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 ((~sa->dp_tx->features & SFC_DP_TX_FEAT_VLAN_INSERT) ||
 139             !encp->enc_hw_tx_insert_vlan_enabled)
 140                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOVLANOFFL;
 141         else
 142                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_VLAN_INSERT;
 143
 144         if (~sa->dp_tx->features & SFC_DP_TX_FEAT_MULTI_SEG)
 145                 dev_info->default_txconf.txq_flags |= ETH_TXQ_FLAGS_NOMULTSEGS;
 146
 147 #if EFSYS_OPT_RX_SCALE
 148         if (sa->rss_support != EFX_RX_SCALE_UNAVAILABLE) {
 149                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 150                 dev_info->hash_key_size = SFC_RSS_KEY_SIZE;
 151                 dev_info->flow_type_rss_offloads = SFC_RSS_OFFLOADS;
 152         }
 153 #endif
 154
 155         if (sa->tso)
 156                 dev_info->tx_offload_capa |= DEV_TX_OFFLOAD_TCP_TSO;
 157
 158         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 159         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 160         /* The RXQ hardware requires that the descriptor count is a power
 161          * of 2, but rx_desc_lim cannot properly describe that constraint.
 162          */
 163         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 164
 165         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 166         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 167         /*
 168          * The TXQ hardware requires that the descriptor count is a power
 169          * of 2, but tx_desc_lim cannot properly describe that constraint
 170          */
 171         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 172 }
 173
 174 static const uint32_t *
 175 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 176 {
 177         struct sfc_adapter *sa = dev->data->dev_private;
 178
 179         return sa->dp_rx->supported_ptypes_get();
 180 }
 181
 182 static int
 183 sfc_dev_configure(struct rte_eth_dev *dev)
 184 {
 185         struct rte_eth_dev_data *dev_data = dev->data;
 186         struct sfc_adapter *sa = dev_data->dev_private;
 187         int rc;
 188
 189         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 190                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 191
 192         sfc_adapter_lock(sa);
 193         switch (sa->state) {
 194         case SFC_ADAPTER_CONFIGURED:
 195                 sfc_close(sa);
 196                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 197                 /* FALLTHROUGH */
 198         case SFC_ADAPTER_INITIALIZED:
 199                 rc = sfc_configure(sa);
 200                 break;
 201         default:
 202                 sfc_err(sa, "unexpected adapter state %u to configure",
 203                         sa->state);
 204                 rc = EINVAL;
 205                 break;
 206         }
 207         sfc_adapter_unlock(sa);
 208
 209         sfc_log_init(sa, "done %d", rc);
 210         SFC_ASSERT(rc >= 0);
 211         return -rc;
 212 }
 213
 214 static int
 215 sfc_dev_start(struct rte_eth_dev *dev)
 216 {
 217         struct sfc_adapter *sa = dev->data->dev_private;
 218         int rc;
 219
 220         sfc_log_init(sa, "entry");
 221
 222         sfc_adapter_lock(sa);
 223         rc = sfc_start(sa);
 224         sfc_adapter_unlock(sa);
 225
 226         sfc_log_init(sa, "done %d", rc);
 227         SFC_ASSERT(rc >= 0);
 228         return -rc;
 229 }
 230
 231 static int
 232 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 233 {
 234         struct sfc_adapter *sa = dev->data->dev_private;
 235         struct rte_eth_link *dev_link = &dev->data->dev_link;
 236         struct rte_eth_link old_link;
 237         struct rte_eth_link current_link;
 238
 239         sfc_log_init(sa, "entry");
 240
 241 retry:
 242         EFX_STATIC_ASSERT(sizeof(*dev_link) == sizeof(rte_atomic64_t));
 243         *(int64_t *)&old_link = rte_atomic64_read((rte_atomic64_t *)dev_link);
 244
 245         if (sa->state != SFC_ADAPTER_STARTED) {
 246                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 247                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 248                                          *(uint64_t *)&old_link,
 249                                          *(uint64_t *)&current_link))
 250                         goto retry;
 251         } else if (wait_to_complete) {
 252                 efx_link_mode_t link_mode;
 253
 254                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 255                         link_mode = EFX_LINK_UNKNOWN;
 256                 sfc_port_link_mode_to_info(link_mode, &current_link);
 257
 258                 if (!rte_atomic64_cmpset((volatile uint64_t *)dev_link,
 259                                          *(uint64_t *)&old_link,
 260                                          *(uint64_t *)&current_link))
 261                         goto retry;
 262         } else {
 263                 sfc_ev_mgmt_qpoll(sa);
 264                 *(int64_t *)&current_link =
 265                         rte_atomic64_read((rte_atomic64_t *)dev_link);
 266         }
 267
 268         if (old_link.link_status != current_link.link_status)
 269                 sfc_info(sa, "Link status is %s",
 270                          current_link.link_status ? "UP" : "DOWN");
 271
 272         return old_link.link_status == current_link.link_status ? 0 : -1;
 273 }
 274
 275 static void
 276 sfc_dev_stop(struct rte_eth_dev *dev)
 277 {
 278         struct sfc_adapter *sa = dev->data->dev_private;
 279
 280         sfc_log_init(sa, "entry");
 281
 282         sfc_adapter_lock(sa);
 283         sfc_stop(sa);
 284         sfc_adapter_unlock(sa);
 285
 286         sfc_log_init(sa, "done");
 287 }
 288
 289 static int
 290 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 291 {
 292         struct sfc_adapter *sa = dev->data->dev_private;
 293         int rc;
 294
 295         sfc_log_init(sa, "entry");
 296
 297         sfc_adapter_lock(sa);
 298         rc = sfc_start(sa);
 299         sfc_adapter_unlock(sa);
 300
 301         SFC_ASSERT(rc >= 0);
 302         return -rc;
 303 }
 304
 305 static int
 306 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 307 {
 308         struct sfc_adapter *sa = dev->data->dev_private;
 309
 310         sfc_log_init(sa, "entry");
 311
 312         sfc_adapter_lock(sa);
 313         sfc_stop(sa);
 314         sfc_adapter_unlock(sa);
 315
 316         return 0;
 317 }
 318
 319 static void
 320 sfc_dev_close(struct rte_eth_dev *dev)
 321 {
 322         struct sfc_adapter *sa = dev->data->dev_private;
 323
 324         sfc_log_init(sa, "entry");
 325
 326         sfc_adapter_lock(sa);
 327         switch (sa->state) {
 328         case SFC_ADAPTER_STARTED:
 329                 sfc_stop(sa);
 330                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 331                 /* FALLTHROUGH */
 332         case SFC_ADAPTER_CONFIGURED:
 333                 sfc_close(sa);
 334                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 335                 /* FALLTHROUGH */
 336         case SFC_ADAPTER_INITIALIZED:
 337                 break;
 338         default:
 339                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 340                 break;
 341         }
 342         sfc_adapter_unlock(sa);
 343
 344         sfc_log_init(sa, "done");
 345 }
 346
 347 static void
 348 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 349                    boolean_t enabled)
 350 {
 351         struct sfc_port *port;
 352         boolean_t *toggle;
 353         struct sfc_adapter *sa = dev->data->dev_private;
 354         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 355         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 356
 357         sfc_adapter_lock(sa);
 358
 359         port = &sa->port;
 360         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 361
 362         if (*toggle != enabled) {
 363                 *toggle = enabled;
 364
 365                 if ((sa->state == SFC_ADAPTER_STARTED) &&
 366                     (sfc_set_rx_mode(sa) != 0)) {
 367                         *toggle = !(enabled);
 368                         sfc_warn(sa, "Failed to %s %s mode",
 369                                  ((enabled) ? "enable" : "disable"), desc);
 370                 }
 371         }
 372
 373         sfc_adapter_unlock(sa);
 374 }
 375
 376 static void
 377 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 378 {
 379         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 380 }
 381
 382 static void
 383 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 384 {
 385         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 386 }
 387
 388 static void
 389 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 390 {
 391         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 392 }
 393
 394 static void
 395 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 396 {
 397         sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 398 }
 399
 400 static int
 401 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 402                    uint16_t nb_rx_desc, unsigned int socket_id,
 403                    const struct rte_eth_rxconf *rx_conf,
 404                    struct rte_mempool *mb_pool)
 405 {
 406         struct sfc_adapter *sa = dev->data->dev_private;
 407         int rc;
 408
 409         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 410                      rx_queue_id, nb_rx_desc, socket_id);
 411
 412         sfc_adapter_lock(sa);
 413
 414         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 415                           rx_conf, mb_pool);
 416         if (rc != 0)
 417                 goto fail_rx_qinit;
 418
 419         dev->data->rx_queues[rx_queue_id] = sa->rxq_info[rx_queue_id].rxq->dp;
 420
 421         sfc_adapter_unlock(sa);
 422
 423         return 0;
 424
 425 fail_rx_qinit:
 426         sfc_adapter_unlock(sa);
 427         SFC_ASSERT(rc > 0);
 428         return -rc;
 429 }
 430
 431 static void
 432 sfc_rx_queue_release(void *queue)
 433 {
 434         struct sfc_dp_rxq *dp_rxq = queue;
 435         struct sfc_rxq *rxq;
 436         struct sfc_adapter *sa;
 437         unsigned int sw_index;
 438
 439         if (dp_rxq == NULL)
 440                 return;
 441
 442         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 443         sa = rxq->evq->sa;
 444         sfc_adapter_lock(sa);
 445
 446         sw_index = sfc_rxq_sw_index(rxq);
 447
 448         sfc_log_init(sa, "RxQ=%u", sw_index);
 449
 450         sa->eth_dev->data->rx_queues[sw_index] = NULL;
 451
 452         sfc_rx_qfini(sa, sw_index);
 453
 454         sfc_adapter_unlock(sa);
 455 }
 456
 457 static int
 458 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 459                    uint16_t nb_tx_desc, unsigned int socket_id,
 460                    const struct rte_eth_txconf *tx_conf)
 461 {
 462         struct sfc_adapter *sa = dev->data->dev_private;
 463         int rc;
 464
 465         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 466                      tx_queue_id, nb_tx_desc, socket_id);
 467
 468         sfc_adapter_lock(sa);
 469
 470         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 471         if (rc != 0)
 472                 goto fail_tx_qinit;
 473
 474         dev->data->tx_queues[tx_queue_id] = sa->txq_info[tx_queue_id].txq->dp;
 475
 476         sfc_adapter_unlock(sa);
 477         return 0;
 478
 479 fail_tx_qinit:
 480         sfc_adapter_unlock(sa);
 481         SFC_ASSERT(rc > 0);
 482         return -rc;
 483 }
 484
 485 static void
 486 sfc_tx_queue_release(void *queue)
 487 {
 488         struct sfc_dp_txq *dp_txq = queue;
 489         struct sfc_txq *txq;
 490         unsigned int sw_index;
 491         struct sfc_adapter *sa;
 492
 493         if (dp_txq == NULL)
 494                 return;
 495
 496         txq = sfc_txq_by_dp_txq(dp_txq);
 497         sw_index = sfc_txq_sw_index(txq);
 498
 499         SFC_ASSERT(txq->evq != NULL);
 500         sa = txq->evq->sa;
 501
 502         sfc_log_init(sa, "TxQ = %u", sw_index);
 503
 504         sfc_adapter_lock(sa);
 505
 506         SFC_ASSERT(sw_index < sa->eth_dev->data->nb_tx_queues);
 507         sa->eth_dev->data->tx_queues[sw_index] = NULL;
 508
 509         sfc_tx_qfini(sa, sw_index);
 510
 511         sfc_adapter_unlock(sa);
 512 }
 513
 514 static void
 515 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 516 {
 517         struct sfc_adapter *sa = dev->data->dev_private;
 518         struct sfc_port *port = &sa->port;
 519         uint64_t *mac_stats;
 520
 521         rte_spinlock_lock(&port->mac_stats_lock);
 522
 523         if (sfc_port_update_mac_stats(sa) != 0)
 524                 goto unlock;
 525
 526         mac_stats = port->mac_stats_buf;
 527
 528         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 529                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 530                 stats->ipackets =
 531                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 532                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 533                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 534                 stats->opackets =
 535                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 536                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 537                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 538                 stats->ibytes =
 539                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 540                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 541                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 542                 stats->obytes =
 543                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 544                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 545                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 546                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_OVERFLOW];
 547                 stats->ierrors = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 548                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 549         } else {
 550                 stats->ipackets = mac_stats[EFX_MAC_RX_PKTS];
 551                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 552                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 553                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 554                 /*
 555                  * Take into account stats which are whenever supported
 556                  * on EF10. If some stat is not supported by current
 557                  * firmware variant or HW revision, it is guaranteed
 558                  * to be zero in mac_stats.
 559                  */
 560                 stats->imissed =
 561                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 562                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 563                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 564                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 565                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 566                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 567                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 568                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 569                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 570                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 571                 stats->ierrors =
 572                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 573                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 574                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 575                 /* no oerrors counters supported on EF10 */
 576         }
 577
 578 unlock:
 579         rte_spinlock_unlock(&port->mac_stats_lock);
 580 }
 581
 582 static void
 583 sfc_stats_reset(struct rte_eth_dev *dev)
 584 {
 585         struct sfc_adapter *sa = dev->data->dev_private;
 586         struct sfc_port *port = &sa->port;
 587         int rc;
 588
 589         if (sa->state != SFC_ADAPTER_STARTED) {
 590                 /*
 591                  * The operation cannot be done if port is not started; it
 592                  * will be scheduled to be done during the next port start
 593                  */
 594                 port->mac_stats_reset_pending = B_TRUE;
 595                 return;
 596         }
 597
 598         rc = sfc_port_reset_mac_stats(sa);
 599         if (rc != 0)
 600                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 601 }
 602
 603 static int
 604 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 605                unsigned int xstats_count)
 606 {
 607         struct sfc_adapter *sa = dev->data->dev_private;
 608         struct sfc_port *port = &sa->port;
 609         uint64_t *mac_stats;
 610         int rc;
 611         unsigned int i;
 612         int nstats = 0;
 613
 614         rte_spinlock_lock(&port->mac_stats_lock);
 615
 616         rc = sfc_port_update_mac_stats(sa);
 617         if (rc != 0) {
 618                 SFC_ASSERT(rc > 0);
 619                 nstats = -rc;
 620                 goto unlock;
 621         }
 622
 623         mac_stats = port->mac_stats_buf;
 624
 625         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 626                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 627                         if (xstats != NULL && nstats < (int)xstats_count) {
 628                                 xstats[nstats].id = nstats;
 629                                 xstats[nstats].value = mac_stats[i];
 630                         }
 631                         nstats++;
 632                 }
 633         }
 634
 635 unlock:
 636         rte_spinlock_unlock(&port->mac_stats_lock);
 637
 638         return nstats;
 639 }
 640
 641 static int
 642 sfc_xstats_get_names(struct rte_eth_dev *dev,
 643                      struct rte_eth_xstat_name *xstats_names,
 644                      unsigned int xstats_count)
 645 {
 646         struct sfc_adapter *sa = dev->data->dev_private;
 647         struct sfc_port *port = &sa->port;
 648         unsigned int i;
 649         unsigned int nstats = 0;
 650
 651         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 652                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 653                         if (xstats_names != NULL && nstats < xstats_count)
 654                                 strncpy(xstats_names[nstats].name,
 655                                         efx_mac_stat_name(sa->nic, i),
 656                                         sizeof(xstats_names[0].name));
 657                         nstats++;
 658                 }
 659         }
 660
 661         return nstats;
 662 }
 663
 664 static int
 665 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 666 {
 667         struct sfc_adapter *sa = dev->data->dev_private;
 668         unsigned int wanted_fc, link_fc;
 669
 670         memset(fc_conf, 0, sizeof(*fc_conf));
 671
 672         sfc_adapter_lock(sa);
 673
 674         if (sa->state == SFC_ADAPTER_STARTED)
 675                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 676         else
 677                 link_fc = sa->port.flow_ctrl;
 678
 679         switch (link_fc) {
 680         case 0:
 681                 fc_conf->mode = RTE_FC_NONE;
 682                 break;
 683         case EFX_FCNTL_RESPOND:
 684                 fc_conf->mode = RTE_FC_RX_PAUSE;
 685                 break;
 686         case EFX_FCNTL_GENERATE:
 687                 fc_conf->mode = RTE_FC_TX_PAUSE;
 688                 break;
 689         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 690                 fc_conf->mode = RTE_FC_FULL;
 691                 break;
 692         default:
 693                 sfc_err(sa, "%s: unexpected flow control value %#x",
 694                         __func__, link_fc);
 695         }
 696
 697         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 698
 699         sfc_adapter_unlock(sa);
 700
 701         return 0;
 702 }
 703
 704 static int
 705 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 706 {
 707         struct sfc_adapter *sa = dev->data->dev_private;
 708         struct sfc_port *port = &sa->port;
 709         unsigned int fcntl;
 710         int rc;
 711
 712         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 713             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 714             fc_conf->mac_ctrl_frame_fwd != 0) {
 715                 sfc_err(sa, "unsupported flow control settings specified");
 716                 rc = EINVAL;
 717                 goto fail_inval;
 718         }
 719
 720         switch (fc_conf->mode) {
 721         case RTE_FC_NONE:
 722                 fcntl = 0;
 723                 break;
 724         case RTE_FC_RX_PAUSE:
 725                 fcntl = EFX_FCNTL_RESPOND;
 726                 break;
 727         case RTE_FC_TX_PAUSE:
 728                 fcntl = EFX_FCNTL_GENERATE;
 729                 break;
 730         case RTE_FC_FULL:
 731                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 732                 break;
 733         default:
 734                 rc = EINVAL;
 735                 goto fail_inval;
 736         }
 737
 738         sfc_adapter_lock(sa);
 739
 740         if (sa->state == SFC_ADAPTER_STARTED) {
 741                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 742                 if (rc != 0)
 743                         goto fail_mac_fcntl_set;
 744         }
 745
 746         port->flow_ctrl = fcntl;
 747         port->flow_ctrl_autoneg = fc_conf->autoneg;
 748
 749         sfc_adapter_unlock(sa);
 750
 751         return 0;
 752
 753 fail_mac_fcntl_set:
 754         sfc_adapter_unlock(sa);
 755 fail_inval:
 756         SFC_ASSERT(rc > 0);
 757         return -rc;
 758 }
 759
 760 static int
 761 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 762 {
 763         struct sfc_adapter *sa = dev->data->dev_private;
 764         size_t pdu = EFX_MAC_PDU(mtu);
 765         size_t old_pdu;
 766         int rc;
 767
 768         sfc_log_init(sa, "mtu=%u", mtu);
 769
 770         rc = EINVAL;
 771         if (pdu < EFX_MAC_PDU_MIN) {
 772                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 773                         (unsigned int)mtu, (unsigned int)pdu,
 774                         EFX_MAC_PDU_MIN);
 775                 goto fail_inval;
 776         }
 777         if (pdu > EFX_MAC_PDU_MAX) {
 778                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 779                         (unsigned int)mtu, (unsigned int)pdu,
 780                         EFX_MAC_PDU_MAX);
 781                 goto fail_inval;
 782         }
 783
 784         sfc_adapter_lock(sa);
 785
 786         if (pdu != sa->port.pdu) {
 787                 if (sa->state == SFC_ADAPTER_STARTED) {
 788                         sfc_stop(sa);
 789
 790                         old_pdu = sa->port.pdu;
 791                         sa->port.pdu = pdu;
 792                         rc = sfc_start(sa);
 793                         if (rc != 0)
 794                                 goto fail_start;
 795                 } else {
 796                         sa->port.pdu = pdu;
 797                 }
 798         }
 799
 800         /*
 801          * The driver does not use it, but other PMDs update jumbo_frame
 802          * flag and max_rx_pkt_len when MTU is set.
 803          */
 804         dev->data->dev_conf.rxmode.jumbo_frame = (mtu > ETHER_MAX_LEN);
 805         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 806
 807         sfc_adapter_unlock(sa);
 808
 809         sfc_log_init(sa, "done");
 810         return 0;
 811
 812 fail_start:
 813         sa->port.pdu = old_pdu;
 814         if (sfc_start(sa) != 0)
 815                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 816                         "PDU max size - port is stopped",
 817                         (unsigned int)pdu, (unsigned int)old_pdu);
 818         sfc_adapter_unlock(sa);
 819
 820 fail_inval:
 821         sfc_log_init(sa, "failed %d", rc);
 822         SFC_ASSERT(rc > 0);
 823         return -rc;
 824 }
 825 static void
 826 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 827 {
 828         struct sfc_adapter *sa = dev->data->dev_private;
 829         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 830         int rc;
 831
 832         sfc_adapter_lock(sa);
 833
 834         if (sa->state != SFC_ADAPTER_STARTED) {
 835                 sfc_info(sa, "the port is not started");
 836                 sfc_info(sa, "the new MAC address will be set on port start");
 837
 838                 goto unlock;
 839         }
 840
 841         if (encp->enc_allow_set_mac_with_installed_filters) {
 842                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 843                 if (rc != 0) {
 844                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 845                         goto unlock;
 846                 }
 847
 848                 /*
 849                  * Changing the MAC address by means of MCDI request
 850                  * has no effect on received traffic, therefore
 851                  * we also need to update unicast filters
 852                  */
 853                 rc = sfc_set_rx_mode(sa);
 854                 if (rc != 0)
 855                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 856         } else {
 857                 sfc_warn(sa, "cannot set MAC address with filters installed");
 858                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 859                 sfc_warn(sa, "(some traffic may be dropped)");
 860
 861                 /*
 862                  * Since setting MAC address with filters installed is not
 863                  * allowed on the adapter, one needs to simply restart adapter
 864                  * so that the new MAC address will be taken from an outer
 865                  * storage and set flawlessly by means of sfc_start() call
 866                  */
 867                 sfc_stop(sa);
 868                 rc = sfc_start(sa);
 869                 if (rc != 0)
 870                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
 871         }
 872
 873 unlock:
 874         sfc_adapter_unlock(sa);
 875 }
 876
 877
 878 static int
 879 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
 880                      uint32_t nb_mc_addr)
 881 {
 882         struct sfc_adapter *sa = dev->data->dev_private;
 883         struct sfc_port *port = &sa->port;
 884         uint8_t *mc_addrs = port->mcast_addrs;
 885         int rc;
 886         unsigned int i;
 887
 888         if (mc_addrs == NULL)
 889                 return -ENOBUFS;
 890
 891         if (nb_mc_addr > port->max_mcast_addrs) {
 892                 sfc_err(sa, "too many multicast addresses: %u > %u",
 893                          nb_mc_addr, port->max_mcast_addrs);
 894                 return -EINVAL;
 895         }
 896
 897         for (i = 0; i < nb_mc_addr; ++i) {
 898                 (void)rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
 899                                  EFX_MAC_ADDR_LEN);
 900                 mc_addrs += EFX_MAC_ADDR_LEN;
 901         }
 902
 903         port->nb_mcast_addrs = nb_mc_addr;
 904
 905         if (sa->state != SFC_ADAPTER_STARTED)
 906                 return 0;
 907
 908         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
 909                                         port->nb_mcast_addrs);
 910         if (rc != 0)
 911                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
 912
 913         SFC_ASSERT(rc > 0);
 914         return -rc;
 915 }
 916
 917 static void
 918 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 919                       struct rte_eth_rxq_info *qinfo)
 920 {
 921         struct sfc_adapter *sa = dev->data->dev_private;
 922         struct sfc_rxq_info *rxq_info;
 923         struct sfc_rxq *rxq;
 924
 925         sfc_adapter_lock(sa);
 926
 927         SFC_ASSERT(rx_queue_id < sa->rxq_count);
 928
 929         rxq_info = &sa->rxq_info[rx_queue_id];
 930         rxq = rxq_info->rxq;
 931         SFC_ASSERT(rxq != NULL);
 932
 933         qinfo->mp = rxq->refill_mb_pool;
 934         qinfo->conf.rx_free_thresh = rxq->refill_threshold;
 935         qinfo->conf.rx_drop_en = 1;
 936         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
 937         qinfo->scattered_rx = (rxq_info->type == EFX_RXQ_TYPE_SCATTER);
 938         qinfo->nb_desc = rxq_info->entries;
 939
 940         sfc_adapter_unlock(sa);
 941 }
 942
 943 static void
 944 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 945                       struct rte_eth_txq_info *qinfo)
 946 {
 947         struct sfc_adapter *sa = dev->data->dev_private;
 948         struct sfc_txq_info *txq_info;
 949
 950         sfc_adapter_lock(sa);
 951
 952         SFC_ASSERT(tx_queue_id < sa->txq_count);
 953
 954         txq_info = &sa->txq_info[tx_queue_id];
 955         SFC_ASSERT(txq_info->txq != NULL);
 956
 957         memset(qinfo, 0, sizeof(*qinfo));
 958
 959         qinfo->conf.txq_flags = txq_info->txq->flags;
 960         qinfo->conf.tx_free_thresh = txq_info->txq->free_thresh;
 961         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
 962         qinfo->nb_desc = txq_info->entries;
 963
 964         sfc_adapter_unlock(sa);
 965 }
 966
 967 static uint32_t
 968 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 969 {
 970         struct sfc_adapter *sa = dev->data->dev_private;
 971
 972         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 973
 974         return sfc_rx_qdesc_npending(sa, rx_queue_id);
 975 }
 976
 977 static int
 978 sfc_rx_descriptor_done(void *queue, uint16_t offset)
 979 {
 980         struct sfc_dp_rxq *dp_rxq = queue;
 981
 982         return sfc_rx_qdesc_done(dp_rxq, offset);
 983 }
 984
 985 static int
 986 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
 987 {
 988         struct sfc_adapter *sa = dev->data->dev_private;
 989         int rc;
 990
 991         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
 992
 993         sfc_adapter_lock(sa);
 994
 995         rc = EINVAL;
 996         if (sa->state != SFC_ADAPTER_STARTED)
 997                 goto fail_not_started;
 998
 999         rc = sfc_rx_qstart(sa, rx_queue_id);
1000         if (rc != 0)
1001                 goto fail_rx_qstart;
1002
1003         sa->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1004
1005         sfc_adapter_unlock(sa);
1006
1007         return 0;
1008
1009 fail_rx_qstart:
1010 fail_not_started:
1011         sfc_adapter_unlock(sa);
1012         SFC_ASSERT(rc > 0);
1013         return -rc;
1014 }
1015
1016 static int
1017 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1018 {
1019         struct sfc_adapter *sa = dev->data->dev_private;
1020
1021         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1022
1023         sfc_adapter_lock(sa);
1024         sfc_rx_qstop(sa, rx_queue_id);
1025
1026         sa->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1027
1028         sfc_adapter_unlock(sa);
1029
1030         return 0;
1031 }
1032
1033 static int
1034 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1035 {
1036         struct sfc_adapter *sa = dev->data->dev_private;
1037         int rc;
1038
1039         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1040
1041         sfc_adapter_lock(sa);
1042
1043         rc = EINVAL;
1044         if (sa->state != SFC_ADAPTER_STARTED)
1045                 goto fail_not_started;
1046
1047         rc = sfc_tx_qstart(sa, tx_queue_id);
1048         if (rc != 0)
1049                 goto fail_tx_qstart;
1050
1051         sa->txq_info[tx_queue_id].deferred_started = B_TRUE;
1052
1053         sfc_adapter_unlock(sa);
1054         return 0;
1055
1056 fail_tx_qstart:
1057
1058 fail_not_started:
1059         sfc_adapter_unlock(sa);
1060         SFC_ASSERT(rc > 0);
1061         return -rc;
1062 }
1063
1064 static int
1065 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1066 {
1067         struct sfc_adapter *sa = dev->data->dev_private;
1068
1069         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1070
1071         sfc_adapter_lock(sa);
1072
1073         sfc_tx_qstop(sa, tx_queue_id);
1074
1075         sa->txq_info[tx_queue_id].deferred_started = B_FALSE;
1076
1077         sfc_adapter_unlock(sa);
1078         return 0;
1079 }
1080
1081 #if EFSYS_OPT_RX_SCALE
1082 static int
1083 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1084                           struct rte_eth_rss_conf *rss_conf)
1085 {
1086         struct sfc_adapter *sa = dev->data->dev_private;
1087
1088         if ((sa->rss_channels == 1) ||
1089             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1090                 return -ENOTSUP;
1091
1092         sfc_adapter_lock(sa);
1093
1094         /*
1095          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1096          * hence, conversion is done here to derive a correct set of ETH_RSS
1097          * flags which corresponds to the active EFX configuration stored
1098          * locally in 'sfc_adapter' and kept up-to-date
1099          */
1100         rss_conf->rss_hf = sfc_efx_to_rte_hash_type(sa->rss_hash_types);
1101         rss_conf->rss_key_len = SFC_RSS_KEY_SIZE;
1102         if (rss_conf->rss_key != NULL)
1103                 rte_memcpy(rss_conf->rss_key, sa->rss_key, SFC_RSS_KEY_SIZE);
1104
1105         sfc_adapter_unlock(sa);
1106
1107         return 0;
1108 }
1109
1110 static int
1111 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1112                         struct rte_eth_rss_conf *rss_conf)
1113 {
1114         struct sfc_adapter *sa = dev->data->dev_private;
1115         unsigned int efx_hash_types;
1116         int rc = 0;
1117
1118         if ((sa->rss_channels == 1) ||
1119             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1120                 sfc_err(sa, "RSS is not available");
1121                 return -ENOTSUP;
1122         }
1123
1124         if ((rss_conf->rss_key != NULL) &&
1125             (rss_conf->rss_key_len != sizeof(sa->rss_key))) {
1126                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1127                         sizeof(sa->rss_key));
1128                 return -EINVAL;
1129         }
1130
1131         if ((rss_conf->rss_hf & ~SFC_RSS_OFFLOADS) != 0) {
1132                 sfc_err(sa, "unsupported hash functions requested");
1133                 return -EINVAL;
1134         }
1135
1136         sfc_adapter_lock(sa);
1137
1138         efx_hash_types = sfc_rte_to_efx_hash_type(rss_conf->rss_hf);
1139
1140         rc = efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1141                                    efx_hash_types, B_TRUE);
1142         if (rc != 0)
1143                 goto fail_scale_mode_set;
1144
1145         if (rss_conf->rss_key != NULL) {
1146                 if (sa->state == SFC_ADAPTER_STARTED) {
1147                         rc = efx_rx_scale_key_set(sa->nic, rss_conf->rss_key,
1148                                                   sizeof(sa->rss_key));
1149                         if (rc != 0)
1150                                 goto fail_scale_key_set;
1151                 }
1152
1153                 rte_memcpy(sa->rss_key, rss_conf->rss_key, sizeof(sa->rss_key));
1154         }
1155
1156         sa->rss_hash_types = efx_hash_types;
1157
1158         sfc_adapter_unlock(sa);
1159
1160         return 0;
1161
1162 fail_scale_key_set:
1163         if (efx_rx_scale_mode_set(sa->nic, EFX_RX_HASHALG_TOEPLITZ,
1164                                   sa->rss_hash_types, B_TRUE) != 0)
1165                 sfc_err(sa, "failed to restore RSS mode");
1166
1167 fail_scale_mode_set:
1168         sfc_adapter_unlock(sa);
1169         return -rc;
1170 }
1171
1172 static int
1173 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1174                        struct rte_eth_rss_reta_entry64 *reta_conf,
1175                        uint16_t reta_size)
1176 {
1177         struct sfc_adapter *sa = dev->data->dev_private;
1178         int entry;
1179
1180         if ((sa->rss_channels == 1) ||
1181             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE))
1182                 return -ENOTSUP;
1183
1184         if (reta_size != EFX_RSS_TBL_SIZE)
1185                 return -EINVAL;
1186
1187         sfc_adapter_lock(sa);
1188
1189         for (entry = 0; entry < reta_size; entry++) {
1190                 int grp = entry / RTE_RETA_GROUP_SIZE;
1191                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1192
1193                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1194                         reta_conf[grp].reta[grp_idx] = sa->rss_tbl[entry];
1195         }
1196
1197         sfc_adapter_unlock(sa);
1198
1199         return 0;
1200 }
1201
1202 static int
1203 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1204                         struct rte_eth_rss_reta_entry64 *reta_conf,
1205                         uint16_t reta_size)
1206 {
1207         struct sfc_adapter *sa = dev->data->dev_private;
1208         unsigned int *rss_tbl_new;
1209         uint16_t entry;
1210         int rc;
1211
1212
1213         if ((sa->rss_channels == 1) ||
1214             (sa->rss_support != EFX_RX_SCALE_EXCLUSIVE)) {
1215                 sfc_err(sa, "RSS is not available");
1216                 return -ENOTSUP;
1217         }
1218
1219         if (reta_size != EFX_RSS_TBL_SIZE) {
1220                 sfc_err(sa, "RETA size is wrong (should be %u)",
1221                         EFX_RSS_TBL_SIZE);
1222                 return -EINVAL;
1223         }
1224
1225         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(sa->rss_tbl), 0);
1226         if (rss_tbl_new == NULL)
1227                 return -ENOMEM;
1228
1229         sfc_adapter_lock(sa);
1230
1231         rte_memcpy(rss_tbl_new, sa->rss_tbl, sizeof(sa->rss_tbl));
1232
1233         for (entry = 0; entry < reta_size; entry++) {
1234                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1235                 struct rte_eth_rss_reta_entry64 *grp;
1236
1237                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1238
1239                 if (grp->mask & (1ull << grp_idx)) {
1240                         if (grp->reta[grp_idx] >= sa->rss_channels) {
1241                                 rc = EINVAL;
1242                                 goto bad_reta_entry;
1243                         }
1244                         rss_tbl_new[entry] = grp->reta[grp_idx];
1245                 }
1246         }
1247
1248         rc = efx_rx_scale_tbl_set(sa->nic, rss_tbl_new, EFX_RSS_TBL_SIZE);
1249         if (rc == 0)
1250                 rte_memcpy(sa->rss_tbl, rss_tbl_new, sizeof(sa->rss_tbl));
1251
1252 bad_reta_entry:
1253         sfc_adapter_unlock(sa);
1254
1255         rte_free(rss_tbl_new);
1256
1257         SFC_ASSERT(rc >= 0);
1258         return -rc;
1259 }
1260 #endif
1261
1262 static int
1263 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1264                     enum rte_filter_op filter_op,
1265                     void *arg)
1266 {
1267         struct sfc_adapter *sa = dev->data->dev_private;
1268         int rc = ENOTSUP;
1269
1270         sfc_log_init(sa, "entry");
1271
1272         switch (filter_type) {
1273         case RTE_ETH_FILTER_NONE:
1274                 sfc_err(sa, "Global filters configuration not supported");
1275                 break;
1276         case RTE_ETH_FILTER_MACVLAN:
1277                 sfc_err(sa, "MACVLAN filters not supported");
1278                 break;
1279         case RTE_ETH_FILTER_ETHERTYPE:
1280                 sfc_err(sa, "EtherType filters not supported");
1281                 break;
1282         case RTE_ETH_FILTER_FLEXIBLE:
1283                 sfc_err(sa, "Flexible filters not supported");
1284                 break;
1285         case RTE_ETH_FILTER_SYN:
1286                 sfc_err(sa, "SYN filters not supported");
1287                 break;
1288         case RTE_ETH_FILTER_NTUPLE:
1289                 sfc_err(sa, "NTUPLE filters not supported");
1290                 break;
1291         case RTE_ETH_FILTER_TUNNEL:
1292                 sfc_err(sa, "Tunnel filters not supported");
1293                 break;
1294         case RTE_ETH_FILTER_FDIR:
1295                 sfc_err(sa, "Flow Director filters not supported");
1296                 break;
1297         case RTE_ETH_FILTER_HASH:
1298                 sfc_err(sa, "Hash filters not supported");
1299                 break;
1300         case RTE_ETH_FILTER_GENERIC:
1301                 if (filter_op != RTE_ETH_FILTER_GET) {
1302                         rc = EINVAL;
1303                 } else {
1304                         *(const void **)arg = &sfc_flow_ops;
1305                         rc = 0;
1306                 }
1307                 break;
1308         default:
1309                 sfc_err(sa, "Unknown filter type %u", filter_type);
1310                 break;
1311         }
1312
1313         sfc_log_init(sa, "exit: %d", -rc);
1314         SFC_ASSERT(rc >= 0);
1315         return -rc;
1316 }
1317
1318 static const struct eth_dev_ops sfc_eth_dev_ops = {
1319         .dev_configure                  = sfc_dev_configure,
1320         .dev_start                      = sfc_dev_start,
1321         .dev_stop                       = sfc_dev_stop,
1322         .dev_set_link_up                = sfc_dev_set_link_up,
1323         .dev_set_link_down              = sfc_dev_set_link_down,
1324         .dev_close                      = sfc_dev_close,
1325         .promiscuous_enable             = sfc_dev_promisc_enable,
1326         .promiscuous_disable            = sfc_dev_promisc_disable,
1327         .allmulticast_enable            = sfc_dev_allmulti_enable,
1328         .allmulticast_disable           = sfc_dev_allmulti_disable,
1329         .link_update                    = sfc_dev_link_update,
1330         .stats_get                      = sfc_stats_get,
1331         .stats_reset                    = sfc_stats_reset,
1332         .xstats_get                     = sfc_xstats_get,
1333         .xstats_reset                   = sfc_stats_reset,
1334         .xstats_get_names               = sfc_xstats_get_names,
1335         .dev_infos_get                  = sfc_dev_infos_get,
1336         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1337         .mtu_set                        = sfc_dev_set_mtu,
1338         .rx_queue_start                 = sfc_rx_queue_start,
1339         .rx_queue_stop                  = sfc_rx_queue_stop,
1340         .tx_queue_start                 = sfc_tx_queue_start,
1341         .tx_queue_stop                  = sfc_tx_queue_stop,
1342         .rx_queue_setup                 = sfc_rx_queue_setup,
1343         .rx_queue_release               = sfc_rx_queue_release,
1344         .rx_queue_count                 = sfc_rx_queue_count,
1345         .rx_descriptor_done             = sfc_rx_descriptor_done,
1346         .tx_queue_setup                 = sfc_tx_queue_setup,
1347         .tx_queue_release               = sfc_tx_queue_release,
1348         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1349         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1350         .mac_addr_set                   = sfc_mac_addr_set,
1351 #if EFSYS_OPT_RX_SCALE
1352         .reta_update                    = sfc_dev_rss_reta_update,
1353         .reta_query                     = sfc_dev_rss_reta_query,
1354         .rss_hash_update                = sfc_dev_rss_hash_update,
1355         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1356 #endif
1357         .filter_ctrl                    = sfc_dev_filter_ctrl,
1358         .set_mc_addr_list               = sfc_set_mc_addr_list,
1359         .rxq_info_get                   = sfc_rx_queue_info_get,
1360         .txq_info_get                   = sfc_tx_queue_info_get,
1361         .fw_version_get                 = sfc_fw_version_get,
1362 };
1363
1364 static int
1365 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1366 {
1367         struct sfc_adapter *sa = dev->data->dev_private;
1368         unsigned int avail_caps = 0;
1369         const char *rx_name = NULL;
1370         const char *tx_name = NULL;
1371         int rc;
1372
1373         if (sa == NULL || sa->state == SFC_ADAPTER_UNINITIALIZED)
1374                 return -E_RTE_SECONDARY;
1375
1376         switch (sa->family) {
1377         case EFX_FAMILY_HUNTINGTON:
1378         case EFX_FAMILY_MEDFORD:
1379                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1380                 break;
1381         default:
1382                 break;
1383         }
1384
1385         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1386                                 sfc_kvarg_string_handler, &rx_name);
1387         if (rc != 0)
1388                 goto fail_kvarg_rx_datapath;
1389
1390         if (rx_name != NULL) {
1391                 sa->dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1392                 if (sa->dp_rx == NULL) {
1393                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1394                         rc = ENOENT;
1395                         goto fail_dp_rx;
1396                 }
1397                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_rx->dp, avail_caps)) {
1398                         sfc_err(sa,
1399                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1400                                 rx_name);
1401                         rc = EINVAL;
1402                         goto fail_dp_rx;
1403                 }
1404         } else {
1405                 sa->dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1406                 if (sa->dp_rx == NULL) {
1407                         sfc_err(sa, "Rx datapath by caps %#x not found",
1408                                 avail_caps);
1409                         rc = ENOENT;
1410                         goto fail_dp_rx;
1411                 }
1412         }
1413
1414         sfc_info(sa, "use %s Rx datapath", sa->dp_rx->dp.name);
1415
1416         dev->rx_pkt_burst = sa->dp_rx->pkt_burst;
1417
1418         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1419                                 sfc_kvarg_string_handler, &tx_name);
1420         if (rc != 0)
1421                 goto fail_kvarg_tx_datapath;
1422
1423         if (tx_name != NULL) {
1424                 sa->dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1425                 if (sa->dp_tx == NULL) {
1426                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1427                         rc = ENOENT;
1428                         goto fail_dp_tx;
1429                 }
1430                 if (!sfc_dp_match_hw_fw_caps(&sa->dp_tx->dp, avail_caps)) {
1431                         sfc_err(sa,
1432                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1433                                 tx_name);
1434                         rc = EINVAL;
1435                         goto fail_dp_tx;
1436                 }
1437         } else {
1438                 sa->dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1439                 if (sa->dp_tx == NULL) {
1440                         sfc_err(sa, "Tx datapath by caps %#x not found",
1441                                 avail_caps);
1442                         rc = ENOENT;
1443                         goto fail_dp_tx;
1444                 }
1445         }
1446
1447         sfc_info(sa, "use %s Tx datapath", sa->dp_tx->dp.name);
1448
1449         dev->tx_pkt_burst = sa->dp_tx->pkt_burst;
1450
1451         dev->dev_ops = &sfc_eth_dev_ops;
1452
1453         return 0;
1454
1455 fail_dp_tx:
1456 fail_kvarg_tx_datapath:
1457 fail_dp_rx:
1458 fail_kvarg_rx_datapath:
1459         return rc;
1460 }
1461
1462 static void
1463 sfc_register_dp(void)
1464 {
1465         /* Register once */
1466         if (TAILQ_EMPTY(&sfc_dp_head)) {
1467                 /* Prefer EF10 datapath */
1468                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1469                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
1470
1471                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
1472                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
1473                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
1474         }
1475 }
1476
1477 static int
1478 sfc_eth_dev_init(struct rte_eth_dev *dev)
1479 {
1480         struct sfc_adapter *sa = dev->data->dev_private;
1481         struct rte_pci_device *pci_dev = SFC_DEV_TO_PCI(dev);
1482         int rc;
1483         const efx_nic_cfg_t *encp;
1484         const struct ether_addr *from;
1485
1486         sfc_register_dp();
1487
1488         /* Required for logging */
1489         sa->eth_dev = dev;
1490
1491         /* Copy PCI device info to the dev->data */
1492         rte_eth_copy_pci_info(dev, pci_dev);
1493
1494         rc = sfc_kvargs_parse(sa);
1495         if (rc != 0)
1496                 goto fail_kvargs_parse;
1497
1498         rc = sfc_kvargs_process(sa, SFC_KVARG_DEBUG_INIT,
1499                                 sfc_kvarg_bool_handler, &sa->debug_init);
1500         if (rc != 0)
1501                 goto fail_kvarg_debug_init;
1502
1503         sfc_log_init(sa, "entry");
1504
1505         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
1506         if (dev->data->mac_addrs == NULL) {
1507                 rc = ENOMEM;
1508                 goto fail_mac_addrs;
1509         }
1510
1511         sfc_adapter_lock_init(sa);
1512         sfc_adapter_lock(sa);
1513
1514         sfc_log_init(sa, "attaching");
1515         rc = sfc_attach(sa);
1516         if (rc != 0)
1517                 goto fail_attach;
1518
1519         encp = efx_nic_cfg_get(sa->nic);
1520
1521         /*
1522          * The arguments are really reverse order in comparison to
1523          * Linux kernel. Copy from NIC config to Ethernet device data.
1524          */
1525         from = (const struct ether_addr *)(encp->enc_mac_addr);
1526         ether_addr_copy(from, &dev->data->mac_addrs[0]);
1527
1528         sfc_adapter_unlock(sa);
1529
1530         sfc_eth_dev_set_ops(dev);
1531
1532         sfc_log_init(sa, "done");
1533         return 0;
1534
1535 fail_attach:
1536         sfc_adapter_unlock(sa);
1537         sfc_adapter_lock_fini(sa);
1538         rte_free(dev->data->mac_addrs);
1539         dev->data->mac_addrs = NULL;
1540
1541 fail_mac_addrs:
1542 fail_kvarg_debug_init:
1543         sfc_kvargs_cleanup(sa);
1544
1545 fail_kvargs_parse:
1546         sfc_log_init(sa, "failed %d", rc);
1547         SFC_ASSERT(rc > 0);
1548         return -rc;
1549 }
1550
1551 static int
1552 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
1553 {
1554         struct sfc_adapter *sa = dev->data->dev_private;
1555
1556         sfc_log_init(sa, "entry");
1557
1558         sfc_adapter_lock(sa);
1559
1560         sfc_detach(sa);
1561
1562         rte_free(dev->data->mac_addrs);
1563         dev->data->mac_addrs = NULL;
1564
1565         dev->dev_ops = NULL;
1566         dev->rx_pkt_burst = NULL;
1567         dev->tx_pkt_burst = NULL;
1568
1569         sfc_kvargs_cleanup(sa);
1570
1571         sfc_adapter_unlock(sa);
1572         sfc_adapter_lock_fini(sa);
1573
1574         sfc_log_init(sa, "done");
1575
1576         /* Required for logging, so cleanup last */
1577         sa->eth_dev = NULL;
1578         return 0;
1579 }
1580
1581 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
1582         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
1583         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
1584         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
1585         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
1586         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
1587         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
1588         { .vendor_id = 0 /* sentinel */ }
1589 };
1590
1591 static struct eth_driver sfc_efx_pmd = {
1592         .pci_drv = {
1593                 .id_table = pci_id_sfc_efx_map,
1594                 .drv_flags =
1595                         RTE_PCI_DRV_INTR_LSC |
1596                         RTE_PCI_DRV_NEED_MAPPING,
1597                 .probe = rte_eth_dev_pci_probe,
1598                 .remove = rte_eth_dev_pci_remove,
1599         },
1600         .eth_dev_init = sfc_eth_dev_init,
1601         .eth_dev_uninit = sfc_eth_dev_uninit,
1602         .dev_private_size = sizeof(struct sfc_adapter),
1603 };
1604
1605 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd.pci_drv);
1606 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
1607 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio");
1608 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
1609         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
1610         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
1611         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
1612         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long> "
1613         SFC_KVARG_MCDI_LOGGING "=" SFC_KVARG_VALUES_BOOL " "
1614         SFC_KVARG_DEBUG_INIT "=" SFC_KVARG_VALUES_BOOL);