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 /*
  25  * The set of PCI devices this driver supports
  26  */
  27 #define CISCO_PCI_VENDOR_ID 0x1137
  28 static const struct rte_pci_id pci_id_enic_map[] = {
  29         {RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET)},
  30         {RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_VF)},
  31         {RTE_PCI_DEVICE(CISCO_PCI_VENDOR_ID, PCI_DEVICE_ID_CISCO_VIC_ENET_SN)},
  32         {.vendor_id = 0, /* sentinel */},
  33 };
  34
  35 /* Supported link speeds of production VIC models */
  36 static const struct vic_speed_capa {
  37         uint16_t sub_devid;
  38         uint32_t capa;
  39 } vic_speed_capa_map[] = {
  40         { 0x0043, ETH_LINK_SPEED_10G }, /* VIC */
  41         { 0x0047, ETH_LINK_SPEED_10G }, /* P81E PCIe */
  42         { 0x0048, ETH_LINK_SPEED_10G }, /* M81KR Mezz */
  43         { 0x004f, ETH_LINK_SPEED_10G }, /* 1280 Mezz */
  44         { 0x0084, ETH_LINK_SPEED_10G }, /* 1240 MLOM */
  45         { 0x0085, ETH_LINK_SPEED_10G }, /* 1225 PCIe */
  46         { 0x00cd, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1285 PCIe */
  47         { 0x00ce, ETH_LINK_SPEED_10G }, /* 1225T PCIe */
  48         { 0x012a, ETH_LINK_SPEED_40G }, /* M4308 */
  49         { 0x012c, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1340 MLOM */
  50         { 0x012e, ETH_LINK_SPEED_10G }, /* 1227 PCIe */
  51         { 0x0137, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1380 Mezz */
  52         { 0x014d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1385 PCIe */
  53         { 0x015d, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_40G }, /* 1387 MLOM */
  54         { 0x0215, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
  55                   ETH_LINK_SPEED_40G }, /* 1440 Mezz */
  56         { 0x0216, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G |
  57                   ETH_LINK_SPEED_40G }, /* 1480 MLOM */
  58         { 0x0217, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1455 PCIe */
  59         { 0x0218, ETH_LINK_SPEED_10G | ETH_LINK_SPEED_25G }, /* 1457 MLOM */
  60         { 0x0219, ETH_LINK_SPEED_40G }, /* 1485 PCIe */
  61         { 0x021a, ETH_LINK_SPEED_40G }, /* 1487 MLOM */
  62         { 0x024a, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1495 PCIe */
  63         { 0x024b, ETH_LINK_SPEED_40G | ETH_LINK_SPEED_100G }, /* 1497 MLOM */
  64         { 0, 0 }, /* End marker */
  65 };
  66
  67 #define ENIC_DEVARG_DISABLE_OVERLAY "disable-overlay"
  68 #define ENIC_DEVARG_ENABLE_AVX2_RX "enable-avx2-rx"
  69 #define ENIC_DEVARG_GENEVE_OPT "geneve-opt"
  70 #define ENIC_DEVARG_IG_VLAN_REWRITE "ig-vlan-rewrite"
  71
  72 RTE_LOG_REGISTER(enic_pmd_logtype, pmd.net.enic, INFO);
  73
  74 static int
  75 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
  76                         enum rte_filter_op filter_op, void *arg)
  77 {
  78         struct enic *enic = pmd_priv(eth_dev);
  79         int ret = 0;
  80
  81         ENICPMD_FUNC_TRACE();
  82         if (filter_op == RTE_ETH_FILTER_NOP)
  83                 return 0;
  84
  85         if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
  86                 return -EINVAL;
  87
  88         switch (filter_op) {
  89         case RTE_ETH_FILTER_ADD:
  90         case RTE_ETH_FILTER_UPDATE:
  91                 ret = enic_fdir_add_fltr(enic,
  92                         (struct rte_eth_fdir_filter *)arg);
  93                 break;
  94
  95         case RTE_ETH_FILTER_DELETE:
  96                 ret = enic_fdir_del_fltr(enic,
  97                         (struct rte_eth_fdir_filter *)arg);
  98                 break;
  99
 100         case RTE_ETH_FILTER_STATS:
 101                 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
 102                 break;
 103
 104         case RTE_ETH_FILTER_FLUSH:
 105                 dev_warning(enic, "unsupported operation %u", filter_op);
 106                 ret = -ENOTSUP;
 107                 break;
 108         case RTE_ETH_FILTER_INFO:
 109                 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
 110                 break;
 111         default:
 112                 dev_err(enic, "unknown operation %u", filter_op);
 113                 ret = -EINVAL;
 114                 break;
 115         }
 116         return ret;
 117 }
 118
 119 static int
 120 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
 121                      enum rte_filter_type filter_type,
 122                      enum rte_filter_op filter_op,
 123                      void *arg)
 124 {
 125         struct enic *enic = pmd_priv(dev);
 126         int ret = 0;
 127
 128         ENICPMD_FUNC_TRACE();
 129
 130         /*
 131          * Currently, when Geneve with options offload is enabled, host
 132          * cannot insert match-action rules.
 133          */
 134         if (enic->geneve_opt_enabled)
 135                 return -ENOTSUP;
 136         switch (filter_type) {
 137         case RTE_ETH_FILTER_GENERIC:
 138                 if (filter_op != RTE_ETH_FILTER_GET)
 139                         return -EINVAL;
 140                 if (enic->flow_filter_mode == FILTER_FLOWMAN)
 141                         *(const void **)arg = &enic_fm_flow_ops;
 142                 else
 143                         *(const void **)arg = &enic_flow_ops;
 144                 break;
 145         case RTE_ETH_FILTER_FDIR:
 146                 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
 147                 break;
 148         default:
 149                 dev_warning(enic, "Filter type (%d) not supported",
 150                         filter_type);
 151                 ret = -EINVAL;
 152                 break;
 153         }
 154
 155         return ret;
 156 }
 157
 158 static void enicpmd_dev_tx_queue_release(void *txq)
 159 {
 160         ENICPMD_FUNC_TRACE();
 161
 162         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 163                 return;
 164
 165         enic_free_wq(txq);
 166 }
 167
 168 static int enicpmd_dev_setup_intr(struct enic *enic)
 169 {
 170         int ret;
 171         unsigned int index;
 172
 173         ENICPMD_FUNC_TRACE();
 174
 175         /* Are we done with the init of all the queues? */
 176         for (index = 0; index < enic->cq_count; index++) {
 177                 if (!enic->cq[index].ctrl)
 178                         break;
 179         }
 180         if (enic->cq_count != index)
 181                 return 0;
 182         for (index = 0; index < enic->wq_count; index++) {
 183                 if (!enic->wq[index].ctrl)
 184                         break;
 185         }
 186         if (enic->wq_count != index)
 187                 return 0;
 188         /* check start of packet (SOP) RQs only in case scatter is disabled. */
 189         for (index = 0; index < enic->rq_count; index++) {
 190                 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
 191                         break;
 192         }
 193         if (enic->rq_count != index)
 194                 return 0;
 195
 196         ret = enic_alloc_intr_resources(enic);
 197         if (ret) {
 198                 dev_err(enic, "alloc intr failed\n");
 199                 return ret;
 200         }
 201         enic_init_vnic_resources(enic);
 202
 203         ret = enic_setup_finish(enic);
 204         if (ret)
 205                 dev_err(enic, "setup could not be finished\n");
 206
 207         return ret;
 208 }
 209
 210 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
 211         uint16_t queue_idx,
 212         uint16_t nb_desc,
 213         unsigned int socket_id,
 214         const struct rte_eth_txconf *tx_conf)
 215 {
 216         int ret;
 217         struct enic *enic = pmd_priv(eth_dev);
 218         struct vnic_wq *wq;
 219
 220         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 221                 return -E_RTE_SECONDARY;
 222
 223         ENICPMD_FUNC_TRACE();
 224         RTE_ASSERT(queue_idx < enic->conf_wq_count);
 225         wq = &enic->wq[queue_idx];
 226         wq->offloads = tx_conf->offloads |
 227                 eth_dev->data->dev_conf.txmode.offloads;
 228         eth_dev->data->tx_queues[queue_idx] = (void *)wq;
 229
 230         ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
 231         if (ret) {
 232                 dev_err(enic, "error in allocating wq\n");
 233                 return ret;
 234         }
 235
 236         return enicpmd_dev_setup_intr(enic);
 237 }
 238
 239 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
 240         uint16_t queue_idx)
 241 {
 242         struct enic *enic = pmd_priv(eth_dev);
 243
 244         ENICPMD_FUNC_TRACE();
 245
 246         enic_start_wq(enic, queue_idx);
 247
 248         return 0;
 249 }
 250
 251 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
 252         uint16_t queue_idx)
 253 {
 254         int ret;
 255         struct enic *enic = pmd_priv(eth_dev);
 256
 257         ENICPMD_FUNC_TRACE();
 258
 259         ret = enic_stop_wq(enic, queue_idx);
 260         if (ret)
 261                 dev_err(enic, "error in stopping wq %d\n", queue_idx);
 262
 263         return ret;
 264 }
 265
 266 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
 267         uint16_t queue_idx)
 268 {
 269         struct enic *enic = pmd_priv(eth_dev);
 270
 271         ENICPMD_FUNC_TRACE();
 272
 273         enic_start_rq(enic, queue_idx);
 274
 275         return 0;
 276 }
 277
 278 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
 279         uint16_t queue_idx)
 280 {
 281         int ret;
 282         struct enic *enic = pmd_priv(eth_dev);
 283
 284         ENICPMD_FUNC_TRACE();
 285
 286         ret = enic_stop_rq(enic, queue_idx);
 287         if (ret)
 288                 dev_err(enic, "error in stopping rq %d\n", queue_idx);
 289
 290         return ret;
 291 }
 292
 293 static void enicpmd_dev_rx_queue_release(void *rxq)
 294 {
 295         ENICPMD_FUNC_TRACE();
 296
 297         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 298                 return;
 299
 300         enic_free_rq(rxq);
 301 }
 302
 303 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
 304                                            uint16_t rx_queue_id)
 305 {
 306         struct enic *enic = pmd_priv(dev);
 307         uint32_t queue_count = 0;
 308         struct vnic_cq *cq;
 309         uint32_t cq_tail;
 310         uint16_t cq_idx;
 311         int rq_num;
 312
 313         rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 314         cq = &enic->cq[enic_cq_rq(enic, rq_num)];
 315         cq_idx = cq->to_clean;
 316
 317         cq_tail = ioread32(&cq->ctrl->cq_tail);
 318
 319         if (cq_tail < cq_idx)
 320                 cq_tail += cq->ring.desc_count;
 321
 322         queue_count = cq_tail - cq_idx;
 323
 324         return queue_count;
 325 }
 326
 327 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
 328         uint16_t queue_idx,
 329         uint16_t nb_desc,
 330         unsigned int socket_id,
 331         const struct rte_eth_rxconf *rx_conf,
 332         struct rte_mempool *mp)
 333 {
 334         int ret;
 335         struct enic *enic = pmd_priv(eth_dev);
 336
 337         ENICPMD_FUNC_TRACE();
 338
 339         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 340                 return -E_RTE_SECONDARY;
 341         RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
 342         eth_dev->data->rx_queues[queue_idx] =
 343                 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 344
 345         ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
 346                             rx_conf->rx_free_thresh);
 347         if (ret) {
 348                 dev_err(enic, "error in allocating rq\n");
 349                 return ret;
 350         }
 351
 352         return enicpmd_dev_setup_intr(enic);
 353 }
 354
 355 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
 356 {
 357         struct enic *enic = pmd_priv(eth_dev);
 358         uint64_t offloads;
 359
 360         ENICPMD_FUNC_TRACE();
 361
 362         offloads = eth_dev->data->dev_conf.rxmode.offloads;
 363         if (mask & ETH_VLAN_STRIP_MASK) {
 364                 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
 365                         enic->ig_vlan_strip_en = 1;
 366                 else
 367                         enic->ig_vlan_strip_en = 0;
 368         }
 369
 370         return enic_set_vlan_strip(enic);
 371 }
 372
 373 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
 374 {
 375         int ret;
 376         int mask;
 377         struct enic *enic = pmd_priv(eth_dev);
 378
 379         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 380                 return -E_RTE_SECONDARY;
 381
 382         ENICPMD_FUNC_TRACE();
 383         ret = enic_set_vnic_res(enic);
 384         if (ret) {
 385                 dev_err(enic, "Set vNIC resource num  failed, aborting\n");
 386                 return ret;
 387         }
 388
 389         if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 390                 eth_dev->data->dev_conf.rxmode.offloads |=
 391                         DEV_RX_OFFLOAD_RSS_HASH;
 392
 393         enic->mc_count = 0;
 394         enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
 395                                   DEV_RX_OFFLOAD_CHECKSUM);
 396         /* All vlan offload masks to apply the current settings */
 397         mask = ETH_VLAN_STRIP_MASK |
 398                 ETH_VLAN_FILTER_MASK |
 399                 ETH_VLAN_EXTEND_MASK;
 400         ret = enicpmd_vlan_offload_set(eth_dev, mask);
 401         if (ret) {
 402                 dev_err(enic, "Failed to configure VLAN offloads\n");
 403                 return ret;
 404         }
 405         /*
 406          * Initialize RSS with the default reta and key. If the user key is
 407          * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
 408          * default key.
 409          */
 410         return enic_init_rss_nic_cfg(enic);
 411 }
 412
 413 /* Start the device.
 414  * It returns 0 on success.
 415  */
 416 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
 417 {
 418         struct enic *enic = pmd_priv(eth_dev);
 419
 420         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 421                 return -E_RTE_SECONDARY;
 422
 423         ENICPMD_FUNC_TRACE();
 424         return enic_enable(enic);
 425 }
 426
 427 /*
 428  * Stop device: disable rx and tx functions to allow for reconfiguring.
 429  */
 430 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
 431 {
 432         struct rte_eth_link link;
 433         struct enic *enic = pmd_priv(eth_dev);
 434
 435         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 436                 return;
 437
 438         ENICPMD_FUNC_TRACE();
 439         enic_disable(enic);
 440
 441         memset(&link, 0, sizeof(link));
 442         rte_eth_linkstatus_set(eth_dev, &link);
 443 }
 444
 445 /*
 446  * Stop device.
 447  */
 448 static void enicpmd_dev_close(struct rte_eth_dev *eth_dev)
 449 {
 450         struct enic *enic = pmd_priv(eth_dev);
 451
 452         ENICPMD_FUNC_TRACE();
 453         enic_remove(enic);
 454 }
 455
 456 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
 457         __rte_unused int wait_to_complete)
 458 {
 459         ENICPMD_FUNC_TRACE();
 460         return enic_link_update(eth_dev);
 461 }
 462
 463 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
 464         struct rte_eth_stats *stats)
 465 {
 466         struct enic *enic = pmd_priv(eth_dev);
 467
 468         ENICPMD_FUNC_TRACE();
 469         return enic_dev_stats_get(enic, stats);
 470 }
 471
 472 static int enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
 473 {
 474         struct enic *enic = pmd_priv(eth_dev);
 475
 476         ENICPMD_FUNC_TRACE();
 477         return enic_dev_stats_clear(enic);
 478 }
 479
 480 static uint32_t speed_capa_from_pci_id(struct rte_eth_dev *eth_dev)
 481 {
 482         const struct vic_speed_capa *m;
 483         struct rte_pci_device *pdev;
 484         uint16_t id;
 485
 486         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
 487         id = pdev->id.subsystem_device_id;
 488         for (m = vic_speed_capa_map; m->sub_devid != 0; m++) {
 489                 if (m->sub_devid == id)
 490                         return m->capa;
 491         }
 492         /* 1300 and later models are at least 40G */
 493         if (id >= 0x0100)
 494                 return ETH_LINK_SPEED_40G;
 495         /* VFs have subsystem id 0, check device id */
 496         if (id == 0) {
 497                 /* Newer VF implies at least 40G model */
 498                 if (pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_SN)
 499                         return ETH_LINK_SPEED_40G;
 500         }
 501         return ETH_LINK_SPEED_10G;
 502 }
 503
 504 static int enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
 505         struct rte_eth_dev_info *device_info)
 506 {
 507         struct enic *enic = pmd_priv(eth_dev);
 508
 509         ENICPMD_FUNC_TRACE();
 510         /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
 511         device_info->max_rx_queues = enic->conf_rq_count / 2;
 512         device_info->max_tx_queues = enic->conf_wq_count;
 513         device_info->min_rx_bufsize = ENIC_MIN_MTU;
 514         /* "Max" mtu is not a typo. HW receives packet sizes up to the
 515          * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
 516          * a hint to the driver to size receive buffers accordingly so that
 517          * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
 518          * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
 519          * ignoring vNIC mtu.
 520          */
 521         device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
 522         device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
 523         device_info->min_mtu = ENIC_MIN_MTU;
 524         device_info->max_mtu = enic->max_mtu;
 525         device_info->rx_offload_capa = enic->rx_offload_capa;
 526         device_info->tx_offload_capa = enic->tx_offload_capa;
 527         device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
 528         device_info->default_rxconf = (struct rte_eth_rxconf) {
 529                 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
 530         };
 531         device_info->reta_size = enic->reta_size;
 532         device_info->hash_key_size = enic->hash_key_size;
 533         device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
 534         device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
 535                 .nb_max = enic->config.rq_desc_count,
 536                 .nb_min = ENIC_MIN_RQ_DESCS,
 537                 .nb_align = ENIC_ALIGN_DESCS,
 538         };
 539         device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
 540                 .nb_max = enic->config.wq_desc_count,
 541                 .nb_min = ENIC_MIN_WQ_DESCS,
 542                 .nb_align = ENIC_ALIGN_DESCS,
 543                 .nb_seg_max = ENIC_TX_XMIT_MAX,
 544                 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
 545         };
 546         device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
 547                 .burst_size = ENIC_DEFAULT_RX_BURST,
 548                 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
 549                         ENIC_DEFAULT_RX_RING_SIZE),
 550                 .nb_queues = ENIC_DEFAULT_RX_RINGS,
 551         };
 552         device_info->default_txportconf = (struct rte_eth_dev_portconf) {
 553                 .burst_size = ENIC_DEFAULT_TX_BURST,
 554                 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
 555                         ENIC_DEFAULT_TX_RING_SIZE),
 556                 .nb_queues = ENIC_DEFAULT_TX_RINGS,
 557         };
 558         device_info->speed_capa = speed_capa_from_pci_id(eth_dev);
 559
 560         return 0;
 561 }
 562
 563 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 564 {
 565         static const uint32_t ptypes[] = {
 566                 RTE_PTYPE_L2_ETHER,
 567                 RTE_PTYPE_L2_ETHER_VLAN,
 568                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 569                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 570                 RTE_PTYPE_L4_TCP,
 571                 RTE_PTYPE_L4_UDP,
 572                 RTE_PTYPE_L4_FRAG,
 573                 RTE_PTYPE_L4_NONFRAG,
 574                 RTE_PTYPE_UNKNOWN
 575         };
 576         static const uint32_t ptypes_overlay[] = {
 577                 RTE_PTYPE_L2_ETHER,
 578                 RTE_PTYPE_L2_ETHER_VLAN,
 579                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 580                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 581                 RTE_PTYPE_L4_TCP,
 582                 RTE_PTYPE_L4_UDP,
 583                 RTE_PTYPE_L4_FRAG,
 584                 RTE_PTYPE_L4_NONFRAG,
 585                 RTE_PTYPE_TUNNEL_GRENAT,
 586                 RTE_PTYPE_INNER_L2_ETHER,
 587                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 588                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 589                 RTE_PTYPE_INNER_L4_TCP,
 590                 RTE_PTYPE_INNER_L4_UDP,
 591                 RTE_PTYPE_INNER_L4_FRAG,
 592                 RTE_PTYPE_INNER_L4_NONFRAG,
 593                 RTE_PTYPE_UNKNOWN
 594         };
 595
 596         if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
 597             dev->rx_pkt_burst != NULL) {
 598                 struct enic *enic = pmd_priv(dev);
 599                 if (enic->overlay_offload)
 600                         return ptypes_overlay;
 601                 else
 602                         return ptypes;
 603         }
 604         return NULL;
 605 }
 606
 607 static int enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 608 {
 609         struct enic *enic = pmd_priv(eth_dev);
 610         int ret;
 611
 612         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 613                 return -E_RTE_SECONDARY;
 614
 615         ENICPMD_FUNC_TRACE();
 616
 617         enic->promisc = 1;
 618         ret = enic_add_packet_filter(enic);
 619         if (ret != 0)
 620                 enic->promisc = 0;
 621
 622         return ret;
 623 }
 624
 625 static int enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 626 {
 627         struct enic *enic = pmd_priv(eth_dev);
 628         int ret;
 629
 630         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 631                 return -E_RTE_SECONDARY;
 632
 633         ENICPMD_FUNC_TRACE();
 634         enic->promisc = 0;
 635         ret = enic_add_packet_filter(enic);
 636         if (ret != 0)
 637                 enic->promisc = 1;
 638
 639         return ret;
 640 }
 641
 642 static int enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 643 {
 644         struct enic *enic = pmd_priv(eth_dev);
 645         int ret;
 646
 647         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 648                 return -E_RTE_SECONDARY;
 649
 650         ENICPMD_FUNC_TRACE();
 651         enic->allmulti = 1;
 652         ret = enic_add_packet_filter(enic);
 653         if (ret != 0)
 654                 enic->allmulti = 0;
 655
 656         return ret;
 657 }
 658
 659 static int enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 660 {
 661         struct enic *enic = pmd_priv(eth_dev);
 662         int ret;
 663
 664         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 665                 return -E_RTE_SECONDARY;
 666
 667         ENICPMD_FUNC_TRACE();
 668         enic->allmulti = 0;
 669         ret = enic_add_packet_filter(enic);
 670         if (ret != 0)
 671                 enic->allmulti = 1;
 672
 673         return ret;
 674 }
 675
 676 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
 677         struct rte_ether_addr *mac_addr,
 678         __rte_unused uint32_t index, __rte_unused uint32_t pool)
 679 {
 680         struct enic *enic = pmd_priv(eth_dev);
 681
 682         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 683                 return -E_RTE_SECONDARY;
 684
 685         ENICPMD_FUNC_TRACE();
 686         return enic_set_mac_address(enic, mac_addr->addr_bytes);
 687 }
 688
 689 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
 690 {
 691         struct enic *enic = pmd_priv(eth_dev);
 692
 693         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 694                 return;
 695
 696         ENICPMD_FUNC_TRACE();
 697         if (enic_del_mac_address(enic, index))
 698                 dev_err(enic, "del mac addr failed\n");
 699 }
 700
 701 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
 702                                 struct rte_ether_addr *addr)
 703 {
 704         struct enic *enic = pmd_priv(eth_dev);
 705         int ret;
 706
 707         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 708                 return -E_RTE_SECONDARY;
 709
 710         ENICPMD_FUNC_TRACE();
 711         ret = enic_del_mac_address(enic, 0);
 712         if (ret)
 713                 return ret;
 714         return enic_set_mac_address(enic, addr->addr_bytes);
 715 }
 716
 717 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
 718 {
 719         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 720
 721         rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
 722         ENICPMD_LOG(DEBUG, " %s address %s\n",
 723                      add ? "add" : "remove", mac_str);
 724 }
 725
 726 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
 727                                     struct rte_ether_addr *mc_addr_set,
 728                                     uint32_t nb_mc_addr)
 729 {
 730         struct enic *enic = pmd_priv(eth_dev);
 731         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 732         struct rte_ether_addr *addr;
 733         uint32_t i, j;
 734         int ret;
 735
 736         ENICPMD_FUNC_TRACE();
 737
 738         /* Validate the given addresses first */
 739         for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
 740                 addr = &mc_addr_set[i];
 741                 if (!rte_is_multicast_ether_addr(addr) ||
 742                     rte_is_broadcast_ether_addr(addr)) {
 743                         rte_ether_format_addr(mac_str,
 744                                         RTE_ETHER_ADDR_FMT_SIZE, addr);
 745                         ENICPMD_LOG(ERR, " invalid multicast address %s\n",
 746                                      mac_str);
 747                         return -EINVAL;
 748                 }
 749         }
 750
 751         /* Flush all if requested */
 752         if (nb_mc_addr == 0 || mc_addr_set == NULL) {
 753                 ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
 754                 for (i = 0; i < enic->mc_count; i++) {
 755                         addr = &enic->mc_addrs[i];
 756                         debug_log_add_del_addr(addr, false);
 757                         ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 758                         if (ret)
 759                                 return ret;
 760                 }
 761                 enic->mc_count = 0;
 762                 return 0;
 763         }
 764
 765         if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
 766                 ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
 767                              ENIC_MULTICAST_PERFECT_FILTERS);
 768                 return -ENOSPC;
 769         }
 770         /*
 771          * devcmd is slow, so apply the difference instead of flushing and
 772          * adding everything.
 773          * 1. Delete addresses on the NIC but not on the host
 774          */
 775         for (i = 0; i < enic->mc_count; i++) {
 776                 addr = &enic->mc_addrs[i];
 777                 for (j = 0; j < nb_mc_addr; j++) {
 778                         if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
 779                                 break;
 780                 }
 781                 if (j < nb_mc_addr)
 782                         continue;
 783                 debug_log_add_del_addr(addr, false);
 784                 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 785                 if (ret)
 786                         return ret;
 787         }
 788         /* 2. Add addresses on the host but not on the NIC */
 789         for (i = 0; i < nb_mc_addr; i++) {
 790                 addr = &mc_addr_set[i];
 791                 for (j = 0; j < enic->mc_count; j++) {
 792                         if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
 793                                 break;
 794                 }
 795                 if (j < enic->mc_count)
 796                         continue;
 797                 debug_log_add_del_addr(addr, true);
 798                 ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
 799                 if (ret)
 800                         return ret;
 801         }
 802         /* Keep a copy so we can flush/apply later on.. */
 803         memcpy(enic->mc_addrs, mc_addr_set,
 804                nb_mc_addr * sizeof(struct rte_ether_addr));
 805         enic->mc_count = nb_mc_addr;
 806         return 0;
 807 }
 808
 809 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 810 {
 811         struct enic *enic = pmd_priv(eth_dev);
 812
 813         ENICPMD_FUNC_TRACE();
 814         return enic_set_mtu(enic, mtu);
 815 }
 816
 817 static int enicpmd_dev_rss_reta_query(struct rte_eth_dev *dev,
 818                                       struct rte_eth_rss_reta_entry64
 819                                       *reta_conf,
 820                                       uint16_t reta_size)
 821 {
 822         struct enic *enic = pmd_priv(dev);
 823         uint16_t i, idx, shift;
 824
 825         ENICPMD_FUNC_TRACE();
 826         if (reta_size != ENIC_RSS_RETA_SIZE) {
 827                 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
 828                         reta_size, ENIC_RSS_RETA_SIZE);
 829                 return -EINVAL;
 830         }
 831
 832         for (i = 0; i < reta_size; i++) {
 833                 idx = i / RTE_RETA_GROUP_SIZE;
 834                 shift = i % RTE_RETA_GROUP_SIZE;
 835                 if (reta_conf[idx].mask & (1ULL << shift))
 836                         reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
 837                                 enic->rss_cpu.cpu[i / 4].b[i % 4]);
 838         }
 839
 840         return 0;
 841 }
 842
 843 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
 844                                        struct rte_eth_rss_reta_entry64
 845                                        *reta_conf,
 846                                        uint16_t reta_size)
 847 {
 848         struct enic *enic = pmd_priv(dev);
 849         union vnic_rss_cpu rss_cpu;
 850         uint16_t i, idx, shift;
 851
 852         ENICPMD_FUNC_TRACE();
 853         if (reta_size != ENIC_RSS_RETA_SIZE) {
 854                 dev_err(enic, "reta_update: wrong reta_size. given=%u"
 855                         " expected=%u\n",
 856                         reta_size, ENIC_RSS_RETA_SIZE);
 857                 return -EINVAL;
 858         }
 859         /*
 860          * Start with the current reta and modify it per reta_conf, as we
 861          * need to push the entire reta even if we only modify one entry.
 862          */
 863         rss_cpu = enic->rss_cpu;
 864         for (i = 0; i < reta_size; i++) {
 865                 idx = i / RTE_RETA_GROUP_SIZE;
 866                 shift = i % RTE_RETA_GROUP_SIZE;
 867                 if (reta_conf[idx].mask & (1ULL << shift))
 868                         rss_cpu.cpu[i / 4].b[i % 4] =
 869                                 enic_rte_rq_idx_to_sop_idx(
 870                                         reta_conf[idx].reta[shift]);
 871         }
 872         return enic_set_rss_reta(enic, &rss_cpu);
 873 }
 874
 875 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
 876                                        struct rte_eth_rss_conf *rss_conf)
 877 {
 878         struct enic *enic = pmd_priv(dev);
 879
 880         ENICPMD_FUNC_TRACE();
 881         return enic_set_rss_conf(enic, rss_conf);
 882 }
 883
 884 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 885                                          struct rte_eth_rss_conf *rss_conf)
 886 {
 887         struct enic *enic = pmd_priv(dev);
 888
 889         ENICPMD_FUNC_TRACE();
 890         if (rss_conf == NULL)
 891                 return -EINVAL;
 892         if (rss_conf->rss_key != NULL &&
 893             rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
 894                 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
 895                         " expected=%u+\n",
 896                         rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
 897                 return -EINVAL;
 898         }
 899         rss_conf->rss_hf = enic->rss_hf;
 900         if (rss_conf->rss_key != NULL) {
 901                 int i;
 902                 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
 903                         rss_conf->rss_key[i] =
 904                                 enic->rss_key.key[i / 10].b[i % 10];
 905                 }
 906                 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
 907         }
 908         return 0;
 909 }
 910
 911 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
 912                                      uint16_t rx_queue_id,
 913                                      struct rte_eth_rxq_info *qinfo)
 914 {
 915         struct enic *enic = pmd_priv(dev);
 916         struct vnic_rq *rq_sop;
 917         struct vnic_rq *rq_data;
 918         struct rte_eth_rxconf *conf;
 919         uint16_t sop_queue_idx;
 920         uint16_t data_queue_idx;
 921
 922         ENICPMD_FUNC_TRACE();
 923         sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 924         data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id, enic);
 925         rq_sop = &enic->rq[sop_queue_idx];
 926         rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
 927         qinfo->mp = rq_sop->mp;
 928         qinfo->scattered_rx = rq_sop->data_queue_enable;
 929         qinfo->nb_desc = rq_sop->ring.desc_count;
 930         if (qinfo->scattered_rx)
 931                 qinfo->nb_desc += rq_data->ring.desc_count;
 932         conf = &qinfo->conf;
 933         memset(conf, 0, sizeof(*conf));
 934         conf->rx_free_thresh = rq_sop->rx_free_thresh;
 935         conf->rx_drop_en = 1;
 936         /*
 937          * Except VLAN stripping (port setting), all the checksum offloads
 938          * are always enabled.
 939          */
 940         conf->offloads = enic->rx_offload_capa;
 941         if (!enic->ig_vlan_strip_en)
 942                 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
 943         /* rx_thresh and other fields are not applicable for enic */
 944 }
 945
 946 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
 947                                      uint16_t tx_queue_id,
 948                                      struct rte_eth_txq_info *qinfo)
 949 {
 950         struct enic *enic = pmd_priv(dev);
 951         struct vnic_wq *wq = &enic->wq[tx_queue_id];
 952
 953         ENICPMD_FUNC_TRACE();
 954         qinfo->nb_desc = wq->ring.desc_count;
 955         memset(&qinfo->conf, 0, sizeof(qinfo->conf));
 956         qinfo->conf.offloads = wq->offloads;
 957         /* tx_thresh, and all the other fields are not applicable for enic */
 958 }
 959
 960 static int enicpmd_dev_rx_burst_mode_get(struct rte_eth_dev *dev,
 961                                          __rte_unused uint16_t queue_id,
 962                                          struct rte_eth_burst_mode *mode)
 963 {
 964         eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
 965         struct enic *enic = pmd_priv(dev);
 966         const char *info_str = NULL;
 967         int ret = -EINVAL;
 968
 969         ENICPMD_FUNC_TRACE();
 970         if (enic->use_noscatter_vec_rx_handler)
 971                 info_str = "Vector AVX2 No Scatter";
 972         else if (pkt_burst == enic_noscatter_recv_pkts)
 973                 info_str = "Scalar No Scatter";
 974         else if (pkt_burst == enic_recv_pkts)
 975                 info_str = "Scalar";
 976         if (info_str) {
 977                 strlcpy(mode->info, info_str, sizeof(mode->info));
 978                 ret = 0;
 979         }
 980         return ret;
 981 }
 982
 983 static int enicpmd_dev_tx_burst_mode_get(struct rte_eth_dev *dev,
 984                                          __rte_unused uint16_t queue_id,
 985                                          struct rte_eth_burst_mode *mode)
 986 {
 987         eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
 988         const char *info_str = NULL;
 989         int ret = -EINVAL;
 990
 991         ENICPMD_FUNC_TRACE();
 992         if (pkt_burst == enic_simple_xmit_pkts)
 993                 info_str = "Scalar Simplified";
 994         else if (pkt_burst == enic_xmit_pkts)
 995                 info_str = "Scalar";
 996         if (info_str) {
 997                 strlcpy(mode->info, info_str, sizeof(mode->info));
 998                 ret = 0;
 999         }
1000         return ret;
1001 }
1002
1003 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
1004                                             uint16_t rx_queue_id)
1005 {
1006         struct enic *enic = pmd_priv(eth_dev);
1007
1008         ENICPMD_FUNC_TRACE();
1009         vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
1010         return 0;
1011 }
1012
1013 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
1014                                              uint16_t rx_queue_id)
1015 {
1016         struct enic *enic = pmd_priv(eth_dev);
1017
1018         ENICPMD_FUNC_TRACE();
1019         vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
1020         return 0;
1021 }
1022
1023 static int udp_tunnel_common_check(struct enic *enic,
1024                                    struct rte_eth_udp_tunnel *tnl)
1025 {
1026         if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
1027                 return -ENOTSUP;
1028         if (!enic->overlay_offload) {
1029                 ENICPMD_LOG(DEBUG, " vxlan (overlay offload) is not "
1030                              "supported\n");
1031                 return -ENOTSUP;
1032         }
1033         return 0;
1034 }
1035
1036 static int update_vxlan_port(struct enic *enic, uint16_t port)
1037 {
1038         if (vnic_dev_overlay_offload_cfg(enic->vdev,
1039                                          OVERLAY_CFG_VXLAN_PORT_UPDATE,
1040                                          port)) {
1041                 ENICPMD_LOG(DEBUG, " failed to update vxlan port\n");
1042                 return -EINVAL;
1043         }
1044         ENICPMD_LOG(DEBUG, " updated vxlan port to %u\n", port);
1045         enic->vxlan_port = port;
1046         return 0;
1047 }
1048
1049 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
1050                                            struct rte_eth_udp_tunnel *tnl)
1051 {
1052         struct enic *enic = pmd_priv(eth_dev);
1053         int ret;
1054
1055         ENICPMD_FUNC_TRACE();
1056         ret = udp_tunnel_common_check(enic, tnl);
1057         if (ret)
1058                 return ret;
1059         /*
1060          * The NIC has 1 configurable VXLAN port number. "Adding" a new port
1061          * number replaces it.
1062          */
1063         if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
1064                 ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
1065                              tnl->udp_port);
1066                 return -EINVAL;
1067         }
1068         return update_vxlan_port(enic, tnl->udp_port);
1069 }
1070
1071 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
1072                                            struct rte_eth_udp_tunnel *tnl)
1073 {
1074         struct enic *enic = pmd_priv(eth_dev);
1075         int ret;
1076
1077         ENICPMD_FUNC_TRACE();
1078         ret = udp_tunnel_common_check(enic, tnl);
1079         if (ret)
1080                 return ret;
1081         /*
1082          * Clear the previously set port number and restore the
1083          * hardware default port number. Some drivers disable VXLAN
1084          * offloads when there are no configured port numbers. But
1085          * enic does not do that as VXLAN is part of overlay offload,
1086          * which is tied to inner RSS and TSO.
1087          */
1088         if (tnl->udp_port != enic->vxlan_port) {
1089                 ENICPMD_LOG(DEBUG, " %u is not a configured vxlan port\n",
1090                              tnl->udp_port);
1091                 return -EINVAL;
1092         }
1093         return update_vxlan_port(enic, RTE_VXLAN_DEFAULT_PORT);
1094 }
1095
1096 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1097                                       char *fw_version, size_t fw_size)
1098 {
1099         struct vnic_devcmd_fw_info *info;
1100         struct enic *enic;
1101         int ret;
1102
1103         ENICPMD_FUNC_TRACE();
1104         if (fw_version == NULL || fw_size <= 0)
1105                 return -EINVAL;
1106         enic = pmd_priv(eth_dev);
1107         ret = vnic_dev_fw_info(enic->vdev, &info);
1108         if (ret)
1109                 return ret;
1110         snprintf(fw_version, fw_size, "%s %s",
1111                  info->fw_version, info->fw_build);
1112         fw_version[fw_size - 1] = '\0';
1113         return 0;
1114 }
1115
1116 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1117         .dev_configure        = enicpmd_dev_configure,
1118         .dev_start            = enicpmd_dev_start,
1119         .dev_stop             = enicpmd_dev_stop,
1120         .dev_set_link_up      = NULL,
1121         .dev_set_link_down    = NULL,
1122         .dev_close            = enicpmd_dev_close,
1123         .promiscuous_enable   = enicpmd_dev_promiscuous_enable,
1124         .promiscuous_disable  = enicpmd_dev_promiscuous_disable,
1125         .allmulticast_enable  = enicpmd_dev_allmulticast_enable,
1126         .allmulticast_disable = enicpmd_dev_allmulticast_disable,
1127         .link_update          = enicpmd_dev_link_update,
1128         .stats_get            = enicpmd_dev_stats_get,
1129         .stats_reset          = enicpmd_dev_stats_reset,
1130         .queue_stats_mapping_set = NULL,
1131         .dev_infos_get        = enicpmd_dev_info_get,
1132         .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1133         .mtu_set              = enicpmd_mtu_set,
1134         .vlan_filter_set      = NULL,
1135         .vlan_tpid_set        = NULL,
1136         .vlan_offload_set     = enicpmd_vlan_offload_set,
1137         .vlan_strip_queue_set = NULL,
1138         .rx_queue_start       = enicpmd_dev_rx_queue_start,
1139         .rx_queue_stop        = enicpmd_dev_rx_queue_stop,
1140         .tx_queue_start       = enicpmd_dev_tx_queue_start,
1141         .tx_queue_stop        = enicpmd_dev_tx_queue_stop,
1142         .rx_queue_setup       = enicpmd_dev_rx_queue_setup,
1143         .rx_queue_release     = enicpmd_dev_rx_queue_release,
1144         .rx_queue_count       = enicpmd_dev_rx_queue_count,
1145         .rx_descriptor_done   = NULL,
1146         .tx_queue_setup       = enicpmd_dev_tx_queue_setup,
1147         .tx_queue_release     = enicpmd_dev_tx_queue_release,
1148         .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1149         .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1150         .rxq_info_get         = enicpmd_dev_rxq_info_get,
1151         .txq_info_get         = enicpmd_dev_txq_info_get,
1152         .rx_burst_mode_get    = enicpmd_dev_rx_burst_mode_get,
1153         .tx_burst_mode_get    = enicpmd_dev_tx_burst_mode_get,
1154         .dev_led_on           = NULL,
1155         .dev_led_off          = NULL,
1156         .flow_ctrl_get        = NULL,
1157         .flow_ctrl_set        = NULL,
1158         .priority_flow_ctrl_set = NULL,
1159         .mac_addr_add         = enicpmd_add_mac_addr,
1160         .mac_addr_remove      = enicpmd_remove_mac_addr,
1161         .mac_addr_set         = enicpmd_set_mac_addr,
1162         .set_mc_addr_list     = enicpmd_set_mc_addr_list,
1163         .filter_ctrl          = enicpmd_dev_filter_ctrl,
1164         .reta_query           = enicpmd_dev_rss_reta_query,
1165         .reta_update          = enicpmd_dev_rss_reta_update,
1166         .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
1167         .rss_hash_update      = enicpmd_dev_rss_hash_update,
1168         .udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
1169         .udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
1170         .fw_version_get       = enicpmd_dev_fw_version_get,
1171 };
1172
1173 static int enic_parse_zero_one(const char *key,
1174                                const char *value,
1175                                void *opaque)
1176 {
1177         struct enic *enic;
1178         bool b;
1179
1180         enic = (struct enic *)opaque;
1181         if (strcmp(value, "0") == 0) {
1182                 b = false;
1183         } else if (strcmp(value, "1") == 0) {
1184                 b = true;
1185         } else {
1186                 dev_err(enic, "Invalid value for %s"
1187                         ": expected=0|1 given=%s\n", key, value);
1188                 return -EINVAL;
1189         }
1190         if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1191                 enic->disable_overlay = b;
1192         if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1193                 enic->enable_avx2_rx = b;
1194         if (strcmp(key, ENIC_DEVARG_GENEVE_OPT) == 0)
1195                 enic->geneve_opt_request = b;
1196         return 0;
1197 }
1198
1199 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1200                                       const char *value,
1201                                       void *opaque)
1202 {
1203         struct enic *enic;
1204
1205         enic = (struct enic *)opaque;
1206         if (strcmp(value, "trunk") == 0) {
1207                 /* Trunk mode: always tag */
1208                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1209         } else if (strcmp(value, "untag") == 0) {
1210                 /* Untag default VLAN mode: untag if VLAN = default VLAN */
1211                 enic->ig_vlan_rewrite_mode =
1212                         IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1213         } else if (strcmp(value, "priority") == 0) {
1214                 /*
1215                  * Priority-tag default VLAN mode: priority tag (VLAN header
1216                  * with ID=0) if VLAN = default
1217                  */
1218                 enic->ig_vlan_rewrite_mode =
1219                         IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1220         } else if (strcmp(value, "pass") == 0) {
1221                 /* Pass through mode: do not touch tags */
1222                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1223         } else {
1224                 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1225                         ": expected=trunk|untag|priority|pass given=%s\n",
1226                         value);
1227                 return -EINVAL;
1228         }
1229         return 0;
1230 }
1231
1232 static int enic_check_devargs(struct rte_eth_dev *dev)
1233 {
1234         static const char *const valid_keys[] = {
1235                 ENIC_DEVARG_DISABLE_OVERLAY,
1236                 ENIC_DEVARG_ENABLE_AVX2_RX,
1237                 ENIC_DEVARG_GENEVE_OPT,
1238                 ENIC_DEVARG_IG_VLAN_REWRITE,
1239                 NULL};
1240         struct enic *enic = pmd_priv(dev);
1241         struct rte_kvargs *kvlist;
1242
1243         ENICPMD_FUNC_TRACE();
1244
1245         enic->disable_overlay = false;
1246         enic->enable_avx2_rx = false;
1247         enic->geneve_opt_request = false;
1248         enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1249         if (!dev->device->devargs)
1250                 return 0;
1251         kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1252         if (!kvlist)
1253                 return -EINVAL;
1254         if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1255                                enic_parse_zero_one, enic) < 0 ||
1256             rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1257                                enic_parse_zero_one, enic) < 0 ||
1258             rte_kvargs_process(kvlist, ENIC_DEVARG_GENEVE_OPT,
1259                                enic_parse_zero_one, enic) < 0 ||
1260             rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1261                                enic_parse_ig_vlan_rewrite, enic) < 0) {
1262                 rte_kvargs_free(kvlist);
1263                 return -EINVAL;
1264         }
1265         rte_kvargs_free(kvlist);
1266         return 0;
1267 }
1268
1269 /* Initialize the driver
1270  * It returns 0 on success.
1271  */
1272 static int eth_enicpmd_dev_init(struct rte_eth_dev *eth_dev)
1273 {
1274         struct rte_pci_device *pdev;
1275         struct rte_pci_addr *addr;
1276         struct enic *enic = pmd_priv(eth_dev);
1277         int err;
1278
1279         ENICPMD_FUNC_TRACE();
1280
1281         eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1282         eth_dev->rx_pkt_burst = &enic_recv_pkts;
1283         eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1284         eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1285         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1286                 enic_pick_tx_handler(eth_dev);
1287                 enic_pick_rx_handler(eth_dev);
1288                 return 0;
1289         }
1290         /* Only the primary sets up adapter and other data in shared memory */
1291         enic->port_id = eth_dev->data->port_id;
1292         enic->rte_dev = eth_dev;
1293         enic->dev_data = eth_dev->data;
1294         /* Let rte_eth_dev_close() release the port resources */
1295         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1296
1297         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1298         rte_eth_copy_pci_info(eth_dev, pdev);
1299         enic->pdev = pdev;
1300         addr = &pdev->addr;
1301
1302         snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1303                 addr->domain, addr->bus, addr->devid, addr->function);
1304
1305         err = enic_check_devargs(eth_dev);
1306         if (err)
1307                 return err;
1308         return enic_probe(enic);
1309 }
1310
1311 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1312         struct rte_pci_device *pci_dev)
1313 {
1314         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct enic),
1315                 eth_enicpmd_dev_init);
1316 }
1317
1318 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1319 {
1320         return rte_eth_dev_pci_generic_remove(pci_dev, NULL);
1321 }
1322
1323 static struct rte_pci_driver rte_enic_pmd = {
1324         .id_table = pci_id_enic_map,
1325         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1326         .probe = eth_enic_pci_probe,
1327         .remove = eth_enic_pci_remove,
1328 };
1329
1330 int dev_is_enic(struct rte_eth_dev *dev)
1331 {
1332         return dev->device->driver == &rte_enic_pmd.driver;
1333 }
1334
1335 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1336 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1337 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1338 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1339         ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1340         ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1341         ENIC_DEVARG_GENEVE_OPT "=0|1 "
1342         ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");