drivers/net/netvsc/hn_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2018 Microsoft Corporation
   3  * Copyright(c) 2013-2016 Brocade Communications Systems, Inc.
   4  * All rights reserved.
   5  */
   6
   7 #include <stdint.h>
   8 #include <string.h>
   9 #include <stdio.h>
  10 #include <errno.h>
  11 #include <unistd.h>
  12
  13 #include <rte_ethdev.h>
  14 #include <rte_memcpy.h>
  15 #include <rte_string_fns.h>
  16 #include <rte_memzone.h>
  17 #include <rte_devargs.h>
  18 #include <rte_malloc.h>
  19 #include <rte_kvargs.h>
  20 #include <rte_atomic.h>
  21 #include <rte_branch_prediction.h>
  22 #include <rte_ether.h>
  23 #include <rte_ethdev_driver.h>
  24 #include <rte_cycles.h>
  25 #include <rte_errno.h>
  26 #include <rte_memory.h>
  27 #include <rte_eal.h>
  28 #include <rte_dev.h>
  29 #include <rte_bus_vmbus.h>
  30
  31 #include "hn_logs.h"
  32 #include "hn_var.h"
  33 #include "hn_rndis.h"
  34 #include "hn_nvs.h"
  35 #include "ndis.h"
  36
  37 #define HN_TX_OFFLOAD_CAPS (DEV_TX_OFFLOAD_IPV4_CKSUM | \
  38                             DEV_TX_OFFLOAD_TCP_CKSUM  | \
  39                             DEV_TX_OFFLOAD_UDP_CKSUM  | \
  40                             DEV_TX_OFFLOAD_TCP_TSO    | \
  41                             DEV_TX_OFFLOAD_MULTI_SEGS | \
  42                             DEV_TX_OFFLOAD_VLAN_INSERT)
  43
  44 #define HN_RX_OFFLOAD_CAPS (DEV_RX_OFFLOAD_CHECKSUM | \
  45                             DEV_RX_OFFLOAD_VLAN_STRIP | \
  46                             DEV_RX_OFFLOAD_RSS_HASH)
  47
  48 int hn_logtype_init;
  49 int hn_logtype_driver;
  50
  51 struct hn_xstats_name_off {
  52         char name[RTE_ETH_XSTATS_NAME_SIZE];
  53         unsigned int offset;
  54 };
  55
  56 static const struct hn_xstats_name_off hn_stat_strings[] = {
  57         { "good_packets",           offsetof(struct hn_stats, packets) },
  58         { "good_bytes",             offsetof(struct hn_stats, bytes) },
  59         { "errors",                 offsetof(struct hn_stats, errors) },
  60         { "ring full",              offsetof(struct hn_stats, ring_full) },
  61         { "multicast_packets",      offsetof(struct hn_stats, multicast) },
  62         { "broadcast_packets",      offsetof(struct hn_stats, broadcast) },
  63         { "undersize_packets",      offsetof(struct hn_stats, size_bins[0]) },
  64         { "size_64_packets",        offsetof(struct hn_stats, size_bins[1]) },
  65         { "size_65_127_packets",    offsetof(struct hn_stats, size_bins[2]) },
  66         { "size_128_255_packets",   offsetof(struct hn_stats, size_bins[3]) },
  67         { "size_256_511_packets",   offsetof(struct hn_stats, size_bins[4]) },
  68         { "size_512_1023_packets",  offsetof(struct hn_stats, size_bins[5]) },
  69         { "size_1024_1518_packets", offsetof(struct hn_stats, size_bins[6]) },
  70         { "size_1519_max_packets",  offsetof(struct hn_stats, size_bins[7]) },
  71 };
  72
  73 /* The default RSS key.
  74  * This value is the same as MLX5 so that flows will be
  75  * received on same path for both VF and synthetic NIC.
  76  */
  77 static const uint8_t rss_default_key[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
  78         0x2c, 0xc6, 0x81, 0xd1, 0x5b, 0xdb, 0xf4, 0xf7,
  79         0xfc, 0xa2, 0x83, 0x19, 0xdb, 0x1a, 0x3e, 0x94,
  80         0x6b, 0x9e, 0x38, 0xd9, 0x2c, 0x9c, 0x03, 0xd1,
  81         0xad, 0x99, 0x44, 0xa7, 0xd9, 0x56, 0x3d, 0x59,
  82         0x06, 0x3c, 0x25, 0xf3, 0xfc, 0x1f, 0xdc, 0x2a,
  83 };
  84
  85 static struct rte_eth_dev *
  86 eth_dev_vmbus_allocate(struct rte_vmbus_device *dev, size_t private_data_size)
  87 {
  88         struct rte_eth_dev *eth_dev;
  89         const char *name;
  90
  91         if (!dev)
  92                 return NULL;
  93
  94         name = dev->device.name;
  95
  96         if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
  97                 eth_dev = rte_eth_dev_allocate(name);
  98                 if (!eth_dev) {
  99                         PMD_DRV_LOG(NOTICE, "can not allocate rte ethdev");
 100                         return NULL;
 101                 }
 102
 103                 if (private_data_size) {
 104                         eth_dev->data->dev_private =
 105                                 rte_zmalloc_socket(name, private_data_size,
 106                                                      RTE_CACHE_LINE_SIZE, dev->device.numa_node);
 107                         if (!eth_dev->data->dev_private) {
 108                                 PMD_DRV_LOG(NOTICE, "can not allocate driver data");
 109                                 rte_eth_dev_release_port(eth_dev);
 110                                 return NULL;
 111                         }
 112                 }
 113         } else {
 114                 eth_dev = rte_eth_dev_attach_secondary(name);
 115                 if (!eth_dev) {
 116                         PMD_DRV_LOG(NOTICE, "can not attach secondary");
 117                         return NULL;
 118                 }
 119         }
 120
 121         eth_dev->device = &dev->device;
 122
 123         /* interrupt is simulated */
 124         dev->intr_handle.type = RTE_INTR_HANDLE_EXT;
 125         eth_dev->data->dev_flags |= RTE_ETH_DEV_INTR_LSC;
 126         eth_dev->intr_handle = &dev->intr_handle;
 127
 128         /* allow ethdev to remove on close */
 129         eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
 130
 131         return eth_dev;
 132 }
 133
 134 static void
 135 eth_dev_vmbus_release(struct rte_eth_dev *eth_dev)
 136 {
 137         /* free ether device */
 138         rte_eth_dev_release_port(eth_dev);
 139
 140         eth_dev->device = NULL;
 141         eth_dev->intr_handle = NULL;
 142 }
 143
 144 /* handle "latency=X" from devargs */
 145 static int hn_set_latency(const char *key, const char *value, void *opaque)
 146 {
 147         struct hn_data *hv = opaque;
 148         char *endp = NULL;
 149         unsigned long lat;
 150
 151         errno = 0;
 152         lat = strtoul(value, &endp, 0);
 153
 154         if (*value == '\0' || *endp != '\0') {
 155                 PMD_DRV_LOG(ERR, "invalid parameter %s=%s", key, value);
 156                 return -EINVAL;
 157         }
 158
 159         PMD_DRV_LOG(DEBUG, "set latency %lu usec", lat);
 160
 161         hv->latency = lat * 1000;       /* usec to nsec */
 162         return 0;
 163 }
 164
 165 /* Parse device arguments */
 166 static int hn_parse_args(const struct rte_eth_dev *dev)
 167 {
 168         struct hn_data *hv = dev->data->dev_private;
 169         struct rte_devargs *devargs = dev->device->devargs;
 170         static const char * const valid_keys[] = {
 171                 "latency",
 172                 NULL
 173         };
 174         struct rte_kvargs *kvlist;
 175         int ret;
 176
 177         if (!devargs)
 178                 return 0;
 179
 180         PMD_INIT_LOG(DEBUG, "device args %s %s",
 181                      devargs->name, devargs->args);
 182
 183         kvlist = rte_kvargs_parse(devargs->args, valid_keys);
 184         if (!kvlist) {
 185                 PMD_DRV_LOG(NOTICE, "invalid parameters");
 186                 return -EINVAL;
 187         }
 188
 189         ret = rte_kvargs_process(kvlist, "latency", hn_set_latency, hv);
 190         if (ret)
 191                 PMD_DRV_LOG(ERR, "Unable to process latency arg\n");
 192
 193         rte_kvargs_free(kvlist);
 194         return ret;
 195 }
 196
 197 /* Update link status.
 198  * Note: the DPDK definition of "wait_to_complete"
 199  *   means block this call until link is up.
 200  *   which is not worth supporting.
 201  */
 202 int
 203 hn_dev_link_update(struct rte_eth_dev *dev,
 204                    int wait_to_complete)
 205 {
 206         struct hn_data *hv = dev->data->dev_private;
 207         struct rte_eth_link link, old;
 208         int error;
 209
 210         old = dev->data->dev_link;
 211
 212         error = hn_rndis_get_linkstatus(hv);
 213         if (error)
 214                 return error;
 215
 216         hn_rndis_get_linkspeed(hv);
 217
 218         hn_vf_link_update(dev, wait_to_complete);
 219
 220         link = (struct rte_eth_link) {
 221                 .link_duplex = ETH_LINK_FULL_DUPLEX,
 222                 .link_autoneg = ETH_LINK_SPEED_FIXED,
 223                 .link_speed = hv->link_speed / 10000,
 224         };
 225
 226         if (hv->link_status == NDIS_MEDIA_STATE_CONNECTED)
 227                 link.link_status = ETH_LINK_UP;
 228         else
 229                 link.link_status = ETH_LINK_DOWN;
 230
 231         if (old.link_status == link.link_status)
 232                 return 0;
 233
 234         PMD_INIT_LOG(DEBUG, "Port %d is %s", dev->data->port_id,
 235                      (link.link_status == ETH_LINK_UP) ? "up" : "down");
 236
 237         return rte_eth_linkstatus_set(dev, &link);
 238 }
 239
 240 static int hn_dev_info_get(struct rte_eth_dev *dev,
 241                            struct rte_eth_dev_info *dev_info)
 242 {
 243         struct hn_data *hv = dev->data->dev_private;
 244         int rc;
 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         dev_info->tx_desc_lim.nb_min = 1;
 259         dev_info->tx_desc_lim.nb_max = 4096;
 260
 261         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 262                 return 0;
 263
 264         /* fills in rx and tx offload capability */
 265         rc = hn_rndis_get_offload(hv, dev_info);
 266         if (rc != 0)
 267                 return rc;
 268
 269         /* merges the offload and queues of vf */
 270         return hn_vf_info_get(hv, dev_info);
 271 }
 272
 273 static int hn_rss_reta_update(struct rte_eth_dev *dev,
 274                               struct rte_eth_rss_reta_entry64 *reta_conf,
 275                               uint16_t reta_size)
 276 {
 277         struct hn_data *hv = dev->data->dev_private;
 278         unsigned int i;
 279         int err;
 280
 281         PMD_INIT_FUNC_TRACE();
 282
 283         if (reta_size != NDIS_HASH_INDCNT) {
 284                 PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
 285                 return -EINVAL;
 286         }
 287
 288         for (i = 0; i < NDIS_HASH_INDCNT; i++) {
 289                 uint16_t idx = i / RTE_RETA_GROUP_SIZE;
 290                 uint16_t shift = i % RTE_RETA_GROUP_SIZE;
 291                 uint64_t mask = (uint64_t)1 << shift;
 292
 293                 if (reta_conf[idx].mask & mask)
 294                         hv->rss_ind[i] = reta_conf[idx].reta[shift];
 295         }
 296
 297         err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 298         if (err) {
 299                 PMD_DRV_LOG(NOTICE,
 300                         "rss disable failed");
 301                 return err;
 302         }
 303
 304         err = hn_rndis_conf_rss(hv, 0);
 305         if (err) {
 306                 PMD_DRV_LOG(NOTICE,
 307                             "reta reconfig failed");
 308                 return err;
 309         }
 310
 311         return hn_vf_reta_hash_update(dev, reta_conf, reta_size);
 312 }
 313
 314 static int hn_rss_reta_query(struct rte_eth_dev *dev,
 315                              struct rte_eth_rss_reta_entry64 *reta_conf,
 316                              uint16_t reta_size)
 317 {
 318         struct hn_data *hv = dev->data->dev_private;
 319         unsigned int i;
 320
 321         PMD_INIT_FUNC_TRACE();
 322
 323         if (reta_size != NDIS_HASH_INDCNT) {
 324                 PMD_DRV_LOG(ERR, "Hash lookup table size does not match NDIS");
 325                 return -EINVAL;
 326         }
 327
 328         for (i = 0; i < NDIS_HASH_INDCNT; i++) {
 329                 uint16_t idx = i / RTE_RETA_GROUP_SIZE;
 330                 uint16_t shift = i % RTE_RETA_GROUP_SIZE;
 331                 uint64_t mask = (uint64_t)1 << shift;
 332
 333                 if (reta_conf[idx].mask & mask)
 334                         reta_conf[idx].reta[shift] = hv->rss_ind[i];
 335         }
 336         return 0;
 337 }
 338
 339 static void hn_rss_hash_init(struct hn_data *hv,
 340                              const struct rte_eth_rss_conf *rss_conf)
 341 {
 342         /* Convert from DPDK RSS hash flags to NDIS hash flags */
 343         hv->rss_hash = NDIS_HASH_FUNCTION_TOEPLITZ;
 344
 345         if (rss_conf->rss_hf & ETH_RSS_IPV4)
 346                 hv->rss_hash |= NDIS_HASH_IPV4;
 347         if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV4_TCP)
 348                 hv->rss_hash |= NDIS_HASH_TCP_IPV4;
 349         if (rss_conf->rss_hf & ETH_RSS_IPV6)
 350                 hv->rss_hash |=  NDIS_HASH_IPV6;
 351         if (rss_conf->rss_hf & ETH_RSS_IPV6_EX)
 352                 hv->rss_hash |=  NDIS_HASH_IPV6_EX;
 353         if (rss_conf->rss_hf & ETH_RSS_NONFRAG_IPV6_TCP)
 354                 hv->rss_hash |= NDIS_HASH_TCP_IPV6;
 355         if (rss_conf->rss_hf & ETH_RSS_IPV6_TCP_EX)
 356                 hv->rss_hash |= NDIS_HASH_TCP_IPV6_EX;
 357
 358         memcpy(hv->rss_key, rss_conf->rss_key ? : rss_default_key,
 359                NDIS_HASH_KEYSIZE_TOEPLITZ);
 360 }
 361
 362 static int hn_rss_hash_update(struct rte_eth_dev *dev,
 363                               struct rte_eth_rss_conf *rss_conf)
 364 {
 365         struct hn_data *hv = dev->data->dev_private;
 366         int err;
 367
 368         PMD_INIT_FUNC_TRACE();
 369
 370         err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 371         if (err) {
 372                 PMD_DRV_LOG(NOTICE,
 373                             "rss disable failed");
 374                 return err;
 375         }
 376
 377         hn_rss_hash_init(hv, rss_conf);
 378
 379         if (rss_conf->rss_hf != 0) {
 380                 err = hn_rndis_conf_rss(hv, 0);
 381                 if (err) {
 382                         PMD_DRV_LOG(NOTICE,
 383                                     "rss reconfig failed (RSS disabled)");
 384                         return err;
 385                 }
 386         }
 387
 388         return hn_vf_rss_hash_update(dev, rss_conf);
 389 }
 390
 391 static int hn_rss_hash_conf_get(struct rte_eth_dev *dev,
 392                                 struct rte_eth_rss_conf *rss_conf)
 393 {
 394         struct hn_data *hv = dev->data->dev_private;
 395
 396         PMD_INIT_FUNC_TRACE();
 397
 398         if (hv->ndis_ver < NDIS_VERSION_6_20) {
 399                 PMD_DRV_LOG(DEBUG, "RSS not supported on this host");
 400                 return -EOPNOTSUPP;
 401         }
 402
 403         rss_conf->rss_key_len = NDIS_HASH_KEYSIZE_TOEPLITZ;
 404         if (rss_conf->rss_key)
 405                 memcpy(rss_conf->rss_key, hv->rss_key,
 406                        NDIS_HASH_KEYSIZE_TOEPLITZ);
 407
 408         rss_conf->rss_hf = 0;
 409         if (hv->rss_hash & NDIS_HASH_IPV4)
 410                 rss_conf->rss_hf |= ETH_RSS_IPV4;
 411
 412         if (hv->rss_hash & NDIS_HASH_TCP_IPV4)
 413                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV4_TCP;
 414
 415         if (hv->rss_hash & NDIS_HASH_IPV6)
 416                 rss_conf->rss_hf |= ETH_RSS_IPV6;
 417
 418         if (hv->rss_hash & NDIS_HASH_IPV6_EX)
 419                 rss_conf->rss_hf |= ETH_RSS_IPV6_EX;
 420
 421         if (hv->rss_hash & NDIS_HASH_TCP_IPV6)
 422                 rss_conf->rss_hf |= ETH_RSS_NONFRAG_IPV6_TCP;
 423
 424         if (hv->rss_hash & NDIS_HASH_TCP_IPV6_EX)
 425                 rss_conf->rss_hf |= ETH_RSS_IPV6_TCP_EX;
 426
 427         return 0;
 428 }
 429
 430 static int
 431 hn_dev_promiscuous_enable(struct rte_eth_dev *dev)
 432 {
 433         struct hn_data *hv = dev->data->dev_private;
 434
 435         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_PROMISCUOUS);
 436         return hn_vf_promiscuous_enable(dev);
 437 }
 438
 439 static int
 440 hn_dev_promiscuous_disable(struct rte_eth_dev *dev)
 441 {
 442         struct hn_data *hv = dev->data->dev_private;
 443         uint32_t filter;
 444
 445         filter = NDIS_PACKET_TYPE_DIRECTED | NDIS_PACKET_TYPE_BROADCAST;
 446         if (dev->data->all_multicast)
 447                 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 448         hn_rndis_set_rxfilter(hv, filter);
 449         return hn_vf_promiscuous_disable(dev);
 450 }
 451
 452 static int
 453 hn_dev_allmulticast_enable(struct rte_eth_dev *dev)
 454 {
 455         struct hn_data *hv = dev->data->dev_private;
 456
 457         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 458                               NDIS_PACKET_TYPE_ALL_MULTICAST |
 459                         NDIS_PACKET_TYPE_BROADCAST);
 460         return hn_vf_allmulticast_enable(dev);
 461 }
 462
 463 static int
 464 hn_dev_allmulticast_disable(struct rte_eth_dev *dev)
 465 {
 466         struct hn_data *hv = dev->data->dev_private;
 467
 468         hn_rndis_set_rxfilter(hv, NDIS_PACKET_TYPE_DIRECTED |
 469                              NDIS_PACKET_TYPE_BROADCAST);
 470         return hn_vf_allmulticast_disable(dev);
 471 }
 472
 473 static int
 474 hn_dev_mc_addr_list(struct rte_eth_dev *dev,
 475                      struct rte_ether_addr *mc_addr_set,
 476                      uint32_t nb_mc_addr)
 477 {
 478         /* No filtering on the synthetic path, but can do it on VF */
 479         return hn_vf_mc_addr_list(dev, mc_addr_set, nb_mc_addr);
 480 }
 481
 482 /* Setup shared rx/tx queue data */
 483 static int hn_subchan_configure(struct hn_data *hv,
 484                                 uint32_t subchan)
 485 {
 486         struct vmbus_channel *primary = hn_primary_chan(hv);
 487         int err;
 488         unsigned int retry = 0;
 489
 490         PMD_DRV_LOG(DEBUG,
 491                     "open %u subchannels", subchan);
 492
 493         /* Send create sub channels command */
 494         err = hn_nvs_alloc_subchans(hv, &subchan);
 495         if (err)
 496                 return  err;
 497
 498         while (subchan > 0) {
 499                 struct vmbus_channel *new_sc;
 500                 uint16_t chn_index;
 501
 502                 err = rte_vmbus_subchan_open(primary, &new_sc);
 503                 if (err == -ENOENT && ++retry < 1000) {
 504                         /* This can happen if not ready yet */
 505                         rte_delay_ms(10);
 506                         continue;
 507                 }
 508
 509                 if (err) {
 510                         PMD_DRV_LOG(ERR,
 511                                     "open subchannel failed: %d", err);
 512                         return err;
 513                 }
 514
 515                 rte_vmbus_set_latency(hv->vmbus, new_sc, hv->latency);
 516
 517                 retry = 0;
 518                 chn_index = rte_vmbus_sub_channel_index(new_sc);
 519                 if (chn_index == 0 || chn_index > hv->max_queues) {
 520                         PMD_DRV_LOG(ERR,
 521                                     "Invalid subchannel offermsg channel %u",
 522                                     chn_index);
 523                         return -EIO;
 524                 }
 525
 526                 PMD_DRV_LOG(DEBUG, "new sub channel %u", chn_index);
 527                 hv->channels[chn_index] = new_sc;
 528                 --subchan;
 529         }
 530
 531         return err;
 532 }
 533
 534 static int hn_dev_configure(struct rte_eth_dev *dev)
 535 {
 536         struct rte_eth_conf *dev_conf = &dev->data->dev_conf;
 537         struct rte_eth_rss_conf *rss_conf = &dev_conf->rx_adv_conf.rss_conf;
 538         const struct rte_eth_rxmode *rxmode = &dev_conf->rxmode;
 539         const struct rte_eth_txmode *txmode = &dev_conf->txmode;
 540         struct hn_data *hv = dev->data->dev_private;
 541         uint64_t unsupported;
 542         int i, err, subchan;
 543
 544         PMD_INIT_FUNC_TRACE();
 545
 546         if (dev_conf->rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
 547                 dev_conf->rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
 548
 549         unsupported = txmode->offloads & ~HN_TX_OFFLOAD_CAPS;
 550         if (unsupported) {
 551                 PMD_DRV_LOG(NOTICE,
 552                             "unsupported TX offload: %#" PRIx64,
 553                             unsupported);
 554                 return -EINVAL;
 555         }
 556
 557         unsupported = rxmode->offloads & ~HN_RX_OFFLOAD_CAPS;
 558         if (unsupported) {
 559                 PMD_DRV_LOG(NOTICE,
 560                             "unsupported RX offload: %#" PRIx64,
 561                             rxmode->offloads);
 562                 return -EINVAL;
 563         }
 564
 565         hv->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
 566
 567         err = hn_rndis_conf_offload(hv, txmode->offloads,
 568                                     rxmode->offloads);
 569         if (err) {
 570                 PMD_DRV_LOG(NOTICE,
 571                             "offload configure failed");
 572                 return err;
 573         }
 574
 575         hv->num_queues = RTE_MAX(dev->data->nb_rx_queues,
 576                                  dev->data->nb_tx_queues);
 577
 578         for (i = 0; i < NDIS_HASH_INDCNT; i++)
 579                 hv->rss_ind[i] = i % dev->data->nb_rx_queues;
 580
 581         hn_rss_hash_init(hv, rss_conf);
 582
 583         subchan = hv->num_queues - 1;
 584         if (subchan > 0) {
 585                 err = hn_subchan_configure(hv, subchan);
 586                 if (err) {
 587                         PMD_DRV_LOG(NOTICE,
 588                                     "subchannel configuration failed");
 589                         return err;
 590                 }
 591
 592                 err = hn_rndis_conf_rss(hv, NDIS_RSS_FLAG_DISABLE);
 593                 if (err) {
 594                         PMD_DRV_LOG(NOTICE,
 595                                 "rss disable failed");
 596                         return err;
 597                 }
 598
 599                 if (rss_conf->rss_hf != 0) {
 600                         err = hn_rndis_conf_rss(hv, 0);
 601                         if (err) {
 602                                 PMD_DRV_LOG(NOTICE,
 603                                             "initial RSS config failed");
 604                                 return err;
 605                         }
 606                 }
 607         }
 608
 609         return hn_vf_configure(dev, dev_conf);
 610 }
 611
 612 static int hn_dev_stats_get(struct rte_eth_dev *dev,
 613                             struct rte_eth_stats *stats)
 614 {
 615         unsigned int i;
 616
 617         hn_vf_stats_get(dev, stats);
 618
 619         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 620                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 621
 622                 if (!txq)
 623                         continue;
 624
 625                 stats->opackets += txq->stats.packets;
 626                 stats->obytes += txq->stats.bytes;
 627                 stats->oerrors += txq->stats.errors;
 628
 629                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 630                         stats->q_opackets[i] = txq->stats.packets;
 631                         stats->q_obytes[i] = txq->stats.bytes;
 632                 }
 633         }
 634
 635         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 636                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 637
 638                 if (!rxq)
 639                         continue;
 640
 641                 stats->ipackets += rxq->stats.packets;
 642                 stats->ibytes += rxq->stats.bytes;
 643                 stats->ierrors += rxq->stats.errors;
 644                 stats->imissed += rxq->stats.ring_full;
 645
 646                 if (i < RTE_ETHDEV_QUEUE_STAT_CNTRS) {
 647                         stats->q_ipackets[i] = rxq->stats.packets;
 648                         stats->q_ibytes[i] = rxq->stats.bytes;
 649                 }
 650         }
 651
 652         stats->rx_nombuf = dev->data->rx_mbuf_alloc_failed;
 653         return 0;
 654 }
 655
 656 static int
 657 hn_dev_stats_reset(struct rte_eth_dev *dev)
 658 {
 659         unsigned int i;
 660
 661         PMD_INIT_FUNC_TRACE();
 662
 663         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 664                 struct hn_tx_queue *txq = dev->data->tx_queues[i];
 665
 666                 if (!txq)
 667                         continue;
 668                 memset(&txq->stats, 0, sizeof(struct hn_stats));
 669         }
 670
 671         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 672                 struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 673
 674                 if (!rxq)
 675                         continue;
 676
 677                 memset(&rxq->stats, 0, sizeof(struct hn_stats));
 678         }
 679
 680         return 0;
 681 }
 682
 683 static int
 684 hn_dev_xstats_reset(struct rte_eth_dev *dev)
 685 {
 686         int ret;
 687
 688         ret = hn_dev_stats_reset(dev);
 689         if (ret != 0)
 690                 return 0;
 691
 692         return hn_vf_xstats_reset(dev);
 693 }
 694
 695 static int
 696 hn_dev_xstats_count(struct rte_eth_dev *dev)
 697 {
 698         int ret, count;
 699
 700         count = dev->data->nb_tx_queues * RTE_DIM(hn_stat_strings);
 701         count += dev->data->nb_rx_queues * RTE_DIM(hn_stat_strings);
 702
 703         ret = hn_vf_xstats_get_names(dev, NULL, 0);
 704         if (ret < 0)
 705                 return ret;
 706
 707         return count + ret;
 708 }
 709
 710 static int
 711 hn_dev_xstats_get_names(struct rte_eth_dev *dev,
 712                         struct rte_eth_xstat_name *xstats_names,
 713                         unsigned int limit)
 714 {
 715         unsigned int i, t, count = 0;
 716         int ret;
 717
 718         if (!xstats_names)
 719                 return hn_dev_xstats_count(dev);
 720
 721         /* Note: limit checked in rte_eth_xstats_names() */
 722         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 723                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 724
 725                 if (!txq)
 726                         continue;
 727
 728                 if (count >= limit)
 729                         break;
 730
 731                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 732                         snprintf(xstats_names[count++].name,
 733                                  RTE_ETH_XSTATS_NAME_SIZE,
 734                                  "tx_q%u_%s", i, hn_stat_strings[t].name);
 735         }
 736
 737         for (i = 0; i < dev->data->nb_rx_queues; i++)  {
 738                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 739
 740                 if (!rxq)
 741                         continue;
 742
 743                 if (count >= limit)
 744                         break;
 745
 746                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++)
 747                         snprintf(xstats_names[count++].name,
 748                                  RTE_ETH_XSTATS_NAME_SIZE,
 749                                  "rx_q%u_%s", i,
 750                                  hn_stat_strings[t].name);
 751         }
 752
 753         ret = hn_vf_xstats_get_names(dev, xstats_names + count,
 754                                      limit - count);
 755         if (ret < 0)
 756                 return ret;
 757
 758         return count + ret;
 759 }
 760
 761 static int
 762 hn_dev_xstats_get(struct rte_eth_dev *dev,
 763                   struct rte_eth_xstat *xstats,
 764                   unsigned int n)
 765 {
 766         unsigned int i, t, count = 0;
 767         const unsigned int nstats = hn_dev_xstats_count(dev);
 768         const char *stats;
 769         int ret;
 770
 771         PMD_INIT_FUNC_TRACE();
 772
 773         if (n < nstats)
 774                 return nstats;
 775
 776         for (i = 0; i < dev->data->nb_tx_queues; i++) {
 777                 const struct hn_tx_queue *txq = dev->data->tx_queues[i];
 778
 779                 if (!txq)
 780                         continue;
 781
 782                 stats = (const char *)&txq->stats;
 783                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 784                         xstats[count].id = count;
 785                         xstats[count].value = *(const uint64_t *)
 786                                 (stats + hn_stat_strings[t].offset);
 787                 }
 788         }
 789
 790         for (i = 0; i < dev->data->nb_rx_queues; i++) {
 791                 const struct hn_rx_queue *rxq = dev->data->rx_queues[i];
 792
 793                 if (!rxq)
 794                         continue;
 795
 796                 stats = (const char *)&rxq->stats;
 797                 for (t = 0; t < RTE_DIM(hn_stat_strings); t++, count++) {
 798                         xstats[count].id = count;
 799                         xstats[count].value = *(const uint64_t *)
 800                                 (stats + hn_stat_strings[t].offset);
 801                 }
 802         }
 803
 804         ret = hn_vf_xstats_get(dev, xstats, count, n);
 805         if (ret < 0)
 806                 return ret;
 807
 808         return count + ret;
 809 }
 810
 811 static int
 812 hn_dev_start(struct rte_eth_dev *dev)
 813 {
 814         struct hn_data *hv = dev->data->dev_private;
 815         int error;
 816
 817         PMD_INIT_FUNC_TRACE();
 818
 819         error = hn_rndis_set_rxfilter(hv,
 820                                       NDIS_PACKET_TYPE_BROADCAST |
 821                                       NDIS_PACKET_TYPE_ALL_MULTICAST |
 822                                       NDIS_PACKET_TYPE_DIRECTED);
 823         if (error)
 824                 return error;
 825
 826         error = hn_vf_start(dev);
 827         if (error)
 828                 hn_rndis_set_rxfilter(hv, 0);
 829
 830         /* Initialize Link state */
 831         if (error == 0)
 832                 hn_dev_link_update(dev, 0);
 833
 834         return error;
 835 }
 836
 837 static void
 838 hn_dev_stop(struct rte_eth_dev *dev)
 839 {
 840         struct hn_data *hv = dev->data->dev_private;
 841
 842         PMD_INIT_FUNC_TRACE();
 843
 844         hn_rndis_set_rxfilter(hv, 0);
 845         hn_vf_stop(dev);
 846 }
 847
 848 static void
 849 hn_dev_close(struct rte_eth_dev *dev)
 850 {
 851         PMD_INIT_FUNC_TRACE();
 852
 853         hn_vf_close(dev);
 854         hn_dev_free_queues(dev);
 855 }
 856
 857 static const struct eth_dev_ops hn_eth_dev_ops = {
 858         .dev_configure          = hn_dev_configure,
 859         .dev_start              = hn_dev_start,
 860         .dev_stop               = hn_dev_stop,
 861         .dev_close              = hn_dev_close,
 862         .dev_infos_get          = hn_dev_info_get,
 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->tx_pkt_burst = &hn_xmit_pkts;
 939         eth_dev->rx_pkt_burst = &hn_recv_pkts;
 940
 941         /*
 942          * for secondary processes, we don't initialize any further as primary
 943          * has already done this work.
 944          */
 945         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 946                 return 0;
 947
 948         /* Since Hyper-V only supports one MAC address */
 949         eth_dev->data->mac_addrs = rte_calloc("hv_mac", HN_MAX_MAC_ADDRS,
 950                                               sizeof(struct rte_ether_addr), 0);
 951         if (eth_dev->data->mac_addrs == NULL) {
 952                 PMD_INIT_LOG(ERR,
 953                              "Failed to allocate memory store MAC addresses");
 954                 return -ENOMEM;
 955         }
 956
 957         hv->vmbus = vmbus;
 958         hv->rxbuf_res = &vmbus->resource[HV_RECV_BUF_MAP];
 959         hv->chim_res  = &vmbus->resource[HV_SEND_BUF_MAP];
 960         hv->port_id = eth_dev->data->port_id;
 961         hv->latency = HN_CHAN_LATENCY_NS;
 962         hv->max_queues = 1;
 963         rte_rwlock_init(&hv->vf_lock);
 964         hv->vf_port = HN_INVALID_PORT;
 965
 966         err = hn_parse_args(eth_dev);
 967         if (err)
 968                 return err;
 969
 970         strlcpy(hv->owner.name, eth_dev->device->name,
 971                 RTE_ETH_MAX_OWNER_NAME_LEN);
 972         err = rte_eth_dev_owner_new(&hv->owner.id);
 973         if (err) {
 974                 PMD_INIT_LOG(ERR, "Can not get owner id");
 975                 return err;
 976         }
 977
 978         /* Initialize primary channel input for control operations */
 979         err = rte_vmbus_chan_open(vmbus, &hv->channels[0]);
 980         if (err)
 981                 return err;
 982
 983         rte_vmbus_set_latency(hv->vmbus, hv->channels[0], hv->latency);
 984
 985         hv->primary = hn_rx_queue_alloc(hv, 0,
 986                                         eth_dev->device->numa_node);
 987
 988         if (!hv->primary)
 989                 return -ENOMEM;
 990
 991         err = hn_attach(hv, RTE_ETHER_MTU);
 992         if  (err)
 993                 goto failed;
 994
 995         err = hn_chim_init(eth_dev);
 996         if (err)
 997                 goto failed;
 998
 999         err = hn_rndis_get_eaddr(hv, eth_dev->data->mac_addrs->addr_bytes);
1000         if (err)
1001                 goto failed;
1002
1003         /* Multi queue requires later versions of windows server */
1004         if (hv->nvs_ver < NVS_VERSION_5)
1005                 return 0;
1006
1007         max_chan = rte_vmbus_max_channels(vmbus);
1008         PMD_INIT_LOG(DEBUG, "VMBus max channels %d", max_chan);
1009         if (max_chan <= 0)
1010                 goto failed;
1011
1012         if (hn_rndis_query_rsscaps(hv, &rxr_cnt) != 0)
1013                 rxr_cnt = 1;
1014
1015         hv->max_queues = RTE_MIN(rxr_cnt, (unsigned int)max_chan);
1016
1017         /* If VF was reported but not added, do it now */
1018         if (hv->vf_present && !hn_vf_attached(hv)) {
1019                 PMD_INIT_LOG(DEBUG, "Adding VF device");
1020
1021                 err = hn_vf_add(eth_dev, hv);
1022                 if (err)
1023                         hv->vf_present = 0;
1024         }
1025
1026         return 0;
1027
1028 failed:
1029         PMD_INIT_LOG(NOTICE, "device init failed");
1030
1031         hn_chim_uninit(eth_dev);
1032         hn_detach(hv);
1033         return err;
1034 }
1035
1036 static int
1037 eth_hn_dev_uninit(struct rte_eth_dev *eth_dev)
1038 {
1039         struct hn_data *hv = eth_dev->data->dev_private;
1040         int ret;
1041
1042         PMD_INIT_FUNC_TRACE();
1043
1044         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1045                 return 0;
1046
1047         hn_dev_stop(eth_dev);
1048         hn_dev_close(eth_dev);
1049
1050         eth_dev->dev_ops = NULL;
1051         eth_dev->tx_pkt_burst = NULL;
1052         eth_dev->rx_pkt_burst = NULL;
1053
1054         hn_detach(hv);
1055         hn_chim_uninit(eth_dev);
1056         rte_vmbus_chan_close(hv->primary->chan);
1057         rte_free(hv->primary);
1058         ret = rte_eth_dev_owner_delete(hv->owner.id);
1059         if (ret != 0)
1060                 return ret;
1061
1062         return 0;
1063 }
1064
1065 static int eth_hn_probe(struct rte_vmbus_driver *drv __rte_unused,
1066                         struct rte_vmbus_device *dev)
1067 {
1068         struct rte_eth_dev *eth_dev;
1069         int ret;
1070
1071         PMD_INIT_FUNC_TRACE();
1072
1073         eth_dev = eth_dev_vmbus_allocate(dev, sizeof(struct hn_data));
1074         if (!eth_dev)
1075                 return -ENOMEM;
1076
1077         ret = eth_hn_dev_init(eth_dev);
1078         if (ret)
1079                 eth_dev_vmbus_release(eth_dev);
1080         else
1081                 rte_eth_dev_probing_finish(eth_dev);
1082
1083         return ret;
1084 }
1085
1086 static int eth_hn_remove(struct rte_vmbus_device *dev)
1087 {
1088         struct rte_eth_dev *eth_dev;
1089         int ret;
1090
1091         PMD_INIT_FUNC_TRACE();
1092
1093         eth_dev = rte_eth_dev_allocated(dev->device.name);
1094         if (!eth_dev)
1095                 return -ENODEV;
1096
1097         ret = eth_hn_dev_uninit(eth_dev);
1098         if (ret)
1099                 return ret;
1100
1101         eth_dev_vmbus_release(eth_dev);
1102         return 0;
1103 }
1104
1105 /* Network device GUID */
1106 static const rte_uuid_t hn_net_ids[] = {
1107         /*  f8615163-df3e-46c5-913f-f2d2f965ed0e */
1108         RTE_UUID_INIT(0xf8615163, 0xdf3e, 0x46c5, 0x913f, 0xf2d2f965ed0eULL),
1109         { 0 }
1110 };
1111
1112 static struct rte_vmbus_driver rte_netvsc_pmd = {
1113         .id_table = hn_net_ids,
1114         .probe = eth_hn_probe,
1115         .remove = eth_hn_remove,
1116 };
1117
1118 RTE_PMD_REGISTER_VMBUS(net_netvsc, rte_netvsc_pmd);
1119 RTE_PMD_REGISTER_KMOD_DEP(net_netvsc, "* uio_hv_generic");
1120
1121 RTE_INIT(hn_init_log)
1122 {
1123         hn_logtype_init = rte_log_register("pmd.net.netvsc.init");
1124         if (hn_logtype_init >= 0)
1125                 rte_log_set_level(hn_logtype_init, RTE_LOG_NOTICE);
1126         hn_logtype_driver = rte_log_register("pmd.net.netvsc.driver");
1127         if (hn_logtype_driver >= 0)
1128                 rte_log_set_level(hn_logtype_driver, RTE_LOG_NOTICE);
1129 }