drivers/net/sfc/sfc_ethdev.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  * Copyright (c) 2016-2017 Solarflare Communications Inc.
   5  * All rights reserved.
   6  *
   7  * This software was jointly developed between OKTET Labs (under contract
   8  * for Solarflare) and Solarflare Communications, Inc.
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright notice,
  14  *    this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright notice,
  16  *    this list of conditions and the following disclaimer in the documentation
  17  *    and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
  20  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
  21  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  22  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
  23  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  24  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  25  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
  26  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  27  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
  28  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
  29  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  30  */
  31
  32 #include <rte_dev.h>
  33 #include <rte_ethdev.h>
  34 #include <rte_ethdev_pci.h>
  35 #include <rte_pci.h>
  36 #include <rte_bus_pci.h>
  37 #include <rte_errno.h>
  38
  39 #include "efx.h"
  40
  41 #include "sfc.h"
  42 #include "sfc_debug.h"
  43 #include "sfc_log.h"
  44 #include "sfc_kvargs.h"
  45 #include "sfc_ev.h"
  46 #include "sfc_rx.h"
  47 #include "sfc_tx.h"
  48 #include "sfc_flow.h"
  49 #include "sfc_dp.h"
  50 #include "sfc_dp_rx.h"
  51
  52 static struct sfc_dp_list sfc_dp_head =
  53         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  54
  55 static int
  56 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  57 {
  58         struct sfc_adapter *sa = dev->data->dev_private;
  59         efx_nic_fw_info_t enfi;
  60         int ret;
  61         int rc;
  62
  63         /*
  64          * Return value of the callback is likely supposed to be
  65          * equal to or greater than 0, nevertheless, if an error
  66          * occurs, it will be desirable to pass it to the caller
  67          */
  68         if ((fw_version == NULL) || (fw_size == 0))
  69                 return -EINVAL;
  70
  71         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  72         if (rc != 0)
  73                 return -rc;
  74
  75         ret = snprintf(fw_version, fw_size,
  76                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  77                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  78                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  79         if (ret < 0)
  80                 return ret;
  81
  82         if (enfi.enfi_dpcpu_fw_ids_valid) {
  83                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  84                 int ret_extra;
  85
  86                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  87                                      fw_size - dpcpu_fw_ids_offset,
  88                                      " rx%" PRIx16 " tx%" PRIx16,
  89                                      enfi.enfi_rx_dpcpu_fw_id,
  90                                      enfi.enfi_tx_dpcpu_fw_id);
  91                 if (ret_extra < 0)
  92                         return ret_extra;
  93
  94                 ret += ret_extra;
  95         }
  96
  97         if (fw_size < (size_t)(++ret))
  98                 return ret;
  99         else
 100                 return 0;
 101 }
 102
 103 static void
 104 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 105 {
 106         struct sfc_adapter *sa = dev->data->dev_private;
 107         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 108
 109         sfc_log_init(sa, "entry");
 110
 111         dev_info->pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 112         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 113
 114         /* Autonegotiation may be disabled */
 115         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 116         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
 117                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 118         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
 119                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 120         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 121                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 122
 123         dev_info->max_rx_queues = sa->rxq_max;
 124         dev_info->max_tx_queues = sa->txq_max;
 125
 126         /* By default packets are dropped if no descriptors are available */
 127         dev_info->default_rxconf.rx_drop_en = 1;
 128
 129         dev_info->rx_offload_capa =
 130                 DEV_RX_OFFLOAD_IPV4_CKSUM |
 131                 DEV_RX_OFFLOAD_UDP_CKSUM |
 132                 DEV_RX_OFFLOAD_TCP_CKSUM;
 133
 134         dev_info->tx_offload_capa =
 135                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 136                 DEV_TX_OFFLOAD_UDP_CKSUM |
 137                 DEV_TX_OFFLOAD_TCP_CKSUM;
 138
 139         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 140         if ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 141             !encp->enc_hw_tx_insert_vlan_enabled)
 142                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 143         else
 144                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 145
 146         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 147                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 148
 149         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_POOL)
 150                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTMEMP;
 151
 152         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_REFCNT)
 153                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOREFCOUNT;
 154
 155 #if EFSYS_OPT_RX_SCALE
 156         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 157                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 158                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 159                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 160         }
 161 #endif
 162
 163         if (sa->tso)
 164                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 165
 166         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 167         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 168         /* The RXQ hardware requires that the descriptor count is a power
 169          * of 2, but rx_desc_lim cannot properly describe that constraint.
 170          */
 171         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 172
 173         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 174         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 175         /*
 176          * The TXQ hardware requires that the descriptor count is a power
 177          * of 2, but tx_desc_lim cannot properly describe that constraint
 178          */
 179         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 180 }
 181
 182 static const uint32_t *
 183 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 184 {
 185         struct sfc_adapter *sa = dev->data->dev_private;
 186
 187         return sa->dp_rx->supported_ptypes_get();
 188 }
 189
 190 static int
 191 sfc_dev_configure(struct rte_eth_dev *dev)
 192 {
 193         struct rte_eth_dev_data *dev_data = dev->data;
 194         struct sfc_adapter *sa = dev_data->dev_private;
 195         int rc;
 196
 197         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 198                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 199
 200         sfc_adapter_lock(sa);
 201         switch (sa->state) {
 202         case SFC_ADAPTER_CONFIGURED:
 203                 /* FALLTHROUGH */
 204         case SFC_ADAPTER_INITIALIZED:
 205                 rc = sfc_configure(sa);
 206                 break;
 207         default:
 208                 sfc_err(sa, "unexpected adapter state %u to configure",
 209                         sa->state);
 210                 rc = EINVAL;
 211                 break;
 212         }
 213         sfc_adapter_unlock(sa);
 214
 215         sfc_log_init(sa, "done %d", rc);
 216         SFC_ASSERT(rc >= 0);
 217         return -rc;
 218 }
 219
 220 static int
 221 sfc_dev_start(struct rte_eth_dev *dev)
 222 {
 223         struct sfc_adapter *sa = dev->data->dev_private;
 224         int rc;
 225
 226         sfc_log_init(sa, "entry");
 227
 228         sfc_adapter_lock(sa);
 229         rc = sfc_start(sa);
 230         sfc_adapter_unlock(sa);
 231
 232         sfc_log_init(sa, "done %d", rc);
 233         SFC_ASSERT(rc >= 0);
 234         return -rc;
 235 }
 236
 237 static int
 238 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 239 {
 240         struct sfc_adapter *sa = dev->data->dev_private;
 241         struct rte_eth_link *dev_link = &dev->data->dev_link;
 242         struct rte_eth_link old_link;
 243         struct rte_eth_link current_link;
 244
 245         sfc_log_init(sa, "entry");
 246
 247 retry:
 248         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 249         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 250
 251         if (sa->state != SFC_ADAPTER_STARTED) {
 252                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 253                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 254                                          *(uint64_t *)&old_link,
 255                                          *(uint64_t *)&current_link))
 256                         goto retry;
 257         } else if (wait_to_complete) {
 258                 efx_link_mode_t link_mode;
 259
 260                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 261                         link_mode = EFX_LINK_UNKNOWN;
 262                 sfc_port_link_mode_to_info(link_mode, &current_link);
 263
 264                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 265                                          *(uint64_t *)&old_link,
 266                                          *(uint64_t *)&current_link))
 267                         goto retry;
 268         } else {
 269                 sfc_ev_mgmt_qpoll(sa);
 270                 *(int64_t *)&current_link =
 271                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 272         }
 273
 274         if (old_link.link_status != current_link.link_status)
 275                 sfc_info(sa, "Link status is %s",
 276                          current_link.link_status ? "UP" : "DOWN");
 277
 278         return old_link.link_status == current_link.link_status ? 0 : -1;
 279 }
 280
 281 static void
 282 sfc_dev_stop(struct rte_eth_dev *dev)
 283 {
 284         struct sfc_adapter *sa = dev->data->dev_private;
 285
 286         sfc_log_init(sa, "entry");
 287
 288         sfc_adapter_lock(sa);
 289         sfc_stop(sa);
 290         sfc_adapter_unlock(sa);
 291
 292         sfc_log_init(sa, "done");
 293 }
 294
 295 static int
 296 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 297 {
 298         struct sfc_adapter *sa = dev->data->dev_private;
 299         int rc;
 300
 301         sfc_log_init(sa, "entry");
 302
 303         sfc_adapter_lock(sa);
 304         rc = sfc_start(sa);
 305         sfc_adapter_unlock(sa);
 306
 307         SFC_ASSERT(rc >= 0);
 308         return -rc;
 309 }
 310
 311 static int
 312 sfc_dev_set_link_down(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         sfc_stop(sa);
 320         sfc_adapter_unlock(sa);
 321
 322         return 0;
 323 }
 324
 325 static void
 326 sfc_dev_close(struct rte_eth_dev *dev)
 327 {
 328         struct sfc_adapter *sa = dev->data->dev_private;
 329
 330         sfc_log_init(sa, "entry");
 331
 332         sfc_adapter_lock(sa);
 333         switch (sa->state) {
 334         case SFC_ADAPTER_STARTED:
 335                 sfc_stop(sa);
 336                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 337                 /* FALLTHROUGH */
 338         case SFC_ADAPTER_CONFIGURED:
 339                 sfc_close(sa);
 340                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 341                 /* FALLTHROUGH */
 342         case SFC_ADAPTER_INITIALIZED:
 343                 break;
 344         default:
 345                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 346                 break;
 347         }
 348         sfc_adapter_unlock(sa);
 349
 350         sfc_log_init(sa, "done");
 351 }
 352
 353 static void
 354 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 355                    boolean_t enabled)
 356 {
 357         struct sfc_port *port;
 358         boolean_t *toggle;
 359         struct sfc_adapter *sa = dev->data->dev_private;
 360         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 361         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 362
 363         sfc_adapter_lock(sa);
 364
 365         port = &sa->port;
 366         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 367
 368         if (*toggle != enabled) {
 369                 *toggle = enabled;
 370
 371                 if (port->isolated) {
 372                         sfc_warn(sa, "isolated mode is active on the port");
 373                         sfc_warn(sa, "the change is to be applied on the next "
 374                                      "start provided that isolated mode is "
 375                                      "disabled prior the next start");
 376                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 377                            (sfc_set_rx_mode(sa) != 0)) {
 378                         *toggle = !(enabled);
 379                         sfc_warn(sa, "Failed to %s %s mode",
 380                                  ((enabled) ? "enable" : "disable"), desc);
 381                 }
 382         }
 383
 384         sfc_adapter_unlock(sa);
 385 }
 386
 387 static void
 388 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 389 {
 390         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 391 }
 392
 393 static void
 394 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 395 {
 396         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 397 }
 398
 399 static void
 400 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 401 {
 402         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 403 }
 404
 405 static void
 406 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 407 {
 408         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 409 }
 410
 411 static int
 412 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 413                    uint16_t nb_rx_desc, unsigned int socket_id,
 414                    const struct rte_eth_rxconf *rx_conf,
 415                    struct rte_mempool *mb_pool)
 416 {
 417         struct sfc_adapter *sa = dev->data->dev_private;
 418         int rc;
 419
 420         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 421                      rx_queue_id, nb_rx_desc, socket_id);
 422
 423         sfc_adapter_lock(sa);
 424
 425         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 426                           rx_conf, mb_pool);
 427         if (rc != 0)
 428                 goto fail_rx_qinit;
 429
 430         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 431
 432         sfc_adapter_unlock(sa);
 433
 434         return 0;
 435
 436 fail_rx_qinit:
 437         sfc_adapter_unlock(sa);
 438         SFC_ASSERT(rc > 0);
 439         return -rc;
 440 }
 441
 442 static void
 443 sfc_rx_queue_release(void *queue)
 444 {
 445         struct sfc_dp_rxq *dp_rxq = queue;
 446         struct sfc_rxq *rxq;
 447         struct sfc_adapter *sa;
 448         unsigned int sw_index;
 449
 450         if (dp_rxq == NULL)
 451                 return;
 452
 453         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 454         sa = rxq->evq->sa;
 455         sfc_adapter_lock(sa);
 456
 457         sw_index = sfc_rxq_sw_index(rxq);
 458
 459         sfc_log_init(sa, "RxQ=%u", sw_index);
 460
 461         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 462
 463         sfc_rx_qfini(sa, sw_index);
 464
 465         sfc_adapter_unlock(sa);
 466 }
 467
 468 static int
 469 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 470                    uint16_t nb_tx_desc, unsigned int socket_id,
 471                    const struct rte_eth_txconf *tx_conf)
 472 {
 473         struct sfc_adapter *sa = dev->data->dev_private;
 474         int rc;
 475
 476         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 477                      tx_queue_id, nb_tx_desc, socket_id);
 478
 479         sfc_adapter_lock(sa);
 480
 481         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 482         if (rc != 0)
 483                 goto fail_tx_qinit;
 484
 485         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 486
 487         sfc_adapter_unlock(sa);
 488         return 0;
 489
 490 fail_tx_qinit:
 491         sfc_adapter_unlock(sa);
 492         SFC_ASSERT(rc > 0);
 493         return -rc;
 494 }
 495
 496 static void
 497 sfc_tx_queue_release(void *queue)
 498 {
 499         struct sfc_dp_txq *dp_txq = queue;
 500         struct sfc_txq *txq;
 501         unsigned int sw_index;
 502         struct sfc_adapter *sa;
 503
 504         if (dp_txq == NULL)
 505                 return;
 506
 507         txq = sfc_txq_by_dp_txq(dp_txq);
 508         sw_index = sfc_txq_sw_index(txq);
 509
 510         SFC_ASSERT(txq->evq != NULL);
 511         sa = txq->evq->sa;
 512
 513         sfc_log_init(sa, "TxQ = %u", sw_index);
 514
 515         sfc_adapter_lock(sa);
 516
 517         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 518         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 519
 520         sfc_tx_qfini(sa, sw_index);
 521
 522         sfc_adapter_unlock(sa);
 523 }
 524
 525 static int
 526 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 527 {
 528         struct sfc_adapter *sa = dev->data->dev_private;
 529         struct sfc_port *port = &sa->port;
 530         uint64_t *mac_stats;
 531         int ret;
 532
 533         rte_spinlock_lock(&port->mac_stats_lock);
 534
 535         ret = sfc_port_update_mac_stats(sa);
 536         if (ret != 0)
 537                 goto unlock;
 538
 539         mac_stats = port->mac_stats_buf;
 540
 541         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 542                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 543                 stats->ipackets =
 544                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 545                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 546                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 547                 stats->opackets =
 548                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 549                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 550                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 551                 stats->ibytes =
 552                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 553                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 554                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 555                 stats->obytes =
 556                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 557                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 558                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 559                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 560                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 561                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 562         } else {
 563                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 564                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 565                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 566                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 567                 /*
 568                  * Take into account stats which are whenever supported
 569                  * on EF10. If some stat is not supported by current
 570                  * firmware variant or HW revision, it is guaranteed
 571                  * to be zero in mac_stats.
 572                  */
 573                 stats->imissed =
 574                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 575                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 576                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 577                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 578                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 579                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 580                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 581                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 582                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 583                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 584                 stats->ierrors =
 585                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 586                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 587                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 588                 /* no oerrors counters supported on EF10 */
 589         }
 590
 591 unlock:
 592         rte_spinlock_unlock(&port->mac_stats_lock);
 593         SFC_ASSERT(ret >= 0);
 594         return -ret;
 595 }
 596
 597 static void
 598 sfc_stats_reset(struct rte_eth_dev *dev)
 599 {
 600         struct sfc_adapter *sa = dev->data->dev_private;
 601         struct sfc_port *port = &sa->port;
 602         int rc;
 603
 604         if (sa->state != SFC_ADAPTER_STARTED) {
 605                 /*
 606                  * The operation cannot be done if port is not started; it
 607                  * will be scheduled to be done during the next port start
 608                  */
 609                 port->mac_stats_reset_pending = B_TRUE;
 610                 return;
 611         }
 612
 613         rc = sfc_port_reset_mac_stats(sa);
 614         if (rc != 0)
 615                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 616 }
 617
 618 static int
 619 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 620                unsigned int xstats_count)
 621 {
 622         struct sfc_adapter *sa = dev->data->dev_private;
 623         struct sfc_port *port = &sa->port;
 624         uint64_t *mac_stats;
 625         int rc;
 626         unsigned int i;
 627         int nstats = 0;
 628
 629         rte_spinlock_lock(&port->mac_stats_lock);
 630
 631         rc = sfc_port_update_mac_stats(sa);
 632         if (rc != 0) {
 633                 SFC_ASSERT(rc > 0);
 634                 nstats = -rc;
 635                 goto unlock;
 636         }
 637
 638         mac_stats = port->mac_stats_buf;
 639
 640         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 641                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 642                         if (xstats != NULL && nstats < (int)xstats_count) {
 643                                 xstats[nstats].id = nstats;
 644                                 xstats[nstats].value = mac_stats[i];
 645                         }
 646                         nstats++;
 647                 }
 648         }
 649
 650 unlock:
 651         rte_spinlock_unlock(&port->mac_stats_lock);
 652
 653         return nstats;
 654 }
 655
 656 static int
 657 sfc_xstats_get_names(struct rte_eth_dev *dev,
 658                      struct rte_eth_xstat_name *xstats_names,
 659                      unsigned int xstats_count)
 660 {
 661         struct sfc_adapter *sa = dev->data->dev_private;
 662         struct sfc_port *port = &sa->port;
 663         unsigned int i;
 664         unsigned int nstats = 0;
 665
 666         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 667                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 668                         if (xstats_names != NULL && nstats < xstats_count)
 669                                 strncpy(xstats_names[nstats].name,
 670                                         efx_mac_stat_name(sa->nic, i),
 671                                         sizeof(xstats_names[0].name));
 672                         nstats++;
 673                 }
 674         }
 675
 676         return nstats;
 677 }
 678
 679 static int
 680 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 681                      uint64_t *values, unsigned int n)
 682 {
 683         struct sfc_adapter *sa = dev->data->dev_private;
 684         struct sfc_port *port = &sa->port;
 685         uint64_t *mac_stats;
 686         unsigned int nb_supported = 0;
 687         unsigned int nb_written = 0;
 688         unsigned int i;
 689         int ret;
 690         int rc;
 691
 692         if (unlikely(values == NULL) ||
 693             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 694                 return port->mac_stats_nb_supported;
 695
 696         rte_spinlock_lock(&port->mac_stats_lock);
 697
 698         rc = sfc_port_update_mac_stats(sa);
 699         if (rc != 0) {
 700                 SFC_ASSERT(rc > 0);
 701                 ret = -rc;
 702                 goto unlock;
 703         }
 704
 705         mac_stats = port->mac_stats_buf;
 706
 707         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 708                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 709                         continue;
 710
 711                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 712                         values[nb_written++] = mac_stats[i];
 713
 714                 ++nb_supported;
 715         }
 716
 717         ret = nb_written;
 718
 719 unlock:
 720         rte_spinlock_unlock(&port->mac_stats_lock);
 721
 722         return ret;
 723 }
 724
 725 static int
 726 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 727                            struct rte_eth_xstat_name *xstats_names,
 728                            const uint64_t *ids, unsigned int size)
 729 {
 730         struct sfc_adapter *sa = dev->data->dev_private;
 731         struct sfc_port *port = &sa->port;
 732         unsigned int nb_supported = 0;
 733         unsigned int nb_written = 0;
 734         unsigned int i;
 735
 736         if (unlikely(xstats_names == NULL) ||
 737             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 738                 return port->mac_stats_nb_supported;
 739
 740         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 741                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 742                         continue;
 743
 744                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 745                         char *name = xstats_names[nb_written++].name;
 746
 747                         strncpy(name, efx_mac_stat_name(sa->nic, i),
 748                                 sizeof(xstats_names[0].name));
 749                         name[sizeof(xstats_names[0].name) - 1] = '\0';
 750                 }
 751
 752                 ++nb_supported;
 753         }
 754
 755         return nb_written;
 756 }
 757
 758 static int
 759 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 760 {
 761         struct sfc_adapter *sa = dev->data->dev_private;
 762         unsigned int wanted_fc, link_fc;
 763
 764         memset(fc_conf, 0, sizeof(*fc_conf));
 765
 766         sfc_adapter_lock(sa);
 767
 768         if (sa->state == SFC_ADAPTER_STARTED)
 769                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 770         else
 771                 link_fc = sa->port.flow_ctrl;
 772
 773         switch (link_fc) {
 774         case 0:
 775                 fc_conf->mode = RTE_FC_NONE;
 776                 break;
 777         case EFX_FCNTL_RESPOND:
 778                 fc_conf->mode = RTE_FC_RX_PAUSE;
 779                 break;
 780         case EFX_FCNTL_GENERATE:
 781                 fc_conf->mode = RTE_FC_TX_PAUSE;
 782                 break;
 783         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 784                 fc_conf->mode = RTE_FC_FULL;
 785                 break;
 786         default:
 787                 sfc_err(sa, "%s: unexpected flow control value %#x",
 788                         __func__, link_fc);
 789         }
 790
 791         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 792
 793         sfc_adapter_unlock(sa);
 794
 795         return 0;
 796 }
 797
 798 static int
 799 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 800 {
 801         struct sfc_adapter *sa = dev->data->dev_private;
 802         struct sfc_port *port = &sa->port;
 803         unsigned int fcntl;
 804         int rc;
 805
 806         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 807             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 808             fc_conf->mac_ctrl_frame_fwd != 0) {
 809                 sfc_err(sa, "unsupported flow control settings specified");
 810                 rc = EINVAL;
 811                 goto fail_inval;
 812         }
 813
 814         switch (fc_conf->mode) {
 815         case RTE_FC_NONE:
 816                 fcntl = 0;
 817                 break;
 818         case RTE_FC_RX_PAUSE:
 819                 fcntl = EFX_FCNTL_RESPOND;
 820                 break;
 821         case RTE_FC_TX_PAUSE:
 822                 fcntl = EFX_FCNTL_GENERATE;
 823                 break;
 824         case RTE_FC_FULL:
 825                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 826                 break;
 827         default:
 828                 rc = EINVAL;
 829                 goto fail_inval;
 830         }
 831
 832         sfc_adapter_lock(sa);
 833
 834         if (sa->state == SFC_ADAPTER_STARTED) {
 835                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 836                 if (rc != 0)
 837                         goto fail_mac_fcntl_set;
 838         }
 839
 840         port->flow_ctrl = fcntl;
 841         port->flow_ctrl_autoneg = fc_conf->autoneg;
 842
 843         sfc_adapter_unlock(sa);
 844
 845         return 0;
 846
 847 fail_mac_fcntl_set:
 848         sfc_adapter_unlock(sa);
 849 fail_inval:
 850         SFC_ASSERT(rc > 0);
 851         return -rc;
 852 }
 853
 854 static int
 855 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 856 {
 857         struct sfc_adapter *sa = dev->data->dev_private;
 858         size_t pdu = EFX_MAC_PDU(mtu);
 859         size_t old_pdu;
 860         int rc;
 861
 862         sfc_log_init(sa, "mtu=%u", mtu);
 863
 864         rc = EINVAL;
 865         if (pdu < EFX_MAC_PDU_MIN) {
 866                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 867                         (unsigned int)mtu, (unsigned int)pdu,
 868                         EFX_MAC_PDU_MIN);
 869                 goto fail_inval;
 870         }
 871         if (pdu > EFX_MAC_PDU_MAX) {
 872                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 873                         (unsigned int)mtu, (unsigned int)pdu,
 874                         EFX_MAC_PDU_MAX);
 875                 goto fail_inval;
 876         }
 877
 878         sfc_adapter_lock(sa);
 879
 880         if (pdu != sa->port.pdu) {
 881                 if (sa->state == SFC_ADAPTER_STARTED) {
 882                         sfc_stop(sa);
 883
 884                         old_pdu = sa->port.pdu;
 885                         sa->port.pdu = pdu;
 886                         rc = sfc_start(sa);
 887                         if (rc != 0)
 888                                 goto fail_start;
 889                 } else {
 890                         sa->port.pdu = pdu;
 891                 }
 892         }
 893
 894         /*
 895          * The driver does not use it, but other PMDs update jumbo_frame
 896          * flag and max_rx_pkt_len when MTU is set.
 897          */
 898         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 899         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 900
 901         sfc_adapter_unlock(sa);
 902
 903         sfc_log_init(sa, "done");
 904         return 0;
 905
 906 fail_start:
 907         sa->port.pdu = old_pdu;
 908         if (sfc_start(sa) != 0)
 909                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 910                         "PDU max size - port is stopped",
 911                         (unsigned int)pdu, (unsigned int)old_pdu);
 912         sfc_adapter_unlock(sa);
 913
 914 fail_inval:
 915         sfc_log_init(sa, "failed %d", rc);
 916         SFC_ASSERT(rc > 0);
 917         return -rc;
 918 }
 919 static void
 920 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 921 {
 922         struct sfc_adapter *sa = dev->data->dev_private;
 923         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 924         struct sfc_port *port = &sa->port;
 925         int rc;
 926
 927         sfc_adapter_lock(sa);
 928
 929         /*
 930          * Copy the address to the device private data so that
 931          * it could be recalled in the case of adapter restart.
 932          */
 933         ether_addr_copy(mac_addr, &port->default_mac_addr);
 934
 935         if (port->isolated) {
 936                 sfc_err(sa, "isolated mode is active on the port");
 937                 sfc_err(sa, "will not set MAC address");
 938                 goto unlock;
 939         }
 940
 941         if (sa->state != SFC_ADAPTER_STARTED) {
 942                 sfc_info(sa, "the port is not started");
 943                 sfc_info(sa, "the new MAC address will be set on port start");
 944
 945                 goto unlock;
 946         }
 947
 948         if (encp->enc_allow_set_mac_with_installed_filters) {
 949                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 950                 if (rc != 0) {
 951                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 952                         goto unlock;
 953                 }
 954
 955                 /*
 956                  * Changing the MAC address by means of MCDI request
 957                  * has no effect on received traffic, therefore
 958                  * we also need to update unicast filters
 959                  */
 960                 rc = sfc_set_rx_mode(sa);
 961                 if (rc != 0)
 962                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 963         } else {
 964                 sfc_warn(sa, "cannot set MAC address with filters installed");
 965                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 966                 sfc_warn(sa, "(some traffic may be dropped)");
 967
 968                 /*
 969                  * Since setting MAC address with filters installed is not
 970                  * allowed on the adapter, the new MAC address will be set
 971                  * by means of adapter restart. sfc_start() shall retrieve
 972                  * the new address from the device private data and set it.
 973                  */
 974                 sfc_stop(sa);
 975                 rc = sfc_start(sa);
 976                 if (rc != 0)
 977                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 978         }
 979
 980 unlock:
 981         /*
 982          * In the case of failure sa->port->default_mac_addr does not
 983          * need rollback since no error code is returned, and the upper
 984          * API will anyway update the external MAC address storage.
 985          * To be consistent with that new value it is better to keep
 986          * the device private value the same.
 987          */
 988         sfc_adapter_unlock(sa);
 989 }
 990
 991
 992 static int
 993 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 994                      uint32_t nb_mc_addr)
 995 {
 996         struct sfc_adapter *sa = dev->data->dev_private;
 997         struct sfc_port *port = &sa->port;
 998         uint8_t *mc_addrs = port->mcast_addrs;
 999         int rc;
1000         unsigned int i;
1001
1002         if (port->isolated) {
1003                 sfc_err(sa, "isolated mode is active on the port");
1004                 sfc_err(sa, "will not set multicast address list");
1005                 return -ENOTSUP;
1006         }
1007
1008         if (mc_addrs == NULL)
1009                 return -ENOBUFS;
1010
1011         if (nb_mc_addr > port->max_mcast_addrs) {
1012                 sfc_err(sa, "too many multicast addresses: %u > %u",
1013                          nb_mc_addr, port->max_mcast_addrs);
1014                 return -EINVAL;
1015         }
1016
1017         for (i = 0; i < nb_mc_addr; ++i) {
1018                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1019                                  EFX_MAC_ADDR_LEN);
1020                 mc_addrs += EFX_MAC_ADDR_LEN;
1021         }
1022
1023         port->nb_mcast_addrs = nb_mc_addr;
1024
1025         if (sa->state != SFC_ADAPTER_STARTED)
1026                 return 0;
1027
1028         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1029                                         port->nb_mcast_addrs);
1030         if (rc != 0)
1031                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1032
1033         SFC_ASSERT(rc > 0);
1034         return -rc;
1035 }
1036
1037 /*
1038  * The function is used by the secondary process as well. It must not
1039  * use any process-local pointers from the adapter data.
1040  */
1041 static void
1042 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1043                       struct rte_eth_rxq_info *qinfo)
1044 {
1045         struct sfc_adapter *sa = dev->data->dev_private;
1046         struct sfc_rxq_info *rxq_info;
1047         struct sfc_rxq *rxq;
1048
1049         sfc_adapter_lock(sa);
1050
1051         SFC_ASSERT(rx_queue_id < sa->rxq_count);
1052
1053         rxq_info = &sa->rxq_info[rx_queue_id];
1054         rxq = rxq_info->rxq;
1055         SFC_ASSERT(rxq != NULL);
1056
1057         qinfo->mp = rxq->refill_mb_pool;
1058         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
1059         qinfo->conf.rx_drop_en = 1;
1060         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1061         qinfo->scattered_rx =
1062                 ((rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) != 0);
1063         qinfo->nb_desc = rxq_info->entries;
1064
1065         sfc_adapter_unlock(sa);
1066 }
1067
1068 /*
1069  * The function is used by the secondary process as well. It must not
1070  * use any process-local pointers from the adapter data.
1071  */
1072 static void
1073 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1074                       struct rte_eth_txq_info *qinfo)
1075 {
1076         struct sfc_adapter *sa = dev->data->dev_private;
1077         struct sfc_txq_info *txq_info;
1078
1079         sfc_adapter_lock(sa);
1080
1081         SFC_ASSERT(tx_queue_id < sa->txq_count);
1082
1083         txq_info = &sa->txq_info[tx_queue_id];
1084         SFC_ASSERT(txq_info->txq != NULL);
1085
1086         memset(qinfo, 0, sizeof(*qinfo));
1087
1088         qinfo->conf.txq_flags = txq_info->txq->flags;
1089         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
1090         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1091         qinfo->nb_desc = txq_info->entries;
1092
1093         sfc_adapter_unlock(sa);
1094 }
1095
1096 static uint32_t
1097 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1098 {
1099         struct sfc_adapter *sa = dev->data->dev_private;
1100
1101         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1102
1103         return sfc_rx_qdesc_npending(sa, rx_queue_id);
1104 }
1105
1106 static int
1107 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1108 {
1109         struct sfc_dp_rxq *dp_rxq = queue;
1110
1111         return sfc_rx_qdesc_done(dp_rxq, offset);
1112 }
1113
1114 static int
1115 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1116 {
1117         struct sfc_dp_rxq *dp_rxq = queue;
1118         struct sfc_rxq *rxq = sfc_rxq_by_dp_rxq(dp_rxq);
1119
1120         return rxq->evq->sa->dp_rx->qdesc_status(dp_rxq, offset);
1121 }
1122
1123 static int
1124 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1125 {
1126         struct sfc_dp_txq *dp_txq = queue;
1127         struct sfc_txq *txq = sfc_txq_by_dp_txq(dp_txq);
1128
1129         return txq->evq->sa->dp_tx->qdesc_status(dp_txq, offset);
1130 }
1131
1132 static int
1133 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1134 {
1135         struct sfc_adapter *sa = dev->data->dev_private;
1136         int rc;
1137
1138         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1139
1140         sfc_adapter_lock(sa);
1141
1142         rc = EINVAL;
1143         if (sa->state != SFC_ADAPTER_STARTED)
1144                 goto fail_not_started;
1145
1146         rc = sfc_rx_qstart(sa, rx_queue_id);
1147         if (rc != 0)
1148                 goto fail_rx_qstart;
1149
1150         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1151
1152         sfc_adapter_unlock(sa);
1153
1154         return 0;
1155
1156 fail_rx_qstart:
1157 fail_not_started:
1158         sfc_adapter_unlock(sa);
1159         SFC_ASSERT(rc > 0);
1160         return -rc;
1161 }
1162
1163 static int
1164 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1165 {
1166         struct sfc_adapter *sa = dev->data->dev_private;
1167
1168         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1169
1170         sfc_adapter_lock(sa);
1171         sfc_rx_qstop(sa, rx_queue_id);
1172
1173         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1174
1175         sfc_adapter_unlock(sa);
1176
1177         return 0;
1178 }
1179
1180 static int
1181 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1182 {
1183         struct sfc_adapter *sa = dev->data->dev_private;
1184         int rc;
1185
1186         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1187
1188         sfc_adapter_lock(sa);
1189
1190         rc = EINVAL;
1191         if (sa->state != SFC_ADAPTER_STARTED)
1192                 goto fail_not_started;
1193
1194         rc = sfc_tx_qstart(sa, tx_queue_id);
1195         if (rc != 0)
1196                 goto fail_tx_qstart;
1197
1198         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1199
1200         sfc_adapter_unlock(sa);
1201         return 0;
1202
1203 fail_tx_qstart:
1204
1205 fail_not_started:
1206         sfc_adapter_unlock(sa);
1207         SFC_ASSERT(rc > 0);
1208         return -rc;
1209 }
1210
1211 static int
1212 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1213 {
1214         struct sfc_adapter *sa = dev->data->dev_private;
1215
1216         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1217
1218         sfc_adapter_lock(sa);
1219
1220         sfc_tx_qstop(sa, tx_queue_id);
1221
1222         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1223
1224         sfc_adapter_unlock(sa);
1225         return 0;
1226 }
1227
1228 #if EFSYS_OPT_RX_SCALE
1229 static int
1230 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1231                           struct rte_eth_rss_conf *rss_conf)
1232 {
1233         struct sfc_adapter *sa = dev->data->dev_private;
1234         struct sfc_port *port = &sa->port;
1235
1236         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1237                 return -ENOTSUP;
1238
1239         if (sa->rss_channels == 0)
1240                 return -EINVAL;
1241
1242         sfc_adapter_lock(sa);
1243
1244         /*
1245          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1246          * hence, conversion is done here to derive a correct set of ETH_RSS
1247          * flags which corresponds to the active EFX configuration stored
1248          * locally in 'sfc_adapter' and kept up-to-date
1249          */
1250         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1251         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1252         if (rss_conf->rss_key != NULL)
1253                 rte_memcpy(rss_conf->rss_key, sa->rss_key, EFX_RSS_KEY_SIZE);
1254
1255         sfc_adapter_unlock(sa);
1256
1257         return 0;
1258 }
1259
1260 static int
1261 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1262                         struct rte_eth_rss_conf *rss_conf)
1263 {
1264         struct sfc_adapter *sa = dev->data->dev_private;
1265         struct sfc_port *port = &sa->port;
1266         unsigned int efx_hash_types;
1267         int rc = 0;
1268
1269         if (port->isolated)
1270                 return -ENOTSUP;
1271
1272         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1273                 sfc_err(sa, "RSS is not available");
1274                 return -ENOTSUP;
1275         }
1276
1277         if (sa->rss_channels == 0) {
1278                 sfc_err(sa, "RSS is not configured");
1279                 return -EINVAL;
1280         }
1281
1282         if ((rss_conf->rss_key != NULL) &&
1283             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1284                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1285                         sizeof(sa->rss_key));
1286                 return -EINVAL;
1287         }
1288
1289         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1290                 sfc_err(sa, "unsupported hash functions requested");
1291                 return -EINVAL;
1292         }
1293
1294         sfc_adapter_lock(sa);
1295
1296         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1297
1298         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1299                                    EFX_RX_HASHALG_TOEPLITZ,
1300                                    efx_hash_types, B_TRUE);
1301         if (rc != 0)
1302                 goto fail_scale_mode_set;
1303
1304         if (rss_conf->rss_key != NULL) {
1305                 if (sa->state == SFC_ADAPTER_STARTED) {
1306                         rc = efx_rx_scale_key_set(sa->nic,
1307                                                   EFX_RSS_CONTEXT_DEFAULT,
1308                                                   rss_conf->rss_key,
1309                                                   sizeof(sa->rss_key));
1310                         if (rc != 0)
1311                                 goto fail_scale_key_set;
1312                 }
1313
1314                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1315         }
1316
1317         sa->rss_hash_types = efx_hash_types;
1318
1319         sfc_adapter_unlock(sa);
1320
1321         return 0;
1322
1323 fail_scale_key_set:
1324         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1325                                   EFX_RX_HASHALG_TOEPLITZ,
1326                                   sa->rss_hash_types, B_TRUE) != 0)
1327                 sfc_err(sa, "failed to restore RSS mode");
1328
1329 fail_scale_mode_set:
1330         sfc_adapter_unlock(sa);
1331         return -rc;
1332 }
1333
1334 static int
1335 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1336                        struct rte_eth_rss_reta_entry64 *reta_conf,
1337                        uint16_t reta_size)
1338 {
1339         struct sfc_adapter *sa = dev->data->dev_private;
1340         struct sfc_port *port = &sa->port;
1341         int entry;
1342
1343         if ((sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) || port->isolated)
1344                 return -ENOTSUP;
1345
1346         if (sa->rss_channels == 0)
1347                 return -EINVAL;
1348
1349         if (reta_size != EFX_RSS_TBL_SIZE)
1350                 return -EINVAL;
1351
1352         sfc_adapter_lock(sa);
1353
1354         for (entry = 0; entry < reta_size; entry++) {
1355                 int grp = entry / RTE_RETA_GROUP_SIZE;
1356                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1357
1358                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1359                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1360         }
1361
1362         sfc_adapter_unlock(sa);
1363
1364         return 0;
1365 }
1366
1367 static int
1368 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1369                         struct rte_eth_rss_reta_entry64 *reta_conf,
1370                         uint16_t reta_size)
1371 {
1372         struct sfc_adapter *sa = dev->data->dev_private;
1373         struct sfc_port *port = &sa->port;
1374         unsigned int *rss_tbl_new;
1375         uint16_t entry;
1376         int rc = 0;
1377
1378
1379         if (port->isolated)
1380                 return -ENOTSUP;
1381
1382         if (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE) {
1383                 sfc_err(sa, "RSS is not available");
1384                 return -ENOTSUP;
1385         }
1386
1387         if (sa->rss_channels == 0) {
1388                 sfc_err(sa, "RSS is not configured");
1389                 return -EINVAL;
1390         }
1391
1392         if (reta_size != EFX_RSS_TBL_SIZE) {
1393                 sfc_err(sa, "RETA size is wrong (should be %u)",
1394                         EFX_RSS_TBL_SIZE);
1395                 return -EINVAL;
1396         }
1397
1398         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1399         if (rss_tbl_new == NULL)
1400                 return -ENOMEM;
1401
1402         sfc_adapter_lock(sa);
1403
1404         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1405
1406         for (entry = 0; entry < reta_size; entry++) {
1407                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1408                 struct rte_eth_rss_reta_entry64 *grp;
1409
1410                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1411
1412                 if (grp->mask & (1ull << grp_idx)) {
1413                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1414                                 rc = EINVAL;
1415                                 goto bad_reta_entry;
1416                         }
1417                         rss_tbl_new[entry] = grp->reta[grp_idx];
1418                 }
1419         }
1420
1421         if (sa->state == SFC_ADAPTER_STARTED) {
1422                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1423                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1424                 if (rc != 0)
1425                         goto fail_scale_tbl_set;
1426         }
1427
1428         rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1429
1430 fail_scale_tbl_set:
1431 bad_reta_entry:
1432         sfc_adapter_unlock(sa);
1433
1434         rte_free(rss_tbl_new);
1435
1436         SFC_ASSERT(rc >= 0);
1437         return -rc;
1438 }
1439 #endif
1440
1441 static int
1442 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1443                     enum rte_filter_op filter_op,
1444                     void *arg)
1445 {
1446         struct sfc_adapter *sa = dev->data->dev_private;
1447         int rc = ENOTSUP;
1448
1449         sfc_log_init(sa, "entry");
1450
1451         switch (filter_type) {
1452         case RTE_ETH_FILTER_NONE:
1453                 sfc_err(sa, "Global filters configuration not supported");
1454                 break;
1455         case RTE_ETH_FILTER_MACVLAN:
1456                 sfc_err(sa, "MACVLAN filters not supported");
1457                 break;
1458         case RTE_ETH_FILTER_ETHERTYPE:
1459                 sfc_err(sa, "EtherType filters not supported");
1460                 break;
1461         case RTE_ETH_FILTER_FLEXIBLE:
1462                 sfc_err(sa, "Flexible filters not supported");
1463                 break;
1464         case RTE_ETH_FILTER_SYN:
1465                 sfc_err(sa, "SYN filters not supported");
1466                 break;
1467         case RTE_ETH_FILTER_NTUPLE:
1468                 sfc_err(sa, "NTUPLE filters not supported");
1469                 break;
1470         case RTE_ETH_FILTER_TUNNEL:
1471                 sfc_err(sa, "Tunnel filters not supported");
1472                 break;
1473         case RTE_ETH_FILTER_FDIR:
1474                 sfc_err(sa, "Flow Director filters not supported");
1475                 break;
1476         case RTE_ETH_FILTER_HASH:
1477                 sfc_err(sa, "Hash filters not supported");
1478                 break;
1479         case RTE_ETH_FILTER_GENERIC:
1480                 if (filter_op != RTE_ETH_FILTER_GET) {
1481                         rc = EINVAL;
1482                 } else {
1483                         *(const void **)arg = &sfc_flow_ops;
1484                         rc = 0;
1485                 }
1486                 break;
1487         default:
1488                 sfc_err(sa, "Unknown filter type %u", filter_type);
1489                 break;
1490         }
1491
1492         sfc_log_init(sa, "exit: %d", -rc);
1493         SFC_ASSERT(rc >= 0);
1494         return -rc;
1495 }
1496
1497 static const struct eth_dev_ops sfc_eth_dev_ops = {
1498         .dev_configure                  = sfc_dev_configure,
1499         .dev_start                      = sfc_dev_start,
1500         .dev_stop                       = sfc_dev_stop,
1501         .dev_set_link_up                = sfc_dev_set_link_up,
1502         .dev_set_link_down              = sfc_dev_set_link_down,
1503         .dev_close                      = sfc_dev_close,
1504         .promiscuous_enable             = sfc_dev_promisc_enable,
1505         .promiscuous_disable            = sfc_dev_promisc_disable,
1506         .allmulticast_enable            = sfc_dev_allmulti_enable,
1507         .allmulticast_disable           = sfc_dev_allmulti_disable,
1508         .link_update                    = sfc_dev_link_update,
1509         .stats_get                      = sfc_stats_get,
1510         .stats_reset                    = sfc_stats_reset,
1511         .xstats_get                     = sfc_xstats_get,
1512         .xstats_reset                   = sfc_stats_reset,
1513         .xstats_get_names               = sfc_xstats_get_names,
1514         .dev_infos_get                  = sfc_dev_infos_get,
1515         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1516         .mtu_set                        = sfc_dev_set_mtu,
1517         .rx_queue_start                 = sfc_rx_queue_start,
1518         .rx_queue_stop                  = sfc_rx_queue_stop,
1519         .tx_queue_start                 = sfc_tx_queue_start,
1520         .tx_queue_stop                  = sfc_tx_queue_stop,
1521         .rx_queue_setup                 = sfc_rx_queue_setup,
1522         .rx_queue_release               = sfc_rx_queue_release,
1523         .rx_queue_count                 = sfc_rx_queue_count,
1524         .rx_descriptor_done             = sfc_rx_descriptor_done,
1525         .rx_descriptor_status           = sfc_rx_descriptor_status,
1526         .tx_descriptor_status           = sfc_tx_descriptor_status,
1527         .tx_queue_setup                 = sfc_tx_queue_setup,
1528         .tx_queue_release               = sfc_tx_queue_release,
1529         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1530         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1531         .mac_addr_set                   = sfc_mac_addr_set,
1532 #if EFSYS_OPT_RX_SCALE
1533         .reta_update                    = sfc_dev_rss_reta_update,
1534         .reta_query                     = sfc_dev_rss_reta_query,
1535         .rss_hash_update                = sfc_dev_rss_hash_update,
1536         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1537 #endif
1538         .filter_ctrl                    = sfc_dev_filter_ctrl,
1539         .set_mc_addr_list               = sfc_set_mc_addr_list,
1540         .rxq_info_get                   = sfc_rx_queue_info_get,
1541         .txq_info_get                   = sfc_tx_queue_info_get,
1542         .fw_version_get                 = sfc_fw_version_get,
1543         .xstats_get_by_id               = sfc_xstats_get_by_id,
1544         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1545 };
1546
1547 /**
1548  * Duplicate a string in potentially shared memory required for
1549  * multi-process support.
1550  *
1551  * strdup() allocates from process-local heap/memory.
1552  */
1553 static char *
1554 sfc_strdup(const char *str)
1555 {
1556         size_t size;
1557         char *copy;
1558
1559         if (str == NULL)
1560                 return NULL;
1561
1562         size = strlen(str) + 1;
1563         copy = rte_malloc(__func__, size, 0);
1564         if (copy != NULL)
1565                 rte_memcpy(copy, str, size);
1566
1567         return copy;
1568 }
1569
1570 static int
1571 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1572 {
1573         struct sfc_adapter *sa = dev->data->dev_private;
1574         unsigned int avail_caps = 0;
1575         const char *rx_name = NULL;
1576         const char *tx_name = NULL;
1577         int rc;
1578
1579         switch (sa->family) {
1580         case EFX_FAMILY_HUNTINGTON:
1581         case EFX_FAMILY_MEDFORD:
1582                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1583                 break;
1584         default:
1585                 break;
1586         }
1587
1588         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1589                                 sfc_kvarg_string_handler, &rx_name);
1590         if (rc != 0)
1591                 goto fail_kvarg_rx_datapath;
1592
1593         if (rx_name != NULL) {
1594                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1595                 if (sa->dp_rx == NULL) {
1596                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1597                         rc = ENOENT;
1598                         goto fail_dp_rx;
1599                 }
1600                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1601                         sfc_err(sa,
1602                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1603                                 rx_name);
1604                         rc = EINVAL;
1605                         goto fail_dp_rx_caps;
1606                 }
1607         } else {
1608                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1609                 if (sa->dp_rx == NULL) {
1610                         sfc_err(sa, "Rx datapath by caps %#x not found",
1611                                 avail_caps);
1612                         rc = ENOENT;
1613                         goto fail_dp_rx;
1614                 }
1615         }
1616
1617         sa->dp_rx_name = sfc_strdup(sa->dp_rx->dp.name);
1618         if (sa->dp_rx_name == NULL) {
1619                 rc = ENOMEM;
1620                 goto fail_dp_rx_name;
1621         }
1622
1623         sfc_info(sa, "use %s Rx datapath", sa->dp_rx_name);
1624
1625         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1626
1627         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1628                                 sfc_kvarg_string_handler, &tx_name);
1629         if (rc != 0)
1630                 goto fail_kvarg_tx_datapath;
1631
1632         if (tx_name != NULL) {
1633                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1634                 if (sa->dp_tx == NULL) {
1635                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1636                         rc = ENOENT;
1637                         goto fail_dp_tx;
1638                 }
1639                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1640                         sfc_err(sa,
1641                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1642                                 tx_name);
1643                         rc = EINVAL;
1644                         goto fail_dp_tx_caps;
1645                 }
1646         } else {
1647                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1648                 if (sa->dp_tx == NULL) {
1649                         sfc_err(sa, "Tx datapath by caps %#x not found",
1650                                 avail_caps);
1651                         rc = ENOENT;
1652                         goto fail_dp_tx;
1653                 }
1654         }
1655
1656         sa->dp_tx_name = sfc_strdup(sa->dp_tx->dp.name);
1657         if (sa->dp_tx_name == NULL) {
1658                 rc = ENOMEM;
1659                 goto fail_dp_tx_name;
1660         }
1661
1662         sfc_info(sa, "use %s Tx datapath", sa->dp_tx_name);
1663
1664         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1665
1666         dev->dev_ops = &sfc_eth_dev_ops;
1667
1668         return 0;
1669
1670 fail_dp_tx_name:
1671 fail_dp_tx_caps:
1672         sa->dp_tx = NULL;
1673
1674 fail_dp_tx:
1675 fail_kvarg_tx_datapath:
1676         rte_free(sa->dp_rx_name);
1677         sa->dp_rx_name = NULL;
1678
1679 fail_dp_rx_name:
1680 fail_dp_rx_caps:
1681         sa->dp_rx = NULL;
1682
1683 fail_dp_rx:
1684 fail_kvarg_rx_datapath:
1685         return rc;
1686 }
1687
1688 static void
1689 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1690 {
1691         struct sfc_adapter *sa = dev->data->dev_private;
1692
1693         dev->dev_ops = NULL;
1694         dev->rx_pkt_burst = NULL;
1695         dev->tx_pkt_burst = NULL;
1696
1697         rte_free(sa->dp_tx_name);
1698         sa->dp_tx_name = NULL;
1699         sa->dp_tx = NULL;
1700
1701         rte_free(sa->dp_rx_name);
1702         sa->dp_rx_name = NULL;
1703         sa->dp_rx = NULL;
1704 }
1705
1706 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1707         .rxq_info_get                   = sfc_rx_queue_info_get,
1708         .txq_info_get                   = sfc_tx_queue_info_get,
1709 };
1710
1711 static int
1712 sfc_eth_dev_secondary_set_ops(struct rte_eth_dev *dev)
1713 {
1714         /*
1715          * Device private data has really many process-local pointers.
1716          * Below code should be extremely careful to use data located
1717          * in shared memory only.
1718          */
1719         struct sfc_adapter *sa = dev->data->dev_private;
1720         const struct sfc_dp_rx *dp_rx;
1721         const struct sfc_dp_tx *dp_tx;
1722         int rc;
1723
1724         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sa->dp_rx_name);
1725         if (dp_rx == NULL) {
1726                 sfc_err(sa, "cannot find %s Rx datapath", sa->dp_tx_name);
1727                 rc = ENOENT;
1728                 goto fail_dp_rx;
1729         }
1730         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1731                 sfc_err(sa, "%s Rx datapath does not support multi-process",
1732                         sa->dp_tx_name);
1733                 rc = EINVAL;
1734                 goto fail_dp_rx_multi_process;
1735         }
1736
1737         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sa->dp_tx_name);
1738         if (dp_tx == NULL) {
1739                 sfc_err(sa, "cannot find %s Tx datapath", sa->dp_tx_name);
1740                 rc = ENOENT;
1741                 goto fail_dp_tx;
1742         }
1743         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1744                 sfc_err(sa, "%s Tx datapath does not support multi-process",
1745                         sa->dp_tx_name);
1746                 rc = EINVAL;
1747                 goto fail_dp_tx_multi_process;
1748         }
1749
1750         dev->rx_pkt_burst = dp_rx->pkt_burst;
1751         dev->tx_pkt_burst = dp_tx->pkt_burst;
1752         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1753
1754         return 0;
1755
1756 fail_dp_tx_multi_process:
1757 fail_dp_tx:
1758 fail_dp_rx_multi_process:
1759 fail_dp_rx:
1760         return rc;
1761 }
1762
1763 static void
1764 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1765 {
1766         dev->dev_ops = NULL;
1767         dev->tx_pkt_burst = NULL;
1768         dev->rx_pkt_burst = NULL;
1769 }
1770
1771 static void
1772 sfc_register_dp(void)
1773 {
1774         /* Register once */
1775         if (TAILQ_EMPTY(&sfc_dp_head)) {
1776                 /* Prefer EF10 datapath */
1777                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1778                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1779
1780                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1781                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1782                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1783         }
1784 }
1785
1786 static int
1787 sfc_eth_dev_init(struct rte_eth_dev *dev)
1788 {
1789         struct sfc_adapter *sa = dev->data->dev_private;
1790         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
1791         int rc;
1792         const efx_nic_cfg_t *encp;
1793         const struct ether_addr *from;
1794
1795         sfc_register_dp();
1796
1797         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1798                 return -sfc_eth_dev_secondary_set_ops(dev);
1799
1800         /* Required for logging */
1801         sa->pci_addr = pci_dev->addr;
1802         sa->port_id = dev->data->port_id;
1803
1804         sa->eth_dev = dev;
1805
1806         /* Copy PCI device info to the dev->data */
1807         rte_eth_copy_pci_info(dev, pci_dev);
1808
1809         rc = sfc_kvargs_parse(sa);
1810         if (rc != 0)
1811                 goto fail_kvargs_parse;
1812
1813         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1814                                 sfc_kvarg_bool_handler, &sa->debug_init);
1815         if (rc != 0)
1816                 goto fail_kvarg_debug_init;
1817
1818         sfc_log_init(sa, "entry");
1819
1820         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1821         if (dev->data->mac_addrs == NULL) {
1822                 rc = ENOMEM;
1823                 goto fail_mac_addrs;
1824         }
1825
1826         sfc_adapter_lock_init(sa);
1827         sfc_adapter_lock(sa);
1828
1829         sfc_log_init(sa, "probing");
1830         rc = sfc_probe(sa);
1831         if (rc != 0)
1832                 goto fail_probe;
1833
1834         sfc_log_init(sa, "set device ops");
1835         rc = sfc_eth_dev_set_ops(dev);
1836         if (rc != 0)
1837                 goto fail_set_ops;
1838
1839         sfc_log_init(sa, "attaching");
1840         rc = sfc_attach(sa);
1841         if (rc != 0)
1842                 goto fail_attach;
1843
1844         encp = efx_nic_cfg_get(sa->nic);
1845
1846         /*
1847          * The arguments are really reverse order in comparison to
1848          * Linux kernel. Copy from NIC config to Ethernet device data.
1849          */
1850         from = (const struct ether_addr *)(encp->enc_mac_addr);
1851         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1852
1853         sfc_adapter_unlock(sa);
1854
1855         sfc_log_init(sa, "done");
1856         return 0;
1857
1858 fail_attach:
1859         sfc_eth_dev_clear_ops(dev);
1860
1861 fail_set_ops:
1862         sfc_unprobe(sa);
1863
1864 fail_probe:
1865         sfc_adapter_unlock(sa);
1866         sfc_adapter_lock_fini(sa);
1867         rte_free(dev->data->mac_addrs);
1868         dev->data->mac_addrs = NULL;
1869
1870 fail_mac_addrs:
1871 fail_kvarg_debug_init:
1872         sfc_kvargs_cleanup(sa);
1873
1874 fail_kvargs_parse:
1875         sfc_log_init(sa, "failed %d", rc);
1876         SFC_ASSERT(rc > 0);
1877         return -rc;
1878 }
1879
1880 static int
1881 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1882 {
1883         struct sfc_adapter *sa;
1884
1885         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1886                 sfc_eth_dev_secondary_clear_ops(dev);
1887                 return 0;
1888         }
1889
1890         sa = dev->data->dev_private;
1891         sfc_log_init(sa, "entry");
1892
1893         sfc_adapter_lock(sa);
1894
1895         sfc_eth_dev_clear_ops(dev);
1896
1897         sfc_detach(sa);
1898         sfc_unprobe(sa);
1899
1900         rte_free(dev->data->mac_addrs);
1901         dev->data->mac_addrs = NULL;
1902
1903         sfc_kvargs_cleanup(sa);
1904
1905         sfc_adapter_unlock(sa);
1906         sfc_adapter_lock_fini(sa);
1907
1908         sfc_log_init(sa, "done");
1909
1910         /* Required for logging, so cleanup last */
1911         sa->eth_dev = NULL;
1912         return 0;
1913 }
1914
1915 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1916         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1917         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1918         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1919         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1920         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1921         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1922         { .vendor_id = 0 /* sentinel */ }
1923 };
1924
1925 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1926         struct rte_pci_device *pci_dev)
1927 {
1928         return rte_eth_dev_pci_generic_probe(pci_dev,
1929                 sizeof(struct sfc_adapter), sfc_eth_dev_init);
1930 }
1931
1932 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
1933 {
1934         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
1935 }
1936
1937 static struct rte_pci_driver sfc_efx_pmd = {
1938         .id_table = pci_id_sfc_efx_map,
1939         .drv_flags =
1940                 RTE_PCI_DRV_INTR_LSC |
1941                 RTE_PCI_DRV_NEED_MAPPING,
1942         .probe = sfc_eth_dev_pci_probe,
1943         .remove = sfc_eth_dev_pci_remove,
1944 };
1945
1946 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
1947 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1948 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
1949 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1950         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1951         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1952         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1953         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1954         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1955         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);