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 #include <linux/if_ether.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         init_waitqueue_head(&net_device->wait_drain);
  46         net_device->start_remove = false;
  47         net_device->destroy = false;
  48         net_device->dev = device;
  49         net_device->ndev = ndev;
  50
  51         hv_set_drvdata(device, net_device);
  52         return net_device;
  53 }
  54
  55 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  56 {
  57         struct netvsc_device *net_device;
  58
  59         net_device = hv_get_drvdata(device);
  60         if (net_device && net_device->destroy)
  61                 net_device = NULL;
  62
  63         return net_device;
  64 }
  65
  66 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  67 {
  68         struct netvsc_device *net_device;
  69
  70         net_device = hv_get_drvdata(device);
  71
  72         if (!net_device)
  73                 goto get_in_err;
  74
  75         if (net_device->destroy &&
  76                 atomic_read(&net_device->num_outstanding_sends) == 0)
  77                 net_device = NULL;
  78
  79 get_in_err:
  80         return net_device;
  81 }
  82
  83
  84 static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
  85 {
  86         struct nvsp_message *revoke_packet;
  87         int ret = 0;
  88         struct net_device *ndev = net_device->ndev;
  89
  90         /*
  91          * If we got a section count, it means we received a
  92          * SendReceiveBufferComplete msg (ie sent
  93          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
  94          * to send a revoke msg here
  95          */
  96         if (net_device->recv_section_cnt) {
  97                 /* Send the revoke receive buffer */
  98                 revoke_packet = &net_device->revoke_packet;
  99                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 100
 101                 revoke_packet->hdr.msg_type =
 102                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 103                 revoke_packet->msg.v1_msg.
 104                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 105
 106                 ret = vmbus_sendpacket(net_device->dev->channel,
 107                                        revoke_packet,
 108                                        sizeof(struct nvsp_message),
 109                                        (unsigned long)revoke_packet,
 110                                        VM_PKT_DATA_INBAND, 0);
 111                 /*
 112                  * If we failed here, we might as well return and
 113                  * have a leak rather than continue and a bugchk
 114                  */
 115                 if (ret != 0) {
 116                         netdev_err(ndev, "unable to send "
 117                                 "revoke receive buffer to netvsp\n");
 118                         return ret;
 119                 }
 120         }
 121
 122         /* Teardown the gpadl on the vsp end */
 123         if (net_device->recv_buf_gpadl_handle) {
 124                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 125                            net_device->recv_buf_gpadl_handle);
 126
 127                 /* If we failed here, we might as well return and have a leak
 128                  * rather than continue and a bugchk
 129                  */
 130                 if (ret != 0) {
 131                         netdev_err(ndev,
 132                                    "unable to teardown receive buffer's gpadl\n");
 133                         return ret;
 134                 }
 135                 net_device->recv_buf_gpadl_handle = 0;
 136         }
 137
 138         if (net_device->recv_buf) {
 139                 /* Free up the receive buffer */
 140                 free_pages((unsigned long)net_device->recv_buf,
 141                         get_order(net_device->recv_buf_size));
 142                 net_device->recv_buf = NULL;
 143         }
 144
 145         if (net_device->recv_section) {
 146                 net_device->recv_section_cnt = 0;
 147                 kfree(net_device->recv_section);
 148                 net_device->recv_section = NULL;
 149         }
 150
 151         return ret;
 152 }
 153
 154 static int netvsc_init_recv_buf(struct hv_device *device)
 155 {
 156         int ret = 0;
 157         int t;
 158         struct netvsc_device *net_device;
 159         struct nvsp_message *init_packet;
 160         struct net_device *ndev;
 161
 162         net_device = get_outbound_net_device(device);
 163         if (!net_device)
 164                 return -ENODEV;
 165         ndev = net_device->ndev;
 166
 167         net_device->recv_buf =
 168                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
 169                                 get_order(net_device->recv_buf_size));
 170         if (!net_device->recv_buf) {
 171                 netdev_err(ndev, "unable to allocate receive "
 172                         "buffer of size %d\n", net_device->recv_buf_size);
 173                 ret = -ENOMEM;
 174                 goto cleanup;
 175         }
 176
 177         /*
 178          * Establish the gpadl handle for this buffer on this
 179          * channel.  Note: This call uses the vmbus connection rather
 180          * than the channel to establish the gpadl handle.
 181          */
 182         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 183                                     net_device->recv_buf_size,
 184                                     &net_device->recv_buf_gpadl_handle);
 185         if (ret != 0) {
 186                 netdev_err(ndev,
 187                         "unable to establish receive buffer's gpadl\n");
 188                 goto cleanup;
 189         }
 190
 191
 192         /* Notify the NetVsp of the gpadl handle */
 193         init_packet = &net_device->channel_init_pkt;
 194
 195         memset(init_packet, 0, sizeof(struct nvsp_message));
 196
 197         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 198         init_packet->msg.v1_msg.send_recv_buf.
 199                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 200         init_packet->msg.v1_msg.
 201                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 202
 203         /* Send the gpadl notification request */
 204         ret = vmbus_sendpacket(device->channel, init_packet,
 205                                sizeof(struct nvsp_message),
 206                                (unsigned long)init_packet,
 207                                VM_PKT_DATA_INBAND,
 208                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 209         if (ret != 0) {
 210                 netdev_err(ndev,
 211                         "unable to send receive buffer's gpadl to netvsp\n");
 212                 goto cleanup;
 213         }
 214
 215         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 216         BUG_ON(t == 0);
 217
 218
 219         /* Check the response */
 220         if (init_packet->msg.v1_msg.
 221             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 222                 netdev_err(ndev, "Unable to complete receive buffer "
 223                            "initialization with NetVsp - status %d\n",
 224                            init_packet->msg.v1_msg.
 225                            send_recv_buf_complete.status);
 226                 ret = -EINVAL;
 227                 goto cleanup;
 228         }
 229
 230         /* Parse the response */
 231
 232         net_device->recv_section_cnt = init_packet->msg.
 233                 v1_msg.send_recv_buf_complete.num_sections;
 234
 235         net_device->recv_section = kmemdup(
 236                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 237                 net_device->recv_section_cnt *
 238                 sizeof(struct nvsp_1_receive_buffer_section),
 239                 GFP_KERNEL);
 240         if (net_device->recv_section == NULL) {
 241                 ret = -EINVAL;
 242                 goto cleanup;
 243         }
 244
 245         /*
 246          * For 1st release, there should only be 1 section that represents the
 247          * entire receive buffer
 248          */
 249         if (net_device->recv_section_cnt != 1 ||
 250             net_device->recv_section->offset != 0) {
 251                 ret = -EINVAL;
 252                 goto cleanup;
 253         }
 254
 255         goto exit;
 256
 257 cleanup:
 258         netvsc_destroy_recv_buf(net_device);
 259
 260 exit:
 261         return ret;
 262 }
 263
 264
 265 /* Negotiate NVSP protocol version */
 266 static int negotiate_nvsp_ver(struct hv_device *device,
 267                               struct netvsc_device *net_device,
 268                               struct nvsp_message *init_packet,
 269                               u32 nvsp_ver)
 270 {
 271         int ret, t;
 272
 273         memset(init_packet, 0, sizeof(struct nvsp_message));
 274         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 275         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 276         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 277
 278         /* Send the init request */
 279         ret = vmbus_sendpacket(device->channel, init_packet,
 280                                sizeof(struct nvsp_message),
 281                                (unsigned long)init_packet,
 282                                VM_PKT_DATA_INBAND,
 283                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 284
 285         if (ret != 0)
 286                 return ret;
 287
 288         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 289
 290         if (t == 0)
 291                 return -ETIMEDOUT;
 292
 293         if (init_packet->msg.init_msg.init_complete.status !=
 294             NVSP_STAT_SUCCESS)
 295                 return -EINVAL;
 296
 297         if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
 298                 return 0;
 299
 300         /* NVSPv2 only: Send NDIS config */
 301         memset(init_packet, 0, sizeof(struct nvsp_message));
 302         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 303         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
 304         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 305
 306         ret = vmbus_sendpacket(device->channel, init_packet,
 307                                 sizeof(struct nvsp_message),
 308                                 (unsigned long)init_packet,
 309                                 VM_PKT_DATA_INBAND, 0);
 310
 311         return ret;
 312 }
 313
 314 static int netvsc_connect_vsp(struct hv_device *device)
 315 {
 316         int ret;
 317         struct netvsc_device *net_device;
 318         struct nvsp_message *init_packet;
 319         int ndis_version;
 320         struct net_device *ndev;
 321
 322         net_device = get_outbound_net_device(device);
 323         if (!net_device)
 324                 return -ENODEV;
 325         ndev = net_device->ndev;
 326
 327         init_packet = &net_device->channel_init_pkt;
 328
 329         /* Negotiate the latest NVSP protocol supported */
 330         if (negotiate_nvsp_ver(device, net_device, init_packet,
 331                                NVSP_PROTOCOL_VERSION_2) == 0) {
 332                 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
 333         } else if (negotiate_nvsp_ver(device, net_device, init_packet,
 334                                     NVSP_PROTOCOL_VERSION_1) == 0) {
 335                 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
 336         } else {
 337                 ret = -EPROTO;
 338                 goto cleanup;
 339         }
 340
 341         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 342
 343         /* Send the ndis version */
 344         memset(init_packet, 0, sizeof(struct nvsp_message));
 345
 346         ndis_version = 0x00050001;
 347
 348         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 349         init_packet->msg.v1_msg.
 350                 send_ndis_ver.ndis_major_ver =
 351                                 (ndis_version & 0xFFFF0000) >> 16;
 352         init_packet->msg.v1_msg.
 353                 send_ndis_ver.ndis_minor_ver =
 354                                 ndis_version & 0xFFFF;
 355
 356         /* Send the init request */
 357         ret = vmbus_sendpacket(device->channel, init_packet,
 358                                 sizeof(struct nvsp_message),
 359                                 (unsigned long)init_packet,
 360                                 VM_PKT_DATA_INBAND, 0);
 361         if (ret != 0)
 362                 goto cleanup;
 363
 364         /* Post the big receive buffer to NetVSP */
 365         ret = netvsc_init_recv_buf(device);
 366
 367 cleanup:
 368         return ret;
 369 }
 370
 371 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 372 {
 373         netvsc_destroy_recv_buf(net_device);
 374 }
 375
 376 /*
 377  * netvsc_device_remove - Callback when the root bus device is removed
 378  */
 379 int netvsc_device_remove(struct hv_device *device)
 380 {
 381         struct netvsc_device *net_device;
 382         struct hv_netvsc_packet *netvsc_packet, *pos;
 383         unsigned long flags;
 384
 385         net_device = hv_get_drvdata(device);
 386         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 387         net_device->destroy = true;
 388         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 389
 390         /* Wait for all send completions */
 391         wait_event(net_device->wait_drain,
 392                    atomic_read(&net_device->num_outstanding_sends) == 0);
 393
 394         netvsc_disconnect_vsp(net_device);
 395
 396         /*
 397          * Since we have already drained, we don't need to busy wait
 398          * as was done in final_release_stor_device()
 399          * Note that we cannot set the ext pointer to NULL until
 400          * we have drained - to drain the outgoing packets, we need to
 401          * allow incoming packets.
 402          */
 403
 404         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 405         hv_set_drvdata(device, NULL);
 406         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 407
 408         /*
 409          * At this point, no one should be accessing net_device
 410          * except in here
 411          */
 412         dev_notice(&device->device, "net device safe to remove\n");
 413
 414         /* Now, we can close the channel safely */
 415         vmbus_close(device->channel);
 416
 417         /* Release all resources */
 418         list_for_each_entry_safe(netvsc_packet, pos,
 419                                  &net_device->recv_pkt_list, list_ent) {
 420                 list_del(&netvsc_packet->list_ent);
 421                 kfree(netvsc_packet);
 422         }
 423
 424         kfree(net_device);
 425         return 0;
 426 }
 427
 428
 429 #define RING_AVAIL_PERCENT_HIWATER 20
 430 #define RING_AVAIL_PERCENT_LOWATER 10
 431
 432 /*
 433  * Get the percentage of available bytes to write in the ring.
 434  * The return value is in range from 0 to 100.
 435  */
 436 static inline u32 hv_ringbuf_avail_percent(
 437                 struct hv_ring_buffer_info *ring_info)
 438 {
 439         u32 avail_read, avail_write;
 440
 441         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 442
 443         return avail_write * 100 / ring_info->ring_datasize;
 444 }
 445
 446 static void netvsc_send_completion(struct hv_device *device,
 447                                    struct vmpacket_descriptor *packet)
 448 {
 449         struct netvsc_device *net_device;
 450         struct nvsp_message *nvsp_packet;
 451         struct hv_netvsc_packet *nvsc_packet;
 452         struct net_device *ndev;
 453
 454         net_device = get_inbound_net_device(device);
 455         if (!net_device)
 456                 return;
 457         ndev = net_device->ndev;
 458
 459         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 460                         (packet->offset8 << 3));
 461
 462         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 463             (nvsp_packet->hdr.msg_type ==
 464              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 465             (nvsp_packet->hdr.msg_type ==
 466              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
 467                 /* Copy the response back */
 468                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 469                        sizeof(struct nvsp_message));
 470                 complete(&net_device->channel_init_wait);
 471         } else if (nvsp_packet->hdr.msg_type ==
 472                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 473                 int num_outstanding_sends;
 474
 475                 /* Get the send context */
 476                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 477                         packet->trans_id;
 478
 479                 /* Notify the layer above us */
 480                 nvsc_packet->completion.send.send_completion(
 481                         nvsc_packet->completion.send.send_completion_ctx);
 482
 483                 num_outstanding_sends =
 484                         atomic_dec_return(&net_device->num_outstanding_sends);
 485
 486                 if (net_device->destroy && num_outstanding_sends == 0)
 487                         wake_up(&net_device->wait_drain);
 488
 489                 if (netif_queue_stopped(ndev) && !net_device->start_remove &&
 490                         (hv_ringbuf_avail_percent(&device->channel->outbound)
 491                         > RING_AVAIL_PERCENT_HIWATER ||
 492                         num_outstanding_sends < 1))
 493                                 netif_wake_queue(ndev);
 494         } else {
 495                 netdev_err(ndev, "Unknown send completion packet type- "
 496                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 497         }
 498
 499 }
 500
 501 int netvsc_send(struct hv_device *device,
 502                         struct hv_netvsc_packet *packet)
 503 {
 504         struct netvsc_device *net_device;
 505         int ret = 0;
 506         struct nvsp_message sendMessage;
 507         struct net_device *ndev;
 508
 509         net_device = get_outbound_net_device(device);
 510         if (!net_device)
 511                 return -ENODEV;
 512         ndev = net_device->ndev;
 513
 514         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 515         if (packet->is_data_pkt) {
 516                 /* 0 is RMC_DATA; */
 517                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 518         } else {
 519                 /* 1 is RMC_CONTROL; */
 520                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 521         }
 522
 523         /* Not using send buffer section */
 524         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 525                 0xFFFFFFFF;
 526         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 527
 528         if (packet->page_buf_cnt) {
 529                 ret = vmbus_sendpacket_pagebuffer(device->channel,
 530                                                   packet->page_buf,
 531                                                   packet->page_buf_cnt,
 532                                                   &sendMessage,
 533                                                   sizeof(struct nvsp_message),
 534                                                   (unsigned long)packet);
 535         } else {
 536                 ret = vmbus_sendpacket(device->channel, &sendMessage,
 537                                 sizeof(struct nvsp_message),
 538                                 (unsigned long)packet,
 539                                 VM_PKT_DATA_INBAND,
 540                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 541
 542         }
 543
 544         if (ret == 0) {
 545                 atomic_inc(&net_device->num_outstanding_sends);
 546                 if (hv_ringbuf_avail_percent(&device->channel->outbound) <
 547                         RING_AVAIL_PERCENT_LOWATER) {
 548                         netif_stop_queue(ndev);
 549                         if (atomic_read(&net_device->
 550                                 num_outstanding_sends) < 1)
 551                                 netif_wake_queue(ndev);
 552                 }
 553         } else if (ret == -EAGAIN) {
 554                 netif_stop_queue(ndev);
 555                 if (atomic_read(&net_device->num_outstanding_sends) < 1) {
 556                         netif_wake_queue(ndev);
 557                         ret = -ENOSPC;
 558                 }
 559         } else {
 560                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 561                            packet, ret);
 562         }
 563
 564         return ret;
 565 }
 566
 567 static void netvsc_send_recv_completion(struct hv_device *device,
 568                                         u64 transaction_id)
 569 {
 570         struct nvsp_message recvcompMessage;
 571         int retries = 0;
 572         int ret;
 573         struct net_device *ndev;
 574         struct netvsc_device *net_device = hv_get_drvdata(device);
 575
 576         ndev = net_device->ndev;
 577
 578         recvcompMessage.hdr.msg_type =
 579                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 580
 581         /* FIXME: Pass in the status */
 582         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
 583                 NVSP_STAT_SUCCESS;
 584
 585 retry_send_cmplt:
 586         /* Send the completion */
 587         ret = vmbus_sendpacket(device->channel, &recvcompMessage,
 588                                sizeof(struct nvsp_message), transaction_id,
 589                                VM_PKT_COMP, 0);
 590         if (ret == 0) {
 591                 /* success */
 592                 /* no-op */
 593         } else if (ret == -EAGAIN) {
 594                 /* no more room...wait a bit and attempt to retry 3 times */
 595                 retries++;
 596                 netdev_err(ndev, "unable to send receive completion pkt"
 597                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 598
 599                 if (retries < 4) {
 600                         udelay(100);
 601                         goto retry_send_cmplt;
 602                 } else {
 603                         netdev_err(ndev, "unable to send receive "
 604                                 "completion pkt (tid %llx)...give up retrying\n",
 605                                 transaction_id);
 606                 }
 607         } else {
 608                 netdev_err(ndev, "unable to send receive "
 609                         "completion pkt - %llx\n", transaction_id);
 610         }
 611 }
 612
 613 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
 614 static void netvsc_receive_completion(void *context)
 615 {
 616         struct hv_netvsc_packet *packet = context;
 617         struct hv_device *device = packet->device;
 618         struct netvsc_device *net_device;
 619         u64 transaction_id = 0;
 620         bool fsend_receive_comp = false;
 621         unsigned long flags;
 622         struct net_device *ndev;
 623
 624         /*
 625          * Even though it seems logical to do a GetOutboundNetDevice() here to
 626          * send out receive completion, we are using GetInboundNetDevice()
 627          * since we may have disable outbound traffic already.
 628          */
 629         net_device = get_inbound_net_device(device);
 630         if (!net_device)
 631                 return;
 632         ndev = net_device->ndev;
 633
 634         /* Overloading use of the lock. */
 635         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 636
 637         packet->xfer_page_pkt->count--;
 638
 639         /*
 640          * Last one in the line that represent 1 xfer page packet.
 641          * Return the xfer page packet itself to the freelist
 642          */
 643         if (packet->xfer_page_pkt->count == 0) {
 644                 fsend_receive_comp = true;
 645                 transaction_id = packet->completion.recv.recv_completion_tid;
 646                 list_add_tail(&packet->xfer_page_pkt->list_ent,
 647                               &net_device->recv_pkt_list);
 648
 649         }
 650
 651         /* Put the packet back */
 652         list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
 653         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
 654
 655         /* Send a receive completion for the xfer page packet */
 656         if (fsend_receive_comp)
 657                 netvsc_send_recv_completion(device, transaction_id);
 658
 659 }
 660
 661 static void netvsc_receive(struct hv_device *device,
 662                             struct vmpacket_descriptor *packet)
 663 {
 664         struct netvsc_device *net_device;
 665         struct vmtransfer_page_packet_header *vmxferpage_packet;
 666         struct nvsp_message *nvsp_packet;
 667         struct hv_netvsc_packet *netvsc_packet = NULL;
 668         /* struct netvsc_driver *netvscDriver; */
 669         struct xferpage_packet *xferpage_packet = NULL;
 670         int i;
 671         int count = 0;
 672         unsigned long flags;
 673         struct net_device *ndev;
 674
 675         LIST_HEAD(listHead);
 676
 677         net_device = get_inbound_net_device(device);
 678         if (!net_device)
 679                 return;
 680         ndev = net_device->ndev;
 681
 682         /*
 683          * All inbound packets other than send completion should be xfer page
 684          * packet
 685          */
 686         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 687                 netdev_err(ndev, "Unknown packet type received - %d\n",
 688                            packet->type);
 689                 return;
 690         }
 691
 692         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 693                         (packet->offset8 << 3));
 694
 695         /* Make sure this is a valid nvsp packet */
 696         if (nvsp_packet->hdr.msg_type !=
 697             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 698                 netdev_err(ndev, "Unknown nvsp packet type received-"
 699                         " %d\n", nvsp_packet->hdr.msg_type);
 700                 return;
 701         }
 702
 703         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 704
 705         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 706                 netdev_err(ndev, "Invalid xfer page set id - "
 707                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
 708                            vmxferpage_packet->xfer_pageset_id);
 709                 return;
 710         }
 711
 712         /*
 713          * Grab free packets (range count + 1) to represent this xfer
 714          * page packet. +1 to represent the xfer page packet itself.
 715          * We grab it here so that we know exactly how many we can
 716          * fulfil
 717          */
 718         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 719         while (!list_empty(&net_device->recv_pkt_list)) {
 720                 list_move_tail(net_device->recv_pkt_list.next, &listHead);
 721                 if (++count == vmxferpage_packet->range_cnt + 1)
 722                         break;
 723         }
 724         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
 725
 726         /*
 727          * We need at least 2 netvsc pkts (1 to represent the xfer
 728          * page and at least 1 for the range) i.e. we can handled
 729          * some of the xfer page packet ranges...
 730          */
 731         if (count < 2) {
 732                 netdev_err(ndev, "Got only %d netvsc pkt...needed "
 733                         "%d pkts. Dropping this xfer page packet completely!\n",
 734                         count, vmxferpage_packet->range_cnt + 1);
 735
 736                 /* Return it to the freelist */
 737                 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
 738                 for (i = count; i != 0; i--) {
 739                         list_move_tail(listHead.next,
 740                                        &net_device->recv_pkt_list);
 741                 }
 742                 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
 743                                        flags);
 744
 745                 netvsc_send_recv_completion(device,
 746                                             vmxferpage_packet->d.trans_id);
 747
 748                 return;
 749         }
 750
 751         /* Remove the 1st packet to represent the xfer page packet itself */
 752         xferpage_packet = (struct xferpage_packet *)listHead.next;
 753         list_del(&xferpage_packet->list_ent);
 754
 755         /* This is how much we can satisfy */
 756         xferpage_packet->count = count - 1;
 757
 758         if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
 759                 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
 760                         "this xfer page...got %d\n",
 761                         vmxferpage_packet->range_cnt, xferpage_packet->count);
 762         }
 763
 764         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 765         for (i = 0; i < (count - 1); i++) {
 766                 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
 767                 list_del(&netvsc_packet->list_ent);
 768
 769                 /* Initialize the netvsc packet */
 770                 netvsc_packet->xfer_page_pkt = xferpage_packet;
 771                 netvsc_packet->completion.recv.recv_completion =
 772                                         netvsc_receive_completion;
 773                 netvsc_packet->completion.recv.recv_completion_ctx =
 774                                         netvsc_packet;
 775                 netvsc_packet->device = device;
 776                 /* Save this so that we can send it back */
 777                 netvsc_packet->completion.recv.recv_completion_tid =
 778                                         vmxferpage_packet->d.trans_id;
 779
 780                 netvsc_packet->data = (void *)((unsigned long)net_device->
 781                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
 782                 netvsc_packet->total_data_buflen =
 783                                         vmxferpage_packet->ranges[i].byte_count;
 784
 785                 /* Pass it to the upper layer */
 786                 rndis_filter_receive(device, netvsc_packet);
 787
 788                 netvsc_receive_completion(netvsc_packet->
 789                                 completion.recv.recv_completion_ctx);
 790         }
 791
 792 }
 793
 794 static void netvsc_channel_cb(void *context)
 795 {
 796         int ret;
 797         struct hv_device *device = context;
 798         struct netvsc_device *net_device;
 799         u32 bytes_recvd;
 800         u64 request_id;
 801         unsigned char *packet;
 802         struct vmpacket_descriptor *desc;
 803         unsigned char *buffer;
 804         int bufferlen = NETVSC_PACKET_SIZE;
 805         struct net_device *ndev;
 806
 807         packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
 808                          GFP_ATOMIC);
 809         if (!packet)
 810                 return;
 811         buffer = packet;
 812
 813         net_device = get_inbound_net_device(device);
 814         if (!net_device)
 815                 goto out;
 816         ndev = net_device->ndev;
 817
 818         do {
 819                 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
 820                                            &bytes_recvd, &request_id);
 821                 if (ret == 0) {
 822                         if (bytes_recvd > 0) {
 823                                 desc = (struct vmpacket_descriptor *)buffer;
 824                                 switch (desc->type) {
 825                                 case VM_PKT_COMP:
 826                                         netvsc_send_completion(device, desc);
 827                                         break;
 828
 829                                 case VM_PKT_DATA_USING_XFER_PAGES:
 830                                         netvsc_receive(device, desc);
 831                                         break;
 832
 833                                 default:
 834                                         netdev_err(ndev,
 835                                                    "unhandled packet type %d, "
 836                                                    "tid %llx len %d\n",
 837                                                    desc->type, request_id,
 838                                                    bytes_recvd);
 839                                         break;
 840                                 }
 841
 842                                 /* reset */
 843                                 if (bufferlen > NETVSC_PACKET_SIZE) {
 844                                         kfree(buffer);
 845                                         buffer = packet;
 846                                         bufferlen = NETVSC_PACKET_SIZE;
 847                                 }
 848                         } else {
 849                                 /* reset */
 850                                 if (bufferlen > NETVSC_PACKET_SIZE) {
 851                                         kfree(buffer);
 852                                         buffer = packet;
 853                                         bufferlen = NETVSC_PACKET_SIZE;
 854                                 }
 855
 856                                 break;
 857                         }
 858                 } else if (ret == -ENOBUFS) {
 859                         /* Handle large packet */
 860                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
 861                         if (buffer == NULL) {
 862                                 /* Try again next time around */
 863                                 netdev_err(ndev,
 864                                            "unable to allocate buffer of size "
 865                                            "(%d)!!\n", bytes_recvd);
 866                                 break;
 867                         }
 868
 869                         bufferlen = bytes_recvd;
 870                 }
 871         } while (1);
 872
 873 out:
 874         kfree(buffer);
 875         return;
 876 }
 877
 878 /*
 879  * netvsc_device_add - Callback when the device belonging to this
 880  * driver is added
 881  */
 882 int netvsc_device_add(struct hv_device *device, void *additional_info)
 883 {
 884         int ret = 0;
 885         int i;
 886         int ring_size =
 887         ((struct netvsc_device_info *)additional_info)->ring_size;
 888         struct netvsc_device *net_device;
 889         struct hv_netvsc_packet *packet, *pos;
 890         struct net_device *ndev;
 891
 892         net_device = alloc_net_device(device);
 893         if (!net_device) {
 894                 ret = -ENOMEM;
 895                 goto cleanup;
 896         }
 897
 898         /*
 899          * Coming into this function, struct net_device * is
 900          * registered as the driver private data.
 901          * In alloc_net_device(), we register struct netvsc_device *
 902          * as the driver private data and stash away struct net_device *
 903          * in struct netvsc_device *.
 904          */
 905         ndev = net_device->ndev;
 906
 907         /* Initialize the NetVSC channel extension */
 908         net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 909         spin_lock_init(&net_device->recv_pkt_list_lock);
 910
 911         INIT_LIST_HEAD(&net_device->recv_pkt_list);
 912
 913         for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
 914                 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 915                                  (NETVSC_RECEIVE_SG_COUNT *
 916                                   sizeof(struct hv_page_buffer)), GFP_KERNEL);
 917                 if (!packet)
 918                         break;
 919
 920                 list_add_tail(&packet->list_ent,
 921                               &net_device->recv_pkt_list);
 922         }
 923         init_completion(&net_device->channel_init_wait);
 924
 925         /* Open the channel */
 926         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
 927                          ring_size * PAGE_SIZE, NULL, 0,
 928                          netvsc_channel_cb, device);
 929
 930         if (ret != 0) {
 931                 netdev_err(ndev, "unable to open channel: %d\n", ret);
 932                 goto cleanup;
 933         }
 934
 935         /* Channel is opened */
 936         pr_info("hv_netvsc channel opened successfully\n");
 937
 938         /* Connect with the NetVsp */
 939         ret = netvsc_connect_vsp(device);
 940         if (ret != 0) {
 941                 netdev_err(ndev,
 942                         "unable to connect to NetVSP - %d\n", ret);
 943                 goto close;
 944         }
 945
 946         return ret;
 947
 948 close:
 949         /* Now, we can close the channel safely */
 950         vmbus_close(device->channel);
 951
 952 cleanup:
 953
 954         if (net_device) {
 955                 list_for_each_entry_safe(packet, pos,
 956                                          &net_device->recv_pkt_list,
 957                                          list_ent) {
 958                         list_del(&packet->list_ent);
 959                         kfree(packet);
 960                 }
 961
 962                 kfree(net_device);
 963         }
 964
 965         return ret;
 966 }