drivers/net/netvsc/hn_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2018 Microsoft Corporation
   3  * Copyright(c) 2013-2016 Brocade Communications Systems, Inc.
   4  * All rights reserved.
   5  */
   6
   7 #include <stdint.h>
   8 #include <string.h>
   9 #include <stdio.h>
  10 #include <errno.h>
  11 #include <unistd.h>
  12
  13 #include <rte_ethdev.h>
  14 #include <rte_memcpy.h>
  15 #include <rte_string_fns.h>
  16 #include <rte_memzone.h>
  17 #include <rte_devargs.h>
  18 #include <rte_malloc.h>
  19 #include <rte_kvargs.h>
  20 #include <rte_atomic.h>
  21 #include <rte_branch_prediction.h>
  22 #include <rte_ether.h>
  23 #include <rte_ethdev_driver.h>
  24 #include <rte_cycles.h>
  25 #include <rte_errno.h>
  26 #include <rte_memory.h>
  27 #include <rte_eal.h>
  28 #include <rte_dev.h>
  29 #include <rte_bus_vmbus.h>
  30
  31 #include "hn_logs.h"
  32 #include "hn_var.h"
  33 #include "hn_rndis.h"
  34 #include "hn_nvs.h"
  35 #include "ndis.h"
  36
  37 #define HN_TX_OFFLOAD_CAPS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
  38                             DEV_TX_OFFLOAD_TCP_CKSUM  | \
  39                             DEV_TX_OFFLOAD_UDP_CKSUM  | \
  40                             DEV_TX_OFFLOAD_TCP_TSO    | \
  41                             DEV_TX_OFFLOAD_MULTI_SEGS | \
  42                             DEV_TX_OFFLOAD_VLAN_INSERT)
  43
  44 #define HN_RX_OFFLOAD_CAPS (DEV_RX_OFFLOAD_CHECKSUM | \
  45                             DEV_RX_OFFLOAD_VLAN_STRIP)
  46
  47 int hn_logtype_init;
  48 int hn_logtype_driver;
  49
  50 struct hn_xstats_name_off {
  51         char name[RTE_ETH_XSTATS_NAME_SIZE];
  52         unsigned int offset;
  53 };
  54
  55 static const struct hn_xstats_name_off hn_stat_strings[] = {
  56         { "good_packets",           offsetof(struct hn_stats, packets) },
  57         { "good_bytes",             offsetof(struct hn_stats, bytes) },
  58         { "errors",                 offsetof(struct hn_stats, errors) },
  59         { "ring full",              offsetof(struct hn_stats, ring_full) },
  60         { "multicast_packets",      offsetof(struct hn_stats, multicast) },
  61         { "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
  62         { "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
  63         { "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
  64         { "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
  65         { "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
  66         { "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
  67         { "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
  68         { "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
  69         { "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
  70 };
  71
  72 static struct rte_eth_dev *
  73 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
  74 {
  75         struct rte_eth_dev *eth_dev;
  76         const char *name;
  77
  78         if (!dev)
  79                 return NULL;
  80
  81         name = dev->device.name;
  82
  83         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
  84                 eth_dev = rte_eth_dev_allocate(name);
  85                 if (!eth_dev) {
  86                         PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
  87                         return NULL;
  88                 }
  89
  90                 if (private_data_size) {
  91                         eth_dev->data->dev_private =
  92                                 rte_zmalloc_socket(name, private_data_size,
  93                                                      RTE_CACHE_LINE_SIZE, dev->device.numa_node);
  94                         if (!eth_dev->data->dev_private) {
  95                                 PMD_DRV_LOG(NOTICE, "can not allocate driver data");
  96                                 rte_eth_dev_release_port(eth_dev);
  97                                 return NULL;
  98                         }
  99                 }
 100         } else {
 101                 eth_dev = rte_eth_dev_attach_secondary(name);
 102                 if (!eth_dev) {
 103                         PMD_DRV_LOG(NOTICE, "can not attach secondary");
 104                         return NULL;
 105                 }
 106         }
 107
 108         eth_dev->device = &dev->device;
 109
 110         /* interrupt is simulated */
 111         dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
 112         eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
 113         eth_dev->intr_handle = &dev->intr_handle;
 114
 115         return eth_dev;
 116 }
 117
 118 static void
 119 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 120 {
 121         /* mac_addrs must not be freed alone because part of dev_private */
 122         eth_dev->data->mac_addrs = NULL;
 123         /* free ether device */
 124         rte_eth_dev_release_port(eth_dev);
 125
 126         eth_dev->device = NULL;
 127         eth_dev->intr_handle = NULL;
 128 }
 129
 130 /* handle "latency=X" from devargs */
 131 static int hn_set_latency(const char *key, const char *value, void *opaque)
 132 {
 133         struct hn_data *hv = opaque;
 134         char *endp = NULL;
 135         unsigned long lat;
 136
 137         errno = 0;
 138         lat = strtoul(value, &endp, 0);
 139
 140         if (*value == '\0' || *endp != '\0') {
 141                 PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
 142                 return -EINVAL;
 143         }
 144
 145         PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
 146
 147         hv->latency = lat * 1000;       /* usec to nsec */
 148         return 0;
 149 }
 150
 151 /* Parse device arguments */
 152 static int hn_parse_args(const struct rte_eth_dev *dev)
 153 {
 154         struct hn_data *hv = dev->data->dev_private;
 155         struct rte_devargs *devargs = dev->device->devargs;
 156         static const char * const valid_keys[] = {
 157                 "latency",
 158                 NULL
 159         };
 160         struct rte_kvargs *kvlist;
 161         int ret;
 162
 163         if (!devargs)
 164                 return 0;
 165
 166         PMD_INIT_LOG(DEBUG, "device args %s %s",
 167                      devargs->name, devargs->args);
 168
 169         kvlist = rte_kvargs_parse(devargs->args, valid_keys);
 170         if (!kvlist) {
 171                 PMD_DRV_LOG(NOTICE, "invalid parameters");
 172                 return -EINVAL;
 173         }
 174
 175         ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
 176         if (ret)
 177                 PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
 178
 179         rte_kvargs_free(kvlist);
 180         return ret;
 181 }
 182
 183 /* Update link status.
 184  * Note: the DPDK definition of "wait_to_complete"
 185  *   means block this call until link is up.
 186  *   which is not worth supporting.
 187  */
 188 int
 189 hn_dev_link_update(struct rte_eth_dev *dev,
 190                    int wait_to_complete)
 191 {
 192         struct hn_data *hv = dev->data->dev_private;
 193         struct rte_eth_link link, old;
 194         int error;
 195
 196         old = dev->data->dev_link;
 197
 198         error = hn_rndis_get_linkstatus(hv);
 199         if (error)
 200                 return error;
 201
 202         hn_rndis_get_linkspeed(hv);
 203
 204         hn_vf_link_update(dev, wait_to_complete);
 205
 206         link = (struct rte_eth_link) {
 207                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 208                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 209                 .link_speed = hv->link_speed / 10000,
 210         };
 211
 212         if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
 213                 link.link_status = ETH_LINK_UP;
 214         else
 215                 link.link_status = ETH_LINK_DOWN;
 216
 217         if (old.link_status == link.link_status)
 218                 return 0;
 219
 220         PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
 221                      (link.link_status == ETH_LINK_UP) ? "up" : "down");
 222
 223         return rte_eth_linkstatus_set(dev, &link);
 224 }
 225
 226 static void hn_dev_info_get(struct rte_eth_dev *dev,
 227                             struct rte_eth_dev_info *dev_info)
 228 {
 229         struct hn_data *hv = dev->data->dev_private;
 230
 231         dev_info->speed_capa = ETH_LINK_SPEED_10G;
 232         dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
 233         dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
 234         dev_info->max_mac_addrs  = 1;
 235
 236         dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
 237         dev_info->flow_type_rss_offloads =
 238                 ETH_RSS_IPV4 | ETH_RSS_IPV6 | ETH_RSS_TCP | ETH_RSS_UDP;
 239
 240         dev_info->max_rx_queues = hv->max_queues;
 241         dev_info->max_tx_queues = hv->max_queues;
 242
 243         hn_rndis_get_offload(hv, dev_info);
 244         hn_vf_info_get(hv, dev_info);
 245 }
 246
 247 static void
 248 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 249 {
 250         struct hn_data *hv = dev->data->dev_private;
 251
 252         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
 253         hn_vf_promiscuous_enable(dev);
 254 }
 255
 256 static void
 257 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 258 {
 259         struct hn_data *hv = dev->data->dev_private;
 260         uint32_t filter;
 261
 262         filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
 263         if (dev->data->all_multicast)
 264                 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 265         hn_rndis_set_rxfilter(hv, filter);
 266         hn_vf_promiscuous_disable(dev);
 267 }
 268
 269 static void
 270 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 271 {
 272         struct hn_data *hv = dev->data->dev_private;
 273
 274         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 275                               NDIS_PACKET_TYPE_ALL_MULTICAST |
 276                         NDIS_PACKET_TYPE_BROADCAST);
 277         hn_vf_allmulticast_enable(dev);
 278 }
 279
 280 static void
 281 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 282 {
 283         struct hn_data *hv = dev->data->dev_private;
 284
 285         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 286                              NDIS_PACKET_TYPE_BROADCAST);
 287         hn_vf_allmulticast_disable(dev);
 288 }
 289
 290 static int
 291 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
 292                      struct ether_addr *mc_addr_set,
 293                      uint32_t nb_mc_addr)
 294 {
 295         /* No filtering on the synthetic path, but can do it on VF */
 296         return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
 297 }
 298
 299 /* Setup shared rx/tx queue data */
 300 static int hn_subchan_configure(struct hn_data *hv,
 301                                 uint32_t subchan)
 302 {
 303         struct vmbus_channel *primary = hn_primary_chan(hv);
 304         int err;
 305         unsigned int retry = 0;
 306
 307         PMD_DRV_LOG(DEBUG,
 308                     "open %u subchannels", subchan);
 309
 310         /* Send create sub channels command */
 311         err = hn_nvs_alloc_subchans(hv, &subchan);
 312         if (err)
 313                 return  err;
 314
 315         while (subchan > 0) {
 316                 struct vmbus_channel *new_sc;
 317                 uint16_t chn_index;
 318
 319                 err = rte_vmbus_subchan_open(primary, &new_sc);
 320                 if (err == -ENOENT && ++retry < 1000) {
 321                         /* This can happen if not ready yet */
 322                         rte_delay_ms(10);
 323                         continue;
 324                 }
 325
 326                 if (err) {
 327                         PMD_DRV_LOG(ERR,
 328                                     "open subchannel failed: %d", err);
 329                         return err;
 330                 }
 331
 332                 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
 333
 334                 retry = 0;
 335                 chn_index = rte_vmbus_sub_channel_index(new_sc);
 336                 if (chn_index == 0 || chn_index > hv->max_queues) {
 337                         PMD_DRV_LOG(ERR,
 338                                     "Invalid subchannel offermsg channel %u",
 339                                     chn_index);
 340                         return -EIO;
 341                 }
 342
 343                 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
 344                 hv->channels[chn_index] = new_sc;
 345                 --subchan;
 346         }
 347
 348         return err;
 349 }
 350
 351 static int hn_dev_configure(struct rte_eth_dev *dev)
 352 {
 353         const struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 354         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 355         const struct rte_eth_txmode *txmode = &dev_conf->txmode;
 356
 357         const struct rte_eth_rss_conf *rss_conf =
 358                 &dev_conf->rx_adv_conf.rss_conf;
 359         struct hn_data *hv = dev->data->dev_private;
 360         uint64_t unsupported;
 361         int err, subchan;
 362
 363         PMD_INIT_FUNC_TRACE();
 364
 365         unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
 366         if (unsupported) {
 367                 PMD_DRV_LOG(NOTICE,
 368                             "unsupported TX offload: %#" PRIx64,
 369                             unsupported);
 370                 return -EINVAL;
 371         }
 372
 373         unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
 374         if (unsupported) {
 375                 PMD_DRV_LOG(NOTICE,
 376                             "unsupported RX offload: %#" PRIx64,
 377                             rxmode->offloads);
 378                 return -EINVAL;
 379         }
 380
 381         hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 382
 383         err = hn_rndis_conf_offload(hv, txmode->offloads,
 384                                     rxmode->offloads);
 385         if (err) {
 386                 PMD_DRV_LOG(NOTICE,
 387                             "offload configure failed");
 388                 return err;
 389         }
 390
 391         hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
 392                                  dev->data->nb_tx_queues);
 393         subchan = hv->num_queues - 1;
 394         if (subchan > 0) {
 395                 err = hn_subchan_configure(hv, subchan);
 396                 if (err) {
 397                         PMD_DRV_LOG(NOTICE,
 398                                     "subchannel configuration failed");
 399                         return err;
 400                 }
 401
 402                 err = hn_rndis_conf_rss(hv, rss_conf);
 403                 if (err) {
 404                         PMD_DRV_LOG(NOTICE,
 405                                     "rss configuration failed");
 406                         return err;
 407                 }
 408         }
 409
 410         return hn_vf_configure(dev, dev_conf);
 411 }
 412
 413 static int hn_dev_stats_get(struct rte_eth_dev *dev,
 414                             struct rte_eth_stats *stats)
 415 {
 416         unsigned int i;
 417
 418         hn_vf_stats_get(dev, stats);
 419
 420         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 421                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 422
 423                 if (!txq)
 424                         continue;
 425
 426                 stats->opackets += txq->stats.packets;
 427                 stats->obytes += txq->stats.bytes;
 428                 stats->oerrors += txq->stats.errors;
 429
 430                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 431                         stats->q_opackets[i] = txq->stats.packets;
 432                         stats->q_obytes[i] = txq->stats.bytes;
 433                 }
 434         }
 435
 436         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 437                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 438
 439                 if (!rxq)
 440                         continue;
 441
 442                 stats->ipackets += rxq->stats.packets;
 443                 stats->ibytes += rxq->stats.bytes;
 444                 stats->ierrors += rxq->stats.errors;
 445                 stats->imissed += rxq->stats.ring_full;
 446
 447                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 448                         stats->q_ipackets[i] = rxq->stats.packets;
 449                         stats->q_ibytes[i] = rxq->stats.bytes;
 450                 }
 451         }
 452
 453         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 454         return 0;
 455 }
 456
 457 static void
 458 hn_dev_stats_reset(struct rte_eth_dev *dev)
 459 {
 460         unsigned int i;
 461
 462         PMD_INIT_FUNC_TRACE();
 463
 464         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 465                 struct hn_tx_queue *txq = dev->data->tx_queues[i];
 466
 467                 if (!txq)
 468                         continue;
 469                 memset(&txq->stats, 0, sizeof(struct hn_stats));
 470         }
 471
 472         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 473                 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 474
 475                 if (!rxq)
 476                         continue;
 477
 478                 memset(&rxq->stats, 0, sizeof(struct hn_stats));
 479         }
 480 }
 481
 482 static void
 483 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 484 {
 485         hn_dev_stats_reset(dev);
 486         hn_vf_xstats_reset(dev);
 487 }
 488
 489 static int
 490 hn_dev_xstats_count(struct rte_eth_dev *dev)
 491 {
 492         int ret, count;
 493
 494         count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
 495         count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
 496
 497         ret = hn_vf_xstats_get_names(dev, NULL, 0);
 498         if (ret < 0)
 499                 return ret;
 500
 501         return count + ret;
 502 }
 503
 504 static int
 505 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 506                         struct rte_eth_xstat_name *xstats_names,
 507                         unsigned int limit)
 508 {
 509         unsigned int i, t, count = 0;
 510         int ret;
 511
 512         if (!xstats_names)
 513                 return hn_dev_xstats_count(dev);
 514
 515         /* Note: limit checked in rte_eth_xstats_names() */
 516         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 517                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 518
 519                 if (!txq)
 520                         continue;
 521
 522                 if (count >= limit)
 523                         break;
 524
 525                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 526                         snprintf(xstats_names[count++].name,
 527                                  RTE_ETH_XSTATS_NAME_SIZE,
 528                                  "tx_q%u_%s", i, hn_stat_strings[t].name);
 529         }
 530
 531         for (i = 0; i < dev->data->nb_rx_queues; i++)  {
 532                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 533
 534                 if (!rxq)
 535                         continue;
 536
 537                 if (count >= limit)
 538                         break;
 539
 540                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 541                         snprintf(xstats_names[count++].name,
 542                                  RTE_ETH_XSTATS_NAME_SIZE,
 543                                  "rx_q%u_%s", i,
 544                                  hn_stat_strings[t].name);
 545         }
 546
 547         ret = hn_vf_xstats_get_names(dev, xstats_names + count,
 548                                      limit - count);
 549         if (ret < 0)
 550                 return ret;
 551
 552         return count + ret;
 553 }
 554
 555 static int
 556 hn_dev_xstats_get(struct rte_eth_dev *dev,
 557                   struct rte_eth_xstat *xstats,
 558                   unsigned int n)
 559 {
 560         unsigned int i, t, count = 0;
 561         const unsigned int nstats = hn_dev_xstats_count(dev);
 562         const char *stats;
 563         int ret;
 564
 565         PMD_INIT_FUNC_TRACE();
 566
 567         if (n < nstats)
 568                 return nstats;
 569
 570         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 571                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 572
 573                 if (!txq)
 574                         continue;
 575
 576                 stats = (const char *)&txq->stats;
 577                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 578                         xstats[count++].value = *(const uint64_t *)
 579                                 (stats + hn_stat_strings[t].offset);
 580         }
 581
 582         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 583                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 584
 585                 if (!rxq)
 586                         continue;
 587
 588                 stats = (const char *)&rxq->stats;
 589                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 590                         xstats[count++].value = *(const uint64_t *)
 591                                 (stats + hn_stat_strings[t].offset);
 592         }
 593
 594         ret = hn_vf_xstats_get(dev, xstats + count, n - count);
 595         if (ret < 0)
 596                 return ret;
 597
 598         return count + ret;
 599 }
 600
 601 static int
 602 hn_dev_start(struct rte_eth_dev *dev)
 603 {
 604         struct hn_data *hv = dev->data->dev_private;
 605         int error;
 606
 607         PMD_INIT_FUNC_TRACE();
 608
 609         error = hn_rndis_set_rxfilter(hv,
 610                                       NDIS_PACKET_TYPE_BROADCAST |
 611                                       NDIS_PACKET_TYPE_ALL_MULTICAST |
 612                                       NDIS_PACKET_TYPE_DIRECTED);
 613         if (error)
 614                 return error;
 615
 616         error = hn_vf_start(dev);
 617         if (error)
 618                 hn_rndis_set_rxfilter(hv, 0);
 619
 620         return error;
 621 }
 622
 623 static void
 624 hn_dev_stop(struct rte_eth_dev *dev)
 625 {
 626         struct hn_data *hv = dev->data->dev_private;
 627
 628         PMD_INIT_FUNC_TRACE();
 629
 630         hn_rndis_set_rxfilter(hv, 0);
 631         hn_vf_stop(dev);
 632 }
 633
 634 static void
 635 hn_dev_close(struct rte_eth_dev *dev __rte_unused)
 636 {
 637         PMD_INIT_LOG(DEBUG, "close");
 638
 639         hn_vf_close(dev);
 640 }
 641
 642 static const struct eth_dev_ops hn_eth_dev_ops = {
 643         .dev_configure          = hn_dev_configure,
 644         .dev_start              = hn_dev_start,
 645         .dev_stop               = hn_dev_stop,
 646         .dev_close              = hn_dev_close,
 647         .dev_infos_get          = hn_dev_info_get,
 648         .dev_supported_ptypes_get = hn_vf_supported_ptypes,
 649         .promiscuous_enable     = hn_dev_promiscuous_enable,
 650         .promiscuous_disable    = hn_dev_promiscuous_disable,
 651         .allmulticast_enable    = hn_dev_allmulticast_enable,
 652         .allmulticast_disable   = hn_dev_allmulticast_disable,
 653         .set_mc_addr_list       = hn_dev_mc_addr_list,
 654         .tx_queue_setup         = hn_dev_tx_queue_setup,
 655         .tx_queue_release       = hn_dev_tx_queue_release,
 656         .tx_done_cleanup        = hn_dev_tx_done_cleanup,
 657         .rx_queue_setup         = hn_dev_rx_queue_setup,
 658         .rx_queue_release       = hn_dev_rx_queue_release,
 659         .link_update            = hn_dev_link_update,
 660         .stats_get              = hn_dev_stats_get,
 661         .stats_reset            = hn_dev_stats_reset,
 662         .xstats_get             = hn_dev_xstats_get,
 663         .xstats_get_names       = hn_dev_xstats_get_names,
 664         .xstats_reset           = hn_dev_xstats_reset,
 665 };
 666
 667 /*
 668  * Setup connection between PMD and kernel.
 669  */
 670 static int
 671 hn_attach(struct hn_data *hv, unsigned int mtu)
 672 {
 673         int error;
 674
 675         /* Attach NVS */
 676         error = hn_nvs_attach(hv, mtu);
 677         if (error)
 678                 goto failed_nvs;
 679
 680         /* Attach RNDIS */
 681         error = hn_rndis_attach(hv);
 682         if (error)
 683                 goto failed_rndis;
 684
 685         /*
 686          * NOTE:
 687          * Under certain conditions on certain versions of Hyper-V,
 688          * the RNDIS rxfilter is _not_ zero on the hypervisor side
 689          * after the successful RNDIS initialization.
 690          */
 691         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
 692         return 0;
 693 failed_rndis:
 694         hn_nvs_detach(hv);
 695 failed_nvs:
 696         return error;
 697 }
 698
 699 static void
 700 hn_detach(struct hn_data *hv)
 701 {
 702         hn_nvs_detach(hv);
 703         hn_rndis_detach(hv);
 704 }
 705
 706 static int
 707 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 708 {
 709         struct hn_data *hv = eth_dev->data->dev_private;
 710         struct rte_device *device = eth_dev->device;
 711         struct rte_vmbus_device *vmbus;
 712         unsigned int rxr_cnt;
 713         int err, max_chan;
 714
 715         PMD_INIT_FUNC_TRACE();
 716
 717         vmbus = container_of(device, struct rte_vmbus_device, device);
 718         eth_dev->dev_ops = &hn_eth_dev_ops;
 719         eth_dev->tx_pkt_burst = &hn_xmit_pkts;
 720         eth_dev->rx_pkt_burst = &hn_recv_pkts;
 721
 722         /*
 723          * for secondary processes, we don't initialize any further as primary
 724          * has already done this work.
 725          */
 726         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 727                 return 0;
 728
 729         /* Since Hyper-V only supports one MAC address, just use local data */
 730         eth_dev->data->mac_addrs = &hv->mac_addr;
 731
 732         hv->vmbus = vmbus;
 733         hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
 734         hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
 735         hv->port_id = eth_dev->data->port_id;
 736         hv->latency = HN_CHAN_LATENCY_NS;
 737         hv->max_queues = 1;
 738         hv->vf_port = HN_INVALID_PORT;
 739
 740         err = hn_parse_args(eth_dev);
 741         if (err)
 742                 return err;
 743
 744         strlcpy(hv->owner.name, eth_dev->device->name,
 745                 RTE_ETH_MAX_OWNER_NAME_LEN);
 746         err = rte_eth_dev_owner_new(&hv->owner.id);
 747         if (err) {
 748                 PMD_INIT_LOG(ERR, "Can not get owner id");
 749                 return err;
 750         }
 751
 752         /* Initialize primary channel input for control operations */
 753         err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 754         if (err)
 755                 return err;
 756
 757         rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
 758
 759         hv->primary = hn_rx_queue_alloc(hv, 0,
 760                                         eth_dev->device->numa_node);
 761
 762         if (!hv->primary)
 763                 return -ENOMEM;
 764
 765         err = hn_attach(hv, ETHER_MTU);
 766         if  (err)
 767                 goto failed;
 768
 769         err = hn_tx_pool_init(eth_dev);
 770         if (err)
 771                 goto failed;
 772
 773         err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
 774         if (err)
 775                 goto failed;
 776
 777         /* Multi queue requires later versions of windows server */
 778         if (hv->nvs_ver < NVS_VERSION_5)
 779                 return 0;
 780
 781         max_chan = rte_vmbus_max_channels(vmbus);
 782         PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
 783         if (max_chan <= 0)
 784                 goto failed;
 785
 786         if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
 787                 rxr_cnt = 1;
 788
 789         hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
 790
 791         /* If VF was reported but not added, do it now */
 792         if (hv->vf_present && !hn_vf_attached(hv)) {
 793                 PMD_INIT_LOG(DEBUG, "Adding VF device");
 794
 795                 err = hn_vf_add(eth_dev, hv);
 796                 if (err)
 797                         hv->vf_present = 0;
 798         }
 799
 800         return 0;
 801
 802 failed:
 803         PMD_INIT_LOG(NOTICE, "device init failed");
 804
 805         hn_tx_pool_uninit(eth_dev);
 806         hn_detach(hv);
 807         return err;
 808 }
 809
 810 static int
 811 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
 812 {
 813         struct hn_data *hv = eth_dev->data->dev_private;
 814
 815         PMD_INIT_FUNC_TRACE();
 816
 817         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 818                 return 0;
 819
 820         hn_dev_stop(eth_dev);
 821         hn_dev_close(eth_dev);
 822
 823         eth_dev->dev_ops = NULL;
 824         eth_dev->tx_pkt_burst = NULL;
 825         eth_dev->rx_pkt_burst = NULL;
 826
 827         hn_detach(hv);
 828         hn_tx_pool_uninit(eth_dev);
 829         rte_vmbus_chan_close(hv->primary->chan);
 830         rte_free(hv->primary);
 831         rte_eth_dev_owner_delete(hv->owner.id);
 832
 833         return 0;
 834 }
 835
 836 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
 837                         struct rte_vmbus_device *dev)
 838 {
 839         struct rte_eth_dev *eth_dev;
 840         int ret;
 841
 842         PMD_INIT_FUNC_TRACE();
 843
 844         eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
 845         if (!eth_dev)
 846                 return -ENOMEM;
 847
 848         ret = eth_hn_dev_init(eth_dev);
 849         if (ret)
 850                 eth_dev_vmbus_release(eth_dev);
 851         else
 852                 rte_eth_dev_probing_finish(eth_dev);
 853
 854         return ret;
 855 }
 856
 857 static int eth_hn_remove(struct rte_vmbus_device *dev)
 858 {
 859         struct rte_eth_dev *eth_dev;
 860         int ret;
 861
 862         PMD_INIT_FUNC_TRACE();
 863
 864         eth_dev = rte_eth_dev_allocated(dev->device.name);
 865         if (!eth_dev)
 866                 return -ENODEV;
 867
 868         ret = eth_hn_dev_uninit(eth_dev);
 869         if (ret)
 870                 return ret;
 871
 872         eth_dev_vmbus_release(eth_dev);
 873         return 0;
 874 }
 875
 876 /* Network device GUID */
 877 static const rte_uuid_t hn_net_ids[] = {
 878         /*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
 879         RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
 880         { 0 }
 881 };
 882
 883 static struct rte_vmbus_driver rte_netvsc_pmd = {
 884         .id_table = hn_net_ids,
 885         .probe = eth_hn_probe,
 886         .remove = eth_hn_remove,
 887 };
 888
 889 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
 890 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
 891
 892 RTE_INIT(hn_init_log)
 893 {
 894         hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
 895         if (hn_logtype_init >= 0)
 896                 rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
 897         hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
 898         if (hn_logtype_driver >= 0)
 899                 rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
 900 }