drivers/net/hyperv/netvsc.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/kernel.h>
  12 #include <linux/sched.h>
  13 #include <linux/wait.h>
  14 #include <linux/mm.h>
  15 #include <linux/delay.h>
  16 #include <linux/io.h>
  17 #include <linux/slab.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/prefetch.h>
  23
  24 #include <asm/sync_bitops.h>
  25
  26 #include "hyperv_net.h"
  27 #include "netvsc_trace.h"
  28
  29 /*
  30  * Switch the data path from the synthetic interface to the VF
  31  * interface.
  32  */
  33 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  34 {
  35         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  36         struct hv_device *dev = net_device_ctx->device_ctx;
  37         struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
  38         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  39
  40         memset(init_pkt, 0, sizeof(struct nvsp_message));
  41         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  42         if (vf)
  43                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  44                         NVSP_DATAPATH_VF;
  45         else
  46                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  47                         NVSP_DATAPATH_SYNTHETIC;
  48
  49         trace_nvsp_send(ndev, init_pkt);
  50
  51         vmbus_sendpacket(dev->channel, init_pkt,
  52                                sizeof(struct nvsp_message),
  53                                (unsigned long)init_pkt,
  54                                VM_PKT_DATA_INBAND, 0);
  55 }
  56
  57 /* Worker to setup sub channels on initial setup
  58  * Initial hotplug event occurs in softirq context
  59  * and can't wait for channels.
  60  */
  61 static void netvsc_subchan_work(struct work_struct *w)
  62 {
  63         struct netvsc_device *nvdev =
  64                 container_of(w, struct netvsc_device, subchan_work);
  65         struct rndis_device *rdev;
  66         int i, ret;
  67
  68         /* Avoid deadlock with device removal already under RTNL */
  69         if (!rtnl_trylock()) {
  70                 schedule_work(w);
  71                 return;
  72         }
  73
  74         rdev = nvdev->extension;
  75         if (rdev) {
  76                 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
  77                 if (ret == 0) {
  78                         netif_device_attach(rdev->ndev);
  79                 } else {
  80                         /* fallback to only primary channel */
  81                         for (i = 1; i < nvdev->num_chn; i++)
  82                                 netif_napi_del(&nvdev->chan_table[i].napi);
  83
  84                         nvdev->max_chn = 1;
  85                         nvdev->num_chn = 1;
  86                 }
  87         }
  88
  89         rtnl_unlock();
  90 }
  91
  92 static struct netvsc_device *alloc_net_device(void)
  93 {
  94         struct netvsc_device *net_device;
  95
  96         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  97         if (!net_device)
  98                 return NULL;
  99
 100         init_waitqueue_head(&net_device->wait_drain);
 101         net_device->destroy = false;
 102         net_device->tx_disable = false;
 103
 104         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
 105         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
 106
 107         init_completion(&net_device->channel_init_wait);
 108         init_waitqueue_head(&net_device->subchan_open);
 109         INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
 110
 111         return net_device;
 112 }
 113
 114 static void free_netvsc_device(struct rcu_head *head)
 115 {
 116         struct netvsc_device *nvdev
 117                 = container_of(head, struct netvsc_device, rcu);
 118         int i;
 119
 120         kfree(nvdev->extension);
 121         vfree(nvdev->recv_buf);
 122         vfree(nvdev->send_buf);
 123         kfree(nvdev->send_section_map);
 124
 125         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
 126                 vfree(nvdev->chan_table[i].mrc.slots);
 127
 128         kfree(nvdev);
 129 }
 130
 131 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
 132 {
 133         call_rcu(&nvdev->rcu, free_netvsc_device);
 134 }
 135
 136 static void netvsc_revoke_recv_buf(struct hv_device *device,
 137                                    struct netvsc_device *net_device,
 138                                    struct net_device *ndev)
 139 {
 140         struct nvsp_message *revoke_packet;
 141         int ret;
 142
 143         /*
 144          * If we got a section count, it means we received a
 145          * SendReceiveBufferComplete msg (ie sent
 146          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 147          * to send a revoke msg here
 148          */
 149         if (net_device->recv_section_cnt) {
 150                 /* Send the revoke receive buffer */
 151                 revoke_packet = &net_device->revoke_packet;
 152                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 153
 154                 revoke_packet->hdr.msg_type =
 155                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 156                 revoke_packet->msg.v1_msg.
 157                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 158
 159                 trace_nvsp_send(ndev, revoke_packet);
 160
 161                 ret = vmbus_sendpacket(device->channel,
 162                                        revoke_packet,
 163                                        sizeof(struct nvsp_message),
 164                                        (unsigned long)revoke_packet,
 165                                        VM_PKT_DATA_INBAND, 0);
 166                 /* If the failure is because the channel is rescinded;
 167                  * ignore the failure since we cannot send on a rescinded
 168                  * channel. This would allow us to properly cleanup
 169                  * even when the channel is rescinded.
 170                  */
 171                 if (device->channel->rescind)
 172                         ret = 0;
 173                 /*
 174                  * If we failed here, we might as well return and
 175                  * have a leak rather than continue and a bugchk
 176                  */
 177                 if (ret != 0) {
 178                         netdev_err(ndev, "unable to send "
 179                                 "revoke receive buffer to netvsp\n");
 180                         return;
 181                 }
 182                 net_device->recv_section_cnt = 0;
 183         }
 184 }
 185
 186 static void netvsc_revoke_send_buf(struct hv_device *device,
 187                                    struct netvsc_device *net_device,
 188                                    struct net_device *ndev)
 189 {
 190         struct nvsp_message *revoke_packet;
 191         int ret;
 192
 193         /* Deal with the send buffer we may have setup.
 194          * If we got a  send section size, it means we received a
 195          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 196          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 197          * to send a revoke msg here
 198          */
 199         if (net_device->send_section_cnt) {
 200                 /* Send the revoke receive buffer */
 201                 revoke_packet = &net_device->revoke_packet;
 202                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 203
 204                 revoke_packet->hdr.msg_type =
 205                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 206                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 207                         NETVSC_SEND_BUFFER_ID;
 208
 209                 trace_nvsp_send(ndev, revoke_packet);
 210
 211                 ret = vmbus_sendpacket(device->channel,
 212                                        revoke_packet,
 213                                        sizeof(struct nvsp_message),
 214                                        (unsigned long)revoke_packet,
 215                                        VM_PKT_DATA_INBAND, 0);
 216
 217                 /* If the failure is because the channel is rescinded;
 218                  * ignore the failure since we cannot send on a rescinded
 219                  * channel. This would allow us to properly cleanup
 220                  * even when the channel is rescinded.
 221                  */
 222                 if (device->channel->rescind)
 223                         ret = 0;
 224
 225                 /* If we failed here, we might as well return and
 226                  * have a leak rather than continue and a bugchk
 227                  */
 228                 if (ret != 0) {
 229                         netdev_err(ndev, "unable to send "
 230                                    "revoke send buffer to netvsp\n");
 231                         return;
 232                 }
 233                 net_device->send_section_cnt = 0;
 234         }
 235 }
 236
 237 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
 238                                        struct netvsc_device *net_device,
 239                                        struct net_device *ndev)
 240 {
 241         int ret;
 242
 243         if (net_device->recv_buf_gpadl_handle) {
 244                 ret = vmbus_teardown_gpadl(device->channel,
 245                                            net_device->recv_buf_gpadl_handle);
 246
 247                 /* If we failed here, we might as well return and have a leak
 248                  * rather than continue and a bugchk
 249                  */
 250                 if (ret != 0) {
 251                         netdev_err(ndev,
 252                                    "unable to teardown receive buffer's gpadl\n");
 253                         return;
 254                 }
 255                 net_device->recv_buf_gpadl_handle = 0;
 256         }
 257 }
 258
 259 static void netvsc_teardown_send_gpadl(struct hv_device *device,
 260                                        struct netvsc_device *net_device,
 261                                        struct net_device *ndev)
 262 {
 263         int ret;
 264
 265         if (net_device->send_buf_gpadl_handle) {
 266                 ret = vmbus_teardown_gpadl(device->channel,
 267                                            net_device->send_buf_gpadl_handle);
 268
 269                 /* If we failed here, we might as well return and have a leak
 270                  * rather than continue and a bugchk
 271                  */
 272                 if (ret != 0) {
 273                         netdev_err(ndev,
 274                                    "unable to teardown send buffer's gpadl\n");
 275                         return;
 276                 }
 277                 net_device->send_buf_gpadl_handle = 0;
 278         }
 279 }
 280
 281 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
 282 {
 283         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 284         int node = cpu_to_node(nvchan->channel->target_cpu);
 285         size_t size;
 286
 287         size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
 288         nvchan->mrc.slots = vzalloc_node(size, node);
 289         if (!nvchan->mrc.slots)
 290                 nvchan->mrc.slots = vzalloc(size);
 291
 292         return nvchan->mrc.slots ? 0 : -ENOMEM;
 293 }
 294
 295 static int netvsc_init_buf(struct hv_device *device,
 296                            struct netvsc_device *net_device,
 297                            const struct netvsc_device_info *device_info)
 298 {
 299         struct nvsp_1_message_send_receive_buffer_complete *resp;
 300         struct net_device *ndev = hv_get_drvdata(device);
 301         struct nvsp_message *init_packet;
 302         unsigned int buf_size;
 303         size_t map_words;
 304         int ret = 0;
 305
 306         /* Get receive buffer area. */
 307         buf_size = device_info->recv_sections * device_info->recv_section_size;
 308         buf_size = roundup(buf_size, PAGE_SIZE);
 309
 310         /* Legacy hosts only allow smaller receive buffer */
 311         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 312                 buf_size = min_t(unsigned int, buf_size,
 313                                  NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
 314
 315         net_device->recv_buf = vzalloc(buf_size);
 316         if (!net_device->recv_buf) {
 317                 netdev_err(ndev,
 318                            "unable to allocate receive buffer of size %u\n",
 319                            buf_size);
 320                 ret = -ENOMEM;
 321                 goto cleanup;
 322         }
 323
 324         net_device->recv_buf_size = buf_size;
 325
 326         /*
 327          * Establish the gpadl handle for this buffer on this
 328          * channel.  Note: This call uses the vmbus connection rather
 329          * than the channel to establish the gpadl handle.
 330          */
 331         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 332                                     buf_size,
 333                                     &net_device->recv_buf_gpadl_handle);
 334         if (ret != 0) {
 335                 netdev_err(ndev,
 336                         "unable to establish receive buffer's gpadl\n");
 337                 goto cleanup;
 338         }
 339
 340         /* Notify the NetVsp of the gpadl handle */
 341         init_packet = &net_device->channel_init_pkt;
 342         memset(init_packet, 0, sizeof(struct nvsp_message));
 343         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 344         init_packet->msg.v1_msg.send_recv_buf.
 345                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 346         init_packet->msg.v1_msg.
 347                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 348
 349         trace_nvsp_send(ndev, init_packet);
 350
 351         /* Send the gpadl notification request */
 352         ret = vmbus_sendpacket(device->channel, init_packet,
 353                                sizeof(struct nvsp_message),
 354                                (unsigned long)init_packet,
 355                                VM_PKT_DATA_INBAND,
 356                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 357         if (ret != 0) {
 358                 netdev_err(ndev,
 359                         "unable to send receive buffer's gpadl to netvsp\n");
 360                 goto cleanup;
 361         }
 362
 363         wait_for_completion(&net_device->channel_init_wait);
 364
 365         /* Check the response */
 366         resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
 367         if (resp->status != NVSP_STAT_SUCCESS) {
 368                 netdev_err(ndev,
 369                            "Unable to complete receive buffer initialization with NetVsp - status %d\n",
 370                            resp->status);
 371                 ret = -EINVAL;
 372                 goto cleanup;
 373         }
 374
 375         /* Parse the response */
 376         netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
 377                    resp->num_sections, resp->sections[0].sub_alloc_size,
 378                    resp->sections[0].num_sub_allocs);
 379
 380         /* There should only be one section for the entire receive buffer */
 381         if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
 382                 ret = -EINVAL;
 383                 goto cleanup;
 384         }
 385
 386         net_device->recv_section_size = resp->sections[0].sub_alloc_size;
 387         net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
 388
 389         /* Setup receive completion ring */
 390         net_device->recv_completion_cnt
 391                 = round_up(net_device->recv_section_cnt + 1,
 392                            PAGE_SIZE / sizeof(u64));
 393         ret = netvsc_alloc_recv_comp_ring(net_device, 0);
 394         if (ret)
 395                 goto cleanup;
 396
 397         /* Now setup the send buffer. */
 398         buf_size = device_info->send_sections * device_info->send_section_size;
 399         buf_size = round_up(buf_size, PAGE_SIZE);
 400
 401         net_device->send_buf = vzalloc(buf_size);
 402         if (!net_device->send_buf) {
 403                 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
 404                            buf_size);
 405                 ret = -ENOMEM;
 406                 goto cleanup;
 407         }
 408
 409         /* Establish the gpadl handle for this buffer on this
 410          * channel.  Note: This call uses the vmbus connection rather
 411          * than the channel to establish the gpadl handle.
 412          */
 413         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 414                                     buf_size,
 415                                     &net_device->send_buf_gpadl_handle);
 416         if (ret != 0) {
 417                 netdev_err(ndev,
 418                            "unable to establish send buffer's gpadl\n");
 419                 goto cleanup;
 420         }
 421
 422         /* Notify the NetVsp of the gpadl handle */
 423         init_packet = &net_device->channel_init_pkt;
 424         memset(init_packet, 0, sizeof(struct nvsp_message));
 425         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 426         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 427                 net_device->send_buf_gpadl_handle;
 428         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 429
 430         trace_nvsp_send(ndev, init_packet);
 431
 432         /* Send the gpadl notification request */
 433         ret = vmbus_sendpacket(device->channel, init_packet,
 434                                sizeof(struct nvsp_message),
 435                                (unsigned long)init_packet,
 436                                VM_PKT_DATA_INBAND,
 437                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 438         if (ret != 0) {
 439                 netdev_err(ndev,
 440                            "unable to send send buffer's gpadl to netvsp\n");
 441                 goto cleanup;
 442         }
 443
 444         wait_for_completion(&net_device->channel_init_wait);
 445
 446         /* Check the response */
 447         if (init_packet->msg.v1_msg.
 448             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 449                 netdev_err(ndev, "Unable to complete send buffer "
 450                            "initialization with NetVsp - status %d\n",
 451                            init_packet->msg.v1_msg.
 452                            send_send_buf_complete.status);
 453                 ret = -EINVAL;
 454                 goto cleanup;
 455         }
 456
 457         /* Parse the response */
 458         net_device->send_section_size = init_packet->msg.
 459                                 v1_msg.send_send_buf_complete.section_size;
 460
 461         /* Section count is simply the size divided by the section size. */
 462         net_device->send_section_cnt = buf_size / net_device->send_section_size;
 463
 464         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 465                    net_device->send_section_size, net_device->send_section_cnt);
 466
 467         /* Setup state for managing the send buffer. */
 468         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
 469
 470         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
 471         if (net_device->send_section_map == NULL) {
 472                 ret = -ENOMEM;
 473                 goto cleanup;
 474         }
 475
 476         goto exit;
 477
 478 cleanup:
 479         netvsc_revoke_recv_buf(device, net_device, ndev);
 480         netvsc_revoke_send_buf(device, net_device, ndev);
 481         netvsc_teardown_recv_gpadl(device, net_device, ndev);
 482         netvsc_teardown_send_gpadl(device, net_device, ndev);
 483
 484 exit:
 485         return ret;
 486 }
 487
 488 /* Negotiate NVSP protocol version */
 489 static int negotiate_nvsp_ver(struct hv_device *device,
 490                               struct netvsc_device *net_device,
 491                               struct nvsp_message *init_packet,
 492                               u32 nvsp_ver)
 493 {
 494         struct net_device *ndev = hv_get_drvdata(device);
 495         int ret;
 496
 497         memset(init_packet, 0, sizeof(struct nvsp_message));
 498         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 499         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 500         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 501         trace_nvsp_send(ndev, init_packet);
 502
 503         /* Send the init request */
 504         ret = vmbus_sendpacket(device->channel, init_packet,
 505                                sizeof(struct nvsp_message),
 506                                (unsigned long)init_packet,
 507                                VM_PKT_DATA_INBAND,
 508                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 509
 510         if (ret != 0)
 511                 return ret;
 512
 513         wait_for_completion(&net_device->channel_init_wait);
 514
 515         if (init_packet->msg.init_msg.init_complete.status !=
 516             NVSP_STAT_SUCCESS)
 517                 return -EINVAL;
 518
 519         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 520                 return 0;
 521
 522         /* NVSPv2 or later: Send NDIS config */
 523         memset(init_packet, 0, sizeof(struct nvsp_message));
 524         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 525         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 526         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 527
 528         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 529                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 530
 531                 /* Teaming bit is needed to receive link speed updates */
 532                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 533         }
 534
 535         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
 536                 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
 537
 538         trace_nvsp_send(ndev, init_packet);
 539
 540         ret = vmbus_sendpacket(device->channel, init_packet,
 541                                 sizeof(struct nvsp_message),
 542                                 (unsigned long)init_packet,
 543                                 VM_PKT_DATA_INBAND, 0);
 544
 545         return ret;
 546 }
 547
 548 static int netvsc_connect_vsp(struct hv_device *device,
 549                               struct netvsc_device *net_device,
 550                               const struct netvsc_device_info *device_info)
 551 {
 552         struct net_device *ndev = hv_get_drvdata(device);
 553         static const u32 ver_list[] = {
 554                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 555                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
 556                 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
 557         };
 558         struct nvsp_message *init_packet;
 559         int ndis_version, i, ret;
 560
 561         init_packet = &net_device->channel_init_pkt;
 562
 563         /* Negotiate the latest NVSP protocol supported */
 564         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 565                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 566                                        ver_list[i])  == 0) {
 567                         net_device->nvsp_version = ver_list[i];
 568                         break;
 569                 }
 570
 571         if (i < 0) {
 572                 ret = -EPROTO;
 573                 goto cleanup;
 574         }
 575
 576         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 577
 578         /* Send the ndis version */
 579         memset(init_packet, 0, sizeof(struct nvsp_message));
 580
 581         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 582                 ndis_version = 0x00060001;
 583         else
 584                 ndis_version = 0x0006001e;
 585
 586         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 587         init_packet->msg.v1_msg.
 588                 send_ndis_ver.ndis_major_ver =
 589                                 (ndis_version & 0xFFFF0000) >> 16;
 590         init_packet->msg.v1_msg.
 591                 send_ndis_ver.ndis_minor_ver =
 592                                 ndis_version & 0xFFFF;
 593
 594         trace_nvsp_send(ndev, init_packet);
 595
 596         /* Send the init request */
 597         ret = vmbus_sendpacket(device->channel, init_packet,
 598                                 sizeof(struct nvsp_message),
 599                                 (unsigned long)init_packet,
 600                                 VM_PKT_DATA_INBAND, 0);
 601         if (ret != 0)
 602                 goto cleanup;
 603
 604
 605         ret = netvsc_init_buf(device, net_device, device_info);
 606
 607 cleanup:
 608         return ret;
 609 }
 610
 611 /*
 612  * netvsc_device_remove - Callback when the root bus device is removed
 613  */
 614 void netvsc_device_remove(struct hv_device *device)
 615 {
 616         struct net_device *ndev = hv_get_drvdata(device);
 617         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 618         struct netvsc_device *net_device
 619                 = rtnl_dereference(net_device_ctx->nvdev);
 620         int i;
 621
 622         /*
 623          * Revoke receive buffer. If host is pre-Win2016 then tear down
 624          * receive buffer GPADL. Do the same for send buffer.
 625          */
 626         netvsc_revoke_recv_buf(device, net_device, ndev);
 627         if (vmbus_proto_version < VERSION_WIN10)
 628                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 629
 630         netvsc_revoke_send_buf(device, net_device, ndev);
 631         if (vmbus_proto_version < VERSION_WIN10)
 632                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 633
 634         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 635
 636         /* And disassociate NAPI context from device */
 637         for (i = 0; i < net_device->num_chn; i++)
 638                 netif_napi_del(&net_device->chan_table[i].napi);
 639
 640         /*
 641          * At this point, no one should be accessing net_device
 642          * except in here
 643          */
 644         netdev_dbg(ndev, "net device safe to remove\n");
 645
 646         /* Now, we can close the channel safely */
 647         vmbus_close(device->channel);
 648
 649         /*
 650          * If host is Win2016 or higher then we do the GPADL tear down
 651          * here after VMBus is closed.
 652         */
 653         if (vmbus_proto_version >= VERSION_WIN10) {
 654                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 655                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 656         }
 657
 658         /* Release all resources */
 659         free_netvsc_device_rcu(net_device);
 660 }
 661
 662 #define RING_AVAIL_PERCENT_HIWATER 20
 663 #define RING_AVAIL_PERCENT_LOWATER 10
 664
 665 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 666                                          u32 index)
 667 {
 668         sync_change_bit(index, net_device->send_section_map);
 669 }
 670
 671 static void netvsc_send_tx_complete(struct net_device *ndev,
 672                                     struct netvsc_device *net_device,
 673                                     struct vmbus_channel *channel,
 674                                     const struct vmpacket_descriptor *desc,
 675                                     int budget)
 676 {
 677         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 678         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 679         u16 q_idx = 0;
 680         int queue_sends;
 681
 682         /* Notify the layer above us */
 683         if (likely(skb)) {
 684                 const struct hv_netvsc_packet *packet
 685                         = (struct hv_netvsc_packet *)skb->cb;
 686                 u32 send_index = packet->send_buf_index;
 687                 struct netvsc_stats *tx_stats;
 688
 689                 if (send_index != NETVSC_INVALID_INDEX)
 690                         netvsc_free_send_slot(net_device, send_index);
 691                 q_idx = packet->q_idx;
 692
 693                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 694
 695                 u64_stats_update_begin(&tx_stats->syncp);
 696                 tx_stats->packets += packet->total_packets;
 697                 tx_stats->bytes += packet->total_bytes;
 698                 u64_stats_update_end(&tx_stats->syncp);
 699
 700                 napi_consume_skb(skb, budget);
 701         }
 702
 703         queue_sends =
 704                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 705
 706         if (unlikely(net_device->destroy)) {
 707                 if (queue_sends == 0)
 708                         wake_up(&net_device->wait_drain);
 709         } else {
 710                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
 711
 712                 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
 713                     (hv_get_avail_to_write_percent(&channel->outbound) >
 714                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
 715                         netif_tx_wake_queue(txq);
 716                         ndev_ctx->eth_stats.wake_queue++;
 717                 }
 718         }
 719 }
 720
 721 static void netvsc_send_completion(struct net_device *ndev,
 722                                    struct netvsc_device *net_device,
 723                                    struct vmbus_channel *incoming_channel,
 724                                    const struct vmpacket_descriptor *desc,
 725                                    int budget)
 726 {
 727         const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 728
 729         switch (nvsp_packet->hdr.msg_type) {
 730         case NVSP_MSG_TYPE_INIT_COMPLETE:
 731         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 732         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 733         case NVSP_MSG5_TYPE_SUBCHANNEL:
 734                 /* Copy the response back */
 735                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 736                        sizeof(struct nvsp_message));
 737                 complete(&net_device->channel_init_wait);
 738                 break;
 739
 740         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 741                 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
 742                                         desc, budget);
 743                 break;
 744
 745         default:
 746                 netdev_err(ndev,
 747                            "Unknown send completion type %d received!!\n",
 748                            nvsp_packet->hdr.msg_type);
 749         }
 750 }
 751
 752 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 753 {
 754         unsigned long *map_addr = net_device->send_section_map;
 755         unsigned int i;
 756
 757         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 758                 if (sync_test_and_set_bit(i, map_addr) == 0)
 759                         return i;
 760         }
 761
 762         return NETVSC_INVALID_INDEX;
 763 }
 764
 765 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 766                                     unsigned int section_index,
 767                                     u32 pend_size,
 768                                     struct hv_netvsc_packet *packet,
 769                                     struct rndis_message *rndis_msg,
 770                                     struct hv_page_buffer *pb,
 771                                     bool xmit_more)
 772 {
 773         char *start = net_device->send_buf;
 774         char *dest = start + (section_index * net_device->send_section_size)
 775                      + pend_size;
 776         int i;
 777         u32 padding = 0;
 778         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 779                 packet->page_buf_cnt;
 780         u32 remain;
 781
 782         /* Add padding */
 783         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
 784         if (xmit_more && remain) {
 785                 padding = net_device->pkt_align - remain;
 786                 rndis_msg->msg_len += padding;
 787                 packet->total_data_buflen += padding;
 788         }
 789
 790         for (i = 0; i < page_count; i++) {
 791                 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
 792                 u32 offset = pb[i].offset;
 793                 u32 len = pb[i].len;
 794
 795                 memcpy(dest, (src + offset), len);
 796                 dest += len;
 797         }
 798
 799         if (padding)
 800                 memset(dest, 0, padding);
 801 }
 802
 803 static inline int netvsc_send_pkt(
 804         struct hv_device *device,
 805         struct hv_netvsc_packet *packet,
 806         struct netvsc_device *net_device,
 807         struct hv_page_buffer *pb,
 808         struct sk_buff *skb)
 809 {
 810         struct nvsp_message nvmsg;
 811         struct nvsp_1_message_send_rndis_packet *rpkt =
 812                 &nvmsg.msg.v1_msg.send_rndis_pkt;
 813         struct netvsc_channel * const nvchan =
 814                 &net_device->chan_table[packet->q_idx];
 815         struct vmbus_channel *out_channel = nvchan->channel;
 816         struct net_device *ndev = hv_get_drvdata(device);
 817         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 818         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 819         u64 req_id;
 820         int ret;
 821         u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
 822
 823         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 824         if (skb)
 825                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
 826         else
 827                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
 828
 829         rpkt->send_buf_section_index = packet->send_buf_index;
 830         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 831                 rpkt->send_buf_section_size = 0;
 832         else
 833                 rpkt->send_buf_section_size = packet->total_data_buflen;
 834
 835         req_id = (ulong)skb;
 836
 837         if (out_channel->rescind)
 838                 return -ENODEV;
 839
 840         trace_nvsp_send_pkt(ndev, out_channel, rpkt);
 841
 842         if (packet->page_buf_cnt) {
 843                 if (packet->cp_partial)
 844                         pb += packet->rmsg_pgcnt;
 845
 846                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 847                                                   pb, packet->page_buf_cnt,
 848                                                   &nvmsg, sizeof(nvmsg),
 849                                                   req_id);
 850         } else {
 851                 ret = vmbus_sendpacket(out_channel,
 852                                        &nvmsg, sizeof(nvmsg),
 853                                        req_id, VM_PKT_DATA_INBAND,
 854                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 855         }
 856
 857         if (ret == 0) {
 858                 atomic_inc_return(&nvchan->queue_sends);
 859
 860                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 861                         netif_tx_stop_queue(txq);
 862                         ndev_ctx->eth_stats.stop_queue++;
 863                 }
 864         } else if (ret == -EAGAIN) {
 865                 netif_tx_stop_queue(txq);
 866                 ndev_ctx->eth_stats.stop_queue++;
 867         } else {
 868                 netdev_err(ndev,
 869                            "Unable to send packet pages %u len %u, ret %d\n",
 870                            packet->page_buf_cnt, packet->total_data_buflen,
 871                            ret);
 872         }
 873
 874         if (netif_tx_queue_stopped(txq) &&
 875             atomic_read(&nvchan->queue_sends) < 1 &&
 876             !net_device->tx_disable) {
 877                 netif_tx_wake_queue(txq);
 878                 ndev_ctx->eth_stats.wake_queue++;
 879                 if (ret == -EAGAIN)
 880                         ret = -ENOSPC;
 881         }
 882
 883         return ret;
 884 }
 885
 886 /* Move packet out of multi send data (msd), and clear msd */
 887 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 888                                 struct sk_buff **msd_skb,
 889                                 struct multi_send_data *msdp)
 890 {
 891         *msd_skb = msdp->skb;
 892         *msd_send = msdp->pkt;
 893         msdp->skb = NULL;
 894         msdp->pkt = NULL;
 895         msdp->count = 0;
 896 }
 897
 898 /* RCU already held by caller */
 899 int netvsc_send(struct net_device *ndev,
 900                 struct hv_netvsc_packet *packet,
 901                 struct rndis_message *rndis_msg,
 902                 struct hv_page_buffer *pb,
 903                 struct sk_buff *skb)
 904 {
 905         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 906         struct netvsc_device *net_device
 907                 = rcu_dereference_bh(ndev_ctx->nvdev);
 908         struct hv_device *device = ndev_ctx->device_ctx;
 909         int ret = 0;
 910         struct netvsc_channel *nvchan;
 911         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 912         unsigned int section_index = NETVSC_INVALID_INDEX;
 913         struct multi_send_data *msdp;
 914         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 915         struct sk_buff *msd_skb = NULL;
 916         bool try_batch, xmit_more;
 917
 918         /* If device is rescinded, return error and packet will get dropped. */
 919         if (unlikely(!net_device || net_device->destroy))
 920                 return -ENODEV;
 921
 922         nvchan = &net_device->chan_table[packet->q_idx];
 923         packet->send_buf_index = NETVSC_INVALID_INDEX;
 924         packet->cp_partial = false;
 925
 926         /* Send control message directly without accessing msd (Multi-Send
 927          * Data) field which may be changed during data packet processing.
 928          */
 929         if (!skb)
 930                 return netvsc_send_pkt(device, packet, net_device, pb, skb);
 931
 932         /* batch packets in send buffer if possible */
 933         msdp = &nvchan->msd;
 934         if (msdp->pkt)
 935                 msd_len = msdp->pkt->total_data_buflen;
 936
 937         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
 938         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 939             net_device->send_section_size) {
 940                 section_index = msdp->pkt->send_buf_index;
 941
 942         } else if (try_batch && msd_len + packet->rmsg_size <
 943                    net_device->send_section_size) {
 944                 section_index = msdp->pkt->send_buf_index;
 945                 packet->cp_partial = true;
 946
 947         } else if (pktlen + net_device->pkt_align <
 948                    net_device->send_section_size) {
 949                 section_index = netvsc_get_next_send_section(net_device);
 950                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
 951                         ++ndev_ctx->eth_stats.tx_send_full;
 952                 } else {
 953                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 954                         msd_len = 0;
 955                 }
 956         }
 957
 958         /* Keep aggregating only if stack says more data is coming
 959          * and not doing mixed modes send and not flow blocked
 960          */
 961         xmit_more = netdev_xmit_more() &&
 962                 !packet->cp_partial &&
 963                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
 964
 965         if (section_index != NETVSC_INVALID_INDEX) {
 966                 netvsc_copy_to_send_buf(net_device,
 967                                         section_index, msd_len,
 968                                         packet, rndis_msg, pb, xmit_more);
 969
 970                 packet->send_buf_index = section_index;
 971
 972                 if (packet->cp_partial) {
 973                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 974                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 975                 } else {
 976                         packet->page_buf_cnt = 0;
 977                         packet->total_data_buflen += msd_len;
 978                 }
 979
 980                 if (msdp->pkt) {
 981                         packet->total_packets += msdp->pkt->total_packets;
 982                         packet->total_bytes += msdp->pkt->total_bytes;
 983                 }
 984
 985                 if (msdp->skb)
 986                         dev_consume_skb_any(msdp->skb);
 987
 988                 if (xmit_more) {
 989                         msdp->skb = skb;
 990                         msdp->pkt = packet;
 991                         msdp->count++;
 992                 } else {
 993                         cur_send = packet;
 994                         msdp->skb = NULL;
 995                         msdp->pkt = NULL;
 996                         msdp->count = 0;
 997                 }
 998         } else {
 999                 move_pkt_msd(&msd_send, &msd_skb, msdp);
1000                 cur_send = packet;
1001         }
1002
1003         if (msd_send) {
1004                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1005                                             NULL, msd_skb);
1006
1007                 if (m_ret != 0) {
1008                         netvsc_free_send_slot(net_device,
1009                                               msd_send->send_buf_index);
1010                         dev_kfree_skb_any(msd_skb);
1011                 }
1012         }
1013
1014         if (cur_send)
1015                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1016
1017         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1018                 netvsc_free_send_slot(net_device, section_index);
1019
1020         return ret;
1021 }
1022
1023 /* Send pending recv completions */
1024 static int send_recv_completions(struct net_device *ndev,
1025                                  struct netvsc_device *nvdev,
1026                                  struct netvsc_channel *nvchan)
1027 {
1028         struct multi_recv_comp *mrc = &nvchan->mrc;
1029         struct recv_comp_msg {
1030                 struct nvsp_message_header hdr;
1031                 u32 status;
1032         }  __packed;
1033         struct recv_comp_msg msg = {
1034                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1035         };
1036         int ret;
1037
1038         while (mrc->first != mrc->next) {
1039                 const struct recv_comp_data *rcd
1040                         = mrc->slots + mrc->first;
1041
1042                 msg.status = rcd->status;
1043                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1044                                        rcd->tid, VM_PKT_COMP, 0);
1045                 if (unlikely(ret)) {
1046                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1047
1048                         ++ndev_ctx->eth_stats.rx_comp_busy;
1049                         return ret;
1050                 }
1051
1052                 if (++mrc->first == nvdev->recv_completion_cnt)
1053                         mrc->first = 0;
1054         }
1055
1056         /* receive completion ring has been emptied */
1057         if (unlikely(nvdev->destroy))
1058                 wake_up(&nvdev->wait_drain);
1059
1060         return 0;
1061 }
1062
1063 /* Count how many receive completions are outstanding */
1064 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1065                                  const struct multi_recv_comp *mrc,
1066                                  u32 *filled, u32 *avail)
1067 {
1068         u32 count = nvdev->recv_completion_cnt;
1069
1070         if (mrc->next >= mrc->first)
1071                 *filled = mrc->next - mrc->first;
1072         else
1073                 *filled = (count - mrc->first) + mrc->next;
1074
1075         *avail = count - *filled - 1;
1076 }
1077
1078 /* Add receive complete to ring to send to host. */
1079 static void enq_receive_complete(struct net_device *ndev,
1080                                  struct netvsc_device *nvdev, u16 q_idx,
1081                                  u64 tid, u32 status)
1082 {
1083         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1084         struct multi_recv_comp *mrc = &nvchan->mrc;
1085         struct recv_comp_data *rcd;
1086         u32 filled, avail;
1087
1088         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1089
1090         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1091                 send_recv_completions(ndev, nvdev, nvchan);
1092                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1093         }
1094
1095         if (unlikely(!avail)) {
1096                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1097                            q_idx, tid);
1098                 return;
1099         }
1100
1101         rcd = mrc->slots + mrc->next;
1102         rcd->tid = tid;
1103         rcd->status = status;
1104
1105         if (++mrc->next == nvdev->recv_completion_cnt)
1106                 mrc->next = 0;
1107 }
1108
1109 static int netvsc_receive(struct net_device *ndev,
1110                           struct netvsc_device *net_device,
1111                           struct netvsc_channel *nvchan,
1112                           const struct vmpacket_descriptor *desc,
1113                           const struct nvsp_message *nvsp)
1114 {
1115         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1116         struct vmbus_channel *channel = nvchan->channel;
1117         const struct vmtransfer_page_packet_header *vmxferpage_packet
1118                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1119         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1120         char *recv_buf = net_device->recv_buf;
1121         u32 status = NVSP_STAT_SUCCESS;
1122         int i;
1123         int count = 0;
1124
1125         /* Make sure this is a valid nvsp packet */
1126         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1127                 netif_err(net_device_ctx, rx_err, ndev,
1128                           "Unknown nvsp packet type received %u\n",
1129                           nvsp->hdr.msg_type);
1130                 return 0;
1131         }
1132
1133         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1134                 netif_err(net_device_ctx, rx_err, ndev,
1135                           "Invalid xfer page set id - expecting %x got %x\n",
1136                           NETVSC_RECEIVE_BUFFER_ID,
1137                           vmxferpage_packet->xfer_pageset_id);
1138                 return 0;
1139         }
1140
1141         count = vmxferpage_packet->range_cnt;
1142
1143         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1144         for (i = 0; i < count; i++) {
1145                 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1146                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1147                 void *data;
1148                 int ret;
1149
1150                 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1151                         nvchan->rsc.cnt = 0;
1152                         status = NVSP_STAT_FAIL;
1153                         netif_err(net_device_ctx, rx_err, ndev,
1154                                   "Packet offset:%u + len:%u too big\n",
1155                                   offset, buflen);
1156
1157                         continue;
1158                 }
1159
1160                 data = recv_buf + offset;
1161
1162                 nvchan->rsc.is_last = (i == count - 1);
1163
1164                 trace_rndis_recv(ndev, q_idx, data);
1165
1166                 /* Pass it to the upper layer */
1167                 ret = rndis_filter_receive(ndev, net_device,
1168                                            nvchan, data, buflen);
1169
1170                 if (unlikely(ret != NVSP_STAT_SUCCESS))
1171                         status = NVSP_STAT_FAIL;
1172         }
1173
1174         enq_receive_complete(ndev, net_device, q_idx,
1175                              vmxferpage_packet->d.trans_id, status);
1176
1177         return count;
1178 }
1179
1180 static void netvsc_send_table(struct net_device *ndev,
1181                               const struct nvsp_message *nvmsg)
1182 {
1183         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1184         u32 count, *tab;
1185         int i;
1186
1187         count = nvmsg->msg.v5_msg.send_table.count;
1188         if (count != VRSS_SEND_TAB_SIZE) {
1189                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1190                 return;
1191         }
1192
1193         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1194                       nvmsg->msg.v5_msg.send_table.offset);
1195
1196         for (i = 0; i < count; i++)
1197                 net_device_ctx->tx_table[i] = tab[i];
1198 }
1199
1200 static void netvsc_send_vf(struct net_device *ndev,
1201                            const struct nvsp_message *nvmsg)
1202 {
1203         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1204
1205         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1206         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1207         netdev_info(ndev, "VF slot %u %s\n",
1208                     net_device_ctx->vf_serial,
1209                     net_device_ctx->vf_alloc ? "added" : "removed");
1210 }
1211
1212 static  void netvsc_receive_inband(struct net_device *ndev,
1213                                    const struct nvsp_message *nvmsg)
1214 {
1215         switch (nvmsg->hdr.msg_type) {
1216         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1217                 netvsc_send_table(ndev, nvmsg);
1218                 break;
1219
1220         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1221                 netvsc_send_vf(ndev, nvmsg);
1222                 break;
1223         }
1224 }
1225
1226 static int netvsc_process_raw_pkt(struct hv_device *device,
1227                                   struct netvsc_channel *nvchan,
1228                                   struct netvsc_device *net_device,
1229                                   struct net_device *ndev,
1230                                   const struct vmpacket_descriptor *desc,
1231                                   int budget)
1232 {
1233         struct vmbus_channel *channel = nvchan->channel;
1234         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1235
1236         trace_nvsp_recv(ndev, channel, nvmsg);
1237
1238         switch (desc->type) {
1239         case VM_PKT_COMP:
1240                 netvsc_send_completion(ndev, net_device, channel,
1241                                        desc, budget);
1242                 break;
1243
1244         case VM_PKT_DATA_USING_XFER_PAGES:
1245                 return netvsc_receive(ndev, net_device, nvchan,
1246                                       desc, nvmsg);
1247                 break;
1248
1249         case VM_PKT_DATA_INBAND:
1250                 netvsc_receive_inband(ndev, nvmsg);
1251                 break;
1252
1253         default:
1254                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1255                            desc->type, desc->trans_id);
1256                 break;
1257         }
1258
1259         return 0;
1260 }
1261
1262 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1263 {
1264         struct vmbus_channel *primary = channel->primary_channel;
1265
1266         return primary ? primary->device_obj : channel->device_obj;
1267 }
1268
1269 /* Network processing softirq
1270  * Process data in incoming ring buffer from host
1271  * Stops when ring is empty or budget is met or exceeded.
1272  */
1273 int netvsc_poll(struct napi_struct *napi, int budget)
1274 {
1275         struct netvsc_channel *nvchan
1276                 = container_of(napi, struct netvsc_channel, napi);
1277         struct netvsc_device *net_device = nvchan->net_device;
1278         struct vmbus_channel *channel = nvchan->channel;
1279         struct hv_device *device = netvsc_channel_to_device(channel);
1280         struct net_device *ndev = hv_get_drvdata(device);
1281         int work_done = 0;
1282         int ret;
1283
1284         /* If starting a new interval */
1285         if (!nvchan->desc)
1286                 nvchan->desc = hv_pkt_iter_first(channel);
1287
1288         while (nvchan->desc && work_done < budget) {
1289                 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1290                                                     ndev, nvchan->desc, budget);
1291                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1292         }
1293
1294         /* Send any pending receive completions */
1295         ret = send_recv_completions(ndev, net_device, nvchan);
1296
1297         /* If it did not exhaust NAPI budget this time
1298          *  and not doing busy poll
1299          * then re-enable host interrupts
1300          *  and reschedule if ring is not empty
1301          *   or sending receive completion failed.
1302          */
1303         if (work_done < budget &&
1304             napi_complete_done(napi, work_done) &&
1305             (ret || hv_end_read(&channel->inbound)) &&
1306             napi_schedule_prep(napi)) {
1307                 hv_begin_read(&channel->inbound);
1308                 __napi_schedule(napi);
1309         }
1310
1311         /* Driver may overshoot since multiple packets per descriptor */
1312         return min(work_done, budget);
1313 }
1314
1315 /* Call back when data is available in host ring buffer.
1316  * Processing is deferred until network softirq (NAPI)
1317  */
1318 void netvsc_channel_cb(void *context)
1319 {
1320         struct netvsc_channel *nvchan = context;
1321         struct vmbus_channel *channel = nvchan->channel;
1322         struct hv_ring_buffer_info *rbi = &channel->inbound;
1323
1324         /* preload first vmpacket descriptor */
1325         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1326
1327         if (napi_schedule_prep(&nvchan->napi)) {
1328                 /* disable interrupts from host */
1329                 hv_begin_read(rbi);
1330
1331                 __napi_schedule_irqoff(&nvchan->napi);
1332         }
1333 }
1334
1335 /*
1336  * netvsc_device_add - Callback when the device belonging to this
1337  * driver is added
1338  */
1339 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1340                                 const struct netvsc_device_info *device_info)
1341 {
1342         int i, ret = 0;
1343         struct netvsc_device *net_device;
1344         struct net_device *ndev = hv_get_drvdata(device);
1345         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1346
1347         net_device = alloc_net_device();
1348         if (!net_device)
1349                 return ERR_PTR(-ENOMEM);
1350
1351         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1352                 net_device_ctx->tx_table[i] = 0;
1353
1354         /* Because the device uses NAPI, all the interrupt batching and
1355          * control is done via Net softirq, not the channel handling
1356          */
1357         set_channel_read_mode(device->channel, HV_CALL_ISR);
1358
1359         /* If we're reopening the device we may have multiple queues, fill the
1360          * chn_table with the default channel to use it before subchannels are
1361          * opened.
1362          * Initialize the channel state before we open;
1363          * we can be interrupted as soon as we open the channel.
1364          */
1365
1366         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1367                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1368
1369                 nvchan->channel = device->channel;
1370                 nvchan->net_device = net_device;
1371                 u64_stats_init(&nvchan->tx_stats.syncp);
1372                 u64_stats_init(&nvchan->rx_stats.syncp);
1373         }
1374
1375         /* Enable NAPI handler before init callbacks */
1376         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1377                        netvsc_poll, NAPI_POLL_WEIGHT);
1378
1379         /* Open the channel */
1380         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1381                          netvsc_ring_bytes,  NULL, 0,
1382                          netvsc_channel_cb, net_device->chan_table);
1383
1384         if (ret != 0) {
1385                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1386                 goto cleanup;
1387         }
1388
1389         /* Channel is opened */
1390         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1391
1392         napi_enable(&net_device->chan_table[0].napi);
1393
1394         /* Connect with the NetVsp */
1395         ret = netvsc_connect_vsp(device, net_device, device_info);
1396         if (ret != 0) {
1397                 netdev_err(ndev,
1398                         "unable to connect to NetVSP - %d\n", ret);
1399                 goto close;
1400         }
1401
1402         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1403          * populated.
1404          */
1405         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1406
1407         return net_device;
1408
1409 close:
1410         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1411         napi_disable(&net_device->chan_table[0].napi);
1412
1413         /* Now, we can close the channel safely */
1414         vmbus_close(device->channel);
1415
1416 cleanup:
1417         netif_napi_del(&net_device->chan_table[0].napi);
1418         free_netvsc_device(&net_device->rcu);
1419
1420         return ERR_PTR(ret);
1421 }