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 struct hn_xstats_name_off {
  49         char name[RTE_ETH_XSTATS_NAME_SIZE];
  50         unsigned int offset;
  51 };
  52
  53 static const struct hn_xstats_name_off hn_stat_strings[] = {
  54         { "good_packets",           offsetof(struct hn_stats, packets) },
  55         { "good_bytes",             offsetof(struct hn_stats, bytes) },
  56         { "errors",                 offsetof(struct hn_stats, errors) },
  57         { "ring full",              offsetof(struct hn_stats, ring_full) },
  58         { "channel full",           offsetof(struct hn_stats, channel_full) },
  59         { "multicast_packets",      offsetof(struct hn_stats, multicast) },
  60         { "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
  61         { "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
  62         { "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
  63         { "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
  64         { "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
  65         { "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
  66         { "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
  67         { "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
  68         { "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
  69 };
  70
  71 /* The default RSS key.
  72  * This value is the same as MLX5 so that flows will be
  73  * received on same path for both VF and synthetic NIC.
  74  */
  75 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
  76         0x2c, 0xc6, 0x81, 0xd1, 0x5b, 0xdb, 0xf4, 0xf7,
  77         0xfc, 0xa2, 0x83, 0x19, 0xdb, 0x1a, 0x3e, 0x94,
  78         0x6b, 0x9e, 0x38, 0xd9, 0x2c, 0x9c, 0x03, 0xd1,
  79         0xad, 0x99, 0x44, 0xa7, 0xd9, 0x56, 0x3d, 0x59,
  80         0x06, 0x3c, 0x25, 0xf3, 0xfc, 0x1f, 0xdc, 0x2a,
  81 };
  82
  83 static struct rte_eth_dev *
  84 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
  85 {
  86         struct rte_eth_dev *eth_dev;
  87         const char *name;
  88
  89         if (!dev)
  90                 return NULL;
  91
  92         name = dev->device.name;
  93
  94         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
  95                 eth_dev = rte_eth_dev_allocate(name);
  96                 if (!eth_dev) {
  97                         PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
  98                         return NULL;
  99                 }
 100
 101                 if (private_data_size) {
 102                         eth_dev->data->dev_private =
 103                                 rte_zmalloc_socket(name, private_data_size,
 104                                                      RTE_CACHE_LINE_SIZE, dev->device.numa_node);
 105                         if (!eth_dev->data->dev_private) {
 106                                 PMD_DRV_LOG(NOTICE, "can not allocate driver data");
 107                                 rte_eth_dev_release_port(eth_dev);
 108                                 return NULL;
 109                         }
 110                 }
 111         } else {
 112                 eth_dev = rte_eth_dev_attach_secondary(name);
 113                 if (!eth_dev) {
 114                         PMD_DRV_LOG(NOTICE, "can not attach secondary");
 115                         return NULL;
 116                 }
 117         }
 118
 119         eth_dev->device = &dev->device;
 120
 121         /* interrupt is simulated */
 122         dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
 123         eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
 124         eth_dev->intr_handle = &dev->intr_handle;
 125
 126         return eth_dev;
 127 }
 128
 129 static void
 130 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 131 {
 132         /* free ether device */
 133         rte_eth_dev_release_port(eth_dev);
 134
 135         eth_dev->device = NULL;
 136         eth_dev->intr_handle = NULL;
 137 }
 138
 139 /* handle "latency=X" from devargs */
 140 static int hn_set_latency(const char *key, const char *value, void *opaque)
 141 {
 142         struct hn_data *hv = opaque;
 143         char *endp = NULL;
 144         unsigned long lat;
 145
 146         errno = 0;
 147         lat = strtoul(value, &endp, 0);
 148
 149         if (*value == '\0' || *endp != '\0') {
 150                 PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
 151                 return -EINVAL;
 152         }
 153
 154         PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
 155
 156         hv->latency = lat * 1000;       /* usec to nsec */
 157         return 0;
 158 }
 159
 160 /* Parse device arguments */
 161 static int hn_parse_args(const struct rte_eth_dev *dev)
 162 {
 163         struct hn_data *hv = dev->data->dev_private;
 164         struct rte_devargs *devargs = dev->device->devargs;
 165         static const char * const valid_keys[] = {
 166                 "latency",
 167                 NULL
 168         };
 169         struct rte_kvargs *kvlist;
 170         int ret;
 171
 172         if (!devargs)
 173                 return 0;
 174
 175         PMD_INIT_LOG(DEBUG, "device args %s %s",
 176                      devargs->name, devargs->args);
 177
 178         kvlist = rte_kvargs_parse(devargs->args, valid_keys);
 179         if (!kvlist) {
 180                 PMD_DRV_LOG(NOTICE, "invalid parameters");
 181                 return -EINVAL;
 182         }
 183
 184         ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
 185         if (ret)
 186                 PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
 187
 188         rte_kvargs_free(kvlist);
 189         return ret;
 190 }
 191
 192 /* Update link status.
 193  * Note: the DPDK definition of "wait_to_complete"
 194  *   means block this call until link is up.
 195  *   which is not worth supporting.
 196  */
 197 int
 198 hn_dev_link_update(struct rte_eth_dev *dev,
 199                    int wait_to_complete __rte_unused)
 200 {
 201         struct hn_data *hv = dev->data->dev_private;
 202         struct rte_eth_link link, old;
 203         int error;
 204
 205         old = dev->data->dev_link;
 206
 207         error = hn_rndis_get_linkstatus(hv);
 208         if (error)
 209                 return error;
 210
 211         hn_rndis_get_linkspeed(hv);
 212
 213         link = (struct rte_eth_link) {
 214                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 215                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 216                 .link_speed = hv->link_speed / 10000,
 217         };
 218
 219         if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
 220                 link.link_status = ETH_LINK_UP;
 221         else
 222                 link.link_status = ETH_LINK_DOWN;
 223
 224         if (old.link_status == link.link_status)
 225                 return 0;
 226
 227         PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
 228                      (link.link_status == ETH_LINK_UP) ? "up" : "down");
 229
 230         return rte_eth_linkstatus_set(dev, &link);
 231 }
 232
 233 static int hn_dev_info_get(struct rte_eth_dev *dev,
 234                            struct rte_eth_dev_info *dev_info)
 235 {
 236         struct hn_data *hv = dev->data->dev_private;
 237         int rc;
 238
 239         dev_info->speed_capa = ETH_LINK_SPEED_10G;
 240         dev_info->min_rx_bufsize = HN_MIN_RX_BUF_SIZE;
 241         dev_info->max_rx_pktlen  = HN_MAX_XFER_LEN;
 242         dev_info->max_mac_addrs  = 1;
 243
 244         dev_info->hash_key_size = NDIS_HASH_KEYSIZE_TOEPLITZ;
 245         dev_info->flow_type_rss_offloads = hv->rss_offloads;
 246         dev_info->reta_size = ETH_RSS_RETA_SIZE_128;
 247
 248         dev_info->max_rx_queues = hv->max_queues;
 249         dev_info->max_tx_queues = hv->max_queues;
 250
 251         dev_info->tx_desc_lim.nb_min = 1;
 252         dev_info->tx_desc_lim.nb_max = 4096;
 253
 254         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 255                 return 0;
 256
 257         /* fills in rx and tx offload capability */
 258         rc = hn_rndis_get_offload(hv, dev_info);
 259         if (rc != 0)
 260                 return rc;
 261
 262         /* merges the offload and queues of vf */
 263         return hn_vf_info_get(hv, dev_info);
 264 }
 265
 266 static int hn_rss_reta_update(struct rte_eth_dev *dev,
 267                               struct rte_eth_rss_reta_entry64 *reta_conf,
 268                               uint16_t reta_size)
 269 {
 270         struct hn_data *hv = dev->data->dev_private;
 271         unsigned int i;
 272         int err;
 273
 274         PMD_INIT_FUNC_TRACE();
 275
 276         if (reta_size != NDIS_HASH_INDCNT) {
 277                 PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
 278                 return -EINVAL;
 279         }
 280
 281         for (i = 0; i < NDIS_HASH_INDCNT; i++) {
 282                 uint16_t idx = i / RTE_RETA_GROUP_SIZE;
 283                 uint16_t shift = i % RTE_RETA_GROUP_SIZE;
 284                 uint64_t mask = (uint64_t)1 << shift;
 285
 286                 if (reta_conf[idx].mask & mask)
 287                         hv->rss_ind[i] = reta_conf[idx].reta[shift];
 288         }
 289
 290         err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 291         if (err) {
 292                 PMD_DRV_LOG(NOTICE,
 293                         "rss disable failed");
 294                 return err;
 295         }
 296
 297         err = hn_rndis_conf_rss(hv, 0);
 298         if (err) {
 299                 PMD_DRV_LOG(NOTICE,
 300                             "reta reconfig failed");
 301                 return err;
 302         }
 303
 304         return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
 305 }
 306
 307 static int hn_rss_reta_query(struct rte_eth_dev *dev,
 308                              struct rte_eth_rss_reta_entry64 *reta_conf,
 309                              uint16_t reta_size)
 310 {
 311         struct hn_data *hv = dev->data->dev_private;
 312         unsigned int i;
 313
 314         PMD_INIT_FUNC_TRACE();
 315
 316         if (reta_size != NDIS_HASH_INDCNT) {
 317                 PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
 318                 return -EINVAL;
 319         }
 320
 321         for (i = 0; i < NDIS_HASH_INDCNT; i++) {
 322                 uint16_t idx = i / RTE_RETA_GROUP_SIZE;
 323                 uint16_t shift = i % RTE_RETA_GROUP_SIZE;
 324                 uint64_t mask = (uint64_t)1 << shift;
 325
 326                 if (reta_conf[idx].mask & mask)
 327                         reta_conf[idx].reta[shift] = hv->rss_ind[i];
 328         }
 329         return 0;
 330 }
 331
 332 static void hn_rss_hash_init(struct hn_data *hv,
 333                              const struct rte_eth_rss_conf *rss_conf)
 334 {
 335         /* Convert from DPDK RSS hash flags to NDIS hash flags */
 336         hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
 337
 338         if (rss_conf->rss_hf & ETH_RSS_IPV4)
 339                 hv->rss_hash |= NDIS_HASH_IPV4;
 340         if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
 341                 hv->rss_hash |= NDIS_HASH_TCP_IPV4;
 342         if (rss_conf->rss_hf & ETH_RSS_IPV6)
 343                 hv->rss_hash |=  NDIS_HASH_IPV6;
 344         if (rss_conf->rss_hf & ETH_RSS_IPV6_EX)
 345                 hv->rss_hash |=  NDIS_HASH_IPV6_EX;
 346         if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
 347                 hv->rss_hash |= NDIS_HASH_TCP_IPV6;
 348         if (rss_conf->rss_hf & ETH_RSS_IPV6_TCP_EX)
 349                 hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
 350
 351         memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
 352                NDIS_HASH_KEYSIZE_TOEPLITZ);
 353 }
 354
 355 static int hn_rss_hash_update(struct rte_eth_dev *dev,
 356                               struct rte_eth_rss_conf *rss_conf)
 357 {
 358         struct hn_data *hv = dev->data->dev_private;
 359         int err;
 360
 361         PMD_INIT_FUNC_TRACE();
 362
 363         err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 364         if (err) {
 365                 PMD_DRV_LOG(NOTICE,
 366                             "rss disable failed");
 367                 return err;
 368         }
 369
 370         hn_rss_hash_init(hv, rss_conf);
 371
 372         if (rss_conf->rss_hf != 0) {
 373                 err = hn_rndis_conf_rss(hv, 0);
 374                 if (err) {
 375                         PMD_DRV_LOG(NOTICE,
 376                                     "rss reconfig failed (RSS disabled)");
 377                         return err;
 378                 }
 379         }
 380
 381         return hn_vf_rss_hash_update(dev, rss_conf);
 382 }
 383
 384 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
 385                                 struct rte_eth_rss_conf *rss_conf)
 386 {
 387         struct hn_data *hv = dev->data->dev_private;
 388
 389         PMD_INIT_FUNC_TRACE();
 390
 391         if (hv->ndis_ver < NDIS_VERSION_6_20) {
 392                 PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
 393                 return -EOPNOTSUPP;
 394         }
 395
 396         rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
 397         if (rss_conf->rss_key)
 398                 memcpy(rss_conf->rss_key, hv->rss_key,
 399                        NDIS_HASH_KEYSIZE_TOEPLITZ);
 400
 401         rss_conf->rss_hf = 0;
 402         if (hv->rss_hash & NDIS_HASH_IPV4)
 403                 rss_conf->rss_hf |= ETH_RSS_IPV4;
 404
 405         if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
 406                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
 407
 408         if (hv->rss_hash & NDIS_HASH_IPV6)
 409                 rss_conf->rss_hf |= ETH_RSS_IPV6;
 410
 411         if (hv->rss_hash & NDIS_HASH_IPV6_EX)
 412                 rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
 413
 414         if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
 415                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
 416
 417         if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
 418                 rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
 419
 420         return 0;
 421 }
 422
 423 static int
 424 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 425 {
 426         struct hn_data *hv = dev->data->dev_private;
 427
 428         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
 429         return hn_vf_promiscuous_enable(dev);
 430 }
 431
 432 static int
 433 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 434 {
 435         struct hn_data *hv = dev->data->dev_private;
 436         uint32_t filter;
 437
 438         filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
 439         if (dev->data->all_multicast)
 440                 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 441         hn_rndis_set_rxfilter(hv, filter);
 442         return hn_vf_promiscuous_disable(dev);
 443 }
 444
 445 static int
 446 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 447 {
 448         struct hn_data *hv = dev->data->dev_private;
 449
 450         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 451                               NDIS_PACKET_TYPE_ALL_MULTICAST |
 452                         NDIS_PACKET_TYPE_BROADCAST);
 453         return hn_vf_allmulticast_enable(dev);
 454 }
 455
 456 static int
 457 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 458 {
 459         struct hn_data *hv = dev->data->dev_private;
 460
 461         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 462                              NDIS_PACKET_TYPE_BROADCAST);
 463         return hn_vf_allmulticast_disable(dev);
 464 }
 465
 466 static int
 467 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
 468                      struct rte_ether_addr *mc_addr_set,
 469                      uint32_t nb_mc_addr)
 470 {
 471         /* No filtering on the synthetic path, but can do it on VF */
 472         return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
 473 }
 474
 475 /* Setup shared rx/tx queue data */
 476 static int hn_subchan_configure(struct hn_data *hv,
 477                                 uint32_t subchan)
 478 {
 479         struct vmbus_channel *primary = hn_primary_chan(hv);
 480         int err;
 481         unsigned int retry = 0;
 482
 483         PMD_DRV_LOG(DEBUG,
 484                     "open %u subchannels", subchan);
 485
 486         /* Send create sub channels command */
 487         err = hn_nvs_alloc_subchans(hv, &subchan);
 488         if (err)
 489                 return  err;
 490
 491         while (subchan > 0) {
 492                 struct vmbus_channel *new_sc;
 493                 uint16_t chn_index;
 494
 495                 err = rte_vmbus_subchan_open(primary, &new_sc);
 496                 if (err == -ENOENT && ++retry < 1000) {
 497                         /* This can happen if not ready yet */
 498                         rte_delay_ms(10);
 499                         continue;
 500                 }
 501
 502                 if (err) {
 503                         PMD_DRV_LOG(ERR,
 504                                     "open subchannel failed: %d", err);
 505                         return err;
 506                 }
 507
 508                 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
 509
 510                 retry = 0;
 511                 chn_index = rte_vmbus_sub_channel_index(new_sc);
 512                 if (chn_index == 0 || chn_index > hv->max_queues) {
 513                         PMD_DRV_LOG(ERR,
 514                                     "Invalid subchannel offermsg channel %u",
 515                                     chn_index);
 516                         return -EIO;
 517                 }
 518
 519                 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
 520                 hv->channels[chn_index] = new_sc;
 521                 --subchan;
 522         }
 523
 524         return err;
 525 }
 526
 527 static int hn_dev_configure(struct rte_eth_dev *dev)
 528 {
 529         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 530         struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
 531         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 532         const struct rte_eth_txmode *txmode = &dev_conf->txmode;
 533         struct hn_data *hv = dev->data->dev_private;
 534         uint64_t unsupported;
 535         int i, err, subchan;
 536
 537         PMD_INIT_FUNC_TRACE();
 538
 539         if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 540                 dev_conf->rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
 541
 542         unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
 543         if (unsupported) {
 544                 PMD_DRV_LOG(NOTICE,
 545                             "unsupported TX offload: %#" PRIx64,
 546                             unsupported);
 547                 return -EINVAL;
 548         }
 549
 550         unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
 551         if (unsupported) {
 552                 PMD_DRV_LOG(NOTICE,
 553                             "unsupported RX offload: %#" PRIx64,
 554                             rxmode->offloads);
 555                 return -EINVAL;
 556         }
 557
 558         hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 559
 560         err = hn_rndis_conf_offload(hv, txmode->offloads,
 561                                     rxmode->offloads);
 562         if (err) {
 563                 PMD_DRV_LOG(NOTICE,
 564                             "offload configure failed");
 565                 return err;
 566         }
 567
 568         hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
 569                                  dev->data->nb_tx_queues);
 570
 571         for (i = 0; i < NDIS_HASH_INDCNT; i++)
 572                 hv->rss_ind[i] = i % dev->data->nb_rx_queues;
 573
 574         hn_rss_hash_init(hv, rss_conf);
 575
 576         subchan = hv->num_queues - 1;
 577         if (subchan > 0) {
 578                 err = hn_subchan_configure(hv, subchan);
 579                 if (err) {
 580                         PMD_DRV_LOG(NOTICE,
 581                                     "subchannel configuration failed");
 582                         return err;
 583                 }
 584
 585                 err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 586                 if (err) {
 587                         PMD_DRV_LOG(NOTICE,
 588                                 "rss disable failed");
 589                         return err;
 590                 }
 591
 592                 if (rss_conf->rss_hf != 0) {
 593                         err = hn_rndis_conf_rss(hv, 0);
 594                         if (err) {
 595                                 PMD_DRV_LOG(NOTICE,
 596                                             "initial RSS config failed");
 597                                 return err;
 598                         }
 599                 }
 600         }
 601
 602         return hn_vf_configure(dev, dev_conf);
 603 }
 604
 605 static int hn_dev_stats_get(struct rte_eth_dev *dev,
 606                             struct rte_eth_stats *stats)
 607 {
 608         unsigned int i;
 609
 610         hn_vf_stats_get(dev, stats);
 611
 612         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 613                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 614
 615                 if (!txq)
 616                         continue;
 617
 618                 stats->opackets += txq->stats.packets;
 619                 stats->obytes += txq->stats.bytes;
 620                 stats->oerrors += txq->stats.errors;
 621
 622                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 623                         stats->q_opackets[i] = txq->stats.packets;
 624                         stats->q_obytes[i] = txq->stats.bytes;
 625                 }
 626         }
 627
 628         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 629                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 630
 631                 if (!rxq)
 632                         continue;
 633
 634                 stats->ipackets += rxq->stats.packets;
 635                 stats->ibytes += rxq->stats.bytes;
 636                 stats->ierrors += rxq->stats.errors;
 637                 stats->imissed += rxq->stats.ring_full;
 638
 639                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 640                         stats->q_ipackets[i] = rxq->stats.packets;
 641                         stats->q_ibytes[i] = rxq->stats.bytes;
 642                 }
 643         }
 644
 645         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 646         return 0;
 647 }
 648
 649 static int
 650 hn_dev_stats_reset(struct rte_eth_dev *dev)
 651 {
 652         unsigned int i;
 653
 654         PMD_INIT_FUNC_TRACE();
 655
 656         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 657                 struct hn_tx_queue *txq = dev->data->tx_queues[i];
 658
 659                 if (!txq)
 660                         continue;
 661                 memset(&txq->stats, 0, sizeof(struct hn_stats));
 662         }
 663
 664         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 665                 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 666
 667                 if (!rxq)
 668                         continue;
 669
 670                 memset(&rxq->stats, 0, sizeof(struct hn_stats));
 671         }
 672
 673         return 0;
 674 }
 675
 676 static int
 677 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 678 {
 679         int ret;
 680
 681         ret = hn_dev_stats_reset(dev);
 682         if (ret != 0)
 683                 return 0;
 684
 685         return hn_vf_xstats_reset(dev);
 686 }
 687
 688 static int
 689 hn_dev_xstats_count(struct rte_eth_dev *dev)
 690 {
 691         int ret, count;
 692
 693         count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
 694         count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
 695
 696         ret = hn_vf_xstats_get_names(dev, NULL, 0);
 697         if (ret < 0)
 698                 return ret;
 699
 700         return count + ret;
 701 }
 702
 703 static int
 704 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 705                         struct rte_eth_xstat_name *xstats_names,
 706                         unsigned int limit)
 707 {
 708         unsigned int i, t, count = 0;
 709         int ret;
 710
 711         if (!xstats_names)
 712                 return hn_dev_xstats_count(dev);
 713
 714         /* Note: limit checked in rte_eth_xstats_names() */
 715         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 716                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 717
 718                 if (!txq)
 719                         continue;
 720
 721                 if (count >= limit)
 722                         break;
 723
 724                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 725                         snprintf(xstats_names[count++].name,
 726                                  RTE_ETH_XSTATS_NAME_SIZE,
 727                                  "tx_q%u_%s", i, hn_stat_strings[t].name);
 728         }
 729
 730         for (i = 0; i < dev->data->nb_rx_queues; i++)  {
 731                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 732
 733                 if (!rxq)
 734                         continue;
 735
 736                 if (count >= limit)
 737                         break;
 738
 739                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 740                         snprintf(xstats_names[count++].name,
 741                                  RTE_ETH_XSTATS_NAME_SIZE,
 742                                  "rx_q%u_%s", i,
 743                                  hn_stat_strings[t].name);
 744         }
 745
 746         ret = hn_vf_xstats_get_names(dev, xstats_names + count,
 747                                      limit - count);
 748         if (ret < 0)
 749                 return ret;
 750
 751         return count + ret;
 752 }
 753
 754 static int
 755 hn_dev_xstats_get(struct rte_eth_dev *dev,
 756                   struct rte_eth_xstat *xstats,
 757                   unsigned int n)
 758 {
 759         unsigned int i, t, count = 0;
 760         const unsigned int nstats = hn_dev_xstats_count(dev);
 761         const char *stats;
 762         int ret;
 763
 764         PMD_INIT_FUNC_TRACE();
 765
 766         if (n < nstats)
 767                 return nstats;
 768
 769         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 770                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 771
 772                 if (!txq)
 773                         continue;
 774
 775                 stats = (const char *)&txq->stats;
 776                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 777                         xstats[count].id = count;
 778                         xstats[count].value = *(const uint64_t *)
 779                                 (stats + hn_stat_strings[t].offset);
 780                 }
 781         }
 782
 783         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 784                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 785
 786                 if (!rxq)
 787                         continue;
 788
 789                 stats = (const char *)&rxq->stats;
 790                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 791                         xstats[count].id = count;
 792                         xstats[count].value = *(const uint64_t *)
 793                                 (stats + hn_stat_strings[t].offset);
 794                 }
 795         }
 796
 797         ret = hn_vf_xstats_get(dev, xstats, count, n);
 798         if (ret < 0)
 799                 return ret;
 800
 801         return count + ret;
 802 }
 803
 804 static int
 805 hn_dev_start(struct rte_eth_dev *dev)
 806 {
 807         struct hn_data *hv = dev->data->dev_private;
 808         int error;
 809
 810         PMD_INIT_FUNC_TRACE();
 811
 812         error = hn_rndis_set_rxfilter(hv,
 813                                       NDIS_PACKET_TYPE_BROADCAST |
 814                                       NDIS_PACKET_TYPE_ALL_MULTICAST |
 815                                       NDIS_PACKET_TYPE_DIRECTED);
 816         if (error)
 817                 return error;
 818
 819         error = hn_vf_start(dev);
 820         if (error)
 821                 hn_rndis_set_rxfilter(hv, 0);
 822
 823         /* Initialize Link state */
 824         if (error == 0)
 825                 hn_dev_link_update(dev, 0);
 826
 827         return error;
 828 }
 829
 830 static void
 831 hn_dev_stop(struct rte_eth_dev *dev)
 832 {
 833         struct hn_data *hv = dev->data->dev_private;
 834
 835         PMD_INIT_FUNC_TRACE();
 836         dev->data->dev_started = 0;
 837
 838         hn_rndis_set_rxfilter(hv, 0);
 839         hn_vf_stop(dev);
 840 }
 841
 842 static int
 843 hn_dev_close(struct rte_eth_dev *dev)
 844 {
 845         PMD_INIT_FUNC_TRACE();
 846         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 847                 return 0;
 848
 849         hn_vf_close(dev);
 850         hn_dev_free_queues(dev);
 851
 852         return 0;
 853 }
 854
 855 static const struct eth_dev_ops hn_eth_dev_ops = {
 856         .dev_configure          = hn_dev_configure,
 857         .dev_start              = hn_dev_start,
 858         .dev_stop               = hn_dev_stop,
 859         .dev_close              = hn_dev_close,
 860         .dev_infos_get          = hn_dev_info_get,
 861         .txq_info_get           = hn_dev_tx_queue_info,
 862         .rxq_info_get           = hn_dev_rx_queue_info,
 863         .dev_supported_ptypes_get = hn_vf_supported_ptypes,
 864         .promiscuous_enable     = hn_dev_promiscuous_enable,
 865         .promiscuous_disable    = hn_dev_promiscuous_disable,
 866         .allmulticast_enable    = hn_dev_allmulticast_enable,
 867         .allmulticast_disable   = hn_dev_allmulticast_disable,
 868         .set_mc_addr_list       = hn_dev_mc_addr_list,
 869         .reta_update            = hn_rss_reta_update,
 870         .reta_query             = hn_rss_reta_query,
 871         .rss_hash_update        = hn_rss_hash_update,
 872         .rss_hash_conf_get      = hn_rss_hash_conf_get,
 873         .tx_queue_setup         = hn_dev_tx_queue_setup,
 874         .tx_queue_release       = hn_dev_tx_queue_release,
 875         .tx_done_cleanup        = hn_dev_tx_done_cleanup,
 876         .rx_queue_setup         = hn_dev_rx_queue_setup,
 877         .rx_queue_release       = hn_dev_rx_queue_release,
 878         .link_update            = hn_dev_link_update,
 879         .stats_get              = hn_dev_stats_get,
 880         .stats_reset            = hn_dev_stats_reset,
 881         .xstats_get             = hn_dev_xstats_get,
 882         .xstats_get_names       = hn_dev_xstats_get_names,
 883         .xstats_reset           = hn_dev_xstats_reset,
 884 };
 885
 886 /*
 887  * Setup connection between PMD and kernel.
 888  */
 889 static int
 890 hn_attach(struct hn_data *hv, unsigned int mtu)
 891 {
 892         int error;
 893
 894         /* Attach NVS */
 895         error = hn_nvs_attach(hv, mtu);
 896         if (error)
 897                 goto failed_nvs;
 898
 899         /* Attach RNDIS */
 900         error = hn_rndis_attach(hv);
 901         if (error)
 902                 goto failed_rndis;
 903
 904         /*
 905          * NOTE:
 906          * Under certain conditions on certain versions of Hyper-V,
 907          * the RNDIS rxfilter is _not_ zero on the hypervisor side
 908          * after the successful RNDIS initialization.
 909          */
 910         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_NONE);
 911         return 0;
 912 failed_rndis:
 913         hn_nvs_detach(hv);
 914 failed_nvs:
 915         return error;
 916 }
 917
 918 static void
 919 hn_detach(struct hn_data *hv)
 920 {
 921         hn_nvs_detach(hv);
 922         hn_rndis_detach(hv);
 923 }
 924
 925 static int
 926 eth_hn_dev_init(struct rte_eth_dev *eth_dev)
 927 {
 928         struct hn_data *hv = eth_dev->data->dev_private;
 929         struct rte_device *device = eth_dev->device;
 930         struct rte_vmbus_device *vmbus;
 931         unsigned int rxr_cnt;
 932         int err, max_chan;
 933
 934         PMD_INIT_FUNC_TRACE();
 935
 936         vmbus = container_of(device, struct rte_vmbus_device, device);
 937         eth_dev->dev_ops = &hn_eth_dev_ops;
 938         eth_dev->rx_queue_count = hn_dev_rx_queue_count;
 939         eth_dev->rx_descriptor_status = hn_dev_rx_queue_status;
 940         eth_dev->tx_descriptor_status = hn_dev_tx_descriptor_status;
 941         eth_dev->tx_pkt_burst = &hn_xmit_pkts;
 942         eth_dev->rx_pkt_burst = &hn_recv_pkts;
 943
 944         /*
 945          * for secondary processes, we don't initialize any further as primary
 946          * has already done this work.
 947          */
 948         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 949                 return 0;
 950
 951         /* Since Hyper-V only supports one MAC address */
 952         eth_dev->data->mac_addrs = rte_calloc("hv_mac", HN_MAX_MAC_ADDRS,
 953                                               sizeof(struct rte_ether_addr), 0);
 954         if (eth_dev->data->mac_addrs == NULL) {
 955                 PMD_INIT_LOG(ERR,
 956                              "Failed to allocate memory store MAC addresses");
 957                 return -ENOMEM;
 958         }
 959
 960         hv->vmbus = vmbus;
 961         hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
 962         hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
 963         hv->port_id = eth_dev->data->port_id;
 964         hv->latency = HN_CHAN_LATENCY_NS;
 965         hv->max_queues = 1;
 966         rte_rwlock_init(&hv->vf_lock);
 967         hv->vf_port = HN_INVALID_PORT;
 968
 969         err = hn_parse_args(eth_dev);
 970         if (err)
 971                 return err;
 972
 973         strlcpy(hv->owner.name, eth_dev->device->name,
 974                 RTE_ETH_MAX_OWNER_NAME_LEN);
 975         err = rte_eth_dev_owner_new(&hv->owner.id);
 976         if (err) {
 977                 PMD_INIT_LOG(ERR, "Can not get owner id");
 978                 return err;
 979         }
 980
 981         /* Initialize primary channel input for control operations */
 982         err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 983         if (err)
 984                 return err;
 985
 986         rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
 987
 988         hv->primary = hn_rx_queue_alloc(hv, 0,
 989                                         eth_dev->device->numa_node);
 990
 991         if (!hv->primary)
 992                 return -ENOMEM;
 993
 994         err = hn_attach(hv, RTE_ETHER_MTU);
 995         if  (err)
 996                 goto failed;
 997
 998         err = hn_chim_init(eth_dev);
 999         if (err)
1000                 goto failed;
1001
1002         err = hn_rndis_get_eaddr(hv, eth_dev->data->mac_addrs->addr_bytes);
1003         if (err)
1004                 goto failed;
1005
1006         /* Multi queue requires later versions of windows server */
1007         if (hv->nvs_ver < NVS_VERSION_5)
1008                 return 0;
1009
1010         max_chan = rte_vmbus_max_channels(vmbus);
1011         PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
1012         if (max_chan <= 0)
1013                 goto failed;
1014
1015         if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
1016                 rxr_cnt = 1;
1017
1018         hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
1019
1020         /* If VF was reported but not added, do it now */
1021         if (hv->vf_present && !hn_vf_attached(hv)) {
1022                 PMD_INIT_LOG(DEBUG, "Adding VF device");
1023
1024                 err = hn_vf_add(eth_dev, hv);
1025                 if (err)
1026                         hv->vf_present = 0;
1027         }
1028
1029         return 0;
1030
1031 failed:
1032         PMD_INIT_LOG(NOTICE, "device init failed");
1033
1034         hn_chim_uninit(eth_dev);
1035         hn_detach(hv);
1036         return err;
1037 }
1038
1039 static int
1040 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1041 {
1042         struct hn_data *hv = eth_dev->data->dev_private;
1043         int ret;
1044
1045         PMD_INIT_FUNC_TRACE();
1046
1047         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1048                 return 0;
1049
1050         hn_dev_stop(eth_dev);
1051         hn_dev_close(eth_dev);
1052
1053         hn_detach(hv);
1054         hn_chim_uninit(eth_dev);
1055         rte_vmbus_chan_close(hv->primary->chan);
1056         rte_free(hv->primary);
1057         ret = rte_eth_dev_owner_delete(hv->owner.id);
1058         if (ret != 0)
1059                 return ret;
1060
1061         return 0;
1062 }
1063
1064 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1065                         struct rte_vmbus_device *dev)
1066 {
1067         struct rte_eth_dev *eth_dev;
1068         int ret;
1069
1070         PMD_INIT_FUNC_TRACE();
1071
1072         eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1073         if (!eth_dev)
1074                 return -ENOMEM;
1075
1076         ret = eth_hn_dev_init(eth_dev);
1077         if (ret)
1078                 eth_dev_vmbus_release(eth_dev);
1079         else
1080                 rte_eth_dev_probing_finish(eth_dev);
1081
1082         return ret;
1083 }
1084
1085 static int eth_hn_remove(struct rte_vmbus_device *dev)
1086 {
1087         struct rte_eth_dev *eth_dev;
1088         int ret;
1089
1090         PMD_INIT_FUNC_TRACE();
1091
1092         eth_dev = rte_eth_dev_allocated(dev->device.name);
1093         if (!eth_dev)
1094                 return 0; /* port already released */
1095
1096         ret = eth_hn_dev_uninit(eth_dev);
1097         if (ret)
1098                 return ret;
1099
1100         eth_dev_vmbus_release(eth_dev);
1101         return 0;
1102 }
1103
1104 /* Network device GUID */
1105 static const rte_uuid_t hn_net_ids[] = {
1106         /*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1107         RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1108         { 0 }
1109 };
1110
1111 static struct rte_vmbus_driver rte_netvsc_pmd = {
1112         .id_table = hn_net_ids,
1113         .probe = eth_hn_probe,
1114         .remove = eth_hn_remove,
1115 };
1116
1117 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1118 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1119 RTE_LOG_REGISTER(hn_logtype_init, pmd.net.netvsc.init, NOTICE);
1120 RTE_LOG_REGISTER(hn_logtype_driver, pmd.net.netvsc.driver, NOTICE);