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 <asm/sync_bitops.h>
  32
  33 #include "hyperv_net.h"
  34
  35
  36 static struct netvsc_device *alloc_net_device(struct hv_device *device)
  37 {
  38         struct netvsc_device *net_device;
  39         struct net_device *ndev = hv_get_drvdata(device);
  40
  41         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  42         if (!net_device)
  43                 return NULL;
  44
  45         net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
  46         if (!net_device->cb_buffer) {
  47                 kfree(net_device);
  48                 return NULL;
  49         }
  50
  51         init_waitqueue_head(&net_device->wait_drain);
  52         net_device->start_remove = false;
  53         net_device->destroy = false;
  54         net_device->dev = device;
  55         net_device->ndev = ndev;
  56
  57         hv_set_drvdata(device, net_device);
  58         return net_device;
  59 }
  60
  61 static void free_netvsc_device(struct netvsc_device *nvdev)
  62 {
  63         kfree(nvdev->cb_buffer);
  64         kfree(nvdev);
  65 }
  66
  67 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  68 {
  69         struct netvsc_device *net_device;
  70
  71         net_device = hv_get_drvdata(device);
  72         if (net_device && net_device->destroy)
  73                 net_device = NULL;
  74
  75         return net_device;
  76 }
  77
  78 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  79 {
  80         struct netvsc_device *net_device;
  81
  82         net_device = hv_get_drvdata(device);
  83
  84         if (!net_device)
  85                 goto get_in_err;
  86
  87         if (net_device->destroy &&
  88                 atomic_read(&net_device->num_outstanding_sends) == 0)
  89                 net_device = NULL;
  90
  91 get_in_err:
  92         return net_device;
  93 }
  94
  95
  96 static int netvsc_destroy_buf(struct netvsc_device *net_device)
  97 {
  98         struct nvsp_message *revoke_packet;
  99         int ret = 0;
 100         struct net_device *ndev = net_device->ndev;
 101
 102         /*
 103          * If we got a section count, it means we received a
 104          * SendReceiveBufferComplete msg (ie sent
 105          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 106          * to send a revoke msg here
 107          */
 108         if (net_device->recv_section_cnt) {
 109                 /* Send the revoke receive buffer */
 110                 revoke_packet = &net_device->revoke_packet;
 111                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 112
 113                 revoke_packet->hdr.msg_type =
 114                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 115                 revoke_packet->msg.v1_msg.
 116                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 117
 118                 ret = vmbus_sendpacket(net_device->dev->channel,
 119                                        revoke_packet,
 120                                        sizeof(struct nvsp_message),
 121                                        (unsigned long)revoke_packet,
 122                                        VM_PKT_DATA_INBAND, 0);
 123                 /*
 124                  * If we failed here, we might as well return and
 125                  * have a leak rather than continue and a bugchk
 126                  */
 127                 if (ret != 0) {
 128                         netdev_err(ndev, "unable to send "
 129                                 "revoke receive buffer to netvsp\n");
 130                         return ret;
 131                 }
 132         }
 133
 134         /* Teardown the gpadl on the vsp end */
 135         if (net_device->recv_buf_gpadl_handle) {
 136                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 137                            net_device->recv_buf_gpadl_handle);
 138
 139                 /* If we failed here, we might as well return and have a leak
 140                  * rather than continue and a bugchk
 141                  */
 142                 if (ret != 0) {
 143                         netdev_err(ndev,
 144                                    "unable to teardown receive buffer's gpadl\n");
 145                         return ret;
 146                 }
 147                 net_device->recv_buf_gpadl_handle = 0;
 148         }
 149
 150         if (net_device->recv_buf) {
 151                 /* Free up the receive buffer */
 152                 vfree(net_device->recv_buf);
 153                 net_device->recv_buf = NULL;
 154         }
 155
 156         if (net_device->recv_section) {
 157                 net_device->recv_section_cnt = 0;
 158                 kfree(net_device->recv_section);
 159                 net_device->recv_section = NULL;
 160         }
 161
 162         /* Deal with the send buffer we may have setup.
 163          * If we got a  send section size, it means we received a
 164          * SendsendBufferComplete msg (ie sent
 165          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 166          * to send a revoke msg here
 167          */
 168         if (net_device->send_section_size) {
 169                 /* Send the revoke receive buffer */
 170                 revoke_packet = &net_device->revoke_packet;
 171                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 172
 173                 revoke_packet->hdr.msg_type =
 174                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 175                 revoke_packet->msg.v1_msg.revoke_recv_buf.id = 0;
 176
 177                 ret = vmbus_sendpacket(net_device->dev->channel,
 178                                        revoke_packet,
 179                                        sizeof(struct nvsp_message),
 180                                        (unsigned long)revoke_packet,
 181                                        VM_PKT_DATA_INBAND, 0);
 182                 /* If we failed here, we might as well return and
 183                  * have a leak rather than continue and a bugchk
 184                  */
 185                 if (ret != 0) {
 186                         netdev_err(ndev, "unable to send "
 187                                    "revoke send buffer to netvsp\n");
 188                         return ret;
 189                 }
 190         }
 191         /* Teardown the gpadl on the vsp end */
 192         if (net_device->send_buf_gpadl_handle) {
 193                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 194                                            net_device->send_buf_gpadl_handle);
 195
 196                 /* If we failed here, we might as well return and have a leak
 197                  * rather than continue and a bugchk
 198                  */
 199                 if (ret != 0) {
 200                         netdev_err(ndev,
 201                                    "unable to teardown send buffer's gpadl\n");
 202                         return ret;
 203                 }
 204                 net_device->send_buf_gpadl_handle = 0;
 205         }
 206         if (net_device->send_buf) {
 207                 /* Free up the receive buffer */
 208                 vfree(net_device->send_buf);
 209                 net_device->send_buf = NULL;
 210         }
 211         kfree(net_device->send_section_map);
 212
 213         return ret;
 214 }
 215
 216 static int netvsc_init_buf(struct hv_device *device)
 217 {
 218         int ret = 0;
 219         int t;
 220         struct netvsc_device *net_device;
 221         struct nvsp_message *init_packet;
 222         struct net_device *ndev;
 223
 224         net_device = get_outbound_net_device(device);
 225         if (!net_device)
 226                 return -ENODEV;
 227         ndev = net_device->ndev;
 228
 229         net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 230         if (!net_device->recv_buf) {
 231                 netdev_err(ndev, "unable to allocate receive "
 232                         "buffer of size %d\n", net_device->recv_buf_size);
 233                 ret = -ENOMEM;
 234                 goto cleanup;
 235         }
 236
 237         /*
 238          * Establish the gpadl handle for this buffer on this
 239          * channel.  Note: This call uses the vmbus connection rather
 240          * than the channel to establish the gpadl handle.
 241          */
 242         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 243                                     net_device->recv_buf_size,
 244                                     &net_device->recv_buf_gpadl_handle);
 245         if (ret != 0) {
 246                 netdev_err(ndev,
 247                         "unable to establish receive buffer's gpadl\n");
 248                 goto cleanup;
 249         }
 250
 251
 252         /* Notify the NetVsp of the gpadl handle */
 253         init_packet = &net_device->channel_init_pkt;
 254
 255         memset(init_packet, 0, sizeof(struct nvsp_message));
 256
 257         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 258         init_packet->msg.v1_msg.send_recv_buf.
 259                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 260         init_packet->msg.v1_msg.
 261                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 262
 263         /* Send the gpadl notification request */
 264         ret = vmbus_sendpacket(device->channel, init_packet,
 265                                sizeof(struct nvsp_message),
 266                                (unsigned long)init_packet,
 267                                VM_PKT_DATA_INBAND,
 268                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 269         if (ret != 0) {
 270                 netdev_err(ndev,
 271                         "unable to send receive buffer's gpadl to netvsp\n");
 272                 goto cleanup;
 273         }
 274
 275         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 276         BUG_ON(t == 0);
 277
 278
 279         /* Check the response */
 280         if (init_packet->msg.v1_msg.
 281             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 282                 netdev_err(ndev, "Unable to complete receive buffer "
 283                            "initialization with NetVsp - status %d\n",
 284                            init_packet->msg.v1_msg.
 285                            send_recv_buf_complete.status);
 286                 ret = -EINVAL;
 287                 goto cleanup;
 288         }
 289
 290         /* Parse the response */
 291
 292         net_device->recv_section_cnt = init_packet->msg.
 293                 v1_msg.send_recv_buf_complete.num_sections;
 294
 295         net_device->recv_section = kmemdup(
 296                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 297                 net_device->recv_section_cnt *
 298                 sizeof(struct nvsp_1_receive_buffer_section),
 299                 GFP_KERNEL);
 300         if (net_device->recv_section == NULL) {
 301                 ret = -EINVAL;
 302                 goto cleanup;
 303         }
 304
 305         /*
 306          * For 1st release, there should only be 1 section that represents the
 307          * entire receive buffer
 308          */
 309         if (net_device->recv_section_cnt != 1 ||
 310             net_device->recv_section->offset != 0) {
 311                 ret = -EINVAL;
 312                 goto cleanup;
 313         }
 314
 315         /* Now setup the send buffer.
 316          */
 317         net_device->send_buf = vzalloc(net_device->send_buf_size);
 318         if (!net_device->send_buf) {
 319                 netdev_err(ndev, "unable to allocate send "
 320                            "buffer of size %d\n", net_device->send_buf_size);
 321                 ret = -ENOMEM;
 322                 goto cleanup;
 323         }
 324
 325         /* Establish the gpadl handle for this buffer on this
 326          * channel.  Note: This call uses the vmbus connection rather
 327          * than the channel to establish the gpadl handle.
 328          */
 329         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 330                                     net_device->send_buf_size,
 331                                     &net_device->send_buf_gpadl_handle);
 332         if (ret != 0) {
 333                 netdev_err(ndev,
 334                            "unable to establish send buffer's gpadl\n");
 335                 goto cleanup;
 336         }
 337
 338         /* Notify the NetVsp of the gpadl handle */
 339         init_packet = &net_device->channel_init_pkt;
 340         memset(init_packet, 0, sizeof(struct nvsp_message));
 341         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 342         init_packet->msg.v1_msg.send_recv_buf.gpadl_handle =
 343                 net_device->send_buf_gpadl_handle;
 344         init_packet->msg.v1_msg.send_recv_buf.id = 0;
 345
 346         /* Send the gpadl notification request */
 347         ret = vmbus_sendpacket(device->channel, init_packet,
 348                                sizeof(struct nvsp_message),
 349                                (unsigned long)init_packet,
 350                                VM_PKT_DATA_INBAND,
 351                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 352         if (ret != 0) {
 353                 netdev_err(ndev,
 354                            "unable to send send buffer's gpadl to netvsp\n");
 355                 goto cleanup;
 356         }
 357
 358         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 359         BUG_ON(t == 0);
 360
 361         /* Check the response */
 362         if (init_packet->msg.v1_msg.
 363             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 364                 netdev_err(ndev, "Unable to complete send buffer "
 365                            "initialization with NetVsp - status %d\n",
 366                            init_packet->msg.v1_msg.
 367                            send_recv_buf_complete.status);
 368                 ret = -EINVAL;
 369                 goto cleanup;
 370         }
 371
 372         /* Parse the response */
 373         net_device->send_section_size = init_packet->msg.
 374                                 v1_msg.send_send_buf_complete.section_size;
 375
 376         /* Section count is simply the size divided by the section size.
 377          */
 378         net_device->send_section_cnt =
 379                 net_device->send_buf_size/net_device->send_section_size;
 380
 381         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 382                  net_device->send_section_size, net_device->send_section_cnt);
 383
 384         /* Setup state for managing the send buffer. */
 385         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 386                                              BITS_PER_LONG);
 387
 388         net_device->send_section_map =
 389                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 390         if (net_device->send_section_map == NULL) {
 391                 ret = -ENOMEM;
 392                 goto cleanup;
 393         }
 394
 395         goto exit;
 396
 397 cleanup:
 398         netvsc_destroy_buf(net_device);
 399
 400 exit:
 401         return ret;
 402 }
 403
 404
 405 /* Negotiate NVSP protocol version */
 406 static int negotiate_nvsp_ver(struct hv_device *device,
 407                               struct netvsc_device *net_device,
 408                               struct nvsp_message *init_packet,
 409                               u32 nvsp_ver)
 410 {
 411         int ret, t;
 412
 413         memset(init_packet, 0, sizeof(struct nvsp_message));
 414         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 415         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 416         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 417
 418         /* Send the init request */
 419         ret = vmbus_sendpacket(device->channel, init_packet,
 420                                sizeof(struct nvsp_message),
 421                                (unsigned long)init_packet,
 422                                VM_PKT_DATA_INBAND,
 423                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 424
 425         if (ret != 0)
 426                 return ret;
 427
 428         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 429
 430         if (t == 0)
 431                 return -ETIMEDOUT;
 432
 433         if (init_packet->msg.init_msg.init_complete.status !=
 434             NVSP_STAT_SUCCESS)
 435                 return -EINVAL;
 436
 437         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 438                 return 0;
 439
 440         /* NVSPv2 only: Send NDIS config */
 441         memset(init_packet, 0, sizeof(struct nvsp_message));
 442         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 443         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
 444                                                        ETH_HLEN;
 445         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 446
 447         ret = vmbus_sendpacket(device->channel, init_packet,
 448                                 sizeof(struct nvsp_message),
 449                                 (unsigned long)init_packet,
 450                                 VM_PKT_DATA_INBAND, 0);
 451
 452         return ret;
 453 }
 454
 455 static int netvsc_connect_vsp(struct hv_device *device)
 456 {
 457         int ret;
 458         struct netvsc_device *net_device;
 459         struct nvsp_message *init_packet;
 460         int ndis_version;
 461         struct net_device *ndev;
 462         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 463                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 464         int i, num_ver = 4; /* number of different NVSP versions */
 465
 466         net_device = get_outbound_net_device(device);
 467         if (!net_device)
 468                 return -ENODEV;
 469         ndev = net_device->ndev;
 470
 471         init_packet = &net_device->channel_init_pkt;
 472
 473         /* Negotiate the latest NVSP protocol supported */
 474         for (i = num_ver - 1; i >= 0; i--)
 475                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 476                                        ver_list[i])  == 0) {
 477                         net_device->nvsp_version = ver_list[i];
 478                         break;
 479                 }
 480
 481         if (i < 0) {
 482                 ret = -EPROTO;
 483                 goto cleanup;
 484         }
 485
 486         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 487
 488         /* Send the ndis version */
 489         memset(init_packet, 0, sizeof(struct nvsp_message));
 490
 491         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 492                 ndis_version = 0x00060001;
 493         else
 494                 ndis_version = 0x0006001e;
 495
 496         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 497         init_packet->msg.v1_msg.
 498                 send_ndis_ver.ndis_major_ver =
 499                                 (ndis_version & 0xFFFF0000) >> 16;
 500         init_packet->msg.v1_msg.
 501                 send_ndis_ver.ndis_minor_ver =
 502                                 ndis_version & 0xFFFF;
 503
 504         /* Send the init request */
 505         ret = vmbus_sendpacket(device->channel, init_packet,
 506                                 sizeof(struct nvsp_message),
 507                                 (unsigned long)init_packet,
 508                                 VM_PKT_DATA_INBAND, 0);
 509         if (ret != 0)
 510                 goto cleanup;
 511
 512         /* Post the big receive buffer to NetVSP */
 513         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 514                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 515         else
 516                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 517         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 518
 519         ret = netvsc_init_buf(device);
 520
 521 cleanup:
 522         return ret;
 523 }
 524
 525 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 526 {
 527         netvsc_destroy_buf(net_device);
 528 }
 529
 530 /*
 531  * netvsc_device_remove - Callback when the root bus device is removed
 532  */
 533 int netvsc_device_remove(struct hv_device *device)
 534 {
 535         struct netvsc_device *net_device;
 536         unsigned long flags;
 537
 538         net_device = hv_get_drvdata(device);
 539
 540         netvsc_disconnect_vsp(net_device);
 541
 542         /*
 543          * Since we have already drained, we don't need to busy wait
 544          * as was done in final_release_stor_device()
 545          * Note that we cannot set the ext pointer to NULL until
 546          * we have drained - to drain the outgoing packets, we need to
 547          * allow incoming packets.
 548          */
 549
 550         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 551         hv_set_drvdata(device, NULL);
 552         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 553
 554         /*
 555          * At this point, no one should be accessing net_device
 556          * except in here
 557          */
 558         dev_notice(&device->device, "net device safe to remove\n");
 559
 560         /* Now, we can close the channel safely */
 561         vmbus_close(device->channel);
 562
 563         /* Release all resources */
 564         vfree(net_device->sub_cb_buf);
 565         free_netvsc_device(net_device);
 566         return 0;
 567 }
 568
 569
 570 #define RING_AVAIL_PERCENT_HIWATER 20
 571 #define RING_AVAIL_PERCENT_LOWATER 10
 572
 573 /*
 574  * Get the percentage of available bytes to write in the ring.
 575  * The return value is in range from 0 to 100.
 576  */
 577 static inline u32 hv_ringbuf_avail_percent(
 578                 struct hv_ring_buffer_info *ring_info)
 579 {
 580         u32 avail_read, avail_write;
 581
 582         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 583
 584         return avail_write * 100 / ring_info->ring_datasize;
 585 }
 586
 587 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 588                                          u32 index)
 589 {
 590         sync_change_bit(index, net_device->send_section_map);
 591 }
 592
 593 static void netvsc_send_completion(struct netvsc_device *net_device,
 594                                    struct hv_device *device,
 595                                    struct vmpacket_descriptor *packet)
 596 {
 597         struct nvsp_message *nvsp_packet;
 598         struct hv_netvsc_packet *nvsc_packet;
 599         struct net_device *ndev;
 600         u32 send_index;
 601
 602         ndev = net_device->ndev;
 603
 604         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 605                         (packet->offset8 << 3));
 606
 607         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 608             (nvsp_packet->hdr.msg_type ==
 609              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 610             (nvsp_packet->hdr.msg_type ==
 611              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
 612             (nvsp_packet->hdr.msg_type ==
 613              NVSP_MSG5_TYPE_SUBCHANNEL)) {
 614                 /* Copy the response back */
 615                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 616                        sizeof(struct nvsp_message));
 617                 complete(&net_device->channel_init_wait);
 618         } else if (nvsp_packet->hdr.msg_type ==
 619                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 620                 int num_outstanding_sends;
 621                 u16 q_idx = 0;
 622                 struct vmbus_channel *channel = device->channel;
 623                 int queue_sends;
 624
 625                 /* Get the send context */
 626                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 627                         packet->trans_id;
 628
 629                 /* Notify the layer above us */
 630                 if (nvsc_packet) {
 631                         send_index = nvsc_packet->send_buf_index;
 632                         if (send_index != NETVSC_INVALID_INDEX)
 633                                 netvsc_free_send_slot(net_device, send_index);
 634                         q_idx = nvsc_packet->q_idx;
 635                         channel = nvsc_packet->channel;
 636                         nvsc_packet->send_completion(nvsc_packet->
 637                                                      send_completion_ctx);
 638                 }
 639
 640                 num_outstanding_sends =
 641                         atomic_dec_return(&net_device->num_outstanding_sends);
 642                 queue_sends = atomic_dec_return(&net_device->
 643                                                 queue_sends[q_idx]);
 644
 645                 if (net_device->destroy && num_outstanding_sends == 0)
 646                         wake_up(&net_device->wait_drain);
 647
 648                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 649                     !net_device->start_remove &&
 650                     (hv_ringbuf_avail_percent(&channel->outbound) >
 651                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 652                                 netif_tx_wake_queue(netdev_get_tx_queue(
 653                                                     ndev, q_idx));
 654         } else {
 655                 netdev_err(ndev, "Unknown send completion packet type- "
 656                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 657         }
 658
 659 }
 660
 661 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 662 {
 663         unsigned long index;
 664         u32 max_words = net_device->map_words;
 665         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 666         u32 section_cnt = net_device->send_section_cnt;
 667         int ret_val = NETVSC_INVALID_INDEX;
 668         int i;
 669         int prev_val;
 670
 671         for (i = 0; i < max_words; i++) {
 672                 if (!~(map_addr[i]))
 673                         continue;
 674                 index = ffz(map_addr[i]);
 675                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 676                 if (prev_val)
 677                         continue;
 678                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 679                         break;
 680                 ret_val = (index + (i * BITS_PER_LONG));
 681                 break;
 682         }
 683         return ret_val;
 684 }
 685
 686 u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 687                             unsigned int section_index,
 688                             struct hv_netvsc_packet *packet)
 689 {
 690         char *start = net_device->send_buf;
 691         char *dest = (start + (section_index * net_device->send_section_size));
 692         int i;
 693         u32 msg_size = 0;
 694
 695         for (i = 0; i < packet->page_buf_cnt; i++) {
 696                 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
 697                 u32 offset = packet->page_buf[i].offset;
 698                 u32 len = packet->page_buf[i].len;
 699
 700                 memcpy(dest, (src + offset), len);
 701                 msg_size += len;
 702                 dest += len;
 703         }
 704         return msg_size;
 705 }
 706
 707 int netvsc_send(struct hv_device *device,
 708                         struct hv_netvsc_packet *packet)
 709 {
 710         struct netvsc_device *net_device;
 711         int ret = 0;
 712         struct nvsp_message sendMessage;
 713         struct net_device *ndev;
 714         struct vmbus_channel *out_channel = NULL;
 715         u64 req_id;
 716         unsigned int section_index = NETVSC_INVALID_INDEX;
 717         u32 msg_size = 0;
 718         struct sk_buff *skb;
 719         u16 q_idx = packet->q_idx;
 720
 721
 722         net_device = get_outbound_net_device(device);
 723         if (!net_device)
 724                 return -ENODEV;
 725         ndev = net_device->ndev;
 726
 727         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 728         if (packet->is_data_pkt) {
 729                 /* 0 is RMC_DATA; */
 730                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 731         } else {
 732                 /* 1 is RMC_CONTROL; */
 733                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 734         }
 735
 736         /* Attempt to send via sendbuf */
 737         if (packet->total_data_buflen < net_device->send_section_size) {
 738                 section_index = netvsc_get_next_send_section(net_device);
 739                 if (section_index != NETVSC_INVALID_INDEX) {
 740                         msg_size = netvsc_copy_to_send_buf(net_device,
 741                                                            section_index,
 742                                                            packet);
 743                         skb = (struct sk_buff *)
 744                               (unsigned long)packet->send_completion_tid;
 745                         if (skb)
 746                                 dev_kfree_skb_any(skb);
 747                         packet->page_buf_cnt = 0;
 748                 }
 749         }
 750         packet->send_buf_index = section_index;
 751
 752
 753         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 754                 section_index;
 755         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = msg_size;
 756
 757         if (packet->send_completion)
 758                 req_id = (ulong)packet;
 759         else
 760                 req_id = 0;
 761
 762         out_channel = net_device->chn_table[packet->q_idx];
 763         if (out_channel == NULL)
 764                 out_channel = device->channel;
 765         packet->channel = out_channel;
 766
 767         if (out_channel->rescind)
 768                 return -ENODEV;
 769
 770         if (packet->page_buf_cnt) {
 771                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 772                                                   packet->page_buf,
 773                                                   packet->page_buf_cnt,
 774                                                   &sendMessage,
 775                                                   sizeof(struct nvsp_message),
 776                                                   req_id);
 777         } else {
 778                 ret = vmbus_sendpacket(out_channel, &sendMessage,
 779                                 sizeof(struct nvsp_message),
 780                                 req_id,
 781                                 VM_PKT_DATA_INBAND,
 782                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 783         }
 784
 785         if (ret == 0) {
 786                 atomic_inc(&net_device->num_outstanding_sends);
 787                 atomic_inc(&net_device->queue_sends[q_idx]);
 788
 789                 if (hv_ringbuf_avail_percent(&out_channel->outbound) <
 790                         RING_AVAIL_PERCENT_LOWATER) {
 791                         netif_tx_stop_queue(netdev_get_tx_queue(
 792                                             ndev, q_idx));
 793
 794                         if (atomic_read(&net_device->
 795                                 queue_sends[q_idx]) < 1)
 796                                 netif_tx_wake_queue(netdev_get_tx_queue(
 797                                                     ndev, q_idx));
 798                 }
 799         } else if (ret == -EAGAIN) {
 800                 netif_tx_stop_queue(netdev_get_tx_queue(
 801                                     ndev, q_idx));
 802                 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
 803                         netif_tx_wake_queue(netdev_get_tx_queue(
 804                                             ndev, q_idx));
 805                         ret = -ENOSPC;
 806                 }
 807         } else {
 808                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 809                            packet, ret);
 810         }
 811
 812         return ret;
 813 }
 814
 815 static void netvsc_send_recv_completion(struct hv_device *device,
 816                                         struct vmbus_channel *channel,
 817                                         struct netvsc_device *net_device,
 818                                         u64 transaction_id, u32 status)
 819 {
 820         struct nvsp_message recvcompMessage;
 821         int retries = 0;
 822         int ret;
 823         struct net_device *ndev;
 824
 825         ndev = net_device->ndev;
 826
 827         recvcompMessage.hdr.msg_type =
 828                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 829
 830         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 831
 832 retry_send_cmplt:
 833         /* Send the completion */
 834         ret = vmbus_sendpacket(channel, &recvcompMessage,
 835                                sizeof(struct nvsp_message), transaction_id,
 836                                VM_PKT_COMP, 0);
 837         if (ret == 0) {
 838                 /* success */
 839                 /* no-op */
 840         } else if (ret == -EAGAIN) {
 841                 /* no more room...wait a bit and attempt to retry 3 times */
 842                 retries++;
 843                 netdev_err(ndev, "unable to send receive completion pkt"
 844                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 845
 846                 if (retries < 4) {
 847                         udelay(100);
 848                         goto retry_send_cmplt;
 849                 } else {
 850                         netdev_err(ndev, "unable to send receive "
 851                                 "completion pkt (tid %llx)...give up retrying\n",
 852                                 transaction_id);
 853                 }
 854         } else {
 855                 netdev_err(ndev, "unable to send receive "
 856                         "completion pkt - %llx\n", transaction_id);
 857         }
 858 }
 859
 860 static void netvsc_receive(struct netvsc_device *net_device,
 861                         struct vmbus_channel *channel,
 862                         struct hv_device *device,
 863                         struct vmpacket_descriptor *packet)
 864 {
 865         struct vmtransfer_page_packet_header *vmxferpage_packet;
 866         struct nvsp_message *nvsp_packet;
 867         struct hv_netvsc_packet nv_pkt;
 868         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
 869         u32 status = NVSP_STAT_SUCCESS;
 870         int i;
 871         int count = 0;
 872         struct net_device *ndev;
 873
 874         ndev = net_device->ndev;
 875
 876         /*
 877          * All inbound packets other than send completion should be xfer page
 878          * packet
 879          */
 880         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 881                 netdev_err(ndev, "Unknown packet type received - %d\n",
 882                            packet->type);
 883                 return;
 884         }
 885
 886         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 887                         (packet->offset8 << 3));
 888
 889         /* Make sure this is a valid nvsp packet */
 890         if (nvsp_packet->hdr.msg_type !=
 891             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 892                 netdev_err(ndev, "Unknown nvsp packet type received-"
 893                         " %d\n", nvsp_packet->hdr.msg_type);
 894                 return;
 895         }
 896
 897         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 898
 899         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 900                 netdev_err(ndev, "Invalid xfer page set id - "
 901                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
 902                            vmxferpage_packet->xfer_pageset_id);
 903                 return;
 904         }
 905
 906         count = vmxferpage_packet->range_cnt;
 907         netvsc_packet->device = device;
 908         netvsc_packet->channel = channel;
 909
 910         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 911         for (i = 0; i < count; i++) {
 912                 /* Initialize the netvsc packet */
 913                 netvsc_packet->status = NVSP_STAT_SUCCESS;
 914                 netvsc_packet->data = (void *)((unsigned long)net_device->
 915                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
 916                 netvsc_packet->total_data_buflen =
 917                                         vmxferpage_packet->ranges[i].byte_count;
 918
 919                 /* Pass it to the upper layer */
 920                 rndis_filter_receive(device, netvsc_packet);
 921
 922                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
 923                         status = NVSP_STAT_FAIL;
 924         }
 925
 926         netvsc_send_recv_completion(device, channel, net_device,
 927                                     vmxferpage_packet->d.trans_id, status);
 928 }
 929
 930
 931 static void netvsc_send_table(struct hv_device *hdev,
 932                               struct vmpacket_descriptor *vmpkt)
 933 {
 934         struct netvsc_device *nvscdev;
 935         struct net_device *ndev;
 936         struct nvsp_message *nvmsg;
 937         int i;
 938         u32 count, *tab;
 939
 940         nvscdev = get_outbound_net_device(hdev);
 941         if (!nvscdev)
 942                 return;
 943         ndev = nvscdev->ndev;
 944
 945         nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
 946                                         (vmpkt->offset8 << 3));
 947
 948         if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
 949                 return;
 950
 951         count = nvmsg->msg.v5_msg.send_table.count;
 952         if (count != VRSS_SEND_TAB_SIZE) {
 953                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
 954                 return;
 955         }
 956
 957         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
 958                       nvmsg->msg.v5_msg.send_table.offset);
 959
 960         for (i = 0; i < count; i++)
 961                 nvscdev->send_table[i] = tab[i];
 962 }
 963
 964 void netvsc_channel_cb(void *context)
 965 {
 966         int ret;
 967         struct vmbus_channel *channel = (struct vmbus_channel *)context;
 968         struct hv_device *device;
 969         struct netvsc_device *net_device;
 970         u32 bytes_recvd;
 971         u64 request_id;
 972         struct vmpacket_descriptor *desc;
 973         unsigned char *buffer;
 974         int bufferlen = NETVSC_PACKET_SIZE;
 975         struct net_device *ndev;
 976
 977         if (channel->primary_channel != NULL)
 978                 device = channel->primary_channel->device_obj;
 979         else
 980                 device = channel->device_obj;
 981
 982         net_device = get_inbound_net_device(device);
 983         if (!net_device)
 984                 return;
 985         ndev = net_device->ndev;
 986         buffer = get_per_channel_state(channel);
 987
 988         do {
 989                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
 990                                            &bytes_recvd, &request_id);
 991                 if (ret == 0) {
 992                         if (bytes_recvd > 0) {
 993                                 desc = (struct vmpacket_descriptor *)buffer;
 994                                 switch (desc->type) {
 995                                 case VM_PKT_COMP:
 996                                         netvsc_send_completion(net_device,
 997                                                                 device, desc);
 998                                         break;
 999
1000                                 case VM_PKT_DATA_USING_XFER_PAGES:
1001                                         netvsc_receive(net_device, channel,
1002                                                        device, desc);
1003                                         break;
1004
1005                                 case VM_PKT_DATA_INBAND:
1006                                         netvsc_send_table(device, desc);
1007                                         break;
1008
1009                                 default:
1010                                         netdev_err(ndev,
1011                                                    "unhandled packet type %d, "
1012                                                    "tid %llx len %d\n",
1013                                                    desc->type, request_id,
1014                                                    bytes_recvd);
1015                                         break;
1016                                 }
1017
1018                         } else {
1019                                 /*
1020                                  * We are done for this pass.
1021                                  */
1022                                 break;
1023                         }
1024
1025                 } else if (ret == -ENOBUFS) {
1026                         if (bufferlen > NETVSC_PACKET_SIZE)
1027                                 kfree(buffer);
1028                         /* Handle large packet */
1029                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1030                         if (buffer == NULL) {
1031                                 /* Try again next time around */
1032                                 netdev_err(ndev,
1033                                            "unable to allocate buffer of size "
1034                                            "(%d)!!\n", bytes_recvd);
1035                                 break;
1036                         }
1037
1038                         bufferlen = bytes_recvd;
1039                 }
1040         } while (1);
1041
1042         if (bufferlen > NETVSC_PACKET_SIZE)
1043                 kfree(buffer);
1044         return;
1045 }
1046
1047 /*
1048  * netvsc_device_add - Callback when the device belonging to this
1049  * driver is added
1050  */
1051 int netvsc_device_add(struct hv_device *device, void *additional_info)
1052 {
1053         int ret = 0;
1054         int ring_size =
1055         ((struct netvsc_device_info *)additional_info)->ring_size;
1056         struct netvsc_device *net_device;
1057         struct net_device *ndev;
1058
1059         net_device = alloc_net_device(device);
1060         if (!net_device)
1061                 return -ENOMEM;
1062
1063         net_device->ring_size = ring_size;
1064
1065         /*
1066          * Coming into this function, struct net_device * is
1067          * registered as the driver private data.
1068          * In alloc_net_device(), we register struct netvsc_device *
1069          * as the driver private data and stash away struct net_device *
1070          * in struct netvsc_device *.
1071          */
1072         ndev = net_device->ndev;
1073
1074         /* Initialize the NetVSC channel extension */
1075         init_completion(&net_device->channel_init_wait);
1076
1077         set_per_channel_state(device->channel, net_device->cb_buffer);
1078
1079         /* Open the channel */
1080         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1081                          ring_size * PAGE_SIZE, NULL, 0,
1082                          netvsc_channel_cb, device->channel);
1083
1084         if (ret != 0) {
1085                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1086                 goto cleanup;
1087         }
1088
1089         /* Channel is opened */
1090         pr_info("hv_netvsc channel opened successfully\n");
1091
1092         net_device->chn_table[0] = device->channel;
1093
1094         /* Connect with the NetVsp */
1095         ret = netvsc_connect_vsp(device);
1096         if (ret != 0) {
1097                 netdev_err(ndev,
1098                         "unable to connect to NetVSP - %d\n", ret);
1099                 goto close;
1100         }
1101
1102         return ret;
1103
1104 close:
1105         /* Now, we can close the channel safely */
1106         vmbus_close(device->channel);
1107
1108 cleanup:
1109         free_netvsc_device(net_device);
1110
1111         return ret;
1112 }