drivers/net/sfc/sfc_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright(c) 2019-2020 Xilinx, Inc.
   4  * Copyright(c) 2016-2019 Solarflare Communications Inc.
   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 #include <rte_ether.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_debug.h"
  23 #include "sfc_log.h"
  24 #include "sfc_kvargs.h"
  25 #include "sfc_ev.h"
  26 #include "sfc_rx.h"
  27 #include "sfc_tx.h"
  28 #include "sfc_flow.h"
  29 #include "sfc_dp.h"
  30 #include "sfc_dp_rx.h"
  31
  32 uint32_t sfc_logtype_driver;
  33
  34 static struct sfc_dp_list sfc_dp_head =
  35         TAILQ_HEAD_INITIALIZER(sfc_dp_head);
  36
  37
  38 static void sfc_eth_dev_clear_ops(struct rte_eth_dev *dev);
  39
  40
  41 static int
  42 sfc_fw_version_get(struct rte_eth_dev *dev, char *fw_version, size_t fw_size)
  43 {
  44         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  45         efx_nic_fw_info_t enfi;
  46         int ret;
  47         int rc;
  48
  49         /*
  50          * Return value of the callback is likely supposed to be
  51          * equal to or greater than 0, nevertheless, if an error
  52          * occurs, it will be desirable to pass it to the caller
  53          */
  54         if ((fw_version == NULL) || (fw_size == 0))
  55                 return -EINVAL;
  56
  57         rc = efx_nic_get_fw_version(sa->nic, &enfi);
  58         if (rc != 0)
  59                 return -rc;
  60
  61         ret = snprintf(fw_version, fw_size,
  62                        "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16,
  63                        enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1],
  64                        enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]);
  65         if (ret < 0)
  66                 return ret;
  67
  68         if (enfi.enfi_dpcpu_fw_ids_valid) {
  69                 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret);
  70                 int ret_extra;
  71
  72                 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset,
  73                                      fw_size - dpcpu_fw_ids_offset,
  74                                      " rx%" PRIx16 " tx%" PRIx16,
  75                                      enfi.enfi_rx_dpcpu_fw_id,
  76                                      enfi.enfi_tx_dpcpu_fw_id);
  77                 if (ret_extra < 0)
  78                         return ret_extra;
  79
  80                 ret += ret_extra;
  81         }
  82
  83         if (fw_size < (size_t)(++ret))
  84                 return ret;
  85         else
  86                 return 0;
  87 }
  88
  89 static int
  90 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
  91 {
  92         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
  93         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
  94         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
  95         struct sfc_rss *rss = &sas->rss;
  96         uint64_t txq_offloads_def = 0;
  97
  98         sfc_log_init(sa, "entry");
  99
 100         dev_info->min_mtu = RTE_ETHER_MIN_MTU;
 101         dev_info->max_mtu = EFX_MAC_SDU_MAX;
 102
 103         dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX;
 104
 105         dev_info->max_vfs = sa->sriov.num_vfs;
 106
 107         /* Autonegotiation may be disabled */
 108         dev_info->speed_capa = ETH_LINK_SPEED_FIXED;
 109         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX))
 110                 dev_info->speed_capa |= ETH_LINK_SPEED_1G;
 111         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX))
 112                 dev_info->speed_capa |= ETH_LINK_SPEED_10G;
 113         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX))
 114                 dev_info->speed_capa |= ETH_LINK_SPEED_25G;
 115         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX))
 116                 dev_info->speed_capa |= ETH_LINK_SPEED_40G;
 117         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX))
 118                 dev_info->speed_capa |= ETH_LINK_SPEED_50G;
 119         if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX))
 120                 dev_info->speed_capa |= ETH_LINK_SPEED_100G;
 121
 122         dev_info->max_rx_queues = sa->rxq_max;
 123         dev_info->max_tx_queues = sa->txq_max;
 124
 125         /* By default packets are dropped if no descriptors are available */
 126         dev_info->default_rxconf.rx_drop_en = 1;
 127
 128         dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa);
 129
 130         /*
 131          * rx_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->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) |
 136                                     dev_info->rx_queue_offload_capa;
 137
 138         dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa);
 139
 140         /*
 141          * tx_offload_capa includes both device and queue offloads since
 142          * the latter may be requested on a per device basis which makes
 143          * sense when some offloads are needed to be set on all queues.
 144          */
 145         dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) |
 146                                     dev_info->tx_queue_offload_capa;
 147
 148         if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 149                 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 150
 151         dev_info->default_txconf.offloads |= txq_offloads_def;
 152
 153         if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) {
 154                 uint64_t rte_hf = 0;
 155                 unsigned int i;
 156
 157                 for (i = 0; i < rss->hf_map_nb_entries; ++i)
 158                         rte_hf |= rss->hf_map[i].rte;
 159
 160                 dev_info->reta_size = EFX_RSS_TBL_SIZE;
 161                 dev_info->hash_key_size = EFX_RSS_KEY_SIZE;
 162                 dev_info->flow_type_rss_offloads = rte_hf;
 163         }
 164
 165         /* Initialize to hardware limits */
 166         dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries;
 167         dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries;
 168         /* The RXQ hardware requires that the descriptor count is a power
 169          * of 2, but rx_desc_lim cannot properly describe that constraint.
 170          */
 171         dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries;
 172
 173         /* Initialize to hardware limits */
 174         dev_info->tx_desc_lim.nb_max = sa->txq_max_entries;
 175         dev_info->tx_desc_lim.nb_min = sa->txq_min_entries;
 176         /*
 177          * The TXQ hardware requires that the descriptor count is a power
 178          * of 2, but tx_desc_lim cannot properly describe that constraint
 179          */
 180         dev_info->tx_desc_lim.nb_align = sa->txq_min_entries;
 181
 182         if (sap->dp_rx->get_dev_info != NULL)
 183                 sap->dp_rx->get_dev_info(dev_info);
 184         if (sap->dp_tx->get_dev_info != NULL)
 185                 sap->dp_tx->get_dev_info(dev_info);
 186
 187         dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP |
 188                              RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP;
 189
 190         return 0;
 191 }
 192
 193 static const uint32_t *
 194 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 195 {
 196         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
 197
 198         return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps);
 199 }
 200
 201 static int
 202 sfc_dev_configure(struct rte_eth_dev *dev)
 203 {
 204         struct rte_eth_dev_data *dev_data = dev->data;
 205         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 206         int rc;
 207
 208         sfc_log_init(sa, "entry n_rxq=%u n_txq=%u",
 209                      dev_data->nb_rx_queues, dev_data->nb_tx_queues);
 210
 211         sfc_adapter_lock(sa);
 212         switch (sa->state) {
 213         case SFC_ADAPTER_CONFIGURED:
 214                 /* FALLTHROUGH */
 215         case SFC_ADAPTER_INITIALIZED:
 216                 rc = sfc_configure(sa);
 217                 break;
 218         default:
 219                 sfc_err(sa, "unexpected adapter state %u to configure",
 220                         sa->state);
 221                 rc = EINVAL;
 222                 break;
 223         }
 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_start(struct rte_eth_dev *dev)
 233 {
 234         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 235         int rc;
 236
 237         sfc_log_init(sa, "entry");
 238
 239         sfc_adapter_lock(sa);
 240         rc = sfc_start(sa);
 241         sfc_adapter_unlock(sa);
 242
 243         sfc_log_init(sa, "done %d", rc);
 244         SFC_ASSERT(rc >= 0);
 245         return -rc;
 246 }
 247
 248 static int
 249 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 250 {
 251         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 252         struct rte_eth_link current_link;
 253         int ret;
 254
 255         sfc_log_init(sa, "entry");
 256
 257         if (sa->state != SFC_ADAPTER_STARTED) {
 258                 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, &current_link);
 259         } else if (wait_to_complete) {
 260                 efx_link_mode_t link_mode;
 261
 262                 if (efx_port_poll(sa->nic, &link_mode) != 0)
 263                         link_mode = EFX_LINK_UNKNOWN;
 264                 sfc_port_link_mode_to_info(link_mode, &current_link);
 265
 266         } else {
 267                 sfc_ev_mgmt_qpoll(sa);
 268                 rte_eth_linkstatus_get(dev, &current_link);
 269         }
 270
 271         ret = rte_eth_linkstatus_set(dev, &current_link);
 272         if (ret == 0)
 273                 sfc_notice(sa, "Link status is %s",
 274                            current_link.link_status ? "UP" : "DOWN");
 275
 276         return ret;
 277 }
 278
 279 static void
 280 sfc_dev_stop(struct rte_eth_dev *dev)
 281 {
 282         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 283
 284         sfc_log_init(sa, "entry");
 285
 286         sfc_adapter_lock(sa);
 287         sfc_stop(sa);
 288         sfc_adapter_unlock(sa);
 289
 290         sfc_log_init(sa, "done");
 291 }
 292
 293 static int
 294 sfc_dev_set_link_up(struct rte_eth_dev *dev)
 295 {
 296         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 297         int rc;
 298
 299         sfc_log_init(sa, "entry");
 300
 301         sfc_adapter_lock(sa);
 302         rc = sfc_start(sa);
 303         sfc_adapter_unlock(sa);
 304
 305         SFC_ASSERT(rc >= 0);
 306         return -rc;
 307 }
 308
 309 static int
 310 sfc_dev_set_link_down(struct rte_eth_dev *dev)
 311 {
 312         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 313
 314         sfc_log_init(sa, "entry");
 315
 316         sfc_adapter_lock(sa);
 317         sfc_stop(sa);
 318         sfc_adapter_unlock(sa);
 319
 320         return 0;
 321 }
 322
 323 static void
 324 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev)
 325 {
 326         free(dev->process_private);
 327         rte_eth_dev_release_port(dev);
 328 }
 329
 330 static int
 331 sfc_dev_close(struct rte_eth_dev *dev)
 332 {
 333         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 334
 335         sfc_log_init(sa, "entry");
 336
 337         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 338                 sfc_eth_dev_secondary_clear_ops(dev);
 339                 return 0;
 340         }
 341
 342         sfc_adapter_lock(sa);
 343         switch (sa->state) {
 344         case SFC_ADAPTER_STARTED:
 345                 sfc_stop(sa);
 346                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 347                 /* FALLTHROUGH */
 348         case SFC_ADAPTER_CONFIGURED:
 349                 sfc_close(sa);
 350                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 351                 /* FALLTHROUGH */
 352         case SFC_ADAPTER_INITIALIZED:
 353                 break;
 354         default:
 355                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 356                 break;
 357         }
 358
 359         /*
 360          * Cleanup all resources.
 361          * Rollback primary process sfc_eth_dev_init() below.
 362          */
 363
 364         sfc_eth_dev_clear_ops(dev);
 365
 366         sfc_detach(sa);
 367         sfc_unprobe(sa);
 368
 369         sfc_kvargs_cleanup(sa);
 370
 371         sfc_adapter_unlock(sa);
 372         sfc_adapter_lock_fini(sa);
 373
 374         sfc_log_init(sa, "done");
 375
 376         /* Required for logging, so cleanup last */
 377         sa->eth_dev = NULL;
 378
 379         free(sa);
 380
 381         return 0;
 382 }
 383
 384 static int
 385 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 386                    boolean_t enabled)
 387 {
 388         struct sfc_port *port;
 389         boolean_t *toggle;
 390         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 391         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 392         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 393         int rc = 0;
 394
 395         sfc_adapter_lock(sa);
 396
 397         port = &sa->port;
 398         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 399
 400         if (*toggle != enabled) {
 401                 *toggle = enabled;
 402
 403                 if (sfc_sa2shared(sa)->isolated) {
 404                         sfc_warn(sa, "isolated mode is active on the port");
 405                         sfc_warn(sa, "the change is to be applied on the next "
 406                                      "start provided that isolated mode is "
 407                                      "disabled prior the next start");
 408                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 409                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 410                         *toggle = !(enabled);
 411                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 412                                  ((enabled) ? "enable" : "disable"), desc, rc);
 413
 414                         /*
 415                          * For promiscuous and all-multicast filters a
 416                          * permission failure should be reported as an
 417                          * unsupported filter.
 418                          */
 419                         if (rc == EPERM)
 420                                 rc = ENOTSUP;
 421                 }
 422         }
 423
 424         sfc_adapter_unlock(sa);
 425         return rc;
 426 }
 427
 428 static int
 429 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 430 {
 431         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 432
 433         SFC_ASSERT(rc >= 0);
 434         return -rc;
 435 }
 436
 437 static int
 438 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 439 {
 440         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 441
 442         SFC_ASSERT(rc >= 0);
 443         return -rc;
 444 }
 445
 446 static int
 447 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 448 {
 449         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 450
 451         SFC_ASSERT(rc >= 0);
 452         return -rc;
 453 }
 454
 455 static int
 456 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 457 {
 458         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 459
 460         SFC_ASSERT(rc >= 0);
 461         return -rc;
 462 }
 463
 464 static int
 465 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 466                    uint16_t nb_rx_desc, unsigned int socket_id,
 467                    const struct rte_eth_rxconf *rx_conf,
 468                    struct rte_mempool *mb_pool)
 469 {
 470         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 471         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 472         int rc;
 473
 474         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 475                      rx_queue_id, nb_rx_desc, socket_id);
 476
 477         sfc_adapter_lock(sa);
 478
 479         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 480                           rx_conf, mb_pool);
 481         if (rc != 0)
 482                 goto fail_rx_qinit;
 483
 484         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 485
 486         sfc_adapter_unlock(sa);
 487
 488         return 0;
 489
 490 fail_rx_qinit:
 491         sfc_adapter_unlock(sa);
 492         SFC_ASSERT(rc > 0);
 493         return -rc;
 494 }
 495
 496 static void
 497 sfc_rx_queue_release(void *queue)
 498 {
 499         struct sfc_dp_rxq *dp_rxq = queue;
 500         struct sfc_rxq *rxq;
 501         struct sfc_adapter *sa;
 502         unsigned int sw_index;
 503
 504         if (dp_rxq == NULL)
 505                 return;
 506
 507         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 508         sa = rxq->evq->sa;
 509         sfc_adapter_lock(sa);
 510
 511         sw_index = dp_rxq->dpq.queue_id;
 512
 513         sfc_log_init(sa, "RxQ=%u", sw_index);
 514
 515         sfc_rx_qfini(sa, sw_index);
 516
 517         sfc_adapter_unlock(sa);
 518 }
 519
 520 static int
 521 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 522                    uint16_t nb_tx_desc, unsigned int socket_id,
 523                    const struct rte_eth_txconf *tx_conf)
 524 {
 525         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 526         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 527         int rc;
 528
 529         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 530                      tx_queue_id, nb_tx_desc, socket_id);
 531
 532         sfc_adapter_lock(sa);
 533
 534         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 535         if (rc != 0)
 536                 goto fail_tx_qinit;
 537
 538         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
 539
 540         sfc_adapter_unlock(sa);
 541         return 0;
 542
 543 fail_tx_qinit:
 544         sfc_adapter_unlock(sa);
 545         SFC_ASSERT(rc > 0);
 546         return -rc;
 547 }
 548
 549 static void
 550 sfc_tx_queue_release(void *queue)
 551 {
 552         struct sfc_dp_txq *dp_txq = queue;
 553         struct sfc_txq *txq;
 554         unsigned int sw_index;
 555         struct sfc_adapter *sa;
 556
 557         if (dp_txq == NULL)
 558                 return;
 559
 560         txq = sfc_txq_by_dp_txq(dp_txq);
 561         sw_index = dp_txq->dpq.queue_id;
 562
 563         SFC_ASSERT(txq->evq != NULL);
 564         sa = txq->evq->sa;
 565
 566         sfc_log_init(sa, "TxQ = %u", sw_index);
 567
 568         sfc_adapter_lock(sa);
 569
 570         sfc_tx_qfini(sa, sw_index);
 571
 572         sfc_adapter_unlock(sa);
 573 }
 574
 575 /*
 576  * Some statistics are computed as A - B where A and B each increase
 577  * monotonically with some hardware counter(s) and the counters are read
 578  * asynchronously.
 579  *
 580  * If packet X is counted in A, but not counted in B yet, computed value is
 581  * greater than real.
 582  *
 583  * If packet X is not counted in A at the moment of reading the counter,
 584  * but counted in B at the moment of reading the counter, computed value
 585  * is less than real.
 586  *
 587  * However, counter which grows backward is worse evil than slightly wrong
 588  * value. So, let's try to guarantee that it never happens except may be
 589  * the case when the MAC stats are zeroed as a result of a NIC reset.
 590  */
 591 static void
 592 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 593 {
 594         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 595                 *stat = newval;
 596 }
 597
 598 static int
 599 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 600 {
 601         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 602         struct sfc_port *port = &sa->port;
 603         uint64_t *mac_stats;
 604         int ret;
 605
 606         rte_spinlock_lock(&port->mac_stats_lock);
 607
 608         ret = sfc_port_update_mac_stats(sa);
 609         if (ret != 0)
 610                 goto unlock;
 611
 612         mac_stats = port->mac_stats_buf;
 613
 614         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 615                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 616                 stats->ipackets =
 617                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 618                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 619                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 620                 stats->opackets =
 621                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 622                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 623                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 624                 stats->ibytes =
 625                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 626                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 627                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 628                 stats->obytes =
 629                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 630                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 631                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 632                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 633                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 634         } else {
 635                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 636                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 637                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 638                 /*
 639                  * Take into account stats which are whenever supported
 640                  * on EF10. If some stat is not supported by current
 641                  * firmware variant or HW revision, it is guaranteed
 642                  * to be zero in mac_stats.
 643                  */
 644                 stats->imissed =
 645                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 646                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 647                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 648                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 649                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 650                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 651                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 652                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 653                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 654                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 655                 stats->ierrors =
 656                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 657                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 658                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 659                 /* no oerrors counters supported on EF10 */
 660
 661                 /* Exclude missed, errors and pauses from Rx packets */
 662                 sfc_update_diff_stat(&port->ipackets,
 663                         mac_stats[EFX_MAC_RX_PKTS] -
 664                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 665                         stats->imissed - stats->ierrors);
 666                 stats->ipackets = port->ipackets;
 667         }
 668
 669 unlock:
 670         rte_spinlock_unlock(&port->mac_stats_lock);
 671         SFC_ASSERT(ret >= 0);
 672         return -ret;
 673 }
 674
 675 static int
 676 sfc_stats_reset(struct rte_eth_dev *dev)
 677 {
 678         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 679         struct sfc_port *port = &sa->port;
 680         int rc;
 681
 682         if (sa->state != SFC_ADAPTER_STARTED) {
 683                 /*
 684                  * The operation cannot be done if port is not started; it
 685                  * will be scheduled to be done during the next port start
 686                  */
 687                 port->mac_stats_reset_pending = B_TRUE;
 688                 return 0;
 689         }
 690
 691         rc = sfc_port_reset_mac_stats(sa);
 692         if (rc != 0)
 693                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 694
 695         SFC_ASSERT(rc >= 0);
 696         return -rc;
 697 }
 698
 699 static int
 700 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 701                unsigned int xstats_count)
 702 {
 703         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 704         struct sfc_port *port = &sa->port;
 705         uint64_t *mac_stats;
 706         int rc;
 707         unsigned int i;
 708         int nstats = 0;
 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                 nstats = -rc;
 716                 goto unlock;
 717         }
 718
 719         mac_stats = port->mac_stats_buf;
 720
 721         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 722                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 723                         if (xstats != NULL && nstats < (int)xstats_count) {
 724                                 xstats[nstats].id = nstats;
 725                                 xstats[nstats].value = mac_stats[i];
 726                         }
 727                         nstats++;
 728                 }
 729         }
 730
 731 unlock:
 732         rte_spinlock_unlock(&port->mac_stats_lock);
 733
 734         return nstats;
 735 }
 736
 737 static int
 738 sfc_xstats_get_names(struct rte_eth_dev *dev,
 739                      struct rte_eth_xstat_name *xstats_names,
 740                      unsigned int xstats_count)
 741 {
 742         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 743         struct sfc_port *port = &sa->port;
 744         unsigned int i;
 745         unsigned int nstats = 0;
 746
 747         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 748                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 749                         if (xstats_names != NULL && nstats < xstats_count)
 750                                 strlcpy(xstats_names[nstats].name,
 751                                         efx_mac_stat_name(sa->nic, i),
 752                                         sizeof(xstats_names[0].name));
 753                         nstats++;
 754                 }
 755         }
 756
 757         return nstats;
 758 }
 759
 760 static int
 761 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 762                      uint64_t *values, unsigned int n)
 763 {
 764         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 765         struct sfc_port *port = &sa->port;
 766         uint64_t *mac_stats;
 767         unsigned int nb_supported = 0;
 768         unsigned int nb_written = 0;
 769         unsigned int i;
 770         int ret;
 771         int rc;
 772
 773         if (unlikely(values == NULL) ||
 774             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 775                 return port->mac_stats_nb_supported;
 776
 777         rte_spinlock_lock(&port->mac_stats_lock);
 778
 779         rc = sfc_port_update_mac_stats(sa);
 780         if (rc != 0) {
 781                 SFC_ASSERT(rc > 0);
 782                 ret = -rc;
 783                 goto unlock;
 784         }
 785
 786         mac_stats = port->mac_stats_buf;
 787
 788         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 789                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 790                         continue;
 791
 792                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 793                         values[nb_written++] = mac_stats[i];
 794
 795                 ++nb_supported;
 796         }
 797
 798         ret = nb_written;
 799
 800 unlock:
 801         rte_spinlock_unlock(&port->mac_stats_lock);
 802
 803         return ret;
 804 }
 805
 806 static int
 807 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 808                            struct rte_eth_xstat_name *xstats_names,
 809                            const uint64_t *ids, unsigned int size)
 810 {
 811         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 812         struct sfc_port *port = &sa->port;
 813         unsigned int nb_supported = 0;
 814         unsigned int nb_written = 0;
 815         unsigned int i;
 816
 817         if (unlikely(xstats_names == NULL) ||
 818             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 819                 return port->mac_stats_nb_supported;
 820
 821         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 822                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 823                         continue;
 824
 825                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 826                         char *name = xstats_names[nb_written++].name;
 827
 828                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 829                                 sizeof(xstats_names[0].name));
 830                 }
 831
 832                 ++nb_supported;
 833         }
 834
 835         return nb_written;
 836 }
 837
 838 static int
 839 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 840 {
 841         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 842         unsigned int wanted_fc, link_fc;
 843
 844         memset(fc_conf, 0, sizeof(*fc_conf));
 845
 846         sfc_adapter_lock(sa);
 847
 848         if (sa->state == SFC_ADAPTER_STARTED)
 849                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 850         else
 851                 link_fc = sa->port.flow_ctrl;
 852
 853         switch (link_fc) {
 854         case 0:
 855                 fc_conf->mode = RTE_FC_NONE;
 856                 break;
 857         case EFX_FCNTL_RESPOND:
 858                 fc_conf->mode = RTE_FC_RX_PAUSE;
 859                 break;
 860         case EFX_FCNTL_GENERATE:
 861                 fc_conf->mode = RTE_FC_TX_PAUSE;
 862                 break;
 863         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 864                 fc_conf->mode = RTE_FC_FULL;
 865                 break;
 866         default:
 867                 sfc_err(sa, "%s: unexpected flow control value %#x",
 868                         __func__, link_fc);
 869         }
 870
 871         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 872
 873         sfc_adapter_unlock(sa);
 874
 875         return 0;
 876 }
 877
 878 static int
 879 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 880 {
 881         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 882         struct sfc_port *port = &sa->port;
 883         unsigned int fcntl;
 884         int rc;
 885
 886         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 887             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 888             fc_conf->mac_ctrl_frame_fwd != 0) {
 889                 sfc_err(sa, "unsupported flow control settings specified");
 890                 rc = EINVAL;
 891                 goto fail_inval;
 892         }
 893
 894         switch (fc_conf->mode) {
 895         case RTE_FC_NONE:
 896                 fcntl = 0;
 897                 break;
 898         case RTE_FC_RX_PAUSE:
 899                 fcntl = EFX_FCNTL_RESPOND;
 900                 break;
 901         case RTE_FC_TX_PAUSE:
 902                 fcntl = EFX_FCNTL_GENERATE;
 903                 break;
 904         case RTE_FC_FULL:
 905                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 906                 break;
 907         default:
 908                 rc = EINVAL;
 909                 goto fail_inval;
 910         }
 911
 912         sfc_adapter_lock(sa);
 913
 914         if (sa->state == SFC_ADAPTER_STARTED) {
 915                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 916                 if (rc != 0)
 917                         goto fail_mac_fcntl_set;
 918         }
 919
 920         port->flow_ctrl = fcntl;
 921         port->flow_ctrl_autoneg = fc_conf->autoneg;
 922
 923         sfc_adapter_unlock(sa);
 924
 925         return 0;
 926
 927 fail_mac_fcntl_set:
 928         sfc_adapter_unlock(sa);
 929 fail_inval:
 930         SFC_ASSERT(rc > 0);
 931         return -rc;
 932 }
 933
 934 static int
 935 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 936 {
 937         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 938         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 939         boolean_t scatter_enabled;
 940         const char *error;
 941         unsigned int i;
 942
 943         for (i = 0; i < sas->rxq_count; i++) {
 944                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 945                         continue;
 946
 947                 scatter_enabled = (sas->rxq_info[i].type_flags &
 948                                    EFX_RXQ_FLAG_SCATTER);
 949
 950                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 951                                           encp->enc_rx_prefix_size,
 952                                           scatter_enabled,
 953                                           encp->enc_rx_scatter_max, &error)) {
 954                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 955                                 error);
 956                         return EINVAL;
 957                 }
 958         }
 959
 960         return 0;
 961 }
 962
 963 static int
 964 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 965 {
 966         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 967         size_t pdu = EFX_MAC_PDU(mtu);
 968         size_t old_pdu;
 969         int rc;
 970
 971         sfc_log_init(sa, "mtu=%u", mtu);
 972
 973         rc = EINVAL;
 974         if (pdu < EFX_MAC_PDU_MIN) {
 975                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 976                         (unsigned int)mtu, (unsigned int)pdu,
 977                         EFX_MAC_PDU_MIN);
 978                 goto fail_inval;
 979         }
 980         if (pdu > EFX_MAC_PDU_MAX) {
 981                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 982                         (unsigned int)mtu, (unsigned int)pdu,
 983                         (unsigned int)EFX_MAC_PDU_MAX);
 984                 goto fail_inval;
 985         }
 986
 987         sfc_adapter_lock(sa);
 988
 989         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
 990         if (rc != 0)
 991                 goto fail_check_scatter;
 992
 993         if (pdu != sa->port.pdu) {
 994                 if (sa->state == SFC_ADAPTER_STARTED) {
 995                         sfc_stop(sa);
 996
 997                         old_pdu = sa->port.pdu;
 998                         sa->port.pdu = pdu;
 999                         rc = sfc_start(sa);
1000                         if (rc != 0)
1001                                 goto fail_start;
1002                 } else {
1003                         sa->port.pdu = pdu;
1004                 }
1005         }
1006
1007         /*
1008          * The driver does not use it, but other PMDs update jumbo frame
1009          * flag and max_rx_pkt_len when MTU is set.
1010          */
1011         if (mtu > RTE_ETHER_MAX_LEN) {
1012                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1013                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1014         }
1015
1016         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1017
1018         sfc_adapter_unlock(sa);
1019
1020         sfc_log_init(sa, "done");
1021         return 0;
1022
1023 fail_start:
1024         sa->port.pdu = old_pdu;
1025         if (sfc_start(sa) != 0)
1026                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1027                         "PDU max size - port is stopped",
1028                         (unsigned int)pdu, (unsigned int)old_pdu);
1029
1030 fail_check_scatter:
1031         sfc_adapter_unlock(sa);
1032
1033 fail_inval:
1034         sfc_log_init(sa, "failed %d", rc);
1035         SFC_ASSERT(rc > 0);
1036         return -rc;
1037 }
1038 static int
1039 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1040 {
1041         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1042         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1043         struct sfc_port *port = &sa->port;
1044         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1045         int rc = 0;
1046
1047         sfc_adapter_lock(sa);
1048
1049         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1050                 goto unlock;
1051
1052         /*
1053          * Copy the address to the device private data so that
1054          * it could be recalled in the case of adapter restart.
1055          */
1056         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1057
1058         /*
1059          * Neither of the two following checks can return
1060          * an error. The new MAC address is preserved in
1061          * the device private data and can be activated
1062          * on the next port start if the user prevents
1063          * isolated mode from being enabled.
1064          */
1065         if (sfc_sa2shared(sa)->isolated) {
1066                 sfc_warn(sa, "isolated mode is active on the port");
1067                 sfc_warn(sa, "will not set MAC address");
1068                 goto unlock;
1069         }
1070
1071         if (sa->state != SFC_ADAPTER_STARTED) {
1072                 sfc_notice(sa, "the port is not started");
1073                 sfc_notice(sa, "the new MAC address will be set on port start");
1074
1075                 goto unlock;
1076         }
1077
1078         if (encp->enc_allow_set_mac_with_installed_filters) {
1079                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1080                 if (rc != 0) {
1081                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1082                         goto unlock;
1083                 }
1084
1085                 /*
1086                  * Changing the MAC address by means of MCDI request
1087                  * has no effect on received traffic, therefore
1088                  * we also need to update unicast filters
1089                  */
1090                 rc = sfc_set_rx_mode_unchecked(sa);
1091                 if (rc != 0) {
1092                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1093                         /* Rollback the old address */
1094                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1095                         (void)sfc_set_rx_mode_unchecked(sa);
1096                 }
1097         } else {
1098                 sfc_warn(sa, "cannot set MAC address with filters installed");
1099                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1100                 sfc_warn(sa, "(some traffic may be dropped)");
1101
1102                 /*
1103                  * Since setting MAC address with filters installed is not
1104                  * allowed on the adapter, the new MAC address will be set
1105                  * by means of adapter restart. sfc_start() shall retrieve
1106                  * the new address from the device private data and set it.
1107                  */
1108                 sfc_stop(sa);
1109                 rc = sfc_start(sa);
1110                 if (rc != 0)
1111                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1112         }
1113
1114 unlock:
1115         if (rc != 0)
1116                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1117
1118         sfc_adapter_unlock(sa);
1119
1120         SFC_ASSERT(rc >= 0);
1121         return -rc;
1122 }
1123
1124
1125 static int
1126 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1127                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1128 {
1129         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1130         struct sfc_port *port = &sa->port;
1131         uint8_t *mc_addrs = port->mcast_addrs;
1132         int rc;
1133         unsigned int i;
1134
1135         if (sfc_sa2shared(sa)->isolated) {
1136                 sfc_err(sa, "isolated mode is active on the port");
1137                 sfc_err(sa, "will not set multicast address list");
1138                 return -ENOTSUP;
1139         }
1140
1141         if (mc_addrs == NULL)
1142                 return -ENOBUFS;
1143
1144         if (nb_mc_addr > port->max_mcast_addrs) {
1145                 sfc_err(sa, "too many multicast addresses: %u > %u",
1146                          nb_mc_addr, port->max_mcast_addrs);
1147                 return -EINVAL;
1148         }
1149
1150         for (i = 0; i < nb_mc_addr; ++i) {
1151                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1152                                  EFX_MAC_ADDR_LEN);
1153                 mc_addrs += EFX_MAC_ADDR_LEN;
1154         }
1155
1156         port->nb_mcast_addrs = nb_mc_addr;
1157
1158         if (sa->state != SFC_ADAPTER_STARTED)
1159                 return 0;
1160
1161         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1162                                         port->nb_mcast_addrs);
1163         if (rc != 0)
1164                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1165
1166         SFC_ASSERT(rc >= 0);
1167         return -rc;
1168 }
1169
1170 /*
1171  * The function is used by the secondary process as well. It must not
1172  * use any process-local pointers from the adapter data.
1173  */
1174 static void
1175 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1176                       struct rte_eth_rxq_info *qinfo)
1177 {
1178         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1179         struct sfc_rxq_info *rxq_info;
1180
1181         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1182
1183         rxq_info = &sas->rxq_info[rx_queue_id];
1184
1185         qinfo->mp = rxq_info->refill_mb_pool;
1186         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1187         qinfo->conf.rx_drop_en = 1;
1188         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1189         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1190         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1191                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1192                 qinfo->scattered_rx = 1;
1193         }
1194         qinfo->nb_desc = rxq_info->entries;
1195 }
1196
1197 /*
1198  * The function is used by the secondary process as well. It must not
1199  * use any process-local pointers from the adapter data.
1200  */
1201 static void
1202 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1203                       struct rte_eth_txq_info *qinfo)
1204 {
1205         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1206         struct sfc_txq_info *txq_info;
1207
1208         SFC_ASSERT(tx_queue_id < sas->txq_count);
1209
1210         txq_info = &sas->txq_info[tx_queue_id];
1211
1212         memset(qinfo, 0, sizeof(*qinfo));
1213
1214         qinfo->conf.offloads = txq_info->offloads;
1215         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1216         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1217         qinfo->nb_desc = txq_info->entries;
1218 }
1219
1220 /*
1221  * The function is used by the secondary process as well. It must not
1222  * use any process-local pointers from the adapter data.
1223  */
1224 static uint32_t
1225 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1226 {
1227         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1228         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1229         struct sfc_rxq_info *rxq_info;
1230
1231         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1232         rxq_info = &sas->rxq_info[rx_queue_id];
1233
1234         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1235                 return 0;
1236
1237         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1238 }
1239
1240 /*
1241  * The function is used by the secondary process as well. It must not
1242  * use any process-local pointers from the adapter data.
1243  */
1244 static int
1245 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1246 {
1247         struct sfc_dp_rxq *dp_rxq = queue;
1248         const struct sfc_dp_rx *dp_rx;
1249
1250         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1251
1252         return offset < dp_rx->qdesc_npending(dp_rxq);
1253 }
1254
1255 /*
1256  * The function is used by the secondary process as well. It must not
1257  * use any process-local pointers from the adapter data.
1258  */
1259 static int
1260 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1261 {
1262         struct sfc_dp_rxq *dp_rxq = queue;
1263         const struct sfc_dp_rx *dp_rx;
1264
1265         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1266
1267         return dp_rx->qdesc_status(dp_rxq, offset);
1268 }
1269
1270 /*
1271  * The function is used by the secondary process as well. It must not
1272  * use any process-local pointers from the adapter data.
1273  */
1274 static int
1275 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1276 {
1277         struct sfc_dp_txq *dp_txq = queue;
1278         const struct sfc_dp_tx *dp_tx;
1279
1280         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1281
1282         return dp_tx->qdesc_status(dp_txq, offset);
1283 }
1284
1285 static int
1286 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1287 {
1288         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1289         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1290         int rc;
1291
1292         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1293
1294         sfc_adapter_lock(sa);
1295
1296         rc = EINVAL;
1297         if (sa->state != SFC_ADAPTER_STARTED)
1298                 goto fail_not_started;
1299
1300         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1301                 goto fail_not_setup;
1302
1303         rc = sfc_rx_qstart(sa, rx_queue_id);
1304         if (rc != 0)
1305                 goto fail_rx_qstart;
1306
1307         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1308
1309         sfc_adapter_unlock(sa);
1310
1311         return 0;
1312
1313 fail_rx_qstart:
1314 fail_not_setup:
1315 fail_not_started:
1316         sfc_adapter_unlock(sa);
1317         SFC_ASSERT(rc > 0);
1318         return -rc;
1319 }
1320
1321 static int
1322 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1323 {
1324         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1325         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1326
1327         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1328
1329         sfc_adapter_lock(sa);
1330         sfc_rx_qstop(sa, rx_queue_id);
1331
1332         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1333
1334         sfc_adapter_unlock(sa);
1335
1336         return 0;
1337 }
1338
1339 static int
1340 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1341 {
1342         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1343         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1344         int rc;
1345
1346         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1347
1348         sfc_adapter_lock(sa);
1349
1350         rc = EINVAL;
1351         if (sa->state != SFC_ADAPTER_STARTED)
1352                 goto fail_not_started;
1353
1354         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1355                 goto fail_not_setup;
1356
1357         rc = sfc_tx_qstart(sa, tx_queue_id);
1358         if (rc != 0)
1359                 goto fail_tx_qstart;
1360
1361         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1362
1363         sfc_adapter_unlock(sa);
1364         return 0;
1365
1366 fail_tx_qstart:
1367
1368 fail_not_setup:
1369 fail_not_started:
1370         sfc_adapter_unlock(sa);
1371         SFC_ASSERT(rc > 0);
1372         return -rc;
1373 }
1374
1375 static int
1376 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1377 {
1378         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1379         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1380
1381         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1382
1383         sfc_adapter_lock(sa);
1384
1385         sfc_tx_qstop(sa, tx_queue_id);
1386
1387         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1388
1389         sfc_adapter_unlock(sa);
1390         return 0;
1391 }
1392
1393 static efx_tunnel_protocol_t
1394 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1395 {
1396         switch (rte_type) {
1397         case RTE_TUNNEL_TYPE_VXLAN:
1398                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1399         case RTE_TUNNEL_TYPE_GENEVE:
1400                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1401         default:
1402                 return EFX_TUNNEL_NPROTOS;
1403         }
1404 }
1405
1406 enum sfc_udp_tunnel_op_e {
1407         SFC_UDP_TUNNEL_ADD_PORT,
1408         SFC_UDP_TUNNEL_DEL_PORT,
1409 };
1410
1411 static int
1412 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1413                       struct rte_eth_udp_tunnel *tunnel_udp,
1414                       enum sfc_udp_tunnel_op_e op)
1415 {
1416         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1417         efx_tunnel_protocol_t tunnel_proto;
1418         int rc;
1419
1420         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1421                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1422                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1423                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1424
1425         tunnel_proto =
1426                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1427         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1428                 rc = ENOTSUP;
1429                 goto fail_bad_proto;
1430         }
1431
1432         sfc_adapter_lock(sa);
1433
1434         switch (op) {
1435         case SFC_UDP_TUNNEL_ADD_PORT:
1436                 rc = efx_tunnel_config_udp_add(sa->nic,
1437                                                tunnel_udp->udp_port,
1438                                                tunnel_proto);
1439                 break;
1440         case SFC_UDP_TUNNEL_DEL_PORT:
1441                 rc = efx_tunnel_config_udp_remove(sa->nic,
1442                                                   tunnel_udp->udp_port,
1443                                                   tunnel_proto);
1444                 break;
1445         default:
1446                 rc = EINVAL;
1447                 goto fail_bad_op;
1448         }
1449
1450         if (rc != 0)
1451                 goto fail_op;
1452
1453         if (sa->state == SFC_ADAPTER_STARTED) {
1454                 rc = efx_tunnel_reconfigure(sa->nic);
1455                 if (rc == EAGAIN) {
1456                         /*
1457                          * Configuration is accepted by FW and MC reboot
1458                          * is initiated to apply the changes. MC reboot
1459                          * will be handled in a usual way (MC reboot
1460                          * event on management event queue and adapter
1461                          * restart).
1462                          */
1463                         rc = 0;
1464                 } else if (rc != 0) {
1465                         goto fail_reconfigure;
1466                 }
1467         }
1468
1469         sfc_adapter_unlock(sa);
1470         return 0;
1471
1472 fail_reconfigure:
1473         /* Remove/restore entry since the change makes the trouble */
1474         switch (op) {
1475         case SFC_UDP_TUNNEL_ADD_PORT:
1476                 (void)efx_tunnel_config_udp_remove(sa->nic,
1477                                                    tunnel_udp->udp_port,
1478                                                    tunnel_proto);
1479                 break;
1480         case SFC_UDP_TUNNEL_DEL_PORT:
1481                 (void)efx_tunnel_config_udp_add(sa->nic,
1482                                                 tunnel_udp->udp_port,
1483                                                 tunnel_proto);
1484                 break;
1485         }
1486
1487 fail_op:
1488 fail_bad_op:
1489         sfc_adapter_unlock(sa);
1490
1491 fail_bad_proto:
1492         SFC_ASSERT(rc > 0);
1493         return -rc;
1494 }
1495
1496 static int
1497 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1498                             struct rte_eth_udp_tunnel *tunnel_udp)
1499 {
1500         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1501 }
1502
1503 static int
1504 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1505                             struct rte_eth_udp_tunnel *tunnel_udp)
1506 {
1507         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1508 }
1509
1510 /*
1511  * The function is used by the secondary process as well. It must not
1512  * use any process-local pointers from the adapter data.
1513  */
1514 static int
1515 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1516                           struct rte_eth_rss_conf *rss_conf)
1517 {
1518         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1519         struct sfc_rss *rss = &sas->rss;
1520
1521         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1522                 return -ENOTSUP;
1523
1524         /*
1525          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1526          * hence, conversion is done here to derive a correct set of ETH_RSS
1527          * flags which corresponds to the active EFX configuration stored
1528          * locally in 'sfc_adapter' and kept up-to-date
1529          */
1530         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1531         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1532         if (rss_conf->rss_key != NULL)
1533                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1534
1535         return 0;
1536 }
1537
1538 static int
1539 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1540                         struct rte_eth_rss_conf *rss_conf)
1541 {
1542         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1543         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1544         unsigned int efx_hash_types;
1545         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1546         unsigned int n_contexts;
1547         unsigned int mode_i = 0;
1548         unsigned int key_i = 0;
1549         unsigned int i = 0;
1550         int rc = 0;
1551
1552         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1553
1554         if (sfc_sa2shared(sa)->isolated)
1555                 return -ENOTSUP;
1556
1557         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1558                 sfc_err(sa, "RSS is not available");
1559                 return -ENOTSUP;
1560         }
1561
1562         if (rss->channels == 0) {
1563                 sfc_err(sa, "RSS is not configured");
1564                 return -EINVAL;
1565         }
1566
1567         if ((rss_conf->rss_key != NULL) &&
1568             (rss_conf->rss_key_len != sizeof(rss->key))) {
1569                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1570                         sizeof(rss->key));
1571                 return -EINVAL;
1572         }
1573
1574         sfc_adapter_lock(sa);
1575
1576         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1577         if (rc != 0)
1578                 goto fail_rx_hf_rte_to_efx;
1579
1580         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1581                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1582                                            rss->hash_alg, efx_hash_types,
1583                                            B_TRUE);
1584                 if (rc != 0)
1585                         goto fail_scale_mode_set;
1586         }
1587
1588         if (rss_conf->rss_key != NULL) {
1589                 if (sa->state == SFC_ADAPTER_STARTED) {
1590                         for (key_i = 0; key_i < n_contexts; key_i++) {
1591                                 rc = efx_rx_scale_key_set(sa->nic,
1592                                                           contexts[key_i],
1593                                                           rss_conf->rss_key,
1594                                                           sizeof(rss->key));
1595                                 if (rc != 0)
1596                                         goto fail_scale_key_set;
1597                         }
1598                 }
1599
1600                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1601         }
1602
1603         rss->hash_types = efx_hash_types;
1604
1605         sfc_adapter_unlock(sa);
1606
1607         return 0;
1608
1609 fail_scale_key_set:
1610         for (i = 0; i < key_i; i++) {
1611                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1612                                          sizeof(rss->key)) != 0)
1613                         sfc_err(sa, "failed to restore RSS key");
1614         }
1615
1616 fail_scale_mode_set:
1617         for (i = 0; i < mode_i; i++) {
1618                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1619                                           EFX_RX_HASHALG_TOEPLITZ,
1620                                           rss->hash_types, B_TRUE) != 0)
1621                         sfc_err(sa, "failed to restore RSS mode");
1622         }
1623
1624 fail_rx_hf_rte_to_efx:
1625         sfc_adapter_unlock(sa);
1626         return -rc;
1627 }
1628
1629 /*
1630  * The function is used by the secondary process as well. It must not
1631  * use any process-local pointers from the adapter data.
1632  */
1633 static int
1634 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1635                        struct rte_eth_rss_reta_entry64 *reta_conf,
1636                        uint16_t reta_size)
1637 {
1638         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1639         struct sfc_rss *rss = &sas->rss;
1640         int entry;
1641
1642         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1643                 return -ENOTSUP;
1644
1645         if (rss->channels == 0)
1646                 return -EINVAL;
1647
1648         if (reta_size != EFX_RSS_TBL_SIZE)
1649                 return -EINVAL;
1650
1651         for (entry = 0; entry < reta_size; entry++) {
1652                 int grp = entry / RTE_RETA_GROUP_SIZE;
1653                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1654
1655                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1656                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1657         }
1658
1659         return 0;
1660 }
1661
1662 static int
1663 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1664                         struct rte_eth_rss_reta_entry64 *reta_conf,
1665                         uint16_t reta_size)
1666 {
1667         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1668         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1669         unsigned int *rss_tbl_new;
1670         uint16_t entry;
1671         int rc = 0;
1672
1673
1674         if (sfc_sa2shared(sa)->isolated)
1675                 return -ENOTSUP;
1676
1677         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1678                 sfc_err(sa, "RSS is not available");
1679                 return -ENOTSUP;
1680         }
1681
1682         if (rss->channels == 0) {
1683                 sfc_err(sa, "RSS is not configured");
1684                 return -EINVAL;
1685         }
1686
1687         if (reta_size != EFX_RSS_TBL_SIZE) {
1688                 sfc_err(sa, "RETA size is wrong (should be %u)",
1689                         EFX_RSS_TBL_SIZE);
1690                 return -EINVAL;
1691         }
1692
1693         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1694         if (rss_tbl_new == NULL)
1695                 return -ENOMEM;
1696
1697         sfc_adapter_lock(sa);
1698
1699         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1700
1701         for (entry = 0; entry < reta_size; entry++) {
1702                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1703                 struct rte_eth_rss_reta_entry64 *grp;
1704
1705                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1706
1707                 if (grp->mask & (1ull << grp_idx)) {
1708                         if (grp->reta[grp_idx] >= rss->channels) {
1709                                 rc = EINVAL;
1710                                 goto bad_reta_entry;
1711                         }
1712                         rss_tbl_new[entry] = grp->reta[grp_idx];
1713                 }
1714         }
1715
1716         if (sa->state == SFC_ADAPTER_STARTED) {
1717                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1718                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1719                 if (rc != 0)
1720                         goto fail_scale_tbl_set;
1721         }
1722
1723         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1724
1725 fail_scale_tbl_set:
1726 bad_reta_entry:
1727         sfc_adapter_unlock(sa);
1728
1729         rte_free(rss_tbl_new);
1730
1731         SFC_ASSERT(rc >= 0);
1732         return -rc;
1733 }
1734
1735 static int
1736 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1737                     enum rte_filter_op filter_op,
1738                     void *arg)
1739 {
1740         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1741         int rc = ENOTSUP;
1742
1743         sfc_log_init(sa, "entry");
1744
1745         switch (filter_type) {
1746         case RTE_ETH_FILTER_NONE:
1747                 sfc_err(sa, "Global filters configuration not supported");
1748                 break;
1749         case RTE_ETH_FILTER_MACVLAN:
1750                 sfc_err(sa, "MACVLAN filters not supported");
1751                 break;
1752         case RTE_ETH_FILTER_ETHERTYPE:
1753                 sfc_err(sa, "EtherType filters not supported");
1754                 break;
1755         case RTE_ETH_FILTER_FLEXIBLE:
1756                 sfc_err(sa, "Flexible filters not supported");
1757                 break;
1758         case RTE_ETH_FILTER_SYN:
1759                 sfc_err(sa, "SYN filters not supported");
1760                 break;
1761         case RTE_ETH_FILTER_NTUPLE:
1762                 sfc_err(sa, "NTUPLE filters not supported");
1763                 break;
1764         case RTE_ETH_FILTER_TUNNEL:
1765                 sfc_err(sa, "Tunnel filters not supported");
1766                 break;
1767         case RTE_ETH_FILTER_FDIR:
1768                 sfc_err(sa, "Flow Director filters not supported");
1769                 break;
1770         case RTE_ETH_FILTER_HASH:
1771                 sfc_err(sa, "Hash filters not supported");
1772                 break;
1773         case RTE_ETH_FILTER_GENERIC:
1774                 if (filter_op != RTE_ETH_FILTER_GET) {
1775                         rc = EINVAL;
1776                 } else {
1777                         *(const void **)arg = &sfc_flow_ops;
1778                         rc = 0;
1779                 }
1780                 break;
1781         default:
1782                 sfc_err(sa, "Unknown filter type %u", filter_type);
1783                 break;
1784         }
1785
1786         sfc_log_init(sa, "exit: %d", -rc);
1787         SFC_ASSERT(rc >= 0);
1788         return -rc;
1789 }
1790
1791 static int
1792 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1793 {
1794         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1795
1796         /*
1797          * If Rx datapath does not provide callback to check mempool,
1798          * all pools are supported.
1799          */
1800         if (sap->dp_rx->pool_ops_supported == NULL)
1801                 return 1;
1802
1803         return sap->dp_rx->pool_ops_supported(pool);
1804 }
1805
1806 static int
1807 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1808 {
1809         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1810         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1811         struct sfc_rxq_info *rxq_info;
1812
1813         SFC_ASSERT(queue_id < sas->rxq_count);
1814         rxq_info = &sas->rxq_info[queue_id];
1815
1816         return sap->dp_rx->intr_enable(rxq_info->dp);
1817 }
1818
1819 static int
1820 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1821 {
1822         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1823         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1824         struct sfc_rxq_info *rxq_info;
1825
1826         SFC_ASSERT(queue_id < sas->rxq_count);
1827         rxq_info = &sas->rxq_info[queue_id];
1828
1829         return sap->dp_rx->intr_disable(rxq_info->dp);
1830 }
1831
1832 static const struct eth_dev_ops sfc_eth_dev_ops = {
1833         .dev_configure                  = sfc_dev_configure,
1834         .dev_start                      = sfc_dev_start,
1835         .dev_stop                       = sfc_dev_stop,
1836         .dev_set_link_up                = sfc_dev_set_link_up,
1837         .dev_set_link_down              = sfc_dev_set_link_down,
1838         .dev_close                      = sfc_dev_close,
1839         .promiscuous_enable             = sfc_dev_promisc_enable,
1840         .promiscuous_disable            = sfc_dev_promisc_disable,
1841         .allmulticast_enable            = sfc_dev_allmulti_enable,
1842         .allmulticast_disable           = sfc_dev_allmulti_disable,
1843         .link_update                    = sfc_dev_link_update,
1844         .stats_get                      = sfc_stats_get,
1845         .stats_reset                    = sfc_stats_reset,
1846         .xstats_get                     = sfc_xstats_get,
1847         .xstats_reset                   = sfc_stats_reset,
1848         .xstats_get_names               = sfc_xstats_get_names,
1849         .dev_infos_get                  = sfc_dev_infos_get,
1850         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1851         .mtu_set                        = sfc_dev_set_mtu,
1852         .rx_queue_start                 = sfc_rx_queue_start,
1853         .rx_queue_stop                  = sfc_rx_queue_stop,
1854         .tx_queue_start                 = sfc_tx_queue_start,
1855         .tx_queue_stop                  = sfc_tx_queue_stop,
1856         .rx_queue_setup                 = sfc_rx_queue_setup,
1857         .rx_queue_release               = sfc_rx_queue_release,
1858         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1859         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1860         .tx_queue_setup                 = sfc_tx_queue_setup,
1861         .tx_queue_release               = sfc_tx_queue_release,
1862         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1863         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1864         .mac_addr_set                   = sfc_mac_addr_set,
1865         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1866         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1867         .reta_update                    = sfc_dev_rss_reta_update,
1868         .reta_query                     = sfc_dev_rss_reta_query,
1869         .rss_hash_update                = sfc_dev_rss_hash_update,
1870         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1871         .filter_ctrl                    = sfc_dev_filter_ctrl,
1872         .set_mc_addr_list               = sfc_set_mc_addr_list,
1873         .rxq_info_get                   = sfc_rx_queue_info_get,
1874         .txq_info_get                   = sfc_tx_queue_info_get,
1875         .fw_version_get                 = sfc_fw_version_get,
1876         .xstats_get_by_id               = sfc_xstats_get_by_id,
1877         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1878         .pool_ops_supported             = sfc_pool_ops_supported,
1879 };
1880
1881 /**
1882  * Duplicate a string in potentially shared memory required for
1883  * multi-process support.
1884  *
1885  * strdup() allocates from process-local heap/memory.
1886  */
1887 static char *
1888 sfc_strdup(const char *str)
1889 {
1890         size_t size;
1891         char *copy;
1892
1893         if (str == NULL)
1894                 return NULL;
1895
1896         size = strlen(str) + 1;
1897         copy = rte_malloc(__func__, size, 0);
1898         if (copy != NULL)
1899                 rte_memcpy(copy, str, size);
1900
1901         return copy;
1902 }
1903
1904 static int
1905 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1906 {
1907         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1908         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1909         const struct sfc_dp_rx *dp_rx;
1910         const struct sfc_dp_tx *dp_tx;
1911         const efx_nic_cfg_t *encp;
1912         unsigned int avail_caps = 0;
1913         const char *rx_name = NULL;
1914         const char *tx_name = NULL;
1915         int rc;
1916
1917         switch (sa->family) {
1918         case EFX_FAMILY_HUNTINGTON:
1919         case EFX_FAMILY_MEDFORD:
1920         case EFX_FAMILY_MEDFORD2:
1921                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1922                 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX;
1923                 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX;
1924                 break;
1925         case EFX_FAMILY_RIVERHEAD:
1926                 avail_caps |= SFC_DP_HW_FW_CAP_EF100;
1927                 break;
1928         default:
1929                 break;
1930         }
1931
1932         encp = efx_nic_cfg_get(sa->nic);
1933         if (encp->enc_rx_es_super_buffer_supported)
1934                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1935
1936         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1937                                 sfc_kvarg_string_handler, &rx_name);
1938         if (rc != 0)
1939                 goto fail_kvarg_rx_datapath;
1940
1941         if (rx_name != NULL) {
1942                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1943                 if (dp_rx == NULL) {
1944                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1945                         rc = ENOENT;
1946                         goto fail_dp_rx;
1947                 }
1948                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1949                         sfc_err(sa,
1950                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1951                                 rx_name);
1952                         rc = EINVAL;
1953                         goto fail_dp_rx_caps;
1954                 }
1955         } else {
1956                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1957                 if (dp_rx == NULL) {
1958                         sfc_err(sa, "Rx datapath by caps %#x not found",
1959                                 avail_caps);
1960                         rc = ENOENT;
1961                         goto fail_dp_rx;
1962                 }
1963         }
1964
1965         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1966         if (sas->dp_rx_name == NULL) {
1967                 rc = ENOMEM;
1968                 goto fail_dp_rx_name;
1969         }
1970
1971         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1972
1973         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1974                                 sfc_kvarg_string_handler, &tx_name);
1975         if (rc != 0)
1976                 goto fail_kvarg_tx_datapath;
1977
1978         if (tx_name != NULL) {
1979                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1980                 if (dp_tx == NULL) {
1981                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1982                         rc = ENOENT;
1983                         goto fail_dp_tx;
1984                 }
1985                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1986                         sfc_err(sa,
1987                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1988                                 tx_name);
1989                         rc = EINVAL;
1990                         goto fail_dp_tx_caps;
1991                 }
1992         } else {
1993                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1994                 if (dp_tx == NULL) {
1995                         sfc_err(sa, "Tx datapath by caps %#x not found",
1996                                 avail_caps);
1997                         rc = ENOENT;
1998                         goto fail_dp_tx;
1999                 }
2000         }
2001
2002         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2003         if (sas->dp_tx_name == NULL) {
2004                 rc = ENOMEM;
2005                 goto fail_dp_tx_name;
2006         }
2007
2008         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2009
2010         sa->priv.dp_rx = dp_rx;
2011         sa->priv.dp_tx = dp_tx;
2012
2013         dev->rx_pkt_burst = dp_rx->pkt_burst;
2014         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2015         dev->tx_pkt_burst = dp_tx->pkt_burst;
2016
2017         dev->rx_queue_count = sfc_rx_queue_count;
2018         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2019         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2020         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2021         dev->dev_ops = &sfc_eth_dev_ops;
2022
2023         return 0;
2024
2025 fail_dp_tx_name:
2026 fail_dp_tx_caps:
2027 fail_dp_tx:
2028 fail_kvarg_tx_datapath:
2029         rte_free(sas->dp_rx_name);
2030         sas->dp_rx_name = NULL;
2031
2032 fail_dp_rx_name:
2033 fail_dp_rx_caps:
2034 fail_dp_rx:
2035 fail_kvarg_rx_datapath:
2036         return rc;
2037 }
2038
2039 static void
2040 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2041 {
2042         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2043         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2044
2045         dev->dev_ops = NULL;
2046         dev->tx_pkt_prepare = NULL;
2047         dev->rx_pkt_burst = NULL;
2048         dev->tx_pkt_burst = NULL;
2049
2050         rte_free(sas->dp_tx_name);
2051         sas->dp_tx_name = NULL;
2052         sa->priv.dp_tx = NULL;
2053
2054         rte_free(sas->dp_rx_name);
2055         sas->dp_rx_name = NULL;
2056         sa->priv.dp_rx = NULL;
2057 }
2058
2059 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2060         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2061         .reta_query                     = sfc_dev_rss_reta_query,
2062         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2063         .rxq_info_get                   = sfc_rx_queue_info_get,
2064         .txq_info_get                   = sfc_tx_queue_info_get,
2065 };
2066
2067 static int
2068 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2069 {
2070         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2071         struct sfc_adapter_priv *sap;
2072         const struct sfc_dp_rx *dp_rx;
2073         const struct sfc_dp_tx *dp_tx;
2074         int rc;
2075
2076         /*
2077          * Allocate process private data from heap, since it should not
2078          * be located in shared memory allocated using rte_malloc() API.
2079          */
2080         sap = calloc(1, sizeof(*sap));
2081         if (sap == NULL) {
2082                 rc = ENOMEM;
2083                 goto fail_alloc_priv;
2084         }
2085
2086         sap->logtype_main = logtype_main;
2087
2088         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2089         if (dp_rx == NULL) {
2090                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2091                         "cannot find %s Rx datapath", sas->dp_rx_name);
2092                 rc = ENOENT;
2093                 goto fail_dp_rx;
2094         }
2095         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2096                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2097                         "%s Rx datapath does not support multi-process",
2098                         sas->dp_rx_name);
2099                 rc = EINVAL;
2100                 goto fail_dp_rx_multi_process;
2101         }
2102
2103         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2104         if (dp_tx == NULL) {
2105                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2106                         "cannot find %s Tx datapath", sas->dp_tx_name);
2107                 rc = ENOENT;
2108                 goto fail_dp_tx;
2109         }
2110         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2111                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2112                         "%s Tx datapath does not support multi-process",
2113                         sas->dp_tx_name);
2114                 rc = EINVAL;
2115                 goto fail_dp_tx_multi_process;
2116         }
2117
2118         sap->dp_rx = dp_rx;
2119         sap->dp_tx = dp_tx;
2120
2121         dev->process_private = sap;
2122         dev->rx_pkt_burst = dp_rx->pkt_burst;
2123         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2124         dev->tx_pkt_burst = dp_tx->pkt_burst;
2125         dev->rx_queue_count = sfc_rx_queue_count;
2126         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2127         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2128         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2129         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2130
2131         return 0;
2132
2133 fail_dp_tx_multi_process:
2134 fail_dp_tx:
2135 fail_dp_rx_multi_process:
2136 fail_dp_rx:
2137         free(sap);
2138
2139 fail_alloc_priv:
2140         return rc;
2141 }
2142
2143 static void
2144 sfc_register_dp(void)
2145 {
2146         /* Register once */
2147         if (TAILQ_EMPTY(&sfc_dp_head)) {
2148                 /* Prefer EF10 datapath */
2149                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp);
2150                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2151                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2152                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2153
2154                 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp);
2155                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2156                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2157                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2158         }
2159 }
2160
2161 static int
2162 sfc_eth_dev_init(struct rte_eth_dev *dev)
2163 {
2164         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2165         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2166         uint32_t logtype_main;
2167         struct sfc_adapter *sa;
2168         int rc;
2169         const efx_nic_cfg_t *encp;
2170         const struct rte_ether_addr *from;
2171         int ret;
2172
2173         sfc_register_dp();
2174
2175         logtype_main = sfc_register_logtype(&pci_dev->addr,
2176                                             SFC_LOGTYPE_MAIN_STR,
2177                                             RTE_LOG_NOTICE);
2178
2179         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2180                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2181
2182         /* Required for logging */
2183         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2184                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2185                         pci_dev->addr.domain, pci_dev->addr.bus,
2186                         pci_dev->addr.devid, pci_dev->addr.function,
2187                         dev->data->port_id);
2188         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2189                 SFC_GENERIC_LOG(ERR,
2190                         "reserved log prefix is too short for " PCI_PRI_FMT,
2191                         pci_dev->addr.domain, pci_dev->addr.bus,
2192                         pci_dev->addr.devid, pci_dev->addr.function);
2193                 return -EINVAL;
2194         }
2195         sas->pci_addr = pci_dev->addr;
2196         sas->port_id = dev->data->port_id;
2197
2198         /*
2199          * Allocate process private data from heap, since it should not
2200          * be located in shared memory allocated using rte_malloc() API.
2201          */
2202         sa = calloc(1, sizeof(*sa));
2203         if (sa == NULL) {
2204                 rc = ENOMEM;
2205                 goto fail_alloc_sa;
2206         }
2207
2208         dev->process_private = sa;
2209
2210         /* Required for logging */
2211         sa->priv.shared = sas;
2212         sa->priv.logtype_main = logtype_main;
2213
2214         sa->eth_dev = dev;
2215
2216         /* Copy PCI device info to the dev->data */
2217         rte_eth_copy_pci_info(dev, pci_dev);
2218
2219         rc = sfc_kvargs_parse(sa);
2220         if (rc != 0)
2221                 goto fail_kvargs_parse;
2222
2223         sfc_log_init(sa, "entry");
2224
2225         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2226         if (dev->data->mac_addrs == NULL) {
2227                 rc = ENOMEM;
2228                 goto fail_mac_addrs;
2229         }
2230
2231         sfc_adapter_lock_init(sa);
2232         sfc_adapter_lock(sa);
2233
2234         sfc_log_init(sa, "probing");
2235         rc = sfc_probe(sa);
2236         if (rc != 0)
2237                 goto fail_probe;
2238
2239         sfc_log_init(sa, "set device ops");
2240         rc = sfc_eth_dev_set_ops(dev);
2241         if (rc != 0)
2242                 goto fail_set_ops;
2243
2244         sfc_log_init(sa, "attaching");
2245         rc = sfc_attach(sa);
2246         if (rc != 0)
2247                 goto fail_attach;
2248
2249         encp = efx_nic_cfg_get(sa->nic);
2250
2251         /*
2252          * The arguments are really reverse order in comparison to
2253          * Linux kernel. Copy from NIC config to Ethernet device data.
2254          */
2255         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2256         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2257
2258         sfc_adapter_unlock(sa);
2259
2260         sfc_log_init(sa, "done");
2261         return 0;
2262
2263 fail_attach:
2264         sfc_eth_dev_clear_ops(dev);
2265
2266 fail_set_ops:
2267         sfc_unprobe(sa);
2268
2269 fail_probe:
2270         sfc_adapter_unlock(sa);
2271         sfc_adapter_lock_fini(sa);
2272         rte_free(dev->data->mac_addrs);
2273         dev->data->mac_addrs = NULL;
2274
2275 fail_mac_addrs:
2276         sfc_kvargs_cleanup(sa);
2277
2278 fail_kvargs_parse:
2279         sfc_log_init(sa, "failed %d", rc);
2280         dev->process_private = NULL;
2281         free(sa);
2282
2283 fail_alloc_sa:
2284         SFC_ASSERT(rc > 0);
2285         return -rc;
2286 }
2287
2288 static int
2289 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2290 {
2291         sfc_dev_close(dev);
2292
2293         return 0;
2294 }
2295
2296 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2297         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2298         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2299         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2300         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2301         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2302         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2303         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2304         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2305         { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) },
2306         { .vendor_id = 0 /* sentinel */ }
2307 };
2308
2309 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2310         struct rte_pci_device *pci_dev)
2311 {
2312         return rte_eth_dev_pci_generic_probe(pci_dev,
2313                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2314 }
2315
2316 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2317 {
2318         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2319 }
2320
2321 static struct rte_pci_driver sfc_efx_pmd = {
2322         .id_table = pci_id_sfc_efx_map,
2323         .drv_flags =
2324                 RTE_PCI_DRV_INTR_LSC |
2325                 RTE_PCI_DRV_NEED_MAPPING,
2326         .probe = sfc_eth_dev_pci_probe,
2327         .remove = sfc_eth_dev_pci_remove,
2328 };
2329
2330 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2331 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2332 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2333 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2334         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2335         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2336         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2337         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2338         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2339         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2340
2341 RTE_INIT(sfc_driver_register_logtype)
2342 {
2343         int ret;
2344
2345         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2346                                                    RTE_LOG_NOTICE);
2347         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2348 }