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