drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright (c) 2016-2018 Solarflare Communications Inc.
   4  * All rights reserved.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_dev.h>
  11 #include <rte_ethdev_driver.h>
  12 #include <rte_ethdev_pci.h>
  13 #include <rte_pci.h>
  14 #include <rte_bus_pci.h>
  15 #include <rte_errno.h>
  16 #include <rte_string_fns.h>
  17
  18 #include "efx.h"
  19
  20 #include "sfc.h"
  21 #include "sfc_debug.h"
  22 #include "sfc_log.h"
  23 #include "sfc_kvargs.h"
  24 #include "sfc_ev.h"
  25 #include "sfc_rx.h"
  26 #include "sfc_tx.h"
  27 #include "sfc_flow.h"
  28 #include "sfc_dp.h"
  29 #include "sfc_dp_rx.h"
  30
  31 uint32_t sfc_logtype_driver;
  32
  33 static struct sfc_dp_list sfc_dp_head =
  34         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  35
  36 static int
  37 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  38 {
  39         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  40         efx_nic_fw_info_t enfi;
  41         int ret;
  42         int rc;
  43
  44         /*
  45          * Return value of the callback is likely supposed to be
  46          * equal to or greater than 0, nevertheless, if an error
  47          * occurs, it will be desirable to pass it to the caller
  48          */
  49         if ((fw_version == NULL) || (fw_size == 0))
  50                 return -EINVAL;
  51
  52         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  53         if (rc != 0)
  54                 return -rc;
  55
  56         ret = snprintf(fw_version, fw_size,
  57                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  58                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  59                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  60         if (ret < 0)
  61                 return ret;
  62
  63         if (enfi.enfi_dpcpu_fw_ids_valid) {
  64                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  65                 int ret_extra;
  66
  67                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  68                                      fw_size - dpcpu_fw_ids_offset,
  69                                      " rx%" PRIx16 " tx%" PRIx16,
  70                                      enfi.enfi_rx_dpcpu_fw_id,
  71                                      enfi.enfi_tx_dpcpu_fw_id);
  72                 if (ret_extra < 0)
  73                         return ret_extra;
  74
  75                 ret += ret_extra;
  76         }
  77
  78         if (fw_size < (size_t)(++ret))
  79                 return ret;
  80         else
  81                 return 0;
  82 }
  83
  84 static void
  85 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  86 {
  87         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  88         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  89         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  90         struct sfc_rss *rss = &sas->rss;
  91         uint64_t txq_offloads_def = 0;
  92
  93         sfc_log_init(sa, "entry");
  94
  95         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
  96
  97         /* Autonegotiation may be disabled */
  98         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
  99         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_1000FDX)
 100                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 101         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_10000FDX)
 102                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 103         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_25000FDX)
 104                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 105         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_40000FDX)
 106                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 107         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_50000FDX)
 108                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 109         if (sa->port.phy_adv_cap_mask & EFX_PHY_CAP_100000FDX)
 110                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 111
 112         dev_info->max_rx_queues = sa->rxq_max;
 113         dev_info->max_tx_queues = sa->txq_max;
 114
 115         /* By default packets are dropped if no descriptors are available */
 116         dev_info->default_rxconf.rx_drop_en = 1;
 117
 118         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 119
 120         /*
 121          * rx_offload_capa includes both device and queue offloads since
 122          * the latter may be requested on a per device basis which makes
 123          * sense when some offloads are needed to be set on all queues.
 124          */
 125         dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 126                                     dev_info->rx_queue_offload_capa;
 127
 128         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 129
 130         /*
 131          * tx_offload_capa includes both device and queue offloads since
 132          * the latter may be requested on a per device basis which makes
 133          * sense when some offloads are needed to be set on all queues.
 134          */
 135         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 136                                     dev_info->tx_queue_offload_capa;
 137
 138         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 139                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 140
 141         dev_info->default_txconf.offloads |= txq_offloads_def;
 142
 143         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 144                 uint64_t rte_hf = 0;
 145                 unsigned int i;
 146
 147                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 148                         rte_hf |= rss->hf_map[i].rte;
 149
 150                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 151                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 152                 dev_info->flow_type_rss_offloads = rte_hf;
 153         }
 154
 155         /* Initialize to hardware limits */
 156         dev_info->rx_desc_lim.nb_max = EFX_RXQ_MAXNDESCS;
 157         dev_info->rx_desc_lim.nb_min = EFX_RXQ_MINNDESCS;
 158         /* The RXQ hardware requires that the descriptor count is a power
 159          * of 2, but rx_desc_lim cannot properly describe that constraint.
 160          */
 161         dev_info->rx_desc_lim.nb_align = EFX_RXQ_MINNDESCS;
 162
 163         /* Initialize to hardware limits */
 164         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 165         dev_info->tx_desc_lim.nb_min = EFX_TXQ_MINNDESCS;
 166         /*
 167          * The TXQ hardware requires that the descriptor count is a power
 168          * of 2, but tx_desc_lim cannot properly describe that constraint
 169          */
 170         dev_info->tx_desc_lim.nb_align = EFX_TXQ_MINNDESCS;
 171
 172         if (sap->dp_rx->get_dev_info != NULL)
 173                 sap->dp_rx->get_dev_info(dev_info);
 174         if (sap->dp_tx->get_dev_info != NULL)
 175                 sap->dp_tx->get_dev_info(dev_info);
 176
 177         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 178                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 179 }
 180
 181 static const uint32_t *
 182 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 183 {
 184         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 185         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 186         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 187         uint32_t tunnel_encaps = encp->enc_tunnel_encapsulations_supported;
 188
 189         return sap->dp_rx->supported_ptypes_get(tunnel_encaps);
 190 }
 191
 192 static int
 193 sfc_dev_configure(struct rte_eth_dev *dev)
 194 {
 195         struct rte_eth_dev_data *dev_data = dev->data;
 196         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 197         int rc;
 198
 199         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 200                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 201
 202         sfc_adapter_lock(sa);
 203         switch (sa->state) {
 204         case SFC_ADAPTER_CONFIGURED:
 205                 /* FALLTHROUGH */
 206         case SFC_ADAPTER_INITIALIZED:
 207                 rc = sfc_configure(sa);
 208                 break;
 209         default:
 210                 sfc_err(sa, "unexpected adapter state %u to configure",
 211                         sa->state);
 212                 rc = EINVAL;
 213                 break;
 214         }
 215         sfc_adapter_unlock(sa);
 216
 217         sfc_log_init(sa, "done %d", rc);
 218         SFC_ASSERT(rc >= 0);
 219         return -rc;
 220 }
 221
 222 static int
 223 sfc_dev_start(struct rte_eth_dev *dev)
 224 {
 225         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 226         int rc;
 227
 228         sfc_log_init(sa, "entry");
 229
 230         sfc_adapter_lock(sa);
 231         rc = sfc_start(sa);
 232         sfc_adapter_unlock(sa);
 233
 234         sfc_log_init(sa, "done %d", rc);
 235         SFC_ASSERT(rc >= 0);
 236         return -rc;
 237 }
 238
 239 static int
 240 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 241 {
 242         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 243         struct rte_eth_link current_link;
 244         int ret;
 245
 246         sfc_log_init(sa, "entry");
 247
 248         if (sa->state != SFC_ADAPTER_STARTED) {
 249                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 250         } else if (wait_to_complete) {
 251                 efx_link_mode_t link_mode;
 252
 253                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 254                         link_mode = EFX_LINK_UNKNOWN;
 255                 sfc_port_link_mode_to_info(link_mode, &current_link);
 256
 257         } else {
 258                 sfc_ev_mgmt_qpoll(sa);
 259                 rte_eth_linkstatus_get(dev, &current_link);
 260         }
 261
 262         ret = rte_eth_linkstatus_set(dev, &current_link);
 263         if (ret == 0)
 264                 sfc_notice(sa, "Link status is %s",
 265                            current_link.link_status ? "UP" : "DOWN");
 266
 267         return ret;
 268 }
 269
 270 static void
 271 sfc_dev_stop(struct rte_eth_dev *dev)
 272 {
 273         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 274
 275         sfc_log_init(sa, "entry");
 276
 277         sfc_adapter_lock(sa);
 278         sfc_stop(sa);
 279         sfc_adapter_unlock(sa);
 280
 281         sfc_log_init(sa, "done");
 282 }
 283
 284 static int
 285 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 286 {
 287         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 288         int rc;
 289
 290         sfc_log_init(sa, "entry");
 291
 292         sfc_adapter_lock(sa);
 293         rc = sfc_start(sa);
 294         sfc_adapter_unlock(sa);
 295
 296         SFC_ASSERT(rc >= 0);
 297         return -rc;
 298 }
 299
 300 static int
 301 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 302 {
 303         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 304
 305         sfc_log_init(sa, "entry");
 306
 307         sfc_adapter_lock(sa);
 308         sfc_stop(sa);
 309         sfc_adapter_unlock(sa);
 310
 311         return 0;
 312 }
 313
 314 static void
 315 sfc_dev_close(struct rte_eth_dev *dev)
 316 {
 317         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 318
 319         sfc_log_init(sa, "entry");
 320
 321         sfc_adapter_lock(sa);
 322         switch (sa->state) {
 323         case SFC_ADAPTER_STARTED:
 324                 sfc_stop(sa);
 325                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 326                 /* FALLTHROUGH */
 327         case SFC_ADAPTER_CONFIGURED:
 328                 sfc_close(sa);
 329                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 330                 /* FALLTHROUGH */
 331         case SFC_ADAPTER_INITIALIZED:
 332                 break;
 333         default:
 334                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 335                 break;
 336         }
 337         sfc_adapter_unlock(sa);
 338
 339         sfc_log_init(sa, "done");
 340 }
 341
 342 static void
 343 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 344                    boolean_t enabled)
 345 {
 346         struct sfc_port *port;
 347         boolean_t *toggle;
 348         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 349         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 350         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 351
 352         sfc_adapter_lock(sa);
 353
 354         port = &sa->port;
 355         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 356
 357         if (*toggle != enabled) {
 358                 *toggle = enabled;
 359
 360                 if (sfc_sa2shared(sa)->isolated) {
 361                         sfc_warn(sa, "isolated mode is active on the port");
 362                         sfc_warn(sa, "the change is to be applied on the next "
 363                                      "start provided that isolated mode is "
 364                                      "disabled prior the next start");
 365                 } else 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_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 407         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 408         int rc;
 409
 410         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 411                      rx_queue_id, nb_rx_desc, socket_id);
 412
 413         sfc_adapter_lock(sa);
 414
 415         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 416                           rx_conf, mb_pool);
 417         if (rc != 0)
 418                 goto fail_rx_qinit;
 419
 420         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 421
 422         sfc_adapter_unlock(sa);
 423
 424         return 0;
 425
 426 fail_rx_qinit:
 427         sfc_adapter_unlock(sa);
 428         SFC_ASSERT(rc > 0);
 429         return -rc;
 430 }
 431
 432 static void
 433 sfc_rx_queue_release(void *queue)
 434 {
 435         struct sfc_dp_rxq *dp_rxq = queue;
 436         struct sfc_rxq *rxq;
 437         struct sfc_adapter *sa;
 438         unsigned int sw_index;
 439
 440         if (dp_rxq == NULL)
 441                 return;
 442
 443         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 444         sa = rxq->evq->sa;
 445         sfc_adapter_lock(sa);
 446
 447         sw_index = dp_rxq->dpq.queue_id;
 448
 449         sfc_log_init(sa, "RxQ=%u", sw_index);
 450
 451         sfc_rx_qfini(sa, sw_index);
 452
 453         sfc_adapter_unlock(sa);
 454 }
 455
 456 static int
 457 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 458                    uint16_t nb_tx_desc, unsigned int socket_id,
 459                    const struct rte_eth_txconf *tx_conf)
 460 {
 461         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 462         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 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] = sas->txq_info[tx_queue_id].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 = dp_txq->dpq.queue_id;
 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_tx_qfini(sa, sw_index);
 507
 508         sfc_adapter_unlock(sa);
 509 }
 510
 511 /*
 512  * Some statistics are computed as A - B where A and B each increase
 513  * monotonically with some hardware counter(s) and the counters are read
 514  * asynchronously.
 515  *
 516  * If packet X is counted in A, but not counted in B yet, computed value is
 517  * greater than real.
 518  *
 519  * If packet X is not counted in A at the moment of reading the counter,
 520  * but counted in B at the moment of reading the counter, computed value
 521  * is less than real.
 522  *
 523  * However, counter which grows backward is worse evil than slightly wrong
 524  * value. So, let's try to guarantee that it never happens except may be
 525  * the case when the MAC stats are zeroed as a result of a NIC reset.
 526  */
 527 static void
 528 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 529 {
 530         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 531                 *stat = newval;
 532 }
 533
 534 static int
 535 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 536 {
 537         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 538         struct sfc_port *port = &sa->port;
 539         uint64_t *mac_stats;
 540         int ret;
 541
 542         rte_spinlock_lock(&port->mac_stats_lock);
 543
 544         ret = sfc_port_update_mac_stats(sa);
 545         if (ret != 0)
 546                 goto unlock;
 547
 548         mac_stats = port->mac_stats_buf;
 549
 550         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 551                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 552                 stats->ipackets =
 553                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 554                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 555                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 556                 stats->opackets =
 557                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 558                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 559                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 560                 stats->ibytes =
 561                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 562                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 563                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 564                 stats->obytes =
 565                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 566                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 567                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 568                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 569                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 570         } else {
 571                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 572                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 573                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 574                 /*
 575                  * Take into account stats which are whenever supported
 576                  * on EF10. If some stat is not supported by current
 577                  * firmware variant or HW revision, it is guaranteed
 578                  * to be zero in mac_stats.
 579                  */
 580                 stats->imissed =
 581                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 582                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 583                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 584                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 585                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 586                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 587                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 588                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 589                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 590                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 591                 stats->ierrors =
 592                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 593                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 594                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 595                 /* no oerrors counters supported on EF10 */
 596
 597                 /* Exclude missed, errors and pauses from Rx packets */
 598                 sfc_update_diff_stat(&port->ipackets,
 599                         mac_stats[EFX_MAC_RX_PKTS] -
 600                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 601                         stats->imissed - stats->ierrors);
 602                 stats->ipackets = port->ipackets;
 603         }
 604
 605 unlock:
 606         rte_spinlock_unlock(&port->mac_stats_lock);
 607         SFC_ASSERT(ret >= 0);
 608         return -ret;
 609 }
 610
 611 static void
 612 sfc_stats_reset(struct rte_eth_dev *dev)
 613 {
 614         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 615         struct sfc_port *port = &sa->port;
 616         int rc;
 617
 618         if (sa->state != SFC_ADAPTER_STARTED) {
 619                 /*
 620                  * The operation cannot be done if port is not started; it
 621                  * will be scheduled to be done during the next port start
 622                  */
 623                 port->mac_stats_reset_pending = B_TRUE;
 624                 return;
 625         }
 626
 627         rc = sfc_port_reset_mac_stats(sa);
 628         if (rc != 0)
 629                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 630 }
 631
 632 static int
 633 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 634                unsigned int xstats_count)
 635 {
 636         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 637         struct sfc_port *port = &sa->port;
 638         uint64_t *mac_stats;
 639         int rc;
 640         unsigned int i;
 641         int nstats = 0;
 642
 643         rte_spinlock_lock(&port->mac_stats_lock);
 644
 645         rc = sfc_port_update_mac_stats(sa);
 646         if (rc != 0) {
 647                 SFC_ASSERT(rc > 0);
 648                 nstats = -rc;
 649                 goto unlock;
 650         }
 651
 652         mac_stats = port->mac_stats_buf;
 653
 654         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 655                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 656                         if (xstats != NULL && nstats < (int)xstats_count) {
 657                                 xstats[nstats].id = nstats;
 658                                 xstats[nstats].value = mac_stats[i];
 659                         }
 660                         nstats++;
 661                 }
 662         }
 663
 664 unlock:
 665         rte_spinlock_unlock(&port->mac_stats_lock);
 666
 667         return nstats;
 668 }
 669
 670 static int
 671 sfc_xstats_get_names(struct rte_eth_dev *dev,
 672                      struct rte_eth_xstat_name *xstats_names,
 673                      unsigned int xstats_count)
 674 {
 675         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 676         struct sfc_port *port = &sa->port;
 677         unsigned int i;
 678         unsigned int nstats = 0;
 679
 680         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 681                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 682                         if (xstats_names != NULL && nstats < xstats_count)
 683                                 strlcpy(xstats_names[nstats].name,
 684                                         efx_mac_stat_name(sa->nic, i),
 685                                         sizeof(xstats_names[0].name));
 686                         nstats++;
 687                 }
 688         }
 689
 690         return nstats;
 691 }
 692
 693 static int
 694 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 695                      uint64_t *values, unsigned int n)
 696 {
 697         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 698         struct sfc_port *port = &sa->port;
 699         uint64_t *mac_stats;
 700         unsigned int nb_supported = 0;
 701         unsigned int nb_written = 0;
 702         unsigned int i;
 703         int ret;
 704         int rc;
 705
 706         if (unlikely(values == NULL) ||
 707             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 708                 return port->mac_stats_nb_supported;
 709
 710         rte_spinlock_lock(&port->mac_stats_lock);
 711
 712         rc = sfc_port_update_mac_stats(sa);
 713         if (rc != 0) {
 714                 SFC_ASSERT(rc > 0);
 715                 ret = -rc;
 716                 goto unlock;
 717         }
 718
 719         mac_stats = port->mac_stats_buf;
 720
 721         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 722                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 723                         continue;
 724
 725                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 726                         values[nb_written++] = mac_stats[i];
 727
 728                 ++nb_supported;
 729         }
 730
 731         ret = nb_written;
 732
 733 unlock:
 734         rte_spinlock_unlock(&port->mac_stats_lock);
 735
 736         return ret;
 737 }
 738
 739 static int
 740 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 741                            struct rte_eth_xstat_name *xstats_names,
 742                            const uint64_t *ids, unsigned int size)
 743 {
 744         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 745         struct sfc_port *port = &sa->port;
 746         unsigned int nb_supported = 0;
 747         unsigned int nb_written = 0;
 748         unsigned int i;
 749
 750         if (unlikely(xstats_names == NULL) ||
 751             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 752                 return port->mac_stats_nb_supported;
 753
 754         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 755                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 756                         continue;
 757
 758                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 759                         char *name = xstats_names[nb_written++].name;
 760
 761                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 762                                 sizeof(xstats_names[0].name));
 763                 }
 764
 765                 ++nb_supported;
 766         }
 767
 768         return nb_written;
 769 }
 770
 771 static int
 772 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 773 {
 774         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 775         unsigned int wanted_fc, link_fc;
 776
 777         memset(fc_conf, 0, sizeof(*fc_conf));
 778
 779         sfc_adapter_lock(sa);
 780
 781         if (sa->state == SFC_ADAPTER_STARTED)
 782                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 783         else
 784                 link_fc = sa->port.flow_ctrl;
 785
 786         switch (link_fc) {
 787         case 0:
 788                 fc_conf->mode = RTE_FC_NONE;
 789                 break;
 790         case EFX_FCNTL_RESPOND:
 791                 fc_conf->mode = RTE_FC_RX_PAUSE;
 792                 break;
 793         case EFX_FCNTL_GENERATE:
 794                 fc_conf->mode = RTE_FC_TX_PAUSE;
 795                 break;
 796         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 797                 fc_conf->mode = RTE_FC_FULL;
 798                 break;
 799         default:
 800                 sfc_err(sa, "%s: unexpected flow control value %#x",
 801                         __func__, link_fc);
 802         }
 803
 804         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 805
 806         sfc_adapter_unlock(sa);
 807
 808         return 0;
 809 }
 810
 811 static int
 812 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 813 {
 814         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 815         struct sfc_port *port = &sa->port;
 816         unsigned int fcntl;
 817         int rc;
 818
 819         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 820             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 821             fc_conf->mac_ctrl_frame_fwd != 0) {
 822                 sfc_err(sa, "unsupported flow control settings specified");
 823                 rc = EINVAL;
 824                 goto fail_inval;
 825         }
 826
 827         switch (fc_conf->mode) {
 828         case RTE_FC_NONE:
 829                 fcntl = 0;
 830                 break;
 831         case RTE_FC_RX_PAUSE:
 832                 fcntl = EFX_FCNTL_RESPOND;
 833                 break;
 834         case RTE_FC_TX_PAUSE:
 835                 fcntl = EFX_FCNTL_GENERATE;
 836                 break;
 837         case RTE_FC_FULL:
 838                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 839                 break;
 840         default:
 841                 rc = EINVAL;
 842                 goto fail_inval;
 843         }
 844
 845         sfc_adapter_lock(sa);
 846
 847         if (sa->state == SFC_ADAPTER_STARTED) {
 848                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 849                 if (rc != 0)
 850                         goto fail_mac_fcntl_set;
 851         }
 852
 853         port->flow_ctrl = fcntl;
 854         port->flow_ctrl_autoneg = fc_conf->autoneg;
 855
 856         sfc_adapter_unlock(sa);
 857
 858         return 0;
 859
 860 fail_mac_fcntl_set:
 861         sfc_adapter_unlock(sa);
 862 fail_inval:
 863         SFC_ASSERT(rc > 0);
 864         return -rc;
 865 }
 866
 867 static int
 868 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 869 {
 870         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 871         size_t pdu = EFX_MAC_PDU(mtu);
 872         size_t old_pdu;
 873         int rc;
 874
 875         sfc_log_init(sa, "mtu=%u", mtu);
 876
 877         rc = EINVAL;
 878         if (pdu < EFX_MAC_PDU_MIN) {
 879                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 880                         (unsigned int)mtu, (unsigned int)pdu,
 881                         EFX_MAC_PDU_MIN);
 882                 goto fail_inval;
 883         }
 884         if (pdu > EFX_MAC_PDU_MAX) {
 885                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 886                         (unsigned int)mtu, (unsigned int)pdu,
 887                         EFX_MAC_PDU_MAX);
 888                 goto fail_inval;
 889         }
 890
 891         sfc_adapter_lock(sa);
 892
 893         if (pdu != sa->port.pdu) {
 894                 if (sa->state == SFC_ADAPTER_STARTED) {
 895                         sfc_stop(sa);
 896
 897                         old_pdu = sa->port.pdu;
 898                         sa->port.pdu = pdu;
 899                         rc = sfc_start(sa);
 900                         if (rc != 0)
 901                                 goto fail_start;
 902                 } else {
 903                         sa->port.pdu = pdu;
 904                 }
 905         }
 906
 907         /*
 908          * The driver does not use it, but other PMDs update jumbo frame
 909          * flag and max_rx_pkt_len when MTU is set.
 910          */
 911         if (mtu > ETHER_MAX_LEN) {
 912                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 913                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
 914         }
 915
 916         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
 917
 918         sfc_adapter_unlock(sa);
 919
 920         sfc_log_init(sa, "done");
 921         return 0;
 922
 923 fail_start:
 924         sa->port.pdu = old_pdu;
 925         if (sfc_start(sa) != 0)
 926                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
 927                         "PDU max size - port is stopped",
 928                         (unsigned int)pdu, (unsigned int)old_pdu);
 929         sfc_adapter_unlock(sa);
 930
 931 fail_inval:
 932         sfc_log_init(sa, "failed %d", rc);
 933         SFC_ASSERT(rc > 0);
 934         return -rc;
 935 }
 936 static int
 937 sfc_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
 938 {
 939         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 940         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 941         struct sfc_port *port = &sa->port;
 942         struct ether_addr *old_addr = &dev->data->mac_addrs[0];
 943         int rc = 0;
 944
 945         sfc_adapter_lock(sa);
 946
 947         /*
 948          * Copy the address to the device private data so that
 949          * it could be recalled in the case of adapter restart.
 950          */
 951         ether_addr_copy(mac_addr, &port->default_mac_addr);
 952
 953         /*
 954          * Neither of the two following checks can return
 955          * an error. The new MAC address is preserved in
 956          * the device private data and can be activated
 957          * on the next port start if the user prevents
 958          * isolated mode from being enabled.
 959          */
 960         if (sfc_sa2shared(sa)->isolated) {
 961                 sfc_warn(sa, "isolated mode is active on the port");
 962                 sfc_warn(sa, "will not set MAC address");
 963                 goto unlock;
 964         }
 965
 966         if (sa->state != SFC_ADAPTER_STARTED) {
 967                 sfc_notice(sa, "the port is not started");
 968                 sfc_notice(sa, "the new MAC address will be set on port start");
 969
 970                 goto unlock;
 971         }
 972
 973         if (encp->enc_allow_set_mac_with_installed_filters) {
 974                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
 975                 if (rc != 0) {
 976                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
 977                         goto unlock;
 978                 }
 979
 980                 /*
 981                  * Changing the MAC address by means of MCDI request
 982                  * has no effect on received traffic, therefore
 983                  * we also need to update unicast filters
 984                  */
 985                 rc = sfc_set_rx_mode(sa);
 986                 if (rc != 0) {
 987                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
 988                         /* Rollback the old address */
 989                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
 990                         (void)sfc_set_rx_mode(sa);
 991                 }
 992         } else {
 993                 sfc_warn(sa, "cannot set MAC address with filters installed");
 994                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
 995                 sfc_warn(sa, "(some traffic may be dropped)");
 996
 997                 /*
 998                  * Since setting MAC address with filters installed is not
 999                  * allowed on the adapter, the new MAC address will be set
1000                  * by means of adapter restart. sfc_start() shall retrieve
1001                  * the new address from the device private data and set it.
1002                  */
1003                 sfc_stop(sa);
1004                 rc = sfc_start(sa);
1005                 if (rc != 0)
1006                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1007         }
1008
1009 unlock:
1010         if (rc != 0)
1011                 ether_addr_copy(old_addr, &port->default_mac_addr);
1012
1013         sfc_adapter_unlock(sa);
1014
1015         SFC_ASSERT(rc >= 0);
1016         return -rc;
1017 }
1018
1019
1020 static int
1021 sfc_set_mc_addr_list(struct rte_eth_dev *dev, struct ether_addr *mc_addr_set,
1022                      uint32_t nb_mc_addr)
1023 {
1024         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1025         struct sfc_port *port = &sa->port;
1026         uint8_t *mc_addrs = port->mcast_addrs;
1027         int rc;
1028         unsigned int i;
1029
1030         if (sfc_sa2shared(sa)->isolated) {
1031                 sfc_err(sa, "isolated mode is active on the port");
1032                 sfc_err(sa, "will not set multicast address list");
1033                 return -ENOTSUP;
1034         }
1035
1036         if (mc_addrs == NULL)
1037                 return -ENOBUFS;
1038
1039         if (nb_mc_addr > port->max_mcast_addrs) {
1040                 sfc_err(sa, "too many multicast addresses: %u > %u",
1041                          nb_mc_addr, port->max_mcast_addrs);
1042                 return -EINVAL;
1043         }
1044
1045         for (i = 0; i < nb_mc_addr; ++i) {
1046                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1047                                  EFX_MAC_ADDR_LEN);
1048                 mc_addrs += EFX_MAC_ADDR_LEN;
1049         }
1050
1051         port->nb_mcast_addrs = nb_mc_addr;
1052
1053         if (sa->state != SFC_ADAPTER_STARTED)
1054                 return 0;
1055
1056         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1057                                         port->nb_mcast_addrs);
1058         if (rc != 0)
1059                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1060
1061         SFC_ASSERT(rc >= 0);
1062         return -rc;
1063 }
1064
1065 /*
1066  * The function is used by the secondary process as well. It must not
1067  * use any process-local pointers from the adapter data.
1068  */
1069 static void
1070 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1071                       struct rte_eth_rxq_info *qinfo)
1072 {
1073         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1074         struct sfc_rxq_info *rxq_info;
1075
1076         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1077
1078         rxq_info = &sas->rxq_info[rx_queue_id];
1079
1080         qinfo->mp = rxq_info->refill_mb_pool;
1081         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1082         qinfo->conf.rx_drop_en = 1;
1083         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1084         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1085         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1086                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1087                 qinfo->scattered_rx = 1;
1088         }
1089         qinfo->nb_desc = rxq_info->entries;
1090 }
1091
1092 /*
1093  * The function is used by the secondary process as well. It must not
1094  * use any process-local pointers from the adapter data.
1095  */
1096 static void
1097 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1098                       struct rte_eth_txq_info *qinfo)
1099 {
1100         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1101         struct sfc_txq_info *txq_info;
1102
1103         SFC_ASSERT(tx_queue_id < sas->txq_count);
1104
1105         txq_info = &sas->txq_info[tx_queue_id];
1106
1107         memset(qinfo, 0, sizeof(*qinfo));
1108
1109         qinfo->conf.offloads = txq_info->offloads;
1110         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1111         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1112         qinfo->nb_desc = txq_info->entries;
1113 }
1114
1115 /*
1116  * The function is used by the secondary process as well. It must not
1117  * use any process-local pointers from the adapter data.
1118  */
1119 static uint32_t
1120 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1121 {
1122         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1123         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1124         struct sfc_rxq_info *rxq_info;
1125
1126         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1127         rxq_info = &sas->rxq_info[rx_queue_id];
1128
1129         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1130                 return 0;
1131
1132         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1133 }
1134
1135 /*
1136  * The function is used by the secondary process as well. It must not
1137  * use any process-local pointers from the adapter data.
1138  */
1139 static int
1140 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1141 {
1142         struct sfc_dp_rxq *dp_rxq = queue;
1143         const struct sfc_dp_rx *dp_rx;
1144
1145         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1146
1147         return offset < dp_rx->qdesc_npending(dp_rxq);
1148 }
1149
1150 /*
1151  * The function is used by the secondary process as well. It must not
1152  * use any process-local pointers from the adapter data.
1153  */
1154 static int
1155 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1156 {
1157         struct sfc_dp_rxq *dp_rxq = queue;
1158         const struct sfc_dp_rx *dp_rx;
1159
1160         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1161
1162         return dp_rx->qdesc_status(dp_rxq, offset);
1163 }
1164
1165 /*
1166  * The function is used by the secondary process as well. It must not
1167  * use any process-local pointers from the adapter data.
1168  */
1169 static int
1170 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1171 {
1172         struct sfc_dp_txq *dp_txq = queue;
1173         const struct sfc_dp_tx *dp_tx;
1174
1175         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1176
1177         return dp_tx->qdesc_status(dp_txq, offset);
1178 }
1179
1180 static int
1181 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1182 {
1183         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1184         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1185         int rc;
1186
1187         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1188
1189         sfc_adapter_lock(sa);
1190
1191         rc = EINVAL;
1192         if (sa->state != SFC_ADAPTER_STARTED)
1193                 goto fail_not_started;
1194
1195         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1196                 goto fail_not_setup;
1197
1198         rc = sfc_rx_qstart(sa, rx_queue_id);
1199         if (rc != 0)
1200                 goto fail_rx_qstart;
1201
1202         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1203
1204         sfc_adapter_unlock(sa);
1205
1206         return 0;
1207
1208 fail_rx_qstart:
1209 fail_not_setup:
1210 fail_not_started:
1211         sfc_adapter_unlock(sa);
1212         SFC_ASSERT(rc > 0);
1213         return -rc;
1214 }
1215
1216 static int
1217 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1218 {
1219         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1220         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1221
1222         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1223
1224         sfc_adapter_lock(sa);
1225         sfc_rx_qstop(sa, rx_queue_id);
1226
1227         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1228
1229         sfc_adapter_unlock(sa);
1230
1231         return 0;
1232 }
1233
1234 static int
1235 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1236 {
1237         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1238         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1239         int rc;
1240
1241         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1242
1243         sfc_adapter_lock(sa);
1244
1245         rc = EINVAL;
1246         if (sa->state != SFC_ADAPTER_STARTED)
1247                 goto fail_not_started;
1248
1249         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1250                 goto fail_not_setup;
1251
1252         rc = sfc_tx_qstart(sa, tx_queue_id);
1253         if (rc != 0)
1254                 goto fail_tx_qstart;
1255
1256         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1257
1258         sfc_adapter_unlock(sa);
1259         return 0;
1260
1261 fail_tx_qstart:
1262
1263 fail_not_setup:
1264 fail_not_started:
1265         sfc_adapter_unlock(sa);
1266         SFC_ASSERT(rc > 0);
1267         return -rc;
1268 }
1269
1270 static int
1271 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1272 {
1273         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1274         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1275
1276         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1277
1278         sfc_adapter_lock(sa);
1279
1280         sfc_tx_qstop(sa, tx_queue_id);
1281
1282         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1283
1284         sfc_adapter_unlock(sa);
1285         return 0;
1286 }
1287
1288 static efx_tunnel_protocol_t
1289 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1290 {
1291         switch (rte_type) {
1292         case RTE_TUNNEL_TYPE_VXLAN:
1293                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1294         case RTE_TUNNEL_TYPE_GENEVE:
1295                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1296         default:
1297                 return EFX_TUNNEL_NPROTOS;
1298         }
1299 }
1300
1301 enum sfc_udp_tunnel_op_e {
1302         SFC_UDP_TUNNEL_ADD_PORT,
1303         SFC_UDP_TUNNEL_DEL_PORT,
1304 };
1305
1306 static int
1307 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1308                       struct rte_eth_udp_tunnel *tunnel_udp,
1309                       enum sfc_udp_tunnel_op_e op)
1310 {
1311         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1312         efx_tunnel_protocol_t tunnel_proto;
1313         int rc;
1314
1315         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1316                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1317                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1318                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1319
1320         tunnel_proto =
1321                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1322         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1323                 rc = ENOTSUP;
1324                 goto fail_bad_proto;
1325         }
1326
1327         sfc_adapter_lock(sa);
1328
1329         switch (op) {
1330         case SFC_UDP_TUNNEL_ADD_PORT:
1331                 rc = efx_tunnel_config_udp_add(sa->nic,
1332                                                tunnel_udp->udp_port,
1333                                                tunnel_proto);
1334                 break;
1335         case SFC_UDP_TUNNEL_DEL_PORT:
1336                 rc = efx_tunnel_config_udp_remove(sa->nic,
1337                                                   tunnel_udp->udp_port,
1338                                                   tunnel_proto);
1339                 break;
1340         default:
1341                 rc = EINVAL;
1342                 goto fail_bad_op;
1343         }
1344
1345         if (rc != 0)
1346                 goto fail_op;
1347
1348         if (sa->state == SFC_ADAPTER_STARTED) {
1349                 rc = efx_tunnel_reconfigure(sa->nic);
1350                 if (rc == EAGAIN) {
1351                         /*
1352                          * Configuration is accepted by FW and MC reboot
1353                          * is initiated to apply the changes. MC reboot
1354                          * will be handled in a usual way (MC reboot
1355                          * event on management event queue and adapter
1356                          * restart).
1357                          */
1358                         rc = 0;
1359                 } else if (rc != 0) {
1360                         goto fail_reconfigure;
1361                 }
1362         }
1363
1364         sfc_adapter_unlock(sa);
1365         return 0;
1366
1367 fail_reconfigure:
1368         /* Remove/restore entry since the change makes the trouble */
1369         switch (op) {
1370         case SFC_UDP_TUNNEL_ADD_PORT:
1371                 (void)efx_tunnel_config_udp_remove(sa->nic,
1372                                                    tunnel_udp->udp_port,
1373                                                    tunnel_proto);
1374                 break;
1375         case SFC_UDP_TUNNEL_DEL_PORT:
1376                 (void)efx_tunnel_config_udp_add(sa->nic,
1377                                                 tunnel_udp->udp_port,
1378                                                 tunnel_proto);
1379                 break;
1380         }
1381
1382 fail_op:
1383 fail_bad_op:
1384         sfc_adapter_unlock(sa);
1385
1386 fail_bad_proto:
1387         SFC_ASSERT(rc > 0);
1388         return -rc;
1389 }
1390
1391 static int
1392 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1393                             struct rte_eth_udp_tunnel *tunnel_udp)
1394 {
1395         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1396 }
1397
1398 static int
1399 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1400                             struct rte_eth_udp_tunnel *tunnel_udp)
1401 {
1402         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1403 }
1404
1405 /*
1406  * The function is used by the secondary process as well. It must not
1407  * use any process-local pointers from the adapter data.
1408  */
1409 static int
1410 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1411                           struct rte_eth_rss_conf *rss_conf)
1412 {
1413         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1414         struct sfc_rss *rss = &sas->rss;
1415
1416         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1417                 return -ENOTSUP;
1418
1419         /*
1420          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1421          * hence, conversion is done here to derive a correct set of ETH_RSS
1422          * flags which corresponds to the active EFX configuration stored
1423          * locally in 'sfc_adapter' and kept up-to-date
1424          */
1425         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1426         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1427         if (rss_conf->rss_key != NULL)
1428                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1429
1430         return 0;
1431 }
1432
1433 static int
1434 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1435                         struct rte_eth_rss_conf *rss_conf)
1436 {
1437         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1438         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1439         unsigned int efx_hash_types;
1440         int rc = 0;
1441
1442         if (sfc_sa2shared(sa)->isolated)
1443                 return -ENOTSUP;
1444
1445         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1446                 sfc_err(sa, "RSS is not available");
1447                 return -ENOTSUP;
1448         }
1449
1450         if (rss->channels == 0) {
1451                 sfc_err(sa, "RSS is not configured");
1452                 return -EINVAL;
1453         }
1454
1455         if ((rss_conf->rss_key != NULL) &&
1456             (rss_conf->rss_key_len != sizeof(rss->key))) {
1457                 sfc_err(sa, "RSS key size is wrong (should be %lu)",
1458                         sizeof(rss->key));
1459                 return -EINVAL;
1460         }
1461
1462         sfc_adapter_lock(sa);
1463
1464         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1465         if (rc != 0)
1466                 goto fail_rx_hf_rte_to_efx;
1467
1468         rc = efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1469                                    rss->hash_alg, efx_hash_types, B_TRUE);
1470         if (rc != 0)
1471                 goto fail_scale_mode_set;
1472
1473         if (rss_conf->rss_key != NULL) {
1474                 if (sa->state == SFC_ADAPTER_STARTED) {
1475                         rc = efx_rx_scale_key_set(sa->nic,
1476                                                   EFX_RSS_CONTEXT_DEFAULT,
1477                                                   rss_conf->rss_key,
1478                                                   sizeof(rss->key));
1479                         if (rc != 0)
1480                                 goto fail_scale_key_set;
1481                 }
1482
1483                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1484         }
1485
1486         rss->hash_types = efx_hash_types;
1487
1488         sfc_adapter_unlock(sa);
1489
1490         return 0;
1491
1492 fail_scale_key_set:
1493         if (efx_rx_scale_mode_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1494                                   EFX_RX_HASHALG_TOEPLITZ,
1495                                   rss->hash_types, B_TRUE) != 0)
1496                 sfc_err(sa, "failed to restore RSS mode");
1497
1498 fail_scale_mode_set:
1499 fail_rx_hf_rte_to_efx:
1500         sfc_adapter_unlock(sa);
1501         return -rc;
1502 }
1503
1504 /*
1505  * The function is used by the secondary process as well. It must not
1506  * use any process-local pointers from the adapter data.
1507  */
1508 static int
1509 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1510                        struct rte_eth_rss_reta_entry64 *reta_conf,
1511                        uint16_t reta_size)
1512 {
1513         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1514         struct sfc_rss *rss = &sas->rss;
1515         int entry;
1516
1517         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1518                 return -ENOTSUP;
1519
1520         if (rss->channels == 0)
1521                 return -EINVAL;
1522
1523         if (reta_size != EFX_RSS_TBL_SIZE)
1524                 return -EINVAL;
1525
1526         for (entry = 0; entry < reta_size; entry++) {
1527                 int grp = entry / RTE_RETA_GROUP_SIZE;
1528                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1529
1530                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1531                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1532         }
1533
1534         return 0;
1535 }
1536
1537 static int
1538 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1539                         struct rte_eth_rss_reta_entry64 *reta_conf,
1540                         uint16_t reta_size)
1541 {
1542         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1543         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1544         unsigned int *rss_tbl_new;
1545         uint16_t entry;
1546         int rc = 0;
1547
1548
1549         if (sfc_sa2shared(sa)->isolated)
1550                 return -ENOTSUP;
1551
1552         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1553                 sfc_err(sa, "RSS is not available");
1554                 return -ENOTSUP;
1555         }
1556
1557         if (rss->channels == 0) {
1558                 sfc_err(sa, "RSS is not configured");
1559                 return -EINVAL;
1560         }
1561
1562         if (reta_size != EFX_RSS_TBL_SIZE) {
1563                 sfc_err(sa, "RETA size is wrong (should be %u)",
1564                         EFX_RSS_TBL_SIZE);
1565                 return -EINVAL;
1566         }
1567
1568         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1569         if (rss_tbl_new == NULL)
1570                 return -ENOMEM;
1571
1572         sfc_adapter_lock(sa);
1573
1574         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1575
1576         for (entry = 0; entry < reta_size; entry++) {
1577                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1578                 struct rte_eth_rss_reta_entry64 *grp;
1579
1580                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1581
1582                 if (grp->mask & (1ull << grp_idx)) {
1583                         if (grp->reta[grp_idx] >= rss->channels) {
1584                                 rc = EINVAL;
1585                                 goto bad_reta_entry;
1586                         }
1587                         rss_tbl_new[entry] = grp->reta[grp_idx];
1588                 }
1589         }
1590
1591         if (sa->state == SFC_ADAPTER_STARTED) {
1592                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1593                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1594                 if (rc != 0)
1595                         goto fail_scale_tbl_set;
1596         }
1597
1598         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1599
1600 fail_scale_tbl_set:
1601 bad_reta_entry:
1602         sfc_adapter_unlock(sa);
1603
1604         rte_free(rss_tbl_new);
1605
1606         SFC_ASSERT(rc >= 0);
1607         return -rc;
1608 }
1609
1610 static int
1611 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1612                     enum rte_filter_op filter_op,
1613                     void *arg)
1614 {
1615         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1616         int rc = ENOTSUP;
1617
1618         sfc_log_init(sa, "entry");
1619
1620         switch (filter_type) {
1621         case RTE_ETH_FILTER_NONE:
1622                 sfc_err(sa, "Global filters configuration not supported");
1623                 break;
1624         case RTE_ETH_FILTER_MACVLAN:
1625                 sfc_err(sa, "MACVLAN filters not supported");
1626                 break;
1627         case RTE_ETH_FILTER_ETHERTYPE:
1628                 sfc_err(sa, "EtherType filters not supported");
1629                 break;
1630         case RTE_ETH_FILTER_FLEXIBLE:
1631                 sfc_err(sa, "Flexible filters not supported");
1632                 break;
1633         case RTE_ETH_FILTER_SYN:
1634                 sfc_err(sa, "SYN filters not supported");
1635                 break;
1636         case RTE_ETH_FILTER_NTUPLE:
1637                 sfc_err(sa, "NTUPLE filters not supported");
1638                 break;
1639         case RTE_ETH_FILTER_TUNNEL:
1640                 sfc_err(sa, "Tunnel filters not supported");
1641                 break;
1642         case RTE_ETH_FILTER_FDIR:
1643                 sfc_err(sa, "Flow Director filters not supported");
1644                 break;
1645         case RTE_ETH_FILTER_HASH:
1646                 sfc_err(sa, "Hash filters not supported");
1647                 break;
1648         case RTE_ETH_FILTER_GENERIC:
1649                 if (filter_op != RTE_ETH_FILTER_GET) {
1650                         rc = EINVAL;
1651                 } else {
1652                         *(const void **)arg = &sfc_flow_ops;
1653                         rc = 0;
1654                 }
1655                 break;
1656         default:
1657                 sfc_err(sa, "Unknown filter type %u", filter_type);
1658                 break;
1659         }
1660
1661         sfc_log_init(sa, "exit: %d", -rc);
1662         SFC_ASSERT(rc >= 0);
1663         return -rc;
1664 }
1665
1666 static int
1667 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1668 {
1669         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1670
1671         /*
1672          * If Rx datapath does not provide callback to check mempool,
1673          * all pools are supported.
1674          */
1675         if (sap->dp_rx->pool_ops_supported == NULL)
1676                 return 1;
1677
1678         return sap->dp_rx->pool_ops_supported(pool);
1679 }
1680
1681 static const struct eth_dev_ops sfc_eth_dev_ops = {
1682         .dev_configure                  = sfc_dev_configure,
1683         .dev_start                      = sfc_dev_start,
1684         .dev_stop                       = sfc_dev_stop,
1685         .dev_set_link_up                = sfc_dev_set_link_up,
1686         .dev_set_link_down              = sfc_dev_set_link_down,
1687         .dev_close                      = sfc_dev_close,
1688         .promiscuous_enable             = sfc_dev_promisc_enable,
1689         .promiscuous_disable            = sfc_dev_promisc_disable,
1690         .allmulticast_enable            = sfc_dev_allmulti_enable,
1691         .allmulticast_disable           = sfc_dev_allmulti_disable,
1692         .link_update                    = sfc_dev_link_update,
1693         .stats_get                      = sfc_stats_get,
1694         .stats_reset                    = sfc_stats_reset,
1695         .xstats_get                     = sfc_xstats_get,
1696         .xstats_reset                   = sfc_stats_reset,
1697         .xstats_get_names               = sfc_xstats_get_names,
1698         .dev_infos_get                  = sfc_dev_infos_get,
1699         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1700         .mtu_set                        = sfc_dev_set_mtu,
1701         .rx_queue_start                 = sfc_rx_queue_start,
1702         .rx_queue_stop                  = sfc_rx_queue_stop,
1703         .tx_queue_start                 = sfc_tx_queue_start,
1704         .tx_queue_stop                  = sfc_tx_queue_stop,
1705         .rx_queue_setup                 = sfc_rx_queue_setup,
1706         .rx_queue_release               = sfc_rx_queue_release,
1707         .rx_queue_count                 = sfc_rx_queue_count,
1708         .rx_descriptor_done             = sfc_rx_descriptor_done,
1709         .rx_descriptor_status           = sfc_rx_descriptor_status,
1710         .tx_descriptor_status           = sfc_tx_descriptor_status,
1711         .tx_queue_setup                 = sfc_tx_queue_setup,
1712         .tx_queue_release               = sfc_tx_queue_release,
1713         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1714         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1715         .mac_addr_set                   = sfc_mac_addr_set,
1716         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1717         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1718         .reta_update                    = sfc_dev_rss_reta_update,
1719         .reta_query                     = sfc_dev_rss_reta_query,
1720         .rss_hash_update                = sfc_dev_rss_hash_update,
1721         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1722         .filter_ctrl                    = sfc_dev_filter_ctrl,
1723         .set_mc_addr_list               = sfc_set_mc_addr_list,
1724         .rxq_info_get                   = sfc_rx_queue_info_get,
1725         .txq_info_get                   = sfc_tx_queue_info_get,
1726         .fw_version_get                 = sfc_fw_version_get,
1727         .xstats_get_by_id               = sfc_xstats_get_by_id,
1728         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1729         .pool_ops_supported             = sfc_pool_ops_supported,
1730 };
1731
1732 /**
1733  * Duplicate a string in potentially shared memory required for
1734  * multi-process support.
1735  *
1736  * strdup() allocates from process-local heap/memory.
1737  */
1738 static char *
1739 sfc_strdup(const char *str)
1740 {
1741         size_t size;
1742         char *copy;
1743
1744         if (str == NULL)
1745                 return NULL;
1746
1747         size = strlen(str) + 1;
1748         copy = rte_malloc(__func__, size, 0);
1749         if (copy != NULL)
1750                 rte_memcpy(copy, str, size);
1751
1752         return copy;
1753 }
1754
1755 static int
1756 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1757 {
1758         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1759         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1760         const struct sfc_dp_rx *dp_rx;
1761         const struct sfc_dp_tx *dp_tx;
1762         const efx_nic_cfg_t *encp;
1763         unsigned int avail_caps = 0;
1764         const char *rx_name = NULL;
1765         const char *tx_name = NULL;
1766         int rc;
1767
1768         switch (sa->family) {
1769         case EFX_FAMILY_HUNTINGTON:
1770         case EFX_FAMILY_MEDFORD:
1771         case EFX_FAMILY_MEDFORD2:
1772                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1773                 break;
1774         default:
1775                 break;
1776         }
1777
1778         encp = efx_nic_cfg_get(sa->nic);
1779         if (encp->enc_rx_es_super_buffer_supported)
1780                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1781
1782         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1783                                 sfc_kvarg_string_handler, &rx_name);
1784         if (rc != 0)
1785                 goto fail_kvarg_rx_datapath;
1786
1787         if (rx_name != NULL) {
1788                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1789                 if (dp_rx == NULL) {
1790                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1791                         rc = ENOENT;
1792                         goto fail_dp_rx;
1793                 }
1794                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1795                         sfc_err(sa,
1796                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1797                                 rx_name);
1798                         rc = EINVAL;
1799                         goto fail_dp_rx_caps;
1800                 }
1801         } else {
1802                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1803                 if (dp_rx == NULL) {
1804                         sfc_err(sa, "Rx datapath by caps %#x not found",
1805                                 avail_caps);
1806                         rc = ENOENT;
1807                         goto fail_dp_rx;
1808                 }
1809         }
1810
1811         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1812         if (sas->dp_rx_name == NULL) {
1813                 rc = ENOMEM;
1814                 goto fail_dp_rx_name;
1815         }
1816
1817         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1818
1819         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1820                                 sfc_kvarg_string_handler, &tx_name);
1821         if (rc != 0)
1822                 goto fail_kvarg_tx_datapath;
1823
1824         if (tx_name != NULL) {
1825                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1826                 if (dp_tx == NULL) {
1827                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1828                         rc = ENOENT;
1829                         goto fail_dp_tx;
1830                 }
1831                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1832                         sfc_err(sa,
1833                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1834                                 tx_name);
1835                         rc = EINVAL;
1836                         goto fail_dp_tx_caps;
1837                 }
1838         } else {
1839                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1840                 if (dp_tx == NULL) {
1841                         sfc_err(sa, "Tx datapath by caps %#x not found",
1842                                 avail_caps);
1843                         rc = ENOENT;
1844                         goto fail_dp_tx;
1845                 }
1846         }
1847
1848         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
1849         if (sas->dp_tx_name == NULL) {
1850                 rc = ENOMEM;
1851                 goto fail_dp_tx_name;
1852         }
1853
1854         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
1855
1856         sa->priv.dp_rx = dp_rx;
1857         sa->priv.dp_tx = dp_tx;
1858
1859         dev->rx_pkt_burst = dp_rx->pkt_burst;
1860         dev->tx_pkt_burst = dp_tx->pkt_burst;
1861
1862         dev->dev_ops = &sfc_eth_dev_ops;
1863
1864         return 0;
1865
1866 fail_dp_tx_name:
1867 fail_dp_tx_caps:
1868 fail_dp_tx:
1869 fail_kvarg_tx_datapath:
1870         rte_free(sas->dp_rx_name);
1871         sas->dp_rx_name = NULL;
1872
1873 fail_dp_rx_name:
1874 fail_dp_rx_caps:
1875 fail_dp_rx:
1876 fail_kvarg_rx_datapath:
1877         return rc;
1878 }
1879
1880 static void
1881 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
1882 {
1883         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1884         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1885
1886         dev->dev_ops = NULL;
1887         dev->rx_pkt_burst = NULL;
1888         dev->tx_pkt_burst = NULL;
1889
1890         rte_free(sas->dp_tx_name);
1891         sas->dp_tx_name = NULL;
1892         sa->priv.dp_tx = NULL;
1893
1894         rte_free(sas->dp_rx_name);
1895         sas->dp_rx_name = NULL;
1896         sa->priv.dp_rx = NULL;
1897 }
1898
1899 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
1900         .rx_queue_count                 = sfc_rx_queue_count,
1901         .rx_descriptor_done             = sfc_rx_descriptor_done,
1902         .rx_descriptor_status           = sfc_rx_descriptor_status,
1903         .tx_descriptor_status           = sfc_tx_descriptor_status,
1904         .reta_query                     = sfc_dev_rss_reta_query,
1905         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1906         .rxq_info_get                   = sfc_rx_queue_info_get,
1907         .txq_info_get                   = sfc_tx_queue_info_get,
1908 };
1909
1910 static int
1911 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
1912 {
1913         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1914         struct sfc_adapter_priv *sap;
1915         const struct sfc_dp_rx *dp_rx;
1916         const struct sfc_dp_tx *dp_tx;
1917         int rc;
1918
1919         /*
1920          * Allocate process private data from heap, since it should not
1921          * be located in shared memory allocated using rte_malloc() API.
1922          */
1923         sap = calloc(1, sizeof(*sap));
1924         if (sap == NULL) {
1925                 rc = ENOMEM;
1926                 goto fail_alloc_priv;
1927         }
1928
1929         sap->logtype_main = logtype_main;
1930
1931         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
1932         if (dp_rx == NULL) {
1933                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1934                         "cannot find %s Rx datapath", sas->dp_rx_name);
1935                 rc = ENOENT;
1936                 goto fail_dp_rx;
1937         }
1938         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
1939                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1940                         "%s Rx datapath does not support multi-process",
1941                         sas->dp_rx_name);
1942                 rc = EINVAL;
1943                 goto fail_dp_rx_multi_process;
1944         }
1945
1946         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
1947         if (dp_tx == NULL) {
1948                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1949                         "cannot find %s Tx datapath", sas->dp_tx_name);
1950                 rc = ENOENT;
1951                 goto fail_dp_tx;
1952         }
1953         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
1954                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
1955                         "%s Tx datapath does not support multi-process",
1956                         sas->dp_tx_name);
1957                 rc = EINVAL;
1958                 goto fail_dp_tx_multi_process;
1959         }
1960
1961         sap->dp_rx = dp_rx;
1962         sap->dp_tx = dp_tx;
1963
1964         dev->process_private = sap;
1965         dev->rx_pkt_burst = dp_rx->pkt_burst;
1966         dev->tx_pkt_burst = dp_tx->pkt_burst;
1967         dev->dev_ops = &sfc_eth_dev_secondary_ops;
1968
1969         return 0;
1970
1971 fail_dp_tx_multi_process:
1972 fail_dp_tx:
1973 fail_dp_rx_multi_process:
1974 fail_dp_rx:
1975         free(sap);
1976
1977 fail_alloc_priv:
1978         return rc;
1979 }
1980
1981 static void
1982 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
1983 {
1984         free(dev->process_private);
1985         dev->process_private = NULL;
1986         dev->dev_ops = NULL;
1987         dev->tx_pkt_burst = NULL;
1988         dev->rx_pkt_burst = NULL;
1989 }
1990
1991 static void
1992 sfc_register_dp(void)
1993 {
1994         /* Register once */
1995         if (TAILQ_EMPTY(&sfc_dp_head)) {
1996                 /* Prefer EF10 datapath */
1997                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
1998                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
1999                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2000
2001                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2002                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2003                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2004         }
2005 }
2006
2007 static int
2008 sfc_eth_dev_init(struct rte_eth_dev *dev)
2009 {
2010         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2011         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2012         uint32_t logtype_main;
2013         struct sfc_adapter *sa;
2014         int rc;
2015         const efx_nic_cfg_t *encp;
2016         const struct ether_addr *from;
2017
2018         sfc_register_dp();
2019
2020         logtype_main = sfc_register_logtype(&pci_dev->addr,
2021                                             SFC_LOGTYPE_MAIN_STR,
2022                                             RTE_LOG_NOTICE);
2023
2024         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2025                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2026
2027         /* Required for logging */
2028         sas->pci_addr = pci_dev->addr;
2029         sas->port_id = dev->data->port_id;
2030
2031         /*
2032          * Allocate process private data from heap, since it should not
2033          * be located in shared memory allocated using rte_malloc() API.
2034          */
2035         sa = calloc(1, sizeof(*sa));
2036         if (sa == NULL) {
2037                 rc = ENOMEM;
2038                 goto fail_alloc_sa;
2039         }
2040
2041         dev->process_private = sa;
2042
2043         /* Required for logging */
2044         sa->priv.shared = sas;
2045         sa->priv.logtype_main = logtype_main;
2046
2047         sa->eth_dev = dev;
2048
2049         /* Copy PCI device info to the dev->data */
2050         rte_eth_copy_pci_info(dev, pci_dev);
2051
2052         rc = sfc_kvargs_parse(sa);
2053         if (rc != 0)
2054                 goto fail_kvargs_parse;
2055
2056         sfc_log_init(sa, "entry");
2057
2058         dev->data->mac_addrs = rte_zmalloc("sfc", ETHER_ADDR_LEN, 0);
2059         if (dev->data->mac_addrs == NULL) {
2060                 rc = ENOMEM;
2061                 goto fail_mac_addrs;
2062         }
2063
2064         sfc_adapter_lock_init(sa);
2065         sfc_adapter_lock(sa);
2066
2067         sfc_log_init(sa, "probing");
2068         rc = sfc_probe(sa);
2069         if (rc != 0)
2070                 goto fail_probe;
2071
2072         sfc_log_init(sa, "set device ops");
2073         rc = sfc_eth_dev_set_ops(dev);
2074         if (rc != 0)
2075                 goto fail_set_ops;
2076
2077         sfc_log_init(sa, "attaching");
2078         rc = sfc_attach(sa);
2079         if (rc != 0)
2080                 goto fail_attach;
2081
2082         encp = efx_nic_cfg_get(sa->nic);
2083
2084         /*
2085          * The arguments are really reverse order in comparison to
2086          * Linux kernel. Copy from NIC config to Ethernet device data.
2087          */
2088         from = (const struct ether_addr *)(encp->enc_mac_addr);
2089         ether_addr_copy(from, &dev->data->mac_addrs[0]);
2090
2091         sfc_adapter_unlock(sa);
2092
2093         sfc_log_init(sa, "done");
2094         return 0;
2095
2096 fail_attach:
2097         sfc_eth_dev_clear_ops(dev);
2098
2099 fail_set_ops:
2100         sfc_unprobe(sa);
2101
2102 fail_probe:
2103         sfc_adapter_unlock(sa);
2104         sfc_adapter_lock_fini(sa);
2105         rte_free(dev->data->mac_addrs);
2106         dev->data->mac_addrs = NULL;
2107
2108 fail_mac_addrs:
2109         sfc_kvargs_cleanup(sa);
2110
2111 fail_kvargs_parse:
2112         sfc_log_init(sa, "failed %d", rc);
2113         dev->process_private = NULL;
2114         free(sa);
2115
2116 fail_alloc_sa:
2117         SFC_ASSERT(rc > 0);
2118         return -rc;
2119 }
2120
2121 static int
2122 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2123 {
2124         struct sfc_adapter *sa;
2125
2126         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
2127                 sfc_eth_dev_secondary_clear_ops(dev);
2128                 return 0;
2129         }
2130
2131         sa = sfc_adapter_by_eth_dev(dev);
2132         sfc_log_init(sa, "entry");
2133
2134         sfc_adapter_lock(sa);
2135
2136         sfc_eth_dev_clear_ops(dev);
2137
2138         sfc_detach(sa);
2139         sfc_unprobe(sa);
2140
2141         sfc_kvargs_cleanup(sa);
2142
2143         sfc_adapter_unlock(sa);
2144         sfc_adapter_lock_fini(sa);
2145
2146         sfc_log_init(sa, "done");
2147
2148         /* Required for logging, so cleanup last */
2149         sa->eth_dev = NULL;
2150
2151         dev->process_private = NULL;
2152         free(sa);
2153
2154         return 0;
2155 }
2156
2157 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2158         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2159         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2160         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2161         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2162         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2163         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2164         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2165         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2166         { .vendor_id = 0 /* sentinel */ }
2167 };
2168
2169 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2170         struct rte_pci_device *pci_dev)
2171 {
2172         return rte_eth_dev_pci_generic_probe(pci_dev,
2173                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2174 }
2175
2176 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2177 {
2178         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2179 }
2180
2181 static struct rte_pci_driver sfc_efx_pmd = {
2182         .id_table = pci_id_sfc_efx_map,
2183         .drv_flags =
2184                 RTE_PCI_DRV_INTR_LSC |
2185                 RTE_PCI_DRV_NEED_MAPPING,
2186         .probe = sfc_eth_dev_pci_probe,
2187         .remove = sfc_eth_dev_pci_remove,
2188 };
2189
2190 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2191 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2192 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2193 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2194         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2195         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2196         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2197         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2198         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2199         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2200
2201 RTE_INIT(sfc_driver_register_logtype)
2202 {
2203         int ret;
2204
2205         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2206                                                    RTE_LOG_NOTICE);
2207         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2208 }