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          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 165          * NVSP_MSG1_TYPE_SEND_SEND_BUF 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_send_buf.id =
 176                         NETVSC_SEND_BUFFER_ID;
 177
 178                 ret = vmbus_sendpacket(net_device->dev->channel,
 179                                        revoke_packet,
 180                                        sizeof(struct nvsp_message),
 181                                        (unsigned long)revoke_packet,
 182                                        VM_PKT_DATA_INBAND, 0);
 183                 /* If we failed here, we might as well return and
 184                  * have a leak rather than continue and a bugchk
 185                  */
 186                 if (ret != 0) {
 187                         netdev_err(ndev, "unable to send "
 188                                    "revoke send buffer to netvsp\n");
 189                         return ret;
 190                 }
 191         }
 192         /* Teardown the gpadl on the vsp end */
 193         if (net_device->send_buf_gpadl_handle) {
 194                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 195                                            net_device->send_buf_gpadl_handle);
 196
 197                 /* If we failed here, we might as well return and have a leak
 198                  * rather than continue and a bugchk
 199                  */
 200                 if (ret != 0) {
 201                         netdev_err(ndev,
 202                                    "unable to teardown send buffer's gpadl\n");
 203                         return ret;
 204                 }
 205                 net_device->send_buf_gpadl_handle = 0;
 206         }
 207         if (net_device->send_buf) {
 208                 /* Free up the send buffer */
 209                 vfree(net_device->send_buf);
 210                 net_device->send_buf = NULL;
 211         }
 212         kfree(net_device->send_section_map);
 213
 214         return ret;
 215 }
 216
 217 static int netvsc_init_buf(struct hv_device *device)
 218 {
 219         int ret = 0;
 220         int t;
 221         struct netvsc_device *net_device;
 222         struct nvsp_message *init_packet;
 223         struct net_device *ndev;
 224
 225         net_device = get_outbound_net_device(device);
 226         if (!net_device)
 227                 return -ENODEV;
 228         ndev = net_device->ndev;
 229
 230         net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 231         if (!net_device->recv_buf) {
 232                 netdev_err(ndev, "unable to allocate receive "
 233                         "buffer of size %d\n", net_device->recv_buf_size);
 234                 ret = -ENOMEM;
 235                 goto cleanup;
 236         }
 237
 238         /*
 239          * Establish the gpadl handle for this buffer on this
 240          * channel.  Note: This call uses the vmbus connection rather
 241          * than the channel to establish the gpadl handle.
 242          */
 243         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 244                                     net_device->recv_buf_size,
 245                                     &net_device->recv_buf_gpadl_handle);
 246         if (ret != 0) {
 247                 netdev_err(ndev,
 248                         "unable to establish receive buffer's gpadl\n");
 249                 goto cleanup;
 250         }
 251
 252
 253         /* Notify the NetVsp of the gpadl handle */
 254         init_packet = &net_device->channel_init_pkt;
 255
 256         memset(init_packet, 0, sizeof(struct nvsp_message));
 257
 258         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 259         init_packet->msg.v1_msg.send_recv_buf.
 260                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 261         init_packet->msg.v1_msg.
 262                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 263
 264         /* Send the gpadl notification request */
 265         ret = vmbus_sendpacket(device->channel, init_packet,
 266                                sizeof(struct nvsp_message),
 267                                (unsigned long)init_packet,
 268                                VM_PKT_DATA_INBAND,
 269                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 270         if (ret != 0) {
 271                 netdev_err(ndev,
 272                         "unable to send receive buffer's gpadl to netvsp\n");
 273                 goto cleanup;
 274         }
 275
 276         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 277         BUG_ON(t == 0);
 278
 279
 280         /* Check the response */
 281         if (init_packet->msg.v1_msg.
 282             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 283                 netdev_err(ndev, "Unable to complete receive buffer "
 284                            "initialization with NetVsp - status %d\n",
 285                            init_packet->msg.v1_msg.
 286                            send_recv_buf_complete.status);
 287                 ret = -EINVAL;
 288                 goto cleanup;
 289         }
 290
 291         /* Parse the response */
 292
 293         net_device->recv_section_cnt = init_packet->msg.
 294                 v1_msg.send_recv_buf_complete.num_sections;
 295
 296         net_device->recv_section = kmemdup(
 297                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 298                 net_device->recv_section_cnt *
 299                 sizeof(struct nvsp_1_receive_buffer_section),
 300                 GFP_KERNEL);
 301         if (net_device->recv_section == NULL) {
 302                 ret = -EINVAL;
 303                 goto cleanup;
 304         }
 305
 306         /*
 307          * For 1st release, there should only be 1 section that represents the
 308          * entire receive buffer
 309          */
 310         if (net_device->recv_section_cnt != 1 ||
 311             net_device->recv_section->offset != 0) {
 312                 ret = -EINVAL;
 313                 goto cleanup;
 314         }
 315
 316         /* Now setup the send buffer.
 317          */
 318         net_device->send_buf = vzalloc(net_device->send_buf_size);
 319         if (!net_device->send_buf) {
 320                 netdev_err(ndev, "unable to allocate send "
 321                            "buffer of size %d\n", net_device->send_buf_size);
 322                 ret = -ENOMEM;
 323                 goto cleanup;
 324         }
 325
 326         /* Establish the gpadl handle for this buffer on this
 327          * channel.  Note: This call uses the vmbus connection rather
 328          * than the channel to establish the gpadl handle.
 329          */
 330         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 331                                     net_device->send_buf_size,
 332                                     &net_device->send_buf_gpadl_handle);
 333         if (ret != 0) {
 334                 netdev_err(ndev,
 335                            "unable to establish send buffer's gpadl\n");
 336                 goto cleanup;
 337         }
 338
 339         /* Notify the NetVsp of the gpadl handle */
 340         init_packet = &net_device->channel_init_pkt;
 341         memset(init_packet, 0, sizeof(struct nvsp_message));
 342         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 343         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 344                 net_device->send_buf_gpadl_handle;
 345         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 346
 347         /* Send the gpadl notification request */
 348         ret = vmbus_sendpacket(device->channel, init_packet,
 349                                sizeof(struct nvsp_message),
 350                                (unsigned long)init_packet,
 351                                VM_PKT_DATA_INBAND,
 352                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 353         if (ret != 0) {
 354                 netdev_err(ndev,
 355                            "unable to send send buffer's gpadl to netvsp\n");
 356                 goto cleanup;
 357         }
 358
 359         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 360         BUG_ON(t == 0);
 361
 362         /* Check the response */
 363         if (init_packet->msg.v1_msg.
 364             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 365                 netdev_err(ndev, "Unable to complete send buffer "
 366                            "initialization with NetVsp - status %d\n",
 367                            init_packet->msg.v1_msg.
 368                            send_send_buf_complete.status);
 369                 ret = -EINVAL;
 370                 goto cleanup;
 371         }
 372
 373         /* Parse the response */
 374         net_device->send_section_size = init_packet->msg.
 375                                 v1_msg.send_send_buf_complete.section_size;
 376
 377         /* Section count is simply the size divided by the section size.
 378          */
 379         net_device->send_section_cnt =
 380                 net_device->send_buf_size/net_device->send_section_size;
 381
 382         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 383                  net_device->send_section_size, net_device->send_section_cnt);
 384
 385         /* Setup state for managing the send buffer. */
 386         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 387                                              BITS_PER_LONG);
 388
 389         net_device->send_section_map =
 390                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 391         if (net_device->send_section_map == NULL) {
 392                 ret = -ENOMEM;
 393                 goto cleanup;
 394         }
 395
 396         goto exit;
 397
 398 cleanup:
 399         netvsc_destroy_buf(net_device);
 400
 401 exit:
 402         return ret;
 403 }
 404
 405
 406 /* Negotiate NVSP protocol version */
 407 static int negotiate_nvsp_ver(struct hv_device *device,
 408                               struct netvsc_device *net_device,
 409                               struct nvsp_message *init_packet,
 410                               u32 nvsp_ver)
 411 {
 412         int ret, t;
 413
 414         memset(init_packet, 0, sizeof(struct nvsp_message));
 415         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 416         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 417         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 418
 419         /* Send the init request */
 420         ret = vmbus_sendpacket(device->channel, init_packet,
 421                                sizeof(struct nvsp_message),
 422                                (unsigned long)init_packet,
 423                                VM_PKT_DATA_INBAND,
 424                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 425
 426         if (ret != 0)
 427                 return ret;
 428
 429         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 430
 431         if (t == 0)
 432                 return -ETIMEDOUT;
 433
 434         if (init_packet->msg.init_msg.init_complete.status !=
 435             NVSP_STAT_SUCCESS)
 436                 return -EINVAL;
 437
 438         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 439                 return 0;
 440
 441         /* NVSPv2 only: Send NDIS config */
 442         memset(init_packet, 0, sizeof(struct nvsp_message));
 443         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 444         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
 445                                                        ETH_HLEN;
 446         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 447
 448         ret = vmbus_sendpacket(device->channel, init_packet,
 449                                 sizeof(struct nvsp_message),
 450                                 (unsigned long)init_packet,
 451                                 VM_PKT_DATA_INBAND, 0);
 452
 453         return ret;
 454 }
 455
 456 static int netvsc_connect_vsp(struct hv_device *device)
 457 {
 458         int ret;
 459         struct netvsc_device *net_device;
 460         struct nvsp_message *init_packet;
 461         int ndis_version;
 462         struct net_device *ndev;
 463         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 464                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 465         int i, num_ver = 4; /* number of different NVSP versions */
 466
 467         net_device = get_outbound_net_device(device);
 468         if (!net_device)
 469                 return -ENODEV;
 470         ndev = net_device->ndev;
 471
 472         init_packet = &net_device->channel_init_pkt;
 473
 474         /* Negotiate the latest NVSP protocol supported */
 475         for (i = num_ver - 1; i >= 0; i--)
 476                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 477                                        ver_list[i])  == 0) {
 478                         net_device->nvsp_version = ver_list[i];
 479                         break;
 480                 }
 481
 482         if (i < 0) {
 483                 ret = -EPROTO;
 484                 goto cleanup;
 485         }
 486
 487         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 488
 489         /* Send the ndis version */
 490         memset(init_packet, 0, sizeof(struct nvsp_message));
 491
 492         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 493                 ndis_version = 0x00060001;
 494         else
 495                 ndis_version = 0x0006001e;
 496
 497         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 498         init_packet->msg.v1_msg.
 499                 send_ndis_ver.ndis_major_ver =
 500                                 (ndis_version & 0xFFFF0000) >> 16;
 501         init_packet->msg.v1_msg.
 502                 send_ndis_ver.ndis_minor_ver =
 503                                 ndis_version & 0xFFFF;
 504
 505         /* Send the init request */
 506         ret = vmbus_sendpacket(device->channel, init_packet,
 507                                 sizeof(struct nvsp_message),
 508                                 (unsigned long)init_packet,
 509                                 VM_PKT_DATA_INBAND, 0);
 510         if (ret != 0)
 511                 goto cleanup;
 512
 513         /* Post the big receive buffer to NetVSP */
 514         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 515                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 516         else
 517                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 518         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 519
 520         ret = netvsc_init_buf(device);
 521
 522 cleanup:
 523         return ret;
 524 }
 525
 526 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 527 {
 528         netvsc_destroy_buf(net_device);
 529 }
 530
 531 /*
 532  * netvsc_device_remove - Callback when the root bus device is removed
 533  */
 534 int netvsc_device_remove(struct hv_device *device)
 535 {
 536         struct netvsc_device *net_device;
 537         unsigned long flags;
 538
 539         net_device = hv_get_drvdata(device);
 540
 541         netvsc_disconnect_vsp(net_device);
 542
 543         /*
 544          * Since we have already drained, we don't need to busy wait
 545          * as was done in final_release_stor_device()
 546          * Note that we cannot set the ext pointer to NULL until
 547          * we have drained - to drain the outgoing packets, we need to
 548          * allow incoming packets.
 549          */
 550
 551         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 552         hv_set_drvdata(device, NULL);
 553         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 554
 555         /*
 556          * At this point, no one should be accessing net_device
 557          * except in here
 558          */
 559         dev_notice(&device->device, "net device safe to remove\n");
 560
 561         /* Now, we can close the channel safely */
 562         vmbus_close(device->channel);
 563
 564         /* Release all resources */
 565         vfree(net_device->sub_cb_buf);
 566         free_netvsc_device(net_device);
 567         return 0;
 568 }
 569
 570
 571 #define RING_AVAIL_PERCENT_HIWATER 20
 572 #define RING_AVAIL_PERCENT_LOWATER 10
 573
 574 /*
 575  * Get the percentage of available bytes to write in the ring.
 576  * The return value is in range from 0 to 100.
 577  */
 578 static inline u32 hv_ringbuf_avail_percent(
 579                 struct hv_ring_buffer_info *ring_info)
 580 {
 581         u32 avail_read, avail_write;
 582
 583         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 584
 585         return avail_write * 100 / ring_info->ring_datasize;
 586 }
 587
 588 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 589                                          u32 index)
 590 {
 591         sync_change_bit(index, net_device->send_section_map);
 592 }
 593
 594 static void netvsc_send_completion(struct netvsc_device *net_device,
 595                                    struct hv_device *device,
 596                                    struct vmpacket_descriptor *packet)
 597 {
 598         struct nvsp_message *nvsp_packet;
 599         struct hv_netvsc_packet *nvsc_packet;
 600         struct net_device *ndev;
 601         u32 send_index;
 602
 603         ndev = net_device->ndev;
 604
 605         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 606                         (packet->offset8 << 3));
 607
 608         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 609             (nvsp_packet->hdr.msg_type ==
 610              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 611             (nvsp_packet->hdr.msg_type ==
 612              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
 613             (nvsp_packet->hdr.msg_type ==
 614              NVSP_MSG5_TYPE_SUBCHANNEL)) {
 615                 /* Copy the response back */
 616                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 617                        sizeof(struct nvsp_message));
 618                 complete(&net_device->channel_init_wait);
 619         } else if (nvsp_packet->hdr.msg_type ==
 620                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 621                 int num_outstanding_sends;
 622                 u16 q_idx = 0;
 623                 struct vmbus_channel *channel = device->channel;
 624                 int queue_sends;
 625
 626                 /* Get the send context */
 627                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 628                         packet->trans_id;
 629
 630                 /* Notify the layer above us */
 631                 if (nvsc_packet) {
 632                         send_index = nvsc_packet->send_buf_index;
 633                         if (send_index != NETVSC_INVALID_INDEX)
 634                                 netvsc_free_send_slot(net_device, send_index);
 635                         q_idx = nvsc_packet->q_idx;
 636                         channel = nvsc_packet->channel;
 637                         nvsc_packet->send_completion(nvsc_packet->
 638                                                      send_completion_ctx);
 639                 }
 640
 641                 num_outstanding_sends =
 642                         atomic_dec_return(&net_device->num_outstanding_sends);
 643                 queue_sends = atomic_dec_return(&net_device->
 644                                                 queue_sends[q_idx]);
 645
 646                 if (net_device->destroy && num_outstanding_sends == 0)
 647                         wake_up(&net_device->wait_drain);
 648
 649                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 650                     !net_device->start_remove &&
 651                     (hv_ringbuf_avail_percent(&channel->outbound) >
 652                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 653                                 netif_tx_wake_queue(netdev_get_tx_queue(
 654                                                     ndev, q_idx));
 655         } else {
 656                 netdev_err(ndev, "Unknown send completion packet type- "
 657                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 658         }
 659
 660 }
 661
 662 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 663 {
 664         unsigned long index;
 665         u32 max_words = net_device->map_words;
 666         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 667         u32 section_cnt = net_device->send_section_cnt;
 668         int ret_val = NETVSC_INVALID_INDEX;
 669         int i;
 670         int prev_val;
 671
 672         for (i = 0; i < max_words; i++) {
 673                 if (!~(map_addr[i]))
 674                         continue;
 675                 index = ffz(map_addr[i]);
 676                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 677                 if (prev_val)
 678                         continue;
 679                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 680                         break;
 681                 ret_val = (index + (i * BITS_PER_LONG));
 682                 break;
 683         }
 684         return ret_val;
 685 }
 686
 687 u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 688                             unsigned int section_index,
 689                             struct hv_netvsc_packet *packet)
 690 {
 691         char *start = net_device->send_buf;
 692         char *dest = (start + (section_index * net_device->send_section_size));
 693         int i;
 694         u32 msg_size = 0;
 695
 696         for (i = 0; i < packet->page_buf_cnt; i++) {
 697                 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
 698                 u32 offset = packet->page_buf[i].offset;
 699                 u32 len = packet->page_buf[i].len;
 700
 701                 memcpy(dest, (src + offset), len);
 702                 msg_size += len;
 703                 dest += len;
 704         }
 705         return msg_size;
 706 }
 707
 708 int netvsc_send(struct hv_device *device,
 709                         struct hv_netvsc_packet *packet)
 710 {
 711         struct netvsc_device *net_device;
 712         int ret = 0;
 713         struct nvsp_message sendMessage;
 714         struct net_device *ndev;
 715         struct vmbus_channel *out_channel = NULL;
 716         u64 req_id;
 717         unsigned int section_index = NETVSC_INVALID_INDEX;
 718         u32 msg_size = 0;
 719         struct sk_buff *skb = NULL;
 720         u16 q_idx = packet->q_idx;
 721
 722
 723         net_device = get_outbound_net_device(device);
 724         if (!net_device)
 725                 return -ENODEV;
 726         ndev = net_device->ndev;
 727
 728         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 729         if (packet->is_data_pkt) {
 730                 /* 0 is RMC_DATA; */
 731                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 732         } else {
 733                 /* 1 is RMC_CONTROL; */
 734                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 735         }
 736
 737         /* Attempt to send via sendbuf */
 738         if (packet->total_data_buflen < net_device->send_section_size) {
 739                 section_index = netvsc_get_next_send_section(net_device);
 740                 if (section_index != NETVSC_INVALID_INDEX) {
 741                         msg_size = netvsc_copy_to_send_buf(net_device,
 742                                                            section_index,
 743                                                            packet);
 744                         skb = (struct sk_buff *)
 745                               (unsigned long)packet->send_completion_tid;
 746                         packet->page_buf_cnt = 0;
 747                 }
 748         }
 749         packet->send_buf_index = section_index;
 750
 751
 752         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 753                 section_index;
 754         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = msg_size;
 755
 756         if (packet->send_completion)
 757                 req_id = (ulong)packet;
 758         else
 759                 req_id = 0;
 760
 761         out_channel = net_device->chn_table[packet->q_idx];
 762         if (out_channel == NULL)
 763                 out_channel = device->channel;
 764         packet->channel = out_channel;
 765
 766         if (out_channel->rescind)
 767                 return -ENODEV;
 768
 769         if (packet->page_buf_cnt) {
 770                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 771                                                   packet->page_buf,
 772                                                   packet->page_buf_cnt,
 773                                                   &sendMessage,
 774                                                   sizeof(struct nvsp_message),
 775                                                   req_id);
 776         } else {
 777                 ret = vmbus_sendpacket(out_channel, &sendMessage,
 778                                 sizeof(struct nvsp_message),
 779                                 req_id,
 780                                 VM_PKT_DATA_INBAND,
 781                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 782         }
 783
 784         if (ret == 0) {
 785                 atomic_inc(&net_device->num_outstanding_sends);
 786                 atomic_inc(&net_device->queue_sends[q_idx]);
 787
 788                 if (hv_ringbuf_avail_percent(&out_channel->outbound) <
 789                         RING_AVAIL_PERCENT_LOWATER) {
 790                         netif_tx_stop_queue(netdev_get_tx_queue(
 791                                             ndev, q_idx));
 792
 793                         if (atomic_read(&net_device->
 794                                 queue_sends[q_idx]) < 1)
 795                                 netif_tx_wake_queue(netdev_get_tx_queue(
 796                                                     ndev, q_idx));
 797                 }
 798         } else if (ret == -EAGAIN) {
 799                 netif_tx_stop_queue(netdev_get_tx_queue(
 800                                     ndev, q_idx));
 801                 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
 802                         netif_tx_wake_queue(netdev_get_tx_queue(
 803                                             ndev, q_idx));
 804                         ret = -ENOSPC;
 805                 }
 806         } else {
 807                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 808                            packet, ret);
 809         }
 810
 811         if (ret != 0) {
 812                 if (section_index != NETVSC_INVALID_INDEX)
 813                         netvsc_free_send_slot(net_device, section_index);
 814         } else if (skb) {
 815                 dev_kfree_skb_any(skb);
 816         }
 817
 818         return ret;
 819 }
 820
 821 static void netvsc_send_recv_completion(struct hv_device *device,
 822                                         struct vmbus_channel *channel,
 823                                         struct netvsc_device *net_device,
 824                                         u64 transaction_id, u32 status)
 825 {
 826         struct nvsp_message recvcompMessage;
 827         int retries = 0;
 828         int ret;
 829         struct net_device *ndev;
 830
 831         ndev = net_device->ndev;
 832
 833         recvcompMessage.hdr.msg_type =
 834                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 835
 836         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 837
 838 retry_send_cmplt:
 839         /* Send the completion */
 840         ret = vmbus_sendpacket(channel, &recvcompMessage,
 841                                sizeof(struct nvsp_message), transaction_id,
 842                                VM_PKT_COMP, 0);
 843         if (ret == 0) {
 844                 /* success */
 845                 /* no-op */
 846         } else if (ret == -EAGAIN) {
 847                 /* no more room...wait a bit and attempt to retry 3 times */
 848                 retries++;
 849                 netdev_err(ndev, "unable to send receive completion pkt"
 850                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 851
 852                 if (retries < 4) {
 853                         udelay(100);
 854                         goto retry_send_cmplt;
 855                 } else {
 856                         netdev_err(ndev, "unable to send receive "
 857                                 "completion pkt (tid %llx)...give up retrying\n",
 858                                 transaction_id);
 859                 }
 860         } else {
 861                 netdev_err(ndev, "unable to send receive "
 862                         "completion pkt - %llx\n", transaction_id);
 863         }
 864 }
 865
 866 static void netvsc_receive(struct netvsc_device *net_device,
 867                         struct vmbus_channel *channel,
 868                         struct hv_device *device,
 869                         struct vmpacket_descriptor *packet)
 870 {
 871         struct vmtransfer_page_packet_header *vmxferpage_packet;
 872         struct nvsp_message *nvsp_packet;
 873         struct hv_netvsc_packet nv_pkt;
 874         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
 875         u32 status = NVSP_STAT_SUCCESS;
 876         int i;
 877         int count = 0;
 878         struct net_device *ndev;
 879
 880         ndev = net_device->ndev;
 881
 882         /*
 883          * All inbound packets other than send completion should be xfer page
 884          * packet
 885          */
 886         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 887                 netdev_err(ndev, "Unknown packet type received - %d\n",
 888                            packet->type);
 889                 return;
 890         }
 891
 892         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 893                         (packet->offset8 << 3));
 894
 895         /* Make sure this is a valid nvsp packet */
 896         if (nvsp_packet->hdr.msg_type !=
 897             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 898                 netdev_err(ndev, "Unknown nvsp packet type received-"
 899                         " %d\n", nvsp_packet->hdr.msg_type);
 900                 return;
 901         }
 902
 903         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 904
 905         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 906                 netdev_err(ndev, "Invalid xfer page set id - "
 907                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
 908                            vmxferpage_packet->xfer_pageset_id);
 909                 return;
 910         }
 911
 912         count = vmxferpage_packet->range_cnt;
 913         netvsc_packet->device = device;
 914         netvsc_packet->channel = channel;
 915
 916         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 917         for (i = 0; i < count; i++) {
 918                 /* Initialize the netvsc packet */
 919                 netvsc_packet->status = NVSP_STAT_SUCCESS;
 920                 netvsc_packet->data = (void *)((unsigned long)net_device->
 921                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
 922                 netvsc_packet->total_data_buflen =
 923                                         vmxferpage_packet->ranges[i].byte_count;
 924
 925                 /* Pass it to the upper layer */
 926                 rndis_filter_receive(device, netvsc_packet);
 927
 928                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
 929                         status = NVSP_STAT_FAIL;
 930         }
 931
 932         netvsc_send_recv_completion(device, channel, net_device,
 933                                     vmxferpage_packet->d.trans_id, status);
 934 }
 935
 936
 937 static void netvsc_send_table(struct hv_device *hdev,
 938                               struct vmpacket_descriptor *vmpkt)
 939 {
 940         struct netvsc_device *nvscdev;
 941         struct net_device *ndev;
 942         struct nvsp_message *nvmsg;
 943         int i;
 944         u32 count, *tab;
 945
 946         nvscdev = get_outbound_net_device(hdev);
 947         if (!nvscdev)
 948                 return;
 949         ndev = nvscdev->ndev;
 950
 951         nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
 952                                         (vmpkt->offset8 << 3));
 953
 954         if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
 955                 return;
 956
 957         count = nvmsg->msg.v5_msg.send_table.count;
 958         if (count != VRSS_SEND_TAB_SIZE) {
 959                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
 960                 return;
 961         }
 962
 963         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
 964                       nvmsg->msg.v5_msg.send_table.offset);
 965
 966         for (i = 0; i < count; i++)
 967                 nvscdev->send_table[i] = tab[i];
 968 }
 969
 970 void netvsc_channel_cb(void *context)
 971 {
 972         int ret;
 973         struct vmbus_channel *channel = (struct vmbus_channel *)context;
 974         struct hv_device *device;
 975         struct netvsc_device *net_device;
 976         u32 bytes_recvd;
 977         u64 request_id;
 978         struct vmpacket_descriptor *desc;
 979         unsigned char *buffer;
 980         int bufferlen = NETVSC_PACKET_SIZE;
 981         struct net_device *ndev;
 982
 983         if (channel->primary_channel != NULL)
 984                 device = channel->primary_channel->device_obj;
 985         else
 986                 device = channel->device_obj;
 987
 988         net_device = get_inbound_net_device(device);
 989         if (!net_device)
 990                 return;
 991         ndev = net_device->ndev;
 992         buffer = get_per_channel_state(channel);
 993
 994         do {
 995                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
 996                                            &bytes_recvd, &request_id);
 997                 if (ret == 0) {
 998                         if (bytes_recvd > 0) {
 999                                 desc = (struct vmpacket_descriptor *)buffer;
1000                                 switch (desc->type) {
1001                                 case VM_PKT_COMP:
1002                                         netvsc_send_completion(net_device,
1003                                                                 device, desc);
1004                                         break;
1005
1006                                 case VM_PKT_DATA_USING_XFER_PAGES:
1007                                         netvsc_receive(net_device, channel,
1008                                                        device, desc);
1009                                         break;
1010
1011                                 case VM_PKT_DATA_INBAND:
1012                                         netvsc_send_table(device, desc);
1013                                         break;
1014
1015                                 default:
1016                                         netdev_err(ndev,
1017                                                    "unhandled packet type %d, "
1018                                                    "tid %llx len %d\n",
1019                                                    desc->type, request_id,
1020                                                    bytes_recvd);
1021                                         break;
1022                                 }
1023
1024                         } else {
1025                                 /*
1026                                  * We are done for this pass.
1027                                  */
1028                                 break;
1029                         }
1030
1031                 } else if (ret == -ENOBUFS) {
1032                         if (bufferlen > NETVSC_PACKET_SIZE)
1033                                 kfree(buffer);
1034                         /* Handle large packet */
1035                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1036                         if (buffer == NULL) {
1037                                 /* Try again next time around */
1038                                 netdev_err(ndev,
1039                                            "unable to allocate buffer of size "
1040                                            "(%d)!!\n", bytes_recvd);
1041                                 break;
1042                         }
1043
1044                         bufferlen = bytes_recvd;
1045                 }
1046         } while (1);
1047
1048         if (bufferlen > NETVSC_PACKET_SIZE)
1049                 kfree(buffer);
1050         return;
1051 }
1052
1053 /*
1054  * netvsc_device_add - Callback when the device belonging to this
1055  * driver is added
1056  */
1057 int netvsc_device_add(struct hv_device *device, void *additional_info)
1058 {
1059         int ret = 0;
1060         int ring_size =
1061         ((struct netvsc_device_info *)additional_info)->ring_size;
1062         struct netvsc_device *net_device;
1063         struct net_device *ndev;
1064
1065         net_device = alloc_net_device(device);
1066         if (!net_device)
1067                 return -ENOMEM;
1068
1069         net_device->ring_size = ring_size;
1070
1071         /*
1072          * Coming into this function, struct net_device * is
1073          * registered as the driver private data.
1074          * In alloc_net_device(), we register struct netvsc_device *
1075          * as the driver private data and stash away struct net_device *
1076          * in struct netvsc_device *.
1077          */
1078         ndev = net_device->ndev;
1079
1080         /* Initialize the NetVSC channel extension */
1081         init_completion(&net_device->channel_init_wait);
1082
1083         set_per_channel_state(device->channel, net_device->cb_buffer);
1084
1085         /* Open the channel */
1086         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1087                          ring_size * PAGE_SIZE, NULL, 0,
1088                          netvsc_channel_cb, device->channel);
1089
1090         if (ret != 0) {
1091                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1092                 goto cleanup;
1093         }
1094
1095         /* Channel is opened */
1096         pr_info("hv_netvsc channel opened successfully\n");
1097
1098         net_device->chn_table[0] = device->channel;
1099
1100         /* Connect with the NetVsp */
1101         ret = netvsc_connect_vsp(device);
1102         if (ret != 0) {
1103                 netdev_err(ndev,
1104                         "unable to connect to NetVSP - %d\n", ret);
1105                 goto close;
1106         }
1107
1108         return ret;
1109
1110 close:
1111         /* Now, we can close the channel safely */
1112         vmbus_close(device->channel);
1113
1114 cleanup:
1115         free_netvsc_device(net_device);
1116
1117         return ret;
1118 }