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, 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 <linux/netdevice.h>
  31
  32 #include "hyperv_net.h"
  33
  34
  35 static struct netvsc_device *alloc_net_device(struct hv_device *device)
  36 {
  37         struct netvsc_device *net_device;
  38         struct net_device *ndev = hv_get_drvdata(device);
  39
  40         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  41         if (!net_device)
  42                 return NULL;
  43
  44
  45         net_device->destroy = false;
  46         net_device->dev = device;
  47         net_device->ndev = ndev;
  48
  49         hv_set_drvdata(device, net_device);
  50         return net_device;
  51 }
  52
  53 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  54 {
  55         struct netvsc_device *net_device;
  56
  57         net_device = hv_get_drvdata(device);
  58         if (net_device && net_device->destroy)
  59                 net_device = NULL;
  60
  61         return net_device;
  62 }
  63
  64 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  65 {
  66         struct netvsc_device *net_device;
  67
  68         net_device = hv_get_drvdata(device);
  69
  70         if (!net_device)
  71                 goto get_in_err;
  72
  73         if (net_device->destroy &&
  74                 atomic_read(&net_device->num_outstanding_sends) == 0)
  75                 net_device = NULL;
  76
  77 get_in_err:
  78         return net_device;
  79 }
  80
  81
  82 static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
  83 {
  84         struct nvsp_message *revoke_packet;
  85         int ret = 0;
  86         struct net_device *ndev = net_device->ndev;
  87
  88         /*
  89          * If we got a section count, it means we received a
  90          * SendReceiveBufferComplete msg (ie sent
  91          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
  92          * to send a revoke msg here
  93          */
  94         if (net_device->recv_section_cnt) {
  95                 /* Send the revoke receive buffer */
  96                 revoke_packet = &net_device->revoke_packet;
  97                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
  98
  99                 revoke_packet->hdr.msg_type =
 100                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 101                 revoke_packet->msg.v1_msg.
 102                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 103
 104                 ret = vmbus_sendpacket(net_device->dev->channel,
 105                                        revoke_packet,
 106                                        sizeof(struct nvsp_message),
 107                                        (unsigned long)revoke_packet,
 108                                        VM_PKT_DATA_INBAND, 0);
 109                 /*
 110                  * If we failed here, we might as well return and
 111                  * have a leak rather than continue and a bugchk
 112                  */
 113                 if (ret != 0) {
 114                         netdev_err(ndev, "unable to send "
 115                                 "revoke receive buffer to netvsp\n");
 116                         return ret;
 117                 }
 118         }
 119
 120         /* Teardown the gpadl on the vsp end */
 121         if (net_device->recv_buf_gpadl_handle) {
 122                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 123                            net_device->recv_buf_gpadl_handle);
 124
 125                 /* If we failed here, we might as well return and have a leak
 126                  * rather than continue and a bugchk
 127                  */
 128                 if (ret != 0) {
 129                         netdev_err(ndev,
 130                                    "unable to teardown receive buffer's gpadl\n");
 131                         return ret;
 132                 }
 133                 net_device->recv_buf_gpadl_handle = 0;
 134         }
 135
 136         if (net_device->recv_buf) {
 137                 /* Free up the receive buffer */
 138                 free_pages((unsigned long)net_device->recv_buf,
 139                         get_order(net_device->recv_buf_size));
 140                 net_device->recv_buf = NULL;
 141         }
 142
 143         if (net_device->recv_section) {
 144                 net_device->recv_section_cnt = 0;
 145                 kfree(net_device->recv_section);
 146                 net_device->recv_section = NULL;
 147         }
 148
 149         return ret;
 150 }
 151
 152 static int netvsc_init_recv_buf(struct hv_device *device)
 153 {
 154         int ret = 0;
 155         int t;
 156         struct netvsc_device *net_device;
 157         struct nvsp_message *init_packet;
 158         struct net_device *ndev;
 159
 160         net_device = get_outbound_net_device(device);
 161         if (!net_device)
 162                 return -ENODEV;
 163         ndev = net_device->ndev;
 164
 165         net_device->recv_buf =
 166                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
 167                                 get_order(net_device->recv_buf_size));
 168         if (!net_device->recv_buf) {
 169                 netdev_err(ndev, "unable to allocate receive "
 170                         "buffer of size %d\n", net_device->recv_buf_size);
 171                 ret = -ENOMEM;
 172                 goto cleanup;
 173         }
 174
 175         /*
 176          * Establish the gpadl handle for this buffer on this
 177          * channel.  Note: This call uses the vmbus connection rather
 178          * than the channel to establish the gpadl handle.
 179          */
 180         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 181                                     net_device->recv_buf_size,
 182                                     &net_device->recv_buf_gpadl_handle);
 183         if (ret != 0) {
 184                 netdev_err(ndev,
 185                         "unable to establish receive buffer's gpadl\n");
 186                 goto cleanup;
 187         }
 188
 189
 190         /* Notify the NetVsp of the gpadl handle */
 191         init_packet = &net_device->channel_init_pkt;
 192
 193         memset(init_packet, 0, sizeof(struct nvsp_message));
 194
 195         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 196         init_packet->msg.v1_msg.send_recv_buf.
 197                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 198         init_packet->msg.v1_msg.
 199                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 200
 201         /* Send the gpadl notification request */
 202         ret = vmbus_sendpacket(device->channel, init_packet,
 203                                sizeof(struct nvsp_message),
 204                                (unsigned long)init_packet,
 205                                VM_PKT_DATA_INBAND,
 206                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 207         if (ret != 0) {
 208                 netdev_err(ndev,
 209                         "unable to send receive buffer's gpadl to netvsp\n");
 210                 goto cleanup;
 211         }
 212
 213         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 214         BUG_ON(t == 0);
 215
 216
 217         /* Check the response */
 218         if (init_packet->msg.v1_msg.
 219             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 220                 netdev_err(ndev, "Unable to complete receive buffer "
 221                            "initialization with NetVsp - status %d\n",
 222                            init_packet->msg.v1_msg.
 223                            send_recv_buf_complete.status);
 224                 ret = -EINVAL;
 225                 goto cleanup;
 226         }
 227
 228         /* Parse the response */
 229
 230         net_device->recv_section_cnt = init_packet->msg.
 231                 v1_msg.send_recv_buf_complete.num_sections;
 232
 233         net_device->recv_section = kmemdup(init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 234                                            net_device->recv_section_cnt * sizeof(struct nvsp_1_receive_buffer_section),
 235                                            GFP_KERNEL);
 236         if (net_device->recv_section == NULL) {
 237                 ret = -EINVAL;
 238                 goto cleanup;
 239         }
 240
 241         /*
 242          * For 1st release, there should only be 1 section that represents the
 243          * entire receive buffer
 244          */
 245         if (net_device->recv_section_cnt != 1 ||
 246             net_device->recv_section->offset != 0) {
 247                 ret = -EINVAL;
 248                 goto cleanup;
 249         }
 250
 251         goto exit;
 252
 253 cleanup:
 254         netvsc_destroy_recv_buf(net_device);
 255
 256 exit:
 257         return ret;
 258 }
 259
 260
 261 static int netvsc_connect_vsp(struct hv_device *device)
 262 {
 263         int ret, t;
 264         struct netvsc_device *net_device;
 265         struct nvsp_message *init_packet;
 266         int ndis_version;
 267         struct net_device *ndev;
 268
 269         net_device = get_outbound_net_device(device);
 270         if (!net_device)
 271                 return -ENODEV;
 272         ndev = net_device->ndev;
 273
 274         init_packet = &net_device->channel_init_pkt;
 275
 276         memset(init_packet, 0, sizeof(struct nvsp_message));
 277         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 278         init_packet->msg.init_msg.init.min_protocol_ver =
 279                 NVSP_MIN_PROTOCOL_VERSION;
 280         init_packet->msg.init_msg.init.max_protocol_ver =
 281                 NVSP_MAX_PROTOCOL_VERSION;
 282
 283         /* Send the init request */
 284         ret = vmbus_sendpacket(device->channel, init_packet,
 285                                sizeof(struct nvsp_message),
 286                                (unsigned long)init_packet,
 287                                VM_PKT_DATA_INBAND,
 288                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 289
 290         if (ret != 0)
 291                 goto cleanup;
 292
 293         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 294
 295         if (t == 0) {
 296                 ret = -ETIMEDOUT;
 297                 goto cleanup;
 298         }
 299
 300         if (init_packet->msg.init_msg.init_complete.status !=
 301             NVSP_STAT_SUCCESS) {
 302                 ret = -EINVAL;
 303                 goto cleanup;
 304         }
 305
 306         if (init_packet->msg.init_msg.init_complete.
 307             negotiated_protocol_ver != NVSP_PROTOCOL_VERSION_1) {
 308                 ret = -EPROTO;
 309                 goto cleanup;
 310         }
 311         /* Send the ndis version */
 312         memset(init_packet, 0, sizeof(struct nvsp_message));
 313
 314         ndis_version = 0x00050000;
 315
 316         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 317         init_packet->msg.v1_msg.
 318                 send_ndis_ver.ndis_major_ver =
 319                                 (ndis_version & 0xFFFF0000) >> 16;
 320         init_packet->msg.v1_msg.
 321                 send_ndis_ver.ndis_minor_ver =
 322                                 ndis_version & 0xFFFF;
 323
 324         /* Send the init request */
 325         ret = vmbus_sendpacket(device->channel, init_packet,
 326                                 sizeof(struct nvsp_message),
 327                                 (unsigned long)init_packet,
 328                                 VM_PKT_DATA_INBAND, 0);
 329         if (ret != 0)
 330                 goto cleanup;
 331
 332         /* Post the big receive buffer to NetVSP */
 333         ret = netvsc_init_recv_buf(device);
 334
 335 cleanup:
 336         return ret;
 337 }
 338
 339 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 340 {
 341         netvsc_destroy_recv_buf(net_device);
 342 }
 343
 344 /*
 345  * netvsc_device_remove - Callback when the root bus device is removed
 346  */
 347 int netvsc_device_remove(struct hv_device *device)
 348 {
 349         struct netvsc_device *net_device;
 350         struct hv_netvsc_packet *netvsc_packet, *pos;
 351         unsigned long flags;
 352
 353         net_device = hv_get_drvdata(device);
 354         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 355         net_device->destroy = true;
 356         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 357
 358         /* Wait for all send completions */
 359         while (atomic_read(&net_device->num_outstanding_sends)) {
 360                 dev_info(&device->device,
 361                         "waiting for %d requests to complete...\n",
 362                         atomic_read(&net_device->num_outstanding_sends));
 363                 udelay(100);
 364         }
 365
 366         netvsc_disconnect_vsp(net_device);
 367
 368         /*
 369          * Since we have already drained, we don't need to busy wait
 370          * as was done in final_release_stor_device()
 371          * Note that we cannot set the ext pointer to NULL until
 372          * we have drained - to drain the outgoing packets, we need to
 373          * allow incoming packets.
 374          */
 375
 376         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 377         hv_set_drvdata(device, NULL);
 378         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 379
 380         /*
 381          * At this point, no one should be accessing net_device
 382          * except in here
 383          */
 384         dev_notice(&device->device, "net device safe to remove\n");
 385
 386         /* Now, we can close the channel safely */
 387         vmbus_close(device->channel);
 388
 389         /* Release all resources */
 390         list_for_each_entry_safe(netvsc_packet, pos,
 391                                  &net_device->recv_pkt_list, list_ent) {
 392                 list_del(&netvsc_packet->list_ent);
 393                 kfree(netvsc_packet);
 394         }
 395
 396         kfree(net_device);
 397         return 0;
 398 }
 399
 400 static void netvsc_send_completion(struct hv_device *device,
 401                                    struct vmpacket_descriptor *packet)
 402 {
 403         struct netvsc_device *net_device;
 404         struct nvsp_message *nvsp_packet;
 405         struct hv_netvsc_packet *nvsc_packet;
 406         struct net_device *ndev;
 407
 408         net_device = get_inbound_net_device(device);
 409         if (!net_device)
 410                 return;
 411         ndev = net_device->ndev;
 412
 413         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 414                         (packet->offset8 << 3));
 415
 416         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 417             (nvsp_packet->hdr.msg_type ==
 418              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 419             (nvsp_packet->hdr.msg_type ==
 420              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
 421                 /* Copy the response back */
 422                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 423                        sizeof(struct nvsp_message));
 424                 complete(&net_device->channel_init_wait);
 425         } else if (nvsp_packet->hdr.msg_type ==
 426                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 427                 /* Get the send context */
 428                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 429                         packet->trans_id;
 430
 431                 /* Notify the layer above us */
 432                 nvsc_packet->completion.send.send_completion(
 433                         nvsc_packet->completion.send.send_completion_ctx);
 434
 435                 atomic_dec(&net_device->num_outstanding_sends);
 436         } else {
 437                 netdev_err(ndev, "Unknown send completion packet type- "
 438                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 439         }
 440
 441 }
 442
 443 int netvsc_send(struct hv_device *device,
 444                         struct hv_netvsc_packet *packet)
 445 {
 446         struct netvsc_device *net_device;
 447         int ret = 0;
 448         struct nvsp_message sendMessage;
 449         struct net_device *ndev;
 450
 451         net_device = get_outbound_net_device(device);
 452         if (!net_device)
 453                 return -ENODEV;
 454         ndev = net_device->ndev;
 455
 456         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 457         if (packet->is_data_pkt) {
 458                 /* 0 is RMC_DATA; */
 459                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 460         } else {
 461                 /* 1 is RMC_CONTROL; */
 462                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 463         }
 464
 465         /* Not using send buffer section */
 466         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 467                 0xFFFFFFFF;
 468         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 469
 470         if (packet->page_buf_cnt) {
 471                 ret = vmbus_sendpacket_pagebuffer(device->channel,
 472                                                   packet->page_buf,
 473                                                   packet->page_buf_cnt,
 474                                                   &sendMessage,
 475                                                   sizeof(struct nvsp_message),
 476                                                   (unsigned long)packet);
 477         } else {
 478                 ret = vmbus_sendpacket(device->channel, &sendMessage,
 479                                 sizeof(struct nvsp_message),
 480                                 (unsigned long)packet,
 481                                 VM_PKT_DATA_INBAND,
 482                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 483
 484         }
 485
 486         if (ret != 0)
 487                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 488                            packet, ret);
 489         else
 490                 atomic_inc(&net_device->num_outstanding_sends);
 491
 492         return ret;
 493 }
 494
 495 static void netvsc_send_recv_completion(struct hv_device *device,
 496                                         u64 transaction_id)
 497 {
 498         struct nvsp_message recvcompMessage;
 499         int retries = 0;
 500         int ret;
 501         struct net_device *ndev;
 502         struct netvsc_device *net_device = hv_get_drvdata(device);
 503
 504         ndev = net_device->ndev;
 505
 506         recvcompMessage.hdr.msg_type =
 507                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 508
 509         /* FIXME: Pass in the status */
 510         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
 511                 NVSP_STAT_SUCCESS;
 512
 513 retry_send_cmplt:
 514         /* Send the completion */
 515         ret = vmbus_sendpacket(device->channel, &recvcompMessage,
 516                                sizeof(struct nvsp_message), transaction_id,
 517                                VM_PKT_COMP, 0);
 518         if (ret == 0) {
 519                 /* success */
 520                 /* no-op */
 521         } else if (ret == -EAGAIN) {
 522                 /* no more room...wait a bit and attempt to retry 3 times */
 523                 retries++;
 524                 netdev_err(ndev, "unable to send receive completion pkt"
 525                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 526
 527                 if (retries < 4) {
 528                         udelay(100);
 529                         goto retry_send_cmplt;
 530                 } else {
 531                         netdev_err(ndev, "unable to send receive "
 532                                 "completion pkt (tid %llx)...give up retrying\n",
 533                                 transaction_id);
 534                 }
 535         } else {
 536                 netdev_err(ndev, "unable to send receive "
 537                         "completion pkt - %llx\n", transaction_id);
 538         }
 539 }
 540
 541 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
 542 static void netvsc_receive_completion(void *context)
 543 {
 544         struct hv_netvsc_packet *packet = context;
 545         struct hv_device *device = (struct hv_device *)packet->device;
 546         struct netvsc_device *net_device;
 547         u64 transaction_id = 0;
 548         bool fsend_receive_comp = false;
 549         unsigned long flags;
 550         struct net_device *ndev;
 551
 552         /*
 553          * Even though it seems logical to do a GetOutboundNetDevice() here to
 554          * send out receive completion, we are using GetInboundNetDevice()
 555          * since we may have disable outbound traffic already.
 556          */
 557         net_device = get_inbound_net_device(device);
 558         if (!net_device)
 559                 return;
 560         ndev = net_device->ndev;
 561
 562         /* Overloading use of the lock. */
 563         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 564
 565         packet->xfer_page_pkt->count--;
 566
 567         /*
 568          * Last one in the line that represent 1 xfer page packet.
 569          * Return the xfer page packet itself to the freelist
 570          */
 571         if (packet->xfer_page_pkt->count == 0) {
 572                 fsend_receive_comp = true;
 573                 transaction_id = packet->completion.recv.recv_completion_tid;
 574                 list_add_tail(&packet->xfer_page_pkt->list_ent,
 575                               &net_device->recv_pkt_list);
 576
 577         }
 578
 579         /* Put the packet back */
 580         list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
 581         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
 582
 583         /* Send a receive completion for the xfer page packet */
 584         if (fsend_receive_comp)
 585                 netvsc_send_recv_completion(device, transaction_id);
 586
 587 }
 588
 589 static void netvsc_receive(struct hv_device *device,
 590                             struct vmpacket_descriptor *packet)
 591 {
 592         struct netvsc_device *net_device;
 593         struct vmtransfer_page_packet_header *vmxferpage_packet;
 594         struct nvsp_message *nvsp_packet;
 595         struct hv_netvsc_packet *netvsc_packet = NULL;
 596         unsigned long start;
 597         unsigned long end, end_virtual;
 598         /* struct netvsc_driver *netvscDriver; */
 599         struct xferpage_packet *xferpage_packet = NULL;
 600         int i, j;
 601         int count = 0, bytes_remain = 0;
 602         unsigned long flags;
 603         struct net_device *ndev;
 604
 605         LIST_HEAD(listHead);
 606
 607         net_device = get_inbound_net_device(device);
 608         if (!net_device)
 609                 return;
 610         ndev = net_device->ndev;
 611
 612         /*
 613          * All inbound packets other than send completion should be xfer page
 614          * packet
 615          */
 616         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 617                 netdev_err(ndev, "Unknown packet type received - %d\n",
 618                            packet->type);
 619                 return;
 620         }
 621
 622         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 623                         (packet->offset8 << 3));
 624
 625         /* Make sure this is a valid nvsp packet */
 626         if (nvsp_packet->hdr.msg_type !=
 627             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 628                 netdev_err(ndev, "Unknown nvsp packet type received-"
 629                         " %d\n", nvsp_packet->hdr.msg_type);
 630                 return;
 631         }
 632
 633         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 634
 635         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 636                 netdev_err(ndev, "Invalid xfer page set id - "
 637                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
 638                            vmxferpage_packet->xfer_pageset_id);
 639                 return;
 640         }
 641
 642         /*
 643          * Grab free packets (range count + 1) to represent this xfer
 644          * page packet. +1 to represent the xfer page packet itself.
 645          * We grab it here so that we know exactly how many we can
 646          * fulfil
 647          */
 648         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 649         while (!list_empty(&net_device->recv_pkt_list)) {
 650                 list_move_tail(net_device->recv_pkt_list.next, &listHead);
 651                 if (++count == vmxferpage_packet->range_cnt + 1)
 652                         break;
 653         }
 654         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
 655
 656         /*
 657          * We need at least 2 netvsc pkts (1 to represent the xfer
 658          * page and at least 1 for the range) i.e. we can handled
 659          * some of the xfer page packet ranges...
 660          */
 661         if (count < 2) {
 662                 netdev_err(ndev, "Got only %d netvsc pkt...needed "
 663                         "%d pkts. Dropping this xfer page packet completely!\n",
 664                         count, vmxferpage_packet->range_cnt + 1);
 665
 666                 /* Return it to the freelist */
 667                 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 668                 for (i = count; i != 0; i--) {
 669                         list_move_tail(listHead.next,
 670                                        &net_device->recv_pkt_list);
 671                 }
 672                 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
 673                                        flags);
 674
 675                 netvsc_send_recv_completion(device,
 676                                             vmxferpage_packet->d.trans_id);
 677
 678                 return;
 679         }
 680
 681         /* Remove the 1st packet to represent the xfer page packet itself */
 682         xferpage_packet = (struct xferpage_packet *)listHead.next;
 683         list_del(&xferpage_packet->list_ent);
 684
 685         /* This is how much we can satisfy */
 686         xferpage_packet->count = count - 1;
 687
 688         if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
 689                 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
 690                         "this xfer page...got %d\n",
 691                         vmxferpage_packet->range_cnt, xferpage_packet->count);
 692         }
 693
 694         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 695         for (i = 0; i < (count - 1); i++) {
 696                 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
 697                 list_del(&netvsc_packet->list_ent);
 698
 699                 /* Initialize the netvsc packet */
 700                 netvsc_packet->xfer_page_pkt = xferpage_packet;
 701                 netvsc_packet->completion.recv.recv_completion =
 702                                         netvsc_receive_completion;
 703                 netvsc_packet->completion.recv.recv_completion_ctx =
 704                                         netvsc_packet;
 705                 netvsc_packet->device = device;
 706                 /* Save this so that we can send it back */
 707                 netvsc_packet->completion.recv.recv_completion_tid =
 708                                         vmxferpage_packet->d.trans_id;
 709
 710                 netvsc_packet->total_data_buflen =
 711                                         vmxferpage_packet->ranges[i].byte_count;
 712                 netvsc_packet->page_buf_cnt = 1;
 713
 714                 netvsc_packet->page_buf[0].len =
 715                                         vmxferpage_packet->ranges[i].byte_count;
 716
 717                 start = virt_to_phys((void *)((unsigned long)net_device->
 718                 recv_buf + vmxferpage_packet->ranges[i].byte_offset));
 719
 720                 netvsc_packet->page_buf[0].pfn = start >> PAGE_SHIFT;
 721                 end_virtual = (unsigned long)net_device->recv_buf
 722                     + vmxferpage_packet->ranges[i].byte_offset
 723                     + vmxferpage_packet->ranges[i].byte_count - 1;
 724                 end = virt_to_phys((void *)end_virtual);
 725
 726                 /* Calculate the page relative offset */
 727                 netvsc_packet->page_buf[0].offset =
 728                         vmxferpage_packet->ranges[i].byte_offset &
 729                         (PAGE_SIZE - 1);
 730                 if ((end >> PAGE_SHIFT) != (start >> PAGE_SHIFT)) {
 731                         /* Handle frame across multiple pages: */
 732                         netvsc_packet->page_buf[0].len =
 733                                 (netvsc_packet->page_buf[0].pfn <<
 734                                  PAGE_SHIFT)
 735                                 + PAGE_SIZE - start;
 736                         bytes_remain = netvsc_packet->total_data_buflen -
 737                                         netvsc_packet->page_buf[0].len;
 738                         for (j = 1; j < NETVSC_PACKET_MAXPAGE; j++) {
 739                                 netvsc_packet->page_buf[j].offset = 0;
 740                                 if (bytes_remain <= PAGE_SIZE) {
 741                                         netvsc_packet->page_buf[j].len =
 742                                                 bytes_remain;
 743                                         bytes_remain = 0;
 744                                 } else {
 745                                         netvsc_packet->page_buf[j].len =
 746                                                 PAGE_SIZE;
 747                                         bytes_remain -= PAGE_SIZE;
 748                                 }
 749                                 netvsc_packet->page_buf[j].pfn =
 750                                     virt_to_phys((void *)(end_virtual -
 751                                                 bytes_remain)) >> PAGE_SHIFT;
 752                                 netvsc_packet->page_buf_cnt++;
 753                                 if (bytes_remain == 0)
 754                                         break;
 755                         }
 756                 }
 757
 758                 /* Pass it to the upper layer */
 759                 rndis_filter_receive(device, netvsc_packet);
 760
 761                 netvsc_receive_completion(netvsc_packet->
 762                                 completion.recv.recv_completion_ctx);
 763         }
 764
 765 }
 766
 767 static void netvsc_channel_cb(void *context)
 768 {
 769         int ret;
 770         struct hv_device *device = context;
 771         struct netvsc_device *net_device;
 772         u32 bytes_recvd;
 773         u64 request_id;
 774         unsigned char *packet;
 775         struct vmpacket_descriptor *desc;
 776         unsigned char *buffer;
 777         int bufferlen = NETVSC_PACKET_SIZE;
 778         struct net_device *ndev;
 779
 780         packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
 781                          GFP_ATOMIC);
 782         if (!packet)
 783                 return;
 784         buffer = packet;
 785
 786         net_device = get_inbound_net_device(device);
 787         if (!net_device)
 788                 goto out;
 789         ndev = net_device->ndev;
 790
 791         do {
 792                 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
 793                                            &bytes_recvd, &request_id);
 794                 if (ret == 0) {
 795                         if (bytes_recvd > 0) {
 796                                 desc = (struct vmpacket_descriptor *)buffer;
 797                                 switch (desc->type) {
 798                                 case VM_PKT_COMP:
 799                                         netvsc_send_completion(device, desc);
 800                                         break;
 801
 802                                 case VM_PKT_DATA_USING_XFER_PAGES:
 803                                         netvsc_receive(device, desc);
 804                                         break;
 805
 806                                 default:
 807                                         netdev_err(ndev,
 808                                                    "unhandled packet type %d, "
 809                                                    "tid %llx len %d\n",
 810                                                    desc->type, request_id,
 811                                                    bytes_recvd);
 812                                         break;
 813                                 }
 814
 815                                 /* reset */
 816                                 if (bufferlen > NETVSC_PACKET_SIZE) {
 817                                         kfree(buffer);
 818                                         buffer = packet;
 819                                         bufferlen = NETVSC_PACKET_SIZE;
 820                                 }
 821                         } else {
 822                                 /* reset */
 823                                 if (bufferlen > NETVSC_PACKET_SIZE) {
 824                                         kfree(buffer);
 825                                         buffer = packet;
 826                                         bufferlen = NETVSC_PACKET_SIZE;
 827                                 }
 828
 829                                 break;
 830                         }
 831                 } else if (ret == -ENOBUFS) {
 832                         /* Handle large packet */
 833                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
 834                         if (buffer == NULL) {
 835                                 /* Try again next time around */
 836                                 netdev_err(ndev,
 837                                            "unable to allocate buffer of size "
 838                                            "(%d)!!\n", bytes_recvd);
 839                                 break;
 840                         }
 841
 842                         bufferlen = bytes_recvd;
 843                 }
 844         } while (1);
 845
 846 out:
 847         kfree(buffer);
 848         return;
 849 }
 850
 851 /*
 852  * netvsc_device_add - Callback when the device belonging to this
 853  * driver is added
 854  */
 855 int netvsc_device_add(struct hv_device *device, void *additional_info)
 856 {
 857         int ret = 0;
 858         int i;
 859         int ring_size =
 860         ((struct netvsc_device_info *)additional_info)->ring_size;
 861         struct netvsc_device *net_device;
 862         struct hv_netvsc_packet *packet, *pos;
 863         struct net_device *ndev;
 864
 865         net_device = alloc_net_device(device);
 866         if (!net_device) {
 867                 ret = -ENOMEM;
 868                 goto cleanup;
 869         }
 870
 871         /*
 872          * Coming into this function, struct net_device * is
 873          * registered as the driver private data.
 874          * In alloc_net_device(), we register struct netvsc_device *
 875          * as the driver private data and stash away struct net_device *
 876          * in struct netvsc_device *.
 877          */
 878         ndev = net_device->ndev;
 879
 880         /* Initialize the NetVSC channel extension */
 881         net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 882         spin_lock_init(&net_device->recv_pkt_list_lock);
 883
 884         INIT_LIST_HEAD(&net_device->recv_pkt_list);
 885
 886         for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
 887                 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 888                                  (NETVSC_RECEIVE_SG_COUNT *
 889                                   sizeof(struct hv_page_buffer)), GFP_KERNEL);
 890                 if (!packet)
 891                         break;
 892
 893                 list_add_tail(&packet->list_ent,
 894                               &net_device->recv_pkt_list);
 895         }
 896         init_completion(&net_device->channel_init_wait);
 897
 898         /* Open the channel */
 899         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
 900                          ring_size * PAGE_SIZE, NULL, 0,
 901                          netvsc_channel_cb, device);
 902
 903         if (ret != 0) {
 904                 netdev_err(ndev, "unable to open channel: %d\n", ret);
 905                 goto cleanup;
 906         }
 907
 908         /* Channel is opened */
 909         pr_info("hv_netvsc channel opened successfully\n");
 910
 911         /* Connect with the NetVsp */
 912         ret = netvsc_connect_vsp(device);
 913         if (ret != 0) {
 914                 netdev_err(ndev,
 915                         "unable to connect to NetVSP - %d\n", ret);
 916                 goto close;
 917         }
 918
 919         return ret;
 920
 921 close:
 922         /* Now, we can close the channel safely */
 923         vmbus_close(device->channel);
 924
 925 cleanup:
 926
 927         if (net_device) {
 928                 list_for_each_entry_safe(packet, pos,
 929                                          &net_device->recv_pkt_list,
 930                                          list_ent) {
 931                         list_del(&packet->list_ent);
 932                         kfree(packet);
 933                 }
 934
 935                 kfree(net_device);
 936         }
 937
 938         return ret;
 939 }