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 enic_pmd_logtype;
  25
  26 /*
  27  * The set of PCI devices this driver supports
  28  */
  29 #define CISCO_PCI_VENDOR_ID 0x1137
  30 static const struct rte_pci_id pci_id_enic_map[] = {
  31         { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  32         { RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF) },
  33         {.vendor_id = 0, /* sentinel */},
  34 };
  35
  36 /* Supported link speeds of production VIC models */
  37 static const struct vic_speed_capa {
  38         uint16_t sub_devid;
  39         uint32_t capa;
  40 } vic_speed_capa_map[] = {
  41         { 0x0043, ETH_LINK_SPEED_10G }, /* VIC */
  42         { 0x0047, ETH_LINK_SPEED_10G }, /* P81E PCIe */
  43         { 0x0048, ETH_LINK_SPEED_10G }, /* M81KR Mezz */
  44         { 0x004f, ETH_LINK_SPEED_10G }, /* 1280 Mezz */
  45         { 0x0084, ETH_LINK_SPEED_10G }, /* 1240 MLOM */
  46         { 0x0085, ETH_LINK_SPEED_10G }, /* 1225 PCIe */
  47         { 0x00cd, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1285 PCIe */
  48         { 0x00ce, ETH_LINK_SPEED_10G }, /* 1225T PCIe */
  49         { 0x012a, ETH_LINK_SPEED_40G }, /* M4308 */
  50         { 0x012c, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1340 MLOM */
  51         { 0x012e, ETH_LINK_SPEED_10G }, /* 1227 PCIe */
  52         { 0x0137, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1380 Mezz */
  53         { 0x014d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1385 PCIe */
  54         { 0x015d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1387 MLOM */
  55         { 0x0215, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
  56                   ETH_LINK_SPEED_40G }, /* 1440 Mezz */
  57         { 0x0216, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
  58                   ETH_LINK_SPEED_40G }, /* 1480 MLOM */
  59         { 0x0217, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1455 PCIe */
  60         { 0x0218, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1457 MLOM */
  61         { 0x0219, ETH_LINK_SPEED_40G }, /* 1485 PCIe */
  62         { 0x021a, ETH_LINK_SPEED_40G }, /* 1487 MLOM */
  63         { 0x024a, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1495 PCIe */
  64         { 0x024b, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1497 MLOM */
  65         { 0, 0 }, /* End marker */
  66 };
  67
  68 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
  69 #define ENIC_DEVARG_ENABLE_AVX2_RX "enable-avx2-rx"
  70 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
  71
  72 RTE_INIT(enicpmd_init_log)
  73 {
  74         enic_pmd_logtype = rte_log_register("pmd.net.enic");
  75         if (enic_pmd_logtype >= 0)
  76                 rte_log_set_level(enic_pmd_logtype, RTE_LOG_INFO);
  77 }
  78
  79 static int
  80 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
  81                         enum rte_filter_op filter_op, void *arg)
  82 {
  83         struct enic *enic = pmd_priv(eth_dev);
  84         int ret = 0;
  85
  86         ENICPMD_FUNC_TRACE();
  87         if (filter_op == RTE_ETH_FILTER_NOP)
  88                 return 0;
  89
  90         if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
  91                 return -EINVAL;
  92
  93         switch (filter_op) {
  94         case RTE_ETH_FILTER_ADD:
  95         case RTE_ETH_FILTER_UPDATE:
  96                 ret = enic_fdir_add_fltr(enic,
  97                         (struct rte_eth_fdir_filter *)arg);
  98                 break;
  99
 100         case RTE_ETH_FILTER_DELETE:
 101                 ret = enic_fdir_del_fltr(enic,
 102                         (struct rte_eth_fdir_filter *)arg);
 103                 break;
 104
 105         case RTE_ETH_FILTER_STATS:
 106                 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
 107                 break;
 108
 109         case RTE_ETH_FILTER_FLUSH:
 110                 dev_warning(enic, "unsupported operation %u", filter_op);
 111                 ret = -ENOTSUP;
 112                 break;
 113         case RTE_ETH_FILTER_INFO:
 114                 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
 115                 break;
 116         default:
 117                 dev_err(enic, "unknown operation %u", filter_op);
 118                 ret = -EINVAL;
 119                 break;
 120         }
 121         return ret;
 122 }
 123
 124 static int
 125 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
 126                      enum rte_filter_type filter_type,
 127                      enum rte_filter_op filter_op,
 128                      void *arg)
 129 {
 130         int ret = 0;
 131
 132         ENICPMD_FUNC_TRACE();
 133
 134         switch (filter_type) {
 135         case RTE_ETH_FILTER_GENERIC:
 136                 if (filter_op != RTE_ETH_FILTER_GET)
 137                         return -EINVAL;
 138                 *(const void **)arg = &enic_flow_ops;
 139                 break;
 140         case RTE_ETH_FILTER_FDIR:
 141                 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
 142                 break;
 143         default:
 144                 dev_warning(enic, "Filter type (%d) not supported",
 145                         filter_type);
 146                 ret = -EINVAL;
 147                 break;
 148         }
 149
 150         return ret;
 151 }
 152
 153 static void enicpmd_dev_tx_queue_release(void *txq)
 154 {
 155         ENICPMD_FUNC_TRACE();
 156
 157         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 158                 return;
 159
 160         enic_free_wq(txq);
 161 }
 162
 163 static int enicpmd_dev_setup_intr(struct enic *enic)
 164 {
 165         int ret;
 166         unsigned int index;
 167
 168         ENICPMD_FUNC_TRACE();
 169
 170         /* Are we done with the init of all the queues? */
 171         for (index = 0; index < enic->cq_count; index++) {
 172                 if (!enic->cq[index].ctrl)
 173                         break;
 174         }
 175         if (enic->cq_count != index)
 176                 return 0;
 177         for (index = 0; index < enic->wq_count; index++) {
 178                 if (!enic->wq[index].ctrl)
 179                         break;
 180         }
 181         if (enic->wq_count != index)
 182                 return 0;
 183         /* check start of packet (SOP) RQs only in case scatter is disabled. */
 184         for (index = 0; index < enic->rq_count; index++) {
 185                 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
 186                         break;
 187         }
 188         if (enic->rq_count != index)
 189                 return 0;
 190
 191         ret = enic_alloc_intr_resources(enic);
 192         if (ret) {
 193                 dev_err(enic, "alloc intr failed\n");
 194                 return ret;
 195         }
 196         enic_init_vnic_resources(enic);
 197
 198         ret = enic_setup_finish(enic);
 199         if (ret)
 200                 dev_err(enic, "setup could not be finished\n");
 201
 202         return ret;
 203 }
 204
 205 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
 206         uint16_t queue_idx,
 207         uint16_t nb_desc,
 208         unsigned int socket_id,
 209         const struct rte_eth_txconf *tx_conf)
 210 {
 211         int ret;
 212         struct enic *enic = pmd_priv(eth_dev);
 213         struct vnic_wq *wq;
 214
 215         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 216                 return -E_RTE_SECONDARY;
 217
 218         ENICPMD_FUNC_TRACE();
 219         RTE_ASSERT(queue_idx < enic->conf_wq_count);
 220         wq = &enic->wq[queue_idx];
 221         wq->offloads = tx_conf->offloads |
 222                 eth_dev->data->dev_conf.txmode.offloads;
 223         eth_dev->data->tx_queues[queue_idx] = (void *)wq;
 224
 225         ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
 226         if (ret) {
 227                 dev_err(enic, "error in allocating wq\n");
 228                 return ret;
 229         }
 230
 231         return enicpmd_dev_setup_intr(enic);
 232 }
 233
 234 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
 235         uint16_t queue_idx)
 236 {
 237         struct enic *enic = pmd_priv(eth_dev);
 238
 239         ENICPMD_FUNC_TRACE();
 240
 241         enic_start_wq(enic, queue_idx);
 242
 243         return 0;
 244 }
 245
 246 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
 247         uint16_t queue_idx)
 248 {
 249         int ret;
 250         struct enic *enic = pmd_priv(eth_dev);
 251
 252         ENICPMD_FUNC_TRACE();
 253
 254         ret = enic_stop_wq(enic, queue_idx);
 255         if (ret)
 256                 dev_err(enic, "error in stopping wq %d\n", queue_idx);
 257
 258         return ret;
 259 }
 260
 261 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
 262         uint16_t queue_idx)
 263 {
 264         struct enic *enic = pmd_priv(eth_dev);
 265
 266         ENICPMD_FUNC_TRACE();
 267
 268         enic_start_rq(enic, queue_idx);
 269
 270         return 0;
 271 }
 272
 273 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
 274         uint16_t queue_idx)
 275 {
 276         int ret;
 277         struct enic *enic = pmd_priv(eth_dev);
 278
 279         ENICPMD_FUNC_TRACE();
 280
 281         ret = enic_stop_rq(enic, queue_idx);
 282         if (ret)
 283                 dev_err(enic, "error in stopping rq %d\n", queue_idx);
 284
 285         return ret;
 286 }
 287
 288 static void enicpmd_dev_rx_queue_release(void *rxq)
 289 {
 290         ENICPMD_FUNC_TRACE();
 291
 292         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 293                 return;
 294
 295         enic_free_rq(rxq);
 296 }
 297
 298 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
 299                                            uint16_t rx_queue_id)
 300 {
 301         struct enic *enic = pmd_priv(dev);
 302         uint32_t queue_count = 0;
 303         struct vnic_cq *cq;
 304         uint32_t cq_tail;
 305         uint16_t cq_idx;
 306         int rq_num;
 307
 308         rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 309         cq = &enic->cq[enic_cq_rq(enic, rq_num)];
 310         cq_idx = cq->to_clean;
 311
 312         cq_tail = ioread32(&cq->ctrl->cq_tail);
 313
 314         if (cq_tail < cq_idx)
 315                 cq_tail += cq->ring.desc_count;
 316
 317         queue_count = cq_tail - cq_idx;
 318
 319         return queue_count;
 320 }
 321
 322 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
 323         uint16_t queue_idx,
 324         uint16_t nb_desc,
 325         unsigned int socket_id,
 326         const struct rte_eth_rxconf *rx_conf,
 327         struct rte_mempool *mp)
 328 {
 329         int ret;
 330         struct enic *enic = pmd_priv(eth_dev);
 331
 332         ENICPMD_FUNC_TRACE();
 333
 334         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 335                 return -E_RTE_SECONDARY;
 336         RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
 337         eth_dev->data->rx_queues[queue_idx] =
 338                 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 339
 340         ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
 341                             rx_conf->rx_free_thresh);
 342         if (ret) {
 343                 dev_err(enic, "error in allocating rq\n");
 344                 return ret;
 345         }
 346
 347         return enicpmd_dev_setup_intr(enic);
 348 }
 349
 350 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
 351 {
 352         struct enic *enic = pmd_priv(eth_dev);
 353         uint64_t offloads;
 354
 355         ENICPMD_FUNC_TRACE();
 356
 357         offloads = eth_dev->data->dev_conf.rxmode.offloads;
 358         if (mask & ETH_VLAN_STRIP_MASK) {
 359                 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
 360                         enic->ig_vlan_strip_en = 1;
 361                 else
 362                         enic->ig_vlan_strip_en = 0;
 363         }
 364
 365         if ((mask & ETH_VLAN_FILTER_MASK) &&
 366             (offloads & DEV_RX_OFFLOAD_VLAN_FILTER)) {
 367                 dev_warning(enic,
 368                         "Configuration of VLAN filter is not supported\n");
 369         }
 370
 371         if ((mask & ETH_VLAN_EXTEND_MASK) &&
 372             (offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)) {
 373                 dev_warning(enic,
 374                         "Configuration of extended VLAN is not supported\n");
 375         }
 376
 377         return enic_set_vlan_strip(enic);
 378 }
 379
 380 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
 381 {
 382         int ret;
 383         int mask;
 384         struct enic *enic = pmd_priv(eth_dev);
 385
 386         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 387                 return -E_RTE_SECONDARY;
 388
 389         ENICPMD_FUNC_TRACE();
 390         ret = enic_set_vnic_res(enic);
 391         if (ret) {
 392                 dev_err(enic, "Set vNIC resource num  failed, aborting\n");
 393                 return ret;
 394         }
 395
 396         enic->mc_count = 0;
 397         enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
 398                                   DEV_RX_OFFLOAD_CHECKSUM);
 399         /* All vlan offload masks to apply the current settings */
 400         mask = ETH_VLAN_STRIP_MASK |
 401                 ETH_VLAN_FILTER_MASK |
 402                 ETH_VLAN_EXTEND_MASK;
 403         ret = enicpmd_vlan_offload_set(eth_dev, mask);
 404         if (ret) {
 405                 dev_err(enic, "Failed to configure VLAN offloads\n");
 406                 return ret;
 407         }
 408         /*
 409          * Initialize RSS with the default reta and key. If the user key is
 410          * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
 411          * default key.
 412          */
 413         return enic_init_rss_nic_cfg(enic);
 414 }
 415
 416 /* Start the device.
 417  * It returns 0 on success.
 418  */
 419 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
 420 {
 421         struct enic *enic = pmd_priv(eth_dev);
 422
 423         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 424                 return -E_RTE_SECONDARY;
 425
 426         ENICPMD_FUNC_TRACE();
 427         return enic_enable(enic);
 428 }
 429
 430 /*
 431  * Stop device: disable rx and tx functions to allow for reconfiguring.
 432  */
 433 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
 434 {
 435         struct rte_eth_link link;
 436         struct enic *enic = pmd_priv(eth_dev);
 437
 438         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 439                 return;
 440
 441         ENICPMD_FUNC_TRACE();
 442         enic_disable(enic);
 443
 444         memset(&link, 0, sizeof(link));
 445         rte_eth_linkstatus_set(eth_dev, &link);
 446 }
 447
 448 /*
 449  * Stop device.
 450  */
 451 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
 452 {
 453         struct enic *enic = pmd_priv(eth_dev);
 454
 455         ENICPMD_FUNC_TRACE();
 456         enic_remove(enic);
 457 }
 458
 459 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
 460         __rte_unused int wait_to_complete)
 461 {
 462         struct enic *enic = pmd_priv(eth_dev);
 463
 464         ENICPMD_FUNC_TRACE();
 465         return enic_link_update(enic);
 466 }
 467
 468 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
 469         struct rte_eth_stats *stats)
 470 {
 471         struct enic *enic = pmd_priv(eth_dev);
 472
 473         ENICPMD_FUNC_TRACE();
 474         return enic_dev_stats_get(enic, stats);
 475 }
 476
 477 static int enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
 478 {
 479         struct enic *enic = pmd_priv(eth_dev);
 480
 481         ENICPMD_FUNC_TRACE();
 482         return enic_dev_stats_clear(enic);
 483 }
 484
 485 static uint32_t speed_capa_from_pci_id(struct rte_eth_dev *eth_dev)
 486 {
 487         const struct vic_speed_capa *m;
 488         struct rte_pci_device *pdev;
 489         uint16_t id;
 490
 491         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
 492         id = pdev->id.subsystem_device_id;
 493         for (m = vic_speed_capa_map; m->sub_devid != 0; m++) {
 494                 if (m->sub_devid == id)
 495                         return m->capa;
 496         }
 497         /* 1300 and later models are at least 40G */
 498         if (id >= 0x0100)
 499                 return ETH_LINK_SPEED_40G;
 500         return ETH_LINK_SPEED_10G;
 501 }
 502
 503 static int enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
 504         struct rte_eth_dev_info *device_info)
 505 {
 506         struct enic *enic = pmd_priv(eth_dev);
 507
 508         ENICPMD_FUNC_TRACE();
 509         /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
 510         device_info->max_rx_queues = enic->conf_rq_count / 2;
 511         device_info->max_tx_queues = enic->conf_wq_count;
 512         device_info->min_rx_bufsize = ENIC_MIN_MTU;
 513         /* "Max" mtu is not a typo. HW receives packet sizes up to the
 514          * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
 515          * a hint to the driver to size receive buffers accordingly so that
 516          * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
 517          * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
 518          * ignoring vNIC mtu.
 519          */
 520         device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
 521         device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
 522         device_info->min_mtu = ENIC_MIN_MTU;
 523         device_info->max_mtu = enic->max_mtu;
 524         device_info->rx_offload_capa = enic->rx_offload_capa;
 525         device_info->tx_offload_capa = enic->tx_offload_capa;
 526         device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
 527         device_info->default_rxconf = (struct rte_eth_rxconf) {
 528                 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
 529         };
 530         device_info->reta_size = enic->reta_size;
 531         device_info->hash_key_size = enic->hash_key_size;
 532         device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
 533         device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
 534                 .nb_max = enic->config.rq_desc_count,
 535                 .nb_min = ENIC_MIN_RQ_DESCS,
 536                 .nb_align = ENIC_ALIGN_DESCS,
 537         };
 538         device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
 539                 .nb_max = enic->config.wq_desc_count,
 540                 .nb_min = ENIC_MIN_WQ_DESCS,
 541                 .nb_align = ENIC_ALIGN_DESCS,
 542                 .nb_seg_max = ENIC_TX_XMIT_MAX,
 543                 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
 544         };
 545         device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
 546                 .burst_size = ENIC_DEFAULT_RX_BURST,
 547                 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
 548                         ENIC_DEFAULT_RX_RING_SIZE),
 549                 .nb_queues = ENIC_DEFAULT_RX_RINGS,
 550         };
 551         device_info->default_txportconf = (struct rte_eth_dev_portconf) {
 552                 .burst_size = ENIC_DEFAULT_TX_BURST,
 553                 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
 554                         ENIC_DEFAULT_TX_RING_SIZE),
 555                 .nb_queues = ENIC_DEFAULT_TX_RINGS,
 556         };
 557         device_info->speed_capa = speed_capa_from_pci_id(eth_dev);
 558
 559         return 0;
 560 }
 561
 562 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 563 {
 564         static const uint32_t ptypes[] = {
 565                 RTE_PTYPE_L2_ETHER,
 566                 RTE_PTYPE_L2_ETHER_VLAN,
 567                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 568                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 569                 RTE_PTYPE_L4_TCP,
 570                 RTE_PTYPE_L4_UDP,
 571                 RTE_PTYPE_L4_FRAG,
 572                 RTE_PTYPE_L4_NONFRAG,
 573                 RTE_PTYPE_UNKNOWN
 574         };
 575         static const uint32_t ptypes_overlay[] = {
 576                 RTE_PTYPE_L2_ETHER,
 577                 RTE_PTYPE_L2_ETHER_VLAN,
 578                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 579                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 580                 RTE_PTYPE_L4_TCP,
 581                 RTE_PTYPE_L4_UDP,
 582                 RTE_PTYPE_L4_FRAG,
 583                 RTE_PTYPE_L4_NONFRAG,
 584                 RTE_PTYPE_TUNNEL_GRENAT,
 585                 RTE_PTYPE_INNER_L2_ETHER,
 586                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 587                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 588                 RTE_PTYPE_INNER_L4_TCP,
 589                 RTE_PTYPE_INNER_L4_UDP,
 590                 RTE_PTYPE_INNER_L4_FRAG,
 591                 RTE_PTYPE_INNER_L4_NONFRAG,
 592                 RTE_PTYPE_UNKNOWN
 593         };
 594
 595         if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
 596             dev->rx_pkt_burst != NULL) {
 597                 struct enic *enic = pmd_priv(dev);
 598                 if (enic->overlay_offload)
 599                         return ptypes_overlay;
 600                 else
 601                         return ptypes;
 602         }
 603         return NULL;
 604 }
 605
 606 static int enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 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
 616         enic->promisc = 1;
 617         ret = enic_add_packet_filter(enic);
 618         if (ret != 0)
 619                 enic->promisc = 0;
 620
 621         return ret;
 622 }
 623
 624 static int enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 625 {
 626         struct enic *enic = pmd_priv(eth_dev);
 627         int ret;
 628
 629         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 630                 return -E_RTE_SECONDARY;
 631
 632         ENICPMD_FUNC_TRACE();
 633         enic->promisc = 0;
 634         ret = enic_add_packet_filter(enic);
 635         if (ret != 0)
 636                 enic->promisc = 1;
 637
 638         return ret;
 639 }
 640
 641 static int enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 642 {
 643         struct enic *enic = pmd_priv(eth_dev);
 644         int ret;
 645
 646         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 647                 return -E_RTE_SECONDARY;
 648
 649         ENICPMD_FUNC_TRACE();
 650         enic->allmulti = 1;
 651         ret = enic_add_packet_filter(enic);
 652         if (ret != 0)
 653                 enic->allmulti = 0;
 654
 655         return ret;
 656 }
 657
 658 static int enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 659 {
 660         struct enic *enic = pmd_priv(eth_dev);
 661         int ret;
 662
 663         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 664                 return -E_RTE_SECONDARY;
 665
 666         ENICPMD_FUNC_TRACE();
 667         enic->allmulti = 0;
 668         ret = enic_add_packet_filter(enic);
 669         if (ret != 0)
 670                 enic->allmulti = 1;
 671
 672         return ret;
 673 }
 674
 675 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
 676         struct rte_ether_addr *mac_addr,
 677         __rte_unused uint32_t index, __rte_unused uint32_t pool)
 678 {
 679         struct enic *enic = pmd_priv(eth_dev);
 680
 681         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 682                 return -E_RTE_SECONDARY;
 683
 684         ENICPMD_FUNC_TRACE();
 685         return enic_set_mac_address(enic, mac_addr->addr_bytes);
 686 }
 687
 688 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
 689 {
 690         struct enic *enic = pmd_priv(eth_dev);
 691
 692         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 693                 return;
 694
 695         ENICPMD_FUNC_TRACE();
 696         if (enic_del_mac_address(enic, index))
 697                 dev_err(enic, "del mac addr failed\n");
 698 }
 699
 700 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
 701                                 struct rte_ether_addr *addr)
 702 {
 703         struct enic *enic = pmd_priv(eth_dev);
 704         int ret;
 705
 706         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 707                 return -E_RTE_SECONDARY;
 708
 709         ENICPMD_FUNC_TRACE();
 710         ret = enic_del_mac_address(enic, 0);
 711         if (ret)
 712                 return ret;
 713         return enic_set_mac_address(enic, addr->addr_bytes);
 714 }
 715
 716 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
 717 {
 718         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 719
 720         rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
 721         ENICPMD_LOG(DEBUG, " %s address %s\n",
 722                      add ? "add" : "remove", mac_str);
 723 }
 724
 725 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
 726                                     struct rte_ether_addr *mc_addr_set,
 727                                     uint32_t nb_mc_addr)
 728 {
 729         struct enic *enic = pmd_priv(eth_dev);
 730         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 731         struct rte_ether_addr *addr;
 732         uint32_t i, j;
 733         int ret;
 734
 735         ENICPMD_FUNC_TRACE();
 736
 737         /* Validate the given addresses first */
 738         for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
 739                 addr = &mc_addr_set[i];
 740                 if (!rte_is_multicast_ether_addr(addr) ||
 741                     rte_is_broadcast_ether_addr(addr)) {
 742                         rte_ether_format_addr(mac_str,
 743                                         RTE_ETHER_ADDR_FMT_SIZE, addr);
 744                         ENICPMD_LOG(ERR, " invalid multicast address %s\n",
 745                                      mac_str);
 746                         return -EINVAL;
 747                 }
 748         }
 749
 750         /* Flush all if requested */
 751         if (nb_mc_addr == 0 || mc_addr_set == NULL) {
 752                 ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
 753                 for (i = 0; i < enic->mc_count; i++) {
 754                         addr = &enic->mc_addrs[i];
 755                         debug_log_add_del_addr(addr, false);
 756                         ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 757                         if (ret)
 758                                 return ret;
 759                 }
 760                 enic->mc_count = 0;
 761                 return 0;
 762         }
 763
 764         if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
 765                 ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
 766                              ENIC_MULTICAST_PERFECT_FILTERS);
 767                 return -ENOSPC;
 768         }
 769         /*
 770          * devcmd is slow, so apply the difference instead of flushing and
 771          * adding everything.
 772          * 1. Delete addresses on the NIC but not on the host
 773          */
 774         for (i = 0; i < enic->mc_count; i++) {
 775                 addr = &enic->mc_addrs[i];
 776                 for (j = 0; j < nb_mc_addr; j++) {
 777                         if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
 778                                 break;
 779                 }
 780                 if (j < nb_mc_addr)
 781                         continue;
 782                 debug_log_add_del_addr(addr, false);
 783                 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 784                 if (ret)
 785                         return ret;
 786         }
 787         /* 2. Add addresses on the host but not on the NIC */
 788         for (i = 0; i < nb_mc_addr; i++) {
 789                 addr = &mc_addr_set[i];
 790                 for (j = 0; j < enic->mc_count; j++) {
 791                         if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
 792                                 break;
 793                 }
 794                 if (j < enic->mc_count)
 795                         continue;
 796                 debug_log_add_del_addr(addr, true);
 797                 ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
 798                 if (ret)
 799                         return ret;
 800         }
 801         /* Keep a copy so we can flush/apply later on.. */
 802         memcpy(enic->mc_addrs, mc_addr_set,
 803                nb_mc_addr * sizeof(struct rte_ether_addr));
 804         enic->mc_count = nb_mc_addr;
 805         return 0;
 806 }
 807
 808 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 809 {
 810         struct enic *enic = pmd_priv(eth_dev);
 811
 812         ENICPMD_FUNC_TRACE();
 813         return enic_set_mtu(enic, mtu);
 814 }
 815
 816 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
 817                                       struct rte_eth_rss_reta_entry64
 818                                       *reta_conf,
 819                                       uint16_t reta_size)
 820 {
 821         struct enic *enic = pmd_priv(dev);
 822         uint16_t i, idx, shift;
 823
 824         ENICPMD_FUNC_TRACE();
 825         if (reta_size != ENIC_RSS_RETA_SIZE) {
 826                 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
 827                         reta_size, ENIC_RSS_RETA_SIZE);
 828                 return -EINVAL;
 829         }
 830
 831         for (i = 0; i < reta_size; i++) {
 832                 idx = i / RTE_RETA_GROUP_SIZE;
 833                 shift = i % RTE_RETA_GROUP_SIZE;
 834                 if (reta_conf[idx].mask & (1ULL << shift))
 835                         reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
 836                                 enic->rss_cpu.cpu[i / 4].b[i % 4]);
 837         }
 838
 839         return 0;
 840 }
 841
 842 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
 843                                        struct rte_eth_rss_reta_entry64
 844                                        *reta_conf,
 845                                        uint16_t reta_size)
 846 {
 847         struct enic *enic = pmd_priv(dev);
 848         union vnic_rss_cpu rss_cpu;
 849         uint16_t i, idx, shift;
 850
 851         ENICPMD_FUNC_TRACE();
 852         if (reta_size != ENIC_RSS_RETA_SIZE) {
 853                 dev_err(enic, "reta_update: wrong reta_size. given=%u"
 854                         " expected=%u\n",
 855                         reta_size, ENIC_RSS_RETA_SIZE);
 856                 return -EINVAL;
 857         }
 858         /*
 859          * Start with the current reta and modify it per reta_conf, as we
 860          * need to push the entire reta even if we only modify one entry.
 861          */
 862         rss_cpu = enic->rss_cpu;
 863         for (i = 0; i < reta_size; i++) {
 864                 idx = i / RTE_RETA_GROUP_SIZE;
 865                 shift = i % RTE_RETA_GROUP_SIZE;
 866                 if (reta_conf[idx].mask & (1ULL << shift))
 867                         rss_cpu.cpu[i / 4].b[i % 4] =
 868                                 enic_rte_rq_idx_to_sop_idx(
 869                                         reta_conf[idx].reta[shift]);
 870         }
 871         return enic_set_rss_reta(enic, &rss_cpu);
 872 }
 873
 874 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
 875                                        struct rte_eth_rss_conf *rss_conf)
 876 {
 877         struct enic *enic = pmd_priv(dev);
 878
 879         ENICPMD_FUNC_TRACE();
 880         return enic_set_rss_conf(enic, rss_conf);
 881 }
 882
 883 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 884                                          struct rte_eth_rss_conf *rss_conf)
 885 {
 886         struct enic *enic = pmd_priv(dev);
 887
 888         ENICPMD_FUNC_TRACE();
 889         if (rss_conf == NULL)
 890                 return -EINVAL;
 891         if (rss_conf->rss_key != NULL &&
 892             rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
 893                 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
 894                         " expected=%u+\n",
 895                         rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
 896                 return -EINVAL;
 897         }
 898         rss_conf->rss_hf = enic->rss_hf;
 899         if (rss_conf->rss_key != NULL) {
 900                 int i;
 901                 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
 902                         rss_conf->rss_key[i] =
 903                                 enic->rss_key.key[i / 10].b[i % 10];
 904                 }
 905                 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
 906         }
 907         return 0;
 908 }
 909
 910 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
 911                                      uint16_t rx_queue_id,
 912                                      struct rte_eth_rxq_info *qinfo)
 913 {
 914         struct enic *enic = pmd_priv(dev);
 915         struct vnic_rq *rq_sop;
 916         struct vnic_rq *rq_data;
 917         struct rte_eth_rxconf *conf;
 918         uint16_t sop_queue_idx;
 919         uint16_t data_queue_idx;
 920
 921         ENICPMD_FUNC_TRACE();
 922         sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 923         data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id);
 924         rq_sop = &enic->rq[sop_queue_idx];
 925         rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
 926         qinfo->mp = rq_sop->mp;
 927         qinfo->scattered_rx = rq_sop->data_queue_enable;
 928         qinfo->nb_desc = rq_sop->ring.desc_count;
 929         if (qinfo->scattered_rx)
 930                 qinfo->nb_desc += rq_data->ring.desc_count;
 931         conf = &qinfo->conf;
 932         memset(conf, 0, sizeof(*conf));
 933         conf->rx_free_thresh = rq_sop->rx_free_thresh;
 934         conf->rx_drop_en = 1;
 935         /*
 936          * Except VLAN stripping (port setting), all the checksum offloads
 937          * are always enabled.
 938          */
 939         conf->offloads = enic->rx_offload_capa;
 940         if (!enic->ig_vlan_strip_en)
 941                 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
 942         /* rx_thresh and other fields are not applicable for enic */
 943 }
 944
 945 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
 946                                      uint16_t tx_queue_id,
 947                                      struct rte_eth_txq_info *qinfo)
 948 {
 949         struct enic *enic = pmd_priv(dev);
 950         struct vnic_wq *wq = &enic->wq[tx_queue_id];
 951
 952         ENICPMD_FUNC_TRACE();
 953         qinfo->nb_desc = wq->ring.desc_count;
 954         memset(&qinfo->conf, 0, sizeof(qinfo->conf));
 955         qinfo->conf.offloads = wq->offloads;
 956         /* tx_thresh, and all the other fields are not applicable for enic */
 957 }
 958
 959 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
 960                                             uint16_t rx_queue_id)
 961 {
 962         struct enic *enic = pmd_priv(eth_dev);
 963
 964         ENICPMD_FUNC_TRACE();
 965         vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
 966         return 0;
 967 }
 968
 969 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
 970                                              uint16_t rx_queue_id)
 971 {
 972         struct enic *enic = pmd_priv(eth_dev);
 973
 974         ENICPMD_FUNC_TRACE();
 975         vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
 976         return 0;
 977 }
 978
 979 static int udp_tunnel_common_check(struct enic *enic,
 980                                    struct rte_eth_udp_tunnel *tnl)
 981 {
 982         if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
 983                 return -ENOTSUP;
 984         if (!enic->overlay_offload) {
 985                 ENICPMD_LOG(DEBUG, " vxlan (overlay offload) is not "
 986                              "supported\n");
 987                 return -ENOTSUP;
 988         }
 989         return 0;
 990 }
 991
 992 static int update_vxlan_port(struct enic *enic, uint16_t port)
 993 {
 994         if (vnic_dev_overlay_offload_cfg(enic->vdev,
 995                                          OVERLAY_CFG_VXLAN_PORT_UPDATE,
 996                                          port)) {
 997                 ENICPMD_LOG(DEBUG, " failed to update vxlan port\n");
 998                 return -EINVAL;
 999         }
1000         ENICPMD_LOG(DEBUG, " updated vxlan port to %u\n", port);
1001         enic->vxlan_port = port;
1002         return 0;
1003 }
1004
1005 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
1006                                            struct rte_eth_udp_tunnel *tnl)
1007 {
1008         struct enic *enic = pmd_priv(eth_dev);
1009         int ret;
1010
1011         ENICPMD_FUNC_TRACE();
1012         ret = udp_tunnel_common_check(enic, tnl);
1013         if (ret)
1014                 return ret;
1015         /*
1016          * The NIC has 1 configurable VXLAN port number. "Adding" a new port
1017          * number replaces it.
1018          */
1019         if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
1020                 ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
1021                              tnl->udp_port);
1022                 return -EINVAL;
1023         }
1024         return update_vxlan_port(enic, tnl->udp_port);
1025 }
1026
1027 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
1028                                            struct rte_eth_udp_tunnel *tnl)
1029 {
1030         struct enic *enic = pmd_priv(eth_dev);
1031         int ret;
1032
1033         ENICPMD_FUNC_TRACE();
1034         ret = udp_tunnel_common_check(enic, tnl);
1035         if (ret)
1036                 return ret;
1037         /*
1038          * Clear the previously set port number and restore the
1039          * hardware default port number. Some drivers disable VXLAN
1040          * offloads when there are no configured port numbers. But
1041          * enic does not do that as VXLAN is part of overlay offload,
1042          * which is tied to inner RSS and TSO.
1043          */
1044         if (tnl->udp_port != enic->vxlan_port) {
1045                 ENICPMD_LOG(DEBUG, " %u is not a configured vxlan port\n",
1046                              tnl->udp_port);
1047                 return -EINVAL;
1048         }
1049         return update_vxlan_port(enic, ENIC_DEFAULT_VXLAN_PORT);
1050 }
1051
1052 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1053                                       char *fw_version, size_t fw_size)
1054 {
1055         struct vnic_devcmd_fw_info *info;
1056         struct enic *enic;
1057         int ret;
1058
1059         ENICPMD_FUNC_TRACE();
1060         if (fw_version == NULL || fw_size <= 0)
1061                 return -EINVAL;
1062         enic = pmd_priv(eth_dev);
1063         ret = vnic_dev_fw_info(enic->vdev, &info);
1064         if (ret)
1065                 return ret;
1066         snprintf(fw_version, fw_size, "%s %s",
1067                  info->fw_version, info->fw_build);
1068         fw_version[fw_size - 1] = '\0';
1069         return 0;
1070 }
1071
1072 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1073         .dev_configure        = enicpmd_dev_configure,
1074         .dev_start            = enicpmd_dev_start,
1075         .dev_stop             = enicpmd_dev_stop,
1076         .dev_set_link_up      = NULL,
1077         .dev_set_link_down    = NULL,
1078         .dev_close            = enicpmd_dev_close,
1079         .promiscuous_enable   = enicpmd_dev_promiscuous_enable,
1080         .promiscuous_disable  = enicpmd_dev_promiscuous_disable,
1081         .allmulticast_enable  = enicpmd_dev_allmulticast_enable,
1082         .allmulticast_disable = enicpmd_dev_allmulticast_disable,
1083         .link_update          = enicpmd_dev_link_update,
1084         .stats_get            = enicpmd_dev_stats_get,
1085         .stats_reset          = enicpmd_dev_stats_reset,
1086         .queue_stats_mapping_set = NULL,
1087         .dev_infos_get        = enicpmd_dev_info_get,
1088         .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1089         .mtu_set              = enicpmd_mtu_set,
1090         .vlan_filter_set      = NULL,
1091         .vlan_tpid_set        = NULL,
1092         .vlan_offload_set     = enicpmd_vlan_offload_set,
1093         .vlan_strip_queue_set = NULL,
1094         .rx_queue_start       = enicpmd_dev_rx_queue_start,
1095         .rx_queue_stop        = enicpmd_dev_rx_queue_stop,
1096         .tx_queue_start       = enicpmd_dev_tx_queue_start,
1097         .tx_queue_stop        = enicpmd_dev_tx_queue_stop,
1098         .rx_queue_setup       = enicpmd_dev_rx_queue_setup,
1099         .rx_queue_release     = enicpmd_dev_rx_queue_release,
1100         .rx_queue_count       = enicpmd_dev_rx_queue_count,
1101         .rx_descriptor_done   = NULL,
1102         .tx_queue_setup       = enicpmd_dev_tx_queue_setup,
1103         .tx_queue_release     = enicpmd_dev_tx_queue_release,
1104         .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1105         .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1106         .rxq_info_get         = enicpmd_dev_rxq_info_get,
1107         .txq_info_get         = enicpmd_dev_txq_info_get,
1108         .dev_led_on           = NULL,
1109         .dev_led_off          = NULL,
1110         .flow_ctrl_get        = NULL,
1111         .flow_ctrl_set        = NULL,
1112         .priority_flow_ctrl_set = NULL,
1113         .mac_addr_add         = enicpmd_add_mac_addr,
1114         .mac_addr_remove      = enicpmd_remove_mac_addr,
1115         .mac_addr_set         = enicpmd_set_mac_addr,
1116         .set_mc_addr_list     = enicpmd_set_mc_addr_list,
1117         .filter_ctrl          = enicpmd_dev_filter_ctrl,
1118         .reta_query           = enicpmd_dev_rss_reta_query,
1119         .reta_update          = enicpmd_dev_rss_reta_update,
1120         .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
1121         .rss_hash_update      = enicpmd_dev_rss_hash_update,
1122         .udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
1123         .udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
1124         .fw_version_get       = enicpmd_dev_fw_version_get,
1125 };
1126
1127 static int enic_parse_zero_one(const char *key,
1128                                const char *value,
1129                                void *opaque)
1130 {
1131         struct enic *enic;
1132         bool b;
1133
1134         enic = (struct enic *)opaque;
1135         if (strcmp(value, "0") == 0) {
1136                 b = false;
1137         } else if (strcmp(value, "1") == 0) {
1138                 b = true;
1139         } else {
1140                 dev_err(enic, "Invalid value for %s"
1141                         ": expected=0|1 given=%s\n", key, value);
1142                 return -EINVAL;
1143         }
1144         if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1145                 enic->disable_overlay = b;
1146         if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1147                 enic->enable_avx2_rx = b;
1148         return 0;
1149 }
1150
1151 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1152                                       const char *value,
1153                                       void *opaque)
1154 {
1155         struct enic *enic;
1156
1157         enic = (struct enic *)opaque;
1158         if (strcmp(value, "trunk") == 0) {
1159                 /* Trunk mode: always tag */
1160                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1161         } else if (strcmp(value, "untag") == 0) {
1162                 /* Untag default VLAN mode: untag if VLAN = default VLAN */
1163                 enic->ig_vlan_rewrite_mode =
1164                         IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1165         } else if (strcmp(value, "priority") == 0) {
1166                 /*
1167                  * Priority-tag default VLAN mode: priority tag (VLAN header
1168                  * with ID=0) if VLAN = default
1169                  */
1170                 enic->ig_vlan_rewrite_mode =
1171                         IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1172         } else if (strcmp(value, "pass") == 0) {
1173                 /* Pass through mode: do not touch tags */
1174                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1175         } else {
1176                 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1177                         ": expected=trunk|untag|priority|pass given=%s\n",
1178                         value);
1179                 return -EINVAL;
1180         }
1181         return 0;
1182 }
1183
1184 static int enic_check_devargs(struct rte_eth_dev *dev)
1185 {
1186         static const char *const valid_keys[] = {
1187                 ENIC_DEVARG_DISABLE_OVERLAY,
1188                 ENIC_DEVARG_ENABLE_AVX2_RX,
1189                 ENIC_DEVARG_IG_VLAN_REWRITE,
1190                 NULL};
1191         struct enic *enic = pmd_priv(dev);
1192         struct rte_kvargs *kvlist;
1193
1194         ENICPMD_FUNC_TRACE();
1195
1196         enic->disable_overlay = false;
1197         enic->enable_avx2_rx = false;
1198         enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1199         if (!dev->device->devargs)
1200                 return 0;
1201         kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1202         if (!kvlist)
1203                 return -EINVAL;
1204         if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1205                                enic_parse_zero_one, enic) < 0 ||
1206             rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1207                                enic_parse_zero_one, enic) < 0 ||
1208             rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1209                                enic_parse_ig_vlan_rewrite, enic) < 0) {
1210                 rte_kvargs_free(kvlist);
1211                 return -EINVAL;
1212         }
1213         rte_kvargs_free(kvlist);
1214         return 0;
1215 }
1216
1217 /* Initialize the driver
1218  * It returns 0 on success.
1219  */
1220 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1221 {
1222         struct rte_pci_device *pdev;
1223         struct rte_pci_addr *addr;
1224         struct enic *enic = pmd_priv(eth_dev);
1225         int err;
1226
1227         ENICPMD_FUNC_TRACE();
1228
1229         enic->port_id = eth_dev->data->port_id;
1230         enic->rte_dev = eth_dev;
1231         eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1232         eth_dev->rx_pkt_burst = &enic_recv_pkts;
1233         eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1234         eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1235         /* Let rte_eth_dev_close() release the port resources */
1236         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1237
1238         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1239         rte_eth_copy_pci_info(eth_dev, pdev);
1240         enic->pdev = pdev;
1241         addr = &pdev->addr;
1242
1243         snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1244                 addr->domain, addr->bus, addr->devid, addr->function);
1245
1246         err = enic_check_devargs(eth_dev);
1247         if (err)
1248                 return err;
1249         return enic_probe(enic);
1250 }
1251
1252 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1253         struct rte_pci_device *pci_dev)
1254 {
1255         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1256                 eth_enicpmd_dev_init);
1257 }
1258
1259 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1260 {
1261         return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1262 }
1263
1264 static struct rte_pci_driver rte_enic_pmd = {
1265         .id_table = pci_id_enic_map,
1266         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1267         .probe = eth_enic_pci_probe,
1268         .remove = eth_enic_pci_remove,
1269 };
1270
1271 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1272 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1273 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1274 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1275         ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1276         ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1277         ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");