drivers/net/hyperv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/init.h>
  23 #include <linux/atomic.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/pci.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/in.h>
  36 #include <linux/slab.h>
  37 #include <linux/rtnetlink.h>
  38 #include <linux/netpoll.h>
  39
  40 #include <net/arp.h>
  41 #include <net/route.h>
  42 #include <net/sock.h>
  43 #include <net/pkt_sched.h>
  44 #include <net/checksum.h>
  45 #include <net/ip6_checksum.h>
  46
  47 #include "hyperv_net.h"
  48
  49 #define RING_SIZE_MIN   64
  50 #define RETRY_US_LO     5000
  51 #define RETRY_US_HI     10000
  52 #define RETRY_MAX       2000    /* >10 sec */
  53
  54 #define LINKCHANGE_INT (2 * HZ)
  55 #define VF_TAKEOVER_INT (HZ / 10)
  56
  57 static unsigned int ring_size __ro_after_init = 128;
  58 module_param(ring_size, uint, 0444);
  59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  60 unsigned int netvsc_ring_bytes __ro_after_init;
  61
  62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  63                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  64                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  65                                 NETIF_MSG_TX_ERR;
  66
  67 static int debug = -1;
  68 module_param(debug, int, 0444);
  69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  70
  71 static LIST_HEAD(netvsc_dev_list);
  72
  73 static void netvsc_change_rx_flags(struct net_device *net, int change)
  74 {
  75         struct net_device_context *ndev_ctx = netdev_priv(net);
  76         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  77         int inc;
  78
  79         if (!vf_netdev)
  80                 return;
  81
  82         if (change & IFF_PROMISC) {
  83                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  84                 dev_set_promiscuity(vf_netdev, inc);
  85         }
  86
  87         if (change & IFF_ALLMULTI) {
  88                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  89                 dev_set_allmulti(vf_netdev, inc);
  90         }
  91 }
  92
  93 static void netvsc_set_rx_mode(struct net_device *net)
  94 {
  95         struct net_device_context *ndev_ctx = netdev_priv(net);
  96         struct net_device *vf_netdev;
  97         struct netvsc_device *nvdev;
  98
  99         rcu_read_lock();
 100         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
 101         if (vf_netdev) {
 102                 dev_uc_sync(vf_netdev, net);
 103                 dev_mc_sync(vf_netdev, net);
 104         }
 105
 106         nvdev = rcu_dereference(ndev_ctx->nvdev);
 107         if (nvdev)
 108                 rndis_filter_update(nvdev);
 109         rcu_read_unlock();
 110 }
 111
 112 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
 113                              struct net_device *ndev)
 114 {
 115         nvscdev->tx_disable = false;
 116         virt_wmb(); /* ensure queue wake up mechanism is on */
 117
 118         netif_tx_wake_all_queues(ndev);
 119 }
 120
 121 static int netvsc_open(struct net_device *net)
 122 {
 123         struct net_device_context *ndev_ctx = netdev_priv(net);
 124         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 125         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 126         struct rndis_device *rdev;
 127         int ret = 0;
 128
 129         netif_carrier_off(net);
 130
 131         /* Open up the device */
 132         ret = rndis_filter_open(nvdev);
 133         if (ret != 0) {
 134                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 135                 return ret;
 136         }
 137
 138         rdev = nvdev->extension;
 139         if (!rdev->link_state) {
 140                 netif_carrier_on(net);
 141                 netvsc_tx_enable(nvdev, net);
 142         }
 143
 144         if (vf_netdev) {
 145                 /* Setting synthetic device up transparently sets
 146                  * slave as up. If open fails, then slave will be
 147                  * still be offline (and not used).
 148                  */
 149                 ret = dev_open(vf_netdev, NULL);
 150                 if (ret)
 151                         netdev_warn(net,
 152                                     "unable to open slave: %s: %d\n",
 153                                     vf_netdev->name, ret);
 154         }
 155         return 0;
 156 }
 157
 158 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 159 {
 160         unsigned int retry = 0;
 161         int i;
 162
 163         /* Ensure pending bytes in ring are read */
 164         for (;;) {
 165                 u32 aread = 0;
 166
 167                 for (i = 0; i < nvdev->num_chn; i++) {
 168                         struct vmbus_channel *chn
 169                                 = nvdev->chan_table[i].channel;
 170
 171                         if (!chn)
 172                                 continue;
 173
 174                         /* make sure receive not running now */
 175                         napi_synchronize(&nvdev->chan_table[i].napi);
 176
 177                         aread = hv_get_bytes_to_read(&chn->inbound);
 178                         if (aread)
 179                                 break;
 180
 181                         aread = hv_get_bytes_to_read(&chn->outbound);
 182                         if (aread)
 183                                 break;
 184                 }
 185
 186                 if (aread == 0)
 187                         return 0;
 188
 189                 if (++retry > RETRY_MAX)
 190                         return -ETIMEDOUT;
 191
 192                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 193         }
 194 }
 195
 196 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
 197                               struct net_device *ndev)
 198 {
 199         if (nvscdev) {
 200                 nvscdev->tx_disable = true;
 201                 virt_wmb(); /* ensure txq will not wake up after stop */
 202         }
 203
 204         netif_tx_disable(ndev);
 205 }
 206
 207 static int netvsc_close(struct net_device *net)
 208 {
 209         struct net_device_context *net_device_ctx = netdev_priv(net);
 210         struct net_device *vf_netdev
 211                 = rtnl_dereference(net_device_ctx->vf_netdev);
 212         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 213         int ret;
 214
 215         netvsc_tx_disable(nvdev, net);
 216
 217         /* No need to close rndis filter if it is removed already */
 218         if (!nvdev)
 219                 return 0;
 220
 221         ret = rndis_filter_close(nvdev);
 222         if (ret != 0) {
 223                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 224                 return ret;
 225         }
 226
 227         ret = netvsc_wait_until_empty(nvdev);
 228         if (ret)
 229                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 230
 231         if (vf_netdev)
 232                 dev_close(vf_netdev);
 233
 234         return ret;
 235 }
 236
 237 static inline void *init_ppi_data(struct rndis_message *msg,
 238                                   u32 ppi_size, u32 pkt_type)
 239 {
 240         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 241         struct rndis_per_packet_info *ppi;
 242
 243         rndis_pkt->data_offset += ppi_size;
 244         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 245                 + rndis_pkt->per_pkt_info_len;
 246
 247         ppi->size = ppi_size;
 248         ppi->type = pkt_type;
 249         ppi->internal = 0;
 250         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 251
 252         rndis_pkt->per_pkt_info_len += ppi_size;
 253
 254         return ppi + 1;
 255 }
 256
 257 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 258  * packets. We can use ethtool to change UDP hash level when necessary.
 259  */
 260 static inline u32 netvsc_get_hash(
 261         struct sk_buff *skb,
 262         const struct net_device_context *ndc)
 263 {
 264         struct flow_keys flow;
 265         u32 hash, pkt_proto = 0;
 266         static u32 hashrnd __read_mostly;
 267
 268         net_get_random_once(&hashrnd, sizeof(hashrnd));
 269
 270         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 271                 return 0;
 272
 273         switch (flow.basic.ip_proto) {
 274         case IPPROTO_TCP:
 275                 if (flow.basic.n_proto == htons(ETH_P_IP))
 276                         pkt_proto = HV_TCP4_L4HASH;
 277                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 278                         pkt_proto = HV_TCP6_L4HASH;
 279
 280                 break;
 281
 282         case IPPROTO_UDP:
 283                 if (flow.basic.n_proto == htons(ETH_P_IP))
 284                         pkt_proto = HV_UDP4_L4HASH;
 285                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 286                         pkt_proto = HV_UDP6_L4HASH;
 287
 288                 break;
 289         }
 290
 291         if (pkt_proto & ndc->l4_hash) {
 292                 return skb_get_hash(skb);
 293         } else {
 294                 if (flow.basic.n_proto == htons(ETH_P_IP))
 295                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 296                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 297                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 298                 else
 299                         hash = 0;
 300
 301                 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
 302         }
 303
 304         return hash;
 305 }
 306
 307 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 308                                       struct sk_buff *skb, int old_idx)
 309 {
 310         const struct net_device_context *ndc = netdev_priv(ndev);
 311         struct sock *sk = skb->sk;
 312         int q_idx;
 313
 314         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 315                               (VRSS_SEND_TAB_SIZE - 1)];
 316
 317         /* If queue index changed record the new value */
 318         if (q_idx != old_idx &&
 319             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 320                 sk_tx_queue_set(sk, q_idx);
 321
 322         return q_idx;
 323 }
 324
 325 /*
 326  * Select queue for transmit.
 327  *
 328  * If a valid queue has already been assigned, then use that.
 329  * Otherwise compute tx queue based on hash and the send table.
 330  *
 331  * This is basically similar to default (netdev_pick_tx) with the added step
 332  * of using the host send_table when no other queue has been assigned.
 333  *
 334  * TODO support XPS - but get_xps_queue not exported
 335  */
 336 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 337 {
 338         int q_idx = sk_tx_queue_get(skb->sk);
 339
 340         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 341                 /* If forwarding a packet, we use the recorded queue when
 342                  * available for better cache locality.
 343                  */
 344                 if (skb_rx_queue_recorded(skb))
 345                         q_idx = skb_get_rx_queue(skb);
 346                 else
 347                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 348         }
 349
 350         return q_idx;
 351 }
 352
 353 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 354                                struct net_device *sb_dev)
 355 {
 356         struct net_device_context *ndc = netdev_priv(ndev);
 357         struct net_device *vf_netdev;
 358         u16 txq;
 359
 360         rcu_read_lock();
 361         vf_netdev = rcu_dereference(ndc->vf_netdev);
 362         if (vf_netdev) {
 363                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 364
 365                 if (vf_ops->ndo_select_queue)
 366                         txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
 367                 else
 368                         txq = netdev_pick_tx(vf_netdev, skb, NULL);
 369
 370                 /* Record the queue selected by VF so that it can be
 371                  * used for common case where VF has more queues than
 372                  * the synthetic device.
 373                  */
 374                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 375         } else {
 376                 txq = netvsc_pick_tx(ndev, skb);
 377         }
 378         rcu_read_unlock();
 379
 380         while (unlikely(txq >= ndev->real_num_tx_queues))
 381                 txq -= ndev->real_num_tx_queues;
 382
 383         return txq;
 384 }
 385
 386 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
 387                        struct hv_page_buffer *pb)
 388 {
 389         int j = 0;
 390
 391         /* Deal with compound pages by ignoring unused part
 392          * of the page.
 393          */
 394         page += (offset >> PAGE_SHIFT);
 395         offset &= ~PAGE_MASK;
 396
 397         while (len > 0) {
 398                 unsigned long bytes;
 399
 400                 bytes = PAGE_SIZE - offset;
 401                 if (bytes > len)
 402                         bytes = len;
 403                 pb[j].pfn = page_to_pfn(page);
 404                 pb[j].offset = offset;
 405                 pb[j].len = bytes;
 406
 407                 offset += bytes;
 408                 len -= bytes;
 409
 410                 if (offset == PAGE_SIZE && len) {
 411                         page++;
 412                         offset = 0;
 413                         j++;
 414                 }
 415         }
 416
 417         return j + 1;
 418 }
 419
 420 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 421                            struct hv_netvsc_packet *packet,
 422                            struct hv_page_buffer *pb)
 423 {
 424         u32 slots_used = 0;
 425         char *data = skb->data;
 426         int frags = skb_shinfo(skb)->nr_frags;
 427         int i;
 428
 429         /* The packet is laid out thus:
 430          * 1. hdr: RNDIS header and PPI
 431          * 2. skb linear data
 432          * 3. skb fragment data
 433          */
 434         slots_used += fill_pg_buf(virt_to_page(hdr),
 435                                   offset_in_page(hdr),
 436                                   len, &pb[slots_used]);
 437
 438         packet->rmsg_size = len;
 439         packet->rmsg_pgcnt = slots_used;
 440
 441         slots_used += fill_pg_buf(virt_to_page(data),
 442                                 offset_in_page(data),
 443                                 skb_headlen(skb), &pb[slots_used]);
 444
 445         for (i = 0; i < frags; i++) {
 446                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 447
 448                 slots_used += fill_pg_buf(skb_frag_page(frag),
 449                                         frag->page_offset,
 450                                         skb_frag_size(frag), &pb[slots_used]);
 451         }
 452         return slots_used;
 453 }
 454
 455 static int count_skb_frag_slots(struct sk_buff *skb)
 456 {
 457         int i, frags = skb_shinfo(skb)->nr_frags;
 458         int pages = 0;
 459
 460         for (i = 0; i < frags; i++) {
 461                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 462                 unsigned long size = skb_frag_size(frag);
 463                 unsigned long offset = frag->page_offset;
 464
 465                 /* Skip unused frames from start of page */
 466                 offset &= ~PAGE_MASK;
 467                 pages += PFN_UP(offset + size);
 468         }
 469         return pages;
 470 }
 471
 472 static int netvsc_get_slots(struct sk_buff *skb)
 473 {
 474         char *data = skb->data;
 475         unsigned int offset = offset_in_page(data);
 476         unsigned int len = skb_headlen(skb);
 477         int slots;
 478         int frag_slots;
 479
 480         slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
 481         frag_slots = count_skb_frag_slots(skb);
 482         return slots + frag_slots;
 483 }
 484
 485 static u32 net_checksum_info(struct sk_buff *skb)
 486 {
 487         if (skb->protocol == htons(ETH_P_IP)) {
 488                 struct iphdr *ip = ip_hdr(skb);
 489
 490                 if (ip->protocol == IPPROTO_TCP)
 491                         return TRANSPORT_INFO_IPV4_TCP;
 492                 else if (ip->protocol == IPPROTO_UDP)
 493                         return TRANSPORT_INFO_IPV4_UDP;
 494         } else {
 495                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 496
 497                 if (ip6->nexthdr == IPPROTO_TCP)
 498                         return TRANSPORT_INFO_IPV6_TCP;
 499                 else if (ip6->nexthdr == IPPROTO_UDP)
 500                         return TRANSPORT_INFO_IPV6_UDP;
 501         }
 502
 503         return TRANSPORT_INFO_NOT_IP;
 504 }
 505
 506 /* Send skb on the slave VF device. */
 507 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 508                           struct sk_buff *skb)
 509 {
 510         struct net_device_context *ndev_ctx = netdev_priv(net);
 511         unsigned int len = skb->len;
 512         int rc;
 513
 514         skb->dev = vf_netdev;
 515         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 516
 517         rc = dev_queue_xmit(skb);
 518         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 519                 struct netvsc_vf_pcpu_stats *pcpu_stats
 520                         = this_cpu_ptr(ndev_ctx->vf_stats);
 521
 522                 u64_stats_update_begin(&pcpu_stats->syncp);
 523                 pcpu_stats->tx_packets++;
 524                 pcpu_stats->tx_bytes += len;
 525                 u64_stats_update_end(&pcpu_stats->syncp);
 526         } else {
 527                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 528         }
 529
 530         return rc;
 531 }
 532
 533 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 534 {
 535         struct net_device_context *net_device_ctx = netdev_priv(net);
 536         struct hv_netvsc_packet *packet = NULL;
 537         int ret;
 538         unsigned int num_data_pgs;
 539         struct rndis_message *rndis_msg;
 540         struct net_device *vf_netdev;
 541         u32 rndis_msg_size;
 542         u32 hash;
 543         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 544
 545         /* if VF is present and up then redirect packets
 546          * already called with rcu_read_lock_bh
 547          */
 548         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 549         if (vf_netdev && netif_running(vf_netdev) &&
 550             !netpoll_tx_running(net))
 551                 return netvsc_vf_xmit(net, vf_netdev, skb);
 552
 553         /* We will atmost need two pages to describe the rndis
 554          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 555          * of pages in a single packet. If skb is scattered around
 556          * more pages we try linearizing it.
 557          */
 558
 559         num_data_pgs = netvsc_get_slots(skb) + 2;
 560
 561         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 562                 ++net_device_ctx->eth_stats.tx_scattered;
 563
 564                 if (skb_linearize(skb))
 565                         goto no_memory;
 566
 567                 num_data_pgs = netvsc_get_slots(skb) + 2;
 568                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 569                         ++net_device_ctx->eth_stats.tx_too_big;
 570                         goto drop;
 571                 }
 572         }
 573
 574         /*
 575          * Place the rndis header in the skb head room and
 576          * the skb->cb will be used for hv_netvsc_packet
 577          * structure.
 578          */
 579         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 580         if (ret)
 581                 goto no_memory;
 582
 583         /* Use the skb control buffer for building up the packet */
 584         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 585                         FIELD_SIZEOF(struct sk_buff, cb));
 586         packet = (struct hv_netvsc_packet *)skb->cb;
 587
 588         packet->q_idx = skb_get_queue_mapping(skb);
 589
 590         packet->total_data_buflen = skb->len;
 591         packet->total_bytes = skb->len;
 592         packet->total_packets = 1;
 593
 594         rndis_msg = (struct rndis_message *)skb->head;
 595
 596         /* Add the rndis header */
 597         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 598         rndis_msg->msg_len = packet->total_data_buflen;
 599
 600         rndis_msg->msg.pkt = (struct rndis_packet) {
 601                 .data_offset = sizeof(struct rndis_packet),
 602                 .data_len = packet->total_data_buflen,
 603                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 604         };
 605
 606         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 607
 608         hash = skb_get_hash_raw(skb);
 609         if (hash != 0 && net->real_num_tx_queues > 1) {
 610                 u32 *hash_info;
 611
 612                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 613                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 614                                           NBL_HASH_VALUE);
 615                 *hash_info = hash;
 616         }
 617
 618         if (skb_vlan_tag_present(skb)) {
 619                 struct ndis_pkt_8021q_info *vlan;
 620
 621                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 622                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 623                                      IEEE_8021Q_INFO);
 624
 625                 vlan->value = 0;
 626                 vlan->vlanid = skb_vlan_tag_get_id(skb);
 627                 vlan->cfi = skb_vlan_tag_get_cfi(skb);
 628                 vlan->pri = skb_vlan_tag_get_prio(skb);
 629         }
 630
 631         if (skb_is_gso(skb)) {
 632                 struct ndis_tcp_lso_info *lso_info;
 633
 634                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 635                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 636                                          TCP_LARGESEND_PKTINFO);
 637
 638                 lso_info->value = 0;
 639                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 640                 if (skb->protocol == htons(ETH_P_IP)) {
 641                         lso_info->lso_v2_transmit.ip_version =
 642                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 643                         ip_hdr(skb)->tot_len = 0;
 644                         ip_hdr(skb)->check = 0;
 645                         tcp_hdr(skb)->check =
 646                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 647                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 648                 } else {
 649                         lso_info->lso_v2_transmit.ip_version =
 650                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 651                         ipv6_hdr(skb)->payload_len = 0;
 652                         tcp_hdr(skb)->check =
 653                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 654                                                  &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 655                 }
 656                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 657                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 658         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 659                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 660                         struct ndis_tcp_ip_checksum_info *csum_info;
 661
 662                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 663                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 664                                                   TCPIP_CHKSUM_PKTINFO);
 665
 666                         csum_info->value = 0;
 667                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 668
 669                         if (skb->protocol == htons(ETH_P_IP)) {
 670                                 csum_info->transmit.is_ipv4 = 1;
 671
 672                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 673                                         csum_info->transmit.tcp_checksum = 1;
 674                                 else
 675                                         csum_info->transmit.udp_checksum = 1;
 676                         } else {
 677                                 csum_info->transmit.is_ipv6 = 1;
 678
 679                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 680                                         csum_info->transmit.tcp_checksum = 1;
 681                                 else
 682                                         csum_info->transmit.udp_checksum = 1;
 683                         }
 684                 } else {
 685                         /* Can't do offload of this type of checksum */
 686                         if (skb_checksum_help(skb))
 687                                 goto drop;
 688                 }
 689         }
 690
 691         /* Start filling in the page buffers with the rndis hdr */
 692         rndis_msg->msg_len += rndis_msg_size;
 693         packet->total_data_buflen = rndis_msg->msg_len;
 694         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 695                                                skb, packet, pb);
 696
 697         /* timestamp packet in software */
 698         skb_tx_timestamp(skb);
 699
 700         ret = netvsc_send(net, packet, rndis_msg, pb, skb);
 701         if (likely(ret == 0))
 702                 return NETDEV_TX_OK;
 703
 704         if (ret == -EAGAIN) {
 705                 ++net_device_ctx->eth_stats.tx_busy;
 706                 return NETDEV_TX_BUSY;
 707         }
 708
 709         if (ret == -ENOSPC)
 710                 ++net_device_ctx->eth_stats.tx_no_space;
 711
 712 drop:
 713         dev_kfree_skb_any(skb);
 714         net->stats.tx_dropped++;
 715
 716         return NETDEV_TX_OK;
 717
 718 no_memory:
 719         ++net_device_ctx->eth_stats.tx_no_memory;
 720         goto drop;
 721 }
 722
 723 /*
 724  * netvsc_linkstatus_callback - Link up/down notification
 725  */
 726 void netvsc_linkstatus_callback(struct net_device *net,
 727                                 struct rndis_message *resp)
 728 {
 729         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 730         struct net_device_context *ndev_ctx = netdev_priv(net);
 731         struct netvsc_reconfig *event;
 732         unsigned long flags;
 733
 734         /* Update the physical link speed when changing to another vSwitch */
 735         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 736                 u32 speed;
 737
 738                 speed = *(u32 *)((void *)indicate
 739                                  + indicate->status_buf_offset) / 10000;
 740                 ndev_ctx->speed = speed;
 741                 return;
 742         }
 743
 744         /* Handle these link change statuses below */
 745         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 746             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 747             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 748                 return;
 749
 750         if (net->reg_state != NETREG_REGISTERED)
 751                 return;
 752
 753         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 754         if (!event)
 755                 return;
 756         event->event = indicate->status;
 757
 758         spin_lock_irqsave(&ndev_ctx->lock, flags);
 759         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 760         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 761
 762         schedule_delayed_work(&ndev_ctx->dwork, 0);
 763 }
 764
 765 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 766 {
 767         struct iphdr *iph = (struct iphdr *)skb->data;
 768
 769         iph->check = 0;
 770         iph->check = ip_fast_csum(iph, iph->ihl);
 771 }
 772
 773 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 774                                              struct netvsc_channel *nvchan)
 775 {
 776         struct napi_struct *napi = &nvchan->napi;
 777         const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
 778         const struct ndis_tcp_ip_checksum_info *csum_info =
 779                                                 nvchan->rsc.csum_info;
 780         struct sk_buff *skb;
 781         int i;
 782
 783         skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 784         if (!skb)
 785                 return skb;
 786
 787         /*
 788          * Copy to skb. This copy is needed here since the memory pointed by
 789          * hv_netvsc_packet cannot be deallocated
 790          */
 791         for (i = 0; i < nvchan->rsc.cnt; i++)
 792                 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]);
 793
 794         skb->protocol = eth_type_trans(skb, net);
 795
 796         /* skb is already created with CHECKSUM_NONE */
 797         skb_checksum_none_assert(skb);
 798
 799         /* Incoming packets may have IP header checksum verified by the host.
 800          * They may not have IP header checksum computed after coalescing.
 801          * We compute it here if the flags are set, because on Linux, the IP
 802          * checksum is always checked.
 803          */
 804         if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
 805             csum_info->receive.ip_checksum_succeeded &&
 806             skb->protocol == htons(ETH_P_IP))
 807                 netvsc_comp_ipcsum(skb);
 808
 809         /* Do L4 checksum offload if enabled and present.
 810          */
 811         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 812                 if (csum_info->receive.tcp_checksum_succeeded ||
 813                     csum_info->receive.udp_checksum_succeeded)
 814                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 815         }
 816
 817         if (vlan) {
 818                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 819                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 820
 821                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 822                                        vlan_tci);
 823         }
 824
 825         return skb;
 826 }
 827
 828 /*
 829  * netvsc_recv_callback -  Callback when we receive a packet from the
 830  * "wire" on the specified device.
 831  */
 832 int netvsc_recv_callback(struct net_device *net,
 833                          struct netvsc_device *net_device,
 834                          struct netvsc_channel *nvchan)
 835 {
 836         struct net_device_context *net_device_ctx = netdev_priv(net);
 837         struct vmbus_channel *channel = nvchan->channel;
 838         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 839         struct sk_buff *skb;
 840         struct netvsc_stats *rx_stats;
 841
 842         if (net->reg_state != NETREG_REGISTERED)
 843                 return NVSP_STAT_FAIL;
 844
 845         /* Allocate a skb - TODO direct I/O to pages? */
 846         skb = netvsc_alloc_recv_skb(net, nvchan);
 847
 848         if (unlikely(!skb)) {
 849                 ++net_device_ctx->eth_stats.rx_no_memory;
 850                 rcu_read_unlock();
 851                 return NVSP_STAT_FAIL;
 852         }
 853
 854         skb_record_rx_queue(skb, q_idx);
 855
 856         /*
 857          * Even if injecting the packet, record the statistics
 858          * on the synthetic device because modifying the VF device
 859          * statistics will not work correctly.
 860          */
 861         rx_stats = &nvchan->rx_stats;
 862         u64_stats_update_begin(&rx_stats->syncp);
 863         rx_stats->packets++;
 864         rx_stats->bytes += nvchan->rsc.pktlen;
 865
 866         if (skb->pkt_type == PACKET_BROADCAST)
 867                 ++rx_stats->broadcast;
 868         else if (skb->pkt_type == PACKET_MULTICAST)
 869                 ++rx_stats->multicast;
 870         u64_stats_update_end(&rx_stats->syncp);
 871
 872         napi_gro_receive(&nvchan->napi, skb);
 873         return NVSP_STAT_SUCCESS;
 874 }
 875
 876 static void netvsc_get_drvinfo(struct net_device *net,
 877                                struct ethtool_drvinfo *info)
 878 {
 879         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 880         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 881 }
 882
 883 static void netvsc_get_channels(struct net_device *net,
 884                                 struct ethtool_channels *channel)
 885 {
 886         struct net_device_context *net_device_ctx = netdev_priv(net);
 887         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 888
 889         if (nvdev) {
 890                 channel->max_combined   = nvdev->max_chn;
 891                 channel->combined_count = nvdev->num_chn;
 892         }
 893 }
 894
 895 /* Alloc struct netvsc_device_info, and initialize it from either existing
 896  * struct netvsc_device, or from default values.
 897  */
 898 static struct netvsc_device_info *netvsc_devinfo_get
 899                         (struct netvsc_device *nvdev)
 900 {
 901         struct netvsc_device_info *dev_info;
 902
 903         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
 904
 905         if (!dev_info)
 906                 return NULL;
 907
 908         if (nvdev) {
 909                 dev_info->num_chn = nvdev->num_chn;
 910                 dev_info->send_sections = nvdev->send_section_cnt;
 911                 dev_info->send_section_size = nvdev->send_section_size;
 912                 dev_info->recv_sections = nvdev->recv_section_cnt;
 913                 dev_info->recv_section_size = nvdev->recv_section_size;
 914
 915                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
 916                        NETVSC_HASH_KEYLEN);
 917         } else {
 918                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
 919                 dev_info->send_sections = NETVSC_DEFAULT_TX;
 920                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
 921                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
 922                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
 923         }
 924
 925         return dev_info;
 926 }
 927
 928 static int netvsc_detach(struct net_device *ndev,
 929                          struct netvsc_device *nvdev)
 930 {
 931         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 932         struct hv_device *hdev = ndev_ctx->device_ctx;
 933         int ret;
 934
 935         /* Don't try continuing to try and setup sub channels */
 936         if (cancel_work_sync(&nvdev->subchan_work))
 937                 nvdev->num_chn = 1;
 938
 939         /* If device was up (receiving) then shutdown */
 940         if (netif_running(ndev)) {
 941                 netvsc_tx_disable(nvdev, ndev);
 942
 943                 ret = rndis_filter_close(nvdev);
 944                 if (ret) {
 945                         netdev_err(ndev,
 946                                    "unable to close device (ret %d).\n", ret);
 947                         return ret;
 948                 }
 949
 950                 ret = netvsc_wait_until_empty(nvdev);
 951                 if (ret) {
 952                         netdev_err(ndev,
 953                                    "Ring buffer not empty after closing rndis\n");
 954                         return ret;
 955                 }
 956         }
 957
 958         netif_device_detach(ndev);
 959
 960         rndis_filter_device_remove(hdev, nvdev);
 961
 962         return 0;
 963 }
 964
 965 static int netvsc_attach(struct net_device *ndev,
 966                          struct netvsc_device_info *dev_info)
 967 {
 968         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 969         struct hv_device *hdev = ndev_ctx->device_ctx;
 970         struct netvsc_device *nvdev;
 971         struct rndis_device *rdev;
 972         int ret;
 973
 974         nvdev = rndis_filter_device_add(hdev, dev_info);
 975         if (IS_ERR(nvdev))
 976                 return PTR_ERR(nvdev);
 977
 978         if (nvdev->num_chn > 1) {
 979                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
 980
 981                 /* if unavailable, just proceed with one queue */
 982                 if (ret) {
 983                         nvdev->max_chn = 1;
 984                         nvdev->num_chn = 1;
 985                 }
 986         }
 987
 988         /* In any case device is now ready */
 989         netif_device_attach(ndev);
 990
 991         /* Note: enable and attach happen when sub-channels setup */
 992         netif_carrier_off(ndev);
 993
 994         if (netif_running(ndev)) {
 995                 ret = rndis_filter_open(nvdev);
 996                 if (ret)
 997                         return ret;
 998
 999                 rdev = nvdev->extension;
1000                 if (!rdev->link_state)
1001                         netif_carrier_on(ndev);
1002         }
1003
1004         return 0;
1005 }
1006
1007 static int netvsc_set_channels(struct net_device *net,
1008                                struct ethtool_channels *channels)
1009 {
1010         struct net_device_context *net_device_ctx = netdev_priv(net);
1011         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1012         unsigned int orig, count = channels->combined_count;
1013         struct netvsc_device_info *device_info;
1014         int ret;
1015
1016         /* We do not support separate count for rx, tx, or other */
1017         if (count == 0 ||
1018             channels->rx_count || channels->tx_count || channels->other_count)
1019                 return -EINVAL;
1020
1021         if (!nvdev || nvdev->destroy)
1022                 return -ENODEV;
1023
1024         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1025                 return -EINVAL;
1026
1027         if (count > nvdev->max_chn)
1028                 return -EINVAL;
1029
1030         orig = nvdev->num_chn;
1031
1032         device_info = netvsc_devinfo_get(nvdev);
1033
1034         if (!device_info)
1035                 return -ENOMEM;
1036
1037         device_info->num_chn = count;
1038
1039         ret = netvsc_detach(net, nvdev);
1040         if (ret)
1041                 goto out;
1042
1043         ret = netvsc_attach(net, device_info);
1044         if (ret) {
1045                 device_info->num_chn = orig;
1046                 if (netvsc_attach(net, device_info))
1047                         netdev_err(net, "restoring channel setting failed\n");
1048         }
1049
1050 out:
1051         kfree(device_info);
1052         return ret;
1053 }
1054
1055 static bool
1056 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
1057 {
1058         struct ethtool_link_ksettings diff1 = *cmd;
1059         struct ethtool_link_ksettings diff2 = {};
1060
1061         diff1.base.speed = 0;
1062         diff1.base.duplex = 0;
1063         /* advertising and cmd are usually set */
1064         ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1065         diff1.base.cmd = 0;
1066         /* We set port to PORT_OTHER */
1067         diff2.base.port = PORT_OTHER;
1068
1069         return !memcmp(&diff1, &diff2, sizeof(diff1));
1070 }
1071
1072 static void netvsc_init_settings(struct net_device *dev)
1073 {
1074         struct net_device_context *ndc = netdev_priv(dev);
1075
1076         ndc->l4_hash = HV_DEFAULT_L4HASH;
1077
1078         ndc->speed = SPEED_UNKNOWN;
1079         ndc->duplex = DUPLEX_FULL;
1080
1081         dev->features = NETIF_F_LRO;
1082 }
1083
1084 static int netvsc_get_link_ksettings(struct net_device *dev,
1085                                      struct ethtool_link_ksettings *cmd)
1086 {
1087         struct net_device_context *ndc = netdev_priv(dev);
1088
1089         cmd->base.speed = ndc->speed;
1090         cmd->base.duplex = ndc->duplex;
1091         cmd->base.port = PORT_OTHER;
1092
1093         return 0;
1094 }
1095
1096 static int netvsc_set_link_ksettings(struct net_device *dev,
1097                                      const struct ethtool_link_ksettings *cmd)
1098 {
1099         struct net_device_context *ndc = netdev_priv(dev);
1100         u32 speed;
1101
1102         speed = cmd->base.speed;
1103         if (!ethtool_validate_speed(speed) ||
1104             !ethtool_validate_duplex(cmd->base.duplex) ||
1105             !netvsc_validate_ethtool_ss_cmd(cmd))
1106                 return -EINVAL;
1107
1108         ndc->speed = speed;
1109         ndc->duplex = cmd->base.duplex;
1110
1111         return 0;
1112 }
1113
1114 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1115 {
1116         struct net_device_context *ndevctx = netdev_priv(ndev);
1117         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1118         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1119         int orig_mtu = ndev->mtu;
1120         struct netvsc_device_info *device_info;
1121         int ret = 0;
1122
1123         if (!nvdev || nvdev->destroy)
1124                 return -ENODEV;
1125
1126         device_info = netvsc_devinfo_get(nvdev);
1127
1128         if (!device_info)
1129                 return -ENOMEM;
1130
1131         /* Change MTU of underlying VF netdev first. */
1132         if (vf_netdev) {
1133                 ret = dev_set_mtu(vf_netdev, mtu);
1134                 if (ret)
1135                         goto out;
1136         }
1137
1138         ret = netvsc_detach(ndev, nvdev);
1139         if (ret)
1140                 goto rollback_vf;
1141
1142         ndev->mtu = mtu;
1143
1144         ret = netvsc_attach(ndev, device_info);
1145         if (!ret)
1146                 goto out;
1147
1148         /* Attempt rollback to original MTU */
1149         ndev->mtu = orig_mtu;
1150
1151         if (netvsc_attach(ndev, device_info))
1152                 netdev_err(ndev, "restoring mtu failed\n");
1153 rollback_vf:
1154         if (vf_netdev)
1155                 dev_set_mtu(vf_netdev, orig_mtu);
1156
1157 out:
1158         kfree(device_info);
1159         return ret;
1160 }
1161
1162 static void netvsc_get_vf_stats(struct net_device *net,
1163                                 struct netvsc_vf_pcpu_stats *tot)
1164 {
1165         struct net_device_context *ndev_ctx = netdev_priv(net);
1166         int i;
1167
1168         memset(tot, 0, sizeof(*tot));
1169
1170         for_each_possible_cpu(i) {
1171                 const struct netvsc_vf_pcpu_stats *stats
1172                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1173                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1174                 unsigned int start;
1175
1176                 do {
1177                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1178                         rx_packets = stats->rx_packets;
1179                         tx_packets = stats->tx_packets;
1180                         rx_bytes = stats->rx_bytes;
1181                         tx_bytes = stats->tx_bytes;
1182                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1183
1184                 tot->rx_packets += rx_packets;
1185                 tot->tx_packets += tx_packets;
1186                 tot->rx_bytes   += rx_bytes;
1187                 tot->tx_bytes   += tx_bytes;
1188                 tot->tx_dropped += stats->tx_dropped;
1189         }
1190 }
1191
1192 static void netvsc_get_pcpu_stats(struct net_device *net,
1193                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1194 {
1195         struct net_device_context *ndev_ctx = netdev_priv(net);
1196         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1197         int i;
1198
1199         /* fetch percpu stats of vf */
1200         for_each_possible_cpu(i) {
1201                 const struct netvsc_vf_pcpu_stats *stats =
1202                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1203                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1204                 unsigned int start;
1205
1206                 do {
1207                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1208                         this_tot->vf_rx_packets = stats->rx_packets;
1209                         this_tot->vf_tx_packets = stats->tx_packets;
1210                         this_tot->vf_rx_bytes = stats->rx_bytes;
1211                         this_tot->vf_tx_bytes = stats->tx_bytes;
1212                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1213                 this_tot->rx_packets = this_tot->vf_rx_packets;
1214                 this_tot->tx_packets = this_tot->vf_tx_packets;
1215                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1216                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1217         }
1218
1219         /* fetch percpu stats of netvsc */
1220         for (i = 0; i < nvdev->num_chn; i++) {
1221                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1222                 const struct netvsc_stats *stats;
1223                 struct netvsc_ethtool_pcpu_stats *this_tot =
1224                         &pcpu_tot[nvchan->channel->target_cpu];
1225                 u64 packets, bytes;
1226                 unsigned int start;
1227
1228                 stats = &nvchan->tx_stats;
1229                 do {
1230                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1231                         packets = stats->packets;
1232                         bytes = stats->bytes;
1233                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1234
1235                 this_tot->tx_bytes      += bytes;
1236                 this_tot->tx_packets    += packets;
1237
1238                 stats = &nvchan->rx_stats;
1239                 do {
1240                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1241                         packets = stats->packets;
1242                         bytes = stats->bytes;
1243                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1244
1245                 this_tot->rx_bytes      += bytes;
1246                 this_tot->rx_packets    += packets;
1247         }
1248 }
1249
1250 static void netvsc_get_stats64(struct net_device *net,
1251                                struct rtnl_link_stats64 *t)
1252 {
1253         struct net_device_context *ndev_ctx = netdev_priv(net);
1254         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1255         struct netvsc_vf_pcpu_stats vf_tot;
1256         int i;
1257
1258         if (!nvdev)
1259                 return;
1260
1261         netdev_stats_to_stats64(t, &net->stats);
1262
1263         netvsc_get_vf_stats(net, &vf_tot);
1264         t->rx_packets += vf_tot.rx_packets;
1265         t->tx_packets += vf_tot.tx_packets;
1266         t->rx_bytes   += vf_tot.rx_bytes;
1267         t->tx_bytes   += vf_tot.tx_bytes;
1268         t->tx_dropped += vf_tot.tx_dropped;
1269
1270         for (i = 0; i < nvdev->num_chn; i++) {
1271                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1272                 const struct netvsc_stats *stats;
1273                 u64 packets, bytes, multicast;
1274                 unsigned int start;
1275
1276                 stats = &nvchan->tx_stats;
1277                 do {
1278                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1279                         packets = stats->packets;
1280                         bytes = stats->bytes;
1281                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1282
1283                 t->tx_bytes     += bytes;
1284                 t->tx_packets   += packets;
1285
1286                 stats = &nvchan->rx_stats;
1287                 do {
1288                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1289                         packets = stats->packets;
1290                         bytes = stats->bytes;
1291                         multicast = stats->multicast + stats->broadcast;
1292                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1293
1294                 t->rx_bytes     += bytes;
1295                 t->rx_packets   += packets;
1296                 t->multicast    += multicast;
1297         }
1298 }
1299
1300 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1301 {
1302         struct net_device_context *ndc = netdev_priv(ndev);
1303         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1304         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1305         struct sockaddr *addr = p;
1306         int err;
1307
1308         err = eth_prepare_mac_addr_change(ndev, p);
1309         if (err)
1310                 return err;
1311
1312         if (!nvdev)
1313                 return -ENODEV;
1314
1315         if (vf_netdev) {
1316                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1317                 if (err)
1318                         return err;
1319         }
1320
1321         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1322         if (!err) {
1323                 eth_commit_mac_addr_change(ndev, p);
1324         } else if (vf_netdev) {
1325                 /* rollback change on VF */
1326                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1327                 dev_set_mac_address(vf_netdev, addr, NULL);
1328         }
1329
1330         return err;
1331 }
1332
1333 static const struct {
1334         char name[ETH_GSTRING_LEN];
1335         u16 offset;
1336 } netvsc_stats[] = {
1337         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1338         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1339         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1340         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1341         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1342         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1343         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1344         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1345         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1346         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1347 }, pcpu_stats[] = {
1348         { "cpu%u_rx_packets",
1349                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1350         { "cpu%u_rx_bytes",
1351                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1352         { "cpu%u_tx_packets",
1353                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1354         { "cpu%u_tx_bytes",
1355                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1356         { "cpu%u_vf_rx_packets",
1357                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1358         { "cpu%u_vf_rx_bytes",
1359                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1360         { "cpu%u_vf_tx_packets",
1361                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1362         { "cpu%u_vf_tx_bytes",
1363                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1364 }, vf_stats[] = {
1365         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1366         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1367         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1368         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1369         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1370 };
1371
1372 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1373 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1374
1375 /* statistics per queue (rx/tx packets/bytes) */
1376 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1377
1378 /* 4 statistics per queue (rx/tx packets/bytes) */
1379 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1380
1381 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1382 {
1383         struct net_device_context *ndc = netdev_priv(dev);
1384         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1385
1386         if (!nvdev)
1387                 return -ENODEV;
1388
1389         switch (string_set) {
1390         case ETH_SS_STATS:
1391                 return NETVSC_GLOBAL_STATS_LEN
1392                         + NETVSC_VF_STATS_LEN
1393                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1394                         + NETVSC_PCPU_STATS_LEN;
1395         default:
1396                 return -EINVAL;
1397         }
1398 }
1399
1400 static void netvsc_get_ethtool_stats(struct net_device *dev,
1401                                      struct ethtool_stats *stats, u64 *data)
1402 {
1403         struct net_device_context *ndc = netdev_priv(dev);
1404         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1405         const void *nds = &ndc->eth_stats;
1406         const struct netvsc_stats *qstats;
1407         struct netvsc_vf_pcpu_stats sum;
1408         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1409         unsigned int start;
1410         u64 packets, bytes;
1411         int i, j, cpu;
1412
1413         if (!nvdev)
1414                 return;
1415
1416         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1417                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1418
1419         netvsc_get_vf_stats(dev, &sum);
1420         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1421                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1422
1423         for (j = 0; j < nvdev->num_chn; j++) {
1424                 qstats = &nvdev->chan_table[j].tx_stats;
1425
1426                 do {
1427                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1428                         packets = qstats->packets;
1429                         bytes = qstats->bytes;
1430                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1431                 data[i++] = packets;
1432                 data[i++] = bytes;
1433
1434                 qstats = &nvdev->chan_table[j].rx_stats;
1435                 do {
1436                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1437                         packets = qstats->packets;
1438                         bytes = qstats->bytes;
1439                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1440                 data[i++] = packets;
1441                 data[i++] = bytes;
1442         }
1443
1444         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1445                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1446                                   GFP_KERNEL);
1447         netvsc_get_pcpu_stats(dev, pcpu_sum);
1448         for_each_present_cpu(cpu) {
1449                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1450
1451                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1452                         data[i++] = *(u64 *)((void *)this_sum
1453                                              + pcpu_stats[j].offset);
1454         }
1455         kvfree(pcpu_sum);
1456 }
1457
1458 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1459 {
1460         struct net_device_context *ndc = netdev_priv(dev);
1461         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1462         u8 *p = data;
1463         int i, cpu;
1464
1465         if (!nvdev)
1466                 return;
1467
1468         switch (stringset) {
1469         case ETH_SS_STATS:
1470                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1471                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1472                         p += ETH_GSTRING_LEN;
1473                 }
1474
1475                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1476                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1477                         p += ETH_GSTRING_LEN;
1478                 }
1479
1480                 for (i = 0; i < nvdev->num_chn; i++) {
1481                         sprintf(p, "tx_queue_%u_packets", i);
1482                         p += ETH_GSTRING_LEN;
1483                         sprintf(p, "tx_queue_%u_bytes", i);
1484                         p += ETH_GSTRING_LEN;
1485                         sprintf(p, "rx_queue_%u_packets", i);
1486                         p += ETH_GSTRING_LEN;
1487                         sprintf(p, "rx_queue_%u_bytes", i);
1488                         p += ETH_GSTRING_LEN;
1489                 }
1490
1491                 for_each_present_cpu(cpu) {
1492                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1493                                 sprintf(p, pcpu_stats[i].name, cpu);
1494                                 p += ETH_GSTRING_LEN;
1495                         }
1496                 }
1497
1498                 break;
1499         }
1500 }
1501
1502 static int
1503 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1504                          struct ethtool_rxnfc *info)
1505 {
1506         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1507
1508         info->data = RXH_IP_SRC | RXH_IP_DST;
1509
1510         switch (info->flow_type) {
1511         case TCP_V4_FLOW:
1512                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1513                         info->data |= l4_flag;
1514
1515                 break;
1516
1517         case TCP_V6_FLOW:
1518                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1519                         info->data |= l4_flag;
1520
1521                 break;
1522
1523         case UDP_V4_FLOW:
1524                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1525                         info->data |= l4_flag;
1526
1527                 break;
1528
1529         case UDP_V6_FLOW:
1530                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1531                         info->data |= l4_flag;
1532
1533                 break;
1534
1535         case IPV4_FLOW:
1536         case IPV6_FLOW:
1537                 break;
1538         default:
1539                 info->data = 0;
1540                 break;
1541         }
1542
1543         return 0;
1544 }
1545
1546 static int
1547 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1548                  u32 *rules)
1549 {
1550         struct net_device_context *ndc = netdev_priv(dev);
1551         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1552
1553         if (!nvdev)
1554                 return -ENODEV;
1555
1556         switch (info->cmd) {
1557         case ETHTOOL_GRXRINGS:
1558                 info->data = nvdev->num_chn;
1559                 return 0;
1560
1561         case ETHTOOL_GRXFH:
1562                 return netvsc_get_rss_hash_opts(ndc, info);
1563         }
1564         return -EOPNOTSUPP;
1565 }
1566
1567 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1568                                     struct ethtool_rxnfc *info)
1569 {
1570         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1571                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1572                 switch (info->flow_type) {
1573                 case TCP_V4_FLOW:
1574                         ndc->l4_hash |= HV_TCP4_L4HASH;
1575                         break;
1576
1577                 case TCP_V6_FLOW:
1578                         ndc->l4_hash |= HV_TCP6_L4HASH;
1579                         break;
1580
1581                 case UDP_V4_FLOW:
1582                         ndc->l4_hash |= HV_UDP4_L4HASH;
1583                         break;
1584
1585                 case UDP_V6_FLOW:
1586                         ndc->l4_hash |= HV_UDP6_L4HASH;
1587                         break;
1588
1589                 default:
1590                         return -EOPNOTSUPP;
1591                 }
1592
1593                 return 0;
1594         }
1595
1596         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1597                 switch (info->flow_type) {
1598                 case TCP_V4_FLOW:
1599                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1600                         break;
1601
1602                 case TCP_V6_FLOW:
1603                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1604                         break;
1605
1606                 case UDP_V4_FLOW:
1607                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1608                         break;
1609
1610                 case UDP_V6_FLOW:
1611                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1612                         break;
1613
1614                 default:
1615                         return -EOPNOTSUPP;
1616                 }
1617
1618                 return 0;
1619         }
1620
1621         return -EOPNOTSUPP;
1622 }
1623
1624 static int
1625 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1626 {
1627         struct net_device_context *ndc = netdev_priv(ndev);
1628
1629         if (info->cmd == ETHTOOL_SRXFH)
1630                 return netvsc_set_rss_hash_opts(ndc, info);
1631
1632         return -EOPNOTSUPP;
1633 }
1634
1635 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1636 {
1637         return NETVSC_HASH_KEYLEN;
1638 }
1639
1640 static u32 netvsc_rss_indir_size(struct net_device *dev)
1641 {
1642         return ITAB_NUM;
1643 }
1644
1645 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1646                            u8 *hfunc)
1647 {
1648         struct net_device_context *ndc = netdev_priv(dev);
1649         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1650         struct rndis_device *rndis_dev;
1651         int i;
1652
1653         if (!ndev)
1654                 return -ENODEV;
1655
1656         if (hfunc)
1657                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1658
1659         rndis_dev = ndev->extension;
1660         if (indir) {
1661                 for (i = 0; i < ITAB_NUM; i++)
1662                         indir[i] = rndis_dev->rx_table[i];
1663         }
1664
1665         if (key)
1666                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1667
1668         return 0;
1669 }
1670
1671 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1672                            const u8 *key, const u8 hfunc)
1673 {
1674         struct net_device_context *ndc = netdev_priv(dev);
1675         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1676         struct rndis_device *rndis_dev;
1677         int i;
1678
1679         if (!ndev)
1680                 return -ENODEV;
1681
1682         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1683                 return -EOPNOTSUPP;
1684
1685         rndis_dev = ndev->extension;
1686         if (indir) {
1687                 for (i = 0; i < ITAB_NUM; i++)
1688                         if (indir[i] >= ndev->num_chn)
1689                                 return -EINVAL;
1690
1691                 for (i = 0; i < ITAB_NUM; i++)
1692                         rndis_dev->rx_table[i] = indir[i];
1693         }
1694
1695         if (!key) {
1696                 if (!indir)
1697                         return 0;
1698
1699                 key = rndis_dev->rss_key;
1700         }
1701
1702         return rndis_filter_set_rss_param(rndis_dev, key);
1703 }
1704
1705 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1706  * It does have pre-allocated receive area which is divided into sections.
1707  */
1708 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1709                                    struct ethtool_ringparam *ring)
1710 {
1711         u32 max_buf_size;
1712
1713         ring->rx_pending = nvdev->recv_section_cnt;
1714         ring->tx_pending = nvdev->send_section_cnt;
1715
1716         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1717                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1718         else
1719                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1720
1721         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1722         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1723                 / nvdev->send_section_size;
1724 }
1725
1726 static void netvsc_get_ringparam(struct net_device *ndev,
1727                                  struct ethtool_ringparam *ring)
1728 {
1729         struct net_device_context *ndevctx = netdev_priv(ndev);
1730         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1731
1732         if (!nvdev)
1733                 return;
1734
1735         __netvsc_get_ringparam(nvdev, ring);
1736 }
1737
1738 static int netvsc_set_ringparam(struct net_device *ndev,
1739                                 struct ethtool_ringparam *ring)
1740 {
1741         struct net_device_context *ndevctx = netdev_priv(ndev);
1742         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1743         struct netvsc_device_info *device_info;
1744         struct ethtool_ringparam orig;
1745         u32 new_tx, new_rx;
1746         int ret = 0;
1747
1748         if (!nvdev || nvdev->destroy)
1749                 return -ENODEV;
1750
1751         memset(&orig, 0, sizeof(orig));
1752         __netvsc_get_ringparam(nvdev, &orig);
1753
1754         new_tx = clamp_t(u32, ring->tx_pending,
1755                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1756         new_rx = clamp_t(u32, ring->rx_pending,
1757                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1758
1759         if (new_tx == orig.tx_pending &&
1760             new_rx == orig.rx_pending)
1761                 return 0;        /* no change */
1762
1763         device_info = netvsc_devinfo_get(nvdev);
1764
1765         if (!device_info)
1766                 return -ENOMEM;
1767
1768         device_info->send_sections = new_tx;
1769         device_info->recv_sections = new_rx;
1770
1771         ret = netvsc_detach(ndev, nvdev);
1772         if (ret)
1773                 goto out;
1774
1775         ret = netvsc_attach(ndev, device_info);
1776         if (ret) {
1777                 device_info->send_sections = orig.tx_pending;
1778                 device_info->recv_sections = orig.rx_pending;
1779
1780                 if (netvsc_attach(ndev, device_info))
1781                         netdev_err(ndev, "restoring ringparam failed");
1782         }
1783
1784 out:
1785         kfree(device_info);
1786         return ret;
1787 }
1788
1789 static int netvsc_set_features(struct net_device *ndev,
1790                                netdev_features_t features)
1791 {
1792         netdev_features_t change = features ^ ndev->features;
1793         struct net_device_context *ndevctx = netdev_priv(ndev);
1794         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1795         struct ndis_offload_params offloads;
1796
1797         if (!nvdev || nvdev->destroy)
1798                 return -ENODEV;
1799
1800         if (!(change & NETIF_F_LRO))
1801                 return 0;
1802
1803         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1804
1805         if (features & NETIF_F_LRO) {
1806                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1807                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1808         } else {
1809                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1810                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1811         }
1812
1813         return rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1814 }
1815
1816 static u32 netvsc_get_msglevel(struct net_device *ndev)
1817 {
1818         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1819
1820         return ndev_ctx->msg_enable;
1821 }
1822
1823 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1824 {
1825         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1826
1827         ndev_ctx->msg_enable = val;
1828 }
1829
1830 static const struct ethtool_ops ethtool_ops = {
1831         .get_drvinfo    = netvsc_get_drvinfo,
1832         .get_msglevel   = netvsc_get_msglevel,
1833         .set_msglevel   = netvsc_set_msglevel,
1834         .get_link       = ethtool_op_get_link,
1835         .get_ethtool_stats = netvsc_get_ethtool_stats,
1836         .get_sset_count = netvsc_get_sset_count,
1837         .get_strings    = netvsc_get_strings,
1838         .get_channels   = netvsc_get_channels,
1839         .set_channels   = netvsc_set_channels,
1840         .get_ts_info    = ethtool_op_get_ts_info,
1841         .get_rxnfc      = netvsc_get_rxnfc,
1842         .set_rxnfc      = netvsc_set_rxnfc,
1843         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1844         .get_rxfh_indir_size = netvsc_rss_indir_size,
1845         .get_rxfh       = netvsc_get_rxfh,
1846         .set_rxfh       = netvsc_set_rxfh,
1847         .get_link_ksettings = netvsc_get_link_ksettings,
1848         .set_link_ksettings = netvsc_set_link_ksettings,
1849         .get_ringparam  = netvsc_get_ringparam,
1850         .set_ringparam  = netvsc_set_ringparam,
1851 };
1852
1853 static const struct net_device_ops device_ops = {
1854         .ndo_open =                     netvsc_open,
1855         .ndo_stop =                     netvsc_close,
1856         .ndo_start_xmit =               netvsc_start_xmit,
1857         .ndo_change_rx_flags =          netvsc_change_rx_flags,
1858         .ndo_set_rx_mode =              netvsc_set_rx_mode,
1859         .ndo_set_features =             netvsc_set_features,
1860         .ndo_change_mtu =               netvsc_change_mtu,
1861         .ndo_validate_addr =            eth_validate_addr,
1862         .ndo_set_mac_address =          netvsc_set_mac_addr,
1863         .ndo_select_queue =             netvsc_select_queue,
1864         .ndo_get_stats64 =              netvsc_get_stats64,
1865 };
1866
1867 /*
1868  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1869  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1870  * present send GARP packet to network peers with netif_notify_peers().
1871  */
1872 static void netvsc_link_change(struct work_struct *w)
1873 {
1874         struct net_device_context *ndev_ctx =
1875                 container_of(w, struct net_device_context, dwork.work);
1876         struct hv_device *device_obj = ndev_ctx->device_ctx;
1877         struct net_device *net = hv_get_drvdata(device_obj);
1878         struct netvsc_device *net_device;
1879         struct rndis_device *rdev;
1880         struct netvsc_reconfig *event = NULL;
1881         bool notify = false, reschedule = false;
1882         unsigned long flags, next_reconfig, delay;
1883
1884         /* if changes are happening, comeback later */
1885         if (!rtnl_trylock()) {
1886                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1887                 return;
1888         }
1889
1890         net_device = rtnl_dereference(ndev_ctx->nvdev);
1891         if (!net_device)
1892                 goto out_unlock;
1893
1894         rdev = net_device->extension;
1895
1896         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1897         if (time_is_after_jiffies(next_reconfig)) {
1898                 /* link_watch only sends one notification with current state
1899                  * per second, avoid doing reconfig more frequently. Handle
1900                  * wrap around.
1901                  */
1902                 delay = next_reconfig - jiffies;
1903                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1904                 schedule_delayed_work(&ndev_ctx->dwork, delay);
1905                 goto out_unlock;
1906         }
1907         ndev_ctx->last_reconfig = jiffies;
1908
1909         spin_lock_irqsave(&ndev_ctx->lock, flags);
1910         if (!list_empty(&ndev_ctx->reconfig_events)) {
1911                 event = list_first_entry(&ndev_ctx->reconfig_events,
1912                                          struct netvsc_reconfig, list);
1913                 list_del(&event->list);
1914                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1915         }
1916         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1917
1918         if (!event)
1919                 goto out_unlock;
1920
1921         switch (event->event) {
1922                 /* Only the following events are possible due to the check in
1923                  * netvsc_linkstatus_callback()
1924                  */
1925         case RNDIS_STATUS_MEDIA_CONNECT:
1926                 if (rdev->link_state) {
1927                         rdev->link_state = false;
1928                         netif_carrier_on(net);
1929                         netvsc_tx_enable(net_device, net);
1930                 } else {
1931                         notify = true;
1932                 }
1933                 kfree(event);
1934                 break;
1935         case RNDIS_STATUS_MEDIA_DISCONNECT:
1936                 if (!rdev->link_state) {
1937                         rdev->link_state = true;
1938                         netif_carrier_off(net);
1939                         netvsc_tx_disable(net_device, net);
1940                 }
1941                 kfree(event);
1942                 break;
1943         case RNDIS_STATUS_NETWORK_CHANGE:
1944                 /* Only makes sense if carrier is present */
1945                 if (!rdev->link_state) {
1946                         rdev->link_state = true;
1947                         netif_carrier_off(net);
1948                         netvsc_tx_disable(net_device, net);
1949                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
1950                         spin_lock_irqsave(&ndev_ctx->lock, flags);
1951                         list_add(&event->list, &ndev_ctx->reconfig_events);
1952                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1953                         reschedule = true;
1954                 }
1955                 break;
1956         }
1957
1958         rtnl_unlock();
1959
1960         if (notify)
1961                 netdev_notify_peers(net);
1962
1963         /* link_watch only sends one notification with current state per
1964          * second, handle next reconfig event in 2 seconds.
1965          */
1966         if (reschedule)
1967                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1968
1969         return;
1970
1971 out_unlock:
1972         rtnl_unlock();
1973 }
1974
1975 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1976 {
1977         struct net_device_context *net_device_ctx;
1978         struct net_device *dev;
1979
1980         dev = netdev_master_upper_dev_get(vf_netdev);
1981         if (!dev || dev->netdev_ops != &device_ops)
1982                 return NULL;    /* not a netvsc device */
1983
1984         net_device_ctx = netdev_priv(dev);
1985         if (!rtnl_dereference(net_device_ctx->nvdev))
1986                 return NULL;    /* device is removed */
1987
1988         return dev;
1989 }
1990
1991 /* Called when VF is injecting data into network stack.
1992  * Change the associated network device from VF to netvsc.
1993  * note: already called with rcu_read_lock
1994  */
1995 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1996 {
1997         struct sk_buff *skb = *pskb;
1998         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1999         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2000         struct netvsc_vf_pcpu_stats *pcpu_stats
2001                  = this_cpu_ptr(ndev_ctx->vf_stats);
2002
2003         skb->dev = ndev;
2004
2005         u64_stats_update_begin(&pcpu_stats->syncp);
2006         pcpu_stats->rx_packets++;
2007         pcpu_stats->rx_bytes += skb->len;
2008         u64_stats_update_end(&pcpu_stats->syncp);
2009
2010         return RX_HANDLER_ANOTHER;
2011 }
2012
2013 static int netvsc_vf_join(struct net_device *vf_netdev,
2014                           struct net_device *ndev)
2015 {
2016         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2017         int ret;
2018
2019         ret = netdev_rx_handler_register(vf_netdev,
2020                                          netvsc_vf_handle_frame, ndev);
2021         if (ret != 0) {
2022                 netdev_err(vf_netdev,
2023                            "can not register netvsc VF receive handler (err = %d)\n",
2024                            ret);
2025                 goto rx_handler_failed;
2026         }
2027
2028         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2029                                            NULL, NULL, NULL);
2030         if (ret != 0) {
2031                 netdev_err(vf_netdev,
2032                            "can not set master device %s (err = %d)\n",
2033                            ndev->name, ret);
2034                 goto upper_link_failed;
2035         }
2036
2037         /* set slave flag before open to prevent IPv6 addrconf */
2038         vf_netdev->flags |= IFF_SLAVE;
2039
2040         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2041
2042         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2043
2044         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2045         return 0;
2046
2047 upper_link_failed:
2048         netdev_rx_handler_unregister(vf_netdev);
2049 rx_handler_failed:
2050         return ret;
2051 }
2052
2053 static void __netvsc_vf_setup(struct net_device *ndev,
2054                               struct net_device *vf_netdev)
2055 {
2056         int ret;
2057
2058         /* Align MTU of VF with master */
2059         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2060         if (ret)
2061                 netdev_warn(vf_netdev,
2062                             "unable to change mtu to %u\n", ndev->mtu);
2063
2064         /* set multicast etc flags on VF */
2065         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2066
2067         /* sync address list from ndev to VF */
2068         netif_addr_lock_bh(ndev);
2069         dev_uc_sync(vf_netdev, ndev);
2070         dev_mc_sync(vf_netdev, ndev);
2071         netif_addr_unlock_bh(ndev);
2072
2073         if (netif_running(ndev)) {
2074                 ret = dev_open(vf_netdev, NULL);
2075                 if (ret)
2076                         netdev_warn(vf_netdev,
2077                                     "unable to open: %d\n", ret);
2078         }
2079 }
2080
2081 /* Setup VF as slave of the synthetic device.
2082  * Runs in workqueue to avoid recursion in netlink callbacks.
2083  */
2084 static void netvsc_vf_setup(struct work_struct *w)
2085 {
2086         struct net_device_context *ndev_ctx
2087                 = container_of(w, struct net_device_context, vf_takeover.work);
2088         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2089         struct net_device *vf_netdev;
2090
2091         if (!rtnl_trylock()) {
2092                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2093                 return;
2094         }
2095
2096         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2097         if (vf_netdev)
2098                 __netvsc_vf_setup(ndev, vf_netdev);
2099
2100         rtnl_unlock();
2101 }
2102
2103 /* Find netvsc by VF serial number.
2104  * The PCI hyperv controller records the serial number as the slot kobj name.
2105  */
2106 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2107 {
2108         struct device *parent = vf_netdev->dev.parent;
2109         struct net_device_context *ndev_ctx;
2110         struct pci_dev *pdev;
2111         u32 serial;
2112
2113         if (!parent || !dev_is_pci(parent))
2114                 return NULL; /* not a PCI device */
2115
2116         pdev = to_pci_dev(parent);
2117         if (!pdev->slot) {
2118                 netdev_notice(vf_netdev, "no PCI slot information\n");
2119                 return NULL;
2120         }
2121
2122         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2123                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2124                               pci_slot_name(pdev->slot));
2125                 return NULL;
2126         }
2127
2128         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2129                 if (!ndev_ctx->vf_alloc)
2130                         continue;
2131
2132                 if (ndev_ctx->vf_serial == serial)
2133                         return hv_get_drvdata(ndev_ctx->device_ctx);
2134         }
2135
2136         netdev_notice(vf_netdev,
2137                       "no netdev found for vf serial:%u\n", serial);
2138         return NULL;
2139 }
2140
2141 static int netvsc_register_vf(struct net_device *vf_netdev)
2142 {
2143         struct net_device_context *net_device_ctx;
2144         struct netvsc_device *netvsc_dev;
2145         struct net_device *ndev;
2146         int ret;
2147
2148         if (vf_netdev->addr_len != ETH_ALEN)
2149                 return NOTIFY_DONE;
2150
2151         ndev = get_netvsc_byslot(vf_netdev);
2152         if (!ndev)
2153                 return NOTIFY_DONE;
2154
2155         net_device_ctx = netdev_priv(ndev);
2156         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2157         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2158                 return NOTIFY_DONE;
2159
2160         /* if synthetic interface is a different namespace,
2161          * then move the VF to that namespace; join will be
2162          * done again in that context.
2163          */
2164         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2165                 ret = dev_change_net_namespace(vf_netdev,
2166                                                dev_net(ndev), "eth%d");
2167                 if (ret)
2168                         netdev_err(vf_netdev,
2169                                    "could not move to same namespace as %s: %d\n",
2170                                    ndev->name, ret);
2171                 else
2172                         netdev_info(vf_netdev,
2173                                     "VF moved to namespace with: %s\n",
2174                                     ndev->name);
2175                 return NOTIFY_DONE;
2176         }
2177
2178         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2179
2180         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2181                 return NOTIFY_DONE;
2182
2183         dev_hold(vf_netdev);
2184         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2185         return NOTIFY_OK;
2186 }
2187
2188 /* VF up/down change detected, schedule to change data path */
2189 static int netvsc_vf_changed(struct net_device *vf_netdev)
2190 {
2191         struct net_device_context *net_device_ctx;
2192         struct netvsc_device *netvsc_dev;
2193         struct net_device *ndev;
2194         bool vf_is_up = netif_running(vf_netdev);
2195
2196         ndev = get_netvsc_byref(vf_netdev);
2197         if (!ndev)
2198                 return NOTIFY_DONE;
2199
2200         net_device_ctx = netdev_priv(ndev);
2201         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2202         if (!netvsc_dev)
2203                 return NOTIFY_DONE;
2204
2205         netvsc_switch_datapath(ndev, vf_is_up);
2206         netdev_info(ndev, "Data path switched %s VF: %s\n",
2207                     vf_is_up ? "to" : "from", vf_netdev->name);
2208
2209         return NOTIFY_OK;
2210 }
2211
2212 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2213 {
2214         struct net_device *ndev;
2215         struct net_device_context *net_device_ctx;
2216
2217         ndev = get_netvsc_byref(vf_netdev);
2218         if (!ndev)
2219                 return NOTIFY_DONE;
2220
2221         net_device_ctx = netdev_priv(ndev);
2222         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2223
2224         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2225
2226         netdev_rx_handler_unregister(vf_netdev);
2227         netdev_upper_dev_unlink(vf_netdev, ndev);
2228         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2229         dev_put(vf_netdev);
2230
2231         return NOTIFY_OK;
2232 }
2233
2234 static int netvsc_probe(struct hv_device *dev,
2235                         const struct hv_vmbus_device_id *dev_id)
2236 {
2237         struct net_device *net = NULL;
2238         struct net_device_context *net_device_ctx;
2239         struct netvsc_device_info *device_info = NULL;
2240         struct netvsc_device *nvdev;
2241         int ret = -ENOMEM;
2242
2243         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2244                                 VRSS_CHANNEL_MAX);
2245         if (!net)
2246                 goto no_net;
2247
2248         netif_carrier_off(net);
2249
2250         netvsc_init_settings(net);
2251
2252         net_device_ctx = netdev_priv(net);
2253         net_device_ctx->device_ctx = dev;
2254         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2255         if (netif_msg_probe(net_device_ctx))
2256                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2257                            net_device_ctx->msg_enable);
2258
2259         hv_set_drvdata(dev, net);
2260
2261         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2262
2263         spin_lock_init(&net_device_ctx->lock);
2264         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2265         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2266
2267         net_device_ctx->vf_stats
2268                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2269         if (!net_device_ctx->vf_stats)
2270                 goto no_stats;
2271
2272         net->netdev_ops = &device_ops;
2273         net->ethtool_ops = &ethtool_ops;
2274         SET_NETDEV_DEV(net, &dev->device);
2275
2276         /* We always need headroom for rndis header */
2277         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2278
2279         /* Initialize the number of queues to be 1, we may change it if more
2280          * channels are offered later.
2281          */
2282         netif_set_real_num_tx_queues(net, 1);
2283         netif_set_real_num_rx_queues(net, 1);
2284
2285         /* Notify the netvsc driver of the new device */
2286         device_info = netvsc_devinfo_get(NULL);
2287
2288         if (!device_info) {
2289                 ret = -ENOMEM;
2290                 goto devinfo_failed;
2291         }
2292
2293         nvdev = rndis_filter_device_add(dev, device_info);
2294         if (IS_ERR(nvdev)) {
2295                 ret = PTR_ERR(nvdev);
2296                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2297                 goto rndis_failed;
2298         }
2299
2300         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2301
2302         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2303          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2304          * all subchannels to show up, but that may not happen because
2305          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2306          * -> ... -> device_add() -> ... -> __device_attach() can't get
2307          * the device lock, so all the subchannels can't be processed --
2308          * finally netvsc_subchan_work() hangs forever.
2309          */
2310         rtnl_lock();
2311
2312         if (nvdev->num_chn > 1)
2313                 schedule_work(&nvdev->subchan_work);
2314
2315         /* hw_features computed in rndis_netdev_set_hwcaps() */
2316         net->features = net->hw_features |
2317                 NETIF_F_HIGHDMA | NETIF_F_SG |
2318                 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2319         net->vlan_features = net->features;
2320
2321         netdev_lockdep_set_classes(net);
2322
2323         /* MTU range: 68 - 1500 or 65521 */
2324         net->min_mtu = NETVSC_MTU_MIN;
2325         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2326                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2327         else
2328                 net->max_mtu = ETH_DATA_LEN;
2329
2330         ret = register_netdevice(net);
2331         if (ret != 0) {
2332                 pr_err("Unable to register netdev.\n");
2333                 goto register_failed;
2334         }
2335
2336         list_add(&net_device_ctx->list, &netvsc_dev_list);
2337         rtnl_unlock();
2338
2339         kfree(device_info);
2340         return 0;
2341
2342 register_failed:
2343         rtnl_unlock();
2344         rndis_filter_device_remove(dev, nvdev);
2345 rndis_failed:
2346         kfree(device_info);
2347 devinfo_failed:
2348         free_percpu(net_device_ctx->vf_stats);
2349 no_stats:
2350         hv_set_drvdata(dev, NULL);
2351         free_netdev(net);
2352 no_net:
2353         return ret;
2354 }
2355
2356 static int netvsc_remove(struct hv_device *dev)
2357 {
2358         struct net_device_context *ndev_ctx;
2359         struct net_device *vf_netdev, *net;
2360         struct netvsc_device *nvdev;
2361
2362         net = hv_get_drvdata(dev);
2363         if (net == NULL) {
2364                 dev_err(&dev->device, "No net device to remove\n");
2365                 return 0;
2366         }
2367
2368         ndev_ctx = netdev_priv(net);
2369
2370         cancel_delayed_work_sync(&ndev_ctx->dwork);
2371
2372         rtnl_lock();
2373         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2374         if (nvdev)
2375                 cancel_work_sync(&nvdev->subchan_work);
2376
2377         /*
2378          * Call to the vsc driver to let it know that the device is being
2379          * removed. Also blocks mtu and channel changes.
2380          */
2381         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2382         if (vf_netdev)
2383                 netvsc_unregister_vf(vf_netdev);
2384
2385         if (nvdev)
2386                 rndis_filter_device_remove(dev, nvdev);
2387
2388         unregister_netdevice(net);
2389         list_del(&ndev_ctx->list);
2390
2391         rtnl_unlock();
2392
2393         hv_set_drvdata(dev, NULL);
2394
2395         free_percpu(ndev_ctx->vf_stats);
2396         free_netdev(net);
2397         return 0;
2398 }
2399
2400 static const struct hv_vmbus_device_id id_table[] = {
2401         /* Network guid */
2402         { HV_NIC_GUID, },
2403         { },
2404 };
2405
2406 MODULE_DEVICE_TABLE(vmbus, id_table);
2407
2408 /* The one and only one */
2409 static struct  hv_driver netvsc_drv = {
2410         .name = KBUILD_MODNAME,
2411         .id_table = id_table,
2412         .probe = netvsc_probe,
2413         .remove = netvsc_remove,
2414         .driver = {
2415                 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
2416         },
2417 };
2418
2419 /*
2420  * On Hyper-V, every VF interface is matched with a corresponding
2421  * synthetic interface. The synthetic interface is presented first
2422  * to the guest. When the corresponding VF instance is registered,
2423  * we will take care of switching the data path.
2424  */
2425 static int netvsc_netdev_event(struct notifier_block *this,
2426                                unsigned long event, void *ptr)
2427 {
2428         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2429
2430         /* Skip our own events */
2431         if (event_dev->netdev_ops == &device_ops)
2432                 return NOTIFY_DONE;
2433
2434         /* Avoid non-Ethernet type devices */
2435         if (event_dev->type != ARPHRD_ETHER)
2436                 return NOTIFY_DONE;
2437
2438         /* Avoid Vlan dev with same MAC registering as VF */
2439         if (is_vlan_dev(event_dev))
2440                 return NOTIFY_DONE;
2441
2442         /* Avoid Bonding master dev with same MAC registering as VF */
2443         if ((event_dev->priv_flags & IFF_BONDING) &&
2444             (event_dev->flags & IFF_MASTER))
2445                 return NOTIFY_DONE;
2446
2447         switch (event) {
2448         case NETDEV_REGISTER:
2449                 return netvsc_register_vf(event_dev);
2450         case NETDEV_UNREGISTER:
2451                 return netvsc_unregister_vf(event_dev);
2452         case NETDEV_UP:
2453         case NETDEV_DOWN:
2454                 return netvsc_vf_changed(event_dev);
2455         default:
2456                 return NOTIFY_DONE;
2457         }
2458 }
2459
2460 static struct notifier_block netvsc_netdev_notifier = {
2461         .notifier_call = netvsc_netdev_event,
2462 };
2463
2464 static void __exit netvsc_drv_exit(void)
2465 {
2466         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2467         vmbus_driver_unregister(&netvsc_drv);
2468 }
2469
2470 static int __init netvsc_drv_init(void)
2471 {
2472         int ret;
2473
2474         if (ring_size < RING_SIZE_MIN) {
2475                 ring_size = RING_SIZE_MIN;
2476                 pr_info("Increased ring_size to %u (min allowed)\n",
2477                         ring_size);
2478         }
2479         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2480
2481         ret = vmbus_driver_register(&netvsc_drv);
2482         if (ret)
2483                 return ret;
2484
2485         register_netdevice_notifier(&netvsc_netdev_notifier);
2486         return 0;
2487 }
2488
2489 MODULE_LICENSE("GPL");
2490 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2491
2492 module_init(netvsc_drv_init);
2493 module_exit(netvsc_drv_exit);