         }
1509
1510         if (sa->rss_channels == 0) {
1511                 sfc_err(sa, "RSS is not configured");
1512                 return -EINVAL;
1513         }
1514
1515         if (reta_size != EFX_RSS_TBL_SIZE) {
1516                 sfc_err(sa, "RETA size is wrong (should be %u)",
1517                         EFX_RSS_TBL_SIZE);
1518                 return -EINVAL;
1519         }
1520
1521         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1522         if (rss_tbl_new == NULL)
1523                 return -ENOMEM;
1524
1525         sfc_adapter_lock(sa);
1526
1527         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1528
1529         for (entry = 0; entry < reta_size; entry++) {
1530                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1531                 struct rte_eth_rss_reta_entry64 *grp;
1532
1533                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1534
1535                 if (grp->mask & (1ull << grp_idx)) {
1536                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1537                                 rc = EINVAL;
1538                                 goto bad_reta_entry;
1539                         }
1540                         rss_tbl_new[entry] = grp->reta[grp_idx];
1541                 }
1542         }
1543
1544         if (sa->state == SFC_ADAPTER_STARTED) {
1545                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1546                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1547                 if (rc != 0)
1548                         goto fail_scale_tbl_set;
1549         }
1550
1551         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1552
1553 fail_scale_tbl_set:
1554 bad_reta_entry:
1555         sfc_adapter_unlock(sa);
1556
1557         rte_free(rss_tbl_new);
1558
1559         SFC_ASSERT(rc >= 0);
1560         return -rc;
1561 }
1562 #endif
1563
1564 static int
1565 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1566                     enum rte_filter_op filter_op,
1567                     void *arg)
1568 {
1569         struct sfc_adapter *sa = dev->data->dev_private;
1570         int rc = ENOTSUP;
1571
1572         sfc_log_init(sa, "entry");
1573
1574         switch (filter_type) {
1575         case RTE_ETH_FILTER_NONE:
1576                 sfc_err(sa, "Global filters configuration not supported");
1577                 break;
1578         case RTE_ETH_FILTER_MACVLAN:
1579                 sfc_err(sa, "MACVLAN filters not supported");
1580                 break;
1581         case RTE_ETH_FILTER_ETHERTYPE:
1582                 sfc_err(sa, "EtherType filters not supported");
1583                 break;
1584         case RTE_ETH_FILTER_FLEXIBLE:
1585                 sfc_err(sa, "Flexible filters not supported");
1586                 break;
1587         case RTE_ETH_FILTER_SYN:
1588                 sfc_err(sa, "SYN filters not supported");
1589                 break;
1590         case RTE_ETH_FILTER_NTUPLE:
1591                 sfc_err(sa, "NTUPLE filters not supported");
1592                 break;
1593         case RTE_ETH_FILTER_TUNNEL:
1594                 sfc_err(sa, "Tunnel filters not supported");
1595                 break;
1596         case RTE_ETH_FILTER_FDIR:
1597                 sfc_err(sa, "Flow Director filters not supported");
1598                 break;
1599         case RTE_ETH_FILTER_HASH:
1600                 sfc_err(sa, "Hash filters not supported");
1601                 break;
1602         case RTE_ETH_FILTER_GENERIC:
1603                 if (filter_op != RTE_ETH_FILTER_GET) {
1604                         rc = EINVAL;
1605                 } else {
1606                         *(const void **)arg = &sfc_flow_ops;
1607                         rc = 0;
1608                 }
1609                 break;
1610         default:
1611                 sfc_err(sa, "Unknown filter type %u", filter_type);
1612                 break;
1613         }
1614
1615         sfc_log_init(sa, "exit: %d", -rc);
1616         SFC_ASSERT(rc >= 0);
1617         return -rc;
1618 }
1619
1620 static const struct eth_dev_ops sfc_eth_dev_ops = {
1621         .dev_configure                  = sfc_dev_configure,
1622         .dev_start                      = sfc_dev_start,
1623         .dev_stop                       = sfc_dev_stop,
1624         .dev_set_link_up                = sfc_dev_set_link_up,
1625         .dev_set_link_down              = sfc_dev_set_link_down,
1626         .dev_close                      = sfc_dev_close,
1627         .promiscuous_enable             = sfc_dev_promisc_enable,
1628         .promiscuous_disable            = sfc_dev_promisc_disable,
1629         .allmulticast_enable            = sfc_dev_allmulti_enable,
1630         .allmulticast_disable           = sfc_dev_allmulti_disable,
1631         .link_update                    = sfc_dev_link_update,
1632         .stats_get                      = sfc_stats_get,
1633         .stats_reset                    = sfc_stats_reset,
1634         .xstats_get                     = sfc_xstats_get,
1635         .xstats_reset                   = sfc_stats_reset,
1636         .xstats_get_names               = sfc_xstats_get_names,
1637         .dev_infos_get                  = sfc_dev_infos_get,
1638         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1639         .mtu_set                        = sfc_dev_set_mtu,
1640         .rx_queue_start                 = sfc_rx_queue_start,
1641         .rx_queue_stop                  = sfc_rx_queue_stop,
1642         .tx_queue_start                 = sfc_tx_queue_start,
1643         .tx_queue_stop                  = sfc_tx_queue_stop,
1644         .rx_queue_setup                 = sfc_rx_queue_setup,
1645         .rx_queue_release               = sfc_rx_queue_release,
1646         .rx_queue_count                 = sfc_rx_queue_count,
1647         .rx_descriptor_done             = sfc_rx_descriptor_done,
1648         .rx_descriptor_status           = sfc_rx_descriptor_status,
1649         .tx_descriptor_status           = sfc_tx_descriptor_status,
1650         .tx_queue_setup                 = sfc_tx_queue_setup,
1651         .tx_queue_release               = sfc_tx_queue_release,
1652         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1653         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1654         .mac_addr_set                   = sfc_mac_addr_set,
1655         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1656         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1657 #if EFSYS_OPT_RX_SCALE
1658         .reta_update                    = sfc_dev_rss_reta_update,
1659         .reta_query                     = sfc_dev_rss_reta_query,
1660         .rss_hash_update                = sfc_dev_rss_hash_update,
1661         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1662 #endif
1663         .filter_ctrl                    = sfc_dev_filter_ctrl,
1664         .set_mc_addr_list               = sfc_set_mc_addr_list,
1665         .rxq_info_get                   = sfc_rx_queue_info_get,
1666         .txq_info_get                   = sfc_tx_queue_info_get,
1667         .fw_version_get                 = sfc_fw_version_get,
1668         .xstats_get_by_id               = sfc_xstats_get_by_id,
1669         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1670 };
1671
1672 /**
1673  * Duplicate a string in potentially shared memory required for
1674  * multi-process support.
1675  *
1676  * strdup() allocates from process-local heap/memory.
1677  */
1678 static char *
1679 sfc_strdup(const char *str)
1680 {
1681         size_t size;
1682         char *copy;
1683
1684         if (str == NULL)
1685                 return NULL;
1686
1687         size = strlen(str) + 1;
1688         copy = rte_malloc(__func__, size, 0);
1689         if (copy != NULL)
1690                 rte_memcpy(copy, str, size);
1691
1692         return copy;
1693 }
1694
1695 static int
1696 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1697 {
1698         struct sfc_adapter *sa = dev->data->dev_private;
1699         unsigned int avail_caps = 0;
1700         const char *rx_name = NULL;
1701         const char *tx_name = NULL;
1702         int rc;
1703
1704         switch (sa->family) {
1705         case EFX_FAMILY_HUNTINGTON:
1706         case EFX_FAMILY_MEDFORD:
1707                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1708                 break;
1709         default:
1710                 break;
1711         }
1712
1713         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1714                                 sfc_kvarg_string_handler, &rx_name);
1715         if (rc != 0)
1716                 goto fail_kvarg_rx_datapath;
1717
1718         if (rx_name != NULL) {
1719                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1720                 if (sa->dp_rx == NULL) {
1721                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1722                         rc = ENOENT;
1723                         goto fail_dp_rx;
1724                 }
1725                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1726                         sfc_err(sa,
1727                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1728                                 rx_name);
1729                         rc = EINVAL;
1730                         goto fail_dp_rx_caps;
1731                 }
1732         } else {
1733                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1734                 if (sa->dp_rx == NULL) {
1735                         sfc_err(sa, "Rx datapath by caps %#x not found",
1736                                 avail_caps);
1737                         rc = ENOENT;
1738                         goto fail_dp_rx;
1739                 }
1740         }
1741
1742         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1743         if (sa->dp_rx_name == NULL) {
1744                 rc = ENOMEM;
1745                 goto fail_dp_rx_name;
1746         }
1747
1748         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1749
1750         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1751
1752         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1753                                 sfc_kvarg_string_handler, &tx_name);
1754         if (rc != 0)
1755                 goto fail_kvarg_tx_datapath;
1756
1757         if (tx_name != NULL) {
1758                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1759                 if (sa->dp_tx == NULL) {
1760                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1761                         rc = ENOENT;
1762                         goto fail_dp_tx;
1763                 }
1764                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1765                         sfc_err(sa,
1766                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1767                                 tx_name);
1768                         rc = EINVAL;
1769                         goto fail_dp_tx_caps;
1770                 }
1771         } else {
1772                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1773                 if (sa->dp_tx == NULL) {
1774                         sfc_err(sa, "Tx datapath by caps %#x not found",
1775                                 avail_caps);
1776                         rc = ENOENT;
1777                         goto fail_dp_tx;
1778                 }
1779         }
1780
1781         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1782         if (sa->dp_tx_name == NULL) {
1783                 rc = ENOMEM;
1784                 goto fail_dp_tx_name;
1785         }
1786
1787         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1788
1789         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1790
1791         dev->dev_ops = &sfc_eth_dev_ops;
1792
1793         return 0;
1794
1795 fail_dp_tx_name:
1796 fail_dp_tx_caps:
1797         sa->dp_tx = NULL;
1798
1799 fail_dp_tx:
1800 fail_kvarg_tx_datapath:
1801         rte_free(sa->dp_rx_name);
1802         sa->dp_rx_name = NULL;
1803
1804 fail_dp_rx_name:
1805 fail_dp_rx_caps:
1806         sa->dp_rx = NULL;
1807
1808 fail_dp_rx:
1809 fail_kvarg_rx_datapath:
1810         return rc;
1811 }
1812
1813 static void
1814 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1815 {
1816         struct sfc_adapter *sa = dev->data->dev_private;
1817
1818         dev->dev_ops = NULL;
1819         dev->rx_pkt_burst = NULL;
1820         dev->tx_pkt_burst = NULL;
1821
1822         rte_free(sa->dp_tx_name);
1823         sa->dp_tx_name = NULL;
1824         sa->dp_tx = NULL;
1825
1826         rte_free(sa->dp_rx_name);
1827         sa->dp_rx_name = NULL;
1828         sa->dp_rx = NULL;
1829 }
1830
1831 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1832         .rxq_info_get                   = sfc_rx_queue_info_get,
1833         .txq_info_get                   = sfc_tx_queue_info_get,
1834 };
1835
1836 static int
1837 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1838 {
1839         /*
1840          * Device private data has really many process-local pointers.
1841          * Below code should be extremely careful to use data located
1842          * in shared memory only.
1843          */
1844         struct sfc_adapter *sa = dev->data->dev_private;
1845         const struct sfc_dp_rx *dp_rx;
1846         const struct sfc_dp_tx *dp_tx;
1847         int rc;
1848
1849         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1850         if (dp_rx == NULL) {
1851                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1852                 rc = ENOENT;
1853                 goto fail_dp_rx;
1854         }
1855         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1856                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1857                         sa->dp_tx_name);
1858                 rc = EINVAL;
1859                 goto fail_dp_rx_multi_process;
1860         }
1861
1862         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1863         if (dp_tx == NULL) {
1864                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1865                 rc = ENOENT;
1866                 goto fail_dp_tx;
1867         }
1868         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1869                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1870                         sa->dp_tx_name);
1871                 rc = EINVAL;
1872                 goto fail_dp_tx_multi_process;
1873         }
1874
1875         dev->rx_pkt_burst = dp_rx->pkt_burst;
1876         dev->tx_pkt_burst = dp_tx->pkt_burst;
1877         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1878
1879         return 0;
1880
1881 fail_dp_tx_multi_process:
1882 fail_dp_tx:
1883 fail_dp_rx_multi_process:
1884 fail_dp_rx:
1885         return rc;
1886 }
1887
1888 static void
1889 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1890 {
1891         dev->dev_ops = NULL;
1892         dev->tx_pkt_burst = NULL;
1893         dev->rx_pkt_burst = NULL;
1894 }
1895
1896 static void
1897 sfc_register_dp(void)
1898 {
1899         /* Register once */
1900         if (TAILQ_EMPTY(&sfc_dp_head)) {
1901                 /* Prefer EF10 datapath */
1902                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1903                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1904
1905                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1906                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1907                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1908         }
1909 }
1910
1911 static int
1912 sfc_eth_dev_init(struct rte_eth_dev *dev)
1913 {
1914         struct sfc_adapter *sa = dev->data->dev_private;
1915         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1916         int rc;
1917         const efx_nic_cfg_t *encp;
1918         const struct ether_addr *from;
1919
1920         sfc_register_dp();
1921
1922         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1923                 return -sfc_eth_dev_secondary_set_ops(dev);
1924
1925         /* Required for logging */
1926         sa->pci_addr = pci_dev->addr;
1927         sa->port_id = dev->data->port_id;
1928
1929         sa->eth_dev = dev;
1930
1931         /* Copy PCI device info to the dev->data */
1932         rte_eth_copy_pci_info(dev, pci_dev);
1933
1934         rc = sfc_kvargs_parse(sa);
1935         if (rc != 0)
1936                 goto fail_kvargs_parse;
1937
1938         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1939                                 sfc_kvarg_bool_handler, &sa->debug_init);
1940         if (rc != 0)
1941                 goto fail_kvarg_debug_init;
1942
1943         sfc_log_init(sa, "entry");
1944
1945         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1946         if (dev->data->mac_addrs == NULL) {
1947                 rc = ENOMEM;
1948                 goto fail_mac_addrs;
1949         }
1950
1951         sfc_adapter_lock_init(sa);
1952         sfc_adapter_lock(sa);
1953
1954         sfc_log_init(sa, "probing");
1955         rc = sfc_probe(sa);
1956         if (rc != 0)
1957                 goto fail_probe;
1958
1959         sfc_log_init(sa, "set device ops");
1960         rc = sfc_eth_dev_set_ops(dev);
1961         if (rc != 0)
1962                 goto fail_set_ops;
1963
1964         sfc_log_init(sa, "attaching");
1965         rc = sfc_attach(sa);
1966         if (rc != 0)
1967                 goto fail_attach;
1968
1969         encp = efx_nic_cfg_get(sa->nic);
1970
1971         /*
1972          * The arguments are really reverse order in comparison to
1973          * Linux kernel. Copy from NIC config to Ethernet device data.
1974          */
1975         from = (const struct ether_addr *)(encp->enc_mac_addr);
1976         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1977
1978         sfc_adapter_unlock(sa);
1979
1980         sfc_log_init(sa, "done");
1981         return 0;
1982
1983 fail_attach:
1984         sfc_eth_dev_clear_ops(dev);
1985
1986 fail_set_ops:
1987         sfc_unprobe(sa);
1988
1989 fail_probe:
1990         sfc_adapter_unlock(sa);
1991         sfc_adapter_lock_fini(sa);
1992         rte_free(dev->data->mac_addrs);
1993         dev->data->mac_addrs = NULL;
1994
1995 fail_mac_addrs:
1996 fail_kvarg_debug_init:
1997         sfc_kvargs_cleanup(sa);
1998
1999 fail_kvargs_parse:
2000         sfc_log_init(sa, "failed %d", rc);
2001         SFC_ASSERT(rc > 0);
2002         return -rc;
2003 }
2004
2005 static int
2006 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2007 {
2008         struct sfc_adapter *sa;
2009
2010         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2011                 sfc_eth_dev_secondary_clear_ops(dev);
2012                 return 0;
2013         }
2014
2015         sa = dev->data->dev_private;
2016         sfc_log_init(sa, "entry");
2017
2018         sfc_adapter_lock(sa);
2019
2020         sfc_eth_dev_clear_ops(dev);
2021
2022         sfc_detach(sa);
2023         sfc_unprobe(sa);
2024
2025         rte_free(dev->data->mac_addrs);
2026         dev->data->mac_addrs = NULL;
2027
2028         sfc_kvargs_cleanup(sa);
2029
2030         sfc_adapter_unlock(sa);
2031         sfc_adapter_lock_fini(sa);
2032
2033         sfc_log_init(sa, "done");
2034
2035         /* Required for logging, so cleanup last */
2036         sa->eth_dev = NULL;
2037         return 0;
2038 }
2039
2040 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2041         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2042         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2043         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2044         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2045         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2046         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2047         { .vendor_id = 0 /* sentinel */ }
2048 };
2049
2050 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2051         struct rte_pci_device *pci_dev)
2052 {
2053         return rte_eth_dev_pci_generic_probe(pci_dev,
2054                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
2055 }
2056
2057 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2058 {
2059         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2060 }
2061
2062 static struct rte_pci_driver sfc_efx_pmd = {
2063         .id_table = pci_id_sfc_efx_map,
2064         .drv_flags =
2065                 RTE_PCI_DRV_INTR_LSC |
2066                 RTE_PCI_DRV_NEED_MAPPING,
2067         .probe = sfc_eth_dev_pci_probe,
2068         .remove = sfc_eth_dev_pci_remove,
2069 };
2070
2071 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2072 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2073 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2074 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2075         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2076         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2077         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2078         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
2079         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
2080         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);