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         { "channel full",           offsetof(struct hn_stats, channel_full) },
  62         { "multicast_packets",      offsetof(struct hn_stats, multicast) },
  63         { "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
  64         { "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
  65         { "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
  66         { "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
  67         { "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
  68         { "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
  69         { "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
  70         { "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
  71         { "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
  72 };
  73
  74 /* The default RSS key.
  75  * This value is the same as MLX5 so that flows will be
  76  * received on same path for both VF and synthetic NIC.
  77  */
  78 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
  79         0x2c, 0xc6, 0x81, 0xd1, 0x5b, 0xdb, 0xf4, 0xf7,
  80         0xfc, 0xa2, 0x83, 0x19, 0xdb, 0x1a, 0x3e, 0x94,
  81         0x6b, 0x9e, 0x38, 0xd9, 0x2c, 0x9c, 0x03, 0xd1,
  82         0xad, 0x99, 0x44, 0xa7, 0xd9, 0x56, 0x3d, 0x59,
  83         0x06, 0x3c, 0x25, 0xf3, 0xfc, 0x1f, 0xdc, 0x2a,
  84 };
  85
  86 static struct rte_eth_dev *
  87 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
  88 {
  89         struct rte_eth_dev *eth_dev;
  90         const char *name;
  91
  92         if (!dev)
  93                 return NULL;
  94
  95         name = dev->device.name;
  96
  97         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
  98                 eth_dev = rte_eth_dev_allocate(name);
  99                 if (!eth_dev) {
 100                         PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
 101                         return NULL;
 102                 }
 103
 104                 if (private_data_size) {
 105                         eth_dev->data->dev_private =
 106                                 rte_zmalloc_socket(name, private_data_size,
 107                                                      RTE_CACHE_LINE_SIZE, dev->device.numa_node);
 108                         if (!eth_dev->data->dev_private) {
 109                                 PMD_DRV_LOG(NOTICE, "can not allocate driver data");
 110                                 rte_eth_dev_release_port(eth_dev);
 111                                 return NULL;
 112                         }
 113                 }
 114         } else {
 115                 eth_dev = rte_eth_dev_attach_secondary(name);
 116                 if (!eth_dev) {
 117                         PMD_DRV_LOG(NOTICE, "can not attach secondary");
 118                         return NULL;
 119                 }
 120         }
 121
 122         eth_dev->device = &dev->device;
 123
 124         /* interrupt is simulated */
 125         dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
 126         eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
 127         eth_dev->intr_handle = &dev->intr_handle;
 128
 129         /* allow ethdev to remove on close */
 130         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 131
 132         return eth_dev;
 133 }
 134
 135 static void
 136 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 137 {
 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 __rte_unused)
 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         link = (struct rte_eth_link) {
 220                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 221                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 222                 .link_speed = hv->link_speed / 10000,
 223         };
 224
 225         if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
 226                 link.link_status = ETH_LINK_UP;
 227         else
 228                 link.link_status = ETH_LINK_DOWN;
 229
 230         if (old.link_status == link.link_status)
 231                 return 0;
 232
 233         PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
 234                      (link.link_status == ETH_LINK_UP) ? "up" : "down");
 235
 236         return rte_eth_linkstatus_set(dev, &link);
 237 }
 238
 239 static int hn_dev_info_get(struct rte_eth_dev *dev,
 240                            struct rte_eth_dev_info *dev_info)
 241 {
 242         struct hn_data *hv = dev->data->dev_private;
 243         int rc;
 244
 245         dev_info->speed_capa = ETH_LINK_SPEED_10G;
 246         dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
 247         dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
 248         dev_info->max_mac_addrs  = 1;
 249
 250         dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
 251         dev_info->flow_type_rss_offloads = hv->rss_offloads;
 252         dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
 253
 254         dev_info->max_rx_queues = hv->max_queues;
 255         dev_info->max_tx_queues = hv->max_queues;
 256
 257         dev_info->tx_desc_lim.nb_min = 1;
 258         dev_info->tx_desc_lim.nb_max = 4096;
 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         if (rss_conf->rss_hf != 0) {
 379                 err = hn_rndis_conf_rss(hv, 0);
 380                 if (err) {
 381                         PMD_DRV_LOG(NOTICE,
 382                                     "rss reconfig failed (RSS disabled)");
 383                         return err;
 384                 }
 385         }
 386
 387         return hn_vf_rss_hash_update(dev, rss_conf);
 388 }
 389
 390 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
 391                                 struct rte_eth_rss_conf *rss_conf)
 392 {
 393         struct hn_data *hv = dev->data->dev_private;
 394
 395         PMD_INIT_FUNC_TRACE();
 396
 397         if (hv->ndis_ver < NDIS_VERSION_6_20) {
 398                 PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
 399                 return -EOPNOTSUPP;
 400         }
 401
 402         rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
 403         if (rss_conf->rss_key)
 404                 memcpy(rss_conf->rss_key, hv->rss_key,
 405                        NDIS_HASH_KEYSIZE_TOEPLITZ);
 406
 407         rss_conf->rss_hf = 0;
 408         if (hv->rss_hash & NDIS_HASH_IPV4)
 409                 rss_conf->rss_hf |= ETH_RSS_IPV4;
 410
 411         if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
 412                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
 413
 414         if (hv->rss_hash & NDIS_HASH_IPV6)
 415                 rss_conf->rss_hf |= ETH_RSS_IPV6;
 416
 417         if (hv->rss_hash & NDIS_HASH_IPV6_EX)
 418                 rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
 419
 420         if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
 421                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
 422
 423         if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
 424                 rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
 425
 426         return 0;
 427 }
 428
 429 static int
 430 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 431 {
 432         struct hn_data *hv = dev->data->dev_private;
 433
 434         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
 435         return hn_vf_promiscuous_enable(dev);
 436 }
 437
 438 static int
 439 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 440 {
 441         struct hn_data *hv = dev->data->dev_private;
 442         uint32_t filter;
 443
 444         filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
 445         if (dev->data->all_multicast)
 446                 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 447         hn_rndis_set_rxfilter(hv, filter);
 448         return hn_vf_promiscuous_disable(dev);
 449 }
 450
 451 static int
 452 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 453 {
 454         struct hn_data *hv = dev->data->dev_private;
 455
 456         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 457                               NDIS_PACKET_TYPE_ALL_MULTICAST |
 458                         NDIS_PACKET_TYPE_BROADCAST);
 459         return hn_vf_allmulticast_enable(dev);
 460 }
 461
 462 static int
 463 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 464 {
 465         struct hn_data *hv = dev->data->dev_private;
 466
 467         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 468                              NDIS_PACKET_TYPE_BROADCAST);
 469         return hn_vf_allmulticast_disable(dev);
 470 }
 471
 472 static int
 473 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
 474                      struct rte_ether_addr *mc_addr_set,
 475                      uint32_t nb_mc_addr)
 476 {
 477         /* No filtering on the synthetic path, but can do it on VF */
 478         return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
 479 }
 480
 481 /* Setup shared rx/tx queue data */
 482 static int hn_subchan_configure(struct hn_data *hv,
 483                                 uint32_t subchan)
 484 {
 485         struct vmbus_channel *primary = hn_primary_chan(hv);
 486         int err;
 487         unsigned int retry = 0;
 488
 489         PMD_DRV_LOG(DEBUG,
 490                     "open %u subchannels", subchan);
 491
 492         /* Send create sub channels command */
 493         err = hn_nvs_alloc_subchans(hv, &subchan);
 494         if (err)
 495                 return  err;
 496
 497         while (subchan > 0) {
 498                 struct vmbus_channel *new_sc;
 499                 uint16_t chn_index;
 500
 501                 err = rte_vmbus_subchan_open(primary, &new_sc);
 502                 if (err == -ENOENT && ++retry < 1000) {
 503                         /* This can happen if not ready yet */
 504                         rte_delay_ms(10);
 505                         continue;
 506                 }
 507
 508                 if (err) {
 509                         PMD_DRV_LOG(ERR,
 510                                     "open subchannel failed: %d", err);
 511                         return err;
 512                 }
 513
 514                 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
 515
 516                 retry = 0;
 517                 chn_index = rte_vmbus_sub_channel_index(new_sc);
 518                 if (chn_index == 0 || chn_index > hv->max_queues) {
 519                         PMD_DRV_LOG(ERR,
 520                                     "Invalid subchannel offermsg channel %u",
 521                                     chn_index);
 522                         return -EIO;
 523                 }
 524
 525                 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
 526                 hv->channels[chn_index] = new_sc;
 527                 --subchan;
 528         }
 529
 530         return err;
 531 }
 532
 533 static int hn_dev_configure(struct rte_eth_dev *dev)
 534 {
 535         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 536         struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
 537         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 538         const struct rte_eth_txmode *txmode = &dev_conf->txmode;
 539         struct hn_data *hv = dev->data->dev_private;
 540         uint64_t unsupported;
 541         int i, err, subchan;
 542
 543         PMD_INIT_FUNC_TRACE();
 544
 545         if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 546                 dev_conf->rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
 547
 548         unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
 549         if (unsupported) {
 550                 PMD_DRV_LOG(NOTICE,
 551                             "unsupported TX offload: %#" PRIx64,
 552                             unsupported);
 553                 return -EINVAL;
 554         }
 555
 556         unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
 557         if (unsupported) {
 558                 PMD_DRV_LOG(NOTICE,
 559                             "unsupported RX offload: %#" PRIx64,
 560                             rxmode->offloads);
 561                 return -EINVAL;
 562         }
 563
 564         hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 565
 566         err = hn_rndis_conf_offload(hv, txmode->offloads,
 567                                     rxmode->offloads);
 568         if (err) {
 569                 PMD_DRV_LOG(NOTICE,
 570                             "offload configure failed");
 571                 return err;
 572         }
 573
 574         hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
 575                                  dev->data->nb_tx_queues);
 576
 577         for (i = 0; i < NDIS_HASH_INDCNT; i++)
 578                 hv->rss_ind[i] = i % dev->data->nb_rx_queues;
 579
 580         hn_rss_hash_init(hv, rss_conf);
 581
 582         subchan = hv->num_queues - 1;
 583         if (subchan > 0) {
 584                 err = hn_subchan_configure(hv, subchan);
 585                 if (err) {
 586                         PMD_DRV_LOG(NOTICE,
 587                                     "subchannel configuration failed");
 588                         return err;
 589                 }
 590
 591                 err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 592                 if (err) {
 593                         PMD_DRV_LOG(NOTICE,
 594                                 "rss disable failed");
 595                         return err;
 596                 }
 597
 598                 if (rss_conf->rss_hf != 0) {
 599                         err = hn_rndis_conf_rss(hv, 0);
 600                         if (err) {
 601                                 PMD_DRV_LOG(NOTICE,
 602                                             "initial RSS config failed");
 603                                 return err;
 604                         }
 605                 }
 606         }
 607
 608         return hn_vf_configure(dev, dev_conf);
 609 }
 610
 611 static int hn_dev_stats_get(struct rte_eth_dev *dev,
 612                             struct rte_eth_stats *stats)
 613 {
 614         unsigned int i;
 615
 616         hn_vf_stats_get(dev, stats);
 617
 618         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 619                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 620
 621                 if (!txq)
 622                         continue;
 623
 624                 stats->opackets += txq->stats.packets;
 625                 stats->obytes += txq->stats.bytes;
 626                 stats->oerrors += txq->stats.errors;
 627
 628                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 629                         stats->q_opackets[i] = txq->stats.packets;
 630                         stats->q_obytes[i] = txq->stats.bytes;
 631                 }
 632         }
 633
 634         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 635                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 636
 637                 if (!rxq)
 638                         continue;
 639
 640                 stats->ipackets += rxq->stats.packets;
 641                 stats->ibytes += rxq->stats.bytes;
 642                 stats->ierrors += rxq->stats.errors;
 643                 stats->imissed += rxq->stats.ring_full;
 644
 645                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 646                         stats->q_ipackets[i] = rxq->stats.packets;
 647                         stats->q_ibytes[i] = rxq->stats.bytes;
 648                 }
 649         }
 650
 651         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 652         return 0;
 653 }
 654
 655 static int
 656 hn_dev_stats_reset(struct rte_eth_dev *dev)
 657 {
 658         unsigned int i;
 659
 660         PMD_INIT_FUNC_TRACE();
 661
 662         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 663                 struct hn_tx_queue *txq = dev->data->tx_queues[i];
 664
 665                 if (!txq)
 666                         continue;
 667                 memset(&txq->stats, 0, sizeof(struct hn_stats));
 668         }
 669
 670         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 671                 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 672
 673                 if (!rxq)
 674                         continue;
 675
 676                 memset(&rxq->stats, 0, sizeof(struct hn_stats));
 677         }
 678
 679         return 0;
 680 }
 681
 682 static int
 683 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 684 {
 685         int ret;
 686
 687         ret = hn_dev_stats_reset(dev);
 688         if (ret != 0)
 689                 return 0;
 690
 691         return hn_vf_xstats_reset(dev);
 692 }
 693
 694 static int
 695 hn_dev_xstats_count(struct rte_eth_dev *dev)
 696 {
 697         int ret, count;
 698
 699         count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
 700         count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
 701
 702         ret = hn_vf_xstats_get_names(dev, NULL, 0);
 703         if (ret < 0)
 704                 return ret;
 705
 706         return count + ret;
 707 }
 708
 709 static int
 710 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 711                         struct rte_eth_xstat_name *xstats_names,
 712                         unsigned int limit)
 713 {
 714         unsigned int i, t, count = 0;
 715         int ret;
 716
 717         if (!xstats_names)
 718                 return hn_dev_xstats_count(dev);
 719
 720         /* Note: limit checked in rte_eth_xstats_names() */
 721         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 722                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 723
 724                 if (!txq)
 725                         continue;
 726
 727                 if (count >= limit)
 728                         break;
 729
 730                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 731                         snprintf(xstats_names[count++].name,
 732                                  RTE_ETH_XSTATS_NAME_SIZE,
 733                                  "tx_q%u_%s", i, hn_stat_strings[t].name);
 734         }
 735
 736         for (i = 0; i < dev->data->nb_rx_queues; i++)  {
 737                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 738
 739                 if (!rxq)
 740                         continue;
 741
 742                 if (count >= limit)
 743                         break;
 744
 745                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 746                         snprintf(xstats_names[count++].name,
 747                                  RTE_ETH_XSTATS_NAME_SIZE,
 748                                  "rx_q%u_%s", i,
 749                                  hn_stat_strings[t].name);
 750         }
 751
 752         ret = hn_vf_xstats_get_names(dev, xstats_names + count,
 753                                      limit - count);
 754         if (ret < 0)
 755                 return ret;
 756
 757         return count + ret;
 758 }
 759
 760 static int
 761 hn_dev_xstats_get(struct rte_eth_dev *dev,
 762                   struct rte_eth_xstat *xstats,
 763                   unsigned int n)
 764 {
 765         unsigned int i, t, count = 0;
 766         const unsigned int nstats = hn_dev_xstats_count(dev);
 767         const char *stats;
 768         int ret;
 769
 770         PMD_INIT_FUNC_TRACE();
 771
 772         if (n < nstats)
 773                 return nstats;
 774
 775         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 776                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 777
 778                 if (!txq)
 779                         continue;
 780
 781                 stats = (const char *)&txq->stats;
 782                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 783                         xstats[count].id = count;
 784                         xstats[count].value = *(const uint64_t *)
 785                                 (stats + hn_stat_strings[t].offset);
 786                 }
 787         }
 788
 789         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 790                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 791
 792                 if (!rxq)
 793                         continue;
 794
 795                 stats = (const char *)&rxq->stats;
 796                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 797                         xstats[count].id = count;
 798                         xstats[count].value = *(const uint64_t *)
 799                                 (stats + hn_stat_strings[t].offset);
 800                 }
 801         }
 802
 803         ret = hn_vf_xstats_get(dev, xstats, count, n);
 804         if (ret < 0)
 805                 return ret;
 806
 807         return count + ret;
 808 }
 809
 810 static int
 811 hn_dev_start(struct rte_eth_dev *dev)
 812 {
 813         struct hn_data *hv = dev->data->dev_private;
 814         int error;
 815
 816         PMD_INIT_FUNC_TRACE();
 817
 818         error = hn_rndis_set_rxfilter(hv,
 819                                       NDIS_PACKET_TYPE_BROADCAST |
 820                                       NDIS_PACKET_TYPE_ALL_MULTICAST |
 821                                       NDIS_PACKET_TYPE_DIRECTED);
 822         if (error)
 823                 return error;
 824
 825         error = hn_vf_start(dev);
 826         if (error)
 827                 hn_rndis_set_rxfilter(hv, 0);
 828
 829         /* Initialize Link state */
 830         if (error == 0)
 831                 hn_dev_link_update(dev, 0);
 832
 833         return error;
 834 }
 835
 836 static void
 837 hn_dev_stop(struct rte_eth_dev *dev)
 838 {
 839         struct hn_data *hv = dev->data->dev_private;
 840
 841         PMD_INIT_FUNC_TRACE();
 842
 843         hn_rndis_set_rxfilter(hv, 0);
 844         hn_vf_stop(dev);
 845 }
 846
 847 static void
 848 hn_dev_close(struct rte_eth_dev *dev)
 849 {
 850         PMD_INIT_FUNC_TRACE();
 851
 852         hn_vf_close(dev);
 853         hn_dev_free_queues(dev);
 854 }
 855
 856 static const struct eth_dev_ops hn_eth_dev_ops = {
 857         .dev_configure          = hn_dev_configure,
 858         .dev_start              = hn_dev_start,
 859         .dev_stop               = hn_dev_stop,
 860         .dev_close              = hn_dev_close,
 861         .dev_infos_get          = hn_dev_info_get,
 862         .txq_info_get           = hn_dev_tx_queue_info,
 863         .rxq_info_get           = hn_dev_rx_queue_info,
 864         .dev_supported_ptypes_get = hn_vf_supported_ptypes,
 865         .promiscuous_enable     = hn_dev_promiscuous_enable,
 866         .promiscuous_disable    = hn_dev_promiscuous_disable,
 867         .allmulticast_enable    = hn_dev_allmulticast_enable,
 868         .allmulticast_disable   = hn_dev_allmulticast_disable,
 869         .set_mc_addr_list       = hn_dev_mc_addr_list,
 870         .reta_update            = hn_rss_reta_update,
 871         .reta_query             = hn_rss_reta_query,
 872         .rss_hash_update        = hn_rss_hash_update,
 873         .rss_hash_conf_get      = hn_rss_hash_conf_get,
 874         .tx_queue_setup         = hn_dev_tx_queue_setup,
 875         .tx_queue_release       = hn_dev_tx_queue_release,
 876         .tx_done_cleanup        = hn_dev_tx_done_cleanup,
 877         .tx_descriptor_status   = hn_dev_tx_descriptor_status,
 878         .rx_queue_setup         = hn_dev_rx_queue_setup,
 879         .rx_queue_release       = hn_dev_rx_queue_release,
 880         .rx_queue_count         = hn_dev_rx_queue_count,
 881         .rx_descriptor_status   = hn_dev_rx_queue_status,
 882         .link_update            = hn_dev_link_update,
 883         .stats_get              = hn_dev_stats_get,
 884         .stats_reset            = hn_dev_stats_reset,
 885         .xstats_get             = hn_dev_xstats_get,
 886         .xstats_get_names       = hn_dev_xstats_get_names,
 887         .xstats_reset           = hn_dev_xstats_reset,
 888 };
 889
 890 /*
 891  * Setup connection between PMD and kernel.
 892  */
 893 static int
 894 hn_attach(struct hn_data *hv, unsigned int mtu)
 895 {
 896         int error;
 897
 898         /* Attach NVS */
 899         error = hn_nvs_attach(hv, mtu);
 900         if (error)
 901                 goto failed_nvs;
 902
 903         /* Attach RNDIS */
 904         error = hn_rndis_attach(hv);
 905         if (error)
 906                 goto failed_rndis;
 907
 908         /*
 909          * NOTE:
 910          * Under certain conditions on certain versions of Hyper-V,
 911          * the RNDIS rxfilter is _not_ zero on the hypervisor side
 912          * after the successful RNDIS initialization.
 913          */
 914         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
 915         return 0;
 916 failed_rndis:
 917         hn_nvs_detach(hv);
 918 failed_nvs:
 919         return error;
 920 }
 921
 922 static void
 923 hn_detach(struct hn_data *hv)
 924 {
 925         hn_nvs_detach(hv);
 926         hn_rndis_detach(hv);
 927 }
 928
 929 static int
 930 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 931 {
 932         struct hn_data *hv = eth_dev->data->dev_private;
 933         struct rte_device *device = eth_dev->device;
 934         struct rte_vmbus_device *vmbus;
 935         unsigned int rxr_cnt;
 936         int err, max_chan;
 937
 938         PMD_INIT_FUNC_TRACE();
 939
 940         vmbus = container_of(device, struct rte_vmbus_device, device);
 941         eth_dev->dev_ops = &hn_eth_dev_ops;
 942         eth_dev->tx_pkt_burst = &hn_xmit_pkts;
 943         eth_dev->rx_pkt_burst = &hn_recv_pkts;
 944
 945         /*
 946          * for secondary processes, we don't initialize any further as primary
 947          * has already done this work.
 948          */
 949         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 950                 return 0;
 951
 952         /* Since Hyper-V only supports one MAC address */
 953         eth_dev->data->mac_addrs = rte_calloc("hv_mac", HN_MAX_MAC_ADDRS,
 954                                               sizeof(struct rte_ether_addr), 0);
 955         if (eth_dev->data->mac_addrs == NULL) {
 956                 PMD_INIT_LOG(ERR,
 957                              "Failed to allocate memory store MAC addresses");
 958                 return -ENOMEM;
 959         }
 960
 961         hv->vmbus = vmbus;
 962         hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
 963         hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
 964         hv->port_id = eth_dev->data->port_id;
 965         hv->latency = HN_CHAN_LATENCY_NS;
 966         hv->max_queues = 1;
 967         rte_rwlock_init(&hv->vf_lock);
 968         hv->vf_port = HN_INVALID_PORT;
 969
 970         err = hn_parse_args(eth_dev);
 971         if (err)
 972                 return err;
 973
 974         strlcpy(hv->owner.name, eth_dev->device->name,
 975                 RTE_ETH_MAX_OWNER_NAME_LEN);
 976         err = rte_eth_dev_owner_new(&hv->owner.id);
 977         if (err) {
 978                 PMD_INIT_LOG(ERR, "Can not get owner id");
 979                 return err;
 980         }
 981
 982         /* Initialize primary channel input for control operations */
 983         err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 984         if (err)
 985                 return err;
 986
 987         rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
 988
 989         hv->primary = hn_rx_queue_alloc(hv, 0,
 990                                         eth_dev->device->numa_node);
 991
 992         if (!hv->primary)
 993                 return -ENOMEM;
 994
 995         err = hn_attach(hv, RTE_ETHER_MTU);
 996         if  (err)
 997                 goto failed;
 998
 999         err = hn_chim_init(eth_dev);
1000         if (err)
1001                 goto failed;
1002
1003         err = hn_rndis_get_eaddr(hv, eth_dev->data->mac_addrs->addr_bytes);
1004         if (err)
1005                 goto failed;
1006
1007         /* Multi queue requires later versions of windows server */
1008         if (hv->nvs_ver < NVS_VERSION_5)
1009                 return 0;
1010
1011         max_chan = rte_vmbus_max_channels(vmbus);
1012         PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
1013         if (max_chan <= 0)
1014                 goto failed;
1015
1016         if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
1017                 rxr_cnt = 1;
1018
1019         hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
1020
1021         /* If VF was reported but not added, do it now */
1022         if (hv->vf_present && !hn_vf_attached(hv)) {
1023                 PMD_INIT_LOG(DEBUG, "Adding VF device");
1024
1025                 err = hn_vf_add(eth_dev, hv);
1026                 if (err)
1027                         hv->vf_present = 0;
1028         }
1029
1030         return 0;
1031
1032 failed:
1033         PMD_INIT_LOG(NOTICE, "device init failed");
1034
1035         hn_chim_uninit(eth_dev);
1036         hn_detach(hv);
1037         return err;
1038 }
1039
1040 static int
1041 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1042 {
1043         struct hn_data *hv = eth_dev->data->dev_private;
1044         int ret;
1045
1046         PMD_INIT_FUNC_TRACE();
1047
1048         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1049                 return 0;
1050
1051         hn_dev_stop(eth_dev);
1052         hn_dev_close(eth_dev);
1053
1054         eth_dev->dev_ops = NULL;
1055         eth_dev->tx_pkt_burst = NULL;
1056         eth_dev->rx_pkt_burst = NULL;
1057
1058         hn_detach(hv);
1059         hn_chim_uninit(eth_dev);
1060         rte_vmbus_chan_close(hv->primary->chan);
1061         rte_free(hv->primary);
1062         ret = rte_eth_dev_owner_delete(hv->owner.id);
1063         if (ret != 0)
1064                 return ret;
1065
1066         return 0;
1067 }
1068
1069 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1070                         struct rte_vmbus_device *dev)
1071 {
1072         struct rte_eth_dev *eth_dev;
1073         int ret;
1074
1075         PMD_INIT_FUNC_TRACE();
1076
1077         eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1078         if (!eth_dev)
1079                 return -ENOMEM;
1080
1081         ret = eth_hn_dev_init(eth_dev);
1082         if (ret)
1083                 eth_dev_vmbus_release(eth_dev);
1084         else
1085                 rte_eth_dev_probing_finish(eth_dev);
1086
1087         return ret;
1088 }
1089
1090 static int eth_hn_remove(struct rte_vmbus_device *dev)
1091 {
1092         struct rte_eth_dev *eth_dev;
1093         int ret;
1094
1095         PMD_INIT_FUNC_TRACE();
1096
1097         eth_dev = rte_eth_dev_allocated(dev->device.name);
1098         if (!eth_dev)
1099                 return -ENODEV;
1100
1101         ret = eth_hn_dev_uninit(eth_dev);
1102         if (ret)
1103                 return ret;
1104
1105         eth_dev_vmbus_release(eth_dev);
1106         return 0;
1107 }
1108
1109 /* Network device GUID */
1110 static const rte_uuid_t hn_net_ids[] = {
1111         /*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1112         RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1113         { 0 }
1114 };
1115
1116 static struct rte_vmbus_driver rte_netvsc_pmd = {
1117         .id_table = hn_net_ids,
1118         .probe = eth_hn_probe,
1119         .remove = eth_hn_remove,
1120 };
1121
1122 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1123 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1124
1125 RTE_INIT(hn_init_log)
1126 {
1127         hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
1128         if (hn_logtype_init >= 0)
1129                 rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
1130         hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
1131         if (hn_logtype_driver >= 0)
1132                 rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
1133 }