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_pci.h>
  35 #include <rte_errno.h>
  36
  37 #include "efx.h"
  38
  39 #include "sfc.h"
  40 #include "sfc_debug.h"
  41 #include "sfc_log.h"
  42 #include "sfc_kvargs.h"
  43 #include "sfc_ev.h"
  44 #include "sfc_rx.h"
  45 #include "sfc_tx.h"
  46 #include "sfc_flow.h"
  47 #include "sfc_dp.h"
  48 #include "sfc_dp_rx.h"
  49
  50 static struct sfc_dp_list sfc_dp_head =
  51         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  52
  53 static int
  54 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  55 {
  56         struct sfc_adapter *sa = dev->data->dev_private;
  57         efx_nic_fw_info_t enfi;
  58         int ret;
  59         int rc;
  60
  61         /*
  62          * Return value of the callback is likely supposed to be
  63          * equal to or greater than 0, nevertheless, if an error
  64          * occurs, it will be desirable to pass it to the caller
  65          */
  66         if ((fw_version == NULL) || (fw_size == 0))
  67                 return -EINVAL;
  68
  69         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  70         if (rc != 0)
  71                 return -rc;
  72
  73         ret = snprintf(fw_version, fw_size,
  74                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  75                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  76                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  77         if (ret < 0)
  78                 return ret;
  79
  80         if (enfi.enfi_dpcpu_fw_ids_valid) {
  81                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  82                 int ret_extra;
  83
  84                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  85                                      fw_size - dpcpu_fw_ids_offset,
  86                                      " rx%" PRIx16 " tx%" PRIx16,
  87                                      enfi.enfi_rx_dpcpu_fw_id,
  88                                      enfi.enfi_tx_dpcpu_fw_id);
  89                 if (ret_extra < 0)
  90                         return ret_extra;
  91
  92                 ret += ret_extra;
  93         }
  94
  95         if (fw_size < (size_t)(++ret))
  96                 return ret;
  97         else
  98                 return 0;
  99 }
 100
 101 static void
 102 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 103 {
 104         struct sfc_adapter *sa = dev->data->dev_private;
 105         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 106
 107         sfc_log_init(sa, "entry");
 108
 109         dev_info->pci_dev = RTE_DEV_TO_PCI(dev->device);
 110         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 111
 112         /* Autonegotiation may be disabled */
 113         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 114         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
 115                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 116         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
 117                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 118         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 119                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 120
 121         dev_info->max_rx_queues = sa->rxq_max;
 122         dev_info->max_tx_queues = sa->txq_max;
 123
 124         /* By default packets are dropped if no descriptors are available */
 125         dev_info->default_rxconf.rx_drop_en = 1;
 126
 127         dev_info->rx_offload_capa =
 128                 DEV_RX_OFFLOAD_IPV4_CKSUM |
 129                 DEV_RX_OFFLOAD_UDP_CKSUM |
 130                 DEV_RX_OFFLOAD_TCP_CKSUM;
 131
 132         dev_info->tx_offload_capa =
 133                 DEV_TX_OFFLOAD_IPV4_CKSUM |
 134                 DEV_TX_OFFLOAD_UDP_CKSUM |
 135                 DEV_TX_OFFLOAD_TCP_CKSUM;
 136
 137         dev_info->default_txconf.txq_flags = ETH_TXQ_FLAGS_NOXSUMSCTP;
 138         if (!encp->enc_hw_tx_insert_vlan_enabled)
 139                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 140         else
 141                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 142
 143 #if EFSYS_OPT_RX_SCALE
 144         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 145                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 146                 dev_info->hash_key_size = SFC_RSS_KEY_SIZE;
 147                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 148         }
 149 #endif
 150
 151         if (sa->tso)
 152                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 153
 154         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 155         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 156         /* The RXQ hardware requires that the descriptor count is a power
 157          * of 2, but rx_desc_lim cannot properly describe that constraint.
 158          */
 159         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 160
 161         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
 170 static const uint32_t *
 171 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 172 {
 173         struct sfc_adapter *sa = dev->data->dev_private;
 174
 175         return sa->dp_rx->supported_ptypes_get();
 176 }
 177
 178 static int
 179 sfc_dev_configure(struct rte_eth_dev *dev)
 180 {
 181         struct rte_eth_dev_data *dev_data = dev->data;
 182         struct sfc_adapter *sa = dev_data->dev_private;
 183         int rc;
 184
 185         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 186                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 187
 188         sfc_adapter_lock(sa);
 189         switch (sa->state) {
 190         case SFC_ADAPTER_CONFIGURED:
 191                 sfc_close(sa);
 192                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 193                 /* FALLTHROUGH */
 194         case SFC_ADAPTER_INITIALIZED:
 195                 rc = sfc_configure(sa);
 196                 break;
 197         default:
 198                 sfc_err(sa, "unexpected adapter state %u to configure",
 199                         sa->state);
 200                 rc = EINVAL;
 201                 break;
 202         }
 203         sfc_adapter_unlock(sa);
 204
 205         sfc_log_init(sa, "done %d", rc);
 206         SFC_ASSERT(rc >= 0);
 207         return -rc;
 208 }
 209
 210 static int
 211 sfc_dev_start(struct rte_eth_dev *dev)
 212 {
 213         struct sfc_adapter *sa = dev->data->dev_private;
 214         int rc;
 215
 216         sfc_log_init(sa, "entry");
 217
 218         sfc_adapter_lock(sa);
 219         rc = sfc_start(sa);
 220         sfc_adapter_unlock(sa);
 221
 222         sfc_log_init(sa, "done %d", rc);
 223         SFC_ASSERT(rc >= 0);
 224         return -rc;
 225 }
 226
 227 static int
 228 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 229 {
 230         struct sfc_adapter *sa = dev->data->dev_private;
 231         struct rte_eth_link *dev_link = &dev->data->dev_link;
 232         struct rte_eth_link old_link;
 233         struct rte_eth_link current_link;
 234
 235         sfc_log_init(sa, "entry");
 236
 237 retry:
 238         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 239         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 240
 241         if (sa->state != SFC_ADAPTER_STARTED) {
 242                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 243                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 244                                          *(uint64_t *)&old_link,
 245                                          *(uint64_t *)&current_link))
 246                         goto retry;
 247         } else if (wait_to_complete) {
 248                 efx_link_mode_t link_mode;
 249
 250                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 251                         link_mode = EFX_LINK_UNKNOWN;
 252                 sfc_port_link_mode_to_info(link_mode, &current_link);
 253
 254                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 255                                          *(uint64_t *)&old_link,
 256                                          *(uint64_t *)&current_link))
 257                         goto retry;
 258         } else {
 259                 sfc_ev_mgmt_qpoll(sa);
 260                 *(int64_t *)&current_link =
 261                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 262         }
 263
 264         if (old_link.link_status != current_link.link_status)
 265                 sfc_info(sa, "Link status is %s",
 266                          current_link.link_status ? "UP" : "DOWN");
 267
 268         return old_link.link_status == current_link.link_status ? 0 : -1;
 269 }
 270
 271 static void
 272 sfc_dev_stop(struct rte_eth_dev *dev)
 273 {
 274         struct sfc_adapter *sa = dev->data->dev_private;
 275
 276         sfc_log_init(sa, "entry");
 277
 278         sfc_adapter_lock(sa);
 279         sfc_stop(sa);
 280         sfc_adapter_unlock(sa);
 281
 282         sfc_log_init(sa, "done");
 283 }
 284
 285 static int
 286 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 287 {
 288         struct sfc_adapter *sa = dev->data->dev_private;
 289         int rc;
 290
 291         sfc_log_init(sa, "entry");
 292
 293         sfc_adapter_lock(sa);
 294         rc = sfc_start(sa);
 295         sfc_adapter_unlock(sa);
 296
 297         SFC_ASSERT(rc >= 0);
 298         return -rc;
 299 }
 300
 301 static int
 302 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 303 {
 304         struct sfc_adapter *sa = dev->data->dev_private;
 305
 306         sfc_log_init(sa, "entry");
 307
 308         sfc_adapter_lock(sa);
 309         sfc_stop(sa);
 310         sfc_adapter_unlock(sa);
 311
 312         return 0;
 313 }
 314
 315 static void
 316 sfc_dev_close(struct rte_eth_dev *dev)
 317 {
 318         struct sfc_adapter *sa = dev->data->dev_private;
 319
 320         sfc_log_init(sa, "entry");
 321
 322         sfc_adapter_lock(sa);
 323         switch (sa->state) {
 324         case SFC_ADAPTER_STARTED:
 325                 sfc_stop(sa);
 326                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 327                 /* FALLTHROUGH */
 328         case SFC_ADAPTER_CONFIGURED:
 329                 sfc_close(sa);
 330                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 331                 /* FALLTHROUGH */
 332         case SFC_ADAPTER_INITIALIZED:
 333                 break;
 334         default:
 335                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 336                 break;
 337         }
 338         sfc_adapter_unlock(sa);
 339
 340         sfc_log_init(sa, "done");
 341 }
 342
 343 static void
 344 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 345                    boolean_t enabled)
 346 {
 347         struct sfc_port *port;
 348         boolean_t *toggle;
 349         struct sfc_adapter *sa = dev->data->dev_private;
 350         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 351         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 352
 353         sfc_adapter_lock(sa);
 354
 355         port = &sa->port;
 356         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 357
 358         if (*toggle != enabled) {
 359                 *toggle = enabled;
 360
 361                 if ((sa->state == SFC_ADAPTER_STARTED) &&
 362                     (sfc_set_rx_mode(sa) != 0)) {
 363                         *toggle = !(enabled);
 364                         sfc_warn(sa, "Failed to %s %s mode",
 365                                  ((enabled) ? "enable" : "disable"), desc);
 366                 }
 367         }
 368
 369         sfc_adapter_unlock(sa);
 370 }
 371
 372 static void
 373 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 374 {
 375         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 376 }
 377
 378 static void
 379 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 380 {
 381         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 382 }
 383
 384 static void
 385 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 386 {
 387         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 388 }
 389
 390 static void
 391 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 392 {
 393         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 394 }
 395
 396 static int
 397 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 398                    uint16_t nb_rx_desc, unsigned int socket_id,
 399                    const struct rte_eth_rxconf *rx_conf,
 400                    struct rte_mempool *mb_pool)
 401 {
 402         struct sfc_adapter *sa = dev->data->dev_private;
 403         int rc;
 404
 405         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 406                      rx_queue_id, nb_rx_desc, socket_id);
 407
 408         sfc_adapter_lock(sa);
 409
 410         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 411                           rx_conf, mb_pool);
 412         if (rc != 0)
 413                 goto fail_rx_qinit;
 414
 415         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 416
 417         sfc_adapter_unlock(sa);
 418
 419         return 0;
 420
 421 fail_rx_qinit:
 422         sfc_adapter_unlock(sa);
 423         SFC_ASSERT(rc > 0);
 424         return -rc;
 425 }
 426
 427 static void
 428 sfc_rx_queue_release(void *queue)
 429 {
 430         struct sfc_dp_rxq *dp_rxq = queue;
 431         struct sfc_rxq *rxq;
 432         struct sfc_adapter *sa;
 433         unsigned int sw_index;
 434
 435         if (dp_rxq == NULL)
 436                 return;
 437
 438         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 439         sa = rxq->evq->sa;
 440         sfc_adapter_lock(sa);
 441
 442         sw_index = sfc_rxq_sw_index(rxq);
 443
 444         sfc_log_init(sa, "RxQ=%u", sw_index);
 445
 446         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 447
 448         sfc_rx_qfini(sa, sw_index);
 449
 450         sfc_adapter_unlock(sa);
 451 }
 452
 453 static int
 454 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 455                    uint16_t nb_tx_desc, unsigned int socket_id,
 456                    const struct rte_eth_txconf *tx_conf)
 457 {
 458         struct sfc_adapter *sa = dev->data->dev_private;
 459         int rc;
 460
 461         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 462                      tx_queue_id, nb_tx_desc, socket_id);
 463
 464         sfc_adapter_lock(sa);
 465
 466         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 467         if (rc != 0)
 468                 goto fail_tx_qinit;
 469
 470         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq;
 471
 472         sfc_adapter_unlock(sa);
 473         return 0;
 474
 475 fail_tx_qinit:
 476         sfc_adapter_unlock(sa);
 477         SFC_ASSERT(rc > 0);
 478         return -rc;
 479 }
 480
 481 static void
 482 sfc_tx_queue_release(void *queue)
 483 {
 484         struct sfc_txq *txq = queue;
 485         unsigned int sw_index;
 486         struct sfc_adapter *sa;
 487
 488         if (txq == NULL)
 489                 return;
 490
 491         sw_index = sfc_txq_sw_index(txq);
 492
 493         SFC_ASSERT(txq->evq != NULL);
 494         sa = txq->evq->sa;
 495
 496         sfc_log_init(sa, "TxQ = %u", sw_index);
 497
 498         sfc_adapter_lock(sa);
 499
 500         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 501         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 502
 503         sfc_tx_qfini(sa, sw_index);
 504
 505         sfc_adapter_unlock(sa);
 506 }
 507
 508 static void
 509 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 510 {
 511         struct sfc_adapter *sa = dev->data->dev_private;
 512         struct sfc_port *port = &sa->port;
 513         uint64_t *mac_stats;
 514
 515         rte_spinlock_lock(&port->mac_stats_lock);
 516
 517         if (sfc_port_update_mac_stats(sa) != 0)
 518                 goto unlock;
 519
 520         mac_stats = port->mac_stats_buf;
 521
 522         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 523                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 524                 stats->ipackets =
 525                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 526                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 527                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 528                 stats->opackets =
 529                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 530                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 531                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 532                 stats->ibytes =
 533                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 534                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 535                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 536                 stats->obytes =
 537                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 538                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 539                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 540                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 541                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 542                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 543         } else {
 544                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 545                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 546                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 547                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 548                 /*
 549                  * Take into account stats which are whenever supported
 550                  * on EF10. If some stat is not supported by current
 551                  * firmware variant or HW revision, it is guaranteed
 552                  * to be zero in mac_stats.
 553                  */
 554                 stats->imissed =
 555                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 556                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 557                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 558                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 559                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 560                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 561                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 562                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 563                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 564                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 565                 stats->ierrors =
 566                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 567                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 568                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 569                 /* no oerrors counters supported on EF10 */
 570         }
 571
 572 unlock:
 573         rte_spinlock_unlock(&port->mac_stats_lock);
 574 }
 575
 576 static void
 577 sfc_stats_reset(struct rte_eth_dev *dev)
 578 {
 579         struct sfc_adapter *sa = dev->data->dev_private;
 580         struct sfc_port *port = &sa->port;
 581         int rc;
 582
 583         if (sa->state != SFC_ADAPTER_STARTED) {
 584                 /*
 585                  * The operation cannot be done if port is not started; it
 586                  * will be scheduled to be done during the next port start
 587                  */
 588                 port->mac_stats_reset_pending = B_TRUE;
 589                 return;
 590         }
 591
 592         rc = sfc_port_reset_mac_stats(sa);
 593         if (rc != 0)
 594                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 595 }
 596
 597 static int
 598 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 599                unsigned int xstats_count)
 600 {
 601         struct sfc_adapter *sa = dev->data->dev_private;
 602         struct sfc_port *port = &sa->port;
 603         uint64_t *mac_stats;
 604         int rc;
 605         unsigned int i;
 606         int nstats = 0;
 607
 608         rte_spinlock_lock(&port->mac_stats_lock);
 609
 610         rc = sfc_port_update_mac_stats(sa);
 611         if (rc != 0) {
 612                 SFC_ASSERT(rc > 0);
 613                 nstats = -rc;
 614                 goto unlock;
 615         }
 616
 617         mac_stats = port->mac_stats_buf;
 618
 619         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 620                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 621                         if (xstats != NULL && nstats < (int)xstats_count) {
 622                                 xstats[nstats].id = nstats;
 623                                 xstats[nstats].value = mac_stats[i];
 624                         }
 625                         nstats++;
 626                 }
 627         }
 628
 629 unlock:
 630         rte_spinlock_unlock(&port->mac_stats_lock);
 631
 632         return nstats;
 633 }
 634
 635 static int
 636 sfc_xstats_get_names(struct rte_eth_dev *dev,
 637                      struct rte_eth_xstat_name *xstats_names,
 638                      unsigned int xstats_count)
 639 {
 640         struct sfc_adapter *sa = dev->data->dev_private;
 641         struct sfc_port *port = &sa->port;
 642         unsigned int i;
 643         unsigned int nstats = 0;
 644
 645         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 646                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 647                         if (xstats_names != NULL && nstats < xstats_count)
 648                                 strncpy(xstats_names[nstats].name,
 649                                         efx_mac_stat_name(sa->nic, i),
 650                                         sizeof(xstats_names[0].name));
 651                         nstats++;
 652                 }
 653         }
 654
 655         return nstats;
 656 }
 657
 658 static int
 659 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 660 {
 661         struct sfc_adapter *sa = dev->data->dev_private;
 662         unsigned int wanted_fc, link_fc;
 663
 664         memset(fc_conf, 0, sizeof(*fc_conf));
 665
 666         sfc_adapter_lock(sa);
 667
 668         if (sa->state == SFC_ADAPTER_STARTED)
 669                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 670         else
 671                 link_fc = sa->port.flow_ctrl;
 672
 673         switch (link_fc) {
 674         case 0:
 675                 fc_conf->mode = RTE_FC_NONE;
 676                 break;
 677         case EFX_FCNTL_RESPOND:
 678                 fc_conf->mode = RTE_FC_RX_PAUSE;
 679                 break;
 680         case EFX_FCNTL_GENERATE:
 681                 fc_conf->mode = RTE_FC_TX_PAUSE;
 682                 break;
 683         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 684                 fc_conf->mode = RTE_FC_FULL;
 685                 break;
 686         default:
 687                 sfc_err(sa, "%s: unexpected flow control value %#x",
 688                         __func__, link_fc);
 689         }
 690
 691         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 692
 693         sfc_adapter_unlock(sa);
 694
 695         return 0;
 696 }
 697
 698 static int
 699 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 700 {
 701         struct sfc_adapter *sa = dev->data->dev_private;
 702         struct sfc_port *port = &sa->port;
 703         unsigned int fcntl;
 704         int rc;
 705
 706         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 707             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 708             fc_conf->mac_ctrl_frame_fwd != 0) {
 709                 sfc_err(sa, "unsupported flow control settings specified");
 710                 rc = EINVAL;
 711                 goto fail_inval;
 712         }
 713
 714         switch (fc_conf->mode) {
 715         case RTE_FC_NONE:
 716                 fcntl = 0;
 717                 break;
 718         case RTE_FC_RX_PAUSE:
 719                 fcntl = EFX_FCNTL_RESPOND;
 720                 break;
 721         case RTE_FC_TX_PAUSE:
 722                 fcntl = EFX_FCNTL_GENERATE;
 723                 break;
 724         case RTE_FC_FULL:
 725                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 726                 break;
 727         default:
 728                 rc = EINVAL;
 729                 goto fail_inval;
 730         }
 731
 732         sfc_adapter_lock(sa);
 733
 734         if (sa->state == SFC_ADAPTER_STARTED) {
 735                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 736                 if (rc != 0)
 737                         goto fail_mac_fcntl_set;
 738         }
 739
 740         port->flow_ctrl = fcntl;
 741         port->flow_ctrl_autoneg = fc_conf->autoneg;
 742
 743         sfc_adapter_unlock(sa);
 744
 745         return 0;
 746
 747 fail_mac_fcntl_set:
 748         sfc_adapter_unlock(sa);
 749 fail_inval:
 750         SFC_ASSERT(rc > 0);
 751         return -rc;
 752 }
 753
 754 static int
 755 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 756 {
 757         struct sfc_adapter *sa = dev->data->dev_private;
 758         size_t pdu = EFX_MAC_PDU(mtu);
 759         size_t old_pdu;
 760         int rc;
 761
 762         sfc_log_init(sa, "mtu=%u", mtu);
 763
 764         rc = EINVAL;
 765         if (pdu < EFX_MAC_PDU_MIN) {
 766                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 767                         (unsigned int)mtu, (unsigned int)pdu,
 768                         EFX_MAC_PDU_MIN);
 769                 goto fail_inval;
 770         }
 771         if (pdu > EFX_MAC_PDU_MAX) {
 772                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 773                         (unsigned int)mtu, (unsigned int)pdu,
 774                         EFX_MAC_PDU_MAX);
 775                 goto fail_inval;
 776         }
 777
 778         sfc_adapter_lock(sa);
 779
 780         if (pdu != sa->port.pdu) {
 781                 if (sa->state == SFC_ADAPTER_STARTED) {
 782                         sfc_stop(sa);
 783
 784                         old_pdu = sa->port.pdu;
 785                         sa->port.pdu = pdu;
 786                         rc = sfc_start(sa);
 787                         if (rc != 0)
 788                                 goto fail_start;
 789                 } else {
 790                         sa->port.pdu = pdu;
 791                 }
 792         }
 793
 794         /*
 795          * The driver does not use it, but other PMDs update jumbo_frame
 796          * flag and max_rx_pkt_len when MTU is set.
 797          */
 798         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 799         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 800
 801         sfc_adapter_unlock(sa);
 802
 803         sfc_log_init(sa, "done");
 804         return 0;
 805
 806 fail_start:
 807         sa->port.pdu = old_pdu;
 808         if (sfc_start(sa) != 0)
 809                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 810                         "PDU max size - port is stopped",
 811                         (unsigned int)pdu, (unsigned int)old_pdu);
 812         sfc_adapter_unlock(sa);
 813
 814 fail_inval:
 815         sfc_log_init(sa, "failed %d", rc);
 816         SFC_ASSERT(rc > 0);
 817         return -rc;
 818 }
 819 static void
 820 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 821 {
 822         struct sfc_adapter *sa = dev->data->dev_private;
 823         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 824         int rc;
 825
 826         sfc_adapter_lock(sa);
 827
 828         if (sa->state != SFC_ADAPTER_STARTED) {
 829                 sfc_info(sa, "the port is not started");
 830                 sfc_info(sa, "the new MAC address will be set on port start");
 831
 832                 goto unlock;
 833         }
 834
 835         if (encp->enc_allow_set_mac_with_installed_filters) {
 836                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 837                 if (rc != 0) {
 838                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 839                         goto unlock;
 840                 }
 841
 842                 /*
 843                  * Changing the MAC address by means of MCDI request
 844                  * has no effect on received traffic, therefore
 845                  * we also need to update unicast filters
 846                  */
 847                 rc = sfc_set_rx_mode(sa);
 848                 if (rc != 0)
 849                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 850         } else {
 851                 sfc_warn(sa, "cannot set MAC address with filters installed");
 852                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 853                 sfc_warn(sa, "(some traffic may be dropped)");
 854
 855                 /*
 856                  * Since setting MAC address with filters installed is not
 857                  * allowed on the adapter, one needs to simply restart adapter
 858                  * so that the new MAC address will be taken from an outer
 859                  * storage and set flawlessly by means of sfc_start() call
 860                  */
 861                 sfc_stop(sa);
 862                 rc = sfc_start(sa);
 863                 if (rc != 0)
 864                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 865         }
 866
 867 unlock:
 868         sfc_adapter_unlock(sa);
 869 }
 870
 871
 872 static int
 873 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 874                      uint32_t nb_mc_addr)
 875 {
 876         struct sfc_adapter *sa = dev->data->dev_private;
 877         struct sfc_port *port = &sa->port;
 878         uint8_t *mc_addrs = port->mcast_addrs;
 879         int rc;
 880         unsigned int i;
 881
 882         if (mc_addrs == NULL)
 883                 return -ENOBUFS;
 884
 885         if (nb_mc_addr > port->max_mcast_addrs) {
 886                 sfc_err(sa, "too many multicast addresses: %u > %u",
 887                          nb_mc_addr, port->max_mcast_addrs);
 888                 return -EINVAL;
 889         }
 890
 891         for (i = 0; i < nb_mc_addr; ++i) {
 892                 (void)rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
 893                                  EFX_MAC_ADDR_LEN);
 894                 mc_addrs += EFX_MAC_ADDR_LEN;
 895         }
 896
 897         port->nb_mcast_addrs = nb_mc_addr;
 898
 899         if (sa->state != SFC_ADAPTER_STARTED)
 900                 return 0;
 901
 902         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
 903                                         port->nb_mcast_addrs);
 904         if (rc != 0)
 905                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
 906
 907         SFC_ASSERT(rc > 0);
 908         return -rc;
 909 }
 910
 911 static void
 912 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 913                       struct rte_eth_rxq_info *qinfo)
 914 {
 915         struct sfc_adapter *sa = dev->data->dev_private;
 916         struct sfc_rxq_info *rxq_info;
 917         struct sfc_rxq *rxq;
 918
 919         sfc_adapter_lock(sa);
 920
 921         SFC_ASSERT(rx_queue_id < sa->rxq_count);
 922
 923         rxq_info = &sa->rxq_info[rx_queue_id];
 924         rxq = rxq_info->rxq;
 925         SFC_ASSERT(rxq != NULL);
 926
 927         qinfo->mp = rxq->refill_mb_pool;
 928         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
 929         qinfo->conf.rx_drop_en = 1;
 930         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
 931         qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
 932         qinfo->nb_desc = rxq_info->entries;
 933
 934         sfc_adapter_unlock(sa);
 935 }
 936
 937 static void
 938 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 939                       struct rte_eth_txq_info *qinfo)
 940 {
 941         struct sfc_adapter *sa = dev->data->dev_private;
 942         struct sfc_txq_info *txq_info;
 943
 944         sfc_adapter_lock(sa);
 945
 946         SFC_ASSERT(tx_queue_id < sa->txq_count);
 947
 948         txq_info = &sa->txq_info[tx_queue_id];
 949         SFC_ASSERT(txq_info->txq != NULL);
 950
 951         memset(qinfo, 0, sizeof(*qinfo));
 952
 953         qinfo->conf.txq_flags = txq_info->txq->flags;
 954         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
 955         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
 956         qinfo->nb_desc = txq_info->entries;
 957
 958         sfc_adapter_unlock(sa);
 959 }
 960
 961 static uint32_t
 962 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 963 {
 964         struct sfc_adapter *sa = dev->data->dev_private;
 965
 966         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 967
 968         return sfc_rx_qdesc_npending(sa, rx_queue_id);
 969 }
 970
 971 static int
 972 sfc_rx_descriptor_done(void *queue, uint16_t offset)
 973 {
 974         struct sfc_dp_rxq *dp_rxq = queue;
 975
 976         return sfc_rx_qdesc_done(dp_rxq, offset);
 977 }
 978
 979 static int
 980 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 981 {
 982         struct sfc_adapter *sa = dev->data->dev_private;
 983         int rc;
 984
 985         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 986
 987         sfc_adapter_lock(sa);
 988
 989         rc = EINVAL;
 990         if (sa->state != SFC_ADAPTER_STARTED)
 991                 goto fail_not_started;
 992
 993         rc = sfc_rx_qstart(sa, rx_queue_id);
 994         if (rc != 0)
 995                 goto fail_rx_qstart;
 996
 997         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
 998
 999         sfc_adapter_unlock(sa);
1000
1001         return 0;
1002
1003 fail_rx_qstart:
1004 fail_not_started:
1005         sfc_adapter_unlock(sa);
1006         SFC_ASSERT(rc > 0);
1007         return -rc;
1008 }
1009
1010 static int
1011 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1012 {
1013         struct sfc_adapter *sa = dev->data->dev_private;
1014
1015         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1016
1017         sfc_adapter_lock(sa);
1018         sfc_rx_qstop(sa, rx_queue_id);
1019
1020         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1021
1022         sfc_adapter_unlock(sa);
1023
1024         return 0;
1025 }
1026
1027 static int
1028 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1029 {
1030         struct sfc_adapter *sa = dev->data->dev_private;
1031         int rc;
1032
1033         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1034
1035         sfc_adapter_lock(sa);
1036
1037         rc = EINVAL;
1038         if (sa->state != SFC_ADAPTER_STARTED)
1039                 goto fail_not_started;
1040
1041         rc = sfc_tx_qstart(sa, tx_queue_id);
1042         if (rc != 0)
1043                 goto fail_tx_qstart;
1044
1045         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1046
1047         sfc_adapter_unlock(sa);
1048         return 0;
1049
1050 fail_tx_qstart:
1051
1052 fail_not_started:
1053         sfc_adapter_unlock(sa);
1054         SFC_ASSERT(rc > 0);
1055         return -rc;
1056 }
1057
1058 static int
1059 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1060 {
1061         struct sfc_adapter *sa = dev->data->dev_private;
1062
1063         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1064
1065         sfc_adapter_lock(sa);
1066
1067         sfc_tx_qstop(sa, tx_queue_id);
1068
1069         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1070
1071         sfc_adapter_unlock(sa);
1072         return 0;
1073 }
1074
1075 #if EFSYS_OPT_RX_SCALE
1076 static int
1077 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1078                           struct rte_eth_rss_conf *rss_conf)
1079 {
1080         struct sfc_adapter *sa = dev->data->dev_private;
1081
1082         if ((sa->rss_channels == 1) ||
1083             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1084                 return -ENOTSUP;
1085
1086         sfc_adapter_lock(sa);
1087
1088         /*
1089          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1090          * hence, conversion is done here to derive a correct set of ETH_RSS
1091          * flags which corresponds to the active EFX configuration stored
1092          * locally in 'sfc_adapter' and kept up-to-date
1093          */
1094         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1095         rss_conf->rss_key_len = SFC_RSS_KEY_SIZE;
1096         if (rss_conf->rss_key != NULL)
1097                 rte_memcpy(rss_conf->rss_key, sa->rss_key, SFC_RSS_KEY_SIZE);
1098
1099         sfc_adapter_unlock(sa);
1100
1101         return 0;
1102 }
1103
1104 static int
1105 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1106                         struct rte_eth_rss_conf *rss_conf)
1107 {
1108         struct sfc_adapter *sa = dev->data->dev_private;
1109         unsigned int efx_hash_types;
1110         int rc = 0;
1111
1112         if ((sa->rss_channels == 1) ||
1113             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1114                 sfc_err(sa, "RSS is not available");
1115                 return -ENOTSUP;
1116         }
1117
1118         if ((rss_conf->rss_key != NULL) &&
1119             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1120                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1121                         sizeof(sa->rss_key));
1122                 return -EINVAL;
1123         }
1124
1125         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1126                 sfc_err(sa, "unsupported hash functions requested");
1127                 return -EINVAL;
1128         }
1129
1130         sfc_adapter_lock(sa);
1131
1132         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1133
1134         rc = efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1135                                    efx_hash_types, B_TRUE);
1136         if (rc != 0)
1137                 goto fail_scale_mode_set;
1138
1139         if (rss_conf->rss_key != NULL) {
1140                 if (sa->state == SFC_ADAPTER_STARTED) {
1141                         rc = efx_rx_scale_key_set(sa->nic, rss_conf->rss_key,
1142                                                   sizeof(sa->rss_key));
1143                         if (rc != 0)
1144                                 goto fail_scale_key_set;
1145                 }
1146
1147                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1148         }
1149
1150         sa->rss_hash_types = efx_hash_types;
1151
1152         sfc_adapter_unlock(sa);
1153
1154         return 0;
1155
1156 fail_scale_key_set:
1157         if (efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1158                                   sa->rss_hash_types, B_TRUE) != 0)
1159                 sfc_err(sa, "failed to restore RSS mode");
1160
1161 fail_scale_mode_set:
1162         sfc_adapter_unlock(sa);
1163         return -rc;
1164 }
1165
1166 static int
1167 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1168                        struct rte_eth_rss_reta_entry64 *reta_conf,
1169                        uint16_t reta_size)
1170 {
1171         struct sfc_adapter *sa = dev->data->dev_private;
1172         int entry;
1173
1174         if ((sa->rss_channels == 1) ||
1175             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1176                 return -ENOTSUP;
1177
1178         if (reta_size != EFX_RSS_TBL_SIZE)
1179                 return -EINVAL;
1180
1181         sfc_adapter_lock(sa);
1182
1183         for (entry = 0; entry < reta_size; entry++) {
1184                 int grp = entry / RTE_RETA_GROUP_SIZE;
1185                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1186
1187                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1188                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1189         }
1190
1191         sfc_adapter_unlock(sa);
1192
1193         return 0;
1194 }
1195
1196 static int
1197 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1198                         struct rte_eth_rss_reta_entry64 *reta_conf,
1199                         uint16_t reta_size)
1200 {
1201         struct sfc_adapter *sa = dev->data->dev_private;
1202         unsigned int *rss_tbl_new;
1203         uint16_t entry;
1204         int rc;
1205
1206
1207         if ((sa->rss_channels == 1) ||
1208             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1209                 sfc_err(sa, "RSS is not available");
1210                 return -ENOTSUP;
1211         }
1212
1213         if (reta_size != EFX_RSS_TBL_SIZE) {
1214                 sfc_err(sa, "RETA size is wrong (should be %u)",
1215                         EFX_RSS_TBL_SIZE);
1216                 return -EINVAL;
1217         }
1218
1219         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1220         if (rss_tbl_new == NULL)
1221                 return -ENOMEM;
1222
1223         sfc_adapter_lock(sa);
1224
1225         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1226
1227         for (entry = 0; entry < reta_size; entry++) {
1228                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1229                 struct rte_eth_rss_reta_entry64 *grp;
1230
1231                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1232
1233                 if (grp->mask & (1ull << grp_idx)) {
1234                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1235                                 rc = EINVAL;
1236                                 goto bad_reta_entry;
1237                         }
1238                         rss_tbl_new[entry] = grp->reta[grp_idx];
1239                 }
1240         }
1241
1242         rc = efx_rx_scale_tbl_set(sa->nic, rss_tbl_new, EFX_RSS_TBL_SIZE);
1243         if (rc == 0)
1244                 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1245
1246 bad_reta_entry:
1247         sfc_adapter_unlock(sa);
1248
1249         rte_free(rss_tbl_new);
1250
1251         SFC_ASSERT(rc >= 0);
1252         return -rc;
1253 }
1254 #endif
1255
1256 static int
1257 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1258                     enum rte_filter_op filter_op,
1259                     void *arg)
1260 {
1261         struct sfc_adapter *sa = dev->data->dev_private;
1262         int rc = ENOTSUP;
1263
1264         sfc_log_init(sa, "entry");
1265
1266         switch (filter_type) {
1267         case RTE_ETH_FILTER_NONE:
1268                 sfc_err(sa, "Global filters configuration not supported");
1269                 break;
1270         case RTE_ETH_FILTER_MACVLAN:
1271                 sfc_err(sa, "MACVLAN filters not supported");
1272                 break;
1273         case RTE_ETH_FILTER_ETHERTYPE:
1274                 sfc_err(sa, "EtherType filters not supported");
1275                 break;
1276         case RTE_ETH_FILTER_FLEXIBLE:
1277                 sfc_err(sa, "Flexible filters not supported");
1278                 break;
1279         case RTE_ETH_FILTER_SYN:
1280                 sfc_err(sa, "SYN filters not supported");
1281                 break;
1282         case RTE_ETH_FILTER_NTUPLE:
1283                 sfc_err(sa, "NTUPLE filters not supported");
1284                 break;
1285         case RTE_ETH_FILTER_TUNNEL:
1286                 sfc_err(sa, "Tunnel filters not supported");
1287                 break;
1288         case RTE_ETH_FILTER_FDIR:
1289                 sfc_err(sa, "Flow Director filters not supported");
1290                 break;
1291         case RTE_ETH_FILTER_HASH:
1292                 sfc_err(sa, "Hash filters not supported");
1293                 break;
1294         case RTE_ETH_FILTER_GENERIC:
1295                 if (filter_op != RTE_ETH_FILTER_GET) {
1296                         rc = EINVAL;
1297                 } else {
1298                         *(const void **)arg = &sfc_flow_ops;
1299                         rc = 0;
1300                 }
1301                 break;
1302         default:
1303                 sfc_err(sa, "Unknown filter type %u", filter_type);
1304                 break;
1305         }
1306
1307         sfc_log_init(sa, "exit: %d", -rc);
1308         SFC_ASSERT(rc >= 0);
1309         return -rc;
1310 }
1311
1312 static const struct eth_dev_ops sfc_eth_dev_ops = {
1313         .dev_configure                  = sfc_dev_configure,
1314         .dev_start                      = sfc_dev_start,
1315         .dev_stop                       = sfc_dev_stop,
1316         .dev_set_link_up                = sfc_dev_set_link_up,
1317         .dev_set_link_down              = sfc_dev_set_link_down,
1318         .dev_close                      = sfc_dev_close,
1319         .promiscuous_enable             = sfc_dev_promisc_enable,
1320         .promiscuous_disable            = sfc_dev_promisc_disable,
1321         .allmulticast_enable            = sfc_dev_allmulti_enable,
1322         .allmulticast_disable           = sfc_dev_allmulti_disable,
1323         .link_update                    = sfc_dev_link_update,
1324         .stats_get                      = sfc_stats_get,
1325         .stats_reset                    = sfc_stats_reset,
1326         .xstats_get                     = sfc_xstats_get,
1327         .xstats_reset                   = sfc_stats_reset,
1328         .xstats_get_names               = sfc_xstats_get_names,
1329         .dev_infos_get                  = sfc_dev_infos_get,
1330         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1331         .mtu_set                        = sfc_dev_set_mtu,
1332         .rx_queue_start                 = sfc_rx_queue_start,
1333         .rx_queue_stop                  = sfc_rx_queue_stop,
1334         .tx_queue_start                 = sfc_tx_queue_start,
1335         .tx_queue_stop                  = sfc_tx_queue_stop,
1336         .rx_queue_setup                 = sfc_rx_queue_setup,
1337         .rx_queue_release               = sfc_rx_queue_release,
1338         .rx_queue_count                 = sfc_rx_queue_count,
1339         .rx_descriptor_done             = sfc_rx_descriptor_done,
1340         .tx_queue_setup                 = sfc_tx_queue_setup,
1341         .tx_queue_release               = sfc_tx_queue_release,
1342         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1343         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1344         .mac_addr_set                   = sfc_mac_addr_set,
1345 #if EFSYS_OPT_RX_SCALE
1346         .reta_update                    = sfc_dev_rss_reta_update,
1347         .reta_query                     = sfc_dev_rss_reta_query,
1348         .rss_hash_update                = sfc_dev_rss_hash_update,
1349         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1350 #endif
1351         .filter_ctrl                    = sfc_dev_filter_ctrl,
1352         .set_mc_addr_list               = sfc_set_mc_addr_list,
1353         .rxq_info_get                   = sfc_rx_queue_info_get,
1354         .txq_info_get                   = sfc_tx_queue_info_get,
1355         .fw_version_get                 = sfc_fw_version_get,
1356 };
1357
1358 static int
1359 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1360 {
1361         struct sfc_adapter *sa = dev->data->dev_private;
1362         unsigned int avail_caps = 0;
1363         const char *rx_name = NULL;
1364         int rc;
1365
1366         if (sa == NULL || sa->state == SFC_ADAPTER_UNINITIALIZED)
1367                 return -E_RTE_SECONDARY;
1368
1369         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1370                                 sfc_kvarg_string_handler, &rx_name);
1371         if (rc != 0)
1372                 goto fail_kvarg_rx_datapath;
1373
1374         if (rx_name != NULL) {
1375                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1376                 if (sa->dp_rx == NULL) {
1377                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1378                         rc = ENOENT;
1379                         goto fail_dp_rx;
1380                 }
1381                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1382                         sfc_err(sa,
1383                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1384                                 rx_name);
1385                         rc = EINVAL;
1386                         goto fail_dp_rx;
1387                 }
1388         } else {
1389                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1390                 if (sa->dp_rx == NULL) {
1391                         sfc_err(sa, "Rx datapath by caps %#x not found",
1392                                 avail_caps);
1393                         rc = ENOENT;
1394                         goto fail_dp_rx;
1395                 }
1396         }
1397
1398         sfc_info(sa, "use %s Rx datapath", sa->dp_rx->dp.name);
1399
1400         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1401
1402         dev->tx_pkt_burst = sfc_xmit_pkts;
1403
1404         dev->dev_ops = &sfc_eth_dev_ops;
1405
1406         return 0;
1407
1408 fail_dp_rx:
1409 fail_kvarg_rx_datapath:
1410         return rc;
1411 }
1412
1413 static void
1414 sfc_register_dp(void)
1415 {
1416         /* Register once */
1417         if (TAILQ_EMPTY(&sfc_dp_head))
1418                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1419 }
1420
1421 static int
1422 sfc_eth_dev_init(struct rte_eth_dev *dev)
1423 {
1424         struct sfc_adapter *sa = dev->data->dev_private;
1425         struct rte_pci_device *pci_dev = SFC_DEV_TO_PCI(dev);
1426         int rc;
1427         const efx_nic_cfg_t *encp;
1428         const struct ether_addr *from;
1429
1430         sfc_register_dp();
1431
1432         /* Required for logging */
1433         sa->eth_dev = dev;
1434
1435         /* Copy PCI device info to the dev->data */
1436         rte_eth_copy_pci_info(dev, pci_dev);
1437
1438         rc = sfc_kvargs_parse(sa);
1439         if (rc != 0)
1440                 goto fail_kvargs_parse;
1441
1442         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1443                                 sfc_kvarg_bool_handler, &sa->debug_init);
1444         if (rc != 0)
1445                 goto fail_kvarg_debug_init;
1446
1447         sfc_log_init(sa, "entry");
1448
1449         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1450         if (dev->data->mac_addrs == NULL) {
1451                 rc = ENOMEM;
1452                 goto fail_mac_addrs;
1453         }
1454
1455         sfc_adapter_lock_init(sa);
1456         sfc_adapter_lock(sa);
1457
1458         sfc_log_init(sa, "attaching");
1459         rc = sfc_attach(sa);
1460         if (rc != 0)
1461                 goto fail_attach;
1462
1463         encp = efx_nic_cfg_get(sa->nic);
1464
1465         /*
1466          * The arguments are really reverse order in comparison to
1467          * Linux kernel. Copy from NIC config to Ethernet device data.
1468          */
1469         from = (const struct ether_addr *)(encp->enc_mac_addr);
1470         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1471
1472         sfc_adapter_unlock(sa);
1473
1474         sfc_eth_dev_set_ops(dev);
1475
1476         sfc_log_init(sa, "done");
1477         return 0;
1478
1479 fail_attach:
1480         sfc_adapter_unlock(sa);
1481         sfc_adapter_lock_fini(sa);
1482         rte_free(dev->data->mac_addrs);
1483         dev->data->mac_addrs = NULL;
1484
1485 fail_mac_addrs:
1486 fail_kvarg_debug_init:
1487         sfc_kvargs_cleanup(sa);
1488
1489 fail_kvargs_parse:
1490         sfc_log_init(sa, "failed %d", rc);
1491         SFC_ASSERT(rc > 0);
1492         return -rc;
1493 }
1494
1495 static int
1496 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1497 {
1498         struct sfc_adapter *sa = dev->data->dev_private;
1499
1500         sfc_log_init(sa, "entry");
1501
1502         sfc_adapter_lock(sa);
1503
1504         sfc_detach(sa);
1505
1506         rte_free(dev->data->mac_addrs);
1507         dev->data->mac_addrs = NULL;
1508
1509         dev->dev_ops = NULL;
1510         dev->rx_pkt_burst = NULL;
1511         dev->tx_pkt_burst = NULL;
1512
1513         sfc_kvargs_cleanup(sa);
1514
1515         sfc_adapter_unlock(sa);
1516         sfc_adapter_lock_fini(sa);
1517
1518         sfc_log_init(sa, "done");
1519
1520         /* Required for logging, so cleanup last */
1521         sa->eth_dev = NULL;
1522         return 0;
1523 }
1524
1525 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1526         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1527         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1528         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1529         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1530         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1531         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1532         { .vendor_id = 0 /* sentinel */ }
1533 };
1534
1535 static struct eth_driver sfc_efx_pmd = {
1536         .pci_drv = {
1537                 .id_table = pci_id_sfc_efx_map,
1538                 .drv_flags =
1539                         RTE_PCI_DRV_INTR_LSC |
1540                         RTE_PCI_DRV_NEED_MAPPING,
1541                 .probe = rte_eth_dev_pci_probe,
1542                 .remove = rte_eth_dev_pci_remove,
1543         },
1544         .eth_dev_init = sfc_eth_dev_init,
1545         .eth_dev_uninit = sfc_eth_dev_uninit,
1546         .dev_private_size = sizeof(struct sfc_adapter),
1547 };
1548
1549 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd.pci_drv);
1550 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1551 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio");
1552 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1553         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1554         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1555         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1556         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1557         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);