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 <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  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 <asm/sync_bitops.h>
  33
  34 #include "hyperv_net.h"
  35
  36 /*
  37  * Switch the data path from the synthetic interface to the VF
  38  * interface.
  39  */
  40 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  41 {
  42         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  43         struct hv_device *dev = net_device_ctx->device_ctx;
  44         struct netvsc_device *nv_dev = net_device_ctx->nvdev;
  45         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  46
  47         memset(init_pkt, 0, sizeof(struct nvsp_message));
  48         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  49         if (vf)
  50                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  51                         NVSP_DATAPATH_VF;
  52         else
  53                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  54                         NVSP_DATAPATH_SYNTHETIC;
  55
  56         vmbus_sendpacket(dev->channel, init_pkt,
  57                                sizeof(struct nvsp_message),
  58                                (unsigned long)init_pkt,
  59                                VM_PKT_DATA_INBAND, 0);
  60 }
  61
  62
  63 static struct netvsc_device *alloc_net_device(void)
  64 {
  65         struct netvsc_device *net_device;
  66
  67         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  68         if (!net_device)
  69                 return NULL;
  70
  71         net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
  72         if (!net_device->cb_buffer) {
  73                 kfree(net_device);
  74                 return NULL;
  75         }
  76
  77         net_device->mrc[0].buf = vzalloc(NETVSC_RECVSLOT_MAX *
  78                                          sizeof(struct recv_comp_data));
  79
  80         init_waitqueue_head(&net_device->wait_drain);
  81         net_device->destroy = false;
  82         atomic_set(&net_device->open_cnt, 0);
  83         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
  84         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
  85
  86         return net_device;
  87 }
  88
  89 static void free_netvsc_device(struct netvsc_device *nvdev)
  90 {
  91         int i;
  92
  93         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
  94                 vfree(nvdev->mrc[i].buf);
  95
  96         kfree(nvdev->cb_buffer);
  97         kfree(nvdev);
  98 }
  99
 100 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
 101 {
 102         struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
 103
 104         if (net_device && net_device->destroy)
 105                 net_device = NULL;
 106
 107         return net_device;
 108 }
 109
 110 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
 111 {
 112         struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
 113
 114         if (!net_device)
 115                 goto get_in_err;
 116
 117         if (net_device->destroy &&
 118             atomic_read(&net_device->num_outstanding_sends) == 0 &&
 119             atomic_read(&net_device->num_outstanding_recvs) == 0)
 120                 net_device = NULL;
 121
 122 get_in_err:
 123         return net_device;
 124 }
 125
 126
 127 static int netvsc_destroy_buf(struct hv_device *device)
 128 {
 129         struct nvsp_message *revoke_packet;
 130         int ret = 0;
 131         struct net_device *ndev = hv_get_drvdata(device);
 132         struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
 133
 134         /*
 135          * If we got a section count, it means we received a
 136          * SendReceiveBufferComplete msg (ie sent
 137          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 138          * to send a revoke msg here
 139          */
 140         if (net_device->recv_section_cnt) {
 141                 /* Send the revoke receive buffer */
 142                 revoke_packet = &net_device->revoke_packet;
 143                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 144
 145                 revoke_packet->hdr.msg_type =
 146                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 147                 revoke_packet->msg.v1_msg.
 148                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 149
 150                 ret = vmbus_sendpacket(device->channel,
 151                                        revoke_packet,
 152                                        sizeof(struct nvsp_message),
 153                                        (unsigned long)revoke_packet,
 154                                        VM_PKT_DATA_INBAND, 0);
 155                 /*
 156                  * If we failed here, we might as well return and
 157                  * have a leak rather than continue and a bugchk
 158                  */
 159                 if (ret != 0) {
 160                         netdev_err(ndev, "unable to send "
 161                                 "revoke receive buffer to netvsp\n");
 162                         return ret;
 163                 }
 164         }
 165
 166         /* Teardown the gpadl on the vsp end */
 167         if (net_device->recv_buf_gpadl_handle) {
 168                 ret = vmbus_teardown_gpadl(device->channel,
 169                                            net_device->recv_buf_gpadl_handle);
 170
 171                 /* If we failed here, we might as well return and have a leak
 172                  * rather than continue and a bugchk
 173                  */
 174                 if (ret != 0) {
 175                         netdev_err(ndev,
 176                                    "unable to teardown receive buffer's gpadl\n");
 177                         return ret;
 178                 }
 179                 net_device->recv_buf_gpadl_handle = 0;
 180         }
 181
 182         if (net_device->recv_buf) {
 183                 /* Free up the receive buffer */
 184                 vfree(net_device->recv_buf);
 185                 net_device->recv_buf = NULL;
 186         }
 187
 188         if (net_device->recv_section) {
 189                 net_device->recv_section_cnt = 0;
 190                 kfree(net_device->recv_section);
 191                 net_device->recv_section = NULL;
 192         }
 193
 194         /* Deal with the send buffer we may have setup.
 195          * If we got a  send section size, it means we received a
 196          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 197          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 198          * to send a revoke msg here
 199          */
 200         if (net_device->send_section_size) {
 201                 /* Send the revoke receive buffer */
 202                 revoke_packet = &net_device->revoke_packet;
 203                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 204
 205                 revoke_packet->hdr.msg_type =
 206                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 207                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 208                         NETVSC_SEND_BUFFER_ID;
 209
 210                 ret = vmbus_sendpacket(device->channel,
 211                                        revoke_packet,
 212                                        sizeof(struct nvsp_message),
 213                                        (unsigned long)revoke_packet,
 214                                        VM_PKT_DATA_INBAND, 0);
 215                 /* If we failed here, we might as well return and
 216                  * have a leak rather than continue and a bugchk
 217                  */
 218                 if (ret != 0) {
 219                         netdev_err(ndev, "unable to send "
 220                                    "revoke send buffer to netvsp\n");
 221                         return ret;
 222                 }
 223         }
 224         /* Teardown the gpadl on the vsp end */
 225         if (net_device->send_buf_gpadl_handle) {
 226                 ret = vmbus_teardown_gpadl(device->channel,
 227                                            net_device->send_buf_gpadl_handle);
 228
 229                 /* If we failed here, we might as well return and have a leak
 230                  * rather than continue and a bugchk
 231                  */
 232                 if (ret != 0) {
 233                         netdev_err(ndev,
 234                                    "unable to teardown send buffer's gpadl\n");
 235                         return ret;
 236                 }
 237                 net_device->send_buf_gpadl_handle = 0;
 238         }
 239         if (net_device->send_buf) {
 240                 /* Free up the send buffer */
 241                 vfree(net_device->send_buf);
 242                 net_device->send_buf = NULL;
 243         }
 244         kfree(net_device->send_section_map);
 245
 246         return ret;
 247 }
 248
 249 static int netvsc_init_buf(struct hv_device *device)
 250 {
 251         int ret = 0;
 252         struct netvsc_device *net_device;
 253         struct nvsp_message *init_packet;
 254         struct net_device *ndev;
 255         int node;
 256
 257         net_device = get_outbound_net_device(device);
 258         if (!net_device)
 259                 return -ENODEV;
 260         ndev = hv_get_drvdata(device);
 261
 262         node = cpu_to_node(device->channel->target_cpu);
 263         net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
 264         if (!net_device->recv_buf)
 265                 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 266
 267         if (!net_device->recv_buf) {
 268                 netdev_err(ndev, "unable to allocate receive "
 269                         "buffer of size %d\n", net_device->recv_buf_size);
 270                 ret = -ENOMEM;
 271                 goto cleanup;
 272         }
 273
 274         /*
 275          * Establish the gpadl handle for this buffer on this
 276          * channel.  Note: This call uses the vmbus connection rather
 277          * than the channel to establish the gpadl handle.
 278          */
 279         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 280                                     net_device->recv_buf_size,
 281                                     &net_device->recv_buf_gpadl_handle);
 282         if (ret != 0) {
 283                 netdev_err(ndev,
 284                         "unable to establish receive buffer's gpadl\n");
 285                 goto cleanup;
 286         }
 287
 288
 289         /* Notify the NetVsp of the gpadl handle */
 290         init_packet = &net_device->channel_init_pkt;
 291
 292         memset(init_packet, 0, sizeof(struct nvsp_message));
 293
 294         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 295         init_packet->msg.v1_msg.send_recv_buf.
 296                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 297         init_packet->msg.v1_msg.
 298                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 299
 300         /* Send the gpadl notification request */
 301         ret = vmbus_sendpacket(device->channel, init_packet,
 302                                sizeof(struct nvsp_message),
 303                                (unsigned long)init_packet,
 304                                VM_PKT_DATA_INBAND,
 305                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 306         if (ret != 0) {
 307                 netdev_err(ndev,
 308                         "unable to send receive buffer's gpadl to netvsp\n");
 309                 goto cleanup;
 310         }
 311
 312         wait_for_completion(&net_device->channel_init_wait);
 313
 314         /* Check the response */
 315         if (init_packet->msg.v1_msg.
 316             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 317                 netdev_err(ndev, "Unable to complete receive buffer "
 318                            "initialization with NetVsp - status %d\n",
 319                            init_packet->msg.v1_msg.
 320                            send_recv_buf_complete.status);
 321                 ret = -EINVAL;
 322                 goto cleanup;
 323         }
 324
 325         /* Parse the response */
 326
 327         net_device->recv_section_cnt = init_packet->msg.
 328                 v1_msg.send_recv_buf_complete.num_sections;
 329
 330         net_device->recv_section = kmemdup(
 331                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 332                 net_device->recv_section_cnt *
 333                 sizeof(struct nvsp_1_receive_buffer_section),
 334                 GFP_KERNEL);
 335         if (net_device->recv_section == NULL) {
 336                 ret = -EINVAL;
 337                 goto cleanup;
 338         }
 339
 340         /*
 341          * For 1st release, there should only be 1 section that represents the
 342          * entire receive buffer
 343          */
 344         if (net_device->recv_section_cnt != 1 ||
 345             net_device->recv_section->offset != 0) {
 346                 ret = -EINVAL;
 347                 goto cleanup;
 348         }
 349
 350         /* Now setup the send buffer.
 351          */
 352         net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
 353         if (!net_device->send_buf)
 354                 net_device->send_buf = vzalloc(net_device->send_buf_size);
 355         if (!net_device->send_buf) {
 356                 netdev_err(ndev, "unable to allocate send "
 357                            "buffer of size %d\n", net_device->send_buf_size);
 358                 ret = -ENOMEM;
 359                 goto cleanup;
 360         }
 361
 362         /* Establish the gpadl handle for this buffer on this
 363          * channel.  Note: This call uses the vmbus connection rather
 364          * than the channel to establish the gpadl handle.
 365          */
 366         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 367                                     net_device->send_buf_size,
 368                                     &net_device->send_buf_gpadl_handle);
 369         if (ret != 0) {
 370                 netdev_err(ndev,
 371                            "unable to establish send buffer's gpadl\n");
 372                 goto cleanup;
 373         }
 374
 375         /* Notify the NetVsp of the gpadl handle */
 376         init_packet = &net_device->channel_init_pkt;
 377         memset(init_packet, 0, sizeof(struct nvsp_message));
 378         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 379         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 380                 net_device->send_buf_gpadl_handle;
 381         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 382
 383         /* Send the gpadl notification request */
 384         ret = vmbus_sendpacket(device->channel, init_packet,
 385                                sizeof(struct nvsp_message),
 386                                (unsigned long)init_packet,
 387                                VM_PKT_DATA_INBAND,
 388                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 389         if (ret != 0) {
 390                 netdev_err(ndev,
 391                            "unable to send send buffer's gpadl to netvsp\n");
 392                 goto cleanup;
 393         }
 394
 395         wait_for_completion(&net_device->channel_init_wait);
 396
 397         /* Check the response */
 398         if (init_packet->msg.v1_msg.
 399             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 400                 netdev_err(ndev, "Unable to complete send buffer "
 401                            "initialization with NetVsp - status %d\n",
 402                            init_packet->msg.v1_msg.
 403                            send_send_buf_complete.status);
 404                 ret = -EINVAL;
 405                 goto cleanup;
 406         }
 407
 408         /* Parse the response */
 409         net_device->send_section_size = init_packet->msg.
 410                                 v1_msg.send_send_buf_complete.section_size;
 411
 412         /* Section count is simply the size divided by the section size.
 413          */
 414         net_device->send_section_cnt =
 415                 net_device->send_buf_size/net_device->send_section_size;
 416
 417         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 418                  net_device->send_section_size, net_device->send_section_cnt);
 419
 420         /* Setup state for managing the send buffer. */
 421         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 422                                              BITS_PER_LONG);
 423
 424         net_device->send_section_map =
 425                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 426         if (net_device->send_section_map == NULL) {
 427                 ret = -ENOMEM;
 428                 goto cleanup;
 429         }
 430
 431         goto exit;
 432
 433 cleanup:
 434         netvsc_destroy_buf(device);
 435
 436 exit:
 437         return ret;
 438 }
 439
 440
 441 /* Negotiate NVSP protocol version */
 442 static int negotiate_nvsp_ver(struct hv_device *device,
 443                               struct netvsc_device *net_device,
 444                               struct nvsp_message *init_packet,
 445                               u32 nvsp_ver)
 446 {
 447         struct net_device *ndev = hv_get_drvdata(device);
 448         int ret;
 449
 450         memset(init_packet, 0, sizeof(struct nvsp_message));
 451         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 452         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 453         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 454
 455         /* Send the init request */
 456         ret = vmbus_sendpacket(device->channel, init_packet,
 457                                sizeof(struct nvsp_message),
 458                                (unsigned long)init_packet,
 459                                VM_PKT_DATA_INBAND,
 460                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 461
 462         if (ret != 0)
 463                 return ret;
 464
 465         wait_for_completion(&net_device->channel_init_wait);
 466
 467         if (init_packet->msg.init_msg.init_complete.status !=
 468             NVSP_STAT_SUCCESS)
 469                 return -EINVAL;
 470
 471         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 472                 return 0;
 473
 474         /* NVSPv2 or later: Send NDIS config */
 475         memset(init_packet, 0, sizeof(struct nvsp_message));
 476         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 477         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 478         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 479
 480         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 481                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 482
 483                 /* Teaming bit is needed to receive link speed updates */
 484                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 485         }
 486
 487         ret = vmbus_sendpacket(device->channel, init_packet,
 488                                 sizeof(struct nvsp_message),
 489                                 (unsigned long)init_packet,
 490                                 VM_PKT_DATA_INBAND, 0);
 491
 492         return ret;
 493 }
 494
 495 static int netvsc_connect_vsp(struct hv_device *device)
 496 {
 497         int ret;
 498         struct netvsc_device *net_device;
 499         struct nvsp_message *init_packet;
 500         int ndis_version;
 501         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 502                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 503         int i, num_ver = 4; /* number of different NVSP versions */
 504
 505         net_device = get_outbound_net_device(device);
 506         if (!net_device)
 507                 return -ENODEV;
 508
 509         init_packet = &net_device->channel_init_pkt;
 510
 511         /* Negotiate the latest NVSP protocol supported */
 512         for (i = num_ver - 1; i >= 0; i--)
 513                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 514                                        ver_list[i])  == 0) {
 515                         net_device->nvsp_version = ver_list[i];
 516                         break;
 517                 }
 518
 519         if (i < 0) {
 520                 ret = -EPROTO;
 521                 goto cleanup;
 522         }
 523
 524         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 525
 526         /* Send the ndis version */
 527         memset(init_packet, 0, sizeof(struct nvsp_message));
 528
 529         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 530                 ndis_version = 0x00060001;
 531         else
 532                 ndis_version = 0x0006001e;
 533
 534         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 535         init_packet->msg.v1_msg.
 536                 send_ndis_ver.ndis_major_ver =
 537                                 (ndis_version & 0xFFFF0000) >> 16;
 538         init_packet->msg.v1_msg.
 539                 send_ndis_ver.ndis_minor_ver =
 540                                 ndis_version & 0xFFFF;
 541
 542         /* Send the init request */
 543         ret = vmbus_sendpacket(device->channel, init_packet,
 544                                 sizeof(struct nvsp_message),
 545                                 (unsigned long)init_packet,
 546                                 VM_PKT_DATA_INBAND, 0);
 547         if (ret != 0)
 548                 goto cleanup;
 549
 550         /* Post the big receive buffer to NetVSP */
 551         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 552                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 553         else
 554                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 555         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 556
 557         ret = netvsc_init_buf(device);
 558
 559 cleanup:
 560         return ret;
 561 }
 562
 563 static void netvsc_disconnect_vsp(struct hv_device *device)
 564 {
 565         netvsc_destroy_buf(device);
 566 }
 567
 568 /*
 569  * netvsc_device_remove - Callback when the root bus device is removed
 570  */
 571 int netvsc_device_remove(struct hv_device *device)
 572 {
 573         struct net_device *ndev = hv_get_drvdata(device);
 574         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 575         struct netvsc_device *net_device = net_device_ctx->nvdev;
 576
 577         netvsc_disconnect_vsp(device);
 578
 579         net_device_ctx->nvdev = NULL;
 580
 581         /*
 582          * At this point, no one should be accessing net_device
 583          * except in here
 584          */
 585         dev_notice(&device->device, "net device safe to remove\n");
 586
 587         /* Now, we can close the channel safely */
 588         vmbus_close(device->channel);
 589
 590         /* Release all resources */
 591         vfree(net_device->sub_cb_buf);
 592         free_netvsc_device(net_device);
 593         return 0;
 594 }
 595
 596
 597 #define RING_AVAIL_PERCENT_HIWATER 20
 598 #define RING_AVAIL_PERCENT_LOWATER 10
 599
 600 /*
 601  * Get the percentage of available bytes to write in the ring.
 602  * The return value is in range from 0 to 100.
 603  */
 604 static inline u32 hv_ringbuf_avail_percent(
 605                 struct hv_ring_buffer_info *ring_info)
 606 {
 607         u32 avail_read, avail_write;
 608
 609         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 610
 611         return avail_write * 100 / ring_info->ring_datasize;
 612 }
 613
 614 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 615                                          u32 index)
 616 {
 617         sync_change_bit(index, net_device->send_section_map);
 618 }
 619
 620 static void netvsc_send_completion(struct netvsc_device *net_device,
 621                                    struct vmbus_channel *incoming_channel,
 622                                    struct hv_device *device,
 623                                    struct vmpacket_descriptor *packet)
 624 {
 625         struct nvsp_message *nvsp_packet;
 626         struct hv_netvsc_packet *nvsc_packet;
 627         struct net_device *ndev = hv_get_drvdata(device);
 628         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 629         u32 send_index;
 630         struct sk_buff *skb;
 631
 632         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 633                         (packet->offset8 << 3));
 634
 635         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 636             (nvsp_packet->hdr.msg_type ==
 637              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 638             (nvsp_packet->hdr.msg_type ==
 639              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
 640             (nvsp_packet->hdr.msg_type ==
 641              NVSP_MSG5_TYPE_SUBCHANNEL)) {
 642                 /* Copy the response back */
 643                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 644                        sizeof(struct nvsp_message));
 645                 complete(&net_device->channel_init_wait);
 646         } else if (nvsp_packet->hdr.msg_type ==
 647                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 648                 int num_outstanding_sends;
 649                 u16 q_idx = 0;
 650                 struct vmbus_channel *channel = device->channel;
 651                 int queue_sends;
 652
 653                 /* Get the send context */
 654                 skb = (struct sk_buff *)(unsigned long)packet->trans_id;
 655
 656                 /* Notify the layer above us */
 657                 if (skb) {
 658                         nvsc_packet = (struct hv_netvsc_packet *) skb->cb;
 659                         send_index = nvsc_packet->send_buf_index;
 660                         if (send_index != NETVSC_INVALID_INDEX)
 661                                 netvsc_free_send_slot(net_device, send_index);
 662                         q_idx = nvsc_packet->q_idx;
 663                         channel = incoming_channel;
 664                         dev_kfree_skb_any(skb);
 665                 }
 666
 667                 num_outstanding_sends =
 668                         atomic_dec_return(&net_device->num_outstanding_sends);
 669                 queue_sends = atomic_dec_return(&net_device->
 670                                                 queue_sends[q_idx]);
 671
 672                 if (net_device->destroy && num_outstanding_sends == 0)
 673                         wake_up(&net_device->wait_drain);
 674
 675                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 676                     !net_device_ctx->start_remove &&
 677                     (hv_ringbuf_avail_percent(&channel->outbound) >
 678                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 679                                 netif_tx_wake_queue(netdev_get_tx_queue(
 680                                                     ndev, q_idx));
 681         } else {
 682                 netdev_err(ndev, "Unknown send completion packet type- "
 683                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 684         }
 685
 686 }
 687
 688 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 689 {
 690         unsigned long index;
 691         u32 max_words = net_device->map_words;
 692         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 693         u32 section_cnt = net_device->send_section_cnt;
 694         int ret_val = NETVSC_INVALID_INDEX;
 695         int i;
 696         int prev_val;
 697
 698         for (i = 0; i < max_words; i++) {
 699                 if (!~(map_addr[i]))
 700                         continue;
 701                 index = ffz(map_addr[i]);
 702                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 703                 if (prev_val)
 704                         continue;
 705                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 706                         break;
 707                 ret_val = (index + (i * BITS_PER_LONG));
 708                 break;
 709         }
 710         return ret_val;
 711 }
 712
 713 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 714                                    unsigned int section_index,
 715                                    u32 pend_size,
 716                                    struct hv_netvsc_packet *packet,
 717                                    struct rndis_message *rndis_msg,
 718                                    struct hv_page_buffer **pb,
 719                                    struct sk_buff *skb)
 720 {
 721         char *start = net_device->send_buf;
 722         char *dest = start + (section_index * net_device->send_section_size)
 723                      + pend_size;
 724         int i;
 725         bool is_data_pkt = (skb != NULL) ? true : false;
 726         bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
 727         u32 msg_size = 0;
 728         u32 padding = 0;
 729         u32 remain = packet->total_data_buflen % net_device->pkt_align;
 730         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 731                 packet->page_buf_cnt;
 732
 733         /* Add padding */
 734         if (is_data_pkt && xmit_more && remain &&
 735             !packet->cp_partial) {
 736                 padding = net_device->pkt_align - remain;
 737                 rndis_msg->msg_len += padding;
 738                 packet->total_data_buflen += padding;
 739         }
 740
 741         for (i = 0; i < page_count; i++) {
 742                 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
 743                 u32 offset = (*pb)[i].offset;
 744                 u32 len = (*pb)[i].len;
 745
 746                 memcpy(dest, (src + offset), len);
 747                 msg_size += len;
 748                 dest += len;
 749         }
 750
 751         if (padding) {
 752                 memset(dest, 0, padding);
 753                 msg_size += padding;
 754         }
 755
 756         return msg_size;
 757 }
 758
 759 static inline int netvsc_send_pkt(
 760         struct hv_device *device,
 761         struct hv_netvsc_packet *packet,
 762         struct netvsc_device *net_device,
 763         struct hv_page_buffer **pb,
 764         struct sk_buff *skb)
 765 {
 766         struct nvsp_message nvmsg;
 767         u16 q_idx = packet->q_idx;
 768         struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
 769         struct net_device *ndev = hv_get_drvdata(device);
 770         u64 req_id;
 771         int ret;
 772         struct hv_page_buffer *pgbuf;
 773         u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
 774         bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
 775
 776         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 777         if (skb != NULL) {
 778                 /* 0 is RMC_DATA; */
 779                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 780         } else {
 781                 /* 1 is RMC_CONTROL; */
 782                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 783         }
 784
 785         nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 786                 packet->send_buf_index;
 787         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 788                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 789         else
 790                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
 791                         packet->total_data_buflen;
 792
 793         req_id = (ulong)skb;
 794
 795         if (out_channel->rescind)
 796                 return -ENODEV;
 797
 798         /*
 799          * It is possible that once we successfully place this packet
 800          * on the ringbuffer, we may stop the queue. In that case, we want
 801          * to notify the host independent of the xmit_more flag. We don't
 802          * need to be precise here; in the worst case we may signal the host
 803          * unnecessarily.
 804          */
 805         if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
 806                 xmit_more = false;
 807
 808         if (packet->page_buf_cnt) {
 809                 pgbuf = packet->cp_partial ? (*pb) +
 810                         packet->rmsg_pgcnt : (*pb);
 811                 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 812                                                       pgbuf,
 813                                                       packet->page_buf_cnt,
 814                                                       &nvmsg,
 815                                                       sizeof(struct nvsp_message),
 816                                                       req_id,
 817                                                       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
 818                                                       !xmit_more);
 819         } else {
 820                 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
 821                                            sizeof(struct nvsp_message),
 822                                            req_id,
 823                                            VM_PKT_DATA_INBAND,
 824                                            VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
 825                                            !xmit_more);
 826         }
 827
 828         if (ret == 0) {
 829                 atomic_inc(&net_device->num_outstanding_sends);
 830                 atomic_inc(&net_device->queue_sends[q_idx]);
 831
 832                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 833                         netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
 834
 835                         if (atomic_read(&net_device->
 836                                 queue_sends[q_idx]) < 1)
 837                                 netif_tx_wake_queue(netdev_get_tx_queue(
 838                                                     ndev, q_idx));
 839                 }
 840         } else if (ret == -EAGAIN) {
 841                 netif_tx_stop_queue(netdev_get_tx_queue(
 842                                     ndev, q_idx));
 843                 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
 844                         netif_tx_wake_queue(netdev_get_tx_queue(
 845                                             ndev, q_idx));
 846                         ret = -ENOSPC;
 847                 }
 848         } else {
 849                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 850                            packet, ret);
 851         }
 852
 853         return ret;
 854 }
 855
 856 /* Move packet out of multi send data (msd), and clear msd */
 857 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 858                                 struct sk_buff **msd_skb,
 859                                 struct multi_send_data *msdp)
 860 {
 861         *msd_skb = msdp->skb;
 862         *msd_send = msdp->pkt;
 863         msdp->skb = NULL;
 864         msdp->pkt = NULL;
 865         msdp->count = 0;
 866 }
 867
 868 int netvsc_send(struct hv_device *device,
 869                 struct hv_netvsc_packet *packet,
 870                 struct rndis_message *rndis_msg,
 871                 struct hv_page_buffer **pb,
 872                 struct sk_buff *skb)
 873 {
 874         struct netvsc_device *net_device;
 875         int ret = 0, m_ret = 0;
 876         struct vmbus_channel *out_channel;
 877         u16 q_idx = packet->q_idx;
 878         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 879         unsigned int section_index = NETVSC_INVALID_INDEX;
 880         struct multi_send_data *msdp;
 881         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 882         struct sk_buff *msd_skb = NULL;
 883         bool try_batch;
 884         bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
 885
 886         net_device = get_outbound_net_device(device);
 887         if (!net_device)
 888                 return -ENODEV;
 889
 890         out_channel = net_device->chn_table[q_idx];
 891
 892         packet->send_buf_index = NETVSC_INVALID_INDEX;
 893         packet->cp_partial = false;
 894
 895         /* Send control message directly without accessing msd (Multi-Send
 896          * Data) field which may be changed during data packet processing.
 897          */
 898         if (!skb) {
 899                 cur_send = packet;
 900                 goto send_now;
 901         }
 902
 903         msdp = &net_device->msd[q_idx];
 904
 905         /* batch packets in send buffer if possible */
 906         if (msdp->pkt)
 907                 msd_len = msdp->pkt->total_data_buflen;
 908
 909         try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
 910                     net_device->max_pkt;
 911
 912         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 913             net_device->send_section_size) {
 914                 section_index = msdp->pkt->send_buf_index;
 915
 916         } else if (try_batch && msd_len + packet->rmsg_size <
 917                    net_device->send_section_size) {
 918                 section_index = msdp->pkt->send_buf_index;
 919                 packet->cp_partial = true;
 920
 921         } else if ((skb != NULL) && pktlen + net_device->pkt_align <
 922                    net_device->send_section_size) {
 923                 section_index = netvsc_get_next_send_section(net_device);
 924                 if (section_index != NETVSC_INVALID_INDEX) {
 925                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 926                         msd_len = 0;
 927                 }
 928         }
 929
 930         if (section_index != NETVSC_INVALID_INDEX) {
 931                 netvsc_copy_to_send_buf(net_device,
 932                                         section_index, msd_len,
 933                                         packet, rndis_msg, pb, skb);
 934
 935                 packet->send_buf_index = section_index;
 936
 937                 if (packet->cp_partial) {
 938                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 939                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 940                 } else {
 941                         packet->page_buf_cnt = 0;
 942                         packet->total_data_buflen += msd_len;
 943                 }
 944
 945                 if (msdp->skb)
 946                         dev_kfree_skb_any(msdp->skb);
 947
 948                 if (xmit_more && !packet->cp_partial) {
 949                         msdp->skb = skb;
 950                         msdp->pkt = packet;
 951                         msdp->count++;
 952                 } else {
 953                         cur_send = packet;
 954                         msdp->skb = NULL;
 955                         msdp->pkt = NULL;
 956                         msdp->count = 0;
 957                 }
 958         } else {
 959                 move_pkt_msd(&msd_send, &msd_skb, msdp);
 960                 cur_send = packet;
 961         }
 962
 963         if (msd_send) {
 964                 m_ret = netvsc_send_pkt(device, msd_send, net_device,
 965                                         NULL, msd_skb);
 966
 967                 if (m_ret != 0) {
 968                         netvsc_free_send_slot(net_device,
 969                                               msd_send->send_buf_index);
 970                         dev_kfree_skb_any(msd_skb);
 971                 }
 972         }
 973
 974 send_now:
 975         if (cur_send)
 976                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
 977
 978         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
 979                 netvsc_free_send_slot(net_device, section_index);
 980
 981         return ret;
 982 }
 983
 984 static int netvsc_send_recv_completion(struct vmbus_channel *channel,
 985                                        u64 transaction_id, u32 status)
 986 {
 987         struct nvsp_message recvcompMessage;
 988         int ret;
 989
 990         recvcompMessage.hdr.msg_type =
 991                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 992
 993         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 994
 995         /* Send the completion */
 996         ret = vmbus_sendpacket(channel, &recvcompMessage,
 997                                sizeof(struct nvsp_message_header) + sizeof(u32),
 998                                transaction_id, VM_PKT_COMP, 0);
 999
1000         return ret;
1001 }
1002
1003 static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
1004                                         u32 *filled, u32 *avail)
1005 {
1006         u32 first = nvdev->mrc[q_idx].first;
1007         u32 next = nvdev->mrc[q_idx].next;
1008
1009         *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
1010                   next - first;
1011
1012         *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
1013 }
1014
1015 /* Read the first filled slot, no change to index */
1016 static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
1017                                                          *nvdev, u16 q_idx)
1018 {
1019         u32 filled, avail;
1020
1021         if (!nvdev->mrc[q_idx].buf)
1022                 return NULL;
1023
1024         count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1025         if (!filled)
1026                 return NULL;
1027
1028         return nvdev->mrc[q_idx].buf + nvdev->mrc[q_idx].first *
1029                sizeof(struct recv_comp_data);
1030 }
1031
1032 /* Put the first filled slot back to available pool */
1033 static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
1034 {
1035         int num_recv;
1036
1037         nvdev->mrc[q_idx].first = (nvdev->mrc[q_idx].first + 1) %
1038                                   NETVSC_RECVSLOT_MAX;
1039
1040         num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1041
1042         if (nvdev->destroy && num_recv == 0)
1043                 wake_up(&nvdev->wait_drain);
1044 }
1045
1046 /* Check and send pending recv completions */
1047 static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1048                                  struct vmbus_channel *channel, u16 q_idx)
1049 {
1050         struct recv_comp_data *rcd;
1051         int ret;
1052
1053         while (true) {
1054                 rcd = read_recv_comp_slot(nvdev, q_idx);
1055                 if (!rcd)
1056                         break;
1057
1058                 ret = netvsc_send_recv_completion(channel, rcd->tid,
1059                                                   rcd->status);
1060                 if (ret)
1061                         break;
1062
1063                 put_recv_comp_slot(nvdev, q_idx);
1064         }
1065 }
1066
1067 #define NETVSC_RCD_WATERMARK 80
1068
1069 /* Get next available slot */
1070 static inline struct recv_comp_data *get_recv_comp_slot(
1071         struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1072 {
1073         u32 filled, avail, next;
1074         struct recv_comp_data *rcd;
1075
1076         if (!nvdev->recv_section)
1077                 return NULL;
1078
1079         if (!nvdev->mrc[q_idx].buf)
1080                 return NULL;
1081
1082         if (atomic_read(&nvdev->num_outstanding_recvs) >
1083             nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1084                 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1085
1086         count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1087         if (!avail)
1088                 return NULL;
1089
1090         next = nvdev->mrc[q_idx].next;
1091         rcd = nvdev->mrc[q_idx].buf + next * sizeof(struct recv_comp_data);
1092         nvdev->mrc[q_idx].next = (next + 1) % NETVSC_RECVSLOT_MAX;
1093
1094         atomic_inc(&nvdev->num_outstanding_recvs);
1095
1096         return rcd;
1097 }
1098
1099 static void netvsc_receive(struct netvsc_device *net_device,
1100                         struct vmbus_channel *channel,
1101                         struct hv_device *device,
1102                         struct vmpacket_descriptor *packet)
1103 {
1104         struct vmtransfer_page_packet_header *vmxferpage_packet;
1105         struct nvsp_message *nvsp_packet;
1106         struct hv_netvsc_packet nv_pkt;
1107         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1108         u32 status = NVSP_STAT_SUCCESS;
1109         int i;
1110         int count = 0;
1111         struct net_device *ndev = hv_get_drvdata(device);
1112         void *data;
1113         int ret;
1114         struct recv_comp_data *rcd;
1115         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1116
1117         /*
1118          * All inbound packets other than send completion should be xfer page
1119          * packet
1120          */
1121         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1122                 netdev_err(ndev, "Unknown packet type received - %d\n",
1123                            packet->type);
1124                 return;
1125         }
1126
1127         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1128                         (packet->offset8 << 3));
1129
1130         /* Make sure this is a valid nvsp packet */
1131         if (nvsp_packet->hdr.msg_type !=
1132             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1133                 netdev_err(ndev, "Unknown nvsp packet type received-"
1134                         " %d\n", nvsp_packet->hdr.msg_type);
1135                 return;
1136         }
1137
1138         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1139
1140         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1141                 netdev_err(ndev, "Invalid xfer page set id - "
1142                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1143                            vmxferpage_packet->xfer_pageset_id);
1144                 return;
1145         }
1146
1147         count = vmxferpage_packet->range_cnt;
1148
1149         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1150         for (i = 0; i < count; i++) {
1151                 /* Initialize the netvsc packet */
1152                 data = (void *)((unsigned long)net_device->
1153                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1154                 netvsc_packet->total_data_buflen =
1155                                         vmxferpage_packet->ranges[i].byte_count;
1156
1157                 /* Pass it to the upper layer */
1158                 status = rndis_filter_receive(device, netvsc_packet, &data,
1159                                               channel);
1160
1161         }
1162
1163         if (!net_device->mrc[q_idx].buf) {
1164                 ret = netvsc_send_recv_completion(channel,
1165                                                   vmxferpage_packet->d.trans_id,
1166                                                   status);
1167                 if (ret)
1168                         netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1169                                    q_idx, vmxferpage_packet->d.trans_id, ret);
1170                 return;
1171         }
1172
1173         rcd = get_recv_comp_slot(net_device, channel, q_idx);
1174
1175         if (!rcd) {
1176                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1177                            q_idx, vmxferpage_packet->d.trans_id);
1178                 return;
1179         }
1180
1181         rcd->tid = vmxferpage_packet->d.trans_id;
1182         rcd->status = status;
1183 }
1184
1185
1186 static void netvsc_send_table(struct hv_device *hdev,
1187                               struct nvsp_message *nvmsg)
1188 {
1189         struct netvsc_device *nvscdev;
1190         struct net_device *ndev = hv_get_drvdata(hdev);
1191         int i;
1192         u32 count, *tab;
1193
1194         nvscdev = get_outbound_net_device(hdev);
1195         if (!nvscdev)
1196                 return;
1197
1198         count = nvmsg->msg.v5_msg.send_table.count;
1199         if (count != VRSS_SEND_TAB_SIZE) {
1200                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1201                 return;
1202         }
1203
1204         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1205                       nvmsg->msg.v5_msg.send_table.offset);
1206
1207         for (i = 0; i < count; i++)
1208                 nvscdev->send_table[i] = tab[i];
1209 }
1210
1211 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1212                            struct nvsp_message *nvmsg)
1213 {
1214         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1215         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1216 }
1217
1218 static inline void netvsc_receive_inband(struct hv_device *hdev,
1219                                  struct net_device_context *net_device_ctx,
1220                                  struct nvsp_message *nvmsg)
1221 {
1222         switch (nvmsg->hdr.msg_type) {
1223         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1224                 netvsc_send_table(hdev, nvmsg);
1225                 break;
1226
1227         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1228                 netvsc_send_vf(net_device_ctx, nvmsg);
1229                 break;
1230         }
1231 }
1232
1233 static void netvsc_process_raw_pkt(struct hv_device *device,
1234                                    struct vmbus_channel *channel,
1235                                    struct netvsc_device *net_device,
1236                                    struct net_device *ndev,
1237                                    u64 request_id,
1238                                    struct vmpacket_descriptor *desc)
1239 {
1240         struct nvsp_message *nvmsg;
1241         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1242
1243         nvmsg = (struct nvsp_message *)((unsigned long)
1244                 desc + (desc->offset8 << 3));
1245
1246         switch (desc->type) {
1247         case VM_PKT_COMP:
1248                 netvsc_send_completion(net_device, channel, device, desc);
1249                 break;
1250
1251         case VM_PKT_DATA_USING_XFER_PAGES:
1252                 netvsc_receive(net_device, channel, device, desc);
1253                 break;
1254
1255         case VM_PKT_DATA_INBAND:
1256                 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1257                 break;
1258
1259         default:
1260                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1261                            desc->type, request_id);
1262                 break;
1263         }
1264 }
1265
1266
1267 void netvsc_channel_cb(void *context)
1268 {
1269         int ret;
1270         struct vmbus_channel *channel = (struct vmbus_channel *)context;
1271         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1272         struct hv_device *device;
1273         struct netvsc_device *net_device;
1274         u32 bytes_recvd;
1275         u64 request_id;
1276         struct vmpacket_descriptor *desc;
1277         unsigned char *buffer;
1278         int bufferlen = NETVSC_PACKET_SIZE;
1279         struct net_device *ndev;
1280         bool need_to_commit = false;
1281
1282         if (channel->primary_channel != NULL)
1283                 device = channel->primary_channel->device_obj;
1284         else
1285                 device = channel->device_obj;
1286
1287         net_device = get_inbound_net_device(device);
1288         if (!net_device)
1289                 return;
1290         ndev = hv_get_drvdata(device);
1291         buffer = get_per_channel_state(channel);
1292
1293         do {
1294                 desc = get_next_pkt_raw(channel);
1295                 if (desc != NULL) {
1296                         netvsc_process_raw_pkt(device,
1297                                                channel,
1298                                                net_device,
1299                                                ndev,
1300                                                desc->trans_id,
1301                                                desc);
1302
1303                         put_pkt_raw(channel, desc);
1304                         need_to_commit = true;
1305                         continue;
1306                 }
1307                 if (need_to_commit) {
1308                         need_to_commit = false;
1309                         commit_rd_index(channel);
1310                 }
1311
1312                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1313                                            &bytes_recvd, &request_id);
1314                 if (ret == 0) {
1315                         if (bytes_recvd > 0) {
1316                                 desc = (struct vmpacket_descriptor *)buffer;
1317                                 netvsc_process_raw_pkt(device,
1318                                                        channel,
1319                                                        net_device,
1320                                                        ndev,
1321                                                        request_id,
1322                                                        desc);
1323
1324
1325                         } else {
1326                                 /*
1327                                  * We are done for this pass.
1328                                  */
1329                                 break;
1330                         }
1331
1332                 } else if (ret == -ENOBUFS) {
1333                         if (bufferlen > NETVSC_PACKET_SIZE)
1334                                 kfree(buffer);
1335                         /* Handle large packet */
1336                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1337                         if (buffer == NULL) {
1338                                 /* Try again next time around */
1339                                 netdev_err(ndev,
1340                                            "unable to allocate buffer of size "
1341                                            "(%d)!!\n", bytes_recvd);
1342                                 break;
1343                         }
1344
1345                         bufferlen = bytes_recvd;
1346                 }
1347         } while (1);
1348
1349         if (bufferlen > NETVSC_PACKET_SIZE)
1350                 kfree(buffer);
1351
1352         netvsc_chk_recv_comp(net_device, channel, q_idx);
1353
1354         return;
1355 }
1356
1357 /*
1358  * netvsc_device_add - Callback when the device belonging to this
1359  * driver is added
1360  */
1361 int netvsc_device_add(struct hv_device *device, void *additional_info)
1362 {
1363         int i, ret = 0;
1364         int ring_size =
1365         ((struct netvsc_device_info *)additional_info)->ring_size;
1366         struct netvsc_device *net_device;
1367         struct net_device *ndev = hv_get_drvdata(device);
1368         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1369
1370         net_device = alloc_net_device();
1371         if (!net_device)
1372                 return -ENOMEM;
1373
1374         net_device->ring_size = ring_size;
1375
1376         /* Initialize the NetVSC channel extension */
1377         init_completion(&net_device->channel_init_wait);
1378
1379         set_per_channel_state(device->channel, net_device->cb_buffer);
1380
1381         /* Open the channel */
1382         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1383                          ring_size * PAGE_SIZE, NULL, 0,
1384                          netvsc_channel_cb, device->channel);
1385
1386         if (ret != 0) {
1387                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1388                 goto cleanup;
1389         }
1390
1391         /* Channel is opened */
1392         pr_info("hv_netvsc channel opened successfully\n");
1393
1394         /* If we're reopening the device we may have multiple queues, fill the
1395          * chn_table with the default channel to use it before subchannels are
1396          * opened.
1397          */
1398         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
1399                 net_device->chn_table[i] = device->channel;
1400
1401         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1402          * populated.
1403          */
1404         wmb();
1405
1406         net_device_ctx->nvdev = net_device;
1407
1408         /* Connect with the NetVsp */
1409         ret = netvsc_connect_vsp(device);
1410         if (ret != 0) {
1411                 netdev_err(ndev,
1412                         "unable to connect to NetVSP - %d\n", ret);
1413                 goto close;
1414         }
1415
1416         return ret;
1417
1418 close:
1419         /* Now, we can close the channel safely */
1420         vmbus_close(device->channel);
1421
1422 cleanup:
1423         free_netvsc_device(net_device);
1424
1425         return ret;
1426 }