drivers/staging/hv/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, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/kernel.h>
  24 #include <linux/sched.h>
  25 #include <linux/wait.h>
  26 #include <linux/mm.h>
  27 #include <linux/delay.h>
  28 #include <linux/io.h>
  29 #include <linux/slab.h>
  30 #include "hv_api.h"
  31 #include "logging.h"
  32 #include "netvsc.h"
  33 #include "rndis_filter.h"
  34 #include "channel.h"
  35
  36
  37 /* Globals */
  38 static const char *driver_name = "netvsc";
  39
  40 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
  41 static const struct hv_guid netvsc_device_type = {
  42         .data = {
  43                 0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
  44                 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E
  45         }
  46 };
  47
  48
  49 static struct netvsc_device *alloc_net_device(struct hv_device *device)
  50 {
  51         struct netvsc_device *net_device;
  52
  53         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  54         if (!net_device)
  55                 return NULL;
  56
  57         /* Set to 2 to allow both inbound and outbound traffic */
  58         atomic_cmpxchg(&net_device->refcnt, 0, 2);
  59
  60         net_device->dev = device;
  61         device->ext = net_device;
  62
  63         return net_device;
  64 }
  65
  66 static void free_net_device(struct netvsc_device *device)
  67 {
  68         WARN_ON(atomic_read(&device->refcnt) != 0);
  69         device->dev->ext = NULL;
  70         kfree(device);
  71 }
  72
  73
  74 /* Get the net device object iff exists and its refcount > 1 */
  75 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  76 {
  77         struct netvsc_device *net_device;
  78
  79         net_device = device->ext;
  80         if (net_device && atomic_read(&net_device->refcnt) > 1)
  81                 atomic_inc(&net_device->refcnt);
  82         else
  83                 net_device = NULL;
  84
  85         return net_device;
  86 }
  87
  88 /* Get the net device object iff exists and its refcount > 0 */
  89 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  90 {
  91         struct netvsc_device *net_device;
  92
  93         net_device = device->ext;
  94         if (net_device && atomic_read(&net_device->refcnt))
  95                 atomic_inc(&net_device->refcnt);
  96         else
  97                 net_device = NULL;
  98
  99         return net_device;
 100 }
 101
 102 static void put_net_device(struct hv_device *device)
 103 {
 104         struct netvsc_device *net_device;
 105
 106         net_device = device->ext;
 107
 108         atomic_dec(&net_device->refcnt);
 109 }
 110
 111 static struct netvsc_device *release_outbound_net_device(
 112                 struct hv_device *device)
 113 {
 114         struct netvsc_device *net_device;
 115
 116         net_device = device->ext;
 117         if (net_device == NULL)
 118                 return NULL;
 119
 120         /* Busy wait until the ref drop to 2, then set it to 1 */
 121         while (atomic_cmpxchg(&net_device->refcnt, 2, 1) != 2)
 122                 udelay(100);
 123
 124         return net_device;
 125 }
 126
 127 static struct netvsc_device *release_inbound_net_device(
 128                 struct hv_device *device)
 129 {
 130         struct netvsc_device *net_device;
 131
 132         net_device = device->ext;
 133         if (net_device == NULL)
 134                 return NULL;
 135
 136         /* Busy wait until the ref drop to 1, then set it to 0 */
 137         while (atomic_cmpxchg(&net_device->refcnt, 1, 0) != 1)
 138                 udelay(100);
 139
 140         device->ext = NULL;
 141         return net_device;
 142 }
 143
 144 static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
 145 {
 146         struct nvsp_message *revoke_packet;
 147         int ret = 0;
 148
 149         /*
 150          * If we got a section count, it means we received a
 151          * SendReceiveBufferComplete msg (ie sent
 152          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 153          * to send a revoke msg here
 154          */
 155         if (net_device->recv_section_cnt) {
 156                 /* Send the revoke receive buffer */
 157                 revoke_packet = &net_device->revoke_packet;
 158                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 159
 160                 revoke_packet->hdr.msg_type =
 161                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 162                 revoke_packet->msg.v1_msg.
 163                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 164
 165                 ret = vmbus_sendpacket(net_device->dev->channel,
 166                                        revoke_packet,
 167                                        sizeof(struct nvsp_message),
 168                                        (unsigned long)revoke_packet,
 169                                        VM_PKT_DATA_INBAND, 0);
 170                 /*
 171                  * If we failed here, we might as well return and
 172                  * have a leak rather than continue and a bugchk
 173                  */
 174                 if (ret != 0) {
 175                         dev_err(&net_device->dev->device, "unable to send "
 176                                 "revoke receive buffer to netvsp");
 177                         return -1;
 178                 }
 179         }
 180
 181         /* Teardown the gpadl on the vsp end */
 182         if (net_device->recv_buf_gpadl_handle) {
 183                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 184                            net_device->recv_buf_gpadl_handle);
 185
 186                 /* If we failed here, we might as well return and have a leak
 187                  * rather than continue and a bugchk
 188                  */
 189                 if (ret != 0) {
 190                         dev_err(&net_device->dev->device,
 191                                    "unable to teardown receive buffer's gpadl");
 192                         return -1;
 193                 }
 194                 net_device->recv_buf_gpadl_handle = 0;
 195         }
 196
 197         if (net_device->recv_buf) {
 198                 /* Free up the receive buffer */
 199                 free_pages((unsigned long)net_device->recv_buf,
 200                         get_order(net_device->recv_buf_size));
 201                 net_device->recv_buf = NULL;
 202         }
 203
 204         if (net_device->recv_section) {
 205                 net_device->recv_section_cnt = 0;
 206                 kfree(net_device->recv_section);
 207                 net_device->recv_section = NULL;
 208         }
 209
 210         return ret;
 211 }
 212
 213 static int netvsc_init_recv_buf(struct hv_device *device)
 214 {
 215         int ret = 0;
 216         struct netvsc_device *net_device;
 217         struct nvsp_message *init_packet;
 218
 219         net_device = get_outbound_net_device(device);
 220         if (!net_device) {
 221                 dev_err(&device->device, "unable to get net device..."
 222                            "device being destroyed?");
 223                 return -1;
 224         }
 225
 226         net_device->recv_buf =
 227                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
 228                                 get_order(net_device->recv_buf_size));
 229         if (!net_device->recv_buf) {
 230                 dev_err(&device->device, "unable to allocate receive "
 231                         "buffer of size %d", net_device->recv_buf_size);
 232                 ret = -1;
 233                 goto cleanup;
 234         }
 235
 236         /*
 237          * Establish the gpadl handle for this buffer on this
 238          * channel.  Note: This call uses the vmbus connection rather
 239          * than the channel to establish the gpadl handle.
 240          */
 241         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 242                                     net_device->recv_buf_size,
 243                                     &net_device->recv_buf_gpadl_handle);
 244         if (ret != 0) {
 245                 dev_err(&device->device,
 246                         "unable to establish receive buffer's gpadl");
 247                 goto cleanup;
 248         }
 249
 250
 251         /* Notify the NetVsp of the gpadl handle */
 252         init_packet = &net_device->channel_init_pkt;
 253
 254         memset(init_packet, 0, sizeof(struct nvsp_message));
 255
 256         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 257         init_packet->msg.v1_msg.send_recv_buf.
 258                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 259         init_packet->msg.v1_msg.
 260                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 261
 262         /* Send the gpadl notification request */
 263         net_device->wait_condition = 0;
 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                 dev_err(&device->device,
 271                         "unable to send receive buffer's gpadl to netvsp");
 272                 goto cleanup;
 273         }
 274
 275         wait_event_timeout(net_device->channel_init_wait,
 276                         net_device->wait_condition,
 277                         msecs_to_jiffies(1000));
 278         BUG_ON(net_device->wait_condition == 0);
 279
 280
 281         /* Check the response */
 282         if (init_packet->msg.v1_msg.
 283             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 284                 dev_err(&device->device, "Unable to complete receive buffer "
 285                            "initialzation with NetVsp - status %d",
 286                            init_packet->msg.v1_msg.
 287                            send_recv_buf_complete.status);
 288                 ret = -1;
 289                 goto cleanup;
 290         }
 291
 292         /* Parse the response */
 293
 294         net_device->recv_section_cnt = init_packet->msg.
 295                 v1_msg.send_recv_buf_complete.num_sections;
 296
 297         net_device->recv_section = kmalloc(net_device->recv_section_cnt
 298                 * sizeof(struct nvsp_1_receive_buffer_section), GFP_KERNEL);
 299         if (net_device->recv_section == NULL) {
 300                 ret = -1;
 301                 goto cleanup;
 302         }
 303
 304         memcpy(net_device->recv_section,
 305                 init_packet->msg.v1_msg.
 306                send_recv_buf_complete.sections,
 307                 net_device->recv_section_cnt *
 308                sizeof(struct nvsp_1_receive_buffer_section));
 309
 310         /*
 311          * For 1st release, there should only be 1 section that represents the
 312          * entire receive buffer
 313          */
 314         if (net_device->recv_section_cnt != 1 ||
 315             net_device->recv_section->offset != 0) {
 316                 ret = -1;
 317                 goto cleanup;
 318         }
 319
 320         goto exit;
 321
 322 cleanup:
 323         netvsc_destroy_recv_buf(net_device);
 324
 325 exit:
 326         put_net_device(device);
 327         return ret;
 328 }
 329
 330 static int netvsc_destroy_send_buf(struct netvsc_device *net_device)
 331 {
 332         struct nvsp_message *revoke_packet;
 333         int ret = 0;
 334
 335         /*
 336          * If we got a section count, it means we received a
 337          *  SendReceiveBufferComplete msg (ie sent
 338          *  NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 339          *  to send a revoke msg here
 340          */
 341         if (net_device->send_section_size) {
 342                 /* Send the revoke send buffer */
 343                 revoke_packet = &net_device->revoke_packet;
 344                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 345
 346                 revoke_packet->hdr.msg_type =
 347                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 348                 revoke_packet->msg.v1_msg.
 349                         revoke_send_buf.id = NETVSC_SEND_BUFFER_ID;
 350
 351                 ret = vmbus_sendpacket(net_device->dev->channel,
 352                                        revoke_packet,
 353                                        sizeof(struct nvsp_message),
 354                                        (unsigned long)revoke_packet,
 355                                        VM_PKT_DATA_INBAND, 0);
 356                 /*
 357                  * If we failed here, we might as well return and have a leak
 358                  * rather than continue and a bugchk
 359                  */
 360                 if (ret != 0) {
 361                         dev_err(&net_device->dev->device, "unable to send "
 362                                 "revoke send buffer to netvsp");
 363                         return -1;
 364                 }
 365         }
 366
 367         /* Teardown the gpadl on the vsp end */
 368         if (net_device->send_buf_gpadl_handle) {
 369                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 370                                            net_device->send_buf_gpadl_handle);
 371
 372                 /*
 373                  * If we failed here, we might as well return and have a leak
 374                  * rather than continue and a bugchk
 375                  */
 376                 if (ret != 0) {
 377                         dev_err(&net_device->dev->device,
 378                                 "unable to teardown send buffer's gpadl");
 379                         return -1;
 380                 }
 381                 net_device->send_buf_gpadl_handle = 0;
 382         }
 383
 384         if (net_device->send_buf) {
 385                 /* Free up the receive buffer */
 386                 free_pages((unsigned long)net_device->send_buf,
 387                                 get_order(net_device->send_buf_size));
 388                 net_device->send_buf = NULL;
 389         }
 390
 391         return ret;
 392 }
 393
 394 static int netvsc_init_send_buf(struct hv_device *device)
 395 {
 396         int ret = 0;
 397         struct netvsc_device *net_device;
 398         struct nvsp_message *init_packet;
 399
 400         net_device = get_outbound_net_device(device);
 401         if (!net_device) {
 402                 dev_err(&device->device, "unable to get net device..."
 403                            "device being destroyed?");
 404                 return -1;
 405         }
 406         if (net_device->send_buf_size <= 0) {
 407                 ret = -EINVAL;
 408                 goto cleanup;
 409         }
 410
 411         net_device->send_buf =
 412                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
 413                                 get_order(net_device->send_buf_size));
 414         if (!net_device->send_buf) {
 415                 dev_err(&device->device, "unable to allocate send "
 416                         "buffer of size %d", net_device->send_buf_size);
 417                 ret = -1;
 418                 goto cleanup;
 419         }
 420
 421         /*
 422          * Establish the gpadl handle for this buffer on this
 423          * channel.  Note: This call uses the vmbus connection rather
 424          * than the channel to establish the gpadl handle.
 425          */
 426         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 427                                     net_device->send_buf_size,
 428                                     &net_device->send_buf_gpadl_handle);
 429         if (ret != 0) {
 430                 dev_err(&device->device, "unable to establish send buffer's gpadl");
 431                 goto cleanup;
 432         }
 433
 434         /* Notify the NetVsp of the gpadl handle */
 435         init_packet = &net_device->channel_init_pkt;
 436
 437         memset(init_packet, 0, sizeof(struct nvsp_message));
 438
 439         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 440         init_packet->msg.v1_msg.send_recv_buf.
 441                 gpadl_handle = net_device->send_buf_gpadl_handle;
 442         init_packet->msg.v1_msg.send_recv_buf.id =
 443                 NETVSC_SEND_BUFFER_ID;
 444
 445         /* Send the gpadl notification request */
 446         net_device->wait_condition = 0;
 447         ret = vmbus_sendpacket(device->channel, init_packet,
 448                                sizeof(struct nvsp_message),
 449                                (unsigned long)init_packet,
 450                                VM_PKT_DATA_INBAND,
 451                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 452         if (ret != 0) {
 453                 dev_err(&device->device,
 454                            "unable to send receive buffer's gpadl to netvsp");
 455                 goto cleanup;
 456         }
 457
 458         wait_event_timeout(net_device->channel_init_wait,
 459                         net_device->wait_condition,
 460                         msecs_to_jiffies(1000));
 461         BUG_ON(net_device->wait_condition == 0);
 462
 463         /* Check the response */
 464         if (init_packet->msg.v1_msg.
 465             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 466                 dev_err(&device->device, "Unable to complete send buffer "
 467                            "initialzation with NetVsp - status %d",
 468                            init_packet->msg.v1_msg.
 469                            send_send_buf_complete.status);
 470                 ret = -1;
 471                 goto cleanup;
 472         }
 473
 474         net_device->send_section_size = init_packet->
 475         msg.v1_msg.send_send_buf_complete.section_size;
 476
 477         goto exit;
 478
 479 cleanup:
 480         netvsc_destroy_send_buf(net_device);
 481
 482 exit:
 483         put_net_device(device);
 484         return ret;
 485 }
 486
 487
 488 static int netvsc_connect_vsp(struct hv_device *device)
 489 {
 490         int ret;
 491         struct netvsc_device *net_device;
 492         struct nvsp_message *init_packet;
 493         int ndis_version;
 494
 495         net_device = get_outbound_net_device(device);
 496         if (!net_device) {
 497                 dev_err(&device->device, "unable to get net device..."
 498                            "device being destroyed?");
 499                 return -1;
 500         }
 501
 502         init_packet = &net_device->channel_init_pkt;
 503
 504         memset(init_packet, 0, sizeof(struct nvsp_message));
 505         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 506         init_packet->msg.init_msg.init.min_protocol_ver =
 507                 NVSP_MIN_PROTOCOL_VERSION;
 508         init_packet->msg.init_msg.init.max_protocol_ver =
 509                 NVSP_MAX_PROTOCOL_VERSION;
 510
 511         /* Send the init request */
 512         net_device->wait_condition = 0;
 513         ret = vmbus_sendpacket(device->channel, init_packet,
 514                                sizeof(struct nvsp_message),
 515                                (unsigned long)init_packet,
 516                                VM_PKT_DATA_INBAND,
 517                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 518
 519         if (ret != 0)
 520                 goto cleanup;
 521
 522         wait_event_timeout(net_device->channel_init_wait,
 523                         net_device->wait_condition,
 524                         msecs_to_jiffies(1000));
 525         if (net_device->wait_condition == 0) {
 526                 ret = -ETIMEDOUT;
 527                 goto cleanup;
 528         }
 529
 530         if (init_packet->msg.init_msg.init_complete.status !=
 531             NVSP_STAT_SUCCESS) {
 532                 ret = -1;
 533                 goto cleanup;
 534         }
 535
 536         if (init_packet->msg.init_msg.init_complete.
 537             negotiated_protocol_ver != NVSP_PROTOCOL_VERSION_1) {
 538                 ret = -1;
 539                 goto cleanup;
 540         }
 541         /* Send the ndis version */
 542         memset(init_packet, 0, sizeof(struct nvsp_message));
 543
 544         ndis_version = 0x00050000;
 545
 546         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 547         init_packet->msg.v1_msg.
 548                 send_ndis_ver.ndis_major_ver =
 549                                 (ndis_version & 0xFFFF0000) >> 16;
 550         init_packet->msg.v1_msg.
 551                 send_ndis_ver.ndis_minor_ver =
 552                                 ndis_version & 0xFFFF;
 553
 554         /* Send the init request */
 555         ret = vmbus_sendpacket(device->channel, init_packet,
 556                                 sizeof(struct nvsp_message),
 557                                 (unsigned long)init_packet,
 558                                 VM_PKT_DATA_INBAND, 0);
 559         if (ret != 0) {
 560                 ret = -1;
 561                 goto cleanup;
 562         }
 563
 564         /* Post the big receive buffer to NetVSP */
 565         ret = netvsc_init_recv_buf(device);
 566         if (ret == 0)
 567                 ret = netvsc_init_send_buf(device);
 568
 569 cleanup:
 570         put_net_device(device);
 571         return ret;
 572 }
 573
 574 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 575 {
 576         netvsc_destroy_recv_buf(net_device);
 577         netvsc_destroy_send_buf(net_device);
 578 }
 579
 580 /*
 581  * netvsc_device_remove - Callback when the root bus device is removed
 582  */
 583 static int netvsc_device_remove(struct hv_device *device)
 584 {
 585         struct netvsc_device *net_device;
 586         struct hv_netvsc_packet *netvsc_packet, *pos;
 587
 588         /* Stop outbound traffic ie sends and receives completions */
 589         net_device = release_outbound_net_device(device);
 590         if (!net_device) {
 591                 dev_err(&device->device, "No net device present!!");
 592                 return -1;
 593         }
 594
 595         /* Wait for all send completions */
 596         while (atomic_read(&net_device->num_outstanding_sends)) {
 597                 dev_err(&device->device,
 598                         "waiting for %d requests to complete...",
 599                         atomic_read(&net_device->num_outstanding_sends));
 600                 udelay(100);
 601         }
 602
 603         netvsc_disconnect_vsp(net_device);
 604
 605         /* Stop inbound traffic ie receives and sends completions */
 606         net_device = release_inbound_net_device(device);
 607
 608         /* At this point, no one should be accessing netDevice except in here */
 609         dev_notice(&device->device, "net device safe to remove");
 610
 611         /* Now, we can close the channel safely */
 612         vmbus_close(device->channel);
 613
 614         /* Release all resources */
 615         list_for_each_entry_safe(netvsc_packet, pos,
 616                                  &net_device->recv_pkt_list, list_ent) {
 617                 list_del(&netvsc_packet->list_ent);
 618                 kfree(netvsc_packet);
 619         }
 620
 621         free_net_device(net_device);
 622         return 0;
 623 }
 624
 625 /*
 626  * netvsc_cleanup - Perform any cleanup when the driver is removed
 627  */
 628 static void netvsc_cleanup(struct hv_driver *drv)
 629 {
 630 }
 631
 632 static void netvsc_send_completion(struct hv_device *device,
 633                                    struct vmpacket_descriptor *packet)
 634 {
 635         struct netvsc_device *net_device;
 636         struct nvsp_message *nvsp_packet;
 637         struct hv_netvsc_packet *nvsc_packet;
 638
 639         net_device = get_inbound_net_device(device);
 640         if (!net_device) {
 641                 dev_err(&device->device, "unable to get net device..."
 642                            "device being destroyed?");
 643                 return;
 644         }
 645
 646         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 647                         (packet->offset8 << 3));
 648
 649         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 650             (nvsp_packet->hdr.msg_type ==
 651              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 652             (nvsp_packet->hdr.msg_type ==
 653              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
 654                 /* Copy the response back */
 655                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 656                        sizeof(struct nvsp_message));
 657                 net_device->wait_condition = 1;
 658                 wake_up(&net_device->channel_init_wait);
 659         } else if (nvsp_packet->hdr.msg_type ==
 660                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 661                 /* Get the send context */
 662                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 663                         packet->trans_id;
 664
 665                 /* Notify the layer above us */
 666                 nvsc_packet->completion.send.send_completion(
 667                         nvsc_packet->completion.send.send_completion_ctx);
 668
 669                 atomic_dec(&net_device->num_outstanding_sends);
 670         } else {
 671                 dev_err(&device->device, "Unknown send completion packet type- "
 672                            "%d received!!", nvsp_packet->hdr.msg_type);
 673         }
 674
 675         put_net_device(device);
 676 }
 677
 678 static int netvsc_send(struct hv_device *device,
 679                         struct hv_netvsc_packet *packet)
 680 {
 681         struct netvsc_device *net_device;
 682         int ret = 0;
 683
 684         struct nvsp_message sendMessage;
 685
 686         net_device = get_outbound_net_device(device);
 687         if (!net_device) {
 688                 dev_err(&device->device, "net device (%p) shutting down..."
 689                            "ignoring outbound packets", net_device);
 690                 return -2;
 691         }
 692
 693         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 694         if (packet->is_data_pkt) {
 695                 /* 0 is RMC_DATA; */
 696                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 697         } else {
 698                 /* 1 is RMC_CONTROL; */
 699                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 700         }
 701
 702         /* Not using send buffer section */
 703         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 704                 0xFFFFFFFF;
 705         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 706
 707         if (packet->page_buf_cnt) {
 708                 ret = vmbus_sendpacket_pagebuffer(device->channel,
 709                                                   packet->page_buf,
 710                                                   packet->page_buf_cnt,
 711                                                   &sendMessage,
 712                                                   sizeof(struct nvsp_message),
 713                                                   (unsigned long)packet);
 714         } else {
 715                 ret = vmbus_sendpacket(device->channel, &sendMessage,
 716                                        sizeof(struct nvsp_message),
 717                                        (unsigned long)packet,
 718                                        VM_PKT_DATA_INBAND,
 719                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 720
 721         }
 722
 723         if (ret != 0)
 724                 dev_err(&device->device, "Unable to send packet %p ret %d",
 725                            packet, ret);
 726
 727         atomic_inc(&net_device->num_outstanding_sends);
 728         put_net_device(device);
 729         return ret;
 730 }
 731
 732 static void netvsc_send_recv_completion(struct hv_device *device,
 733                                         u64 transaction_id)
 734 {
 735         struct nvsp_message recvcompMessage;
 736         int retries = 0;
 737         int ret;
 738
 739         recvcompMessage.hdr.msg_type =
 740                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 741
 742         /* FIXME: Pass in the status */
 743         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
 744                 NVSP_STAT_SUCCESS;
 745
 746 retry_send_cmplt:
 747         /* Send the completion */
 748         ret = vmbus_sendpacket(device->channel, &recvcompMessage,
 749                                sizeof(struct nvsp_message), transaction_id,
 750                                VM_PKT_COMP, 0);
 751         if (ret == 0) {
 752                 /* success */
 753                 /* no-op */
 754         } else if (ret == -1) {
 755                 /* no more room...wait a bit and attempt to retry 3 times */
 756                 retries++;
 757                 dev_err(&device->device, "unable to send receive completion pkt"
 758                         " (tid %llx)...retrying %d", transaction_id, retries);
 759
 760                 if (retries < 4) {
 761                         udelay(100);
 762                         goto retry_send_cmplt;
 763                 } else {
 764                         dev_err(&device->device, "unable to send receive "
 765                                 "completion pkt (tid %llx)...give up retrying",
 766                                 transaction_id);
 767                 }
 768         } else {
 769                 dev_err(&device->device, "unable to send receive "
 770                         "completion pkt - %llx", transaction_id);
 771         }
 772 }
 773
 774 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
 775 static void netvsc_receive_completion(void *context)
 776 {
 777         struct hv_netvsc_packet *packet = context;
 778         struct hv_device *device = (struct hv_device *)packet->device;
 779         struct netvsc_device *net_device;
 780         u64 transaction_id = 0;
 781         bool fsend_receive_comp = false;
 782         unsigned long flags;
 783
 784         /*
 785          * Even though it seems logical to do a GetOutboundNetDevice() here to
 786          * send out receive completion, we are using GetInboundNetDevice()
 787          * since we may have disable outbound traffic already.
 788          */
 789         net_device = get_inbound_net_device(device);
 790         if (!net_device) {
 791                 dev_err(&device->device, "unable to get net device..."
 792                            "device being destroyed?");
 793                 return;
 794         }
 795
 796         /* Overloading use of the lock. */
 797         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 798
 799         packet->xfer_page_pkt->count--;
 800
 801         /*
 802          * Last one in the line that represent 1 xfer page packet.
 803          * Return the xfer page packet itself to the freelist
 804          */
 805         if (packet->xfer_page_pkt->count == 0) {
 806                 fsend_receive_comp = true;
 807                 transaction_id = packet->completion.recv.recv_completion_tid;
 808                 list_add_tail(&packet->xfer_page_pkt->list_ent,
 809                               &net_device->recv_pkt_list);
 810
 811         }
 812
 813         /* Put the packet back */
 814         list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
 815         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
 816
 817         /* Send a receive completion for the xfer page packet */
 818         if (fsend_receive_comp)
 819                 netvsc_send_recv_completion(device, transaction_id);
 820
 821         put_net_device(device);
 822 }
 823
 824 static void netvsc_receive(struct hv_device *device,
 825                             struct vmpacket_descriptor *packet)
 826 {
 827         struct netvsc_device *net_device;
 828         struct vmtransfer_page_packet_header *vmxferpage_packet;
 829         struct nvsp_message *nvsp_packet;
 830         struct hv_netvsc_packet *netvsc_packet = NULL;
 831         unsigned long start;
 832         unsigned long end, end_virtual;
 833         /* struct netvsc_driver *netvscDriver; */
 834         struct xferpage_packet *xferpage_packet = NULL;
 835         int i, j;
 836         int count = 0, bytes_remain = 0;
 837         unsigned long flags;
 838         struct netvsc_driver *netvsc_drv =
 839                  drv_to_netvscdrv(device->device.driver);
 840
 841         LIST_HEAD(listHead);
 842
 843         net_device = get_inbound_net_device(device);
 844         if (!net_device) {
 845                 dev_err(&device->device, "unable to get net device..."
 846                            "device being destroyed?");
 847                 return;
 848         }
 849
 850         /*
 851          * All inbound packets other than send completion should be xfer page
 852          * packet
 853          */
 854         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 855                 dev_err(&device->device, "Unknown packet type received - %d",
 856                            packet->type);
 857                 put_net_device(device);
 858                 return;
 859         }
 860
 861         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 862                         (packet->offset8 << 3));
 863
 864         /* Make sure this is a valid nvsp packet */
 865         if (nvsp_packet->hdr.msg_type !=
 866             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 867                 dev_err(&device->device, "Unknown nvsp packet type received-"
 868                         " %d", nvsp_packet->hdr.msg_type);
 869                 put_net_device(device);
 870                 return;
 871         }
 872
 873         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 874
 875         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 876                 dev_err(&device->device, "Invalid xfer page set id - "
 877                            "expecting %x got %x", NETVSC_RECEIVE_BUFFER_ID,
 878                            vmxferpage_packet->xfer_pageset_id);
 879                 put_net_device(device);
 880                 return;
 881         }
 882
 883         /*
 884          * Grab free packets (range count + 1) to represent this xfer
 885          * page packet. +1 to represent the xfer page packet itself.
 886          * We grab it here so that we know exactly how many we can
 887          * fulfil
 888          */
 889         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 890         while (!list_empty(&net_device->recv_pkt_list)) {
 891                 list_move_tail(net_device->recv_pkt_list.next, &listHead);
 892                 if (++count == vmxferpage_packet->range_cnt + 1)
 893                         break;
 894         }
 895         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
 896
 897         /*
 898          * We need at least 2 netvsc pkts (1 to represent the xfer
 899          * page and at least 1 for the range) i.e. we can handled
 900          * some of the xfer page packet ranges...
 901          */
 902         if (count < 2) {
 903                 dev_err(&device->device, "Got only %d netvsc pkt...needed "
 904                         "%d pkts. Dropping this xfer page packet completely!",
 905                         count, vmxferpage_packet->range_cnt + 1);
 906
 907                 /* Return it to the freelist */
 908                 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 909                 for (i = count; i != 0; i--) {
 910                         list_move_tail(listHead.next,
 911                                        &net_device->recv_pkt_list);
 912                 }
 913                 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
 914                                        flags);
 915
 916                 netvsc_send_recv_completion(device,
 917                                             vmxferpage_packet->d.trans_id);
 918
 919                 put_net_device(device);
 920                 return;
 921         }
 922
 923         /* Remove the 1st packet to represent the xfer page packet itself */
 924         xferpage_packet = (struct xferpage_packet *)listHead.next;
 925         list_del(&xferpage_packet->list_ent);
 926
 927         /* This is how much we can satisfy */
 928         xferpage_packet->count = count - 1;
 929
 930         if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
 931                 dev_err(&device->device, "Needed %d netvsc pkts to satisy "
 932                         "this xfer page...got %d",
 933                         vmxferpage_packet->range_cnt, xferpage_packet->count);
 934         }
 935
 936         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 937         for (i = 0; i < (count - 1); i++) {
 938                 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
 939                 list_del(&netvsc_packet->list_ent);
 940
 941                 /* Initialize the netvsc packet */
 942                 netvsc_packet->xfer_page_pkt = xferpage_packet;
 943                 netvsc_packet->completion.recv.recv_completion =
 944                                         netvsc_receive_completion;
 945                 netvsc_packet->completion.recv.recv_completion_ctx =
 946                                         netvsc_packet;
 947                 netvsc_packet->device = device;
 948                 /* Save this so that we can send it back */
 949                 netvsc_packet->completion.recv.recv_completion_tid =
 950                                         vmxferpage_packet->d.trans_id;
 951
 952                 netvsc_packet->total_data_buflen =
 953                                         vmxferpage_packet->ranges[i].byte_count;
 954                 netvsc_packet->page_buf_cnt = 1;
 955
 956                 netvsc_packet->page_buf[0].len =
 957                                         vmxferpage_packet->ranges[i].byte_count;
 958
 959                 start = virt_to_phys((void *)((unsigned long)net_device->
 960                 recv_buf + vmxferpage_packet->ranges[i].byte_offset));
 961
 962                 netvsc_packet->page_buf[0].pfn = start >> PAGE_SHIFT;
 963                 end_virtual = (unsigned long)net_device->recv_buf
 964                     + vmxferpage_packet->ranges[i].byte_offset
 965                     + vmxferpage_packet->ranges[i].byte_count - 1;
 966                 end = virt_to_phys((void *)end_virtual);
 967
 968                 /* Calculate the page relative offset */
 969                 netvsc_packet->page_buf[0].offset =
 970                         vmxferpage_packet->ranges[i].byte_offset &
 971                         (PAGE_SIZE - 1);
 972                 if ((end >> PAGE_SHIFT) != (start >> PAGE_SHIFT)) {
 973                         /* Handle frame across multiple pages: */
 974                         netvsc_packet->page_buf[0].len =
 975                                 (netvsc_packet->page_buf[0].pfn <<
 976                                  PAGE_SHIFT)
 977                                 + PAGE_SIZE - start;
 978                         bytes_remain = netvsc_packet->total_data_buflen -
 979                                         netvsc_packet->page_buf[0].len;
 980                         for (j = 1; j < NETVSC_PACKET_MAXPAGE; j++) {
 981                                 netvsc_packet->page_buf[j].offset = 0;
 982                                 if (bytes_remain <= PAGE_SIZE) {
 983                                         netvsc_packet->page_buf[j].len =
 984                                                 bytes_remain;
 985                                         bytes_remain = 0;
 986                                 } else {
 987                                         netvsc_packet->page_buf[j].len =
 988                                                 PAGE_SIZE;
 989                                         bytes_remain -= PAGE_SIZE;
 990                                 }
 991                                 netvsc_packet->page_buf[j].pfn =
 992                                     virt_to_phys((void *)(end_virtual -
 993                                                 bytes_remain)) >> PAGE_SHIFT;
 994                                 netvsc_packet->page_buf_cnt++;
 995                                 if (bytes_remain == 0)
 996                                         break;
 997                         }
 998                 }
 999
1000                 /* Pass it to the upper layer */
1001                 netvsc_drv->recv_cb(device, netvsc_packet);
1002
1003                 netvsc_receive_completion(netvsc_packet->
1004                                 completion.recv.recv_completion_ctx);
1005         }
1006
1007         put_net_device(device);
1008 }
1009
1010 static void netvsc_channel_cb(void *context)
1011 {
1012         int ret;
1013         struct hv_device *device = context;
1014         struct netvsc_device *net_device;
1015         u32 bytes_recvd;
1016         u64 request_id;
1017         unsigned char *packet;
1018         struct vmpacket_descriptor *desc;
1019         unsigned char *buffer;
1020         int bufferlen = NETVSC_PACKET_SIZE;
1021
1022         packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
1023                          GFP_ATOMIC);
1024         if (!packet)
1025                 return;
1026         buffer = packet;
1027
1028         net_device = get_inbound_net_device(device);
1029         if (!net_device) {
1030                 dev_err(&device->device, "net device (%p) shutting down..."
1031                            "ignoring inbound packets", net_device);
1032                 goto out;
1033         }
1034
1035         do {
1036                 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
1037                                            &bytes_recvd, &request_id);
1038                 if (ret == 0) {
1039                         if (bytes_recvd > 0) {
1040                                 desc = (struct vmpacket_descriptor *)buffer;
1041                                 switch (desc->type) {
1042                                 case VM_PKT_COMP:
1043                                         netvsc_send_completion(device, desc);
1044                                         break;
1045
1046                                 case VM_PKT_DATA_USING_XFER_PAGES:
1047                                         netvsc_receive(device, desc);
1048                                         break;
1049
1050                                 default:
1051                                         dev_err(&device->device,
1052                                                    "unhandled packet type %d, "
1053                                                    "tid %llx len %d\n",
1054                                                    desc->type, request_id,
1055                                                    bytes_recvd);
1056                                         break;
1057                                 }
1058
1059                                 /* reset */
1060                                 if (bufferlen > NETVSC_PACKET_SIZE) {
1061                                         kfree(buffer);
1062                                         buffer = packet;
1063                                         bufferlen = NETVSC_PACKET_SIZE;
1064                                 }
1065                         } else {
1066                                 /* reset */
1067                                 if (bufferlen > NETVSC_PACKET_SIZE) {
1068                                         kfree(buffer);
1069                                         buffer = packet;
1070                                         bufferlen = NETVSC_PACKET_SIZE;
1071                                 }
1072
1073                                 break;
1074                         }
1075                 } else if (ret == -2) {
1076                         /* Handle large packet */
1077                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1078                         if (buffer == NULL) {
1079                                 /* Try again next time around */
1080                                 dev_err(&device->device,
1081                                            "unable to allocate buffer of size "
1082                                            "(%d)!!", bytes_recvd);
1083                                 break;
1084                         }
1085
1086                         bufferlen = bytes_recvd;
1087                 }
1088         } while (1);
1089
1090         put_net_device(device);
1091 out:
1092         kfree(buffer);
1093         return;
1094 }
1095
1096 /*
1097  * netvsc_device_add - Callback when the device belonging to this
1098  * driver is added
1099  */
1100 static int netvsc_device_add(struct hv_device *device, void *additional_info)
1101 {
1102         int ret = 0;
1103         int i;
1104         struct netvsc_device *net_device;
1105         struct hv_netvsc_packet *packet, *pos;
1106         struct netvsc_driver *net_driver =
1107                 drv_to_netvscdrv(device->device.driver);
1108
1109         net_device = alloc_net_device(device);
1110         if (!net_device) {
1111                 ret = -1;
1112                 goto cleanup;
1113         }
1114
1115         /* Initialize the NetVSC channel extension */
1116         net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1117         spin_lock_init(&net_device->recv_pkt_list_lock);
1118
1119         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
1120
1121         INIT_LIST_HEAD(&net_device->recv_pkt_list);
1122
1123         for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
1124                 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
1125                                  (NETVSC_RECEIVE_SG_COUNT *
1126                                   sizeof(struct hv_page_buffer)), GFP_KERNEL);
1127                 if (!packet)
1128                         break;
1129
1130                 list_add_tail(&packet->list_ent,
1131                               &net_device->recv_pkt_list);
1132         }
1133         init_waitqueue_head(&net_device->channel_init_wait);
1134
1135         /* Open the channel */
1136         ret = vmbus_open(device->channel, net_driver->ring_buf_size,
1137                          net_driver->ring_buf_size, NULL, 0,
1138                          netvsc_channel_cb, device);
1139
1140         if (ret != 0) {
1141                 dev_err(&device->device, "unable to open channel: %d", ret);
1142                 ret = -1;
1143                 goto cleanup;
1144         }
1145
1146         /* Channel is opened */
1147         pr_info("hv_netvsc channel opened successfully");
1148
1149         /* Connect with the NetVsp */
1150         ret = netvsc_connect_vsp(device);
1151         if (ret != 0) {
1152                 dev_err(&device->device,
1153                         "unable to connect to NetVSP - %d", ret);
1154                 ret = -1;
1155                 goto close;
1156         }
1157
1158         return ret;
1159
1160 close:
1161         /* Now, we can close the channel safely */
1162         vmbus_close(device->channel);
1163
1164 cleanup:
1165
1166         if (net_device) {
1167                 list_for_each_entry_safe(packet, pos,
1168                                          &net_device->recv_pkt_list,
1169                                          list_ent) {
1170                         list_del(&packet->list_ent);
1171                         kfree(packet);
1172                 }
1173
1174                 release_outbound_net_device(device);
1175                 release_inbound_net_device(device);
1176
1177                 free_net_device(net_device);
1178         }
1179
1180         return ret;
1181 }
1182
1183 /*
1184  * netvsc_initialize - Main entry point
1185  */
1186 int netvsc_initialize(struct hv_driver *drv)
1187 {
1188         struct netvsc_driver *driver =
1189                 drv_to_netvscdrv(&drv->driver);
1190
1191         drv->name = driver_name;
1192         memcpy(&drv->dev_type, &netvsc_device_type, sizeof(struct hv_guid));
1193
1194         /* Setup the dispatch table */
1195         driver->base.dev_add    = netvsc_device_add;
1196         driver->base.dev_rm     = netvsc_device_remove;
1197         driver->base.cleanup            = netvsc_cleanup;
1198
1199         driver->send                    = netvsc_send;
1200
1201         rndis_filter_init(driver);
1202         return 0;
1203 }