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                             DEV_RX_OFFLOAD_RSS_HASH)
  47
  48 int hn_logtype_init;
  49 int hn_logtype_driver;
  50
  51 struct hn_xstats_name_off {
  52         char name[RTE_ETH_XSTATS_NAME_SIZE];
  53         unsigned int offset;
  54 };
  55
  56 static const struct hn_xstats_name_off hn_stat_strings[] = {
  57         { "good_packets",           offsetof(struct hn_stats, packets) },
  58         { "good_bytes",             offsetof(struct hn_stats, bytes) },
  59         { "errors",                 offsetof(struct hn_stats, errors) },
  60         { "ring full",              offsetof(struct hn_stats, ring_full) },
  61         { "multicast_packets",      offsetof(struct hn_stats, multicast) },
  62         { "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
  63         { "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
  64         { "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
  65         { "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
  66         { "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
  67         { "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
  68         { "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
  69         { "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
  70         { "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
  71 };
  72
  73 /* The default RSS key.
  74  * This value is the same as MLX5 so that flows will be
  75  * received on same path for both VF ans synthetic NIC.
  76  */
  77 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
  78         0x2c, 0xc6, 0x81, 0xd1, 0x5b, 0xdb, 0xf4, 0xf7,
  79         0xfc, 0xa2, 0x83, 0x19, 0xdb, 0x1a, 0x3e, 0x94,
  80         0x6b, 0x9e, 0x38, 0xd9, 0x2c, 0x9c, 0x03, 0xd1,
  81         0xad, 0x99, 0x44, 0xa7, 0xd9, 0x56, 0x3d, 0x59,
  82         0x06, 0x3c, 0x25, 0xf3, 0xfc, 0x1f, 0xdc, 0x2a,
  83 };
  84
  85 static struct rte_eth_dev *
  86 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
  87 {
  88         struct rte_eth_dev *eth_dev;
  89         const char *name;
  90
  91         if (!dev)
  92                 return NULL;
  93
  94         name = dev->device.name;
  95
  96         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
  97                 eth_dev = rte_eth_dev_allocate(name);
  98                 if (!eth_dev) {
  99                         PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
 100                         return NULL;
 101                 }
 102
 103                 if (private_data_size) {
 104                         eth_dev->data->dev_private =
 105                                 rte_zmalloc_socket(name, private_data_size,
 106                                                      RTE_CACHE_LINE_SIZE, dev->device.numa_node);
 107                         if (!eth_dev->data->dev_private) {
 108                                 PMD_DRV_LOG(NOTICE, "can not allocate driver data");
 109                                 rte_eth_dev_release_port(eth_dev);
 110                                 return NULL;
 111                         }
 112                 }
 113         } else {
 114                 eth_dev = rte_eth_dev_attach_secondary(name);
 115                 if (!eth_dev) {
 116                         PMD_DRV_LOG(NOTICE, "can not attach secondary");
 117                         return NULL;
 118                 }
 119         }
 120
 121         eth_dev->device = &dev->device;
 122
 123         /* interrupt is simulated */
 124         dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
 125         eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
 126         eth_dev->intr_handle = &dev->intr_handle;
 127
 128         /* allow ethdev to remove on close */
 129         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 130
 131         return eth_dev;
 132 }
 133
 134 static void
 135 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 136 {
 137         /* mac_addrs must not be freed alone because part of dev_private */
 138         eth_dev->data->mac_addrs = NULL;
 139         /* free ether device */
 140         rte_eth_dev_release_port(eth_dev);
 141
 142         eth_dev->device = NULL;
 143         eth_dev->intr_handle = NULL;
 144 }
 145
 146 /* handle "latency=X" from devargs */
 147 static int hn_set_latency(const char *key, const char *value, void *opaque)
 148 {
 149         struct hn_data *hv = opaque;
 150         char *endp = NULL;
 151         unsigned long lat;
 152
 153         errno = 0;
 154         lat = strtoul(value, &endp, 0);
 155
 156         if (*value == '\0' || *endp != '\0') {
 157                 PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
 158                 return -EINVAL;
 159         }
 160
 161         PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
 162
 163         hv->latency = lat * 1000;       /* usec to nsec */
 164         return 0;
 165 }
 166
 167 /* Parse device arguments */
 168 static int hn_parse_args(const struct rte_eth_dev *dev)
 169 {
 170         struct hn_data *hv = dev->data->dev_private;
 171         struct rte_devargs *devargs = dev->device->devargs;
 172         static const char * const valid_keys[] = {
 173                 "latency",
 174                 NULL
 175         };
 176         struct rte_kvargs *kvlist;
 177         int ret;
 178
 179         if (!devargs)
 180                 return 0;
 181
 182         PMD_INIT_LOG(DEBUG, "device args %s %s",
 183                      devargs->name, devargs->args);
 184
 185         kvlist = rte_kvargs_parse(devargs->args, valid_keys);
 186         if (!kvlist) {
 187                 PMD_DRV_LOG(NOTICE, "invalid parameters");
 188                 return -EINVAL;
 189         }
 190
 191         ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
 192         if (ret)
 193                 PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
 194
 195         rte_kvargs_free(kvlist);
 196         return ret;
 197 }
 198
 199 /* Update link status.
 200  * Note: the DPDK definition of "wait_to_complete"
 201  *   means block this call until link is up.
 202  *   which is not worth supporting.
 203  */
 204 int
 205 hn_dev_link_update(struct rte_eth_dev *dev,
 206                    int wait_to_complete)
 207 {
 208         struct hn_data *hv = dev->data->dev_private;
 209         struct rte_eth_link link, old;
 210         int error;
 211
 212         old = dev->data->dev_link;
 213
 214         error = hn_rndis_get_linkstatus(hv);
 215         if (error)
 216                 return error;
 217
 218         hn_rndis_get_linkspeed(hv);
 219
 220         hn_vf_link_update(dev, wait_to_complete);
 221
 222         link = (struct rte_eth_link) {
 223                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 224                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 225                 .link_speed = hv->link_speed / 10000,
 226         };
 227
 228         if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
 229                 link.link_status = ETH_LINK_UP;
 230         else
 231                 link.link_status = ETH_LINK_DOWN;
 232
 233         if (old.link_status == link.link_status)
 234                 return 0;
 235
 236         PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
 237                      (link.link_status == ETH_LINK_UP) ? "up" : "down");
 238
 239         return rte_eth_linkstatus_set(dev, &link);
 240 }
 241
 242 static int hn_dev_info_get(struct rte_eth_dev *dev,
 243                            struct rte_eth_dev_info *dev_info)
 244 {
 245         struct hn_data *hv = dev->data->dev_private;
 246         int rc;
 247
 248         dev_info->speed_capa = ETH_LINK_SPEED_10G;
 249         dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
 250         dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
 251         dev_info->max_mac_addrs  = 1;
 252
 253         dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
 254         dev_info->flow_type_rss_offloads = hv->rss_offloads;
 255         dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
 256
 257         dev_info->max_rx_queues = hv->max_queues;
 258         dev_info->max_tx_queues = hv->max_queues;
 259
 260         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 261                 return 0;
 262
 263         /* fills in rx and tx offload capability */
 264         rc = hn_rndis_get_offload(hv, dev_info);
 265         if (rc != 0)
 266                 return rc;
 267
 268         /* merges the offload and queues of vf */
 269         return hn_vf_info_get(hv, dev_info);
 270 }
 271
 272 static int hn_rss_reta_update(struct rte_eth_dev *dev,
 273                               struct rte_eth_rss_reta_entry64 *reta_conf,
 274                               uint16_t reta_size)
 275 {
 276         struct hn_data *hv = dev->data->dev_private;
 277         unsigned int i;
 278         int err;
 279
 280         PMD_INIT_FUNC_TRACE();
 281
 282         if (reta_size != NDIS_HASH_INDCNT) {
 283                 PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
 284                 return -EINVAL;
 285         }
 286
 287         for (i = 0; i < NDIS_HASH_INDCNT; i++) {
 288                 uint16_t idx = i / RTE_RETA_GROUP_SIZE;
 289                 uint16_t shift = i % RTE_RETA_GROUP_SIZE;
 290                 uint64_t mask = (uint64_t)1 << shift;
 291
 292                 if (reta_conf[idx].mask & mask)
 293                         hv->rss_ind[i] = reta_conf[idx].reta[shift];
 294         }
 295
 296         err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 297         if (err) {
 298                 PMD_DRV_LOG(NOTICE,
 299                         "rss disable failed");
 300                 return err;
 301         }
 302
 303         err = hn_rndis_conf_rss(hv, 0);
 304         if (err) {
 305                 PMD_DRV_LOG(NOTICE,
 306                             "reta reconfig failed");
 307                 return err;
 308         }
 309
 310         return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
 311 }
 312
 313 static int hn_rss_reta_query(struct rte_eth_dev *dev,
 314                              struct rte_eth_rss_reta_entry64 *reta_conf,
 315                              uint16_t reta_size)
 316 {
 317         struct hn_data *hv = dev->data->dev_private;
 318         unsigned int i;
 319
 320         PMD_INIT_FUNC_TRACE();
 321
 322         if (reta_size != NDIS_HASH_INDCNT) {
 323                 PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
 324                 return -EINVAL;
 325         }
 326
 327         for (i = 0; i < NDIS_HASH_INDCNT; i++) {
 328                 uint16_t idx = i / RTE_RETA_GROUP_SIZE;
 329                 uint16_t shift = i % RTE_RETA_GROUP_SIZE;
 330                 uint64_t mask = (uint64_t)1 << shift;
 331
 332                 if (reta_conf[idx].mask & mask)
 333                         reta_conf[idx].reta[shift] = hv->rss_ind[i];
 334         }
 335         return 0;
 336 }
 337
 338 static void hn_rss_hash_init(struct hn_data *hv,
 339                              const struct rte_eth_rss_conf *rss_conf)
 340 {
 341         /* Convert from DPDK RSS hash flags to NDIS hash flags */
 342         hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
 343
 344         if (rss_conf->rss_hf & ETH_RSS_IPV4)
 345                 hv->rss_hash |= NDIS_HASH_IPV4;
 346         if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
 347                 hv->rss_hash |= NDIS_HASH_TCP_IPV4;
 348         if (rss_conf->rss_hf & ETH_RSS_IPV6)
 349                 hv->rss_hash |=  NDIS_HASH_IPV6;
 350         if (rss_conf->rss_hf & ETH_RSS_IPV6_EX)
 351                 hv->rss_hash |=  NDIS_HASH_IPV6_EX;
 352         if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
 353                 hv->rss_hash |= NDIS_HASH_TCP_IPV6;
 354         if (rss_conf->rss_hf & ETH_RSS_IPV6_TCP_EX)
 355                 hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
 356
 357         memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
 358                NDIS_HASH_KEYSIZE_TOEPLITZ);
 359 }
 360
 361 static int hn_rss_hash_update(struct rte_eth_dev *dev,
 362                               struct rte_eth_rss_conf *rss_conf)
 363 {
 364         struct hn_data *hv = dev->data->dev_private;
 365         int err;
 366
 367         PMD_INIT_FUNC_TRACE();
 368
 369         err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 370         if (err) {
 371                 PMD_DRV_LOG(NOTICE,
 372                             "rss disable failed");
 373                 return err;
 374         }
 375
 376         hn_rss_hash_init(hv, rss_conf);
 377
 378         err = hn_rndis_conf_rss(hv, 0);
 379         if (err) {
 380                 PMD_DRV_LOG(NOTICE,
 381                             "rss reconfig failed (RSS disabled)");
 382                 return err;
 383         }
 384
 385
 386         return hn_vf_rss_hash_update(dev, rss_conf);
 387 }
 388
 389 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
 390                                 struct rte_eth_rss_conf *rss_conf)
 391 {
 392         struct hn_data *hv = dev->data->dev_private;
 393
 394         PMD_INIT_FUNC_TRACE();
 395
 396         if (hv->ndis_ver < NDIS_VERSION_6_20) {
 397                 PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
 398                 return -EOPNOTSUPP;
 399         }
 400
 401         rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
 402         if (rss_conf->rss_key)
 403                 memcpy(rss_conf->rss_key, hv->rss_key,
 404                        NDIS_HASH_KEYSIZE_TOEPLITZ);
 405
 406         rss_conf->rss_hf = 0;
 407         if (hv->rss_hash & NDIS_HASH_IPV4)
 408                 rss_conf->rss_hf |= ETH_RSS_IPV4;
 409
 410         if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
 411                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
 412
 413         if (hv->rss_hash & NDIS_HASH_IPV6)
 414                 rss_conf->rss_hf |= ETH_RSS_IPV6;
 415
 416         if (hv->rss_hash & NDIS_HASH_IPV6_EX)
 417                 rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
 418
 419         if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
 420                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
 421
 422         if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
 423                 rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
 424
 425         return 0;
 426 }
 427
 428 static int
 429 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 430 {
 431         struct hn_data *hv = dev->data->dev_private;
 432
 433         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
 434         return hn_vf_promiscuous_enable(dev);
 435 }
 436
 437 static int
 438 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 439 {
 440         struct hn_data *hv = dev->data->dev_private;
 441         uint32_t filter;
 442
 443         filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
 444         if (dev->data->all_multicast)
 445                 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 446         hn_rndis_set_rxfilter(hv, filter);
 447         return hn_vf_promiscuous_disable(dev);
 448 }
 449
 450 static int
 451 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 452 {
 453         struct hn_data *hv = dev->data->dev_private;
 454
 455         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 456                               NDIS_PACKET_TYPE_ALL_MULTICAST |
 457                         NDIS_PACKET_TYPE_BROADCAST);
 458         return hn_vf_allmulticast_enable(dev);
 459 }
 460
 461 static int
 462 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 463 {
 464         struct hn_data *hv = dev->data->dev_private;
 465
 466         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 467                              NDIS_PACKET_TYPE_BROADCAST);
 468         return hn_vf_allmulticast_disable(dev);
 469 }
 470
 471 static int
 472 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
 473                      struct rte_ether_addr *mc_addr_set,
 474                      uint32_t nb_mc_addr)
 475 {
 476         /* No filtering on the synthetic path, but can do it on VF */
 477         return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
 478 }
 479
 480 /* Setup shared rx/tx queue data */
 481 static int hn_subchan_configure(struct hn_data *hv,
 482                                 uint32_t subchan)
 483 {
 484         struct vmbus_channel *primary = hn_primary_chan(hv);
 485         int err;
 486         unsigned int retry = 0;
 487
 488         PMD_DRV_LOG(DEBUG,
 489                     "open %u subchannels", subchan);
 490
 491         /* Send create sub channels command */
 492         err = hn_nvs_alloc_subchans(hv, &subchan);
 493         if (err)
 494                 return  err;
 495
 496         while (subchan > 0) {
 497                 struct vmbus_channel *new_sc;
 498                 uint16_t chn_index;
 499
 500                 err = rte_vmbus_subchan_open(primary, &new_sc);
 501                 if (err == -ENOENT && ++retry < 1000) {
 502                         /* This can happen if not ready yet */
 503                         rte_delay_ms(10);
 504                         continue;
 505                 }
 506
 507                 if (err) {
 508                         PMD_DRV_LOG(ERR,
 509                                     "open subchannel failed: %d", err);
 510                         return err;
 511                 }
 512
 513                 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
 514
 515                 retry = 0;
 516                 chn_index = rte_vmbus_sub_channel_index(new_sc);
 517                 if (chn_index == 0 || chn_index > hv->max_queues) {
 518                         PMD_DRV_LOG(ERR,
 519                                     "Invalid subchannel offermsg channel %u",
 520                                     chn_index);
 521                         return -EIO;
 522                 }
 523
 524                 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
 525                 hv->channels[chn_index] = new_sc;
 526                 --subchan;
 527         }
 528
 529         return err;
 530 }
 531
 532 static int hn_dev_configure(struct rte_eth_dev *dev)
 533 {
 534         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 535         struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
 536         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 537         const struct rte_eth_txmode *txmode = &dev_conf->txmode;
 538         struct hn_data *hv = dev->data->dev_private;
 539         uint64_t unsupported;
 540         int i, err, subchan;
 541
 542         PMD_INIT_FUNC_TRACE();
 543
 544         if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 545                 dev_conf->rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
 546
 547         unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
 548         if (unsupported) {
 549                 PMD_DRV_LOG(NOTICE,
 550                             "unsupported TX offload: %#" PRIx64,
 551                             unsupported);
 552                 return -EINVAL;
 553         }
 554
 555         unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
 556         if (unsupported) {
 557                 PMD_DRV_LOG(NOTICE,
 558                             "unsupported RX offload: %#" PRIx64,
 559                             rxmode->offloads);
 560                 return -EINVAL;
 561         }
 562
 563         hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 564
 565         err = hn_rndis_conf_offload(hv, txmode->offloads,
 566                                     rxmode->offloads);
 567         if (err) {
 568                 PMD_DRV_LOG(NOTICE,
 569                             "offload configure failed");
 570                 return err;
 571         }
 572
 573         hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
 574                                  dev->data->nb_tx_queues);
 575
 576         for (i = 0; i < NDIS_HASH_INDCNT; i++)
 577                 hv->rss_ind[i] = i % hv->num_queues;
 578
 579         hn_rss_hash_init(hv, rss_conf);
 580
 581         subchan = hv->num_queues - 1;
 582         if (subchan > 0) {
 583                 err = hn_subchan_configure(hv, subchan);
 584                 if (err) {
 585                         PMD_DRV_LOG(NOTICE,
 586                                     "subchannel configuration failed");
 587                         return err;
 588                 }
 589
 590                 err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 591                 if (err) {
 592                         PMD_DRV_LOG(NOTICE,
 593                                 "rss disable failed");
 594                         return err;
 595                 }
 596
 597                 err = hn_rndis_conf_rss(hv, 0);
 598                 if (err) {
 599                         PMD_DRV_LOG(NOTICE,
 600                                     "initial RSS config failed");
 601                         return err;
 602                 }
 603         }
 604
 605         return hn_vf_configure(dev, dev_conf);
 606 }
 607
 608 static int hn_dev_stats_get(struct rte_eth_dev *dev,
 609                             struct rte_eth_stats *stats)
 610 {
 611         unsigned int i;
 612
 613         hn_vf_stats_get(dev, stats);
 614
 615         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 616                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 617
 618                 if (!txq)
 619                         continue;
 620
 621                 stats->opackets += txq->stats.packets;
 622                 stats->obytes += txq->stats.bytes;
 623                 stats->oerrors += txq->stats.errors;
 624
 625                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 626                         stats->q_opackets[i] = txq->stats.packets;
 627                         stats->q_obytes[i] = txq->stats.bytes;
 628                 }
 629         }
 630
 631         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 632                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 633
 634                 if (!rxq)
 635                         continue;
 636
 637                 stats->ipackets += rxq->stats.packets;
 638                 stats->ibytes += rxq->stats.bytes;
 639                 stats->ierrors += rxq->stats.errors;
 640                 stats->imissed += rxq->stats.ring_full;
 641
 642                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 643                         stats->q_ipackets[i] = rxq->stats.packets;
 644                         stats->q_ibytes[i] = rxq->stats.bytes;
 645                 }
 646         }
 647
 648         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 649         return 0;
 650 }
 651
 652 static int
 653 hn_dev_stats_reset(struct rte_eth_dev *dev)
 654 {
 655         unsigned int i;
 656
 657         PMD_INIT_FUNC_TRACE();
 658
 659         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 660                 struct hn_tx_queue *txq = dev->data->tx_queues[i];
 661
 662                 if (!txq)
 663                         continue;
 664                 memset(&txq->stats, 0, sizeof(struct hn_stats));
 665         }
 666
 667         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 668                 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 669
 670                 if (!rxq)
 671                         continue;
 672
 673                 memset(&rxq->stats, 0, sizeof(struct hn_stats));
 674         }
 675
 676         return 0;
 677 }
 678
 679 static int
 680 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 681 {
 682         int ret;
 683
 684         ret = hn_dev_stats_reset(dev);
 685         if (ret != 0)
 686                 return 0;
 687
 688         return hn_vf_xstats_reset(dev);
 689 }
 690
 691 static int
 692 hn_dev_xstats_count(struct rte_eth_dev *dev)
 693 {
 694         int ret, count;
 695
 696         count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
 697         count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
 698
 699         ret = hn_vf_xstats_get_names(dev, NULL, 0);
 700         if (ret < 0)
 701                 return ret;
 702
 703         return count + ret;
 704 }
 705
 706 static int
 707 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 708                         struct rte_eth_xstat_name *xstats_names,
 709                         unsigned int limit)
 710 {
 711         unsigned int i, t, count = 0;
 712         int ret;
 713
 714         if (!xstats_names)
 715                 return hn_dev_xstats_count(dev);
 716
 717         /* Note: limit checked in rte_eth_xstats_names() */
 718         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 719                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 720
 721                 if (!txq)
 722                         continue;
 723
 724                 if (count >= limit)
 725                         break;
 726
 727                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 728                         snprintf(xstats_names[count++].name,
 729                                  RTE_ETH_XSTATS_NAME_SIZE,
 730                                  "tx_q%u_%s", i, hn_stat_strings[t].name);
 731         }
 732
 733         for (i = 0; i < dev->data->nb_rx_queues; i++)  {
 734                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 735
 736                 if (!rxq)
 737                         continue;
 738
 739                 if (count >= limit)
 740                         break;
 741
 742                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 743                         snprintf(xstats_names[count++].name,
 744                                  RTE_ETH_XSTATS_NAME_SIZE,
 745                                  "rx_q%u_%s", i,
 746                                  hn_stat_strings[t].name);
 747         }
 748
 749         ret = hn_vf_xstats_get_names(dev, xstats_names + count,
 750                                      limit - count);
 751         if (ret < 0)
 752                 return ret;
 753
 754         return count + ret;
 755 }
 756
 757 static int
 758 hn_dev_xstats_get(struct rte_eth_dev *dev,
 759                   struct rte_eth_xstat *xstats,
 760                   unsigned int n)
 761 {
 762         unsigned int i, t, count = 0;
 763         const unsigned int nstats = hn_dev_xstats_count(dev);
 764         const char *stats;
 765         int ret;
 766
 767         PMD_INIT_FUNC_TRACE();
 768
 769         if (n < nstats)
 770                 return nstats;
 771
 772         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 773                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 774
 775                 if (!txq)
 776                         continue;
 777
 778                 stats = (const char *)&txq->stats;
 779                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 780                         xstats[count].id = count;
 781                         xstats[count].value = *(const uint64_t *)
 782                                 (stats + hn_stat_strings[t].offset);
 783                 }
 784         }
 785
 786         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 787                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 788
 789                 if (!rxq)
 790                         continue;
 791
 792                 stats = (const char *)&rxq->stats;
 793                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 794                         xstats[count].id = count;
 795                         xstats[count].value = *(const uint64_t *)
 796                                 (stats + hn_stat_strings[t].offset);
 797                 }
 798         }
 799
 800         ret = hn_vf_xstats_get(dev, xstats, count, n);
 801         if (ret < 0)
 802                 return ret;
 803
 804         return count + ret;
 805 }
 806
 807 static int
 808 hn_dev_start(struct rte_eth_dev *dev)
 809 {
 810         struct hn_data *hv = dev->data->dev_private;
 811         int error;
 812
 813         PMD_INIT_FUNC_TRACE();
 814
 815         error = hn_rndis_set_rxfilter(hv,
 816                                       NDIS_PACKET_TYPE_BROADCAST |
 817                                       NDIS_PACKET_TYPE_ALL_MULTICAST |
 818                                       NDIS_PACKET_TYPE_DIRECTED);
 819         if (error)
 820                 return error;
 821
 822         error = hn_vf_start(dev);
 823         if (error)
 824                 hn_rndis_set_rxfilter(hv, 0);
 825
 826         return error;
 827 }
 828
 829 static void
 830 hn_dev_stop(struct rte_eth_dev *dev)
 831 {
 832         struct hn_data *hv = dev->data->dev_private;
 833
 834         PMD_INIT_FUNC_TRACE();
 835
 836         hn_rndis_set_rxfilter(hv, 0);
 837         hn_vf_stop(dev);
 838 }
 839
 840 static void
 841 hn_dev_close(struct rte_eth_dev *dev)
 842 {
 843         PMD_INIT_FUNC_TRACE();
 844
 845         hn_vf_close(dev);
 846         hn_dev_free_queues(dev);
 847 }
 848
 849 static const struct eth_dev_ops hn_eth_dev_ops = {
 850         .dev_configure          = hn_dev_configure,
 851         .dev_start              = hn_dev_start,
 852         .dev_stop               = hn_dev_stop,
 853         .dev_close              = hn_dev_close,
 854         .dev_infos_get          = hn_dev_info_get,
 855         .dev_supported_ptypes_get = hn_vf_supported_ptypes,
 856         .promiscuous_enable     = hn_dev_promiscuous_enable,
 857         .promiscuous_disable    = hn_dev_promiscuous_disable,
 858         .allmulticast_enable    = hn_dev_allmulticast_enable,
 859         .allmulticast_disable   = hn_dev_allmulticast_disable,
 860         .set_mc_addr_list       = hn_dev_mc_addr_list,
 861         .reta_update            = hn_rss_reta_update,
 862         .reta_query             = hn_rss_reta_query,
 863         .rss_hash_update        = hn_rss_hash_update,
 864         .rss_hash_conf_get      = hn_rss_hash_conf_get,
 865         .tx_queue_setup         = hn_dev_tx_queue_setup,
 866         .tx_queue_release       = hn_dev_tx_queue_release,
 867         .tx_done_cleanup        = hn_dev_tx_done_cleanup,
 868         .rx_queue_setup         = hn_dev_rx_queue_setup,
 869         .rx_queue_release       = hn_dev_rx_queue_release,
 870         .link_update            = hn_dev_link_update,
 871         .stats_get              = hn_dev_stats_get,
 872         .stats_reset            = hn_dev_stats_reset,
 873         .xstats_get             = hn_dev_xstats_get,
 874         .xstats_get_names       = hn_dev_xstats_get_names,
 875         .xstats_reset           = hn_dev_xstats_reset,
 876 };
 877
 878 /*
 879  * Setup connection between PMD and kernel.
 880  */
 881 static int
 882 hn_attach(struct hn_data *hv, unsigned int mtu)
 883 {
 884         int error;
 885
 886         /* Attach NVS */
 887         error = hn_nvs_attach(hv, mtu);
 888         if (error)
 889                 goto failed_nvs;
 890
 891         /* Attach RNDIS */
 892         error = hn_rndis_attach(hv);
 893         if (error)
 894                 goto failed_rndis;
 895
 896         /*
 897          * NOTE:
 898          * Under certain conditions on certain versions of Hyper-V,
 899          * the RNDIS rxfilter is _not_ zero on the hypervisor side
 900          * after the successful RNDIS initialization.
 901          */
 902         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
 903         return 0;
 904 failed_rndis:
 905         hn_nvs_detach(hv);
 906 failed_nvs:
 907         return error;
 908 }
 909
 910 static void
 911 hn_detach(struct hn_data *hv)
 912 {
 913         hn_nvs_detach(hv);
 914         hn_rndis_detach(hv);
 915 }
 916
 917 static int
 918 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 919 {
 920         struct hn_data *hv = eth_dev->data->dev_private;
 921         struct rte_device *device = eth_dev->device;
 922         struct rte_vmbus_device *vmbus;
 923         unsigned int rxr_cnt;
 924         int err, max_chan;
 925
 926         PMD_INIT_FUNC_TRACE();
 927
 928         vmbus = container_of(device, struct rte_vmbus_device, device);
 929         eth_dev->dev_ops = &hn_eth_dev_ops;
 930         eth_dev->tx_pkt_burst = &hn_xmit_pkts;
 931         eth_dev->rx_pkt_burst = &hn_recv_pkts;
 932
 933         /* Since Hyper-V only supports one MAC address, just use local data */
 934         eth_dev->data->mac_addrs = &hv->mac_addr;
 935
 936         /*
 937          * for secondary processes, we don't initialize any further as primary
 938          * has already done this work.
 939          */
 940         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 941                 return 0;
 942
 943         hv->vmbus = vmbus;
 944         hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
 945         hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
 946         hv->port_id = eth_dev->data->port_id;
 947         hv->latency = HN_CHAN_LATENCY_NS;
 948         hv->max_queues = 1;
 949         rte_spinlock_init(&hv->vf_lock);
 950         hv->vf_port = HN_INVALID_PORT;
 951
 952         err = hn_parse_args(eth_dev);
 953         if (err)
 954                 return err;
 955
 956         strlcpy(hv->owner.name, eth_dev->device->name,
 957                 RTE_ETH_MAX_OWNER_NAME_LEN);
 958         err = rte_eth_dev_owner_new(&hv->owner.id);
 959         if (err) {
 960                 PMD_INIT_LOG(ERR, "Can not get owner id");
 961                 return err;
 962         }
 963
 964         /* Initialize primary channel input for control operations */
 965         err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 966         if (err)
 967                 return err;
 968
 969         rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
 970
 971         hv->primary = hn_rx_queue_alloc(hv, 0,
 972                                         eth_dev->device->numa_node);
 973
 974         if (!hv->primary)
 975                 return -ENOMEM;
 976
 977         err = hn_attach(hv, RTE_ETHER_MTU);
 978         if  (err)
 979                 goto failed;
 980
 981         err = hn_tx_pool_init(eth_dev);
 982         if (err)
 983                 goto failed;
 984
 985         err = hn_rndis_get_eaddr(hv, hv->mac_addr.addr_bytes);
 986         if (err)
 987                 goto failed;
 988
 989         /* Multi queue requires later versions of windows server */
 990         if (hv->nvs_ver < NVS_VERSION_5)
 991                 return 0;
 992
 993         max_chan = rte_vmbus_max_channels(vmbus);
 994         PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
 995         if (max_chan <= 0)
 996                 goto failed;
 997
 998         if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
 999                 rxr_cnt = 1;
1000
1001         hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
1002
1003         /* If VF was reported but not added, do it now */
1004         if (hv->vf_present && !hn_vf_attached(hv)) {
1005                 PMD_INIT_LOG(DEBUG, "Adding VF device");
1006
1007                 err = hn_vf_add(eth_dev, hv);
1008                 if (err)
1009                         hv->vf_present = 0;
1010         }
1011
1012         return 0;
1013
1014 failed:
1015         PMD_INIT_LOG(NOTICE, "device init failed");
1016
1017         hn_tx_pool_uninit(eth_dev);
1018         hn_detach(hv);
1019         return err;
1020 }
1021
1022 static int
1023 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1024 {
1025         struct hn_data *hv = eth_dev->data->dev_private;
1026         int ret;
1027
1028         PMD_INIT_FUNC_TRACE();
1029
1030         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1031                 return 0;
1032
1033         hn_dev_stop(eth_dev);
1034         hn_dev_close(eth_dev);
1035
1036         eth_dev->dev_ops = NULL;
1037         eth_dev->tx_pkt_burst = NULL;
1038         eth_dev->rx_pkt_burst = NULL;
1039
1040         hn_detach(hv);
1041         hn_tx_pool_uninit(eth_dev);
1042         rte_vmbus_chan_close(hv->primary->chan);
1043         rte_free(hv->primary);
1044         ret = rte_eth_dev_owner_delete(hv->owner.id);
1045         if (ret != 0)
1046                 return ret;
1047
1048         return 0;
1049 }
1050
1051 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1052                         struct rte_vmbus_device *dev)
1053 {
1054         struct rte_eth_dev *eth_dev;
1055         int ret;
1056
1057         PMD_INIT_FUNC_TRACE();
1058
1059         eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1060         if (!eth_dev)
1061                 return -ENOMEM;
1062
1063         ret = eth_hn_dev_init(eth_dev);
1064         if (ret)
1065                 eth_dev_vmbus_release(eth_dev);
1066         else
1067                 rte_eth_dev_probing_finish(eth_dev);
1068
1069         return ret;
1070 }
1071
1072 static int eth_hn_remove(struct rte_vmbus_device *dev)
1073 {
1074         struct rte_eth_dev *eth_dev;
1075         int ret;
1076
1077         PMD_INIT_FUNC_TRACE();
1078
1079         eth_dev = rte_eth_dev_allocated(dev->device.name);
1080         if (!eth_dev)
1081                 return -ENODEV;
1082
1083         ret = eth_hn_dev_uninit(eth_dev);
1084         if (ret)
1085                 return ret;
1086
1087         eth_dev_vmbus_release(eth_dev);
1088         return 0;
1089 }
1090
1091 /* Network device GUID */
1092 static const rte_uuid_t hn_net_ids[] = {
1093         /*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1094         RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1095         { 0 }
1096 };
1097
1098 static struct rte_vmbus_driver rte_netvsc_pmd = {
1099         .id_table = hn_net_ids,
1100         .probe = eth_hn_probe,
1101         .remove = eth_hn_remove,
1102 };
1103
1104 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1105 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1106
1107 RTE_INIT(hn_init_log)
1108 {
1109         hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
1110         if (hn_logtype_init >= 0)
1111                 rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
1112         hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
1113         if (hn_logtype_driver >= 0)
1114                 rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
1115 }