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_NOTICE);
  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 void 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         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 void 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
 560 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 561 {
 562         static const uint32_t ptypes[] = {
 563                 RTE_PTYPE_L2_ETHER,
 564                 RTE_PTYPE_L2_ETHER_VLAN,
 565                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 566                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 567                 RTE_PTYPE_L4_TCP,
 568                 RTE_PTYPE_L4_UDP,
 569                 RTE_PTYPE_L4_FRAG,
 570                 RTE_PTYPE_L4_NONFRAG,
 571                 RTE_PTYPE_UNKNOWN
 572         };
 573         static const uint32_t ptypes_overlay[] = {
 574                 RTE_PTYPE_L2_ETHER,
 575                 RTE_PTYPE_L2_ETHER_VLAN,
 576                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 577                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 578                 RTE_PTYPE_L4_TCP,
 579                 RTE_PTYPE_L4_UDP,
 580                 RTE_PTYPE_L4_FRAG,
 581                 RTE_PTYPE_L4_NONFRAG,
 582                 RTE_PTYPE_TUNNEL_GRENAT,
 583                 RTE_PTYPE_INNER_L2_ETHER,
 584                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 585                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 586                 RTE_PTYPE_INNER_L4_TCP,
 587                 RTE_PTYPE_INNER_L4_UDP,
 588                 RTE_PTYPE_INNER_L4_FRAG,
 589                 RTE_PTYPE_INNER_L4_NONFRAG,
 590                 RTE_PTYPE_UNKNOWN
 591         };
 592
 593         if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
 594             dev->rx_pkt_burst != NULL) {
 595                 struct enic *enic = pmd_priv(dev);
 596                 if (enic->overlay_offload)
 597                         return ptypes_overlay;
 598                 else
 599                         return ptypes;
 600         }
 601         return NULL;
 602 }
 603
 604 static void enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 605 {
 606         struct enic *enic = pmd_priv(eth_dev);
 607
 608         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 609                 return;
 610
 611         ENICPMD_FUNC_TRACE();
 612
 613         enic->promisc = 1;
 614         enic_add_packet_filter(enic);
 615 }
 616
 617 static void enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 618 {
 619         struct enic *enic = pmd_priv(eth_dev);
 620
 621         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 622                 return;
 623
 624         ENICPMD_FUNC_TRACE();
 625         enic->promisc = 0;
 626         enic_add_packet_filter(enic);
 627 }
 628
 629 static void enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 630 {
 631         struct enic *enic = pmd_priv(eth_dev);
 632
 633         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 634                 return;
 635
 636         ENICPMD_FUNC_TRACE();
 637         enic->allmulti = 1;
 638         enic_add_packet_filter(enic);
 639 }
 640
 641 static void enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 642 {
 643         struct enic *enic = pmd_priv(eth_dev);
 644
 645         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 646                 return;
 647
 648         ENICPMD_FUNC_TRACE();
 649         enic->allmulti = 0;
 650         enic_add_packet_filter(enic);
 651 }
 652
 653 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
 654         struct rte_ether_addr *mac_addr,
 655         __rte_unused uint32_t index, __rte_unused uint32_t pool)
 656 {
 657         struct enic *enic = pmd_priv(eth_dev);
 658
 659         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 660                 return -E_RTE_SECONDARY;
 661
 662         ENICPMD_FUNC_TRACE();
 663         return enic_set_mac_address(enic, mac_addr->addr_bytes);
 664 }
 665
 666 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
 667 {
 668         struct enic *enic = pmd_priv(eth_dev);
 669
 670         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 671                 return;
 672
 673         ENICPMD_FUNC_TRACE();
 674         if (enic_del_mac_address(enic, index))
 675                 dev_err(enic, "del mac addr failed\n");
 676 }
 677
 678 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
 679                                 struct rte_ether_addr *addr)
 680 {
 681         struct enic *enic = pmd_priv(eth_dev);
 682         int ret;
 683
 684         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 685                 return -E_RTE_SECONDARY;
 686
 687         ENICPMD_FUNC_TRACE();
 688         ret = enic_del_mac_address(enic, 0);
 689         if (ret)
 690                 return ret;
 691         return enic_set_mac_address(enic, addr->addr_bytes);
 692 }
 693
 694 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
 695 {
 696         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 697
 698         rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
 699         ENICPMD_LOG(DEBUG, " %s address %s\n",
 700                      add ? "add" : "remove", mac_str);
 701 }
 702
 703 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
 704                                     struct rte_ether_addr *mc_addr_set,
 705                                     uint32_t nb_mc_addr)
 706 {
 707         struct enic *enic = pmd_priv(eth_dev);
 708         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 709         struct rte_ether_addr *addr;
 710         uint32_t i, j;
 711         int ret;
 712
 713         ENICPMD_FUNC_TRACE();
 714
 715         /* Validate the given addresses first */
 716         for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
 717                 addr = &mc_addr_set[i];
 718                 if (!rte_is_multicast_ether_addr(addr) ||
 719                     rte_is_broadcast_ether_addr(addr)) {
 720                         rte_ether_format_addr(mac_str,
 721                                         RTE_ETHER_ADDR_FMT_SIZE, addr);
 722                         ENICPMD_LOG(ERR, " invalid multicast address %s\n",
 723                                      mac_str);
 724                         return -EINVAL;
 725                 }
 726         }
 727
 728         /* Flush all if requested */
 729         if (nb_mc_addr == 0 || mc_addr_set == NULL) {
 730                 ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
 731                 for (i = 0; i < enic->mc_count; i++) {
 732                         addr = &enic->mc_addrs[i];
 733                         debug_log_add_del_addr(addr, false);
 734                         ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 735                         if (ret)
 736                                 return ret;
 737                 }
 738                 enic->mc_count = 0;
 739                 return 0;
 740         }
 741
 742         if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
 743                 ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
 744                              ENIC_MULTICAST_PERFECT_FILTERS);
 745                 return -ENOSPC;
 746         }
 747         /*
 748          * devcmd is slow, so apply the difference instead of flushing and
 749          * adding everything.
 750          * 1. Delete addresses on the NIC but not on the host
 751          */
 752         for (i = 0; i < enic->mc_count; i++) {
 753                 addr = &enic->mc_addrs[i];
 754                 for (j = 0; j < nb_mc_addr; j++) {
 755                         if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
 756                                 break;
 757                 }
 758                 if (j < nb_mc_addr)
 759                         continue;
 760                 debug_log_add_del_addr(addr, false);
 761                 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 762                 if (ret)
 763                         return ret;
 764         }
 765         /* 2. Add addresses on the host but not on the NIC */
 766         for (i = 0; i < nb_mc_addr; i++) {
 767                 addr = &mc_addr_set[i];
 768                 for (j = 0; j < enic->mc_count; j++) {
 769                         if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
 770                                 break;
 771                 }
 772                 if (j < enic->mc_count)
 773                         continue;
 774                 debug_log_add_del_addr(addr, true);
 775                 ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
 776                 if (ret)
 777                         return ret;
 778         }
 779         /* Keep a copy so we can flush/apply later on.. */
 780         memcpy(enic->mc_addrs, mc_addr_set,
 781                nb_mc_addr * sizeof(struct rte_ether_addr));
 782         enic->mc_count = nb_mc_addr;
 783         return 0;
 784 }
 785
 786 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 787 {
 788         struct enic *enic = pmd_priv(eth_dev);
 789
 790         ENICPMD_FUNC_TRACE();
 791         return enic_set_mtu(enic, mtu);
 792 }
 793
 794 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
 795                                       struct rte_eth_rss_reta_entry64
 796                                       *reta_conf,
 797                                       uint16_t reta_size)
 798 {
 799         struct enic *enic = pmd_priv(dev);
 800         uint16_t i, idx, shift;
 801
 802         ENICPMD_FUNC_TRACE();
 803         if (reta_size != ENIC_RSS_RETA_SIZE) {
 804                 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
 805                         reta_size, ENIC_RSS_RETA_SIZE);
 806                 return -EINVAL;
 807         }
 808
 809         for (i = 0; i < reta_size; i++) {
 810                 idx = i / RTE_RETA_GROUP_SIZE;
 811                 shift = i % RTE_RETA_GROUP_SIZE;
 812                 if (reta_conf[idx].mask & (1ULL << shift))
 813                         reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
 814                                 enic->rss_cpu.cpu[i / 4].b[i % 4]);
 815         }
 816
 817         return 0;
 818 }
 819
 820 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
 821                                        struct rte_eth_rss_reta_entry64
 822                                        *reta_conf,
 823                                        uint16_t reta_size)
 824 {
 825         struct enic *enic = pmd_priv(dev);
 826         union vnic_rss_cpu rss_cpu;
 827         uint16_t i, idx, shift;
 828
 829         ENICPMD_FUNC_TRACE();
 830         if (reta_size != ENIC_RSS_RETA_SIZE) {
 831                 dev_err(enic, "reta_update: wrong reta_size. given=%u"
 832                         " expected=%u\n",
 833                         reta_size, ENIC_RSS_RETA_SIZE);
 834                 return -EINVAL;
 835         }
 836         /*
 837          * Start with the current reta and modify it per reta_conf, as we
 838          * need to push the entire reta even if we only modify one entry.
 839          */
 840         rss_cpu = enic->rss_cpu;
 841         for (i = 0; i < reta_size; i++) {
 842                 idx = i / RTE_RETA_GROUP_SIZE;
 843                 shift = i % RTE_RETA_GROUP_SIZE;
 844                 if (reta_conf[idx].mask & (1ULL << shift))
 845                         rss_cpu.cpu[i / 4].b[i % 4] =
 846                                 enic_rte_rq_idx_to_sop_idx(
 847                                         reta_conf[idx].reta[shift]);
 848         }
 849         return enic_set_rss_reta(enic, &rss_cpu);
 850 }
 851
 852 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
 853                                        struct rte_eth_rss_conf *rss_conf)
 854 {
 855         struct enic *enic = pmd_priv(dev);
 856
 857         ENICPMD_FUNC_TRACE();
 858         return enic_set_rss_conf(enic, rss_conf);
 859 }
 860
 861 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 862                                          struct rte_eth_rss_conf *rss_conf)
 863 {
 864         struct enic *enic = pmd_priv(dev);
 865
 866         ENICPMD_FUNC_TRACE();
 867         if (rss_conf == NULL)
 868                 return -EINVAL;
 869         if (rss_conf->rss_key != NULL &&
 870             rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
 871                 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
 872                         " expected=%u+\n",
 873                         rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
 874                 return -EINVAL;
 875         }
 876         rss_conf->rss_hf = enic->rss_hf;
 877         if (rss_conf->rss_key != NULL) {
 878                 int i;
 879                 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
 880                         rss_conf->rss_key[i] =
 881                                 enic->rss_key.key[i / 10].b[i % 10];
 882                 }
 883                 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
 884         }
 885         return 0;
 886 }
 887
 888 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
 889                                      uint16_t rx_queue_id,
 890                                      struct rte_eth_rxq_info *qinfo)
 891 {
 892         struct enic *enic = pmd_priv(dev);
 893         struct vnic_rq *rq_sop;
 894         struct vnic_rq *rq_data;
 895         struct rte_eth_rxconf *conf;
 896         uint16_t sop_queue_idx;
 897         uint16_t data_queue_idx;
 898
 899         ENICPMD_FUNC_TRACE();
 900         sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 901         data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id);
 902         rq_sop = &enic->rq[sop_queue_idx];
 903         rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
 904         qinfo->mp = rq_sop->mp;
 905         qinfo->scattered_rx = rq_sop->data_queue_enable;
 906         qinfo->nb_desc = rq_sop->ring.desc_count;
 907         if (qinfo->scattered_rx)
 908                 qinfo->nb_desc += rq_data->ring.desc_count;
 909         conf = &qinfo->conf;
 910         memset(conf, 0, sizeof(*conf));
 911         conf->rx_free_thresh = rq_sop->rx_free_thresh;
 912         conf->rx_drop_en = 1;
 913         /*
 914          * Except VLAN stripping (port setting), all the checksum offloads
 915          * are always enabled.
 916          */
 917         conf->offloads = enic->rx_offload_capa;
 918         if (!enic->ig_vlan_strip_en)
 919                 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
 920         /* rx_thresh and other fields are not applicable for enic */
 921 }
 922
 923 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
 924                                      uint16_t tx_queue_id,
 925                                      struct rte_eth_txq_info *qinfo)
 926 {
 927         struct enic *enic = pmd_priv(dev);
 928         struct vnic_wq *wq = &enic->wq[tx_queue_id];
 929
 930         ENICPMD_FUNC_TRACE();
 931         qinfo->nb_desc = wq->ring.desc_count;
 932         memset(&qinfo->conf, 0, sizeof(qinfo->conf));
 933         qinfo->conf.offloads = wq->offloads;
 934         /* tx_thresh, and all the other fields are not applicable for enic */
 935 }
 936
 937 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
 938                                             uint16_t rx_queue_id)
 939 {
 940         struct enic *enic = pmd_priv(eth_dev);
 941
 942         ENICPMD_FUNC_TRACE();
 943         vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
 944         return 0;
 945 }
 946
 947 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
 948                                              uint16_t rx_queue_id)
 949 {
 950         struct enic *enic = pmd_priv(eth_dev);
 951
 952         ENICPMD_FUNC_TRACE();
 953         vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
 954         return 0;
 955 }
 956
 957 static int udp_tunnel_common_check(struct enic *enic,
 958                                    struct rte_eth_udp_tunnel *tnl)
 959 {
 960         if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
 961                 return -ENOTSUP;
 962         if (!enic->overlay_offload) {
 963                 ENICPMD_LOG(DEBUG, " vxlan (overlay offload) is not "
 964                              "supported\n");
 965                 return -ENOTSUP;
 966         }
 967         return 0;
 968 }
 969
 970 static int update_vxlan_port(struct enic *enic, uint16_t port)
 971 {
 972         if (vnic_dev_overlay_offload_cfg(enic->vdev,
 973                                          OVERLAY_CFG_VXLAN_PORT_UPDATE,
 974                                          port)) {
 975                 ENICPMD_LOG(DEBUG, " failed to update vxlan port\n");
 976                 return -EINVAL;
 977         }
 978         ENICPMD_LOG(DEBUG, " updated vxlan port to %u\n", port);
 979         enic->vxlan_port = port;
 980         return 0;
 981 }
 982
 983 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
 984                                            struct rte_eth_udp_tunnel *tnl)
 985 {
 986         struct enic *enic = pmd_priv(eth_dev);
 987         int ret;
 988
 989         ENICPMD_FUNC_TRACE();
 990         ret = udp_tunnel_common_check(enic, tnl);
 991         if (ret)
 992                 return ret;
 993         /*
 994          * The NIC has 1 configurable VXLAN port number. "Adding" a new port
 995          * number replaces it.
 996          */
 997         if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
 998                 ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
 999                              tnl->udp_port);
1000                 return -EINVAL;
1001         }
1002         return update_vxlan_port(enic, tnl->udp_port);
1003 }
1004
1005 static int enicpmd_dev_udp_tunnel_port_del(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          * Clear the previously set port number and restore the
1017          * hardware default port number. Some drivers disable VXLAN
1018          * offloads when there are no configured port numbers. But
1019          * enic does not do that as VXLAN is part of overlay offload,
1020          * which is tied to inner RSS and TSO.
1021          */
1022         if (tnl->udp_port != enic->vxlan_port) {
1023                 ENICPMD_LOG(DEBUG, " %u is not a configured vxlan port\n",
1024                              tnl->udp_port);
1025                 return -EINVAL;
1026         }
1027         return update_vxlan_port(enic, ENIC_DEFAULT_VXLAN_PORT);
1028 }
1029
1030 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1031                                       char *fw_version, size_t fw_size)
1032 {
1033         struct vnic_devcmd_fw_info *info;
1034         struct enic *enic;
1035         int ret;
1036
1037         ENICPMD_FUNC_TRACE();
1038         if (fw_version == NULL || fw_size <= 0)
1039                 return -EINVAL;
1040         enic = pmd_priv(eth_dev);
1041         ret = vnic_dev_fw_info(enic->vdev, &info);
1042         if (ret)
1043                 return ret;
1044         snprintf(fw_version, fw_size, "%s %s",
1045                  info->fw_version, info->fw_build);
1046         fw_version[fw_size - 1] = '\0';
1047         return 0;
1048 }
1049
1050 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1051         .dev_configure        = enicpmd_dev_configure,
1052         .dev_start            = enicpmd_dev_start,
1053         .dev_stop             = enicpmd_dev_stop,
1054         .dev_set_link_up      = NULL,
1055         .dev_set_link_down    = NULL,
1056         .dev_close            = enicpmd_dev_close,
1057         .promiscuous_enable   = enicpmd_dev_promiscuous_enable,
1058         .promiscuous_disable  = enicpmd_dev_promiscuous_disable,
1059         .allmulticast_enable  = enicpmd_dev_allmulticast_enable,
1060         .allmulticast_disable = enicpmd_dev_allmulticast_disable,
1061         .link_update          = enicpmd_dev_link_update,
1062         .stats_get            = enicpmd_dev_stats_get,
1063         .stats_reset          = enicpmd_dev_stats_reset,
1064         .queue_stats_mapping_set = NULL,
1065         .dev_infos_get        = enicpmd_dev_info_get,
1066         .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1067         .mtu_set              = enicpmd_mtu_set,
1068         .vlan_filter_set      = NULL,
1069         .vlan_tpid_set        = NULL,
1070         .vlan_offload_set     = enicpmd_vlan_offload_set,
1071         .vlan_strip_queue_set = NULL,
1072         .rx_queue_start       = enicpmd_dev_rx_queue_start,
1073         .rx_queue_stop        = enicpmd_dev_rx_queue_stop,
1074         .tx_queue_start       = enicpmd_dev_tx_queue_start,
1075         .tx_queue_stop        = enicpmd_dev_tx_queue_stop,
1076         .rx_queue_setup       = enicpmd_dev_rx_queue_setup,
1077         .rx_queue_release     = enicpmd_dev_rx_queue_release,
1078         .rx_queue_count       = enicpmd_dev_rx_queue_count,
1079         .rx_descriptor_done   = NULL,
1080         .tx_queue_setup       = enicpmd_dev_tx_queue_setup,
1081         .tx_queue_release     = enicpmd_dev_tx_queue_release,
1082         .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1083         .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1084         .rxq_info_get         = enicpmd_dev_rxq_info_get,
1085         .txq_info_get         = enicpmd_dev_txq_info_get,
1086         .dev_led_on           = NULL,
1087         .dev_led_off          = NULL,
1088         .flow_ctrl_get        = NULL,
1089         .flow_ctrl_set        = NULL,
1090         .priority_flow_ctrl_set = NULL,
1091         .mac_addr_add         = enicpmd_add_mac_addr,
1092         .mac_addr_remove      = enicpmd_remove_mac_addr,
1093         .mac_addr_set         = enicpmd_set_mac_addr,
1094         .set_mc_addr_list     = enicpmd_set_mc_addr_list,
1095         .filter_ctrl          = enicpmd_dev_filter_ctrl,
1096         .reta_query           = enicpmd_dev_rss_reta_query,
1097         .reta_update          = enicpmd_dev_rss_reta_update,
1098         .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
1099         .rss_hash_update      = enicpmd_dev_rss_hash_update,
1100         .udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
1101         .udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
1102         .fw_version_get       = enicpmd_dev_fw_version_get,
1103 };
1104
1105 static int enic_parse_zero_one(const char *key,
1106                                const char *value,
1107                                void *opaque)
1108 {
1109         struct enic *enic;
1110         bool b;
1111
1112         enic = (struct enic *)opaque;
1113         if (strcmp(value, "0") == 0) {
1114                 b = false;
1115         } else if (strcmp(value, "1") == 0) {
1116                 b = true;
1117         } else {
1118                 dev_err(enic, "Invalid value for %s"
1119                         ": expected=0|1 given=%s\n", key, value);
1120                 return -EINVAL;
1121         }
1122         if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1123                 enic->disable_overlay = b;
1124         if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1125                 enic->enable_avx2_rx = b;
1126         return 0;
1127 }
1128
1129 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1130                                       const char *value,
1131                                       void *opaque)
1132 {
1133         struct enic *enic;
1134
1135         enic = (struct enic *)opaque;
1136         if (strcmp(value, "trunk") == 0) {
1137                 /* Trunk mode: always tag */
1138                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1139         } else if (strcmp(value, "untag") == 0) {
1140                 /* Untag default VLAN mode: untag if VLAN = default VLAN */
1141                 enic->ig_vlan_rewrite_mode =
1142                         IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1143         } else if (strcmp(value, "priority") == 0) {
1144                 /*
1145                  * Priority-tag default VLAN mode: priority tag (VLAN header
1146                  * with ID=0) if VLAN = default
1147                  */
1148                 enic->ig_vlan_rewrite_mode =
1149                         IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1150         } else if (strcmp(value, "pass") == 0) {
1151                 /* Pass through mode: do not touch tags */
1152                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1153         } else {
1154                 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1155                         ": expected=trunk|untag|priority|pass given=%s\n",
1156                         value);
1157                 return -EINVAL;
1158         }
1159         return 0;
1160 }
1161
1162 static int enic_check_devargs(struct rte_eth_dev *dev)
1163 {
1164         static const char *const valid_keys[] = {
1165                 ENIC_DEVARG_DISABLE_OVERLAY,
1166                 ENIC_DEVARG_ENABLE_AVX2_RX,
1167                 ENIC_DEVARG_IG_VLAN_REWRITE,
1168                 NULL};
1169         struct enic *enic = pmd_priv(dev);
1170         struct rte_kvargs *kvlist;
1171
1172         ENICPMD_FUNC_TRACE();
1173
1174         enic->disable_overlay = false;
1175         enic->enable_avx2_rx = false;
1176         enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1177         if (!dev->device->devargs)
1178                 return 0;
1179         kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1180         if (!kvlist)
1181                 return -EINVAL;
1182         if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1183                                enic_parse_zero_one, enic) < 0 ||
1184             rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1185                                enic_parse_zero_one, enic) < 0 ||
1186             rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1187                                enic_parse_ig_vlan_rewrite, enic) < 0) {
1188                 rte_kvargs_free(kvlist);
1189                 return -EINVAL;
1190         }
1191         rte_kvargs_free(kvlist);
1192         return 0;
1193 }
1194
1195 /* Initialize the driver
1196  * It returns 0 on success.
1197  */
1198 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1199 {
1200         struct rte_pci_device *pdev;
1201         struct rte_pci_addr *addr;
1202         struct enic *enic = pmd_priv(eth_dev);
1203         int err;
1204
1205         ENICPMD_FUNC_TRACE();
1206
1207         enic->port_id = eth_dev->data->port_id;
1208         enic->rte_dev = eth_dev;
1209         eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1210         eth_dev->rx_pkt_burst = &enic_recv_pkts;
1211         eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1212         eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1213         /* Let rte_eth_dev_close() release the port resources */
1214         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1215
1216         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1217         rte_eth_copy_pci_info(eth_dev, pdev);
1218         enic->pdev = pdev;
1219         addr = &pdev->addr;
1220
1221         snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1222                 addr->domain, addr->bus, addr->devid, addr->function);
1223
1224         err = enic_check_devargs(eth_dev);
1225         if (err)
1226                 return err;
1227         return enic_probe(enic);
1228 }
1229
1230 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1231         struct rte_pci_device *pci_dev)
1232 {
1233         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1234                 eth_enicpmd_dev_init);
1235 }
1236
1237 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1238 {
1239         return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1240 }
1241
1242 static struct rte_pci_driver rte_enic_pmd = {
1243         .id_table = pci_id_enic_map,
1244         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1245         .probe = eth_enic_pci_probe,
1246         .remove = eth_enic_pci_remove,
1247 };
1248
1249 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1250 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1251 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1252 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1253         ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1254         ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1255         ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");