drivers/net/enic/enic_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2008-2017 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  */
   5
   6 #include <stdio.h>
   7 #include <stdint.h>
   8
   9 #include <rte_dev.h>
  10 #include <rte_pci.h>
  11 #include <rte_bus_pci.h>
  12 #include <rte_ethdev_driver.h>
  13 #include <rte_ethdev_pci.h>
  14 #include <rte_kvargs.h>
  15 #include <rte_string_fns.h>
  16
  17 #include "vnic_intr.h"
  18 #include "vnic_cq.h"
  19 #include "vnic_wq.h"
  20 #include "vnic_rq.h"
  21 #include "vnic_enet.h"
  22 #include "enic.h"
  23
  24 int enicpmd_logtype_init;
  25 int enicpmd_logtype_flow;
  26
  27 #define PMD_INIT_LOG(level, fmt, args...) \
  28         rte_log(RTE_LOG_ ## level, enicpmd_logtype_init, \
  29                 "%s" fmt "\n", __func__, ##args)
  30
  31 #define ENICPMD_FUNC_TRACE() PMD_INIT_LOG(DEBUG, " >>")
  32
  33 /*
  34  * The set of PCI devices this driver supports
  35  */
  36 #define CISCO_PCI_VENDOR_ID 0x1137
  37 static const struct rte_pci_id pci_id_enic_map[] = {
  38         { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  39         { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
  40         {.vendor_id = 0, /* sentinel */},
  41 };
  42
  43 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
  44 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
  45
  46 RTE_INIT(enicpmd_init_log);
  47 static void
  48 enicpmd_init_log(void)
  49 {
  50         enicpmd_logtype_init = rte_log_register("pmd.net.enic.init");
  51         if (enicpmd_logtype_init >= 0)
  52                 rte_log_set_level(enicpmd_logtype_init, RTE_LOG_NOTICE);
  53         enicpmd_logtype_flow = rte_log_register("pmd.net.enic.flow");
  54         if (enicpmd_logtype_flow >= 0)
  55                 rte_log_set_level(enicpmd_logtype_flow, RTE_LOG_NOTICE);
  56 }
  57
  58 static int
  59 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
  60                         enum rte_filter_op filter_op, void *arg)
  61 {
  62         struct enic *enic = pmd_priv(eth_dev);
  63         int ret = 0;
  64
  65         ENICPMD_FUNC_TRACE();
  66         if (filter_op == RTE_ETH_FILTER_NOP)
  67                 return 0;
  68
  69         if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
  70                 return -EINVAL;
  71
  72         switch (filter_op) {
  73         case RTE_ETH_FILTER_ADD:
  74         case RTE_ETH_FILTER_UPDATE:
  75                 ret = enic_fdir_add_fltr(enic,
  76                         (struct rte_eth_fdir_filter *)arg);
  77                 break;
  78
  79         case RTE_ETH_FILTER_DELETE:
  80                 ret = enic_fdir_del_fltr(enic,
  81                         (struct rte_eth_fdir_filter *)arg);
  82                 break;
  83
  84         case RTE_ETH_FILTER_STATS:
  85                 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
  86                 break;
  87
  88         case RTE_ETH_FILTER_FLUSH:
  89                 dev_warning(enic, "unsupported operation %u", filter_op);
  90                 ret = -ENOTSUP;
  91                 break;
  92         case RTE_ETH_FILTER_INFO:
  93                 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
  94                 break;
  95         default:
  96                 dev_err(enic, "unknown operation %u", filter_op);
  97                 ret = -EINVAL;
  98                 break;
  99         }
 100         return ret;
 101 }
 102
 103 static int
 104 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
 105                      enum rte_filter_type filter_type,
 106                      enum rte_filter_op filter_op,
 107                      void *arg)
 108 {
 109         int ret = 0;
 110
 111         ENICPMD_FUNC_TRACE();
 112
 113         switch (filter_type) {
 114         case RTE_ETH_FILTER_GENERIC:
 115                 if (filter_op != RTE_ETH_FILTER_GET)
 116                         return -EINVAL;
 117                 *(const void **)arg = &enic_flow_ops;
 118                 break;
 119         case RTE_ETH_FILTER_FDIR:
 120                 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
 121                 break;
 122         default:
 123                 dev_warning(enic, "Filter type (%d) not supported",
 124                         filter_type);
 125                 ret = -EINVAL;
 126                 break;
 127         }
 128
 129         return ret;
 130 }
 131
 132 static void enicpmd_dev_tx_queue_release(void *txq)
 133 {
 134         ENICPMD_FUNC_TRACE();
 135
 136         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 137                 return;
 138
 139         enic_free_wq(txq);
 140 }
 141
 142 static int enicpmd_dev_setup_intr(struct enic *enic)
 143 {
 144         int ret;
 145         unsigned int index;
 146
 147         ENICPMD_FUNC_TRACE();
 148
 149         /* Are we done with the init of all the queues? */
 150         for (index = 0; index < enic->cq_count; index++) {
 151                 if (!enic->cq[index].ctrl)
 152                         break;
 153         }
 154         if (enic->cq_count != index)
 155                 return 0;
 156         for (index = 0; index < enic->wq_count; index++) {
 157                 if (!enic->wq[index].ctrl)
 158                         break;
 159         }
 160         if (enic->wq_count != index)
 161                 return 0;
 162         /* check start of packet (SOP) RQs only in case scatter is disabled. */
 163         for (index = 0; index < enic->rq_count; index++) {
 164                 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
 165                         break;
 166         }
 167         if (enic->rq_count != index)
 168                 return 0;
 169
 170         ret = enic_alloc_intr_resources(enic);
 171         if (ret) {
 172                 dev_err(enic, "alloc intr failed\n");
 173                 return ret;
 174         }
 175         enic_init_vnic_resources(enic);
 176
 177         ret = enic_setup_finish(enic);
 178         if (ret)
 179                 dev_err(enic, "setup could not be finished\n");
 180
 181         return ret;
 182 }
 183
 184 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
 185         uint16_t queue_idx,
 186         uint16_t nb_desc,
 187         unsigned int socket_id,
 188         __rte_unused const struct rte_eth_txconf *tx_conf)
 189 {
 190         int ret;
 191         struct enic *enic = pmd_priv(eth_dev);
 192
 193         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 194                 return -E_RTE_SECONDARY;
 195
 196         ENICPMD_FUNC_TRACE();
 197         RTE_ASSERT(queue_idx < enic->conf_wq_count);
 198         eth_dev->data->tx_queues[queue_idx] = (void *)&enic->wq[queue_idx];
 199
 200         ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
 201         if (ret) {
 202                 dev_err(enic, "error in allocating wq\n");
 203                 return ret;
 204         }
 205
 206         return enicpmd_dev_setup_intr(enic);
 207 }
 208
 209 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
 210         uint16_t queue_idx)
 211 {
 212         struct enic *enic = pmd_priv(eth_dev);
 213
 214         ENICPMD_FUNC_TRACE();
 215
 216         enic_start_wq(enic, queue_idx);
 217
 218         return 0;
 219 }
 220
 221 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
 222         uint16_t queue_idx)
 223 {
 224         int ret;
 225         struct enic *enic = pmd_priv(eth_dev);
 226
 227         ENICPMD_FUNC_TRACE();
 228
 229         ret = enic_stop_wq(enic, queue_idx);
 230         if (ret)
 231                 dev_err(enic, "error in stopping wq %d\n", queue_idx);
 232
 233         return ret;
 234 }
 235
 236 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
 237         uint16_t queue_idx)
 238 {
 239         struct enic *enic = pmd_priv(eth_dev);
 240
 241         ENICPMD_FUNC_TRACE();
 242
 243         enic_start_rq(enic, queue_idx);
 244
 245         return 0;
 246 }
 247
 248 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
 249         uint16_t queue_idx)
 250 {
 251         int ret;
 252         struct enic *enic = pmd_priv(eth_dev);
 253
 254         ENICPMD_FUNC_TRACE();
 255
 256         ret = enic_stop_rq(enic, queue_idx);
 257         if (ret)
 258                 dev_err(enic, "error in stopping rq %d\n", queue_idx);
 259
 260         return ret;
 261 }
 262
 263 static void enicpmd_dev_rx_queue_release(void *rxq)
 264 {
 265         ENICPMD_FUNC_TRACE();
 266
 267         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 268                 return;
 269
 270         enic_free_rq(rxq);
 271 }
 272
 273 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
 274                                            uint16_t rx_queue_id)
 275 {
 276         struct enic *enic = pmd_priv(dev);
 277         uint32_t queue_count = 0;
 278         struct vnic_cq *cq;
 279         uint32_t cq_tail;
 280         uint16_t cq_idx;
 281         int rq_num;
 282
 283         rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 284         cq = &enic->cq[enic_cq_rq(enic, rq_num)];
 285         cq_idx = cq->to_clean;
 286
 287         cq_tail = ioread32(&cq->ctrl->cq_tail);
 288
 289         if (cq_tail < cq_idx)
 290                 cq_tail += cq->ring.desc_count;
 291
 292         queue_count = cq_tail - cq_idx;
 293
 294         return queue_count;
 295 }
 296
 297 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
 298         uint16_t queue_idx,
 299         uint16_t nb_desc,
 300         unsigned int socket_id,
 301         const struct rte_eth_rxconf *rx_conf,
 302         struct rte_mempool *mp)
 303 {
 304         int ret;
 305         struct enic *enic = pmd_priv(eth_dev);
 306
 307         ENICPMD_FUNC_TRACE();
 308
 309         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 310                 return -E_RTE_SECONDARY;
 311         RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
 312         eth_dev->data->rx_queues[queue_idx] =
 313                 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 314
 315         ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
 316                             rx_conf->rx_free_thresh);
 317         if (ret) {
 318                 dev_err(enic, "error in allocating rq\n");
 319                 return ret;
 320         }
 321
 322         return enicpmd_dev_setup_intr(enic);
 323 }
 324
 325 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
 326 {
 327         struct enic *enic = pmd_priv(eth_dev);
 328         uint64_t offloads;
 329
 330         ENICPMD_FUNC_TRACE();
 331
 332         offloads = eth_dev->data->dev_conf.rxmode.offloads;
 333         if (mask & ETH_VLAN_STRIP_MASK) {
 334                 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
 335                         enic->ig_vlan_strip_en = 1;
 336                 else
 337                         enic->ig_vlan_strip_en = 0;
 338         }
 339
 340         if ((mask & ETH_VLAN_FILTER_MASK) &&
 341             (offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
 342                 dev_warning(enic,
 343                         "Configuration of VLAN filter is not supported\n");
 344         }
 345
 346         if ((mask & ETH_VLAN_EXTEND_MASK) &&
 347             (offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)) {
 348                 dev_warning(enic,
 349                         "Configuration of extended VLAN is not supported\n");
 350         }
 351
 352         return enic_set_vlan_strip(enic);
 353 }
 354
 355 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
 356 {
 357         int ret;
 358         int mask;
 359         struct enic *enic = pmd_priv(eth_dev);
 360
 361         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 362                 return -E_RTE_SECONDARY;
 363
 364         ENICPMD_FUNC_TRACE();
 365         ret = enic_set_vnic_res(enic);
 366         if (ret) {
 367                 dev_err(enic, "Set vNIC resource num  failed, aborting\n");
 368                 return ret;
 369         }
 370
 371         enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
 372                                   DEV_RX_OFFLOAD_CHECKSUM);
 373         /* All vlan offload masks to apply the current settings */
 374         mask = ETH_VLAN_STRIP_MASK |
 375                 ETH_VLAN_FILTER_MASK |
 376                 ETH_VLAN_EXTEND_MASK;
 377         ret = enicpmd_vlan_offload_set(eth_dev, mask);
 378         if (ret) {
 379                 dev_err(enic, "Failed to configure VLAN offloads\n");
 380                 return ret;
 381         }
 382         /*
 383          * Initialize RSS with the default reta and key. If the user key is
 384          * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
 385          * default key.
 386          */
 387         return enic_init_rss_nic_cfg(enic);
 388 }
 389
 390 /* Start the device.
 391  * It returns 0 on success.
 392  */
 393 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
 394 {
 395         struct enic *enic = pmd_priv(eth_dev);
 396
 397         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 398                 return -E_RTE_SECONDARY;
 399
 400         ENICPMD_FUNC_TRACE();
 401         return enic_enable(enic);
 402 }
 403
 404 /*
 405  * Stop device: disable rx and tx functions to allow for reconfiguring.
 406  */
 407 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
 408 {
 409         struct rte_eth_link link;
 410         struct enic *enic = pmd_priv(eth_dev);
 411
 412         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 413                 return;
 414
 415         ENICPMD_FUNC_TRACE();
 416         enic_disable(enic);
 417
 418         memset(&link, 0, sizeof(link));
 419         rte_eth_linkstatus_set(eth_dev, &link);
 420 }
 421
 422 /*
 423  * Stop device.
 424  */
 425 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
 426 {
 427         struct enic *enic = pmd_priv(eth_dev);
 428
 429         ENICPMD_FUNC_TRACE();
 430         enic_remove(enic);
 431 }
 432
 433 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
 434         __rte_unused int wait_to_complete)
 435 {
 436         struct enic *enic = pmd_priv(eth_dev);
 437
 438         ENICPMD_FUNC_TRACE();
 439         return enic_link_update(enic);
 440 }
 441
 442 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
 443         struct rte_eth_stats *stats)
 444 {
 445         struct enic *enic = pmd_priv(eth_dev);
 446
 447         ENICPMD_FUNC_TRACE();
 448         return enic_dev_stats_get(enic, stats);
 449 }
 450
 451 static void enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
 452 {
 453         struct enic *enic = pmd_priv(eth_dev);
 454
 455         ENICPMD_FUNC_TRACE();
 456         enic_dev_stats_clear(enic);
 457 }
 458
 459 static void enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
 460         struct rte_eth_dev_info *device_info)
 461 {
 462         struct enic *enic = pmd_priv(eth_dev);
 463
 464         ENICPMD_FUNC_TRACE();
 465         /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
 466         device_info->max_rx_queues = enic->conf_rq_count / 2;
 467         device_info->max_tx_queues = enic->conf_wq_count;
 468         device_info->min_rx_bufsize = ENIC_MIN_MTU;
 469         /* "Max" mtu is not a typo. HW receives packet sizes up to the
 470          * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
 471          * a hint to the driver to size receive buffers accordingly so that
 472          * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
 473          * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
 474          * ignoring vNIC mtu.
 475          */
 476         device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
 477         device_info->max_mac_addrs = ENIC_MAX_MAC_ADDR;
 478         device_info->rx_offload_capa = enic->rx_offload_capa;
 479         device_info->tx_offload_capa = enic->tx_offload_capa;
 480         device_info->default_rxconf = (struct rte_eth_rxconf) {
 481                 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
 482         };
 483         device_info->reta_size = enic->reta_size;
 484         device_info->hash_key_size = enic->hash_key_size;
 485         device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
 486         device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
 487                 .nb_max = enic->config.rq_desc_count,
 488                 .nb_min = ENIC_MIN_RQ_DESCS,
 489                 .nb_align = ENIC_ALIGN_DESCS,
 490         };
 491         device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
 492                 .nb_max = enic->config.wq_desc_count,
 493                 .nb_min = ENIC_MIN_WQ_DESCS,
 494                 .nb_align = ENIC_ALIGN_DESCS,
 495                 .nb_seg_max = ENIC_TX_XMIT_MAX,
 496                 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
 497         };
 498         device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
 499                 .burst_size = ENIC_DEFAULT_RX_BURST,
 500                 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
 501                         ENIC_DEFAULT_RX_RING_SIZE),
 502                 .nb_queues = ENIC_DEFAULT_RX_RINGS,
 503         };
 504         device_info->default_txportconf = (struct rte_eth_dev_portconf) {
 505                 .burst_size = ENIC_DEFAULT_TX_BURST,
 506                 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
 507                         ENIC_DEFAULT_TX_RING_SIZE),
 508                 .nb_queues = ENIC_DEFAULT_TX_RINGS,
 509         };
 510 }
 511
 512 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 513 {
 514         static const uint32_t ptypes[] = {
 515                 RTE_PTYPE_L2_ETHER,
 516                 RTE_PTYPE_L2_ETHER_VLAN,
 517                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 518                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 519                 RTE_PTYPE_L4_TCP,
 520                 RTE_PTYPE_L4_UDP,
 521                 RTE_PTYPE_L4_FRAG,
 522                 RTE_PTYPE_L4_NONFRAG,
 523                 RTE_PTYPE_UNKNOWN
 524         };
 525
 526         if (dev->rx_pkt_burst == enic_recv_pkts)
 527                 return ptypes;
 528         return NULL;
 529 }
 530
 531 static void enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 532 {
 533         struct enic *enic = pmd_priv(eth_dev);
 534
 535         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 536                 return;
 537
 538         ENICPMD_FUNC_TRACE();
 539
 540         enic->promisc = 1;
 541         enic_add_packet_filter(enic);
 542 }
 543
 544 static void enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 545 {
 546         struct enic *enic = pmd_priv(eth_dev);
 547
 548         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 549                 return;
 550
 551         ENICPMD_FUNC_TRACE();
 552         enic->promisc = 0;
 553         enic_add_packet_filter(enic);
 554 }
 555
 556 static void enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 557 {
 558         struct enic *enic = pmd_priv(eth_dev);
 559
 560         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 561                 return;
 562
 563         ENICPMD_FUNC_TRACE();
 564         enic->allmulti = 1;
 565         enic_add_packet_filter(enic);
 566 }
 567
 568 static void enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 569 {
 570         struct enic *enic = pmd_priv(eth_dev);
 571
 572         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 573                 return;
 574
 575         ENICPMD_FUNC_TRACE();
 576         enic->allmulti = 0;
 577         enic_add_packet_filter(enic);
 578 }
 579
 580 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
 581         struct ether_addr *mac_addr,
 582         __rte_unused uint32_t index, __rte_unused uint32_t pool)
 583 {
 584         struct enic *enic = pmd_priv(eth_dev);
 585
 586         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 587                 return -E_RTE_SECONDARY;
 588
 589         ENICPMD_FUNC_TRACE();
 590         return enic_set_mac_address(enic, mac_addr->addr_bytes);
 591 }
 592
 593 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
 594 {
 595         struct enic *enic = pmd_priv(eth_dev);
 596
 597         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 598                 return;
 599
 600         ENICPMD_FUNC_TRACE();
 601         if (enic_del_mac_address(enic, index))
 602                 dev_err(enic, "del mac addr failed\n");
 603 }
 604
 605 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
 606                                 struct ether_addr *addr)
 607 {
 608         struct enic *enic = pmd_priv(eth_dev);
 609         int ret;
 610
 611         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 612                 return -E_RTE_SECONDARY;
 613
 614         ENICPMD_FUNC_TRACE();
 615         ret = enic_del_mac_address(enic, 0);
 616         if (ret)
 617                 return ret;
 618         return enic_set_mac_address(enic, addr->addr_bytes);
 619 }
 620
 621 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 622 {
 623         struct enic *enic = pmd_priv(eth_dev);
 624
 625         ENICPMD_FUNC_TRACE();
 626         return enic_set_mtu(enic, mtu);
 627 }
 628
 629 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
 630                                       struct rte_eth_rss_reta_entry64
 631                                       *reta_conf,
 632                                       uint16_t reta_size)
 633 {
 634         struct enic *enic = pmd_priv(dev);
 635         uint16_t i, idx, shift;
 636
 637         ENICPMD_FUNC_TRACE();
 638         if (reta_size != ENIC_RSS_RETA_SIZE) {
 639                 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
 640                         reta_size, ENIC_RSS_RETA_SIZE);
 641                 return -EINVAL;
 642         }
 643
 644         for (i = 0; i < reta_size; i++) {
 645                 idx = i / RTE_RETA_GROUP_SIZE;
 646                 shift = i % RTE_RETA_GROUP_SIZE;
 647                 if (reta_conf[idx].mask & (1ULL << shift))
 648                         reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
 649                                 enic->rss_cpu.cpu[i / 4].b[i % 4]);
 650         }
 651
 652         return 0;
 653 }
 654
 655 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
 656                                        struct rte_eth_rss_reta_entry64
 657                                        *reta_conf,
 658                                        uint16_t reta_size)
 659 {
 660         struct enic *enic = pmd_priv(dev);
 661         union vnic_rss_cpu rss_cpu;
 662         uint16_t i, idx, shift;
 663
 664         ENICPMD_FUNC_TRACE();
 665         if (reta_size != ENIC_RSS_RETA_SIZE) {
 666                 dev_err(enic, "reta_update: wrong reta_size. given=%u"
 667                         " expected=%u\n",
 668                         reta_size, ENIC_RSS_RETA_SIZE);
 669                 return -EINVAL;
 670         }
 671         /*
 672          * Start with the current reta and modify it per reta_conf, as we
 673          * need to push the entire reta even if we only modify one entry.
 674          */
 675         rss_cpu = enic->rss_cpu;
 676         for (i = 0; i < reta_size; i++) {
 677                 idx = i / RTE_RETA_GROUP_SIZE;
 678                 shift = i % RTE_RETA_GROUP_SIZE;
 679                 if (reta_conf[idx].mask & (1ULL << shift))
 680                         rss_cpu.cpu[i / 4].b[i % 4] =
 681                                 enic_rte_rq_idx_to_sop_idx(
 682                                         reta_conf[idx].reta[shift]);
 683         }
 684         return enic_set_rss_reta(enic, &rss_cpu);
 685 }
 686
 687 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
 688                                        struct rte_eth_rss_conf *rss_conf)
 689 {
 690         struct enic *enic = pmd_priv(dev);
 691
 692         ENICPMD_FUNC_TRACE();
 693         return enic_set_rss_conf(enic, rss_conf);
 694 }
 695
 696 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 697                                          struct rte_eth_rss_conf *rss_conf)
 698 {
 699         struct enic *enic = pmd_priv(dev);
 700
 701         ENICPMD_FUNC_TRACE();
 702         if (rss_conf == NULL)
 703                 return -EINVAL;
 704         if (rss_conf->rss_key != NULL &&
 705             rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
 706                 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
 707                         " expected=%u+\n",
 708                         rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
 709                 return -EINVAL;
 710         }
 711         rss_conf->rss_hf = enic->rss_hf;
 712         if (rss_conf->rss_key != NULL) {
 713                 int i;
 714                 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
 715                         rss_conf->rss_key[i] =
 716                                 enic->rss_key.key[i / 10].b[i % 10];
 717                 }
 718                 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
 719         }
 720         return 0;
 721 }
 722
 723 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
 724                                      uint16_t rx_queue_id,
 725                                      struct rte_eth_rxq_info *qinfo)
 726 {
 727         struct enic *enic = pmd_priv(dev);
 728         struct vnic_rq *rq_sop;
 729         struct vnic_rq *rq_data;
 730         struct rte_eth_rxconf *conf;
 731         uint16_t sop_queue_idx;
 732         uint16_t data_queue_idx;
 733
 734         ENICPMD_FUNC_TRACE();
 735         sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 736         data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id);
 737         rq_sop = &enic->rq[sop_queue_idx];
 738         rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
 739         qinfo->mp = rq_sop->mp;
 740         qinfo->scattered_rx = rq_sop->data_queue_enable;
 741         qinfo->nb_desc = rq_sop->ring.desc_count;
 742         if (qinfo->scattered_rx)
 743                 qinfo->nb_desc += rq_data->ring.desc_count;
 744         conf = &qinfo->conf;
 745         memset(conf, 0, sizeof(*conf));
 746         conf->rx_free_thresh = rq_sop->rx_free_thresh;
 747         conf->rx_drop_en = 1;
 748         /*
 749          * Except VLAN stripping (port setting), all the checksum offloads
 750          * are always enabled.
 751          */
 752         conf->offloads = enic->rx_offload_capa;
 753         if (!enic->ig_vlan_strip_en)
 754                 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
 755         /* rx_thresh and other fields are not applicable for enic */
 756 }
 757
 758 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
 759                                      uint16_t tx_queue_id,
 760                                      struct rte_eth_txq_info *qinfo)
 761 {
 762         struct enic *enic = pmd_priv(dev);
 763         struct vnic_wq *wq = &enic->wq[tx_queue_id];
 764
 765         ENICPMD_FUNC_TRACE();
 766         qinfo->nb_desc = wq->ring.desc_count;
 767         memset(&qinfo->conf, 0, sizeof(qinfo->conf));
 768         qinfo->conf.offloads = enic->tx_offload_capa;
 769         /* tx_thresh, and all the other fields are not applicable for enic */
 770 }
 771
 772 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
 773                                             uint16_t rx_queue_id)
 774 {
 775         struct enic *enic = pmd_priv(eth_dev);
 776
 777         ENICPMD_FUNC_TRACE();
 778         vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
 779         return 0;
 780 }
 781
 782 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
 783                                              uint16_t rx_queue_id)
 784 {
 785         struct enic *enic = pmd_priv(eth_dev);
 786
 787         ENICPMD_FUNC_TRACE();
 788         vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
 789         return 0;
 790 }
 791
 792 static int udp_tunnel_common_check(struct enic *enic,
 793                                    struct rte_eth_udp_tunnel *tnl)
 794 {
 795         if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
 796                 return -ENOTSUP;
 797         if (!enic->overlay_offload) {
 798                 PMD_INIT_LOG(DEBUG, " vxlan (overlay offload) is not "
 799                              "supported\n");
 800                 return -ENOTSUP;
 801         }
 802         return 0;
 803 }
 804
 805 static int update_vxlan_port(struct enic *enic, uint16_t port)
 806 {
 807         if (vnic_dev_overlay_offload_cfg(enic->vdev,
 808                                          OVERLAY_CFG_VXLAN_PORT_UPDATE,
 809                                          port)) {
 810                 PMD_INIT_LOG(DEBUG, " failed to update vxlan port\n");
 811                 return -EINVAL;
 812         }
 813         PMD_INIT_LOG(DEBUG, " updated vxlan port to %u\n", port);
 814         enic->vxlan_port = port;
 815         return 0;
 816 }
 817
 818 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
 819                                            struct rte_eth_udp_tunnel *tnl)
 820 {
 821         struct enic *enic = pmd_priv(eth_dev);
 822         int ret;
 823
 824         ENICPMD_FUNC_TRACE();
 825         ret = udp_tunnel_common_check(enic, tnl);
 826         if (ret)
 827                 return ret;
 828         /*
 829          * The NIC has 1 configurable VXLAN port number. "Adding" a new port
 830          * number replaces it.
 831          */
 832         if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
 833                 PMD_INIT_LOG(DEBUG, " %u is already configured or invalid\n",
 834                              tnl->udp_port);
 835                 return -EINVAL;
 836         }
 837         return update_vxlan_port(enic, tnl->udp_port);
 838 }
 839
 840 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
 841                                            struct rte_eth_udp_tunnel *tnl)
 842 {
 843         struct enic *enic = pmd_priv(eth_dev);
 844         int ret;
 845
 846         ENICPMD_FUNC_TRACE();
 847         ret = udp_tunnel_common_check(enic, tnl);
 848         if (ret)
 849                 return ret;
 850         /*
 851          * Clear the previously set port number and restore the
 852          * hardware default port number. Some drivers disable VXLAN
 853          * offloads when there are no configured port numbers. But
 854          * enic does not do that as VXLAN is part of overlay offload,
 855          * which is tied to inner RSS and TSO.
 856          */
 857         if (tnl->udp_port != enic->vxlan_port) {
 858                 PMD_INIT_LOG(DEBUG, " %u is not a configured vxlan port\n",
 859                              tnl->udp_port);
 860                 return -EINVAL;
 861         }
 862         return update_vxlan_port(enic, ENIC_DEFAULT_VXLAN_PORT);
 863 }
 864
 865 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
 866         .dev_configure        = enicpmd_dev_configure,
 867         .dev_start            = enicpmd_dev_start,
 868         .dev_stop             = enicpmd_dev_stop,
 869         .dev_set_link_up      = NULL,
 870         .dev_set_link_down    = NULL,
 871         .dev_close            = enicpmd_dev_close,
 872         .promiscuous_enable   = enicpmd_dev_promiscuous_enable,
 873         .promiscuous_disable  = enicpmd_dev_promiscuous_disable,
 874         .allmulticast_enable  = enicpmd_dev_allmulticast_enable,
 875         .allmulticast_disable = enicpmd_dev_allmulticast_disable,
 876         .link_update          = enicpmd_dev_link_update,
 877         .stats_get            = enicpmd_dev_stats_get,
 878         .stats_reset          = enicpmd_dev_stats_reset,
 879         .queue_stats_mapping_set = NULL,
 880         .dev_infos_get        = enicpmd_dev_info_get,
 881         .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
 882         .mtu_set              = enicpmd_mtu_set,
 883         .vlan_filter_set      = NULL,
 884         .vlan_tpid_set        = NULL,
 885         .vlan_offload_set     = enicpmd_vlan_offload_set,
 886         .vlan_strip_queue_set = NULL,
 887         .rx_queue_start       = enicpmd_dev_rx_queue_start,
 888         .rx_queue_stop        = enicpmd_dev_rx_queue_stop,
 889         .tx_queue_start       = enicpmd_dev_tx_queue_start,
 890         .tx_queue_stop        = enicpmd_dev_tx_queue_stop,
 891         .rx_queue_setup       = enicpmd_dev_rx_queue_setup,
 892         .rx_queue_release     = enicpmd_dev_rx_queue_release,
 893         .rx_queue_count       = enicpmd_dev_rx_queue_count,
 894         .rx_descriptor_done   = NULL,
 895         .tx_queue_setup       = enicpmd_dev_tx_queue_setup,
 896         .tx_queue_release     = enicpmd_dev_tx_queue_release,
 897         .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
 898         .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
 899         .rxq_info_get         = enicpmd_dev_rxq_info_get,
 900         .txq_info_get         = enicpmd_dev_txq_info_get,
 901         .dev_led_on           = NULL,
 902         .dev_led_off          = NULL,
 903         .flow_ctrl_get        = NULL,
 904         .flow_ctrl_set        = NULL,
 905         .priority_flow_ctrl_set = NULL,
 906         .mac_addr_add         = enicpmd_add_mac_addr,
 907         .mac_addr_remove      = enicpmd_remove_mac_addr,
 908         .mac_addr_set         = enicpmd_set_mac_addr,
 909         .filter_ctrl          = enicpmd_dev_filter_ctrl,
 910         .reta_query           = enicpmd_dev_rss_reta_query,
 911         .reta_update          = enicpmd_dev_rss_reta_update,
 912         .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
 913         .rss_hash_update      = enicpmd_dev_rss_hash_update,
 914         .udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
 915         .udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
 916 };
 917
 918 static int enic_parse_disable_overlay(__rte_unused const char *key,
 919                                       const char *value,
 920                                       void *opaque)
 921 {
 922         struct enic *enic;
 923
 924         enic = (struct enic *)opaque;
 925         if (strcmp(value, "0") == 0) {
 926                 enic->disable_overlay = false;
 927         } else if (strcmp(value, "1") == 0) {
 928                 enic->disable_overlay = true;
 929         } else {
 930                 dev_err(enic, "Invalid value for " ENIC_DEVARG_DISABLE_OVERLAY
 931                         ": expected=0|1 given=%s\n", value);
 932                 return -EINVAL;
 933         }
 934         return 0;
 935 }
 936
 937 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
 938                                       const char *value,
 939                                       void *opaque)
 940 {
 941         struct enic *enic;
 942
 943         enic = (struct enic *)opaque;
 944         if (strcmp(value, "trunk") == 0) {
 945                 /* Trunk mode: always tag */
 946                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
 947         } else if (strcmp(value, "untag") == 0) {
 948                 /* Untag default VLAN mode: untag if VLAN = default VLAN */
 949                 enic->ig_vlan_rewrite_mode =
 950                         IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
 951         } else if (strcmp(value, "priority") == 0) {
 952                 /*
 953                  * Priority-tag default VLAN mode: priority tag (VLAN header
 954                  * with ID=0) if VLAN = default
 955                  */
 956                 enic->ig_vlan_rewrite_mode =
 957                         IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
 958         } else if (strcmp(value, "pass") == 0) {
 959                 /* Pass through mode: do not touch tags */
 960                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
 961         } else {
 962                 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
 963                         ": expected=trunk|untag|priority|pass given=%s\n",
 964                         value);
 965                 return -EINVAL;
 966         }
 967         return 0;
 968 }
 969
 970 static int enic_check_devargs(struct rte_eth_dev *dev)
 971 {
 972         static const char *const valid_keys[] = {
 973                 ENIC_DEVARG_DISABLE_OVERLAY,
 974                 ENIC_DEVARG_IG_VLAN_REWRITE,
 975                 NULL};
 976         struct enic *enic = pmd_priv(dev);
 977         struct rte_kvargs *kvlist;
 978
 979         ENICPMD_FUNC_TRACE();
 980
 981         enic->disable_overlay = false;
 982         enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
 983         if (!dev->device->devargs)
 984                 return 0;
 985         kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
 986         if (!kvlist)
 987                 return -EINVAL;
 988         if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
 989                                enic_parse_disable_overlay, enic) < 0 ||
 990             rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
 991                                enic_parse_ig_vlan_rewrite, enic) < 0) {
 992                 rte_kvargs_free(kvlist);
 993                 return -EINVAL;
 994         }
 995         rte_kvargs_free(kvlist);
 996         return 0;
 997 }
 998
 999 struct enic *enicpmd_list_head = NULL;
1000 /* Initialize the driver
1001  * It returns 0 on success.
1002  */
1003 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1004 {
1005         struct rte_pci_device *pdev;
1006         struct rte_pci_addr *addr;
1007         struct enic *enic = pmd_priv(eth_dev);
1008         int err;
1009
1010         ENICPMD_FUNC_TRACE();
1011
1012         enic->port_id = eth_dev->data->port_id;
1013         enic->rte_dev = eth_dev;
1014         eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1015         eth_dev->rx_pkt_burst = &enic_recv_pkts;
1016         eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1017         eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1018
1019         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1020         rte_eth_copy_pci_info(eth_dev, pdev);
1021         enic->pdev = pdev;
1022         addr = &pdev->addr;
1023
1024         snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1025                 addr->domain, addr->bus, addr->devid, addr->function);
1026
1027         err = enic_check_devargs(eth_dev);
1028         if (err)
1029                 return err;
1030         return enic_probe(enic);
1031 }
1032
1033 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1034         struct rte_pci_device *pci_dev)
1035 {
1036         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1037                 eth_enicpmd_dev_init);
1038 }
1039
1040 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1041 {
1042         return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1043 }
1044
1045 static struct rte_pci_driver rte_enic_pmd = {
1046         .id_table = pci_id_enic_map,
1047         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1048         .probe = eth_enic_pci_probe,
1049         .remove = eth_enic_pci_remove,
1050 };
1051
1052 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1053 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1054 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1055 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1056         ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1057         ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");