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 #define ENIC_DEVARG_REPRESENTOR "representor"
  72
  73 RTE_LOG_REGISTER(enic_pmd_logtype, pmd.net.enic, INFO);
  74
  75 static int
  76 enicpmd_fdir_ctrl_func(struct rte_eth_dev *eth_dev,
  77                         enum rte_filter_op filter_op, void *arg)
  78 {
  79         struct enic *enic = pmd_priv(eth_dev);
  80         int ret = 0;
  81
  82         ENICPMD_FUNC_TRACE();
  83         if (filter_op == RTE_ETH_FILTER_NOP)
  84                 return 0;
  85
  86         if (arg == NULL && filter_op != RTE_ETH_FILTER_FLUSH)
  87                 return -EINVAL;
  88
  89         switch (filter_op) {
  90         case RTE_ETH_FILTER_ADD:
  91         case RTE_ETH_FILTER_UPDATE:
  92                 ret = enic_fdir_add_fltr(enic,
  93                         (struct rte_eth_fdir_filter *)arg);
  94                 break;
  95
  96         case RTE_ETH_FILTER_DELETE:
  97                 ret = enic_fdir_del_fltr(enic,
  98                         (struct rte_eth_fdir_filter *)arg);
  99                 break;
 100
 101         case RTE_ETH_FILTER_STATS:
 102                 enic_fdir_stats_get(enic, (struct rte_eth_fdir_stats *)arg);
 103                 break;
 104
 105         case RTE_ETH_FILTER_FLUSH:
 106                 dev_warning(enic, "unsupported operation %u", filter_op);
 107                 ret = -ENOTSUP;
 108                 break;
 109         case RTE_ETH_FILTER_INFO:
 110                 enic_fdir_info_get(enic, (struct rte_eth_fdir_info *)arg);
 111                 break;
 112         default:
 113                 dev_err(enic, "unknown operation %u", filter_op);
 114                 ret = -EINVAL;
 115                 break;
 116         }
 117         return ret;
 118 }
 119
 120 static int
 121 enicpmd_dev_filter_ctrl(struct rte_eth_dev *dev,
 122                      enum rte_filter_type filter_type,
 123                      enum rte_filter_op filter_op,
 124                      void *arg)
 125 {
 126         struct enic *enic = pmd_priv(dev);
 127         int ret = 0;
 128
 129         ENICPMD_FUNC_TRACE();
 130
 131         /*
 132          * Currently, when Geneve with options offload is enabled, host
 133          * cannot insert match-action rules.
 134          */
 135         if (enic->geneve_opt_enabled)
 136                 return -ENOTSUP;
 137         switch (filter_type) {
 138         case RTE_ETH_FILTER_GENERIC:
 139                 if (filter_op != RTE_ETH_FILTER_GET)
 140                         return -EINVAL;
 141                 if (enic->flow_filter_mode == FILTER_FLOWMAN)
 142                         *(const void **)arg = &enic_fm_flow_ops;
 143                 else
 144                         *(const void **)arg = &enic_flow_ops;
 145                 break;
 146         case RTE_ETH_FILTER_FDIR:
 147                 ret = enicpmd_fdir_ctrl_func(dev, filter_op, arg);
 148                 break;
 149         default:
 150                 dev_warning(enic, "Filter type (%d) not supported",
 151                         filter_type);
 152                 ret = -EINVAL;
 153                 break;
 154         }
 155
 156         return ret;
 157 }
 158
 159 static void enicpmd_dev_tx_queue_release(void *txq)
 160 {
 161         ENICPMD_FUNC_TRACE();
 162
 163         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 164                 return;
 165
 166         enic_free_wq(txq);
 167 }
 168
 169 static int enicpmd_dev_setup_intr(struct enic *enic)
 170 {
 171         int ret;
 172         unsigned int index;
 173
 174         ENICPMD_FUNC_TRACE();
 175
 176         /* Are we done with the init of all the queues? */
 177         for (index = 0; index < enic->cq_count; index++) {
 178                 if (!enic->cq[index].ctrl)
 179                         break;
 180         }
 181         if (enic->cq_count != index)
 182                 return 0;
 183         for (index = 0; index < enic->wq_count; index++) {
 184                 if (!enic->wq[index].ctrl)
 185                         break;
 186         }
 187         if (enic->wq_count != index)
 188                 return 0;
 189         /* check start of packet (SOP) RQs only in case scatter is disabled. */
 190         for (index = 0; index < enic->rq_count; index++) {
 191                 if (!enic->rq[enic_rte_rq_idx_to_sop_idx(index)].ctrl)
 192                         break;
 193         }
 194         if (enic->rq_count != index)
 195                 return 0;
 196
 197         ret = enic_alloc_intr_resources(enic);
 198         if (ret) {
 199                 dev_err(enic, "alloc intr failed\n");
 200                 return ret;
 201         }
 202         enic_init_vnic_resources(enic);
 203
 204         ret = enic_setup_finish(enic);
 205         if (ret)
 206                 dev_err(enic, "setup could not be finished\n");
 207
 208         return ret;
 209 }
 210
 211 static int enicpmd_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
 212         uint16_t queue_idx,
 213         uint16_t nb_desc,
 214         unsigned int socket_id,
 215         const struct rte_eth_txconf *tx_conf)
 216 {
 217         int ret;
 218         struct enic *enic = pmd_priv(eth_dev);
 219         struct vnic_wq *wq;
 220
 221         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 222                 return -E_RTE_SECONDARY;
 223
 224         ENICPMD_FUNC_TRACE();
 225         RTE_ASSERT(queue_idx < enic->conf_wq_count);
 226         wq = &enic->wq[queue_idx];
 227         wq->offloads = tx_conf->offloads |
 228                 eth_dev->data->dev_conf.txmode.offloads;
 229         eth_dev->data->tx_queues[queue_idx] = (void *)wq;
 230
 231         ret = enic_alloc_wq(enic, queue_idx, socket_id, nb_desc);
 232         if (ret) {
 233                 dev_err(enic, "error in allocating wq\n");
 234                 return ret;
 235         }
 236
 237         return enicpmd_dev_setup_intr(enic);
 238 }
 239
 240 static int enicpmd_dev_tx_queue_start(struct rte_eth_dev *eth_dev,
 241         uint16_t queue_idx)
 242 {
 243         struct enic *enic = pmd_priv(eth_dev);
 244
 245         ENICPMD_FUNC_TRACE();
 246
 247         enic_start_wq(enic, queue_idx);
 248
 249         return 0;
 250 }
 251
 252 static int enicpmd_dev_tx_queue_stop(struct rte_eth_dev *eth_dev,
 253         uint16_t queue_idx)
 254 {
 255         int ret;
 256         struct enic *enic = pmd_priv(eth_dev);
 257
 258         ENICPMD_FUNC_TRACE();
 259
 260         ret = enic_stop_wq(enic, queue_idx);
 261         if (ret)
 262                 dev_err(enic, "error in stopping wq %d\n", queue_idx);
 263
 264         return ret;
 265 }
 266
 267 static int enicpmd_dev_rx_queue_start(struct rte_eth_dev *eth_dev,
 268         uint16_t queue_idx)
 269 {
 270         struct enic *enic = pmd_priv(eth_dev);
 271
 272         ENICPMD_FUNC_TRACE();
 273
 274         enic_start_rq(enic, queue_idx);
 275
 276         return 0;
 277 }
 278
 279 static int enicpmd_dev_rx_queue_stop(struct rte_eth_dev *eth_dev,
 280         uint16_t queue_idx)
 281 {
 282         int ret;
 283         struct enic *enic = pmd_priv(eth_dev);
 284
 285         ENICPMD_FUNC_TRACE();
 286
 287         ret = enic_stop_rq(enic, queue_idx);
 288         if (ret)
 289                 dev_err(enic, "error in stopping rq %d\n", queue_idx);
 290
 291         return ret;
 292 }
 293
 294 static void enicpmd_dev_rx_queue_release(void *rxq)
 295 {
 296         ENICPMD_FUNC_TRACE();
 297
 298         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 299                 return;
 300
 301         enic_free_rq(rxq);
 302 }
 303
 304 static uint32_t enicpmd_dev_rx_queue_count(struct rte_eth_dev *dev,
 305                                            uint16_t rx_queue_id)
 306 {
 307         struct enic *enic = pmd_priv(dev);
 308         uint32_t queue_count = 0;
 309         struct vnic_cq *cq;
 310         uint32_t cq_tail;
 311         uint16_t cq_idx;
 312         int rq_num;
 313
 314         rq_num = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 315         cq = &enic->cq[enic_cq_rq(enic, rq_num)];
 316         cq_idx = cq->to_clean;
 317
 318         cq_tail = ioread32(&cq->ctrl->cq_tail);
 319
 320         if (cq_tail < cq_idx)
 321                 cq_tail += cq->ring.desc_count;
 322
 323         queue_count = cq_tail - cq_idx;
 324
 325         return queue_count;
 326 }
 327
 328 static int enicpmd_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
 329         uint16_t queue_idx,
 330         uint16_t nb_desc,
 331         unsigned int socket_id,
 332         const struct rte_eth_rxconf *rx_conf,
 333         struct rte_mempool *mp)
 334 {
 335         int ret;
 336         struct enic *enic = pmd_priv(eth_dev);
 337
 338         ENICPMD_FUNC_TRACE();
 339
 340         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 341                 return -E_RTE_SECONDARY;
 342         RTE_ASSERT(enic_rte_rq_idx_to_sop_idx(queue_idx) < enic->conf_rq_count);
 343         eth_dev->data->rx_queues[queue_idx] =
 344                 (void *)&enic->rq[enic_rte_rq_idx_to_sop_idx(queue_idx)];
 345
 346         ret = enic_alloc_rq(enic, queue_idx, socket_id, mp, nb_desc,
 347                             rx_conf->rx_free_thresh);
 348         if (ret) {
 349                 dev_err(enic, "error in allocating rq\n");
 350                 return ret;
 351         }
 352
 353         return enicpmd_dev_setup_intr(enic);
 354 }
 355
 356 static int enicpmd_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
 357 {
 358         struct enic *enic = pmd_priv(eth_dev);
 359         uint64_t offloads;
 360
 361         ENICPMD_FUNC_TRACE();
 362
 363         offloads = eth_dev->data->dev_conf.rxmode.offloads;
 364         if (mask & ETH_VLAN_STRIP_MASK) {
 365                 if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
 366                         enic->ig_vlan_strip_en = 1;
 367                 else
 368                         enic->ig_vlan_strip_en = 0;
 369         }
 370
 371         return enic_set_vlan_strip(enic);
 372 }
 373
 374 static int enicpmd_dev_configure(struct rte_eth_dev *eth_dev)
 375 {
 376         int ret;
 377         int mask;
 378         struct enic *enic = pmd_priv(eth_dev);
 379
 380         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 381                 return -E_RTE_SECONDARY;
 382
 383         ENICPMD_FUNC_TRACE();
 384         ret = enic_set_vnic_res(enic);
 385         if (ret) {
 386                 dev_err(enic, "Set vNIC resource num  failed, aborting\n");
 387                 return ret;
 388         }
 389
 390         if (eth_dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 391                 eth_dev->data->dev_conf.rxmode.offloads |=
 392                         DEV_RX_OFFLOAD_RSS_HASH;
 393
 394         enic->mc_count = 0;
 395         enic->hw_ip_checksum = !!(eth_dev->data->dev_conf.rxmode.offloads &
 396                                   DEV_RX_OFFLOAD_CHECKSUM);
 397         /* All vlan offload masks to apply the current settings */
 398         mask = ETH_VLAN_STRIP_MASK |
 399                 ETH_VLAN_FILTER_MASK |
 400                 ETH_VLAN_EXTEND_MASK;
 401         ret = enicpmd_vlan_offload_set(eth_dev, mask);
 402         if (ret) {
 403                 dev_err(enic, "Failed to configure VLAN offloads\n");
 404                 return ret;
 405         }
 406         /*
 407          * Initialize RSS with the default reta and key. If the user key is
 408          * given (rx_adv_conf.rss_conf.rss_key), will use that instead of the
 409          * default key.
 410          */
 411         return enic_init_rss_nic_cfg(enic);
 412 }
 413
 414 /* Start the device.
 415  * It returns 0 on success.
 416  */
 417 static int enicpmd_dev_start(struct rte_eth_dev *eth_dev)
 418 {
 419         struct enic *enic = pmd_priv(eth_dev);
 420
 421         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 422                 return -E_RTE_SECONDARY;
 423
 424         ENICPMD_FUNC_TRACE();
 425         return enic_enable(enic);
 426 }
 427
 428 /*
 429  * Stop device: disable rx and tx functions to allow for reconfiguring.
 430  */
 431 static void enicpmd_dev_stop(struct rte_eth_dev *eth_dev)
 432 {
 433         struct rte_eth_link link;
 434         struct enic *enic = pmd_priv(eth_dev);
 435
 436         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 437                 return;
 438
 439         ENICPMD_FUNC_TRACE();
 440         enic_disable(enic);
 441
 442         memset(&link, 0, sizeof(link));
 443         rte_eth_linkstatus_set(eth_dev, &link);
 444 }
 445
 446 /*
 447  * Stop device.
 448  */
 449 static int enicpmd_dev_close(struct rte_eth_dev *eth_dev)
 450 {
 451         struct enic *enic = pmd_priv(eth_dev);
 452
 453         ENICPMD_FUNC_TRACE();
 454         enic_remove(enic);
 455
 456         return 0;
 457 }
 458
 459 static int enicpmd_dev_link_update(struct rte_eth_dev *eth_dev,
 460         __rte_unused int wait_to_complete)
 461 {
 462         ENICPMD_FUNC_TRACE();
 463         return enic_link_update(eth_dev);
 464 }
 465
 466 static int enicpmd_dev_stats_get(struct rte_eth_dev *eth_dev,
 467         struct rte_eth_stats *stats)
 468 {
 469         struct enic *enic = pmd_priv(eth_dev);
 470
 471         ENICPMD_FUNC_TRACE();
 472         return enic_dev_stats_get(enic, stats);
 473 }
 474
 475 static int enicpmd_dev_stats_reset(struct rte_eth_dev *eth_dev)
 476 {
 477         struct enic *enic = pmd_priv(eth_dev);
 478
 479         ENICPMD_FUNC_TRACE();
 480         return enic_dev_stats_clear(enic);
 481 }
 482
 483 static uint32_t speed_capa_from_pci_id(struct rte_eth_dev *eth_dev)
 484 {
 485         const struct vic_speed_capa *m;
 486         struct rte_pci_device *pdev;
 487         uint16_t id;
 488
 489         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
 490         id = pdev->id.subsystem_device_id;
 491         for (m = vic_speed_capa_map; m->sub_devid != 0; m++) {
 492                 if (m->sub_devid == id)
 493                         return m->capa;
 494         }
 495         /* 1300 and later models are at least 40G */
 496         if (id >= 0x0100)
 497                 return ETH_LINK_SPEED_40G;
 498         /* VFs have subsystem id 0, check device id */
 499         if (id == 0) {
 500                 /* Newer VF implies at least 40G model */
 501                 if (pdev->id.device_id == PCI_DEVICE_ID_CISCO_VIC_ENET_SN)
 502                         return ETH_LINK_SPEED_40G;
 503         }
 504         return ETH_LINK_SPEED_10G;
 505 }
 506
 507 static int enicpmd_dev_info_get(struct rte_eth_dev *eth_dev,
 508         struct rte_eth_dev_info *device_info)
 509 {
 510         struct enic *enic = pmd_priv(eth_dev);
 511
 512         ENICPMD_FUNC_TRACE();
 513         /* Scattered Rx uses two receive queues per rx queue exposed to dpdk */
 514         device_info->max_rx_queues = enic->conf_rq_count / 2;
 515         device_info->max_tx_queues = enic->conf_wq_count;
 516         device_info->min_rx_bufsize = ENIC_MIN_MTU;
 517         /* "Max" mtu is not a typo. HW receives packet sizes up to the
 518          * max mtu regardless of the current mtu (vNIC's mtu). vNIC mtu is
 519          * a hint to the driver to size receive buffers accordingly so that
 520          * larger-than-vnic-mtu packets get truncated.. For DPDK, we let
 521          * the user decide the buffer size via rxmode.max_rx_pkt_len, basically
 522          * ignoring vNIC mtu.
 523          */
 524         device_info->max_rx_pktlen = enic_mtu_to_max_rx_pktlen(enic->max_mtu);
 525         device_info->max_mac_addrs = ENIC_UNICAST_PERFECT_FILTERS;
 526         device_info->min_mtu = ENIC_MIN_MTU;
 527         device_info->max_mtu = enic->max_mtu;
 528         device_info->rx_offload_capa = enic->rx_offload_capa;
 529         device_info->tx_offload_capa = enic->tx_offload_capa;
 530         device_info->tx_queue_offload_capa = enic->tx_queue_offload_capa;
 531         device_info->default_rxconf = (struct rte_eth_rxconf) {
 532                 .rx_free_thresh = ENIC_DEFAULT_RX_FREE_THRESH
 533         };
 534         device_info->reta_size = enic->reta_size;
 535         device_info->hash_key_size = enic->hash_key_size;
 536         device_info->flow_type_rss_offloads = enic->flow_type_rss_offloads;
 537         device_info->rx_desc_lim = (struct rte_eth_desc_lim) {
 538                 .nb_max = enic->config.rq_desc_count,
 539                 .nb_min = ENIC_MIN_RQ_DESCS,
 540                 .nb_align = ENIC_ALIGN_DESCS,
 541         };
 542         device_info->tx_desc_lim = (struct rte_eth_desc_lim) {
 543                 .nb_max = enic->config.wq_desc_count,
 544                 .nb_min = ENIC_MIN_WQ_DESCS,
 545                 .nb_align = ENIC_ALIGN_DESCS,
 546                 .nb_seg_max = ENIC_TX_XMIT_MAX,
 547                 .nb_mtu_seg_max = ENIC_NON_TSO_MAX_DESC,
 548         };
 549         device_info->default_rxportconf = (struct rte_eth_dev_portconf) {
 550                 .burst_size = ENIC_DEFAULT_RX_BURST,
 551                 .ring_size = RTE_MIN(device_info->rx_desc_lim.nb_max,
 552                         ENIC_DEFAULT_RX_RING_SIZE),
 553                 .nb_queues = ENIC_DEFAULT_RX_RINGS,
 554         };
 555         device_info->default_txportconf = (struct rte_eth_dev_portconf) {
 556                 .burst_size = ENIC_DEFAULT_TX_BURST,
 557                 .ring_size = RTE_MIN(device_info->tx_desc_lim.nb_max,
 558                         ENIC_DEFAULT_TX_RING_SIZE),
 559                 .nb_queues = ENIC_DEFAULT_TX_RINGS,
 560         };
 561         device_info->speed_capa = speed_capa_from_pci_id(eth_dev);
 562
 563         return 0;
 564 }
 565
 566 static const uint32_t *enicpmd_dev_supported_ptypes_get(struct rte_eth_dev *dev)
 567 {
 568         static const uint32_t ptypes[] = {
 569                 RTE_PTYPE_L2_ETHER,
 570                 RTE_PTYPE_L2_ETHER_VLAN,
 571                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 572                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 573                 RTE_PTYPE_L4_TCP,
 574                 RTE_PTYPE_L4_UDP,
 575                 RTE_PTYPE_L4_FRAG,
 576                 RTE_PTYPE_L4_NONFRAG,
 577                 RTE_PTYPE_UNKNOWN
 578         };
 579         static const uint32_t ptypes_overlay[] = {
 580                 RTE_PTYPE_L2_ETHER,
 581                 RTE_PTYPE_L2_ETHER_VLAN,
 582                 RTE_PTYPE_L3_IPV4_EXT_UNKNOWN,
 583                 RTE_PTYPE_L3_IPV6_EXT_UNKNOWN,
 584                 RTE_PTYPE_L4_TCP,
 585                 RTE_PTYPE_L4_UDP,
 586                 RTE_PTYPE_L4_FRAG,
 587                 RTE_PTYPE_L4_NONFRAG,
 588                 RTE_PTYPE_TUNNEL_GRENAT,
 589                 RTE_PTYPE_INNER_L2_ETHER,
 590                 RTE_PTYPE_INNER_L3_IPV4_EXT_UNKNOWN,
 591                 RTE_PTYPE_INNER_L3_IPV6_EXT_UNKNOWN,
 592                 RTE_PTYPE_INNER_L4_TCP,
 593                 RTE_PTYPE_INNER_L4_UDP,
 594                 RTE_PTYPE_INNER_L4_FRAG,
 595                 RTE_PTYPE_INNER_L4_NONFRAG,
 596                 RTE_PTYPE_UNKNOWN
 597         };
 598
 599         if (dev->rx_pkt_burst != enic_dummy_recv_pkts &&
 600             dev->rx_pkt_burst != NULL) {
 601                 struct enic *enic = pmd_priv(dev);
 602                 if (enic->overlay_offload)
 603                         return ptypes_overlay;
 604                 else
 605                         return ptypes;
 606         }
 607         return NULL;
 608 }
 609
 610 static int enicpmd_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
 611 {
 612         struct enic *enic = pmd_priv(eth_dev);
 613         int ret;
 614
 615         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 616                 return -E_RTE_SECONDARY;
 617
 618         ENICPMD_FUNC_TRACE();
 619
 620         enic->promisc = 1;
 621         ret = enic_add_packet_filter(enic);
 622         if (ret != 0)
 623                 enic->promisc = 0;
 624
 625         return ret;
 626 }
 627
 628 static int enicpmd_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
 629 {
 630         struct enic *enic = pmd_priv(eth_dev);
 631         int ret;
 632
 633         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 634                 return -E_RTE_SECONDARY;
 635
 636         ENICPMD_FUNC_TRACE();
 637         enic->promisc = 0;
 638         ret = enic_add_packet_filter(enic);
 639         if (ret != 0)
 640                 enic->promisc = 1;
 641
 642         return ret;
 643 }
 644
 645 static int enicpmd_dev_allmulticast_enable(struct rte_eth_dev *eth_dev)
 646 {
 647         struct enic *enic = pmd_priv(eth_dev);
 648         int ret;
 649
 650         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 651                 return -E_RTE_SECONDARY;
 652
 653         ENICPMD_FUNC_TRACE();
 654         enic->allmulti = 1;
 655         ret = enic_add_packet_filter(enic);
 656         if (ret != 0)
 657                 enic->allmulti = 0;
 658
 659         return ret;
 660 }
 661
 662 static int enicpmd_dev_allmulticast_disable(struct rte_eth_dev *eth_dev)
 663 {
 664         struct enic *enic = pmd_priv(eth_dev);
 665         int ret;
 666
 667         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 668                 return -E_RTE_SECONDARY;
 669
 670         ENICPMD_FUNC_TRACE();
 671         enic->allmulti = 0;
 672         ret = enic_add_packet_filter(enic);
 673         if (ret != 0)
 674                 enic->allmulti = 1;
 675
 676         return ret;
 677 }
 678
 679 static int enicpmd_add_mac_addr(struct rte_eth_dev *eth_dev,
 680         struct rte_ether_addr *mac_addr,
 681         __rte_unused uint32_t index, __rte_unused uint32_t pool)
 682 {
 683         struct enic *enic = pmd_priv(eth_dev);
 684
 685         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 686                 return -E_RTE_SECONDARY;
 687
 688         ENICPMD_FUNC_TRACE();
 689         return enic_set_mac_address(enic, mac_addr->addr_bytes);
 690 }
 691
 692 static void enicpmd_remove_mac_addr(struct rte_eth_dev *eth_dev, uint32_t index)
 693 {
 694         struct enic *enic = pmd_priv(eth_dev);
 695
 696         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 697                 return;
 698
 699         ENICPMD_FUNC_TRACE();
 700         if (enic_del_mac_address(enic, index))
 701                 dev_err(enic, "del mac addr failed\n");
 702 }
 703
 704 static int enicpmd_set_mac_addr(struct rte_eth_dev *eth_dev,
 705                                 struct rte_ether_addr *addr)
 706 {
 707         struct enic *enic = pmd_priv(eth_dev);
 708         int ret;
 709
 710         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 711                 return -E_RTE_SECONDARY;
 712
 713         ENICPMD_FUNC_TRACE();
 714         ret = enic_del_mac_address(enic, 0);
 715         if (ret)
 716                 return ret;
 717         return enic_set_mac_address(enic, addr->addr_bytes);
 718 }
 719
 720 static void debug_log_add_del_addr(struct rte_ether_addr *addr, bool add)
 721 {
 722         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 723
 724         rte_ether_format_addr(mac_str, RTE_ETHER_ADDR_FMT_SIZE, addr);
 725         ENICPMD_LOG(DEBUG, " %s address %s\n",
 726                      add ? "add" : "remove", mac_str);
 727 }
 728
 729 static int enicpmd_set_mc_addr_list(struct rte_eth_dev *eth_dev,
 730                                     struct rte_ether_addr *mc_addr_set,
 731                                     uint32_t nb_mc_addr)
 732 {
 733         struct enic *enic = pmd_priv(eth_dev);
 734         char mac_str[RTE_ETHER_ADDR_FMT_SIZE];
 735         struct rte_ether_addr *addr;
 736         uint32_t i, j;
 737         int ret;
 738
 739         ENICPMD_FUNC_TRACE();
 740
 741         /* Validate the given addresses first */
 742         for (i = 0; i < nb_mc_addr && mc_addr_set != NULL; i++) {
 743                 addr = &mc_addr_set[i];
 744                 if (!rte_is_multicast_ether_addr(addr) ||
 745                     rte_is_broadcast_ether_addr(addr)) {
 746                         rte_ether_format_addr(mac_str,
 747                                         RTE_ETHER_ADDR_FMT_SIZE, addr);
 748                         ENICPMD_LOG(ERR, " invalid multicast address %s\n",
 749                                      mac_str);
 750                         return -EINVAL;
 751                 }
 752         }
 753
 754         /* Flush all if requested */
 755         if (nb_mc_addr == 0 || mc_addr_set == NULL) {
 756                 ENICPMD_LOG(DEBUG, " flush multicast addresses\n");
 757                 for (i = 0; i < enic->mc_count; i++) {
 758                         addr = &enic->mc_addrs[i];
 759                         debug_log_add_del_addr(addr, false);
 760                         ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 761                         if (ret)
 762                                 return ret;
 763                 }
 764                 enic->mc_count = 0;
 765                 return 0;
 766         }
 767
 768         if (nb_mc_addr > ENIC_MULTICAST_PERFECT_FILTERS) {
 769                 ENICPMD_LOG(ERR, " too many multicast addresses: max=%d\n",
 770                              ENIC_MULTICAST_PERFECT_FILTERS);
 771                 return -ENOSPC;
 772         }
 773         /*
 774          * devcmd is slow, so apply the difference instead of flushing and
 775          * adding everything.
 776          * 1. Delete addresses on the NIC but not on the host
 777          */
 778         for (i = 0; i < enic->mc_count; i++) {
 779                 addr = &enic->mc_addrs[i];
 780                 for (j = 0; j < nb_mc_addr; j++) {
 781                         if (rte_is_same_ether_addr(addr, &mc_addr_set[j]))
 782                                 break;
 783                 }
 784                 if (j < nb_mc_addr)
 785                         continue;
 786                 debug_log_add_del_addr(addr, false);
 787                 ret = vnic_dev_del_addr(enic->vdev, addr->addr_bytes);
 788                 if (ret)
 789                         return ret;
 790         }
 791         /* 2. Add addresses on the host but not on the NIC */
 792         for (i = 0; i < nb_mc_addr; i++) {
 793                 addr = &mc_addr_set[i];
 794                 for (j = 0; j < enic->mc_count; j++) {
 795                         if (rte_is_same_ether_addr(addr, &enic->mc_addrs[j]))
 796                                 break;
 797                 }
 798                 if (j < enic->mc_count)
 799                         continue;
 800                 debug_log_add_del_addr(addr, true);
 801                 ret = vnic_dev_add_addr(enic->vdev, addr->addr_bytes);
 802                 if (ret)
 803                         return ret;
 804         }
 805         /* Keep a copy so we can flush/apply later on.. */
 806         memcpy(enic->mc_addrs, mc_addr_set,
 807                nb_mc_addr * sizeof(struct rte_ether_addr));
 808         enic->mc_count = nb_mc_addr;
 809         return 0;
 810 }
 811
 812 static int enicpmd_mtu_set(struct rte_eth_dev *eth_dev, uint16_t mtu)
 813 {
 814         struct enic *enic = pmd_priv(eth_dev);
 815
 816         ENICPMD_FUNC_TRACE();
 817         return enic_set_mtu(enic, mtu);
 818 }
 819
 820 static int enicpmd_dev_rss_reta_query(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         uint16_t i, idx, shift;
 827
 828         ENICPMD_FUNC_TRACE();
 829         if (reta_size != ENIC_RSS_RETA_SIZE) {
 830                 dev_err(enic, "reta_query: wrong reta_size. given=%u expected=%u\n",
 831                         reta_size, ENIC_RSS_RETA_SIZE);
 832                 return -EINVAL;
 833         }
 834
 835         for (i = 0; i < reta_size; i++) {
 836                 idx = i / RTE_RETA_GROUP_SIZE;
 837                 shift = i % RTE_RETA_GROUP_SIZE;
 838                 if (reta_conf[idx].mask & (1ULL << shift))
 839                         reta_conf[idx].reta[shift] = enic_sop_rq_idx_to_rte_idx(
 840                                 enic->rss_cpu.cpu[i / 4].b[i % 4]);
 841         }
 842
 843         return 0;
 844 }
 845
 846 static int enicpmd_dev_rss_reta_update(struct rte_eth_dev *dev,
 847                                        struct rte_eth_rss_reta_entry64
 848                                        *reta_conf,
 849                                        uint16_t reta_size)
 850 {
 851         struct enic *enic = pmd_priv(dev);
 852         union vnic_rss_cpu rss_cpu;
 853         uint16_t i, idx, shift;
 854
 855         ENICPMD_FUNC_TRACE();
 856         if (reta_size != ENIC_RSS_RETA_SIZE) {
 857                 dev_err(enic, "reta_update: wrong reta_size. given=%u"
 858                         " expected=%u\n",
 859                         reta_size, ENIC_RSS_RETA_SIZE);
 860                 return -EINVAL;
 861         }
 862         /*
 863          * Start with the current reta and modify it per reta_conf, as we
 864          * need to push the entire reta even if we only modify one entry.
 865          */
 866         rss_cpu = enic->rss_cpu;
 867         for (i = 0; i < reta_size; i++) {
 868                 idx = i / RTE_RETA_GROUP_SIZE;
 869                 shift = i % RTE_RETA_GROUP_SIZE;
 870                 if (reta_conf[idx].mask & (1ULL << shift))
 871                         rss_cpu.cpu[i / 4].b[i % 4] =
 872                                 enic_rte_rq_idx_to_sop_idx(
 873                                         reta_conf[idx].reta[shift]);
 874         }
 875         return enic_set_rss_reta(enic, &rss_cpu);
 876 }
 877
 878 static int enicpmd_dev_rss_hash_update(struct rte_eth_dev *dev,
 879                                        struct rte_eth_rss_conf *rss_conf)
 880 {
 881         struct enic *enic = pmd_priv(dev);
 882
 883         ENICPMD_FUNC_TRACE();
 884         return enic_set_rss_conf(enic, rss_conf);
 885 }
 886
 887 static int enicpmd_dev_rss_hash_conf_get(struct rte_eth_dev *dev,
 888                                          struct rte_eth_rss_conf *rss_conf)
 889 {
 890         struct enic *enic = pmd_priv(dev);
 891
 892         ENICPMD_FUNC_TRACE();
 893         if (rss_conf == NULL)
 894                 return -EINVAL;
 895         if (rss_conf->rss_key != NULL &&
 896             rss_conf->rss_key_len < ENIC_RSS_HASH_KEY_SIZE) {
 897                 dev_err(enic, "rss_hash_conf_get: wrong rss_key_len. given=%u"
 898                         " expected=%u+\n",
 899                         rss_conf->rss_key_len, ENIC_RSS_HASH_KEY_SIZE);
 900                 return -EINVAL;
 901         }
 902         rss_conf->rss_hf = enic->rss_hf;
 903         if (rss_conf->rss_key != NULL) {
 904                 int i;
 905                 for (i = 0; i < ENIC_RSS_HASH_KEY_SIZE; i++) {
 906                         rss_conf->rss_key[i] =
 907                                 enic->rss_key.key[i / 10].b[i % 10];
 908                 }
 909                 rss_conf->rss_key_len = ENIC_RSS_HASH_KEY_SIZE;
 910         }
 911         return 0;
 912 }
 913
 914 static void enicpmd_dev_rxq_info_get(struct rte_eth_dev *dev,
 915                                      uint16_t rx_queue_id,
 916                                      struct rte_eth_rxq_info *qinfo)
 917 {
 918         struct enic *enic = pmd_priv(dev);
 919         struct vnic_rq *rq_sop;
 920         struct vnic_rq *rq_data;
 921         struct rte_eth_rxconf *conf;
 922         uint16_t sop_queue_idx;
 923         uint16_t data_queue_idx;
 924
 925         ENICPMD_FUNC_TRACE();
 926         sop_queue_idx = enic_rte_rq_idx_to_sop_idx(rx_queue_id);
 927         data_queue_idx = enic_rte_rq_idx_to_data_idx(rx_queue_id, enic);
 928         rq_sop = &enic->rq[sop_queue_idx];
 929         rq_data = &enic->rq[data_queue_idx]; /* valid if data_queue_enable */
 930         qinfo->mp = rq_sop->mp;
 931         qinfo->scattered_rx = rq_sop->data_queue_enable;
 932         qinfo->nb_desc = rq_sop->ring.desc_count;
 933         if (qinfo->scattered_rx)
 934                 qinfo->nb_desc += rq_data->ring.desc_count;
 935         conf = &qinfo->conf;
 936         memset(conf, 0, sizeof(*conf));
 937         conf->rx_free_thresh = rq_sop->rx_free_thresh;
 938         conf->rx_drop_en = 1;
 939         /*
 940          * Except VLAN stripping (port setting), all the checksum offloads
 941          * are always enabled.
 942          */
 943         conf->offloads = enic->rx_offload_capa;
 944         if (!enic->ig_vlan_strip_en)
 945                 conf->offloads &= ~DEV_RX_OFFLOAD_VLAN_STRIP;
 946         /* rx_thresh and other fields are not applicable for enic */
 947 }
 948
 949 static void enicpmd_dev_txq_info_get(struct rte_eth_dev *dev,
 950                                      uint16_t tx_queue_id,
 951                                      struct rte_eth_txq_info *qinfo)
 952 {
 953         struct enic *enic = pmd_priv(dev);
 954         struct vnic_wq *wq = &enic->wq[tx_queue_id];
 955
 956         ENICPMD_FUNC_TRACE();
 957         qinfo->nb_desc = wq->ring.desc_count;
 958         memset(&qinfo->conf, 0, sizeof(qinfo->conf));
 959         qinfo->conf.offloads = wq->offloads;
 960         /* tx_thresh, and all the other fields are not applicable for enic */
 961 }
 962
 963 static int enicpmd_dev_rx_burst_mode_get(struct rte_eth_dev *dev,
 964                                          __rte_unused uint16_t queue_id,
 965                                          struct rte_eth_burst_mode *mode)
 966 {
 967         eth_rx_burst_t pkt_burst = dev->rx_pkt_burst;
 968         struct enic *enic = pmd_priv(dev);
 969         const char *info_str = NULL;
 970         int ret = -EINVAL;
 971
 972         ENICPMD_FUNC_TRACE();
 973         if (enic->use_noscatter_vec_rx_handler)
 974                 info_str = "Vector AVX2 No Scatter";
 975         else if (pkt_burst == enic_noscatter_recv_pkts)
 976                 info_str = "Scalar No Scatter";
 977         else if (pkt_burst == enic_recv_pkts)
 978                 info_str = "Scalar";
 979         if (info_str) {
 980                 strlcpy(mode->info, info_str, sizeof(mode->info));
 981                 ret = 0;
 982         }
 983         return ret;
 984 }
 985
 986 static int enicpmd_dev_tx_burst_mode_get(struct rte_eth_dev *dev,
 987                                          __rte_unused uint16_t queue_id,
 988                                          struct rte_eth_burst_mode *mode)
 989 {
 990         eth_tx_burst_t pkt_burst = dev->tx_pkt_burst;
 991         const char *info_str = NULL;
 992         int ret = -EINVAL;
 993
 994         ENICPMD_FUNC_TRACE();
 995         if (pkt_burst == enic_simple_xmit_pkts)
 996                 info_str = "Scalar Simplified";
 997         else if (pkt_burst == enic_xmit_pkts)
 998                 info_str = "Scalar";
 999         if (info_str) {
1000                 strlcpy(mode->info, info_str, sizeof(mode->info));
1001                 ret = 0;
1002         }
1003         return ret;
1004 }
1005
1006 static int enicpmd_dev_rx_queue_intr_enable(struct rte_eth_dev *eth_dev,
1007                                             uint16_t rx_queue_id)
1008 {
1009         struct enic *enic = pmd_priv(eth_dev);
1010
1011         ENICPMD_FUNC_TRACE();
1012         vnic_intr_unmask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
1013         return 0;
1014 }
1015
1016 static int enicpmd_dev_rx_queue_intr_disable(struct rte_eth_dev *eth_dev,
1017                                              uint16_t rx_queue_id)
1018 {
1019         struct enic *enic = pmd_priv(eth_dev);
1020
1021         ENICPMD_FUNC_TRACE();
1022         vnic_intr_mask(&enic->intr[rx_queue_id + ENICPMD_RXQ_INTR_OFFSET]);
1023         return 0;
1024 }
1025
1026 static int udp_tunnel_common_check(struct enic *enic,
1027                                    struct rte_eth_udp_tunnel *tnl)
1028 {
1029         if (tnl->prot_type != RTE_TUNNEL_TYPE_VXLAN)
1030                 return -ENOTSUP;
1031         if (!enic->overlay_offload) {
1032                 ENICPMD_LOG(DEBUG, " vxlan (overlay offload) is not "
1033                              "supported\n");
1034                 return -ENOTSUP;
1035         }
1036         return 0;
1037 }
1038
1039 static int update_vxlan_port(struct enic *enic, uint16_t port)
1040 {
1041         if (vnic_dev_overlay_offload_cfg(enic->vdev,
1042                                          OVERLAY_CFG_VXLAN_PORT_UPDATE,
1043                                          port)) {
1044                 ENICPMD_LOG(DEBUG, " failed to update vxlan port\n");
1045                 return -EINVAL;
1046         }
1047         ENICPMD_LOG(DEBUG, " updated vxlan port to %u\n", port);
1048         enic->vxlan_port = port;
1049         return 0;
1050 }
1051
1052 static int enicpmd_dev_udp_tunnel_port_add(struct rte_eth_dev *eth_dev,
1053                                            struct rte_eth_udp_tunnel *tnl)
1054 {
1055         struct enic *enic = pmd_priv(eth_dev);
1056         int ret;
1057
1058         ENICPMD_FUNC_TRACE();
1059         ret = udp_tunnel_common_check(enic, tnl);
1060         if (ret)
1061                 return ret;
1062         /*
1063          * The NIC has 1 configurable VXLAN port number. "Adding" a new port
1064          * number replaces it.
1065          */
1066         if (tnl->udp_port == enic->vxlan_port || tnl->udp_port == 0) {
1067                 ENICPMD_LOG(DEBUG, " %u is already configured or invalid\n",
1068                              tnl->udp_port);
1069                 return -EINVAL;
1070         }
1071         return update_vxlan_port(enic, tnl->udp_port);
1072 }
1073
1074 static int enicpmd_dev_udp_tunnel_port_del(struct rte_eth_dev *eth_dev,
1075                                            struct rte_eth_udp_tunnel *tnl)
1076 {
1077         struct enic *enic = pmd_priv(eth_dev);
1078         int ret;
1079
1080         ENICPMD_FUNC_TRACE();
1081         ret = udp_tunnel_common_check(enic, tnl);
1082         if (ret)
1083                 return ret;
1084         /*
1085          * Clear the previously set port number and restore the
1086          * hardware default port number. Some drivers disable VXLAN
1087          * offloads when there are no configured port numbers. But
1088          * enic does not do that as VXLAN is part of overlay offload,
1089          * which is tied to inner RSS and TSO.
1090          */
1091         if (tnl->udp_port != enic->vxlan_port) {
1092                 ENICPMD_LOG(DEBUG, " %u is not a configured vxlan port\n",
1093                              tnl->udp_port);
1094                 return -EINVAL;
1095         }
1096         return update_vxlan_port(enic, RTE_VXLAN_DEFAULT_PORT);
1097 }
1098
1099 static int enicpmd_dev_fw_version_get(struct rte_eth_dev *eth_dev,
1100                                       char *fw_version, size_t fw_size)
1101 {
1102         struct vnic_devcmd_fw_info *info;
1103         struct enic *enic;
1104         int ret;
1105
1106         ENICPMD_FUNC_TRACE();
1107         if (fw_version == NULL || fw_size <= 0)
1108                 return -EINVAL;
1109         enic = pmd_priv(eth_dev);
1110         ret = vnic_dev_fw_info(enic->vdev, &info);
1111         if (ret)
1112                 return ret;
1113         snprintf(fw_version, fw_size, "%s %s",
1114                  info->fw_version, info->fw_build);
1115         fw_version[fw_size - 1] = '\0';
1116         return 0;
1117 }
1118
1119 static const struct eth_dev_ops enicpmd_eth_dev_ops = {
1120         .dev_configure        = enicpmd_dev_configure,
1121         .dev_start            = enicpmd_dev_start,
1122         .dev_stop             = enicpmd_dev_stop,
1123         .dev_set_link_up      = NULL,
1124         .dev_set_link_down    = NULL,
1125         .dev_close            = enicpmd_dev_close,
1126         .promiscuous_enable   = enicpmd_dev_promiscuous_enable,
1127         .promiscuous_disable  = enicpmd_dev_promiscuous_disable,
1128         .allmulticast_enable  = enicpmd_dev_allmulticast_enable,
1129         .allmulticast_disable = enicpmd_dev_allmulticast_disable,
1130         .link_update          = enicpmd_dev_link_update,
1131         .stats_get            = enicpmd_dev_stats_get,
1132         .stats_reset          = enicpmd_dev_stats_reset,
1133         .queue_stats_mapping_set = NULL,
1134         .dev_infos_get        = enicpmd_dev_info_get,
1135         .dev_supported_ptypes_get = enicpmd_dev_supported_ptypes_get,
1136         .mtu_set              = enicpmd_mtu_set,
1137         .vlan_filter_set      = NULL,
1138         .vlan_tpid_set        = NULL,
1139         .vlan_offload_set     = enicpmd_vlan_offload_set,
1140         .vlan_strip_queue_set = NULL,
1141         .rx_queue_start       = enicpmd_dev_rx_queue_start,
1142         .rx_queue_stop        = enicpmd_dev_rx_queue_stop,
1143         .tx_queue_start       = enicpmd_dev_tx_queue_start,
1144         .tx_queue_stop        = enicpmd_dev_tx_queue_stop,
1145         .rx_queue_setup       = enicpmd_dev_rx_queue_setup,
1146         .rx_queue_release     = enicpmd_dev_rx_queue_release,
1147         .tx_queue_setup       = enicpmd_dev_tx_queue_setup,
1148         .tx_queue_release     = enicpmd_dev_tx_queue_release,
1149         .rx_queue_intr_enable = enicpmd_dev_rx_queue_intr_enable,
1150         .rx_queue_intr_disable = enicpmd_dev_rx_queue_intr_disable,
1151         .rxq_info_get         = enicpmd_dev_rxq_info_get,
1152         .txq_info_get         = enicpmd_dev_txq_info_get,
1153         .rx_burst_mode_get    = enicpmd_dev_rx_burst_mode_get,
1154         .tx_burst_mode_get    = enicpmd_dev_tx_burst_mode_get,
1155         .dev_led_on           = NULL,
1156         .dev_led_off          = NULL,
1157         .flow_ctrl_get        = NULL,
1158         .flow_ctrl_set        = NULL,
1159         .priority_flow_ctrl_set = NULL,
1160         .mac_addr_add         = enicpmd_add_mac_addr,
1161         .mac_addr_remove      = enicpmd_remove_mac_addr,
1162         .mac_addr_set         = enicpmd_set_mac_addr,
1163         .set_mc_addr_list     = enicpmd_set_mc_addr_list,
1164         .filter_ctrl          = enicpmd_dev_filter_ctrl,
1165         .reta_query           = enicpmd_dev_rss_reta_query,
1166         .reta_update          = enicpmd_dev_rss_reta_update,
1167         .rss_hash_conf_get    = enicpmd_dev_rss_hash_conf_get,
1168         .rss_hash_update      = enicpmd_dev_rss_hash_update,
1169         .udp_tunnel_port_add  = enicpmd_dev_udp_tunnel_port_add,
1170         .udp_tunnel_port_del  = enicpmd_dev_udp_tunnel_port_del,
1171         .fw_version_get       = enicpmd_dev_fw_version_get,
1172 };
1173
1174 static int enic_parse_zero_one(const char *key,
1175                                const char *value,
1176                                void *opaque)
1177 {
1178         struct enic *enic;
1179         bool b;
1180
1181         enic = (struct enic *)opaque;
1182         if (strcmp(value, "0") == 0) {
1183                 b = false;
1184         } else if (strcmp(value, "1") == 0) {
1185                 b = true;
1186         } else {
1187                 dev_err(enic, "Invalid value for %s"
1188                         ": expected=0|1 given=%s\n", key, value);
1189                 return -EINVAL;
1190         }
1191         if (strcmp(key, ENIC_DEVARG_DISABLE_OVERLAY) == 0)
1192                 enic->disable_overlay = b;
1193         if (strcmp(key, ENIC_DEVARG_ENABLE_AVX2_RX) == 0)
1194                 enic->enable_avx2_rx = b;
1195         if (strcmp(key, ENIC_DEVARG_GENEVE_OPT) == 0)
1196                 enic->geneve_opt_request = b;
1197         return 0;
1198 }
1199
1200 static int enic_parse_ig_vlan_rewrite(__rte_unused const char *key,
1201                                       const char *value,
1202                                       void *opaque)
1203 {
1204         struct enic *enic;
1205
1206         enic = (struct enic *)opaque;
1207         if (strcmp(value, "trunk") == 0) {
1208                 /* Trunk mode: always tag */
1209                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_DEFAULT_TRUNK;
1210         } else if (strcmp(value, "untag") == 0) {
1211                 /* Untag default VLAN mode: untag if VLAN = default VLAN */
1212                 enic->ig_vlan_rewrite_mode =
1213                         IG_VLAN_REWRITE_MODE_UNTAG_DEFAULT_VLAN;
1214         } else if (strcmp(value, "priority") == 0) {
1215                 /*
1216                  * Priority-tag default VLAN mode: priority tag (VLAN header
1217                  * with ID=0) if VLAN = default
1218                  */
1219                 enic->ig_vlan_rewrite_mode =
1220                         IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN;
1221         } else if (strcmp(value, "pass") == 0) {
1222                 /* Pass through mode: do not touch tags */
1223                 enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1224         } else {
1225                 dev_err(enic, "Invalid value for " ENIC_DEVARG_IG_VLAN_REWRITE
1226                         ": expected=trunk|untag|priority|pass given=%s\n",
1227                         value);
1228                 return -EINVAL;
1229         }
1230         return 0;
1231 }
1232
1233 static int enic_check_devargs(struct rte_eth_dev *dev)
1234 {
1235         static const char *const valid_keys[] = {
1236                 ENIC_DEVARG_DISABLE_OVERLAY,
1237                 ENIC_DEVARG_ENABLE_AVX2_RX,
1238                 ENIC_DEVARG_GENEVE_OPT,
1239                 ENIC_DEVARG_IG_VLAN_REWRITE,
1240                 ENIC_DEVARG_REPRESENTOR,
1241                 NULL};
1242         struct enic *enic = pmd_priv(dev);
1243         struct rte_kvargs *kvlist;
1244
1245         ENICPMD_FUNC_TRACE();
1246
1247         enic->disable_overlay = false;
1248         enic->enable_avx2_rx = false;
1249         enic->geneve_opt_request = false;
1250         enic->ig_vlan_rewrite_mode = IG_VLAN_REWRITE_MODE_PASS_THRU;
1251         if (!dev->device->devargs)
1252                 return 0;
1253         kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
1254         if (!kvlist)
1255                 return -EINVAL;
1256         if (rte_kvargs_process(kvlist, ENIC_DEVARG_DISABLE_OVERLAY,
1257                                enic_parse_zero_one, enic) < 0 ||
1258             rte_kvargs_process(kvlist, ENIC_DEVARG_ENABLE_AVX2_RX,
1259                                enic_parse_zero_one, enic) < 0 ||
1260             rte_kvargs_process(kvlist, ENIC_DEVARG_GENEVE_OPT,
1261                                enic_parse_zero_one, enic) < 0 ||
1262             rte_kvargs_process(kvlist, ENIC_DEVARG_IG_VLAN_REWRITE,
1263                                enic_parse_ig_vlan_rewrite, enic) < 0) {
1264                 rte_kvargs_free(kvlist);
1265                 return -EINVAL;
1266         }
1267         rte_kvargs_free(kvlist);
1268         return 0;
1269 }
1270
1271 /* Initialize the driver for PF */
1272 static int eth_enic_dev_init(struct rte_eth_dev *eth_dev,
1273                              void *init_params __rte_unused)
1274 {
1275         struct rte_pci_device *pdev;
1276         struct rte_pci_addr *addr;
1277         struct enic *enic = pmd_priv(eth_dev);
1278         int err;
1279
1280         ENICPMD_FUNC_TRACE();
1281         eth_dev->dev_ops = &enicpmd_eth_dev_ops;
1282         eth_dev->rx_queue_count = enicpmd_dev_rx_queue_count;
1283         eth_dev->rx_pkt_burst = &enic_recv_pkts;
1284         eth_dev->tx_pkt_burst = &enic_xmit_pkts;
1285         eth_dev->tx_pkt_prepare = &enic_prep_pkts;
1286         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1287                 enic_pick_tx_handler(eth_dev);
1288                 enic_pick_rx_handler(eth_dev);
1289                 return 0;
1290         }
1291         /* Only the primary sets up adapter and other data in shared memory */
1292         enic->port_id = eth_dev->data->port_id;
1293         enic->rte_dev = eth_dev;
1294         enic->dev_data = eth_dev->data;
1295         /* Let rte_eth_dev_close() release the port resources */
1296         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
1297
1298         pdev = RTE_ETH_DEV_TO_PCI(eth_dev);
1299         rte_eth_copy_pci_info(eth_dev, pdev);
1300         enic->pdev = pdev;
1301         addr = &pdev->addr;
1302
1303         snprintf(enic->bdf_name, ENICPMD_BDF_LENGTH, "%04x:%02x:%02x.%x",
1304                 addr->domain, addr->bus, addr->devid, addr->function);
1305
1306         err = enic_check_devargs(eth_dev);
1307         if (err)
1308                 return err;
1309         err = enic_probe(enic);
1310         if (!err && enic->fm) {
1311                 err = enic_fm_allocate_switch_domain(enic);
1312                 if (err)
1313                         ENICPMD_LOG(ERR, "failed to allocate switch domain id");
1314         }
1315         return err;
1316 }
1317
1318 static int eth_enic_dev_uninit(struct rte_eth_dev *eth_dev)
1319 {
1320         struct enic *enic = pmd_priv(eth_dev);
1321         int err;
1322
1323         ENICPMD_FUNC_TRACE();
1324         eth_dev->device = NULL;
1325         eth_dev->intr_handle = NULL;
1326         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1327                 return 0;
1328         err = rte_eth_switch_domain_free(enic->switch_domain_id);
1329         if (err)
1330                 ENICPMD_LOG(WARNING, "failed to free switch domain: %d", err);
1331         return 0;
1332 }
1333
1334 static int eth_enic_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1335         struct rte_pci_device *pci_dev)
1336 {
1337         char name[RTE_ETH_NAME_MAX_LEN];
1338         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1339         struct rte_eth_dev *pf_ethdev;
1340         struct enic *pf_enic;
1341         int i, retval;
1342
1343         ENICPMD_FUNC_TRACE();
1344         if (pci_dev->device.devargs) {
1345                 retval = rte_eth_devargs_parse(pci_dev->device.devargs->args,
1346                                 &eth_da);
1347                 if (retval)
1348                         return retval;
1349         }
1350         retval = rte_eth_dev_create(&pci_dev->device, pci_dev->device.name,
1351                 sizeof(struct enic),
1352                 eth_dev_pci_specific_init, pci_dev,
1353                 eth_enic_dev_init, NULL);
1354         if (retval || eth_da.nb_representor_ports < 1)
1355                 return retval;
1356
1357         /* Probe VF representor */
1358         pf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1359         if (pf_ethdev == NULL)
1360                 return -ENODEV;
1361         /* Representors require flowman */
1362         pf_enic = pmd_priv(pf_ethdev);
1363         if (pf_enic->fm == NULL) {
1364                 ENICPMD_LOG(ERR, "VF representors require flowman");
1365                 return -ENOTSUP;
1366         }
1367         /*
1368          * For now representors imply switchdev, as firmware does not support
1369          * legacy mode SR-IOV
1370          */
1371         pf_enic->switchdev_mode = 1;
1372         /* Calculate max VF ID before initializing representor*/
1373         pf_enic->max_vf_id = 0;
1374         for (i = 0; i < eth_da.nb_representor_ports; i++) {
1375                 pf_enic->max_vf_id = RTE_MAX(pf_enic->max_vf_id,
1376                                              eth_da.representor_ports[i]);
1377         }
1378         for (i = 0; i < eth_da.nb_representor_ports; i++) {
1379                 struct enic_vf_representor representor;
1380
1381                 representor.vf_id = eth_da.representor_ports[i];
1382                                 representor.switch_domain_id =
1383                         pmd_priv(pf_ethdev)->switch_domain_id;
1384                 representor.pf = pmd_priv(pf_ethdev);
1385                 snprintf(name, sizeof(name), "net_%s_representor_%d",
1386                         pci_dev->device.name, eth_da.representor_ports[i]);
1387                 retval = rte_eth_dev_create(&pci_dev->device, name,
1388                         sizeof(struct enic_vf_representor), NULL, NULL,
1389                         enic_vf_representor_init, &representor);
1390                 if (retval) {
1391                         ENICPMD_LOG(ERR, "failed to create enic vf representor %s",
1392                                     name);
1393                         return retval;
1394                 }
1395         }
1396         return 0;
1397 }
1398
1399 static int eth_enic_pci_remove(struct rte_pci_device *pci_dev)
1400 {
1401         struct rte_eth_dev *ethdev;
1402
1403         ENICPMD_FUNC_TRACE();
1404         ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1405         if (!ethdev)
1406                 return -ENODEV;
1407         if (ethdev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1408                 return rte_eth_dev_destroy(ethdev, enic_vf_representor_uninit);
1409         else
1410                 return rte_eth_dev_destroy(ethdev, eth_enic_dev_uninit);
1411 }
1412
1413 static struct rte_pci_driver rte_enic_pmd = {
1414         .id_table = pci_id_enic_map,
1415         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
1416         .probe = eth_enic_pci_probe,
1417         .remove = eth_enic_pci_remove,
1418 };
1419
1420 int dev_is_enic(struct rte_eth_dev *dev)
1421 {
1422         return dev->device->driver == &rte_enic_pmd.driver;
1423 }
1424
1425 RTE_PMD_REGISTER_PCI(net_enic, rte_enic_pmd);
1426 RTE_PMD_REGISTER_PCI_TABLE(net_enic, pci_id_enic_map);
1427 RTE_PMD_REGISTER_KMOD_DEP(net_enic, "* igb_uio | uio_pci_generic | vfio-pci");
1428 RTE_PMD_REGISTER_PARAM_STRING(net_enic,
1429         ENIC_DEVARG_DISABLE_OVERLAY "=0|1 "
1430         ENIC_DEVARG_ENABLE_AVX2_RX "=0|1 "
1431         ENIC_DEVARG_GENEVE_OPT "=0|1 "
1432         ENIC_DEVARG_IG_VLAN_REWRITE "=trunk|untag|priority|pass");