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