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         dev->process_private = NULL;
 328         dev->dev_ops = NULL;
 329         dev->tx_pkt_prepare = NULL;
 330         dev->tx_pkt_burst = NULL;
 331         dev->rx_pkt_burst = NULL;
 332 }
 333
 334 static int
 335 sfc_dev_close(struct rte_eth_dev *dev)
 336 {
 337         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 338
 339         sfc_log_init(sa, "entry");
 340
 341         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 342                 sfc_eth_dev_secondary_clear_ops(dev);
 343                 return 0;
 344         }
 345
 346         sfc_adapter_lock(sa);
 347         switch (sa->state) {
 348         case SFC_ADAPTER_STARTED:
 349                 sfc_stop(sa);
 350                 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED);
 351                 /* FALLTHROUGH */
 352         case SFC_ADAPTER_CONFIGURED:
 353                 sfc_close(sa);
 354                 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED);
 355                 /* FALLTHROUGH */
 356         case SFC_ADAPTER_INITIALIZED:
 357                 break;
 358         default:
 359                 sfc_err(sa, "unexpected adapter state %u on close", sa->state);
 360                 break;
 361         }
 362
 363         /*
 364          * Cleanup all resources.
 365          * Rollback primary process sfc_eth_dev_init() below.
 366          */
 367
 368         sfc_eth_dev_clear_ops(dev);
 369
 370         sfc_detach(sa);
 371         sfc_unprobe(sa);
 372
 373         sfc_kvargs_cleanup(sa);
 374
 375         sfc_adapter_unlock(sa);
 376         sfc_adapter_lock_fini(sa);
 377
 378         sfc_log_init(sa, "done");
 379
 380         /* Required for logging, so cleanup last */
 381         sa->eth_dev = NULL;
 382
 383         dev->process_private = NULL;
 384         free(sa);
 385
 386         return 0;
 387 }
 388
 389 static int
 390 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode,
 391                    boolean_t enabled)
 392 {
 393         struct sfc_port *port;
 394         boolean_t *toggle;
 395         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 396         boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI);
 397         const char *desc = (allmulti) ? "all-multi" : "promiscuous";
 398         int rc = 0;
 399
 400         sfc_adapter_lock(sa);
 401
 402         port = &sa->port;
 403         toggle = (allmulti) ? (&port->allmulti) : (&port->promisc);
 404
 405         if (*toggle != enabled) {
 406                 *toggle = enabled;
 407
 408                 if (sfc_sa2shared(sa)->isolated) {
 409                         sfc_warn(sa, "isolated mode is active on the port");
 410                         sfc_warn(sa, "the change is to be applied on the next "
 411                                      "start provided that isolated mode is "
 412                                      "disabled prior the next start");
 413                 } else if ((sa->state == SFC_ADAPTER_STARTED) &&
 414                            ((rc = sfc_set_rx_mode(sa)) != 0)) {
 415                         *toggle = !(enabled);
 416                         sfc_warn(sa, "Failed to %s %s mode, rc = %d",
 417                                  ((enabled) ? "enable" : "disable"), desc, rc);
 418
 419                         /*
 420                          * For promiscuous and all-multicast filters a
 421                          * permission failure should be reported as an
 422                          * unsupported filter.
 423                          */
 424                         if (rc == EPERM)
 425                                 rc = ENOTSUP;
 426                 }
 427         }
 428
 429         sfc_adapter_unlock(sa);
 430         return rc;
 431 }
 432
 433 static int
 434 sfc_dev_promisc_enable(struct rte_eth_dev *dev)
 435 {
 436         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE);
 437
 438         SFC_ASSERT(rc >= 0);
 439         return -rc;
 440 }
 441
 442 static int
 443 sfc_dev_promisc_disable(struct rte_eth_dev *dev)
 444 {
 445         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE);
 446
 447         SFC_ASSERT(rc >= 0);
 448         return -rc;
 449 }
 450
 451 static int
 452 sfc_dev_allmulti_enable(struct rte_eth_dev *dev)
 453 {
 454         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE);
 455
 456         SFC_ASSERT(rc >= 0);
 457         return -rc;
 458 }
 459
 460 static int
 461 sfc_dev_allmulti_disable(struct rte_eth_dev *dev)
 462 {
 463         int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE);
 464
 465         SFC_ASSERT(rc >= 0);
 466         return -rc;
 467 }
 468
 469 static int
 470 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
 471                    uint16_t nb_rx_desc, unsigned int socket_id,
 472                    const struct rte_eth_rxconf *rx_conf,
 473                    struct rte_mempool *mb_pool)
 474 {
 475         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 476         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 477         int rc;
 478
 479         sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u",
 480                      rx_queue_id, nb_rx_desc, socket_id);
 481
 482         sfc_adapter_lock(sa);
 483
 484         rc = sfc_rx_qinit(sa, rx_queue_id, nb_rx_desc, socket_id,
 485                           rx_conf, mb_pool);
 486         if (rc != 0)
 487                 goto fail_rx_qinit;
 488
 489         dev->data->rx_queues[rx_queue_id] = sas->rxq_info[rx_queue_id].dp;
 490
 491         sfc_adapter_unlock(sa);
 492
 493         return 0;
 494
 495 fail_rx_qinit:
 496         sfc_adapter_unlock(sa);
 497         SFC_ASSERT(rc > 0);
 498         return -rc;
 499 }
 500
 501 static void
 502 sfc_rx_queue_release(void *queue)
 503 {
 504         struct sfc_dp_rxq *dp_rxq = queue;
 505         struct sfc_rxq *rxq;
 506         struct sfc_adapter *sa;
 507         unsigned int sw_index;
 508
 509         if (dp_rxq == NULL)
 510                 return;
 511
 512         rxq = sfc_rxq_by_dp_rxq(dp_rxq);
 513         sa = rxq->evq->sa;
 514         sfc_adapter_lock(sa);
 515
 516         sw_index = dp_rxq->dpq.queue_id;
 517
 518         sfc_log_init(sa, "RxQ=%u", sw_index);
 519
 520         sfc_rx_qfini(sa, sw_index);
 521
 522         sfc_adapter_unlock(sa);
 523 }
 524
 525 static int
 526 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
 527                    uint16_t nb_tx_desc, unsigned int socket_id,
 528                    const struct rte_eth_txconf *tx_conf)
 529 {
 530         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
 531         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 532         int rc;
 533
 534         sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u",
 535                      tx_queue_id, nb_tx_desc, socket_id);
 536
 537         sfc_adapter_lock(sa);
 538
 539         rc = sfc_tx_qinit(sa, tx_queue_id, nb_tx_desc, socket_id, tx_conf);
 540         if (rc != 0)
 541                 goto fail_tx_qinit;
 542
 543         dev->data->tx_queues[tx_queue_id] = sas->txq_info[tx_queue_id].dp;
 544
 545         sfc_adapter_unlock(sa);
 546         return 0;
 547
 548 fail_tx_qinit:
 549         sfc_adapter_unlock(sa);
 550         SFC_ASSERT(rc > 0);
 551         return -rc;
 552 }
 553
 554 static void
 555 sfc_tx_queue_release(void *queue)
 556 {
 557         struct sfc_dp_txq *dp_txq = queue;
 558         struct sfc_txq *txq;
 559         unsigned int sw_index;
 560         struct sfc_adapter *sa;
 561
 562         if (dp_txq == NULL)
 563                 return;
 564
 565         txq = sfc_txq_by_dp_txq(dp_txq);
 566         sw_index = dp_txq->dpq.queue_id;
 567
 568         SFC_ASSERT(txq->evq != NULL);
 569         sa = txq->evq->sa;
 570
 571         sfc_log_init(sa, "TxQ = %u", sw_index);
 572
 573         sfc_adapter_lock(sa);
 574
 575         sfc_tx_qfini(sa, sw_index);
 576
 577         sfc_adapter_unlock(sa);
 578 }
 579
 580 /*
 581  * Some statistics are computed as A - B where A and B each increase
 582  * monotonically with some hardware counter(s) and the counters are read
 583  * asynchronously.
 584  *
 585  * If packet X is counted in A, but not counted in B yet, computed value is
 586  * greater than real.
 587  *
 588  * If packet X is not counted in A at the moment of reading the counter,
 589  * but counted in B at the moment of reading the counter, computed value
 590  * is less than real.
 591  *
 592  * However, counter which grows backward is worse evil than slightly wrong
 593  * value. So, let's try to guarantee that it never happens except may be
 594  * the case when the MAC stats are zeroed as a result of a NIC reset.
 595  */
 596 static void
 597 sfc_update_diff_stat(uint64_t *stat, uint64_t newval)
 598 {
 599         if ((int64_t)(newval - *stat) > 0 || newval == 0)
 600                 *stat = newval;
 601 }
 602
 603 static int
 604 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 605 {
 606         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 607         struct sfc_port *port = &sa->port;
 608         uint64_t *mac_stats;
 609         int ret;
 610
 611         rte_spinlock_lock(&port->mac_stats_lock);
 612
 613         ret = sfc_port_update_mac_stats(sa);
 614         if (ret != 0)
 615                 goto unlock;
 616
 617         mac_stats = port->mac_stats_buf;
 618
 619         if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask,
 620                                    EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) {
 621                 stats->ipackets =
 622                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] +
 623                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] +
 624                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS];
 625                 stats->opackets =
 626                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] +
 627                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] +
 628                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS];
 629                 stats->ibytes =
 630                         mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] +
 631                         mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] +
 632                         mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES];
 633                 stats->obytes =
 634                         mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] +
 635                         mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] +
 636                         mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES];
 637                 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS];
 638                 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS];
 639         } else {
 640                 stats->opackets = mac_stats[EFX_MAC_TX_PKTS];
 641                 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS];
 642                 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS];
 643                 /*
 644                  * Take into account stats which are whenever supported
 645                  * on EF10. If some stat is not supported by current
 646                  * firmware variant or HW revision, it is guaranteed
 647                  * to be zero in mac_stats.
 648                  */
 649                 stats->imissed =
 650                         mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] +
 651                         mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] +
 652                         mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] +
 653                         mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] +
 654                         mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] +
 655                         mac_stats[EFX_MAC_PM_TRUNC_QBB] +
 656                         mac_stats[EFX_MAC_PM_DISCARD_QBB] +
 657                         mac_stats[EFX_MAC_PM_DISCARD_MAPPING] +
 658                         mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] +
 659                         mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS];
 660                 stats->ierrors =
 661                         mac_stats[EFX_MAC_RX_FCS_ERRORS] +
 662                         mac_stats[EFX_MAC_RX_ALIGN_ERRORS] +
 663                         mac_stats[EFX_MAC_RX_JABBER_PKTS];
 664                 /* no oerrors counters supported on EF10 */
 665
 666                 /* Exclude missed, errors and pauses from Rx packets */
 667                 sfc_update_diff_stat(&port->ipackets,
 668                         mac_stats[EFX_MAC_RX_PKTS] -
 669                         mac_stats[EFX_MAC_RX_PAUSE_PKTS] -
 670                         stats->imissed - stats->ierrors);
 671                 stats->ipackets = port->ipackets;
 672         }
 673
 674 unlock:
 675         rte_spinlock_unlock(&port->mac_stats_lock);
 676         SFC_ASSERT(ret >= 0);
 677         return -ret;
 678 }
 679
 680 static int
 681 sfc_stats_reset(struct rte_eth_dev *dev)
 682 {
 683         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 684         struct sfc_port *port = &sa->port;
 685         int rc;
 686
 687         if (sa->state != SFC_ADAPTER_STARTED) {
 688                 /*
 689                  * The operation cannot be done if port is not started; it
 690                  * will be scheduled to be done during the next port start
 691                  */
 692                 port->mac_stats_reset_pending = B_TRUE;
 693                 return 0;
 694         }
 695
 696         rc = sfc_port_reset_mac_stats(sa);
 697         if (rc != 0)
 698                 sfc_err(sa, "failed to reset statistics (rc = %d)", rc);
 699
 700         SFC_ASSERT(rc >= 0);
 701         return -rc;
 702 }
 703
 704 static int
 705 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
 706                unsigned int xstats_count)
 707 {
 708         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 709         struct sfc_port *port = &sa->port;
 710         uint64_t *mac_stats;
 711         int rc;
 712         unsigned int i;
 713         int nstats = 0;
 714
 715         rte_spinlock_lock(&port->mac_stats_lock);
 716
 717         rc = sfc_port_update_mac_stats(sa);
 718         if (rc != 0) {
 719                 SFC_ASSERT(rc > 0);
 720                 nstats = -rc;
 721                 goto unlock;
 722         }
 723
 724         mac_stats = port->mac_stats_buf;
 725
 726         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 727                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 728                         if (xstats != NULL && nstats < (int)xstats_count) {
 729                                 xstats[nstats].id = nstats;
 730                                 xstats[nstats].value = mac_stats[i];
 731                         }
 732                         nstats++;
 733                 }
 734         }
 735
 736 unlock:
 737         rte_spinlock_unlock(&port->mac_stats_lock);
 738
 739         return nstats;
 740 }
 741
 742 static int
 743 sfc_xstats_get_names(struct rte_eth_dev *dev,
 744                      struct rte_eth_xstat_name *xstats_names,
 745                      unsigned int xstats_count)
 746 {
 747         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 748         struct sfc_port *port = &sa->port;
 749         unsigned int i;
 750         unsigned int nstats = 0;
 751
 752         for (i = 0; i < EFX_MAC_NSTATS; ++i) {
 753                 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) {
 754                         if (xstats_names != NULL && nstats < xstats_count)
 755                                 strlcpy(xstats_names[nstats].name,
 756                                         efx_mac_stat_name(sa->nic, i),
 757                                         sizeof(xstats_names[0].name));
 758                         nstats++;
 759                 }
 760         }
 761
 762         return nstats;
 763 }
 764
 765 static int
 766 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids,
 767                      uint64_t *values, unsigned int n)
 768 {
 769         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 770         struct sfc_port *port = &sa->port;
 771         uint64_t *mac_stats;
 772         unsigned int nb_supported = 0;
 773         unsigned int nb_written = 0;
 774         unsigned int i;
 775         int ret;
 776         int rc;
 777
 778         if (unlikely(values == NULL) ||
 779             unlikely((ids == NULL) && (n < port->mac_stats_nb_supported)))
 780                 return port->mac_stats_nb_supported;
 781
 782         rte_spinlock_lock(&port->mac_stats_lock);
 783
 784         rc = sfc_port_update_mac_stats(sa);
 785         if (rc != 0) {
 786                 SFC_ASSERT(rc > 0);
 787                 ret = -rc;
 788                 goto unlock;
 789         }
 790
 791         mac_stats = port->mac_stats_buf;
 792
 793         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < n); ++i) {
 794                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 795                         continue;
 796
 797                 if ((ids == NULL) || (ids[nb_written] == nb_supported))
 798                         values[nb_written++] = mac_stats[i];
 799
 800                 ++nb_supported;
 801         }
 802
 803         ret = nb_written;
 804
 805 unlock:
 806         rte_spinlock_unlock(&port->mac_stats_lock);
 807
 808         return ret;
 809 }
 810
 811 static int
 812 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev,
 813                            struct rte_eth_xstat_name *xstats_names,
 814                            const uint64_t *ids, unsigned int size)
 815 {
 816         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 817         struct sfc_port *port = &sa->port;
 818         unsigned int nb_supported = 0;
 819         unsigned int nb_written = 0;
 820         unsigned int i;
 821
 822         if (unlikely(xstats_names == NULL) ||
 823             unlikely((ids == NULL) && (size < port->mac_stats_nb_supported)))
 824                 return port->mac_stats_nb_supported;
 825
 826         for (i = 0; (i < EFX_MAC_NSTATS) && (nb_written < size); ++i) {
 827                 if (!EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i))
 828                         continue;
 829
 830                 if ((ids == NULL) || (ids[nb_written] == nb_supported)) {
 831                         char *name = xstats_names[nb_written++].name;
 832
 833                         strlcpy(name, efx_mac_stat_name(sa->nic, i),
 834                                 sizeof(xstats_names[0].name));
 835                 }
 836
 837                 ++nb_supported;
 838         }
 839
 840         return nb_written;
 841 }
 842
 843 static int
 844 sfc_flow_ctrl_get(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 845 {
 846         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 847         unsigned int wanted_fc, link_fc;
 848
 849         memset(fc_conf, 0, sizeof(*fc_conf));
 850
 851         sfc_adapter_lock(sa);
 852
 853         if (sa->state == SFC_ADAPTER_STARTED)
 854                 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc);
 855         else
 856                 link_fc = sa->port.flow_ctrl;
 857
 858         switch (link_fc) {
 859         case 0:
 860                 fc_conf->mode = RTE_FC_NONE;
 861                 break;
 862         case EFX_FCNTL_RESPOND:
 863                 fc_conf->mode = RTE_FC_RX_PAUSE;
 864                 break;
 865         case EFX_FCNTL_GENERATE:
 866                 fc_conf->mode = RTE_FC_TX_PAUSE;
 867                 break;
 868         case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE):
 869                 fc_conf->mode = RTE_FC_FULL;
 870                 break;
 871         default:
 872                 sfc_err(sa, "%s: unexpected flow control value %#x",
 873                         __func__, link_fc);
 874         }
 875
 876         fc_conf->autoneg = sa->port.flow_ctrl_autoneg;
 877
 878         sfc_adapter_unlock(sa);
 879
 880         return 0;
 881 }
 882
 883 static int
 884 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf)
 885 {
 886         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 887         struct sfc_port *port = &sa->port;
 888         unsigned int fcntl;
 889         int rc;
 890
 891         if (fc_conf->high_water != 0 || fc_conf->low_water != 0 ||
 892             fc_conf->pause_time != 0 || fc_conf->send_xon != 0 ||
 893             fc_conf->mac_ctrl_frame_fwd != 0) {
 894                 sfc_err(sa, "unsupported flow control settings specified");
 895                 rc = EINVAL;
 896                 goto fail_inval;
 897         }
 898
 899         switch (fc_conf->mode) {
 900         case RTE_FC_NONE:
 901                 fcntl = 0;
 902                 break;
 903         case RTE_FC_RX_PAUSE:
 904                 fcntl = EFX_FCNTL_RESPOND;
 905                 break;
 906         case RTE_FC_TX_PAUSE:
 907                 fcntl = EFX_FCNTL_GENERATE;
 908                 break;
 909         case RTE_FC_FULL:
 910                 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE;
 911                 break;
 912         default:
 913                 rc = EINVAL;
 914                 goto fail_inval;
 915         }
 916
 917         sfc_adapter_lock(sa);
 918
 919         if (sa->state == SFC_ADAPTER_STARTED) {
 920                 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg);
 921                 if (rc != 0)
 922                         goto fail_mac_fcntl_set;
 923         }
 924
 925         port->flow_ctrl = fcntl;
 926         port->flow_ctrl_autoneg = fc_conf->autoneg;
 927
 928         sfc_adapter_unlock(sa);
 929
 930         return 0;
 931
 932 fail_mac_fcntl_set:
 933         sfc_adapter_unlock(sa);
 934 fail_inval:
 935         SFC_ASSERT(rc > 0);
 936         return -rc;
 937 }
 938
 939 static int
 940 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu)
 941 {
 942         struct sfc_adapter_shared * const sas = sfc_sa2shared(sa);
 943         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
 944         boolean_t scatter_enabled;
 945         const char *error;
 946         unsigned int i;
 947
 948         for (i = 0; i < sas->rxq_count; i++) {
 949                 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0)
 950                         continue;
 951
 952                 scatter_enabled = (sas->rxq_info[i].type_flags &
 953                                    EFX_RXQ_FLAG_SCATTER);
 954
 955                 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size,
 956                                           encp->enc_rx_prefix_size,
 957                                           scatter_enabled, &error)) {
 958                         sfc_err(sa, "MTU check for RxQ %u failed: %s", i,
 959                                 error);
 960                         return EINVAL;
 961                 }
 962         }
 963
 964         return 0;
 965 }
 966
 967 static int
 968 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu)
 969 {
 970         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
 971         size_t pdu = EFX_MAC_PDU(mtu);
 972         size_t old_pdu;
 973         int rc;
 974
 975         sfc_log_init(sa, "mtu=%u", mtu);
 976
 977         rc = EINVAL;
 978         if (pdu < EFX_MAC_PDU_MIN) {
 979                 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)",
 980                         (unsigned int)mtu, (unsigned int)pdu,
 981                         EFX_MAC_PDU_MIN);
 982                 goto fail_inval;
 983         }
 984         if (pdu > EFX_MAC_PDU_MAX) {
 985                 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)",
 986                         (unsigned int)mtu, (unsigned int)pdu,
 987                         (unsigned int)EFX_MAC_PDU_MAX);
 988                 goto fail_inval;
 989         }
 990
 991         sfc_adapter_lock(sa);
 992
 993         rc = sfc_check_scatter_on_all_rx_queues(sa, pdu);
 994         if (rc != 0)
 995                 goto fail_check_scatter;
 996
 997         if (pdu != sa->port.pdu) {
 998                 if (sa->state == SFC_ADAPTER_STARTED) {
 999                         sfc_stop(sa);
1000
1001                         old_pdu = sa->port.pdu;
1002                         sa->port.pdu = pdu;
1003                         rc = sfc_start(sa);
1004                         if (rc != 0)
1005                                 goto fail_start;
1006                 } else {
1007                         sa->port.pdu = pdu;
1008                 }
1009         }
1010
1011         /*
1012          * The driver does not use it, but other PMDs update jumbo frame
1013          * flag and max_rx_pkt_len when MTU is set.
1014          */
1015         if (mtu > RTE_ETHER_MAX_LEN) {
1016                 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
1017                 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1018         }
1019
1020         dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu;
1021
1022         sfc_adapter_unlock(sa);
1023
1024         sfc_log_init(sa, "done");
1025         return 0;
1026
1027 fail_start:
1028         sa->port.pdu = old_pdu;
1029         if (sfc_start(sa) != 0)
1030                 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) "
1031                         "PDU max size - port is stopped",
1032                         (unsigned int)pdu, (unsigned int)old_pdu);
1033
1034 fail_check_scatter:
1035         sfc_adapter_unlock(sa);
1036
1037 fail_inval:
1038         sfc_log_init(sa, "failed %d", rc);
1039         SFC_ASSERT(rc > 0);
1040         return -rc;
1041 }
1042 static int
1043 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1044 {
1045         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1046         const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic);
1047         struct sfc_port *port = &sa->port;
1048         struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0];
1049         int rc = 0;
1050
1051         sfc_adapter_lock(sa);
1052
1053         if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr))
1054                 goto unlock;
1055
1056         /*
1057          * Copy the address to the device private data so that
1058          * it could be recalled in the case of adapter restart.
1059          */
1060         rte_ether_addr_copy(mac_addr, &port->default_mac_addr);
1061
1062         /*
1063          * Neither of the two following checks can return
1064          * an error. The new MAC address is preserved in
1065          * the device private data and can be activated
1066          * on the next port start if the user prevents
1067          * isolated mode from being enabled.
1068          */
1069         if (sfc_sa2shared(sa)->isolated) {
1070                 sfc_warn(sa, "isolated mode is active on the port");
1071                 sfc_warn(sa, "will not set MAC address");
1072                 goto unlock;
1073         }
1074
1075         if (sa->state != SFC_ADAPTER_STARTED) {
1076                 sfc_notice(sa, "the port is not started");
1077                 sfc_notice(sa, "the new MAC address will be set on port start");
1078
1079                 goto unlock;
1080         }
1081
1082         if (encp->enc_allow_set_mac_with_installed_filters) {
1083                 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes);
1084                 if (rc != 0) {
1085                         sfc_err(sa, "cannot set MAC address (rc = %u)", rc);
1086                         goto unlock;
1087                 }
1088
1089                 /*
1090                  * Changing the MAC address by means of MCDI request
1091                  * has no effect on received traffic, therefore
1092                  * we also need to update unicast filters
1093                  */
1094                 rc = sfc_set_rx_mode_unchecked(sa);
1095                 if (rc != 0) {
1096                         sfc_err(sa, "cannot set filter (rc = %u)", rc);
1097                         /* Rollback the old address */
1098                         (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes);
1099                         (void)sfc_set_rx_mode_unchecked(sa);
1100                 }
1101         } else {
1102                 sfc_warn(sa, "cannot set MAC address with filters installed");
1103                 sfc_warn(sa, "adapter will be restarted to pick the new MAC");
1104                 sfc_warn(sa, "(some traffic may be dropped)");
1105
1106                 /*
1107                  * Since setting MAC address with filters installed is not
1108                  * allowed on the adapter, the new MAC address will be set
1109                  * by means of adapter restart. sfc_start() shall retrieve
1110                  * the new address from the device private data and set it.
1111                  */
1112                 sfc_stop(sa);
1113                 rc = sfc_start(sa);
1114                 if (rc != 0)
1115                         sfc_err(sa, "cannot restart adapter (rc = %u)", rc);
1116         }
1117
1118 unlock:
1119         if (rc != 0)
1120                 rte_ether_addr_copy(old_addr, &port->default_mac_addr);
1121
1122         sfc_adapter_unlock(sa);
1123
1124         SFC_ASSERT(rc >= 0);
1125         return -rc;
1126 }
1127
1128
1129 static int
1130 sfc_set_mc_addr_list(struct rte_eth_dev *dev,
1131                 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr)
1132 {
1133         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1134         struct sfc_port *port = &sa->port;
1135         uint8_t *mc_addrs = port->mcast_addrs;
1136         int rc;
1137         unsigned int i;
1138
1139         if (sfc_sa2shared(sa)->isolated) {
1140                 sfc_err(sa, "isolated mode is active on the port");
1141                 sfc_err(sa, "will not set multicast address list");
1142                 return -ENOTSUP;
1143         }
1144
1145         if (mc_addrs == NULL)
1146                 return -ENOBUFS;
1147
1148         if (nb_mc_addr > port->max_mcast_addrs) {
1149                 sfc_err(sa, "too many multicast addresses: %u > %u",
1150                          nb_mc_addr, port->max_mcast_addrs);
1151                 return -EINVAL;
1152         }
1153
1154         for (i = 0; i < nb_mc_addr; ++i) {
1155                 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes,
1156                                  EFX_MAC_ADDR_LEN);
1157                 mc_addrs += EFX_MAC_ADDR_LEN;
1158         }
1159
1160         port->nb_mcast_addrs = nb_mc_addr;
1161
1162         if (sa->state != SFC_ADAPTER_STARTED)
1163                 return 0;
1164
1165         rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs,
1166                                         port->nb_mcast_addrs);
1167         if (rc != 0)
1168                 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc);
1169
1170         SFC_ASSERT(rc >= 0);
1171         return -rc;
1172 }
1173
1174 /*
1175  * The function is used by the secondary process as well. It must not
1176  * use any process-local pointers from the adapter data.
1177  */
1178 static void
1179 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t rx_queue_id,
1180                       struct rte_eth_rxq_info *qinfo)
1181 {
1182         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1183         struct sfc_rxq_info *rxq_info;
1184
1185         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1186
1187         rxq_info = &sas->rxq_info[rx_queue_id];
1188
1189         qinfo->mp = rxq_info->refill_mb_pool;
1190         qinfo->conf.rx_free_thresh = rxq_info->refill_threshold;
1191         qinfo->conf.rx_drop_en = 1;
1192         qinfo->conf.rx_deferred_start = rxq_info->deferred_start;
1193         qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
1194         if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) {
1195                 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER;
1196                 qinfo->scattered_rx = 1;
1197         }
1198         qinfo->nb_desc = rxq_info->entries;
1199 }
1200
1201 /*
1202  * The function is used by the secondary process as well. It must not
1203  * use any process-local pointers from the adapter data.
1204  */
1205 static void
1206 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t tx_queue_id,
1207                       struct rte_eth_txq_info *qinfo)
1208 {
1209         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1210         struct sfc_txq_info *txq_info;
1211
1212         SFC_ASSERT(tx_queue_id < sas->txq_count);
1213
1214         txq_info = &sas->txq_info[tx_queue_id];
1215
1216         memset(qinfo, 0, sizeof(*qinfo));
1217
1218         qinfo->conf.offloads = txq_info->offloads;
1219         qinfo->conf.tx_free_thresh = txq_info->free_thresh;
1220         qinfo->conf.tx_deferred_start = txq_info->deferred_start;
1221         qinfo->nb_desc = txq_info->entries;
1222 }
1223
1224 /*
1225  * The function is used by the secondary process as well. It must not
1226  * use any process-local pointers from the adapter data.
1227  */
1228 static uint32_t
1229 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1230 {
1231         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1232         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1233         struct sfc_rxq_info *rxq_info;
1234
1235         SFC_ASSERT(rx_queue_id < sas->rxq_count);
1236         rxq_info = &sas->rxq_info[rx_queue_id];
1237
1238         if ((rxq_info->state & SFC_RXQ_STARTED) == 0)
1239                 return 0;
1240
1241         return sap->dp_rx->qdesc_npending(rxq_info->dp);
1242 }
1243
1244 /*
1245  * The function is used by the secondary process as well. It must not
1246  * use any process-local pointers from the adapter data.
1247  */
1248 static int
1249 sfc_rx_descriptor_done(void *queue, uint16_t offset)
1250 {
1251         struct sfc_dp_rxq *dp_rxq = queue;
1252         const struct sfc_dp_rx *dp_rx;
1253
1254         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1255
1256         return offset < dp_rx->qdesc_npending(dp_rxq);
1257 }
1258
1259 /*
1260  * The function is used by the secondary process as well. It must not
1261  * use any process-local pointers from the adapter data.
1262  */
1263 static int
1264 sfc_rx_descriptor_status(void *queue, uint16_t offset)
1265 {
1266         struct sfc_dp_rxq *dp_rxq = queue;
1267         const struct sfc_dp_rx *dp_rx;
1268
1269         dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq);
1270
1271         return dp_rx->qdesc_status(dp_rxq, offset);
1272 }
1273
1274 /*
1275  * The function is used by the secondary process as well. It must not
1276  * use any process-local pointers from the adapter data.
1277  */
1278 static int
1279 sfc_tx_descriptor_status(void *queue, uint16_t offset)
1280 {
1281         struct sfc_dp_txq *dp_txq = queue;
1282         const struct sfc_dp_tx *dp_tx;
1283
1284         dp_tx = sfc_dp_tx_by_dp_txq(dp_txq);
1285
1286         return dp_tx->qdesc_status(dp_txq, offset);
1287 }
1288
1289 static int
1290 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1291 {
1292         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1293         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1294         int rc;
1295
1296         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1297
1298         sfc_adapter_lock(sa);
1299
1300         rc = EINVAL;
1301         if (sa->state != SFC_ADAPTER_STARTED)
1302                 goto fail_not_started;
1303
1304         if (sas->rxq_info[rx_queue_id].state != SFC_RXQ_INITIALIZED)
1305                 goto fail_not_setup;
1306
1307         rc = sfc_rx_qstart(sa, rx_queue_id);
1308         if (rc != 0)
1309                 goto fail_rx_qstart;
1310
1311         sas->rxq_info[rx_queue_id].deferred_started = B_TRUE;
1312
1313         sfc_adapter_unlock(sa);
1314
1315         return 0;
1316
1317 fail_rx_qstart:
1318 fail_not_setup:
1319 fail_not_started:
1320         sfc_adapter_unlock(sa);
1321         SFC_ASSERT(rc > 0);
1322         return -rc;
1323 }
1324
1325 static int
1326 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
1327 {
1328         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1329         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1330
1331         sfc_log_init(sa, "RxQ=%u", rx_queue_id);
1332
1333         sfc_adapter_lock(sa);
1334         sfc_rx_qstop(sa, rx_queue_id);
1335
1336         sas->rxq_info[rx_queue_id].deferred_started = B_FALSE;
1337
1338         sfc_adapter_unlock(sa);
1339
1340         return 0;
1341 }
1342
1343 static int
1344 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1345 {
1346         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1347         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1348         int rc;
1349
1350         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1351
1352         sfc_adapter_lock(sa);
1353
1354         rc = EINVAL;
1355         if (sa->state != SFC_ADAPTER_STARTED)
1356                 goto fail_not_started;
1357
1358         if (sas->txq_info[tx_queue_id].state != SFC_TXQ_INITIALIZED)
1359                 goto fail_not_setup;
1360
1361         rc = sfc_tx_qstart(sa, tx_queue_id);
1362         if (rc != 0)
1363                 goto fail_tx_qstart;
1364
1365         sas->txq_info[tx_queue_id].deferred_started = B_TRUE;
1366
1367         sfc_adapter_unlock(sa);
1368         return 0;
1369
1370 fail_tx_qstart:
1371
1372 fail_not_setup:
1373 fail_not_started:
1374         sfc_adapter_unlock(sa);
1375         SFC_ASSERT(rc > 0);
1376         return -rc;
1377 }
1378
1379 static int
1380 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
1381 {
1382         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1383         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1384
1385         sfc_log_init(sa, "TxQ = %u", tx_queue_id);
1386
1387         sfc_adapter_lock(sa);
1388
1389         sfc_tx_qstop(sa, tx_queue_id);
1390
1391         sas->txq_info[tx_queue_id].deferred_started = B_FALSE;
1392
1393         sfc_adapter_unlock(sa);
1394         return 0;
1395 }
1396
1397 static efx_tunnel_protocol_t
1398 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type)
1399 {
1400         switch (rte_type) {
1401         case RTE_TUNNEL_TYPE_VXLAN:
1402                 return EFX_TUNNEL_PROTOCOL_VXLAN;
1403         case RTE_TUNNEL_TYPE_GENEVE:
1404                 return EFX_TUNNEL_PROTOCOL_GENEVE;
1405         default:
1406                 return EFX_TUNNEL_NPROTOS;
1407         }
1408 }
1409
1410 enum sfc_udp_tunnel_op_e {
1411         SFC_UDP_TUNNEL_ADD_PORT,
1412         SFC_UDP_TUNNEL_DEL_PORT,
1413 };
1414
1415 static int
1416 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev,
1417                       struct rte_eth_udp_tunnel *tunnel_udp,
1418                       enum sfc_udp_tunnel_op_e op)
1419 {
1420         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1421         efx_tunnel_protocol_t tunnel_proto;
1422         int rc;
1423
1424         sfc_log_init(sa, "%s udp_port=%u prot_type=%u",
1425                      (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" :
1426                      (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown",
1427                      tunnel_udp->udp_port, tunnel_udp->prot_type);
1428
1429         tunnel_proto =
1430                 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type);
1431         if (tunnel_proto >= EFX_TUNNEL_NPROTOS) {
1432                 rc = ENOTSUP;
1433                 goto fail_bad_proto;
1434         }
1435
1436         sfc_adapter_lock(sa);
1437
1438         switch (op) {
1439         case SFC_UDP_TUNNEL_ADD_PORT:
1440                 rc = efx_tunnel_config_udp_add(sa->nic,
1441                                                tunnel_udp->udp_port,
1442                                                tunnel_proto);
1443                 break;
1444         case SFC_UDP_TUNNEL_DEL_PORT:
1445                 rc = efx_tunnel_config_udp_remove(sa->nic,
1446                                                   tunnel_udp->udp_port,
1447                                                   tunnel_proto);
1448                 break;
1449         default:
1450                 rc = EINVAL;
1451                 goto fail_bad_op;
1452         }
1453
1454         if (rc != 0)
1455                 goto fail_op;
1456
1457         if (sa->state == SFC_ADAPTER_STARTED) {
1458                 rc = efx_tunnel_reconfigure(sa->nic);
1459                 if (rc == EAGAIN) {
1460                         /*
1461                          * Configuration is accepted by FW and MC reboot
1462                          * is initiated to apply the changes. MC reboot
1463                          * will be handled in a usual way (MC reboot
1464                          * event on management event queue and adapter
1465                          * restart).
1466                          */
1467                         rc = 0;
1468                 } else if (rc != 0) {
1469                         goto fail_reconfigure;
1470                 }
1471         }
1472
1473         sfc_adapter_unlock(sa);
1474         return 0;
1475
1476 fail_reconfigure:
1477         /* Remove/restore entry since the change makes the trouble */
1478         switch (op) {
1479         case SFC_UDP_TUNNEL_ADD_PORT:
1480                 (void)efx_tunnel_config_udp_remove(sa->nic,
1481                                                    tunnel_udp->udp_port,
1482                                                    tunnel_proto);
1483                 break;
1484         case SFC_UDP_TUNNEL_DEL_PORT:
1485                 (void)efx_tunnel_config_udp_add(sa->nic,
1486                                                 tunnel_udp->udp_port,
1487                                                 tunnel_proto);
1488                 break;
1489         }
1490
1491 fail_op:
1492 fail_bad_op:
1493         sfc_adapter_unlock(sa);
1494
1495 fail_bad_proto:
1496         SFC_ASSERT(rc > 0);
1497         return -rc;
1498 }
1499
1500 static int
1501 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
1502                             struct rte_eth_udp_tunnel *tunnel_udp)
1503 {
1504         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT);
1505 }
1506
1507 static int
1508 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1509                             struct rte_eth_udp_tunnel *tunnel_udp)
1510 {
1511         return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT);
1512 }
1513
1514 /*
1515  * The function is used by the secondary process as well. It must not
1516  * use any process-local pointers from the adapter data.
1517  */
1518 static int
1519 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
1520                           struct rte_eth_rss_conf *rss_conf)
1521 {
1522         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1523         struct sfc_rss *rss = &sas->rss;
1524
1525         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE)
1526                 return -ENOTSUP;
1527
1528         /*
1529          * Mapping of hash configuration between RTE and EFX is not one-to-one,
1530          * hence, conversion is done here to derive a correct set of ETH_RSS
1531          * flags which corresponds to the active EFX configuration stored
1532          * locally in 'sfc_adapter' and kept up-to-date
1533          */
1534         rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types);
1535         rss_conf->rss_key_len = EFX_RSS_KEY_SIZE;
1536         if (rss_conf->rss_key != NULL)
1537                 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE);
1538
1539         return 0;
1540 }
1541
1542 static int
1543 sfc_dev_rss_hash_update(struct rte_eth_dev *dev,
1544                         struct rte_eth_rss_conf *rss_conf)
1545 {
1546         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1547         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1548         unsigned int efx_hash_types;
1549         uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context};
1550         unsigned int n_contexts;
1551         unsigned int mode_i = 0;
1552         unsigned int key_i = 0;
1553         unsigned int i = 0;
1554         int rc = 0;
1555
1556         n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2;
1557
1558         if (sfc_sa2shared(sa)->isolated)
1559                 return -ENOTSUP;
1560
1561         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1562                 sfc_err(sa, "RSS is not available");
1563                 return -ENOTSUP;
1564         }
1565
1566         if (rss->channels == 0) {
1567                 sfc_err(sa, "RSS is not configured");
1568                 return -EINVAL;
1569         }
1570
1571         if ((rss_conf->rss_key != NULL) &&
1572             (rss_conf->rss_key_len != sizeof(rss->key))) {
1573                 sfc_err(sa, "RSS key size is wrong (should be %zu)",
1574                         sizeof(rss->key));
1575                 return -EINVAL;
1576         }
1577
1578         sfc_adapter_lock(sa);
1579
1580         rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types);
1581         if (rc != 0)
1582                 goto fail_rx_hf_rte_to_efx;
1583
1584         for (mode_i = 0; mode_i < n_contexts; mode_i++) {
1585                 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i],
1586                                            rss->hash_alg, efx_hash_types,
1587                                            B_TRUE);
1588                 if (rc != 0)
1589                         goto fail_scale_mode_set;
1590         }
1591
1592         if (rss_conf->rss_key != NULL) {
1593                 if (sa->state == SFC_ADAPTER_STARTED) {
1594                         for (key_i = 0; key_i < n_contexts; key_i++) {
1595                                 rc = efx_rx_scale_key_set(sa->nic,
1596                                                           contexts[key_i],
1597                                                           rss_conf->rss_key,
1598                                                           sizeof(rss->key));
1599                                 if (rc != 0)
1600                                         goto fail_scale_key_set;
1601                         }
1602                 }
1603
1604                 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key));
1605         }
1606
1607         rss->hash_types = efx_hash_types;
1608
1609         sfc_adapter_unlock(sa);
1610
1611         return 0;
1612
1613 fail_scale_key_set:
1614         for (i = 0; i < key_i; i++) {
1615                 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key,
1616                                          sizeof(rss->key)) != 0)
1617                         sfc_err(sa, "failed to restore RSS key");
1618         }
1619
1620 fail_scale_mode_set:
1621         for (i = 0; i < mode_i; i++) {
1622                 if (efx_rx_scale_mode_set(sa->nic, contexts[i],
1623                                           EFX_RX_HASHALG_TOEPLITZ,
1624                                           rss->hash_types, B_TRUE) != 0)
1625                         sfc_err(sa, "failed to restore RSS mode");
1626         }
1627
1628 fail_rx_hf_rte_to_efx:
1629         sfc_adapter_unlock(sa);
1630         return -rc;
1631 }
1632
1633 /*
1634  * The function is used by the secondary process as well. It must not
1635  * use any process-local pointers from the adapter data.
1636  */
1637 static int
1638 sfc_dev_rss_reta_query(struct rte_eth_dev *dev,
1639                        struct rte_eth_rss_reta_entry64 *reta_conf,
1640                        uint16_t reta_size)
1641 {
1642         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1643         struct sfc_rss *rss = &sas->rss;
1644         int entry;
1645
1646         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated)
1647                 return -ENOTSUP;
1648
1649         if (rss->channels == 0)
1650                 return -EINVAL;
1651
1652         if (reta_size != EFX_RSS_TBL_SIZE)
1653                 return -EINVAL;
1654
1655         for (entry = 0; entry < reta_size; entry++) {
1656                 int grp = entry / RTE_RETA_GROUP_SIZE;
1657                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1658
1659                 if ((reta_conf[grp].mask >> grp_idx) & 1)
1660                         reta_conf[grp].reta[grp_idx] = rss->tbl[entry];
1661         }
1662
1663         return 0;
1664 }
1665
1666 static int
1667 sfc_dev_rss_reta_update(struct rte_eth_dev *dev,
1668                         struct rte_eth_rss_reta_entry64 *reta_conf,
1669                         uint16_t reta_size)
1670 {
1671         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1672         struct sfc_rss *rss = &sfc_sa2shared(sa)->rss;
1673         unsigned int *rss_tbl_new;
1674         uint16_t entry;
1675         int rc = 0;
1676
1677
1678         if (sfc_sa2shared(sa)->isolated)
1679                 return -ENOTSUP;
1680
1681         if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) {
1682                 sfc_err(sa, "RSS is not available");
1683                 return -ENOTSUP;
1684         }
1685
1686         if (rss->channels == 0) {
1687                 sfc_err(sa, "RSS is not configured");
1688                 return -EINVAL;
1689         }
1690
1691         if (reta_size != EFX_RSS_TBL_SIZE) {
1692                 sfc_err(sa, "RETA size is wrong (should be %u)",
1693                         EFX_RSS_TBL_SIZE);
1694                 return -EINVAL;
1695         }
1696
1697         rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0);
1698         if (rss_tbl_new == NULL)
1699                 return -ENOMEM;
1700
1701         sfc_adapter_lock(sa);
1702
1703         rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl));
1704
1705         for (entry = 0; entry < reta_size; entry++) {
1706                 int grp_idx = entry % RTE_RETA_GROUP_SIZE;
1707                 struct rte_eth_rss_reta_entry64 *grp;
1708
1709                 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE];
1710
1711                 if (grp->mask & (1ull << grp_idx)) {
1712                         if (grp->reta[grp_idx] >= rss->channels) {
1713                                 rc = EINVAL;
1714                                 goto bad_reta_entry;
1715                         }
1716                         rss_tbl_new[entry] = grp->reta[grp_idx];
1717                 }
1718         }
1719
1720         if (sa->state == SFC_ADAPTER_STARTED) {
1721                 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT,
1722                                           rss_tbl_new, EFX_RSS_TBL_SIZE);
1723                 if (rc != 0)
1724                         goto fail_scale_tbl_set;
1725         }
1726
1727         rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl));
1728
1729 fail_scale_tbl_set:
1730 bad_reta_entry:
1731         sfc_adapter_unlock(sa);
1732
1733         rte_free(rss_tbl_new);
1734
1735         SFC_ASSERT(rc >= 0);
1736         return -rc;
1737 }
1738
1739 static int
1740 sfc_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type,
1741                     enum rte_filter_op filter_op,
1742                     void *arg)
1743 {
1744         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1745         int rc = ENOTSUP;
1746
1747         sfc_log_init(sa, "entry");
1748
1749         switch (filter_type) {
1750         case RTE_ETH_FILTER_NONE:
1751                 sfc_err(sa, "Global filters configuration not supported");
1752                 break;
1753         case RTE_ETH_FILTER_MACVLAN:
1754                 sfc_err(sa, "MACVLAN filters not supported");
1755                 break;
1756         case RTE_ETH_FILTER_ETHERTYPE:
1757                 sfc_err(sa, "EtherType filters not supported");
1758                 break;
1759         case RTE_ETH_FILTER_FLEXIBLE:
1760                 sfc_err(sa, "Flexible filters not supported");
1761                 break;
1762         case RTE_ETH_FILTER_SYN:
1763                 sfc_err(sa, "SYN filters not supported");
1764                 break;
1765         case RTE_ETH_FILTER_NTUPLE:
1766                 sfc_err(sa, "NTUPLE filters not supported");
1767                 break;
1768         case RTE_ETH_FILTER_TUNNEL:
1769                 sfc_err(sa, "Tunnel filters not supported");
1770                 break;
1771         case RTE_ETH_FILTER_FDIR:
1772                 sfc_err(sa, "Flow Director filters not supported");
1773                 break;
1774         case RTE_ETH_FILTER_HASH:
1775                 sfc_err(sa, "Hash filters not supported");
1776                 break;
1777         case RTE_ETH_FILTER_GENERIC:
1778                 if (filter_op != RTE_ETH_FILTER_GET) {
1779                         rc = EINVAL;
1780                 } else {
1781                         *(const void **)arg = &sfc_flow_ops;
1782                         rc = 0;
1783                 }
1784                 break;
1785         default:
1786                 sfc_err(sa, "Unknown filter type %u", filter_type);
1787                 break;
1788         }
1789
1790         sfc_log_init(sa, "exit: %d", -rc);
1791         SFC_ASSERT(rc >= 0);
1792         return -rc;
1793 }
1794
1795 static int
1796 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool)
1797 {
1798         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1799
1800         /*
1801          * If Rx datapath does not provide callback to check mempool,
1802          * all pools are supported.
1803          */
1804         if (sap->dp_rx->pool_ops_supported == NULL)
1805                 return 1;
1806
1807         return sap->dp_rx->pool_ops_supported(pool);
1808 }
1809
1810 static int
1811 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t queue_id)
1812 {
1813         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1814         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1815         struct sfc_rxq_info *rxq_info;
1816
1817         SFC_ASSERT(queue_id < sas->rxq_count);
1818         rxq_info = &sas->rxq_info[queue_id];
1819
1820         return sap->dp_rx->intr_enable(rxq_info->dp);
1821 }
1822
1823 static int
1824 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t queue_id)
1825 {
1826         const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev);
1827         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1828         struct sfc_rxq_info *rxq_info;
1829
1830         SFC_ASSERT(queue_id < sas->rxq_count);
1831         rxq_info = &sas->rxq_info[queue_id];
1832
1833         return sap->dp_rx->intr_disable(rxq_info->dp);
1834 }
1835
1836 static const struct eth_dev_ops sfc_eth_dev_ops = {
1837         .dev_configure                  = sfc_dev_configure,
1838         .dev_start                      = sfc_dev_start,
1839         .dev_stop                       = sfc_dev_stop,
1840         .dev_set_link_up                = sfc_dev_set_link_up,
1841         .dev_set_link_down              = sfc_dev_set_link_down,
1842         .dev_close                      = sfc_dev_close,
1843         .promiscuous_enable             = sfc_dev_promisc_enable,
1844         .promiscuous_disable            = sfc_dev_promisc_disable,
1845         .allmulticast_enable            = sfc_dev_allmulti_enable,
1846         .allmulticast_disable           = sfc_dev_allmulti_disable,
1847         .link_update                    = sfc_dev_link_update,
1848         .stats_get                      = sfc_stats_get,
1849         .stats_reset                    = sfc_stats_reset,
1850         .xstats_get                     = sfc_xstats_get,
1851         .xstats_reset                   = sfc_stats_reset,
1852         .xstats_get_names               = sfc_xstats_get_names,
1853         .dev_infos_get                  = sfc_dev_infos_get,
1854         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
1855         .mtu_set                        = sfc_dev_set_mtu,
1856         .rx_queue_start                 = sfc_rx_queue_start,
1857         .rx_queue_stop                  = sfc_rx_queue_stop,
1858         .tx_queue_start                 = sfc_tx_queue_start,
1859         .tx_queue_stop                  = sfc_tx_queue_stop,
1860         .rx_queue_setup                 = sfc_rx_queue_setup,
1861         .rx_queue_release               = sfc_rx_queue_release,
1862         .rx_queue_intr_enable           = sfc_rx_queue_intr_enable,
1863         .rx_queue_intr_disable          = sfc_rx_queue_intr_disable,
1864         .tx_queue_setup                 = sfc_tx_queue_setup,
1865         .tx_queue_release               = sfc_tx_queue_release,
1866         .flow_ctrl_get                  = sfc_flow_ctrl_get,
1867         .flow_ctrl_set                  = sfc_flow_ctrl_set,
1868         .mac_addr_set                   = sfc_mac_addr_set,
1869         .udp_tunnel_port_add            = sfc_dev_udp_tunnel_port_add,
1870         .udp_tunnel_port_del            = sfc_dev_udp_tunnel_port_del,
1871         .reta_update                    = sfc_dev_rss_reta_update,
1872         .reta_query                     = sfc_dev_rss_reta_query,
1873         .rss_hash_update                = sfc_dev_rss_hash_update,
1874         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
1875         .filter_ctrl                    = sfc_dev_filter_ctrl,
1876         .set_mc_addr_list               = sfc_set_mc_addr_list,
1877         .rxq_info_get                   = sfc_rx_queue_info_get,
1878         .txq_info_get                   = sfc_tx_queue_info_get,
1879         .fw_version_get                 = sfc_fw_version_get,
1880         .xstats_get_by_id               = sfc_xstats_get_by_id,
1881         .xstats_get_names_by_id         = sfc_xstats_get_names_by_id,
1882         .pool_ops_supported             = sfc_pool_ops_supported,
1883 };
1884
1885 /**
1886  * Duplicate a string in potentially shared memory required for
1887  * multi-process support.
1888  *
1889  * strdup() allocates from process-local heap/memory.
1890  */
1891 static char *
1892 sfc_strdup(const char *str)
1893 {
1894         size_t size;
1895         char *copy;
1896
1897         if (str == NULL)
1898                 return NULL;
1899
1900         size = strlen(str) + 1;
1901         copy = rte_malloc(__func__, size, 0);
1902         if (copy != NULL)
1903                 rte_memcpy(copy, str, size);
1904
1905         return copy;
1906 }
1907
1908 static int
1909 sfc_eth_dev_set_ops(struct rte_eth_dev *dev)
1910 {
1911         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
1912         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
1913         const struct sfc_dp_rx *dp_rx;
1914         const struct sfc_dp_tx *dp_tx;
1915         const efx_nic_cfg_t *encp;
1916         unsigned int avail_caps = 0;
1917         const char *rx_name = NULL;
1918         const char *tx_name = NULL;
1919         int rc;
1920
1921         switch (sa->family) {
1922         case EFX_FAMILY_HUNTINGTON:
1923         case EFX_FAMILY_MEDFORD:
1924         case EFX_FAMILY_MEDFORD2:
1925                 avail_caps |= SFC_DP_HW_FW_CAP_EF10;
1926                 break;
1927         default:
1928                 break;
1929         }
1930
1931         encp = efx_nic_cfg_get(sa->nic);
1932         if (encp->enc_rx_es_super_buffer_supported)
1933                 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER;
1934
1935         rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH,
1936                                 sfc_kvarg_string_handler, &rx_name);
1937         if (rc != 0)
1938                 goto fail_kvarg_rx_datapath;
1939
1940         if (rx_name != NULL) {
1941                 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name);
1942                 if (dp_rx == NULL) {
1943                         sfc_err(sa, "Rx datapath %s not found", rx_name);
1944                         rc = ENOENT;
1945                         goto fail_dp_rx;
1946                 }
1947                 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) {
1948                         sfc_err(sa,
1949                                 "Insufficient Hw/FW capabilities to use Rx datapath %s",
1950                                 rx_name);
1951                         rc = EINVAL;
1952                         goto fail_dp_rx_caps;
1953                 }
1954         } else {
1955                 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps);
1956                 if (dp_rx == NULL) {
1957                         sfc_err(sa, "Rx datapath by caps %#x not found",
1958                                 avail_caps);
1959                         rc = ENOENT;
1960                         goto fail_dp_rx;
1961                 }
1962         }
1963
1964         sas->dp_rx_name = sfc_strdup(dp_rx->dp.name);
1965         if (sas->dp_rx_name == NULL) {
1966                 rc = ENOMEM;
1967                 goto fail_dp_rx_name;
1968         }
1969
1970         sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name);
1971
1972         rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH,
1973                                 sfc_kvarg_string_handler, &tx_name);
1974         if (rc != 0)
1975                 goto fail_kvarg_tx_datapath;
1976
1977         if (tx_name != NULL) {
1978                 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name);
1979                 if (dp_tx == NULL) {
1980                         sfc_err(sa, "Tx datapath %s not found", tx_name);
1981                         rc = ENOENT;
1982                         goto fail_dp_tx;
1983                 }
1984                 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) {
1985                         sfc_err(sa,
1986                                 "Insufficient Hw/FW capabilities to use Tx datapath %s",
1987                                 tx_name);
1988                         rc = EINVAL;
1989                         goto fail_dp_tx_caps;
1990                 }
1991         } else {
1992                 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps);
1993                 if (dp_tx == NULL) {
1994                         sfc_err(sa, "Tx datapath by caps %#x not found",
1995                                 avail_caps);
1996                         rc = ENOENT;
1997                         goto fail_dp_tx;
1998                 }
1999         }
2000
2001         sas->dp_tx_name = sfc_strdup(dp_tx->dp.name);
2002         if (sas->dp_tx_name == NULL) {
2003                 rc = ENOMEM;
2004                 goto fail_dp_tx_name;
2005         }
2006
2007         sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name);
2008
2009         sa->priv.dp_rx = dp_rx;
2010         sa->priv.dp_tx = dp_tx;
2011
2012         dev->rx_pkt_burst = dp_rx->pkt_burst;
2013         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2014         dev->tx_pkt_burst = dp_tx->pkt_burst;
2015
2016         dev->rx_queue_count = sfc_rx_queue_count;
2017         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2018         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2019         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2020         dev->dev_ops = &sfc_eth_dev_ops;
2021
2022         return 0;
2023
2024 fail_dp_tx_name:
2025 fail_dp_tx_caps:
2026 fail_dp_tx:
2027 fail_kvarg_tx_datapath:
2028         rte_free(sas->dp_rx_name);
2029         sas->dp_rx_name = NULL;
2030
2031 fail_dp_rx_name:
2032 fail_dp_rx_caps:
2033 fail_dp_rx:
2034 fail_kvarg_rx_datapath:
2035         return rc;
2036 }
2037
2038 static void
2039 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev)
2040 {
2041         struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev);
2042         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2043
2044         dev->dev_ops = NULL;
2045         dev->tx_pkt_prepare = NULL;
2046         dev->rx_pkt_burst = NULL;
2047         dev->tx_pkt_burst = NULL;
2048
2049         rte_free(sas->dp_tx_name);
2050         sas->dp_tx_name = NULL;
2051         sa->priv.dp_tx = NULL;
2052
2053         rte_free(sas->dp_rx_name);
2054         sas->dp_rx_name = NULL;
2055         sa->priv.dp_rx = NULL;
2056 }
2057
2058 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = {
2059         .dev_supported_ptypes_get       = sfc_dev_supported_ptypes_get,
2060         .reta_query                     = sfc_dev_rss_reta_query,
2061         .rss_hash_conf_get              = sfc_dev_rss_hash_conf_get,
2062         .rxq_info_get                   = sfc_rx_queue_info_get,
2063         .txq_info_get                   = sfc_tx_queue_info_get,
2064 };
2065
2066 static int
2067 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main)
2068 {
2069         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2070         struct sfc_adapter_priv *sap;
2071         const struct sfc_dp_rx *dp_rx;
2072         const struct sfc_dp_tx *dp_tx;
2073         int rc;
2074
2075         /*
2076          * Allocate process private data from heap, since it should not
2077          * be located in shared memory allocated using rte_malloc() API.
2078          */
2079         sap = calloc(1, sizeof(*sap));
2080         if (sap == NULL) {
2081                 rc = ENOMEM;
2082                 goto fail_alloc_priv;
2083         }
2084
2085         sap->logtype_main = logtype_main;
2086
2087         dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name);
2088         if (dp_rx == NULL) {
2089                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2090                         "cannot find %s Rx datapath", sas->dp_rx_name);
2091                 rc = ENOENT;
2092                 goto fail_dp_rx;
2093         }
2094         if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) {
2095                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2096                         "%s Rx datapath does not support multi-process",
2097                         sas->dp_rx_name);
2098                 rc = EINVAL;
2099                 goto fail_dp_rx_multi_process;
2100         }
2101
2102         dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name);
2103         if (dp_tx == NULL) {
2104                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2105                         "cannot find %s Tx datapath", sas->dp_tx_name);
2106                 rc = ENOENT;
2107                 goto fail_dp_tx;
2108         }
2109         if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) {
2110                 SFC_LOG(sas, RTE_LOG_ERR, logtype_main,
2111                         "%s Tx datapath does not support multi-process",
2112                         sas->dp_tx_name);
2113                 rc = EINVAL;
2114                 goto fail_dp_tx_multi_process;
2115         }
2116
2117         sap->dp_rx = dp_rx;
2118         sap->dp_tx = dp_tx;
2119
2120         dev->process_private = sap;
2121         dev->rx_pkt_burst = dp_rx->pkt_burst;
2122         dev->tx_pkt_prepare = dp_tx->pkt_prepare;
2123         dev->tx_pkt_burst = dp_tx->pkt_burst;
2124         dev->rx_queue_count = sfc_rx_queue_count;
2125         dev->rx_descriptor_done = sfc_rx_descriptor_done;
2126         dev->rx_descriptor_status = sfc_rx_descriptor_status;
2127         dev->tx_descriptor_status = sfc_tx_descriptor_status;
2128         dev->dev_ops = &sfc_eth_dev_secondary_ops;
2129
2130         return 0;
2131
2132 fail_dp_tx_multi_process:
2133 fail_dp_tx:
2134 fail_dp_rx_multi_process:
2135 fail_dp_rx:
2136         free(sap);
2137
2138 fail_alloc_priv:
2139         return rc;
2140 }
2141
2142 static void
2143 sfc_register_dp(void)
2144 {
2145         /* Register once */
2146         if (TAILQ_EMPTY(&sfc_dp_head)) {
2147                 /* Prefer EF10 datapath */
2148                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp);
2149                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp);
2150                 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp);
2151
2152                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp);
2153                 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp);
2154                 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp);
2155         }
2156 }
2157
2158 static int
2159 sfc_eth_dev_init(struct rte_eth_dev *dev)
2160 {
2161         struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev);
2162         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
2163         uint32_t logtype_main;
2164         struct sfc_adapter *sa;
2165         int rc;
2166         const efx_nic_cfg_t *encp;
2167         const struct rte_ether_addr *from;
2168         int ret;
2169
2170         sfc_register_dp();
2171
2172         logtype_main = sfc_register_logtype(&pci_dev->addr,
2173                                             SFC_LOGTYPE_MAIN_STR,
2174                                             RTE_LOG_NOTICE);
2175
2176         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2177                 return -sfc_eth_dev_secondary_init(dev, logtype_main);
2178
2179         /* Required for logging */
2180         ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix),
2181                         "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ",
2182                         pci_dev->addr.domain, pci_dev->addr.bus,
2183                         pci_dev->addr.devid, pci_dev->addr.function,
2184                         dev->data->port_id);
2185         if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) {
2186                 SFC_GENERIC_LOG(ERR,
2187                         "reserved log prefix is too short for " PCI_PRI_FMT,
2188                         pci_dev->addr.domain, pci_dev->addr.bus,
2189                         pci_dev->addr.devid, pci_dev->addr.function);
2190                 return -EINVAL;
2191         }
2192         sas->pci_addr = pci_dev->addr;
2193         sas->port_id = dev->data->port_id;
2194
2195         /*
2196          * Allocate process private data from heap, since it should not
2197          * be located in shared memory allocated using rte_malloc() API.
2198          */
2199         sa = calloc(1, sizeof(*sa));
2200         if (sa == NULL) {
2201                 rc = ENOMEM;
2202                 goto fail_alloc_sa;
2203         }
2204
2205         dev->process_private = sa;
2206
2207         /* Required for logging */
2208         sa->priv.shared = sas;
2209         sa->priv.logtype_main = logtype_main;
2210
2211         sa->eth_dev = dev;
2212
2213         /* Copy PCI device info to the dev->data */
2214         rte_eth_copy_pci_info(dev, pci_dev);
2215
2216         rc = sfc_kvargs_parse(sa);
2217         if (rc != 0)
2218                 goto fail_kvargs_parse;
2219
2220         sfc_log_init(sa, "entry");
2221
2222         dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0);
2223         if (dev->data->mac_addrs == NULL) {
2224                 rc = ENOMEM;
2225                 goto fail_mac_addrs;
2226         }
2227
2228         sfc_adapter_lock_init(sa);
2229         sfc_adapter_lock(sa);
2230
2231         sfc_log_init(sa, "probing");
2232         rc = sfc_probe(sa);
2233         if (rc != 0)
2234                 goto fail_probe;
2235
2236         sfc_log_init(sa, "set device ops");
2237         rc = sfc_eth_dev_set_ops(dev);
2238         if (rc != 0)
2239                 goto fail_set_ops;
2240
2241         sfc_log_init(sa, "attaching");
2242         rc = sfc_attach(sa);
2243         if (rc != 0)
2244                 goto fail_attach;
2245
2246         encp = efx_nic_cfg_get(sa->nic);
2247
2248         /*
2249          * The arguments are really reverse order in comparison to
2250          * Linux kernel. Copy from NIC config to Ethernet device data.
2251          */
2252         from = (const struct rte_ether_addr *)(encp->enc_mac_addr);
2253         rte_ether_addr_copy(from, &dev->data->mac_addrs[0]);
2254
2255         sfc_adapter_unlock(sa);
2256
2257         sfc_log_init(sa, "done");
2258         return 0;
2259
2260 fail_attach:
2261         sfc_eth_dev_clear_ops(dev);
2262
2263 fail_set_ops:
2264         sfc_unprobe(sa);
2265
2266 fail_probe:
2267         sfc_adapter_unlock(sa);
2268         sfc_adapter_lock_fini(sa);
2269         rte_free(dev->data->mac_addrs);
2270         dev->data->mac_addrs = NULL;
2271
2272 fail_mac_addrs:
2273         sfc_kvargs_cleanup(sa);
2274
2275 fail_kvargs_parse:
2276         sfc_log_init(sa, "failed %d", rc);
2277         dev->process_private = NULL;
2278         free(sa);
2279
2280 fail_alloc_sa:
2281         SFC_ASSERT(rc > 0);
2282         return -rc;
2283 }
2284
2285 static int
2286 sfc_eth_dev_uninit(struct rte_eth_dev *dev)
2287 {
2288         sfc_dev_close(dev);
2289
2290         return 0;
2291 }
2292
2293 static const struct rte_pci_id pci_id_sfc_efx_map[] = {
2294         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) },
2295         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) },
2296         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) },
2297         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) },
2298         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) },
2299         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) },
2300         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) },
2301         { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) },
2302         { .vendor_id = 0 /* sentinel */ }
2303 };
2304
2305 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2306         struct rte_pci_device *pci_dev)
2307 {
2308         return rte_eth_dev_pci_generic_probe(pci_dev,
2309                 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init);
2310 }
2311
2312 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev)
2313 {
2314         return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit);
2315 }
2316
2317 static struct rte_pci_driver sfc_efx_pmd = {
2318         .id_table = pci_id_sfc_efx_map,
2319         .drv_flags =
2320                 RTE_PCI_DRV_INTR_LSC |
2321                 RTE_PCI_DRV_NEED_MAPPING,
2322         .probe = sfc_eth_dev_pci_probe,
2323         .remove = sfc_eth_dev_pci_remove,
2324 };
2325
2326 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd);
2327 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map);
2328 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci");
2329 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx,
2330         SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " "
2331         SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " "
2332         SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " "
2333         SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " "
2334         SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> "
2335         SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>");
2336
2337 RTE_INIT(sfc_driver_register_logtype)
2338 {
2339         int ret;
2340
2341         ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver",
2342                                                    RTE_LOG_NOTICE);
2343         sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret;
2344 }