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         /* allow ethdev to remove on close */
 116         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 117
 118         return eth_dev;
 119 }
 120
 121 static void
 122 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 123 {
 124         /* mac_addrs must not be freed alone because part of dev_private */
 125         eth_dev->data->mac_addrs = NULL;
 126         /* free ether device */
 127         rte_eth_dev_release_port(eth_dev);
 128
 129         eth_dev->device = NULL;
 130         eth_dev->intr_handle = NULL;
 131 }
 132
 133 /* handle "latency=X" from devargs */
 134 static int hn_set_latency(const char *key, const char *value, void *opaque)
 135 {
 136         struct hn_data *hv = opaque;
 137         char *endp = NULL;
 138         unsigned long lat;
 139
 140         errno = 0;
 141         lat = strtoul(value, &endp, 0);
 142
 143         if (*value == '\0' || *endp != '\0') {
 144                 PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
 145                 return -EINVAL;
 146         }
 147
 148         PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
 149
 150         hv->latency = lat * 1000;       /* usec to nsec */
 151         return 0;
 152 }
 153
 154 /* Parse device arguments */
 155 static int hn_parse_args(const struct rte_eth_dev *dev)
 156 {
 157         struct hn_data *hv = dev->data->dev_private;
 158         struct rte_devargs *devargs = dev->device->devargs;
 159         static const char * const valid_keys[] = {
 160                 "latency",
 161                 NULL
 162         };
 163         struct rte_kvargs *kvlist;
 164         int ret;
 165
 166         if (!devargs)
 167                 return 0;
 168
 169         PMD_INIT_LOG(DEBUG, "device args %s %s",
 170                      devargs->name, devargs->args);
 171
 172         kvlist = rte_kvargs_parse(devargs->args, valid_keys);
 173         if (!kvlist) {
 174                 PMD_DRV_LOG(NOTICE, "invalid parameters");
 175                 return -EINVAL;
 176         }
 177
 178         ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
 179         if (ret)
 180                 PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
 181
 182         rte_kvargs_free(kvlist);
 183         return ret;
 184 }
 185
 186 /* Update link status.
 187  * Note: the DPDK definition of "wait_to_complete"
 188  *   means block this call until link is up.
 189  *   which is not worth supporting.
 190  */
 191 int
 192 hn_dev_link_update(struct rte_eth_dev *dev,
 193                    int wait_to_complete)
 194 {
 195         struct hn_data *hv = dev->data->dev_private;
 196         struct rte_eth_link link, old;
 197         int error;
 198
 199         old = dev->data->dev_link;
 200
 201         error = hn_rndis_get_linkstatus(hv);
 202         if (error)
 203                 return error;
 204
 205         hn_rndis_get_linkspeed(hv);
 206
 207         hn_vf_link_update(dev, wait_to_complete);
 208
 209         link = (struct rte_eth_link) {
 210                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 211                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 212                 .link_speed = hv->link_speed / 10000,
 213         };
 214
 215         if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
 216                 link.link_status = ETH_LINK_UP;
 217         else
 218                 link.link_status = ETH_LINK_DOWN;
 219
 220         if (old.link_status == link.link_status)
 221                 return 0;
 222
 223         PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
 224                      (link.link_status == ETH_LINK_UP) ? "up" : "down");
 225
 226         return rte_eth_linkstatus_set(dev, &link);
 227 }
 228
 229 static void hn_dev_info_get(struct rte_eth_dev *dev,
 230                             struct rte_eth_dev_info *dev_info)
 231 {
 232         struct hn_data *hv = dev->data->dev_private;
 233
 234         dev_info->speed_capa = ETH_LINK_SPEED_10G;
 235         dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
 236         dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
 237         dev_info->max_mac_addrs  = 1;
 238
 239         dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
 240         dev_info->flow_type_rss_offloads =
 241                 ETH_RSS_IPV4 | ETH_RSS_IPV6 | ETH_RSS_TCP | ETH_RSS_UDP;
 242
 243         dev_info->max_rx_queues = hv->max_queues;
 244         dev_info->max_tx_queues = hv->max_queues;
 245
 246         hn_rndis_get_offload(hv, dev_info);
 247         hn_vf_info_get(hv, dev_info);
 248 }
 249
 250 static void
 251 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 252 {
 253         struct hn_data *hv = dev->data->dev_private;
 254
 255         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
 256         hn_vf_promiscuous_enable(dev);
 257 }
 258
 259 static void
 260 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 261 {
 262         struct hn_data *hv = dev->data->dev_private;
 263         uint32_t filter;
 264
 265         filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
 266         if (dev->data->all_multicast)
 267                 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 268         hn_rndis_set_rxfilter(hv, filter);
 269         hn_vf_promiscuous_disable(dev);
 270 }
 271
 272 static void
 273 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 274 {
 275         struct hn_data *hv = dev->data->dev_private;
 276
 277         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 278                               NDIS_PACKET_TYPE_ALL_MULTICAST |
 279                         NDIS_PACKET_TYPE_BROADCAST);
 280         hn_vf_allmulticast_enable(dev);
 281 }
 282
 283 static void
 284 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 285 {
 286         struct hn_data *hv = dev->data->dev_private;
 287
 288         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 289                              NDIS_PACKET_TYPE_BROADCAST);
 290         hn_vf_allmulticast_disable(dev);
 291 }
 292
 293 static int
 294 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
 295                      struct rte_ether_addr *mc_addr_set,
 296                      uint32_t nb_mc_addr)
 297 {
 298         /* No filtering on the synthetic path, but can do it on VF */
 299         return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
 300 }
 301
 302 /* Setup shared rx/tx queue data */
 303 static int hn_subchan_configure(struct hn_data *hv,
 304                                 uint32_t subchan)
 305 {
 306         struct vmbus_channel *primary = hn_primary_chan(hv);
 307         int err;
 308         unsigned int retry = 0;
 309
 310         PMD_DRV_LOG(DEBUG,
 311                     "open %u subchannels", subchan);
 312
 313         /* Send create sub channels command */
 314         err = hn_nvs_alloc_subchans(hv, &subchan);
 315         if (err)
 316                 return  err;
 317
 318         while (subchan > 0) {
 319                 struct vmbus_channel *new_sc;
 320                 uint16_t chn_index;
 321
 322                 err = rte_vmbus_subchan_open(primary, &new_sc);
 323                 if (err == -ENOENT && ++retry < 1000) {
 324                         /* This can happen if not ready yet */
 325                         rte_delay_ms(10);
 326                         continue;
 327                 }
 328
 329                 if (err) {
 330                         PMD_DRV_LOG(ERR,
 331                                     "open subchannel failed: %d", err);
 332                         return err;
 333                 }
 334
 335                 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
 336
 337                 retry = 0;
 338                 chn_index = rte_vmbus_sub_channel_index(new_sc);
 339                 if (chn_index == 0 || chn_index > hv->max_queues) {
 340                         PMD_DRV_LOG(ERR,
 341                                     "Invalid subchannel offermsg channel %u",
 342                                     chn_index);
 343                         return -EIO;
 344                 }
 345
 346                 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
 347                 hv->channels[chn_index] = new_sc;
 348                 --subchan;
 349         }
 350
 351         return err;
 352 }
 353
 354 static int hn_dev_configure(struct rte_eth_dev *dev)
 355 {
 356         const struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 357         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 358         const struct rte_eth_txmode *txmode = &dev_conf->txmode;
 359
 360         const struct rte_eth_rss_conf *rss_conf =
 361                 &dev_conf->rx_adv_conf.rss_conf;
 362         struct hn_data *hv = dev->data->dev_private;
 363         uint64_t unsupported;
 364         int err, subchan;
 365
 366         PMD_INIT_FUNC_TRACE();
 367
 368         unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
 369         if (unsupported) {
 370                 PMD_DRV_LOG(NOTICE,
 371                             "unsupported TX offload: %#" PRIx64,
 372                             unsupported);
 373                 return -EINVAL;
 374         }
 375
 376         unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
 377         if (unsupported) {
 378                 PMD_DRV_LOG(NOTICE,
 379                             "unsupported RX offload: %#" PRIx64,
 380                             rxmode->offloads);
 381                 return -EINVAL;
 382         }
 383
 384         hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 385
 386         err = hn_rndis_conf_offload(hv, txmode->offloads,
 387                                     rxmode->offloads);
 388         if (err) {
 389                 PMD_DRV_LOG(NOTICE,
 390                             "offload configure failed");
 391                 return err;
 392         }
 393
 394         hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
 395                                  dev->data->nb_tx_queues);
 396         subchan = hv->num_queues - 1;
 397         if (subchan > 0) {
 398                 err = hn_subchan_configure(hv, subchan);
 399                 if (err) {
 400                         PMD_DRV_LOG(NOTICE,
 401                                     "subchannel configuration failed");
 402                         return err;
 403                 }
 404
 405                 err = hn_rndis_conf_rss(hv, rss_conf);
 406                 if (err) {
 407                         PMD_DRV_LOG(NOTICE,
 408                                     "rss configuration failed");
 409                         return err;
 410                 }
 411         }
 412
 413         return hn_vf_configure(dev, dev_conf);
 414 }
 415
 416 static int hn_dev_stats_get(struct rte_eth_dev *dev,
 417                             struct rte_eth_stats *stats)
 418 {
 419         unsigned int i;
 420
 421         hn_vf_stats_get(dev, stats);
 422
 423         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 424                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 425
 426                 if (!txq)
 427                         continue;
 428
 429                 stats->opackets += txq->stats.packets;
 430                 stats->obytes += txq->stats.bytes;
 431                 stats->oerrors += txq->stats.errors;
 432
 433                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 434                         stats->q_opackets[i] = txq->stats.packets;
 435                         stats->q_obytes[i] = txq->stats.bytes;
 436                 }
 437         }
 438
 439         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 440                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 441
 442                 if (!rxq)
 443                         continue;
 444
 445                 stats->ipackets += rxq->stats.packets;
 446                 stats->ibytes += rxq->stats.bytes;
 447                 stats->ierrors += rxq->stats.errors;
 448                 stats->imissed += rxq->stats.ring_full;
 449
 450                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 451                         stats->q_ipackets[i] = rxq->stats.packets;
 452                         stats->q_ibytes[i] = rxq->stats.bytes;
 453                 }
 454         }
 455
 456         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 457         return 0;
 458 }
 459
 460 static void
 461 hn_dev_stats_reset(struct rte_eth_dev *dev)
 462 {
 463         unsigned int i;
 464
 465         PMD_INIT_FUNC_TRACE();
 466
 467         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 468                 struct hn_tx_queue *txq = dev->data->tx_queues[i];
 469
 470                 if (!txq)
 471                         continue;
 472                 memset(&txq->stats, 0, sizeof(struct hn_stats));
 473         }
 474
 475         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 476                 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 477
 478                 if (!rxq)
 479                         continue;
 480
 481                 memset(&rxq->stats, 0, sizeof(struct hn_stats));
 482         }
 483 }
 484
 485 static void
 486 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 487 {
 488         hn_dev_stats_reset(dev);
 489         hn_vf_xstats_reset(dev);
 490 }
 491
 492 static int
 493 hn_dev_xstats_count(struct rte_eth_dev *dev)
 494 {
 495         int ret, count;
 496
 497         count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
 498         count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
 499
 500         ret = hn_vf_xstats_get_names(dev, NULL, 0);
 501         if (ret < 0)
 502                 return ret;
 503
 504         return count + ret;
 505 }
 506
 507 static int
 508 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 509                         struct rte_eth_xstat_name *xstats_names,
 510                         unsigned int limit)
 511 {
 512         unsigned int i, t, count = 0;
 513         int ret;
 514
 515         if (!xstats_names)
 516                 return hn_dev_xstats_count(dev);
 517
 518         /* Note: limit checked in rte_eth_xstats_names() */
 519         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 520                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 521
 522                 if (!txq)
 523                         continue;
 524
 525                 if (count >= limit)
 526                         break;
 527
 528                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 529                         snprintf(xstats_names[count++].name,
 530                                  RTE_ETH_XSTATS_NAME_SIZE,
 531                                  "tx_q%u_%s", i, hn_stat_strings[t].name);
 532         }
 533
 534         for (i = 0; i < dev->data->nb_rx_queues; i++)  {
 535                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 536
 537                 if (!rxq)
 538                         continue;
 539
 540                 if (count >= limit)
 541                         break;
 542
 543                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 544                         snprintf(xstats_names[count++].name,
 545                                  RTE_ETH_XSTATS_NAME_SIZE,
 546                                  "rx_q%u_%s", i,
 547                                  hn_stat_strings[t].name);
 548         }
 549
 550         ret = hn_vf_xstats_get_names(dev, xstats_names + count,
 551                                      limit - count);
 552         if (ret < 0)
 553                 return ret;
 554
 555         return count + ret;
 556 }
 557
 558 static int
 559 hn_dev_xstats_get(struct rte_eth_dev *dev,
 560                   struct rte_eth_xstat *xstats,
 561                   unsigned int n)
 562 {
 563         unsigned int i, t, count = 0;
 564         const unsigned int nstats = hn_dev_xstats_count(dev);
 565         const char *stats;
 566         int ret;
 567
 568         PMD_INIT_FUNC_TRACE();
 569
 570         if (n < nstats)
 571                 return nstats;
 572
 573         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 574                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 575
 576                 if (!txq)
 577                         continue;
 578
 579                 stats = (const char *)&txq->stats;
 580                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 581                         xstats[count++].value = *(const uint64_t *)
 582                                 (stats + hn_stat_strings[t].offset);
 583         }
 584
 585         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 586                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 587
 588                 if (!rxq)
 589                         continue;
 590
 591                 stats = (const char *)&rxq->stats;
 592                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 593                         xstats[count++].value = *(const uint64_t *)
 594                                 (stats + hn_stat_strings[t].offset);
 595         }
 596
 597         ret = hn_vf_xstats_get(dev, xstats + count, n - count);
 598         if (ret < 0)
 599                 return ret;
 600
 601         return count + ret;
 602 }
 603
 604 static int
 605 hn_dev_start(struct rte_eth_dev *dev)
 606 {
 607         struct hn_data *hv = dev->data->dev_private;
 608         int error;
 609
 610         PMD_INIT_FUNC_TRACE();
 611
 612         error = hn_rndis_set_rxfilter(hv,
 613                                       NDIS_PACKET_TYPE_BROADCAST |
 614                                       NDIS_PACKET_TYPE_ALL_MULTICAST |
 615                                       NDIS_PACKET_TYPE_DIRECTED);
 616         if (error)
 617                 return error;
 618
 619         error = hn_vf_start(dev);
 620         if (error)
 621                 hn_rndis_set_rxfilter(hv, 0);
 622
 623         return error;
 624 }
 625
 626 static void
 627 hn_dev_stop(struct rte_eth_dev *dev)
 628 {
 629         struct hn_data *hv = dev->data->dev_private;
 630
 631         PMD_INIT_FUNC_TRACE();
 632
 633         hn_rndis_set_rxfilter(hv, 0);
 634         hn_vf_stop(dev);
 635 }
 636
 637 static void
 638 hn_dev_close(struct rte_eth_dev *dev)
 639 {
 640         PMD_INIT_FUNC_TRACE();
 641
 642         hn_vf_close(dev);
 643         hn_dev_free_queues(dev);
 644 }
 645
 646 static const struct eth_dev_ops hn_eth_dev_ops = {
 647         .dev_configure          = hn_dev_configure,
 648         .dev_start              = hn_dev_start,
 649         .dev_stop               = hn_dev_stop,
 650         .dev_close              = hn_dev_close,
 651         .dev_infos_get          = hn_dev_info_get,
 652         .dev_supported_ptypes_get = hn_vf_supported_ptypes,
 653         .promiscuous_enable     = hn_dev_promiscuous_enable,
 654         .promiscuous_disable    = hn_dev_promiscuous_disable,
 655         .allmulticast_enable    = hn_dev_allmulticast_enable,
 656         .allmulticast_disable   = hn_dev_allmulticast_disable,
 657         .set_mc_addr_list       = hn_dev_mc_addr_list,
 658         .tx_queue_setup         = hn_dev_tx_queue_setup,
 659         .tx_queue_release       = hn_dev_tx_queue_release,
 660         .tx_done_cleanup        = hn_dev_tx_done_cleanup,
 661         .rx_queue_setup         = hn_dev_rx_queue_setup,
 662         .rx_queue_release       = hn_dev_rx_queue_release,
 663         .link_update            = hn_dev_link_update,
 664         .stats_get              = hn_dev_stats_get,
 665         .stats_reset            = hn_dev_stats_reset,
 666         .xstats_get             = hn_dev_xstats_get,
 667         .xstats_get_names       = hn_dev_xstats_get_names,
 668         .xstats_reset           = hn_dev_xstats_reset,
 669 };
 670
 671 /*
 672  * Setup connection between PMD and kernel.
 673  */
 674 static int
 675 hn_attach(struct hn_data *hv, unsigned int mtu)
 676 {
 677         int error;
 678
 679         /* Attach NVS */
 680         error = hn_nvs_attach(hv, mtu);
 681         if (error)
 682                 goto failed_nvs;
 683
 684         /* Attach RNDIS */
 685         error = hn_rndis_attach(hv);
 686         if (error)
 687                 goto failed_rndis;
 688
 689         /*
 690          * NOTE:
 691          * Under certain conditions on certain versions of Hyper-V,
 692          * the RNDIS rxfilter is _not_ zero on the hypervisor side
 693          * after the successful RNDIS initialization.
 694          */
 695         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
 696         return 0;
 697 failed_rndis:
 698         hn_nvs_detach(hv);
 699 failed_nvs:
 700         return error;
 701 }
 702
 703 static void
 704 hn_detach(struct hn_data *hv)
 705 {
 706         hn_nvs_detach(hv);
 707         hn_rndis_detach(hv);
 708 }
 709
 710 static int
 711 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 712 {
 713         struct hn_data *hv = eth_dev->data->dev_private;
 714         struct rte_device *device = eth_dev->device;
 715         struct rte_vmbus_device *vmbus;
 716         unsigned int rxr_cnt;
 717         int err, max_chan;
 718
 719         PMD_INIT_FUNC_TRACE();
 720
 721         vmbus = container_of(device, struct rte_vmbus_device, device);
 722         eth_dev->dev_ops = &hn_eth_dev_ops;
 723         eth_dev->tx_pkt_burst = &hn_xmit_pkts;
 724         eth_dev->rx_pkt_burst = &hn_recv_pkts;
 725
 726         /*
 727          * for secondary processes, we don't initialize any further as primary
 728          * has already done this work.
 729          */
 730         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 731                 return 0;
 732
 733         /* Since Hyper-V only supports one MAC address, just use local data */
 734         eth_dev->data->mac_addrs = &hv->mac_addr;
 735
 736         hv->vmbus = vmbus;
 737         hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
 738         hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
 739         hv->port_id = eth_dev->data->port_id;
 740         hv->latency = HN_CHAN_LATENCY_NS;
 741         hv->max_queues = 1;
 742         hv->vf_port = HN_INVALID_PORT;
 743
 744         err = hn_parse_args(eth_dev);
 745         if (err)
 746                 return err;
 747
 748         strlcpy(hv->owner.name, eth_dev->device->name,
 749                 RTE_ETH_MAX_OWNER_NAME_LEN);
 750         err = rte_eth_dev_owner_new(&hv->owner.id);
 751         if (err) {
 752                 PMD_INIT_LOG(ERR, "Can not get owner id");
 753                 return err;
 754         }
 755
 756         /* Initialize primary channel input for control operations */
 757         err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 758         if (err)
 759                 return err;
 760
 761         rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
 762
 763         hv->primary = hn_rx_queue_alloc(hv, 0,
 764                                         eth_dev->device->numa_node);
 765
 766         if (!hv->primary)
 767                 return -ENOMEM;
 768
 769         err = hn_attach(hv, RTE_ETHER_MTU);
 770         if  (err)
 771                 goto failed;
 772
 773         err = hn_tx_pool_init(eth_dev);
 774         if (err)
 775                 goto failed;
 776
 777         err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
 778         if (err)
 779                 goto failed;
 780
 781         /* Multi queue requires later versions of windows server */
 782         if (hv->nvs_ver < NVS_VERSION_5)
 783                 return 0;
 784
 785         max_chan = rte_vmbus_max_channels(vmbus);
 786         PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
 787         if (max_chan <= 0)
 788                 goto failed;
 789
 790         if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
 791                 rxr_cnt = 1;
 792
 793         hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
 794
 795         /* If VF was reported but not added, do it now */
 796         if (hv->vf_present && !hn_vf_attached(hv)) {
 797                 PMD_INIT_LOG(DEBUG, "Adding VF device");
 798
 799                 err = hn_vf_add(eth_dev, hv);
 800                 if (err)
 801                         hv->vf_present = 0;
 802         }
 803
 804         return 0;
 805
 806 failed:
 807         PMD_INIT_LOG(NOTICE, "device init failed");
 808
 809         hn_tx_pool_uninit(eth_dev);
 810         hn_detach(hv);
 811         return err;
 812 }
 813
 814 static int
 815 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
 816 {
 817         struct hn_data *hv = eth_dev->data->dev_private;
 818
 819         PMD_INIT_FUNC_TRACE();
 820
 821         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 822                 return 0;
 823
 824         hn_dev_stop(eth_dev);
 825         hn_dev_close(eth_dev);
 826
 827         eth_dev->dev_ops = NULL;
 828         eth_dev->tx_pkt_burst = NULL;
 829         eth_dev->rx_pkt_burst = NULL;
 830
 831         hn_detach(hv);
 832         hn_tx_pool_uninit(eth_dev);
 833         rte_vmbus_chan_close(hv->primary->chan);
 834         rte_free(hv->primary);
 835         rte_eth_dev_owner_delete(hv->owner.id);
 836
 837         return 0;
 838 }
 839
 840 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
 841                         struct rte_vmbus_device *dev)
 842 {
 843         struct rte_eth_dev *eth_dev;
 844         int ret;
 845
 846         PMD_INIT_FUNC_TRACE();
 847
 848         eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
 849         if (!eth_dev)
 850                 return -ENOMEM;
 851
 852         ret = eth_hn_dev_init(eth_dev);
 853         if (ret)
 854                 eth_dev_vmbus_release(eth_dev);
 855         else
 856                 rte_eth_dev_probing_finish(eth_dev);
 857
 858         return ret;
 859 }
 860
 861 static int eth_hn_remove(struct rte_vmbus_device *dev)
 862 {
 863         struct rte_eth_dev *eth_dev;
 864         int ret;
 865
 866         PMD_INIT_FUNC_TRACE();
 867
 868         eth_dev = rte_eth_dev_allocated(dev->device.name);
 869         if (!eth_dev)
 870                 return -ENODEV;
 871
 872         ret = eth_hn_dev_uninit(eth_dev);
 873         if (ret)
 874                 return ret;
 875
 876         eth_dev_vmbus_release(eth_dev);
 877         return 0;
 878 }
 879
 880 /* Network device GUID */
 881 static const rte_uuid_t hn_net_ids[] = {
 882         /*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
 883         RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
 884         { 0 }
 885 };
 886
 887 static struct rte_vmbus_driver rte_netvsc_pmd = {
 888         .id_table = hn_net_ids,
 889         .probe = eth_hn_probe,
 890         .remove = eth_hn_remove,
 891 };
 892
 893 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
 894 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
 895
 896 RTE_INIT(hn_init_log)
 897 {
 898         hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
 899         if (hn_logtype_init >= 0)
 900                 rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
 901         hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
 902         if (hn_logtype_driver >= 0)
 903                 rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
 904 }