1 /* SPDX-License-Identifier: BSD-3-Clause 2 * 3 * Copyright(c) 2019-2021 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 <ethdev_driver.h> 12 #include <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 rc = efx_nic_get_fw_version(sa->nic, &enfi); 50 if (rc != 0) 51 return -rc; 52 53 ret = snprintf(fw_version, fw_size, 54 "%" PRIu16 ".%" PRIu16 ".%" PRIu16 ".%" PRIu16, 55 enfi.enfi_mc_fw_version[0], enfi.enfi_mc_fw_version[1], 56 enfi.enfi_mc_fw_version[2], enfi.enfi_mc_fw_version[3]); 57 if (ret < 0) 58 return ret; 59 60 if (enfi.enfi_dpcpu_fw_ids_valid) { 61 size_t dpcpu_fw_ids_offset = MIN(fw_size - 1, (size_t)ret); 62 int ret_extra; 63 64 ret_extra = snprintf(fw_version + dpcpu_fw_ids_offset, 65 fw_size - dpcpu_fw_ids_offset, 66 " rx%" PRIx16 " tx%" PRIx16, 67 enfi.enfi_rx_dpcpu_fw_id, 68 enfi.enfi_tx_dpcpu_fw_id); 69 if (ret_extra < 0) 70 return ret_extra; 71 72 ret += ret_extra; 73 } 74 75 if (fw_size < (size_t)(++ret)) 76 return ret; 77 else 78 return 0; 79 } 80 81 static int 82 sfc_dev_infos_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info) 83 { 84 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev); 85 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 86 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 87 struct sfc_rss *rss = &sas->rss; 88 struct sfc_mae *mae = &sa->mae; 89 uint64_t txq_offloads_def = 0; 90 91 sfc_log_init(sa, "entry"); 92 93 dev_info->min_mtu = RTE_ETHER_MIN_MTU; 94 dev_info->max_mtu = EFX_MAC_SDU_MAX; 95 96 dev_info->max_rx_pktlen = EFX_MAC_PDU_MAX; 97 98 dev_info->max_vfs = sa->sriov.num_vfs; 99 100 /* Autonegotiation may be disabled */ 101 dev_info->speed_capa = ETH_LINK_SPEED_FIXED; 102 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_1000FDX)) 103 dev_info->speed_capa |= ETH_LINK_SPEED_1G; 104 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_10000FDX)) 105 dev_info->speed_capa |= ETH_LINK_SPEED_10G; 106 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_25000FDX)) 107 dev_info->speed_capa |= ETH_LINK_SPEED_25G; 108 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_40000FDX)) 109 dev_info->speed_capa |= ETH_LINK_SPEED_40G; 110 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_50000FDX)) 111 dev_info->speed_capa |= ETH_LINK_SPEED_50G; 112 if (sa->port.phy_adv_cap_mask & (1u << EFX_PHY_CAP_100000FDX)) 113 dev_info->speed_capa |= ETH_LINK_SPEED_100G; 114 115 dev_info->max_rx_queues = sa->rxq_max; 116 dev_info->max_tx_queues = sa->txq_max; 117 118 /* By default packets are dropped if no descriptors are available */ 119 dev_info->default_rxconf.rx_drop_en = 1; 120 121 dev_info->rx_queue_offload_capa = sfc_rx_get_queue_offload_caps(sa); 122 123 /* 124 * rx_offload_capa includes both device and queue offloads since 125 * the latter may be requested on a per device basis which makes 126 * sense when some offloads are needed to be set on all queues. 127 */ 128 dev_info->rx_offload_capa = sfc_rx_get_dev_offload_caps(sa) | 129 dev_info->rx_queue_offload_capa; 130 131 dev_info->tx_queue_offload_capa = sfc_tx_get_queue_offload_caps(sa); 132 133 /* 134 * tx_offload_capa includes both device and queue offloads since 135 * the latter may be requested on a per device basis which makes 136 * sense when some offloads are needed to be set on all queues. 137 */ 138 dev_info->tx_offload_capa = sfc_tx_get_dev_offload_caps(sa) | 139 dev_info->tx_queue_offload_capa; 140 141 if (dev_info->tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE) 142 txq_offloads_def |= DEV_TX_OFFLOAD_MBUF_FAST_FREE; 143 144 dev_info->default_txconf.offloads |= txq_offloads_def; 145 146 if (rss->context_type != EFX_RX_SCALE_UNAVAILABLE) { 147 uint64_t rte_hf = 0; 148 unsigned int i; 149 150 for (i = 0; i < rss->hf_map_nb_entries; ++i) 151 rte_hf |= rss->hf_map[i].rte; 152 153 dev_info->reta_size = EFX_RSS_TBL_SIZE; 154 dev_info->hash_key_size = EFX_RSS_KEY_SIZE; 155 dev_info->flow_type_rss_offloads = rte_hf; 156 } 157 158 /* Initialize to hardware limits */ 159 dev_info->rx_desc_lim.nb_max = sa->rxq_max_entries; 160 dev_info->rx_desc_lim.nb_min = sa->rxq_min_entries; 161 /* The RXQ hardware requires that the descriptor count is a power 162 * of 2, but rx_desc_lim cannot properly describe that constraint. 163 */ 164 dev_info->rx_desc_lim.nb_align = sa->rxq_min_entries; 165 166 /* Initialize to hardware limits */ 167 dev_info->tx_desc_lim.nb_max = sa->txq_max_entries; 168 dev_info->tx_desc_lim.nb_min = sa->txq_min_entries; 169 /* 170 * The TXQ hardware requires that the descriptor count is a power 171 * of 2, but tx_desc_lim cannot properly describe that constraint 172 */ 173 dev_info->tx_desc_lim.nb_align = sa->txq_min_entries; 174 175 if (sap->dp_rx->get_dev_info != NULL) 176 sap->dp_rx->get_dev_info(dev_info); 177 if (sap->dp_tx->get_dev_info != NULL) 178 sap->dp_tx->get_dev_info(dev_info); 179 180 dev_info->dev_capa = RTE_ETH_DEV_CAPA_RUNTIME_RX_QUEUE_SETUP | 181 RTE_ETH_DEV_CAPA_RUNTIME_TX_QUEUE_SETUP; 182 183 if (mae->status == SFC_MAE_STATUS_SUPPORTED) { 184 dev_info->switch_info.name = dev->device->driver->name; 185 dev_info->switch_info.domain_id = mae->switch_domain_id; 186 dev_info->switch_info.port_id = mae->switch_port_id; 187 } 188 189 return 0; 190 } 191 192 static const uint32_t * 193 sfc_dev_supported_ptypes_get(struct rte_eth_dev *dev) 194 { 195 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev); 196 197 return sap->dp_rx->supported_ptypes_get(sap->shared->tunnel_encaps); 198 } 199 200 static int 201 sfc_dev_configure(struct rte_eth_dev *dev) 202 { 203 struct rte_eth_dev_data *dev_data = dev->data; 204 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 205 int rc; 206 207 sfc_log_init(sa, "entry n_rxq=%u n_txq=%u", 208 dev_data->nb_rx_queues, dev_data->nb_tx_queues); 209 210 sfc_adapter_lock(sa); 211 switch (sa->state) { 212 case SFC_ADAPTER_CONFIGURED: 213 /* FALLTHROUGH */ 214 case SFC_ADAPTER_INITIALIZED: 215 rc = sfc_configure(sa); 216 break; 217 default: 218 sfc_err(sa, "unexpected adapter state %u to configure", 219 sa->state); 220 rc = EINVAL; 221 break; 222 } 223 sfc_adapter_unlock(sa); 224 225 sfc_log_init(sa, "done %d", rc); 226 SFC_ASSERT(rc >= 0); 227 return -rc; 228 } 229 230 static int 231 sfc_dev_start(struct rte_eth_dev *dev) 232 { 233 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 234 int rc; 235 236 sfc_log_init(sa, "entry"); 237 238 sfc_adapter_lock(sa); 239 rc = sfc_start(sa); 240 sfc_adapter_unlock(sa); 241 242 sfc_log_init(sa, "done %d", rc); 243 SFC_ASSERT(rc >= 0); 244 return -rc; 245 } 246 247 static int 248 sfc_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete) 249 { 250 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 251 struct rte_eth_link current_link; 252 int ret; 253 254 sfc_log_init(sa, "entry"); 255 256 if (sa->state != SFC_ADAPTER_STARTED) { 257 sfc_port_link_mode_to_info(EFX_LINK_UNKNOWN, ¤t_link); 258 } else if (wait_to_complete) { 259 efx_link_mode_t link_mode; 260 261 if (efx_port_poll(sa->nic, &link_mode) != 0) 262 link_mode = EFX_LINK_UNKNOWN; 263 sfc_port_link_mode_to_info(link_mode, ¤t_link); 264 265 } else { 266 sfc_ev_mgmt_qpoll(sa); 267 rte_eth_linkstatus_get(dev, ¤t_link); 268 } 269 270 ret = rte_eth_linkstatus_set(dev, ¤t_link); 271 if (ret == 0) 272 sfc_notice(sa, "Link status is %s", 273 current_link.link_status ? "UP" : "DOWN"); 274 275 return ret; 276 } 277 278 static int 279 sfc_dev_stop(struct rte_eth_dev *dev) 280 { 281 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 282 283 sfc_log_init(sa, "entry"); 284 285 sfc_adapter_lock(sa); 286 sfc_stop(sa); 287 sfc_adapter_unlock(sa); 288 289 sfc_log_init(sa, "done"); 290 291 return 0; 292 } 293 294 static int 295 sfc_dev_set_link_up(struct rte_eth_dev *dev) 296 { 297 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 298 int rc; 299 300 sfc_log_init(sa, "entry"); 301 302 sfc_adapter_lock(sa); 303 rc = sfc_start(sa); 304 sfc_adapter_unlock(sa); 305 306 SFC_ASSERT(rc >= 0); 307 return -rc; 308 } 309 310 static int 311 sfc_dev_set_link_down(struct rte_eth_dev *dev) 312 { 313 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 314 315 sfc_log_init(sa, "entry"); 316 317 sfc_adapter_lock(sa); 318 sfc_stop(sa); 319 sfc_adapter_unlock(sa); 320 321 return 0; 322 } 323 324 static void 325 sfc_eth_dev_secondary_clear_ops(struct rte_eth_dev *dev) 326 { 327 free(dev->process_private); 328 rte_eth_dev_release_port(dev); 329 } 330 331 static int 332 sfc_dev_close(struct rte_eth_dev *dev) 333 { 334 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 335 336 sfc_log_init(sa, "entry"); 337 338 if (rte_eal_process_type() != RTE_PROC_PRIMARY) { 339 sfc_eth_dev_secondary_clear_ops(dev); 340 return 0; 341 } 342 343 sfc_adapter_lock(sa); 344 switch (sa->state) { 345 case SFC_ADAPTER_STARTED: 346 sfc_stop(sa); 347 SFC_ASSERT(sa->state == SFC_ADAPTER_CONFIGURED); 348 /* FALLTHROUGH */ 349 case SFC_ADAPTER_CONFIGURED: 350 sfc_close(sa); 351 SFC_ASSERT(sa->state == SFC_ADAPTER_INITIALIZED); 352 /* FALLTHROUGH */ 353 case SFC_ADAPTER_INITIALIZED: 354 break; 355 default: 356 sfc_err(sa, "unexpected adapter state %u on close", sa->state); 357 break; 358 } 359 360 /* 361 * Cleanup all resources. 362 * Rollback primary process sfc_eth_dev_init() below. 363 */ 364 365 sfc_eth_dev_clear_ops(dev); 366 367 sfc_detach(sa); 368 sfc_unprobe(sa); 369 370 sfc_kvargs_cleanup(sa); 371 372 sfc_adapter_unlock(sa); 373 sfc_adapter_lock_fini(sa); 374 375 sfc_log_init(sa, "done"); 376 377 /* Required for logging, so cleanup last */ 378 sa->eth_dev = NULL; 379 380 free(sa); 381 382 return 0; 383 } 384 385 static int 386 sfc_dev_filter_set(struct rte_eth_dev *dev, enum sfc_dev_filter_mode mode, 387 boolean_t enabled) 388 { 389 struct sfc_port *port; 390 boolean_t *toggle; 391 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 392 boolean_t allmulti = (mode == SFC_DEV_FILTER_MODE_ALLMULTI); 393 const char *desc = (allmulti) ? "all-multi" : "promiscuous"; 394 int rc = 0; 395 396 sfc_adapter_lock(sa); 397 398 port = &sa->port; 399 toggle = (allmulti) ? (&port->allmulti) : (&port->promisc); 400 401 if (*toggle != enabled) { 402 *toggle = enabled; 403 404 if (sfc_sa2shared(sa)->isolated) { 405 sfc_warn(sa, "isolated mode is active on the port"); 406 sfc_warn(sa, "the change is to be applied on the next " 407 "start provided that isolated mode is " 408 "disabled prior the next start"); 409 } else if ((sa->state == SFC_ADAPTER_STARTED) && 410 ((rc = sfc_set_rx_mode(sa)) != 0)) { 411 *toggle = !(enabled); 412 sfc_warn(sa, "Failed to %s %s mode, rc = %d", 413 ((enabled) ? "enable" : "disable"), desc, rc); 414 415 /* 416 * For promiscuous and all-multicast filters a 417 * permission failure should be reported as an 418 * unsupported filter. 419 */ 420 if (rc == EPERM) 421 rc = ENOTSUP; 422 } 423 } 424 425 sfc_adapter_unlock(sa); 426 return rc; 427 } 428 429 static int 430 sfc_dev_promisc_enable(struct rte_eth_dev *dev) 431 { 432 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_TRUE); 433 434 SFC_ASSERT(rc >= 0); 435 return -rc; 436 } 437 438 static int 439 sfc_dev_promisc_disable(struct rte_eth_dev *dev) 440 { 441 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_PROMISC, B_FALSE); 442 443 SFC_ASSERT(rc >= 0); 444 return -rc; 445 } 446 447 static int 448 sfc_dev_allmulti_enable(struct rte_eth_dev *dev) 449 { 450 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_TRUE); 451 452 SFC_ASSERT(rc >= 0); 453 return -rc; 454 } 455 456 static int 457 sfc_dev_allmulti_disable(struct rte_eth_dev *dev) 458 { 459 int rc = sfc_dev_filter_set(dev, SFC_DEV_FILTER_MODE_ALLMULTI, B_FALSE); 460 461 SFC_ASSERT(rc >= 0); 462 return -rc; 463 } 464 465 static int 466 sfc_rx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid, 467 uint16_t nb_rx_desc, unsigned int socket_id, 468 const struct rte_eth_rxconf *rx_conf, 469 struct rte_mempool *mb_pool) 470 { 471 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 472 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 473 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 474 struct sfc_rxq_info *rxq_info; 475 sfc_sw_index_t sw_index; 476 int rc; 477 478 sfc_log_init(sa, "RxQ=%u nb_rx_desc=%u socket_id=%u", 479 ethdev_qid, nb_rx_desc, socket_id); 480 481 sfc_adapter_lock(sa); 482 483 sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid); 484 rc = sfc_rx_qinit(sa, sw_index, nb_rx_desc, socket_id, 485 rx_conf, mb_pool); 486 if (rc != 0) 487 goto fail_rx_qinit; 488 489 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 490 dev->data->rx_queues[ethdev_qid] = rxq_info->dp; 491 492 sfc_adapter_unlock(sa); 493 494 return 0; 495 496 fail_rx_qinit: 497 sfc_adapter_unlock(sa); 498 SFC_ASSERT(rc > 0); 499 return -rc; 500 } 501 502 static void 503 sfc_rx_queue_release(void *queue) 504 { 505 struct sfc_dp_rxq *dp_rxq = queue; 506 struct sfc_rxq *rxq; 507 struct sfc_adapter *sa; 508 sfc_sw_index_t sw_index; 509 510 if (dp_rxq == NULL) 511 return; 512 513 rxq = sfc_rxq_by_dp_rxq(dp_rxq); 514 sa = rxq->evq->sa; 515 sfc_adapter_lock(sa); 516 517 sw_index = dp_rxq->dpq.queue_id; 518 519 sfc_log_init(sa, "RxQ=%u", sw_index); 520 521 sfc_rx_qfini(sa, sw_index); 522 523 sfc_adapter_unlock(sa); 524 } 525 526 static int 527 sfc_tx_queue_setup(struct rte_eth_dev *dev, uint16_t ethdev_qid, 528 uint16_t nb_tx_desc, unsigned int socket_id, 529 const struct rte_eth_txconf *tx_conf) 530 { 531 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 532 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 533 struct sfc_txq_info *txq_info; 534 sfc_sw_index_t sw_index; 535 int rc; 536 537 sfc_log_init(sa, "TxQ = %u, nb_tx_desc = %u, socket_id = %u", 538 ethdev_qid, nb_tx_desc, socket_id); 539 540 sfc_adapter_lock(sa); 541 542 sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid); 543 rc = sfc_tx_qinit(sa, sw_index, nb_tx_desc, socket_id, tx_conf); 544 if (rc != 0) 545 goto fail_tx_qinit; 546 547 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid); 548 dev->data->tx_queues[ethdev_qid] = txq_info->dp; 549 550 sfc_adapter_unlock(sa); 551 return 0; 552 553 fail_tx_qinit: 554 sfc_adapter_unlock(sa); 555 SFC_ASSERT(rc > 0); 556 return -rc; 557 } 558 559 static void 560 sfc_tx_queue_release(void *queue) 561 { 562 struct sfc_dp_txq *dp_txq = queue; 563 struct sfc_txq *txq; 564 sfc_sw_index_t sw_index; 565 struct sfc_adapter *sa; 566 567 if (dp_txq == NULL) 568 return; 569 570 txq = sfc_txq_by_dp_txq(dp_txq); 571 sw_index = dp_txq->dpq.queue_id; 572 573 SFC_ASSERT(txq->evq != NULL); 574 sa = txq->evq->sa; 575 576 sfc_log_init(sa, "TxQ = %u", sw_index); 577 578 sfc_adapter_lock(sa); 579 580 sfc_tx_qfini(sa, sw_index); 581 582 sfc_adapter_unlock(sa); 583 } 584 585 /* 586 * Some statistics are computed as A - B where A and B each increase 587 * monotonically with some hardware counter(s) and the counters are read 588 * asynchronously. 589 * 590 * If packet X is counted in A, but not counted in B yet, computed value is 591 * greater than real. 592 * 593 * If packet X is not counted in A at the moment of reading the counter, 594 * but counted in B at the moment of reading the counter, computed value 595 * is less than real. 596 * 597 * However, counter which grows backward is worse evil than slightly wrong 598 * value. So, let's try to guarantee that it never happens except may be 599 * the case when the MAC stats are zeroed as a result of a NIC reset. 600 */ 601 static void 602 sfc_update_diff_stat(uint64_t *stat, uint64_t newval) 603 { 604 if ((int64_t)(newval - *stat) > 0 || newval == 0) 605 *stat = newval; 606 } 607 608 static int 609 sfc_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats) 610 { 611 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 612 struct sfc_port *port = &sa->port; 613 uint64_t *mac_stats; 614 int ret; 615 616 sfc_adapter_lock(sa); 617 618 ret = sfc_port_update_mac_stats(sa, B_FALSE); 619 if (ret != 0) 620 goto unlock; 621 622 mac_stats = port->mac_stats_buf; 623 624 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, 625 EFX_MAC_VADAPTER_RX_UNICAST_PACKETS)) { 626 stats->ipackets = 627 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_PACKETS] + 628 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_PACKETS] + 629 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_PACKETS]; 630 stats->opackets = 631 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_PACKETS] + 632 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_PACKETS] + 633 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_PACKETS]; 634 stats->ibytes = 635 mac_stats[EFX_MAC_VADAPTER_RX_UNICAST_BYTES] + 636 mac_stats[EFX_MAC_VADAPTER_RX_MULTICAST_BYTES] + 637 mac_stats[EFX_MAC_VADAPTER_RX_BROADCAST_BYTES]; 638 stats->obytes = 639 mac_stats[EFX_MAC_VADAPTER_TX_UNICAST_BYTES] + 640 mac_stats[EFX_MAC_VADAPTER_TX_MULTICAST_BYTES] + 641 mac_stats[EFX_MAC_VADAPTER_TX_BROADCAST_BYTES]; 642 stats->imissed = mac_stats[EFX_MAC_VADAPTER_RX_BAD_PACKETS]; 643 stats->oerrors = mac_stats[EFX_MAC_VADAPTER_TX_BAD_PACKETS]; 644 645 /* CRC is included in these stats, but shouldn't be */ 646 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN; 647 stats->obytes -= stats->opackets * RTE_ETHER_CRC_LEN; 648 } else { 649 stats->opackets = mac_stats[EFX_MAC_TX_PKTS]; 650 stats->ibytes = mac_stats[EFX_MAC_RX_OCTETS]; 651 stats->obytes = mac_stats[EFX_MAC_TX_OCTETS]; 652 653 /* CRC is included in these stats, but shouldn't be */ 654 stats->ibytes -= mac_stats[EFX_MAC_RX_PKTS] * RTE_ETHER_CRC_LEN; 655 stats->obytes -= mac_stats[EFX_MAC_TX_PKTS] * RTE_ETHER_CRC_LEN; 656 657 /* 658 * Take into account stats which are whenever supported 659 * on EF10. If some stat is not supported by current 660 * firmware variant or HW revision, it is guaranteed 661 * to be zero in mac_stats. 662 */ 663 stats->imissed = 664 mac_stats[EFX_MAC_RX_NODESC_DROP_CNT] + 665 mac_stats[EFX_MAC_PM_TRUNC_BB_OVERFLOW] + 666 mac_stats[EFX_MAC_PM_DISCARD_BB_OVERFLOW] + 667 mac_stats[EFX_MAC_PM_TRUNC_VFIFO_FULL] + 668 mac_stats[EFX_MAC_PM_DISCARD_VFIFO_FULL] + 669 mac_stats[EFX_MAC_PM_TRUNC_QBB] + 670 mac_stats[EFX_MAC_PM_DISCARD_QBB] + 671 mac_stats[EFX_MAC_PM_DISCARD_MAPPING] + 672 mac_stats[EFX_MAC_RXDP_Q_DISABLED_PKTS] + 673 mac_stats[EFX_MAC_RXDP_DI_DROPPED_PKTS]; 674 stats->ierrors = 675 mac_stats[EFX_MAC_RX_FCS_ERRORS] + 676 mac_stats[EFX_MAC_RX_ALIGN_ERRORS] + 677 mac_stats[EFX_MAC_RX_JABBER_PKTS]; 678 /* no oerrors counters supported on EF10 */ 679 680 /* Exclude missed, errors and pauses from Rx packets */ 681 sfc_update_diff_stat(&port->ipackets, 682 mac_stats[EFX_MAC_RX_PKTS] - 683 mac_stats[EFX_MAC_RX_PAUSE_PKTS] - 684 stats->imissed - stats->ierrors); 685 stats->ipackets = port->ipackets; 686 } 687 688 unlock: 689 sfc_adapter_unlock(sa); 690 SFC_ASSERT(ret >= 0); 691 return -ret; 692 } 693 694 static int 695 sfc_stats_reset(struct rte_eth_dev *dev) 696 { 697 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 698 struct sfc_port *port = &sa->port; 699 int rc; 700 701 sfc_adapter_lock(sa); 702 703 if (sa->state != SFC_ADAPTER_STARTED) { 704 /* 705 * The operation cannot be done if port is not started; it 706 * will be scheduled to be done during the next port start 707 */ 708 port->mac_stats_reset_pending = B_TRUE; 709 sfc_adapter_unlock(sa); 710 return 0; 711 } 712 713 rc = sfc_port_reset_mac_stats(sa); 714 if (rc != 0) 715 sfc_err(sa, "failed to reset statistics (rc = %d)", rc); 716 717 sfc_adapter_unlock(sa); 718 719 SFC_ASSERT(rc >= 0); 720 return -rc; 721 } 722 723 static int 724 sfc_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats, 725 unsigned int xstats_count) 726 { 727 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 728 struct sfc_port *port = &sa->port; 729 uint64_t *mac_stats; 730 int rc; 731 unsigned int i; 732 int nstats = 0; 733 734 sfc_adapter_lock(sa); 735 736 if (unlikely(xstats == NULL)) { 737 nstats = port->mac_stats_nb_supported; 738 goto unlock; 739 } 740 741 rc = sfc_port_update_mac_stats(sa, B_FALSE); 742 if (rc != 0) { 743 SFC_ASSERT(rc > 0); 744 nstats = -rc; 745 goto unlock; 746 } 747 748 mac_stats = port->mac_stats_buf; 749 750 for (i = 0; i < EFX_MAC_NSTATS; ++i) { 751 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) { 752 if (nstats < (int)xstats_count) { 753 xstats[nstats].id = nstats; 754 xstats[nstats].value = mac_stats[i]; 755 } 756 nstats++; 757 } 758 } 759 760 unlock: 761 sfc_adapter_unlock(sa); 762 763 return nstats; 764 } 765 766 static int 767 sfc_xstats_get_names(struct rte_eth_dev *dev, 768 struct rte_eth_xstat_name *xstats_names, 769 unsigned int xstats_count) 770 { 771 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 772 struct sfc_port *port = &sa->port; 773 unsigned int i; 774 unsigned int nstats = 0; 775 776 if (unlikely(xstats_names == NULL)) { 777 sfc_adapter_lock(sa); 778 nstats = port->mac_stats_nb_supported; 779 sfc_adapter_unlock(sa); 780 return nstats; 781 } 782 783 for (i = 0; i < EFX_MAC_NSTATS; ++i) { 784 if (EFX_MAC_STAT_SUPPORTED(port->mac_stats_mask, i)) { 785 if (nstats < xstats_count) 786 strlcpy(xstats_names[nstats].name, 787 efx_mac_stat_name(sa->nic, i), 788 sizeof(xstats_names[0].name)); 789 nstats++; 790 } 791 } 792 793 return nstats; 794 } 795 796 static int 797 sfc_xstats_get_by_id(struct rte_eth_dev *dev, const uint64_t *ids, 798 uint64_t *values, unsigned int n) 799 { 800 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 801 struct sfc_port *port = &sa->port; 802 uint64_t *mac_stats; 803 unsigned int i; 804 int ret; 805 int rc; 806 807 if (unlikely(ids == NULL || values == NULL)) 808 return -EINVAL; 809 810 sfc_adapter_lock(sa); 811 812 rc = sfc_port_update_mac_stats(sa, B_FALSE); 813 if (rc != 0) { 814 SFC_ASSERT(rc > 0); 815 ret = -rc; 816 goto unlock; 817 } 818 819 mac_stats = port->mac_stats_buf; 820 821 SFC_ASSERT(port->mac_stats_nb_supported <= 822 RTE_DIM(port->mac_stats_by_id)); 823 824 for (i = 0; i < n; i++) { 825 if (ids[i] < port->mac_stats_nb_supported) { 826 values[i] = mac_stats[port->mac_stats_by_id[ids[i]]]; 827 } else { 828 ret = i; 829 goto unlock; 830 } 831 } 832 833 ret = n; 834 835 unlock: 836 sfc_adapter_unlock(sa); 837 838 return ret; 839 } 840 841 static int 842 sfc_xstats_get_names_by_id(struct rte_eth_dev *dev, 843 struct rte_eth_xstat_name *xstats_names, 844 const uint64_t *ids, unsigned int size) 845 { 846 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 847 struct sfc_port *port = &sa->port; 848 unsigned int nb_supported; 849 unsigned int i; 850 851 if (unlikely(xstats_names == NULL && ids != NULL) || 852 unlikely(xstats_names != NULL && ids == NULL)) 853 return -EINVAL; 854 855 sfc_adapter_lock(sa); 856 857 if (unlikely(xstats_names == NULL && ids == NULL)) { 858 nb_supported = port->mac_stats_nb_supported; 859 sfc_adapter_unlock(sa); 860 return nb_supported; 861 } 862 863 SFC_ASSERT(port->mac_stats_nb_supported <= 864 RTE_DIM(port->mac_stats_by_id)); 865 866 for (i = 0; i < size; i++) { 867 if (ids[i] < port->mac_stats_nb_supported) { 868 strlcpy(xstats_names[i].name, 869 efx_mac_stat_name(sa->nic, 870 port->mac_stats_by_id[ids[i]]), 871 sizeof(xstats_names[0].name)); 872 } else { 873 sfc_adapter_unlock(sa); 874 return i; 875 } 876 } 877 878 sfc_adapter_unlock(sa); 879 880 return size; 881 } 882 883 static int 884 sfc_flow_ctrl_get(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 unsigned int wanted_fc, link_fc; 888 889 memset(fc_conf, 0, sizeof(*fc_conf)); 890 891 sfc_adapter_lock(sa); 892 893 if (sa->state == SFC_ADAPTER_STARTED) 894 efx_mac_fcntl_get(sa->nic, &wanted_fc, &link_fc); 895 else 896 link_fc = sa->port.flow_ctrl; 897 898 switch (link_fc) { 899 case 0: 900 fc_conf->mode = RTE_FC_NONE; 901 break; 902 case EFX_FCNTL_RESPOND: 903 fc_conf->mode = RTE_FC_RX_PAUSE; 904 break; 905 case EFX_FCNTL_GENERATE: 906 fc_conf->mode = RTE_FC_TX_PAUSE; 907 break; 908 case (EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE): 909 fc_conf->mode = RTE_FC_FULL; 910 break; 911 default: 912 sfc_err(sa, "%s: unexpected flow control value %#x", 913 __func__, link_fc); 914 } 915 916 fc_conf->autoneg = sa->port.flow_ctrl_autoneg; 917 918 sfc_adapter_unlock(sa); 919 920 return 0; 921 } 922 923 static int 924 sfc_flow_ctrl_set(struct rte_eth_dev *dev, struct rte_eth_fc_conf *fc_conf) 925 { 926 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 927 struct sfc_port *port = &sa->port; 928 unsigned int fcntl; 929 int rc; 930 931 if (fc_conf->high_water != 0 || fc_conf->low_water != 0 || 932 fc_conf->pause_time != 0 || fc_conf->send_xon != 0 || 933 fc_conf->mac_ctrl_frame_fwd != 0) { 934 sfc_err(sa, "unsupported flow control settings specified"); 935 rc = EINVAL; 936 goto fail_inval; 937 } 938 939 switch (fc_conf->mode) { 940 case RTE_FC_NONE: 941 fcntl = 0; 942 break; 943 case RTE_FC_RX_PAUSE: 944 fcntl = EFX_FCNTL_RESPOND; 945 break; 946 case RTE_FC_TX_PAUSE: 947 fcntl = EFX_FCNTL_GENERATE; 948 break; 949 case RTE_FC_FULL: 950 fcntl = EFX_FCNTL_RESPOND | EFX_FCNTL_GENERATE; 951 break; 952 default: 953 rc = EINVAL; 954 goto fail_inval; 955 } 956 957 sfc_adapter_lock(sa); 958 959 if (sa->state == SFC_ADAPTER_STARTED) { 960 rc = efx_mac_fcntl_set(sa->nic, fcntl, fc_conf->autoneg); 961 if (rc != 0) 962 goto fail_mac_fcntl_set; 963 } 964 965 port->flow_ctrl = fcntl; 966 port->flow_ctrl_autoneg = fc_conf->autoneg; 967 968 sfc_adapter_unlock(sa); 969 970 return 0; 971 972 fail_mac_fcntl_set: 973 sfc_adapter_unlock(sa); 974 fail_inval: 975 SFC_ASSERT(rc > 0); 976 return -rc; 977 } 978 979 static int 980 sfc_check_scatter_on_all_rx_queues(struct sfc_adapter *sa, size_t pdu) 981 { 982 struct sfc_adapter_shared * const sas = sfc_sa2shared(sa); 983 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic); 984 boolean_t scatter_enabled; 985 const char *error; 986 unsigned int i; 987 988 for (i = 0; i < sas->rxq_count; i++) { 989 if ((sas->rxq_info[i].state & SFC_RXQ_INITIALIZED) == 0) 990 continue; 991 992 scatter_enabled = (sas->rxq_info[i].type_flags & 993 EFX_RXQ_FLAG_SCATTER); 994 995 if (!sfc_rx_check_scatter(pdu, sa->rxq_ctrl[i].buf_size, 996 encp->enc_rx_prefix_size, 997 scatter_enabled, 998 encp->enc_rx_scatter_max, &error)) { 999 sfc_err(sa, "MTU check for RxQ %u failed: %s", i, 1000 error); 1001 return EINVAL; 1002 } 1003 } 1004 1005 return 0; 1006 } 1007 1008 static int 1009 sfc_dev_set_mtu(struct rte_eth_dev *dev, uint16_t mtu) 1010 { 1011 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1012 size_t pdu = EFX_MAC_PDU(mtu); 1013 size_t old_pdu; 1014 int rc; 1015 1016 sfc_log_init(sa, "mtu=%u", mtu); 1017 1018 rc = EINVAL; 1019 if (pdu < EFX_MAC_PDU_MIN) { 1020 sfc_err(sa, "too small MTU %u (PDU size %u less than min %u)", 1021 (unsigned int)mtu, (unsigned int)pdu, 1022 EFX_MAC_PDU_MIN); 1023 goto fail_inval; 1024 } 1025 if (pdu > EFX_MAC_PDU_MAX) { 1026 sfc_err(sa, "too big MTU %u (PDU size %u greater than max %u)", 1027 (unsigned int)mtu, (unsigned int)pdu, 1028 (unsigned int)EFX_MAC_PDU_MAX); 1029 goto fail_inval; 1030 } 1031 1032 sfc_adapter_lock(sa); 1033 1034 rc = sfc_check_scatter_on_all_rx_queues(sa, pdu); 1035 if (rc != 0) 1036 goto fail_check_scatter; 1037 1038 if (pdu != sa->port.pdu) { 1039 if (sa->state == SFC_ADAPTER_STARTED) { 1040 sfc_stop(sa); 1041 1042 old_pdu = sa->port.pdu; 1043 sa->port.pdu = pdu; 1044 rc = sfc_start(sa); 1045 if (rc != 0) 1046 goto fail_start; 1047 } else { 1048 sa->port.pdu = pdu; 1049 } 1050 } 1051 1052 /* 1053 * The driver does not use it, but other PMDs update jumbo frame 1054 * flag and max_rx_pkt_len when MTU is set. 1055 */ 1056 if (mtu > RTE_ETHER_MTU) { 1057 struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode; 1058 rxmode->offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME; 1059 } 1060 1061 dev->data->dev_conf.rxmode.max_rx_pkt_len = sa->port.pdu; 1062 1063 sfc_adapter_unlock(sa); 1064 1065 sfc_log_init(sa, "done"); 1066 return 0; 1067 1068 fail_start: 1069 sa->port.pdu = old_pdu; 1070 if (sfc_start(sa) != 0) 1071 sfc_err(sa, "cannot start with neither new (%u) nor old (%u) " 1072 "PDU max size - port is stopped", 1073 (unsigned int)pdu, (unsigned int)old_pdu); 1074 1075 fail_check_scatter: 1076 sfc_adapter_unlock(sa); 1077 1078 fail_inval: 1079 sfc_log_init(sa, "failed %d", rc); 1080 SFC_ASSERT(rc > 0); 1081 return -rc; 1082 } 1083 static int 1084 sfc_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr) 1085 { 1086 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1087 const efx_nic_cfg_t *encp = efx_nic_cfg_get(sa->nic); 1088 struct sfc_port *port = &sa->port; 1089 struct rte_ether_addr *old_addr = &dev->data->mac_addrs[0]; 1090 int rc = 0; 1091 1092 sfc_adapter_lock(sa); 1093 1094 if (rte_is_same_ether_addr(mac_addr, &port->default_mac_addr)) 1095 goto unlock; 1096 1097 /* 1098 * Copy the address to the device private data so that 1099 * it could be recalled in the case of adapter restart. 1100 */ 1101 rte_ether_addr_copy(mac_addr, &port->default_mac_addr); 1102 1103 /* 1104 * Neither of the two following checks can return 1105 * an error. The new MAC address is preserved in 1106 * the device private data and can be activated 1107 * on the next port start if the user prevents 1108 * isolated mode from being enabled. 1109 */ 1110 if (sfc_sa2shared(sa)->isolated) { 1111 sfc_warn(sa, "isolated mode is active on the port"); 1112 sfc_warn(sa, "will not set MAC address"); 1113 goto unlock; 1114 } 1115 1116 if (sa->state != SFC_ADAPTER_STARTED) { 1117 sfc_notice(sa, "the port is not started"); 1118 sfc_notice(sa, "the new MAC address will be set on port start"); 1119 1120 goto unlock; 1121 } 1122 1123 if (encp->enc_allow_set_mac_with_installed_filters) { 1124 rc = efx_mac_addr_set(sa->nic, mac_addr->addr_bytes); 1125 if (rc != 0) { 1126 sfc_err(sa, "cannot set MAC address (rc = %u)", rc); 1127 goto unlock; 1128 } 1129 1130 /* 1131 * Changing the MAC address by means of MCDI request 1132 * has no effect on received traffic, therefore 1133 * we also need to update unicast filters 1134 */ 1135 rc = sfc_set_rx_mode_unchecked(sa); 1136 if (rc != 0) { 1137 sfc_err(sa, "cannot set filter (rc = %u)", rc); 1138 /* Rollback the old address */ 1139 (void)efx_mac_addr_set(sa->nic, old_addr->addr_bytes); 1140 (void)sfc_set_rx_mode_unchecked(sa); 1141 } 1142 } else { 1143 sfc_warn(sa, "cannot set MAC address with filters installed"); 1144 sfc_warn(sa, "adapter will be restarted to pick the new MAC"); 1145 sfc_warn(sa, "(some traffic may be dropped)"); 1146 1147 /* 1148 * Since setting MAC address with filters installed is not 1149 * allowed on the adapter, the new MAC address will be set 1150 * by means of adapter restart. sfc_start() shall retrieve 1151 * the new address from the device private data and set it. 1152 */ 1153 sfc_stop(sa); 1154 rc = sfc_start(sa); 1155 if (rc != 0) 1156 sfc_err(sa, "cannot restart adapter (rc = %u)", rc); 1157 } 1158 1159 unlock: 1160 if (rc != 0) 1161 rte_ether_addr_copy(old_addr, &port->default_mac_addr); 1162 1163 sfc_adapter_unlock(sa); 1164 1165 SFC_ASSERT(rc >= 0); 1166 return -rc; 1167 } 1168 1169 1170 static int 1171 sfc_set_mc_addr_list(struct rte_eth_dev *dev, 1172 struct rte_ether_addr *mc_addr_set, uint32_t nb_mc_addr) 1173 { 1174 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1175 struct sfc_port *port = &sa->port; 1176 uint8_t *mc_addrs = port->mcast_addrs; 1177 int rc; 1178 unsigned int i; 1179 1180 if (sfc_sa2shared(sa)->isolated) { 1181 sfc_err(sa, "isolated mode is active on the port"); 1182 sfc_err(sa, "will not set multicast address list"); 1183 return -ENOTSUP; 1184 } 1185 1186 if (mc_addrs == NULL) 1187 return -ENOBUFS; 1188 1189 if (nb_mc_addr > port->max_mcast_addrs) { 1190 sfc_err(sa, "too many multicast addresses: %u > %u", 1191 nb_mc_addr, port->max_mcast_addrs); 1192 return -EINVAL; 1193 } 1194 1195 for (i = 0; i < nb_mc_addr; ++i) { 1196 rte_memcpy(mc_addrs, mc_addr_set[i].addr_bytes, 1197 EFX_MAC_ADDR_LEN); 1198 mc_addrs += EFX_MAC_ADDR_LEN; 1199 } 1200 1201 port->nb_mcast_addrs = nb_mc_addr; 1202 1203 if (sa->state != SFC_ADAPTER_STARTED) 1204 return 0; 1205 1206 rc = efx_mac_multicast_list_set(sa->nic, port->mcast_addrs, 1207 port->nb_mcast_addrs); 1208 if (rc != 0) 1209 sfc_err(sa, "cannot set multicast address list (rc = %u)", rc); 1210 1211 SFC_ASSERT(rc >= 0); 1212 return -rc; 1213 } 1214 1215 /* 1216 * The function is used by the secondary process as well. It must not 1217 * use any process-local pointers from the adapter data. 1218 */ 1219 static void 1220 sfc_rx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid, 1221 struct rte_eth_rxq_info *qinfo) 1222 { 1223 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1224 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 1225 struct sfc_rxq_info *rxq_info; 1226 1227 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 1228 1229 qinfo->mp = rxq_info->refill_mb_pool; 1230 qinfo->conf.rx_free_thresh = rxq_info->refill_threshold; 1231 qinfo->conf.rx_drop_en = 1; 1232 qinfo->conf.rx_deferred_start = rxq_info->deferred_start; 1233 qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads; 1234 if (rxq_info->type_flags & EFX_RXQ_FLAG_SCATTER) { 1235 qinfo->conf.offloads |= DEV_RX_OFFLOAD_SCATTER; 1236 qinfo->scattered_rx = 1; 1237 } 1238 qinfo->nb_desc = rxq_info->entries; 1239 } 1240 1241 /* 1242 * The function is used by the secondary process as well. It must not 1243 * use any process-local pointers from the adapter data. 1244 */ 1245 static void 1246 sfc_tx_queue_info_get(struct rte_eth_dev *dev, uint16_t ethdev_qid, 1247 struct rte_eth_txq_info *qinfo) 1248 { 1249 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1250 struct sfc_txq_info *txq_info; 1251 1252 SFC_ASSERT(ethdev_qid < sas->ethdev_txq_count); 1253 1254 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid); 1255 1256 memset(qinfo, 0, sizeof(*qinfo)); 1257 1258 qinfo->conf.offloads = txq_info->offloads; 1259 qinfo->conf.tx_free_thresh = txq_info->free_thresh; 1260 qinfo->conf.tx_deferred_start = txq_info->deferred_start; 1261 qinfo->nb_desc = txq_info->entries; 1262 } 1263 1264 /* 1265 * The function is used by the secondary process as well. It must not 1266 * use any process-local pointers from the adapter data. 1267 */ 1268 static uint32_t 1269 sfc_rx_queue_count(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1270 { 1271 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev); 1272 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1273 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 1274 struct sfc_rxq_info *rxq_info; 1275 1276 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 1277 1278 if ((rxq_info->state & SFC_RXQ_STARTED) == 0) 1279 return 0; 1280 1281 return sap->dp_rx->qdesc_npending(rxq_info->dp); 1282 } 1283 1284 /* 1285 * The function is used by the secondary process as well. It must not 1286 * use any process-local pointers from the adapter data. 1287 */ 1288 static int 1289 sfc_rx_descriptor_done(void *queue, uint16_t offset) 1290 { 1291 struct sfc_dp_rxq *dp_rxq = queue; 1292 const struct sfc_dp_rx *dp_rx; 1293 1294 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq); 1295 1296 return offset < dp_rx->qdesc_npending(dp_rxq); 1297 } 1298 1299 /* 1300 * The function is used by the secondary process as well. It must not 1301 * use any process-local pointers from the adapter data. 1302 */ 1303 static int 1304 sfc_rx_descriptor_status(void *queue, uint16_t offset) 1305 { 1306 struct sfc_dp_rxq *dp_rxq = queue; 1307 const struct sfc_dp_rx *dp_rx; 1308 1309 dp_rx = sfc_dp_rx_by_dp_rxq(dp_rxq); 1310 1311 return dp_rx->qdesc_status(dp_rxq, offset); 1312 } 1313 1314 /* 1315 * The function is used by the secondary process as well. It must not 1316 * use any process-local pointers from the adapter data. 1317 */ 1318 static int 1319 sfc_tx_descriptor_status(void *queue, uint16_t offset) 1320 { 1321 struct sfc_dp_txq *dp_txq = queue; 1322 const struct sfc_dp_tx *dp_tx; 1323 1324 dp_tx = sfc_dp_tx_by_dp_txq(dp_txq); 1325 1326 return dp_tx->qdesc_status(dp_txq, offset); 1327 } 1328 1329 static int 1330 sfc_rx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1331 { 1332 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1333 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1334 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 1335 struct sfc_rxq_info *rxq_info; 1336 sfc_sw_index_t sw_index; 1337 int rc; 1338 1339 sfc_log_init(sa, "RxQ=%u", ethdev_qid); 1340 1341 sfc_adapter_lock(sa); 1342 1343 rc = EINVAL; 1344 if (sa->state != SFC_ADAPTER_STARTED) 1345 goto fail_not_started; 1346 1347 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 1348 if (rxq_info->state != SFC_RXQ_INITIALIZED) 1349 goto fail_not_setup; 1350 1351 sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid); 1352 rc = sfc_rx_qstart(sa, sw_index); 1353 if (rc != 0) 1354 goto fail_rx_qstart; 1355 1356 rxq_info->deferred_started = B_TRUE; 1357 1358 sfc_adapter_unlock(sa); 1359 1360 return 0; 1361 1362 fail_rx_qstart: 1363 fail_not_setup: 1364 fail_not_started: 1365 sfc_adapter_unlock(sa); 1366 SFC_ASSERT(rc > 0); 1367 return -rc; 1368 } 1369 1370 static int 1371 sfc_rx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1372 { 1373 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1374 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1375 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 1376 struct sfc_rxq_info *rxq_info; 1377 sfc_sw_index_t sw_index; 1378 1379 sfc_log_init(sa, "RxQ=%u", ethdev_qid); 1380 1381 sfc_adapter_lock(sa); 1382 1383 sw_index = sfc_rxq_sw_index_by_ethdev_rx_qid(sas, sfc_ethdev_qid); 1384 sfc_rx_qstop(sa, sw_index); 1385 1386 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 1387 rxq_info->deferred_started = B_FALSE; 1388 1389 sfc_adapter_unlock(sa); 1390 1391 return 0; 1392 } 1393 1394 static int 1395 sfc_tx_queue_start(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1396 { 1397 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1398 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1399 struct sfc_txq_info *txq_info; 1400 sfc_sw_index_t sw_index; 1401 int rc; 1402 1403 sfc_log_init(sa, "TxQ = %u", ethdev_qid); 1404 1405 sfc_adapter_lock(sa); 1406 1407 rc = EINVAL; 1408 if (sa->state != SFC_ADAPTER_STARTED) 1409 goto fail_not_started; 1410 1411 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid); 1412 if (txq_info->state != SFC_TXQ_INITIALIZED) 1413 goto fail_not_setup; 1414 1415 sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid); 1416 rc = sfc_tx_qstart(sa, sw_index); 1417 if (rc != 0) 1418 goto fail_tx_qstart; 1419 1420 txq_info->deferred_started = B_TRUE; 1421 1422 sfc_adapter_unlock(sa); 1423 return 0; 1424 1425 fail_tx_qstart: 1426 1427 fail_not_setup: 1428 fail_not_started: 1429 sfc_adapter_unlock(sa); 1430 SFC_ASSERT(rc > 0); 1431 return -rc; 1432 } 1433 1434 static int 1435 sfc_tx_queue_stop(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1436 { 1437 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1438 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1439 struct sfc_txq_info *txq_info; 1440 sfc_sw_index_t sw_index; 1441 1442 sfc_log_init(sa, "TxQ = %u", ethdev_qid); 1443 1444 sfc_adapter_lock(sa); 1445 1446 sw_index = sfc_txq_sw_index_by_ethdev_tx_qid(sas, ethdev_qid); 1447 sfc_tx_qstop(sa, sw_index); 1448 1449 txq_info = sfc_txq_info_by_ethdev_qid(sas, ethdev_qid); 1450 txq_info->deferred_started = B_FALSE; 1451 1452 sfc_adapter_unlock(sa); 1453 return 0; 1454 } 1455 1456 static efx_tunnel_protocol_t 1457 sfc_tunnel_rte_type_to_efx_udp_proto(enum rte_eth_tunnel_type rte_type) 1458 { 1459 switch (rte_type) { 1460 case RTE_TUNNEL_TYPE_VXLAN: 1461 return EFX_TUNNEL_PROTOCOL_VXLAN; 1462 case RTE_TUNNEL_TYPE_GENEVE: 1463 return EFX_TUNNEL_PROTOCOL_GENEVE; 1464 default: 1465 return EFX_TUNNEL_NPROTOS; 1466 } 1467 } 1468 1469 enum sfc_udp_tunnel_op_e { 1470 SFC_UDP_TUNNEL_ADD_PORT, 1471 SFC_UDP_TUNNEL_DEL_PORT, 1472 }; 1473 1474 static int 1475 sfc_dev_udp_tunnel_op(struct rte_eth_dev *dev, 1476 struct rte_eth_udp_tunnel *tunnel_udp, 1477 enum sfc_udp_tunnel_op_e op) 1478 { 1479 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1480 efx_tunnel_protocol_t tunnel_proto; 1481 int rc; 1482 1483 sfc_log_init(sa, "%s udp_port=%u prot_type=%u", 1484 (op == SFC_UDP_TUNNEL_ADD_PORT) ? "add" : 1485 (op == SFC_UDP_TUNNEL_DEL_PORT) ? "delete" : "unknown", 1486 tunnel_udp->udp_port, tunnel_udp->prot_type); 1487 1488 tunnel_proto = 1489 sfc_tunnel_rte_type_to_efx_udp_proto(tunnel_udp->prot_type); 1490 if (tunnel_proto >= EFX_TUNNEL_NPROTOS) { 1491 rc = ENOTSUP; 1492 goto fail_bad_proto; 1493 } 1494 1495 sfc_adapter_lock(sa); 1496 1497 switch (op) { 1498 case SFC_UDP_TUNNEL_ADD_PORT: 1499 rc = efx_tunnel_config_udp_add(sa->nic, 1500 tunnel_udp->udp_port, 1501 tunnel_proto); 1502 break; 1503 case SFC_UDP_TUNNEL_DEL_PORT: 1504 rc = efx_tunnel_config_udp_remove(sa->nic, 1505 tunnel_udp->udp_port, 1506 tunnel_proto); 1507 break; 1508 default: 1509 rc = EINVAL; 1510 goto fail_bad_op; 1511 } 1512 1513 if (rc != 0) 1514 goto fail_op; 1515 1516 if (sa->state == SFC_ADAPTER_STARTED) { 1517 rc = efx_tunnel_reconfigure(sa->nic); 1518 if (rc == EAGAIN) { 1519 /* 1520 * Configuration is accepted by FW and MC reboot 1521 * is initiated to apply the changes. MC reboot 1522 * will be handled in a usual way (MC reboot 1523 * event on management event queue and adapter 1524 * restart). 1525 */ 1526 rc = 0; 1527 } else if (rc != 0) { 1528 goto fail_reconfigure; 1529 } 1530 } 1531 1532 sfc_adapter_unlock(sa); 1533 return 0; 1534 1535 fail_reconfigure: 1536 /* Remove/restore entry since the change makes the trouble */ 1537 switch (op) { 1538 case SFC_UDP_TUNNEL_ADD_PORT: 1539 (void)efx_tunnel_config_udp_remove(sa->nic, 1540 tunnel_udp->udp_port, 1541 tunnel_proto); 1542 break; 1543 case SFC_UDP_TUNNEL_DEL_PORT: 1544 (void)efx_tunnel_config_udp_add(sa->nic, 1545 tunnel_udp->udp_port, 1546 tunnel_proto); 1547 break; 1548 } 1549 1550 fail_op: 1551 fail_bad_op: 1552 sfc_adapter_unlock(sa); 1553 1554 fail_bad_proto: 1555 SFC_ASSERT(rc > 0); 1556 return -rc; 1557 } 1558 1559 static int 1560 sfc_dev_udp_tunnel_port_add(struct rte_eth_dev *dev, 1561 struct rte_eth_udp_tunnel *tunnel_udp) 1562 { 1563 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_ADD_PORT); 1564 } 1565 1566 static int 1567 sfc_dev_udp_tunnel_port_del(struct rte_eth_dev *dev, 1568 struct rte_eth_udp_tunnel *tunnel_udp) 1569 { 1570 return sfc_dev_udp_tunnel_op(dev, tunnel_udp, SFC_UDP_TUNNEL_DEL_PORT); 1571 } 1572 1573 /* 1574 * The function is used by the secondary process as well. It must not 1575 * use any process-local pointers from the adapter data. 1576 */ 1577 static int 1578 sfc_dev_rss_hash_conf_get(struct rte_eth_dev *dev, 1579 struct rte_eth_rss_conf *rss_conf) 1580 { 1581 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1582 struct sfc_rss *rss = &sas->rss; 1583 1584 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) 1585 return -ENOTSUP; 1586 1587 /* 1588 * Mapping of hash configuration between RTE and EFX is not one-to-one, 1589 * hence, conversion is done here to derive a correct set of ETH_RSS 1590 * flags which corresponds to the active EFX configuration stored 1591 * locally in 'sfc_adapter' and kept up-to-date 1592 */ 1593 rss_conf->rss_hf = sfc_rx_hf_efx_to_rte(rss, rss->hash_types); 1594 rss_conf->rss_key_len = EFX_RSS_KEY_SIZE; 1595 if (rss_conf->rss_key != NULL) 1596 rte_memcpy(rss_conf->rss_key, rss->key, EFX_RSS_KEY_SIZE); 1597 1598 return 0; 1599 } 1600 1601 static int 1602 sfc_dev_rss_hash_update(struct rte_eth_dev *dev, 1603 struct rte_eth_rss_conf *rss_conf) 1604 { 1605 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1606 struct sfc_rss *rss = &sfc_sa2shared(sa)->rss; 1607 unsigned int efx_hash_types; 1608 uint32_t contexts[] = {EFX_RSS_CONTEXT_DEFAULT, rss->dummy_rss_context}; 1609 unsigned int n_contexts; 1610 unsigned int mode_i = 0; 1611 unsigned int key_i = 0; 1612 unsigned int i = 0; 1613 int rc = 0; 1614 1615 n_contexts = rss->dummy_rss_context == EFX_RSS_CONTEXT_DEFAULT ? 1 : 2; 1616 1617 if (sfc_sa2shared(sa)->isolated) 1618 return -ENOTSUP; 1619 1620 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) { 1621 sfc_err(sa, "RSS is not available"); 1622 return -ENOTSUP; 1623 } 1624 1625 if (rss->channels == 0) { 1626 sfc_err(sa, "RSS is not configured"); 1627 return -EINVAL; 1628 } 1629 1630 if ((rss_conf->rss_key != NULL) && 1631 (rss_conf->rss_key_len != sizeof(rss->key))) { 1632 sfc_err(sa, "RSS key size is wrong (should be %zu)", 1633 sizeof(rss->key)); 1634 return -EINVAL; 1635 } 1636 1637 sfc_adapter_lock(sa); 1638 1639 rc = sfc_rx_hf_rte_to_efx(sa, rss_conf->rss_hf, &efx_hash_types); 1640 if (rc != 0) 1641 goto fail_rx_hf_rte_to_efx; 1642 1643 for (mode_i = 0; mode_i < n_contexts; mode_i++) { 1644 rc = efx_rx_scale_mode_set(sa->nic, contexts[mode_i], 1645 rss->hash_alg, efx_hash_types, 1646 B_TRUE); 1647 if (rc != 0) 1648 goto fail_scale_mode_set; 1649 } 1650 1651 if (rss_conf->rss_key != NULL) { 1652 if (sa->state == SFC_ADAPTER_STARTED) { 1653 for (key_i = 0; key_i < n_contexts; key_i++) { 1654 rc = efx_rx_scale_key_set(sa->nic, 1655 contexts[key_i], 1656 rss_conf->rss_key, 1657 sizeof(rss->key)); 1658 if (rc != 0) 1659 goto fail_scale_key_set; 1660 } 1661 } 1662 1663 rte_memcpy(rss->key, rss_conf->rss_key, sizeof(rss->key)); 1664 } 1665 1666 rss->hash_types = efx_hash_types; 1667 1668 sfc_adapter_unlock(sa); 1669 1670 return 0; 1671 1672 fail_scale_key_set: 1673 for (i = 0; i < key_i; i++) { 1674 if (efx_rx_scale_key_set(sa->nic, contexts[i], rss->key, 1675 sizeof(rss->key)) != 0) 1676 sfc_err(sa, "failed to restore RSS key"); 1677 } 1678 1679 fail_scale_mode_set: 1680 for (i = 0; i < mode_i; i++) { 1681 if (efx_rx_scale_mode_set(sa->nic, contexts[i], 1682 EFX_RX_HASHALG_TOEPLITZ, 1683 rss->hash_types, B_TRUE) != 0) 1684 sfc_err(sa, "failed to restore RSS mode"); 1685 } 1686 1687 fail_rx_hf_rte_to_efx: 1688 sfc_adapter_unlock(sa); 1689 return -rc; 1690 } 1691 1692 /* 1693 * The function is used by the secondary process as well. It must not 1694 * use any process-local pointers from the adapter data. 1695 */ 1696 static int 1697 sfc_dev_rss_reta_query(struct rte_eth_dev *dev, 1698 struct rte_eth_rss_reta_entry64 *reta_conf, 1699 uint16_t reta_size) 1700 { 1701 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1702 struct sfc_rss *rss = &sas->rss; 1703 int entry; 1704 1705 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE || sas->isolated) 1706 return -ENOTSUP; 1707 1708 if (rss->channels == 0) 1709 return -EINVAL; 1710 1711 if (reta_size != EFX_RSS_TBL_SIZE) 1712 return -EINVAL; 1713 1714 for (entry = 0; entry < reta_size; entry++) { 1715 int grp = entry / RTE_RETA_GROUP_SIZE; 1716 int grp_idx = entry % RTE_RETA_GROUP_SIZE; 1717 1718 if ((reta_conf[grp].mask >> grp_idx) & 1) 1719 reta_conf[grp].reta[grp_idx] = rss->tbl[entry]; 1720 } 1721 1722 return 0; 1723 } 1724 1725 static int 1726 sfc_dev_rss_reta_update(struct rte_eth_dev *dev, 1727 struct rte_eth_rss_reta_entry64 *reta_conf, 1728 uint16_t reta_size) 1729 { 1730 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1731 struct sfc_rss *rss = &sfc_sa2shared(sa)->rss; 1732 unsigned int *rss_tbl_new; 1733 uint16_t entry; 1734 int rc = 0; 1735 1736 1737 if (sfc_sa2shared(sa)->isolated) 1738 return -ENOTSUP; 1739 1740 if (rss->context_type != EFX_RX_SCALE_EXCLUSIVE) { 1741 sfc_err(sa, "RSS is not available"); 1742 return -ENOTSUP; 1743 } 1744 1745 if (rss->channels == 0) { 1746 sfc_err(sa, "RSS is not configured"); 1747 return -EINVAL; 1748 } 1749 1750 if (reta_size != EFX_RSS_TBL_SIZE) { 1751 sfc_err(sa, "RETA size is wrong (should be %u)", 1752 EFX_RSS_TBL_SIZE); 1753 return -EINVAL; 1754 } 1755 1756 rss_tbl_new = rte_zmalloc("rss_tbl_new", sizeof(rss->tbl), 0); 1757 if (rss_tbl_new == NULL) 1758 return -ENOMEM; 1759 1760 sfc_adapter_lock(sa); 1761 1762 rte_memcpy(rss_tbl_new, rss->tbl, sizeof(rss->tbl)); 1763 1764 for (entry = 0; entry < reta_size; entry++) { 1765 int grp_idx = entry % RTE_RETA_GROUP_SIZE; 1766 struct rte_eth_rss_reta_entry64 *grp; 1767 1768 grp = &reta_conf[entry / RTE_RETA_GROUP_SIZE]; 1769 1770 if (grp->mask & (1ull << grp_idx)) { 1771 if (grp->reta[grp_idx] >= rss->channels) { 1772 rc = EINVAL; 1773 goto bad_reta_entry; 1774 } 1775 rss_tbl_new[entry] = grp->reta[grp_idx]; 1776 } 1777 } 1778 1779 if (sa->state == SFC_ADAPTER_STARTED) { 1780 rc = efx_rx_scale_tbl_set(sa->nic, EFX_RSS_CONTEXT_DEFAULT, 1781 rss_tbl_new, EFX_RSS_TBL_SIZE); 1782 if (rc != 0) 1783 goto fail_scale_tbl_set; 1784 } 1785 1786 rte_memcpy(rss->tbl, rss_tbl_new, sizeof(rss->tbl)); 1787 1788 fail_scale_tbl_set: 1789 bad_reta_entry: 1790 sfc_adapter_unlock(sa); 1791 1792 rte_free(rss_tbl_new); 1793 1794 SFC_ASSERT(rc >= 0); 1795 return -rc; 1796 } 1797 1798 static int 1799 sfc_dev_flow_ops_get(struct rte_eth_dev *dev __rte_unused, 1800 const struct rte_flow_ops **ops) 1801 { 1802 *ops = &sfc_flow_ops; 1803 return 0; 1804 } 1805 1806 static int 1807 sfc_pool_ops_supported(struct rte_eth_dev *dev, const char *pool) 1808 { 1809 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev); 1810 1811 /* 1812 * If Rx datapath does not provide callback to check mempool, 1813 * all pools are supported. 1814 */ 1815 if (sap->dp_rx->pool_ops_supported == NULL) 1816 return 1; 1817 1818 return sap->dp_rx->pool_ops_supported(pool); 1819 } 1820 1821 static int 1822 sfc_rx_queue_intr_enable(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1823 { 1824 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev); 1825 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1826 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 1827 struct sfc_rxq_info *rxq_info; 1828 1829 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 1830 1831 return sap->dp_rx->intr_enable(rxq_info->dp); 1832 } 1833 1834 static int 1835 sfc_rx_queue_intr_disable(struct rte_eth_dev *dev, uint16_t ethdev_qid) 1836 { 1837 const struct sfc_adapter_priv *sap = sfc_adapter_priv_by_eth_dev(dev); 1838 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1839 sfc_ethdev_qid_t sfc_ethdev_qid = ethdev_qid; 1840 struct sfc_rxq_info *rxq_info; 1841 1842 rxq_info = sfc_rxq_info_by_ethdev_qid(sas, sfc_ethdev_qid); 1843 1844 return sap->dp_rx->intr_disable(rxq_info->dp); 1845 } 1846 1847 static const struct eth_dev_ops sfc_eth_dev_ops = { 1848 .dev_configure = sfc_dev_configure, 1849 .dev_start = sfc_dev_start, 1850 .dev_stop = sfc_dev_stop, 1851 .dev_set_link_up = sfc_dev_set_link_up, 1852 .dev_set_link_down = sfc_dev_set_link_down, 1853 .dev_close = sfc_dev_close, 1854 .promiscuous_enable = sfc_dev_promisc_enable, 1855 .promiscuous_disable = sfc_dev_promisc_disable, 1856 .allmulticast_enable = sfc_dev_allmulti_enable, 1857 .allmulticast_disable = sfc_dev_allmulti_disable, 1858 .link_update = sfc_dev_link_update, 1859 .stats_get = sfc_stats_get, 1860 .stats_reset = sfc_stats_reset, 1861 .xstats_get = sfc_xstats_get, 1862 .xstats_reset = sfc_stats_reset, 1863 .xstats_get_names = sfc_xstats_get_names, 1864 .dev_infos_get = sfc_dev_infos_get, 1865 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get, 1866 .mtu_set = sfc_dev_set_mtu, 1867 .rx_queue_start = sfc_rx_queue_start, 1868 .rx_queue_stop = sfc_rx_queue_stop, 1869 .tx_queue_start = sfc_tx_queue_start, 1870 .tx_queue_stop = sfc_tx_queue_stop, 1871 .rx_queue_setup = sfc_rx_queue_setup, 1872 .rx_queue_release = sfc_rx_queue_release, 1873 .rx_queue_intr_enable = sfc_rx_queue_intr_enable, 1874 .rx_queue_intr_disable = sfc_rx_queue_intr_disable, 1875 .tx_queue_setup = sfc_tx_queue_setup, 1876 .tx_queue_release = sfc_tx_queue_release, 1877 .flow_ctrl_get = sfc_flow_ctrl_get, 1878 .flow_ctrl_set = sfc_flow_ctrl_set, 1879 .mac_addr_set = sfc_mac_addr_set, 1880 .udp_tunnel_port_add = sfc_dev_udp_tunnel_port_add, 1881 .udp_tunnel_port_del = sfc_dev_udp_tunnel_port_del, 1882 .reta_update = sfc_dev_rss_reta_update, 1883 .reta_query = sfc_dev_rss_reta_query, 1884 .rss_hash_update = sfc_dev_rss_hash_update, 1885 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get, 1886 .flow_ops_get = sfc_dev_flow_ops_get, 1887 .set_mc_addr_list = sfc_set_mc_addr_list, 1888 .rxq_info_get = sfc_rx_queue_info_get, 1889 .txq_info_get = sfc_tx_queue_info_get, 1890 .fw_version_get = sfc_fw_version_get, 1891 .xstats_get_by_id = sfc_xstats_get_by_id, 1892 .xstats_get_names_by_id = sfc_xstats_get_names_by_id, 1893 .pool_ops_supported = sfc_pool_ops_supported, 1894 }; 1895 1896 /** 1897 * Duplicate a string in potentially shared memory required for 1898 * multi-process support. 1899 * 1900 * strdup() allocates from process-local heap/memory. 1901 */ 1902 static char * 1903 sfc_strdup(const char *str) 1904 { 1905 size_t size; 1906 char *copy; 1907 1908 if (str == NULL) 1909 return NULL; 1910 1911 size = strlen(str) + 1; 1912 copy = rte_malloc(__func__, size, 0); 1913 if (copy != NULL) 1914 rte_memcpy(copy, str, size); 1915 1916 return copy; 1917 } 1918 1919 static int 1920 sfc_eth_dev_set_ops(struct rte_eth_dev *dev) 1921 { 1922 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 1923 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 1924 const struct sfc_dp_rx *dp_rx; 1925 const struct sfc_dp_tx *dp_tx; 1926 const efx_nic_cfg_t *encp; 1927 unsigned int avail_caps = 0; 1928 const char *rx_name = NULL; 1929 const char *tx_name = NULL; 1930 int rc; 1931 1932 switch (sa->family) { 1933 case EFX_FAMILY_HUNTINGTON: 1934 case EFX_FAMILY_MEDFORD: 1935 case EFX_FAMILY_MEDFORD2: 1936 avail_caps |= SFC_DP_HW_FW_CAP_EF10; 1937 avail_caps |= SFC_DP_HW_FW_CAP_RX_EFX; 1938 avail_caps |= SFC_DP_HW_FW_CAP_TX_EFX; 1939 break; 1940 case EFX_FAMILY_RIVERHEAD: 1941 avail_caps |= SFC_DP_HW_FW_CAP_EF100; 1942 break; 1943 default: 1944 break; 1945 } 1946 1947 encp = efx_nic_cfg_get(sa->nic); 1948 if (encp->enc_rx_es_super_buffer_supported) 1949 avail_caps |= SFC_DP_HW_FW_CAP_RX_ES_SUPER_BUFFER; 1950 1951 rc = sfc_kvargs_process(sa, SFC_KVARG_RX_DATAPATH, 1952 sfc_kvarg_string_handler, &rx_name); 1953 if (rc != 0) 1954 goto fail_kvarg_rx_datapath; 1955 1956 if (rx_name != NULL) { 1957 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, rx_name); 1958 if (dp_rx == NULL) { 1959 sfc_err(sa, "Rx datapath %s not found", rx_name); 1960 rc = ENOENT; 1961 goto fail_dp_rx; 1962 } 1963 if (!sfc_dp_match_hw_fw_caps(&dp_rx->dp, avail_caps)) { 1964 sfc_err(sa, 1965 "Insufficient Hw/FW capabilities to use Rx datapath %s", 1966 rx_name); 1967 rc = EINVAL; 1968 goto fail_dp_rx_caps; 1969 } 1970 } else { 1971 dp_rx = sfc_dp_find_rx_by_caps(&sfc_dp_head, avail_caps); 1972 if (dp_rx == NULL) { 1973 sfc_err(sa, "Rx datapath by caps %#x not found", 1974 avail_caps); 1975 rc = ENOENT; 1976 goto fail_dp_rx; 1977 } 1978 } 1979 1980 sas->dp_rx_name = sfc_strdup(dp_rx->dp.name); 1981 if (sas->dp_rx_name == NULL) { 1982 rc = ENOMEM; 1983 goto fail_dp_rx_name; 1984 } 1985 1986 sfc_notice(sa, "use %s Rx datapath", sas->dp_rx_name); 1987 1988 rc = sfc_kvargs_process(sa, SFC_KVARG_TX_DATAPATH, 1989 sfc_kvarg_string_handler, &tx_name); 1990 if (rc != 0) 1991 goto fail_kvarg_tx_datapath; 1992 1993 if (tx_name != NULL) { 1994 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, tx_name); 1995 if (dp_tx == NULL) { 1996 sfc_err(sa, "Tx datapath %s not found", tx_name); 1997 rc = ENOENT; 1998 goto fail_dp_tx; 1999 } 2000 if (!sfc_dp_match_hw_fw_caps(&dp_tx->dp, avail_caps)) { 2001 sfc_err(sa, 2002 "Insufficient Hw/FW capabilities to use Tx datapath %s", 2003 tx_name); 2004 rc = EINVAL; 2005 goto fail_dp_tx_caps; 2006 } 2007 } else { 2008 dp_tx = sfc_dp_find_tx_by_caps(&sfc_dp_head, avail_caps); 2009 if (dp_tx == NULL) { 2010 sfc_err(sa, "Tx datapath by caps %#x not found", 2011 avail_caps); 2012 rc = ENOENT; 2013 goto fail_dp_tx; 2014 } 2015 } 2016 2017 sas->dp_tx_name = sfc_strdup(dp_tx->dp.name); 2018 if (sas->dp_tx_name == NULL) { 2019 rc = ENOMEM; 2020 goto fail_dp_tx_name; 2021 } 2022 2023 sfc_notice(sa, "use %s Tx datapath", sas->dp_tx_name); 2024 2025 sa->priv.dp_rx = dp_rx; 2026 sa->priv.dp_tx = dp_tx; 2027 2028 dev->rx_pkt_burst = dp_rx->pkt_burst; 2029 dev->tx_pkt_prepare = dp_tx->pkt_prepare; 2030 dev->tx_pkt_burst = dp_tx->pkt_burst; 2031 2032 dev->rx_queue_count = sfc_rx_queue_count; 2033 dev->rx_descriptor_done = sfc_rx_descriptor_done; 2034 dev->rx_descriptor_status = sfc_rx_descriptor_status; 2035 dev->tx_descriptor_status = sfc_tx_descriptor_status; 2036 dev->dev_ops = &sfc_eth_dev_ops; 2037 2038 return 0; 2039 2040 fail_dp_tx_name: 2041 fail_dp_tx_caps: 2042 fail_dp_tx: 2043 fail_kvarg_tx_datapath: 2044 rte_free(sas->dp_rx_name); 2045 sas->dp_rx_name = NULL; 2046 2047 fail_dp_rx_name: 2048 fail_dp_rx_caps: 2049 fail_dp_rx: 2050 fail_kvarg_rx_datapath: 2051 return rc; 2052 } 2053 2054 static void 2055 sfc_eth_dev_clear_ops(struct rte_eth_dev *dev) 2056 { 2057 struct sfc_adapter *sa = sfc_adapter_by_eth_dev(dev); 2058 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 2059 2060 dev->dev_ops = NULL; 2061 dev->tx_pkt_prepare = NULL; 2062 dev->rx_pkt_burst = NULL; 2063 dev->tx_pkt_burst = NULL; 2064 2065 rte_free(sas->dp_tx_name); 2066 sas->dp_tx_name = NULL; 2067 sa->priv.dp_tx = NULL; 2068 2069 rte_free(sas->dp_rx_name); 2070 sas->dp_rx_name = NULL; 2071 sa->priv.dp_rx = NULL; 2072 } 2073 2074 static const struct eth_dev_ops sfc_eth_dev_secondary_ops = { 2075 .dev_supported_ptypes_get = sfc_dev_supported_ptypes_get, 2076 .reta_query = sfc_dev_rss_reta_query, 2077 .rss_hash_conf_get = sfc_dev_rss_hash_conf_get, 2078 .rxq_info_get = sfc_rx_queue_info_get, 2079 .txq_info_get = sfc_tx_queue_info_get, 2080 }; 2081 2082 static int 2083 sfc_eth_dev_secondary_init(struct rte_eth_dev *dev, uint32_t logtype_main) 2084 { 2085 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 2086 struct sfc_adapter_priv *sap; 2087 const struct sfc_dp_rx *dp_rx; 2088 const struct sfc_dp_tx *dp_tx; 2089 int rc; 2090 2091 /* 2092 * Allocate process private data from heap, since it should not 2093 * be located in shared memory allocated using rte_malloc() API. 2094 */ 2095 sap = calloc(1, sizeof(*sap)); 2096 if (sap == NULL) { 2097 rc = ENOMEM; 2098 goto fail_alloc_priv; 2099 } 2100 2101 sap->logtype_main = logtype_main; 2102 2103 dp_rx = sfc_dp_find_rx_by_name(&sfc_dp_head, sas->dp_rx_name); 2104 if (dp_rx == NULL) { 2105 SFC_LOG(sas, RTE_LOG_ERR, logtype_main, 2106 "cannot find %s Rx datapath", sas->dp_rx_name); 2107 rc = ENOENT; 2108 goto fail_dp_rx; 2109 } 2110 if (~dp_rx->features & SFC_DP_RX_FEAT_MULTI_PROCESS) { 2111 SFC_LOG(sas, RTE_LOG_ERR, logtype_main, 2112 "%s Rx datapath does not support multi-process", 2113 sas->dp_rx_name); 2114 rc = EINVAL; 2115 goto fail_dp_rx_multi_process; 2116 } 2117 2118 dp_tx = sfc_dp_find_tx_by_name(&sfc_dp_head, sas->dp_tx_name); 2119 if (dp_tx == NULL) { 2120 SFC_LOG(sas, RTE_LOG_ERR, logtype_main, 2121 "cannot find %s Tx datapath", sas->dp_tx_name); 2122 rc = ENOENT; 2123 goto fail_dp_tx; 2124 } 2125 if (~dp_tx->features & SFC_DP_TX_FEAT_MULTI_PROCESS) { 2126 SFC_LOG(sas, RTE_LOG_ERR, logtype_main, 2127 "%s Tx datapath does not support multi-process", 2128 sas->dp_tx_name); 2129 rc = EINVAL; 2130 goto fail_dp_tx_multi_process; 2131 } 2132 2133 sap->dp_rx = dp_rx; 2134 sap->dp_tx = dp_tx; 2135 2136 dev->process_private = sap; 2137 dev->rx_pkt_burst = dp_rx->pkt_burst; 2138 dev->tx_pkt_prepare = dp_tx->pkt_prepare; 2139 dev->tx_pkt_burst = dp_tx->pkt_burst; 2140 dev->rx_queue_count = sfc_rx_queue_count; 2141 dev->rx_descriptor_done = sfc_rx_descriptor_done; 2142 dev->rx_descriptor_status = sfc_rx_descriptor_status; 2143 dev->tx_descriptor_status = sfc_tx_descriptor_status; 2144 dev->dev_ops = &sfc_eth_dev_secondary_ops; 2145 2146 return 0; 2147 2148 fail_dp_tx_multi_process: 2149 fail_dp_tx: 2150 fail_dp_rx_multi_process: 2151 fail_dp_rx: 2152 free(sap); 2153 2154 fail_alloc_priv: 2155 return rc; 2156 } 2157 2158 static void 2159 sfc_register_dp(void) 2160 { 2161 /* Register once */ 2162 if (TAILQ_EMPTY(&sfc_dp_head)) { 2163 /* Prefer EF10 datapath */ 2164 sfc_dp_register(&sfc_dp_head, &sfc_ef100_rx.dp); 2165 sfc_dp_register(&sfc_dp_head, &sfc_ef10_essb_rx.dp); 2166 sfc_dp_register(&sfc_dp_head, &sfc_ef10_rx.dp); 2167 sfc_dp_register(&sfc_dp_head, &sfc_efx_rx.dp); 2168 2169 sfc_dp_register(&sfc_dp_head, &sfc_ef100_tx.dp); 2170 sfc_dp_register(&sfc_dp_head, &sfc_ef10_tx.dp); 2171 sfc_dp_register(&sfc_dp_head, &sfc_efx_tx.dp); 2172 sfc_dp_register(&sfc_dp_head, &sfc_ef10_simple_tx.dp); 2173 } 2174 } 2175 2176 static int 2177 sfc_eth_dev_init(struct rte_eth_dev *dev) 2178 { 2179 struct sfc_adapter_shared *sas = sfc_adapter_shared_by_eth_dev(dev); 2180 struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev); 2181 uint32_t logtype_main; 2182 struct sfc_adapter *sa; 2183 int rc; 2184 const efx_nic_cfg_t *encp; 2185 const struct rte_ether_addr *from; 2186 int ret; 2187 2188 if (sfc_efx_dev_class_get(pci_dev->device.devargs) != 2189 SFC_EFX_DEV_CLASS_NET) { 2190 SFC_GENERIC_LOG(DEBUG, 2191 "Incompatible device class: skip probing, should be probed by other sfc driver."); 2192 return 1; 2193 } 2194 2195 sfc_register_dp(); 2196 2197 logtype_main = sfc_register_logtype(&pci_dev->addr, 2198 SFC_LOGTYPE_MAIN_STR, 2199 RTE_LOG_NOTICE); 2200 2201 if (rte_eal_process_type() != RTE_PROC_PRIMARY) 2202 return -sfc_eth_dev_secondary_init(dev, logtype_main); 2203 2204 /* Required for logging */ 2205 ret = snprintf(sas->log_prefix, sizeof(sas->log_prefix), 2206 "PMD: sfc_efx " PCI_PRI_FMT " #%" PRIu16 ": ", 2207 pci_dev->addr.domain, pci_dev->addr.bus, 2208 pci_dev->addr.devid, pci_dev->addr.function, 2209 dev->data->port_id); 2210 if (ret < 0 || ret >= (int)sizeof(sas->log_prefix)) { 2211 SFC_GENERIC_LOG(ERR, 2212 "reserved log prefix is too short for " PCI_PRI_FMT, 2213 pci_dev->addr.domain, pci_dev->addr.bus, 2214 pci_dev->addr.devid, pci_dev->addr.function); 2215 return -EINVAL; 2216 } 2217 sas->pci_addr = pci_dev->addr; 2218 sas->port_id = dev->data->port_id; 2219 2220 /* 2221 * Allocate process private data from heap, since it should not 2222 * be located in shared memory allocated using rte_malloc() API. 2223 */ 2224 sa = calloc(1, sizeof(*sa)); 2225 if (sa == NULL) { 2226 rc = ENOMEM; 2227 goto fail_alloc_sa; 2228 } 2229 2230 dev->process_private = sa; 2231 2232 /* Required for logging */ 2233 sa->priv.shared = sas; 2234 sa->priv.logtype_main = logtype_main; 2235 2236 sa->eth_dev = dev; 2237 2238 /* Copy PCI device info to the dev->data */ 2239 rte_eth_copy_pci_info(dev, pci_dev); 2240 dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS; 2241 dev->data->dev_flags |= RTE_ETH_DEV_FLOW_OPS_THREAD_SAFE; 2242 2243 rc = sfc_kvargs_parse(sa); 2244 if (rc != 0) 2245 goto fail_kvargs_parse; 2246 2247 sfc_log_init(sa, "entry"); 2248 2249 dev->data->mac_addrs = rte_zmalloc("sfc", RTE_ETHER_ADDR_LEN, 0); 2250 if (dev->data->mac_addrs == NULL) { 2251 rc = ENOMEM; 2252 goto fail_mac_addrs; 2253 } 2254 2255 sfc_adapter_lock_init(sa); 2256 sfc_adapter_lock(sa); 2257 2258 sfc_log_init(sa, "probing"); 2259 rc = sfc_probe(sa); 2260 if (rc != 0) 2261 goto fail_probe; 2262 2263 sfc_log_init(sa, "set device ops"); 2264 rc = sfc_eth_dev_set_ops(dev); 2265 if (rc != 0) 2266 goto fail_set_ops; 2267 2268 sfc_log_init(sa, "attaching"); 2269 rc = sfc_attach(sa); 2270 if (rc != 0) 2271 goto fail_attach; 2272 2273 encp = efx_nic_cfg_get(sa->nic); 2274 2275 /* 2276 * The arguments are really reverse order in comparison to 2277 * Linux kernel. Copy from NIC config to Ethernet device data. 2278 */ 2279 from = (const struct rte_ether_addr *)(encp->enc_mac_addr); 2280 rte_ether_addr_copy(from, &dev->data->mac_addrs[0]); 2281 2282 sfc_adapter_unlock(sa); 2283 2284 sfc_log_init(sa, "done"); 2285 return 0; 2286 2287 fail_attach: 2288 sfc_eth_dev_clear_ops(dev); 2289 2290 fail_set_ops: 2291 sfc_unprobe(sa); 2292 2293 fail_probe: 2294 sfc_adapter_unlock(sa); 2295 sfc_adapter_lock_fini(sa); 2296 rte_free(dev->data->mac_addrs); 2297 dev->data->mac_addrs = NULL; 2298 2299 fail_mac_addrs: 2300 sfc_kvargs_cleanup(sa); 2301 2302 fail_kvargs_parse: 2303 sfc_log_init(sa, "failed %d", rc); 2304 dev->process_private = NULL; 2305 free(sa); 2306 2307 fail_alloc_sa: 2308 SFC_ASSERT(rc > 0); 2309 return -rc; 2310 } 2311 2312 static int 2313 sfc_eth_dev_uninit(struct rte_eth_dev *dev) 2314 { 2315 sfc_dev_close(dev); 2316 2317 return 0; 2318 } 2319 2320 static const struct rte_pci_id pci_id_sfc_efx_map[] = { 2321 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE) }, 2322 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_FARMINGDALE_VF) }, 2323 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT) }, 2324 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_GREENPORT_VF) }, 2325 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD) }, 2326 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD_VF) }, 2327 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2) }, 2328 { RTE_PCI_DEVICE(EFX_PCI_VENID_SFC, EFX_PCI_DEVID_MEDFORD2_VF) }, 2329 { RTE_PCI_DEVICE(EFX_PCI_VENID_XILINX, EFX_PCI_DEVID_RIVERHEAD) }, 2330 { .vendor_id = 0 /* sentinel */ } 2331 }; 2332 2333 static int sfc_eth_dev_pci_probe(struct rte_pci_driver *pci_drv __rte_unused, 2334 struct rte_pci_device *pci_dev) 2335 { 2336 return rte_eth_dev_pci_generic_probe(pci_dev, 2337 sizeof(struct sfc_adapter_shared), sfc_eth_dev_init); 2338 } 2339 2340 static int sfc_eth_dev_pci_remove(struct rte_pci_device *pci_dev) 2341 { 2342 return rte_eth_dev_pci_generic_remove(pci_dev, sfc_eth_dev_uninit); 2343 } 2344 2345 static struct rte_pci_driver sfc_efx_pmd = { 2346 .id_table = pci_id_sfc_efx_map, 2347 .drv_flags = 2348 RTE_PCI_DRV_INTR_LSC | 2349 RTE_PCI_DRV_NEED_MAPPING, 2350 .probe = sfc_eth_dev_pci_probe, 2351 .remove = sfc_eth_dev_pci_remove, 2352 }; 2353 2354 RTE_PMD_REGISTER_PCI(net_sfc_efx, sfc_efx_pmd); 2355 RTE_PMD_REGISTER_PCI_TABLE(net_sfc_efx, pci_id_sfc_efx_map); 2356 RTE_PMD_REGISTER_KMOD_DEP(net_sfc_efx, "* igb_uio | uio_pci_generic | vfio-pci"); 2357 RTE_PMD_REGISTER_PARAM_STRING(net_sfc_efx, 2358 SFC_KVARG_RX_DATAPATH "=" SFC_KVARG_VALUES_RX_DATAPATH " " 2359 SFC_KVARG_TX_DATAPATH "=" SFC_KVARG_VALUES_TX_DATAPATH " " 2360 SFC_KVARG_PERF_PROFILE "=" SFC_KVARG_VALUES_PERF_PROFILE " " 2361 SFC_KVARG_FW_VARIANT "=" SFC_KVARG_VALUES_FW_VARIANT " " 2362 SFC_KVARG_RXD_WAIT_TIMEOUT_NS "=<long> " 2363 SFC_KVARG_STATS_UPDATE_PERIOD_MS "=<long>"); 2364 2365 RTE_INIT(sfc_driver_register_logtype) 2366 { 2367 int ret; 2368 2369 ret = rte_log_register_type_and_pick_level(SFC_LOGTYPE_PREFIX "driver", 2370 RTE_LOG_NOTICE); 2371 sfc_logtype_driver = (ret < 0) ? RTE_LOGTYPE_PMD : ret; 2372 } 2373