drivers/net/hyperv/netvsc.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/wait.h>
  25 #include <linux/mm.h>
  26 #include <linux/delay.h>
  27 #include <linux/io.h>
  28 #include <linux/slab.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <asm/sync_bitops.h>
  32
  33 #include "hyperv_net.h"
  34
  35
  36 static struct netvsc_device *alloc_net_device(struct hv_device *device)
  37 {
  38         struct netvsc_device *net_device;
  39         struct net_device *ndev = hv_get_drvdata(device);
  40
  41         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  42         if (!net_device)
  43                 return NULL;
  44
  45         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_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                 vfree(net_device->recv_buf);
 141                 net_device->recv_buf = NULL;
 142         }
 143
 144         if (net_device->recv_section) {
 145                 net_device->recv_section_cnt = 0;
 146                 kfree(net_device->recv_section);
 147                 net_device->recv_section = NULL;
 148         }
 149
 150         /* Deal with the send buffer we may have setup.
 151          * If we got a  send section size, it means we received a
 152          * SendsendBufferComplete msg (ie sent
 153          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 154          * to send a revoke msg here
 155          */
 156         if (net_device->send_section_size) {
 157                 /* Send the revoke receive buffer */
 158                 revoke_packet = &net_device->revoke_packet;
 159                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 160
 161                 revoke_packet->hdr.msg_type =
 162                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 163                 revoke_packet->msg.v1_msg.revoke_recv_buf.id = 0;
 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                 /* If we failed here, we might as well return and
 171                  * have a leak rather than continue and a bugchk
 172                  */
 173                 if (ret != 0) {
 174                         netdev_err(ndev, "unable to send "
 175                                    "revoke send buffer to netvsp\n");
 176                         return ret;
 177                 }
 178         }
 179         /* Teardown the gpadl on the vsp end */
 180         if (net_device->send_buf_gpadl_handle) {
 181                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 182                                            net_device->send_buf_gpadl_handle);
 183
 184                 /* If we failed here, we might as well return and have a leak
 185                  * rather than continue and a bugchk
 186                  */
 187                 if (ret != 0) {
 188                         netdev_err(ndev,
 189                                    "unable to teardown send buffer's gpadl\n");
 190                         return ret;
 191                 }
 192                 net_device->send_buf_gpadl_handle = 0;
 193         }
 194         if (net_device->send_buf) {
 195                 /* Free up the receive buffer */
 196                 free_pages((unsigned long)net_device->send_buf,
 197                            get_order(net_device->send_buf_size));
 198                 net_device->send_buf = NULL;
 199         }
 200         kfree(net_device->send_section_map);
 201
 202         return ret;
 203 }
 204
 205 static int netvsc_init_buf(struct hv_device *device)
 206 {
 207         int ret = 0;
 208         int t;
 209         struct netvsc_device *net_device;
 210         struct nvsp_message *init_packet;
 211         struct net_device *ndev;
 212
 213         net_device = get_outbound_net_device(device);
 214         if (!net_device)
 215                 return -ENODEV;
 216         ndev = net_device->ndev;
 217
 218         net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 219         if (!net_device->recv_buf) {
 220                 netdev_err(ndev, "unable to allocate receive "
 221                         "buffer of size %d\n", net_device->recv_buf_size);
 222                 ret = -ENOMEM;
 223                 goto cleanup;
 224         }
 225
 226         /*
 227          * Establish the gpadl handle for this buffer on this
 228          * channel.  Note: This call uses the vmbus connection rather
 229          * than the channel to establish the gpadl handle.
 230          */
 231         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 232                                     net_device->recv_buf_size,
 233                                     &net_device->recv_buf_gpadl_handle);
 234         if (ret != 0) {
 235                 netdev_err(ndev,
 236                         "unable to establish receive buffer's gpadl\n");
 237                 goto cleanup;
 238         }
 239
 240
 241         /* Notify the NetVsp of the gpadl handle */
 242         init_packet = &net_device->channel_init_pkt;
 243
 244         memset(init_packet, 0, sizeof(struct nvsp_message));
 245
 246         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 247         init_packet->msg.v1_msg.send_recv_buf.
 248                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 249         init_packet->msg.v1_msg.
 250                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 251
 252         /* Send the gpadl notification request */
 253         ret = vmbus_sendpacket(device->channel, init_packet,
 254                                sizeof(struct nvsp_message),
 255                                (unsigned long)init_packet,
 256                                VM_PKT_DATA_INBAND,
 257                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 258         if (ret != 0) {
 259                 netdev_err(ndev,
 260                         "unable to send receive buffer's gpadl to netvsp\n");
 261                 goto cleanup;
 262         }
 263
 264         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 265         BUG_ON(t == 0);
 266
 267
 268         /* Check the response */
 269         if (init_packet->msg.v1_msg.
 270             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 271                 netdev_err(ndev, "Unable to complete receive buffer "
 272                            "initialization with NetVsp - status %d\n",
 273                            init_packet->msg.v1_msg.
 274                            send_recv_buf_complete.status);
 275                 ret = -EINVAL;
 276                 goto cleanup;
 277         }
 278
 279         /* Parse the response */
 280
 281         net_device->recv_section_cnt = init_packet->msg.
 282                 v1_msg.send_recv_buf_complete.num_sections;
 283
 284         net_device->recv_section = kmemdup(
 285                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 286                 net_device->recv_section_cnt *
 287                 sizeof(struct nvsp_1_receive_buffer_section),
 288                 GFP_KERNEL);
 289         if (net_device->recv_section == NULL) {
 290                 ret = -EINVAL;
 291                 goto cleanup;
 292         }
 293
 294         /*
 295          * For 1st release, there should only be 1 section that represents the
 296          * entire receive buffer
 297          */
 298         if (net_device->recv_section_cnt != 1 ||
 299             net_device->recv_section->offset != 0) {
 300                 ret = -EINVAL;
 301                 goto cleanup;
 302         }
 303
 304         /* Now setup the send buffer.
 305          */
 306         net_device->send_buf =
 307                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
 308                                          get_order(net_device->send_buf_size));
 309         if (!net_device->send_buf) {
 310                 netdev_err(ndev, "unable to allocate send "
 311                            "buffer of size %d\n", net_device->send_buf_size);
 312                 ret = -ENOMEM;
 313                 goto cleanup;
 314         }
 315
 316         /* Establish the gpadl handle for this buffer on this
 317          * channel.  Note: This call uses the vmbus connection rather
 318          * than the channel to establish the gpadl handle.
 319          */
 320         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 321                                     net_device->send_buf_size,
 322                                     &net_device->send_buf_gpadl_handle);
 323         if (ret != 0) {
 324                 netdev_err(ndev,
 325                            "unable to establish send buffer's gpadl\n");
 326                 goto cleanup;
 327         }
 328
 329         /* Notify the NetVsp of the gpadl handle */
 330         init_packet = &net_device->channel_init_pkt;
 331         memset(init_packet, 0, sizeof(struct nvsp_message));
 332         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 333         init_packet->msg.v1_msg.send_recv_buf.gpadl_handle =
 334                 net_device->send_buf_gpadl_handle;
 335         init_packet->msg.v1_msg.send_recv_buf.id = 0;
 336
 337         /* Send the gpadl notification request */
 338         ret = vmbus_sendpacket(device->channel, init_packet,
 339                                sizeof(struct nvsp_message),
 340                                (unsigned long)init_packet,
 341                                VM_PKT_DATA_INBAND,
 342                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 343         if (ret != 0) {
 344                 netdev_err(ndev,
 345                            "unable to send send buffer's gpadl to netvsp\n");
 346                 goto cleanup;
 347         }
 348
 349         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 350         BUG_ON(t == 0);
 351
 352         /* Check the response */
 353         if (init_packet->msg.v1_msg.
 354             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 355                 netdev_err(ndev, "Unable to complete send buffer "
 356                            "initialization with NetVsp - status %d\n",
 357                            init_packet->msg.v1_msg.
 358                            send_recv_buf_complete.status);
 359                 ret = -EINVAL;
 360                 goto cleanup;
 361         }
 362
 363         /* Parse the response */
 364         net_device->send_section_size = init_packet->msg.
 365                                 v1_msg.send_send_buf_complete.section_size;
 366
 367         /* Section count is simply the size divided by the section size.
 368          */
 369         net_device->send_section_cnt =
 370                 net_device->send_buf_size/net_device->send_section_size;
 371
 372         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 373                  net_device->send_section_size, net_device->send_section_cnt);
 374
 375         /* Setup state for managing the send buffer. */
 376         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 377                                              BITS_PER_LONG);
 378
 379         net_device->send_section_map =
 380                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 381         if (net_device->send_section_map == NULL) {
 382                 ret = -ENOMEM;
 383                 goto cleanup;
 384         }
 385
 386         goto exit;
 387
 388 cleanup:
 389         netvsc_destroy_buf(net_device);
 390
 391 exit:
 392         return ret;
 393 }
 394
 395
 396 /* Negotiate NVSP protocol version */
 397 static int negotiate_nvsp_ver(struct hv_device *device,
 398                               struct netvsc_device *net_device,
 399                               struct nvsp_message *init_packet,
 400                               u32 nvsp_ver)
 401 {
 402         int ret, t;
 403
 404         memset(init_packet, 0, sizeof(struct nvsp_message));
 405         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 406         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 407         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 408
 409         /* Send the init request */
 410         ret = vmbus_sendpacket(device->channel, init_packet,
 411                                sizeof(struct nvsp_message),
 412                                (unsigned long)init_packet,
 413                                VM_PKT_DATA_INBAND,
 414                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 415
 416         if (ret != 0)
 417                 return ret;
 418
 419         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 420
 421         if (t == 0)
 422                 return -ETIMEDOUT;
 423
 424         if (init_packet->msg.init_msg.init_complete.status !=
 425             NVSP_STAT_SUCCESS)
 426                 return -EINVAL;
 427
 428         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 429                 return 0;
 430
 431         /* NVSPv2 only: Send NDIS config */
 432         memset(init_packet, 0, sizeof(struct nvsp_message));
 433         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 434         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
 435         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 436
 437         ret = vmbus_sendpacket(device->channel, init_packet,
 438                                 sizeof(struct nvsp_message),
 439                                 (unsigned long)init_packet,
 440                                 VM_PKT_DATA_INBAND, 0);
 441
 442         return ret;
 443 }
 444
 445 static int netvsc_connect_vsp(struct hv_device *device)
 446 {
 447         int ret;
 448         struct netvsc_device *net_device;
 449         struct nvsp_message *init_packet;
 450         int ndis_version;
 451         struct net_device *ndev;
 452         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 453                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 454         int i, num_ver = 4; /* number of different NVSP versions */
 455
 456         net_device = get_outbound_net_device(device);
 457         if (!net_device)
 458                 return -ENODEV;
 459         ndev = net_device->ndev;
 460
 461         init_packet = &net_device->channel_init_pkt;
 462
 463         /* Negotiate the latest NVSP protocol supported */
 464         for (i = num_ver - 1; i >= 0; i--)
 465                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 466                                        ver_list[i])  == 0) {
 467                         net_device->nvsp_version = ver_list[i];
 468                         break;
 469                 }
 470
 471         if (i < 0) {
 472                 ret = -EPROTO;
 473                 goto cleanup;
 474         }
 475
 476         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 477
 478         /* Send the ndis version */
 479         memset(init_packet, 0, sizeof(struct nvsp_message));
 480
 481         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 482                 ndis_version = 0x00060001;
 483         else
 484                 ndis_version = 0x0006001e;
 485
 486         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 487         init_packet->msg.v1_msg.
 488                 send_ndis_ver.ndis_major_ver =
 489                                 (ndis_version & 0xFFFF0000) >> 16;
 490         init_packet->msg.v1_msg.
 491                 send_ndis_ver.ndis_minor_ver =
 492                                 ndis_version & 0xFFFF;
 493
 494         /* Send the init request */
 495         ret = vmbus_sendpacket(device->channel, init_packet,
 496                                 sizeof(struct nvsp_message),
 497                                 (unsigned long)init_packet,
 498                                 VM_PKT_DATA_INBAND, 0);
 499         if (ret != 0)
 500                 goto cleanup;
 501
 502         /* Post the big receive buffer to NetVSP */
 503         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 504                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 505         else
 506                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 507         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 508
 509         ret = netvsc_init_buf(device);
 510
 511 cleanup:
 512         return ret;
 513 }
 514
 515 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 516 {
 517         netvsc_destroy_buf(net_device);
 518 }
 519
 520 /*
 521  * netvsc_device_remove - Callback when the root bus device is removed
 522  */
 523 int netvsc_device_remove(struct hv_device *device)
 524 {
 525         struct netvsc_device *net_device;
 526         unsigned long flags;
 527
 528         net_device = hv_get_drvdata(device);
 529
 530         netvsc_disconnect_vsp(net_device);
 531
 532         /*
 533          * Since we have already drained, we don't need to busy wait
 534          * as was done in final_release_stor_device()
 535          * Note that we cannot set the ext pointer to NULL until
 536          * we have drained - to drain the outgoing packets, we need to
 537          * allow incoming packets.
 538          */
 539
 540         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 541         hv_set_drvdata(device, NULL);
 542         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 543
 544         /*
 545          * At this point, no one should be accessing net_device
 546          * except in here
 547          */
 548         dev_notice(&device->device, "net device safe to remove\n");
 549
 550         /* Now, we can close the channel safely */
 551         vmbus_close(device->channel);
 552
 553         /* Release all resources */
 554         if (net_device->sub_cb_buf)
 555                 vfree(net_device->sub_cb_buf);
 556
 557         kfree(net_device);
 558         return 0;
 559 }
 560
 561
 562 #define RING_AVAIL_PERCENT_HIWATER 20
 563 #define RING_AVAIL_PERCENT_LOWATER 10
 564
 565 /*
 566  * Get the percentage of available bytes to write in the ring.
 567  * The return value is in range from 0 to 100.
 568  */
 569 static inline u32 hv_ringbuf_avail_percent(
 570                 struct hv_ring_buffer_info *ring_info)
 571 {
 572         u32 avail_read, avail_write;
 573
 574         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 575
 576         return avail_write * 100 / ring_info->ring_datasize;
 577 }
 578
 579 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 580                                          u32 index)
 581 {
 582         sync_change_bit(index, net_device->send_section_map);
 583 }
 584
 585 static void netvsc_send_completion(struct netvsc_device *net_device,
 586                                    struct hv_device *device,
 587                                    struct vmpacket_descriptor *packet)
 588 {
 589         struct nvsp_message *nvsp_packet;
 590         struct hv_netvsc_packet *nvsc_packet;
 591         struct net_device *ndev;
 592         u32 send_index;
 593
 594         ndev = net_device->ndev;
 595
 596         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 597                         (packet->offset8 << 3));
 598
 599         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 600             (nvsp_packet->hdr.msg_type ==
 601              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 602             (nvsp_packet->hdr.msg_type ==
 603              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
 604             (nvsp_packet->hdr.msg_type ==
 605              NVSP_MSG5_TYPE_SUBCHANNEL)) {
 606                 /* Copy the response back */
 607                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 608                        sizeof(struct nvsp_message));
 609                 complete(&net_device->channel_init_wait);
 610         } else if (nvsp_packet->hdr.msg_type ==
 611                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 612                 int num_outstanding_sends;
 613                 u16 q_idx = 0;
 614                 struct vmbus_channel *channel = device->channel;
 615                 int queue_sends;
 616
 617                 /* Get the send context */
 618                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 619                         packet->trans_id;
 620
 621                 /* Notify the layer above us */
 622                 if (nvsc_packet) {
 623                         send_index = nvsc_packet->send_buf_index;
 624                         if (send_index != NETVSC_INVALID_INDEX)
 625                                 netvsc_free_send_slot(net_device, send_index);
 626                         q_idx = nvsc_packet->q_idx;
 627                         channel = nvsc_packet->channel;
 628                         nvsc_packet->send_completion(nvsc_packet->
 629                                                      send_completion_ctx);
 630                 }
 631
 632                 num_outstanding_sends =
 633                         atomic_dec_return(&net_device->num_outstanding_sends);
 634                 queue_sends = atomic_dec_return(&net_device->
 635                                                 queue_sends[q_idx]);
 636
 637                 if (net_device->destroy && num_outstanding_sends == 0)
 638                         wake_up(&net_device->wait_drain);
 639
 640                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 641                     !net_device->start_remove &&
 642                     (hv_ringbuf_avail_percent(&channel->outbound) >
 643                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 644                                 netif_tx_wake_queue(netdev_get_tx_queue(
 645                                                     ndev, q_idx));
 646         } else {
 647                 netdev_err(ndev, "Unknown send completion packet type- "
 648                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 649         }
 650
 651 }
 652
 653 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 654 {
 655         unsigned long index;
 656         u32 max_words = net_device->map_words;
 657         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 658         u32 section_cnt = net_device->send_section_cnt;
 659         int ret_val = NETVSC_INVALID_INDEX;
 660         int i;
 661         int prev_val;
 662
 663         for (i = 0; i < max_words; i++) {
 664                 if (!~(map_addr[i]))
 665                         continue;
 666                 index = ffz(map_addr[i]);
 667                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 668                 if (prev_val)
 669                         continue;
 670                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 671                         break;
 672                 ret_val = (index + (i * BITS_PER_LONG));
 673                 break;
 674         }
 675         return ret_val;
 676 }
 677
 678 u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 679                             unsigned int section_index,
 680                             struct hv_netvsc_packet *packet)
 681 {
 682         char *start = net_device->send_buf;
 683         char *dest = (start + (section_index * net_device->send_section_size));
 684         int i;
 685         u32 msg_size = 0;
 686
 687         for (i = 0; i < packet->page_buf_cnt; i++) {
 688                 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
 689                 u32 offset = packet->page_buf[i].offset;
 690                 u32 len = packet->page_buf[i].len;
 691
 692                 memcpy(dest, (src + offset), len);
 693                 msg_size += len;
 694                 dest += len;
 695         }
 696         return msg_size;
 697 }
 698
 699 int netvsc_send(struct hv_device *device,
 700                         struct hv_netvsc_packet *packet)
 701 {
 702         struct netvsc_device *net_device;
 703         int ret = 0;
 704         struct nvsp_message sendMessage;
 705         struct net_device *ndev;
 706         struct vmbus_channel *out_channel = NULL;
 707         u64 req_id;
 708         unsigned int section_index = NETVSC_INVALID_INDEX;
 709         u32 msg_size = 0;
 710         struct sk_buff *skb;
 711
 712
 713         net_device = get_outbound_net_device(device);
 714         if (!net_device)
 715                 return -ENODEV;
 716         ndev = net_device->ndev;
 717
 718         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 719         if (packet->is_data_pkt) {
 720                 /* 0 is RMC_DATA; */
 721                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 722         } else {
 723                 /* 1 is RMC_CONTROL; */
 724                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 725         }
 726
 727         /* Attempt to send via sendbuf */
 728         if (packet->total_data_buflen < net_device->send_section_size) {
 729                 section_index = netvsc_get_next_send_section(net_device);
 730                 if (section_index != NETVSC_INVALID_INDEX) {
 731                         msg_size = netvsc_copy_to_send_buf(net_device,
 732                                                            section_index,
 733                                                            packet);
 734                         skb = (struct sk_buff *)
 735                               (unsigned long)packet->send_completion_tid;
 736                         if (skb)
 737                                 dev_kfree_skb_any(skb);
 738                         packet->page_buf_cnt = 0;
 739                 }
 740         }
 741         packet->send_buf_index = section_index;
 742
 743
 744         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 745                 section_index;
 746         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = msg_size;
 747
 748         if (packet->send_completion)
 749                 req_id = (ulong)packet;
 750         else
 751                 req_id = 0;
 752
 753         out_channel = net_device->chn_table[packet->q_idx];
 754         if (out_channel == NULL)
 755                 out_channel = device->channel;
 756         packet->channel = out_channel;
 757
 758         if (packet->page_buf_cnt) {
 759                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 760                                                   packet->page_buf,
 761                                                   packet->page_buf_cnt,
 762                                                   &sendMessage,
 763                                                   sizeof(struct nvsp_message),
 764                                                   req_id);
 765         } else {
 766                 ret = vmbus_sendpacket(out_channel, &sendMessage,
 767                                 sizeof(struct nvsp_message),
 768                                 req_id,
 769                                 VM_PKT_DATA_INBAND,
 770                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 771         }
 772
 773         if (ret == 0) {
 774                 atomic_inc(&net_device->num_outstanding_sends);
 775                 atomic_inc(&net_device->queue_sends[packet->q_idx]);
 776
 777                 if (hv_ringbuf_avail_percent(&out_channel->outbound) <
 778                         RING_AVAIL_PERCENT_LOWATER) {
 779                         netif_tx_stop_queue(netdev_get_tx_queue(
 780                                             ndev, packet->q_idx));
 781
 782                         if (atomic_read(&net_device->
 783                                 queue_sends[packet->q_idx]) < 1)
 784                                 netif_tx_wake_queue(netdev_get_tx_queue(
 785                                                     ndev, packet->q_idx));
 786                 }
 787         } else if (ret == -EAGAIN) {
 788                 netif_tx_stop_queue(netdev_get_tx_queue(
 789                                     ndev, packet->q_idx));
 790                 if (atomic_read(&net_device->queue_sends[packet->q_idx]) < 1) {
 791                         netif_tx_wake_queue(netdev_get_tx_queue(
 792                                             ndev, packet->q_idx));
 793                         ret = -ENOSPC;
 794                 }
 795         } else {
 796                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 797                            packet, ret);
 798         }
 799
 800         return ret;
 801 }
 802
 803 static void netvsc_send_recv_completion(struct hv_device *device,
 804                                         struct vmbus_channel *channel,
 805                                         struct netvsc_device *net_device,
 806                                         u64 transaction_id, u32 status)
 807 {
 808         struct nvsp_message recvcompMessage;
 809         int retries = 0;
 810         int ret;
 811         struct net_device *ndev;
 812
 813         ndev = net_device->ndev;
 814
 815         recvcompMessage.hdr.msg_type =
 816                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 817
 818         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 819
 820 retry_send_cmplt:
 821         /* Send the completion */
 822         ret = vmbus_sendpacket(channel, &recvcompMessage,
 823                                sizeof(struct nvsp_message), transaction_id,
 824                                VM_PKT_COMP, 0);
 825         if (ret == 0) {
 826                 /* success */
 827                 /* no-op */
 828         } else if (ret == -EAGAIN) {
 829                 /* no more room...wait a bit and attempt to retry 3 times */
 830                 retries++;
 831                 netdev_err(ndev, "unable to send receive completion pkt"
 832                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 833
 834                 if (retries < 4) {
 835                         udelay(100);
 836                         goto retry_send_cmplt;
 837                 } else {
 838                         netdev_err(ndev, "unable to send receive "
 839                                 "completion pkt (tid %llx)...give up retrying\n",
 840                                 transaction_id);
 841                 }
 842         } else {
 843                 netdev_err(ndev, "unable to send receive "
 844                         "completion pkt - %llx\n", transaction_id);
 845         }
 846 }
 847
 848 static void netvsc_receive(struct netvsc_device *net_device,
 849                         struct vmbus_channel *channel,
 850                         struct hv_device *device,
 851                         struct vmpacket_descriptor *packet)
 852 {
 853         struct vmtransfer_page_packet_header *vmxferpage_packet;
 854         struct nvsp_message *nvsp_packet;
 855         struct hv_netvsc_packet nv_pkt;
 856         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
 857         u32 status = NVSP_STAT_SUCCESS;
 858         int i;
 859         int count = 0;
 860         struct net_device *ndev;
 861
 862         ndev = net_device->ndev;
 863
 864         /*
 865          * All inbound packets other than send completion should be xfer page
 866          * packet
 867          */
 868         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 869                 netdev_err(ndev, "Unknown packet type received - %d\n",
 870                            packet->type);
 871                 return;
 872         }
 873
 874         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 875                         (packet->offset8 << 3));
 876
 877         /* Make sure this is a valid nvsp packet */
 878         if (nvsp_packet->hdr.msg_type !=
 879             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 880                 netdev_err(ndev, "Unknown nvsp packet type received-"
 881                         " %d\n", nvsp_packet->hdr.msg_type);
 882                 return;
 883         }
 884
 885         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 886
 887         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 888                 netdev_err(ndev, "Invalid xfer page set id - "
 889                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
 890                            vmxferpage_packet->xfer_pageset_id);
 891                 return;
 892         }
 893
 894         count = vmxferpage_packet->range_cnt;
 895         netvsc_packet->device = device;
 896         netvsc_packet->channel = channel;
 897
 898         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 899         for (i = 0; i < count; i++) {
 900                 /* Initialize the netvsc packet */
 901                 netvsc_packet->status = NVSP_STAT_SUCCESS;
 902                 netvsc_packet->data = (void *)((unsigned long)net_device->
 903                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
 904                 netvsc_packet->total_data_buflen =
 905                                         vmxferpage_packet->ranges[i].byte_count;
 906
 907                 /* Pass it to the upper layer */
 908                 rndis_filter_receive(device, netvsc_packet);
 909
 910                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
 911                         status = NVSP_STAT_FAIL;
 912         }
 913
 914         netvsc_send_recv_completion(device, channel, net_device,
 915                                     vmxferpage_packet->d.trans_id, status);
 916 }
 917
 918
 919 static void netvsc_send_table(struct hv_device *hdev,
 920                               struct vmpacket_descriptor *vmpkt)
 921 {
 922         struct netvsc_device *nvscdev;
 923         struct net_device *ndev;
 924         struct nvsp_message *nvmsg;
 925         int i;
 926         u32 count, *tab;
 927
 928         nvscdev = get_outbound_net_device(hdev);
 929         if (!nvscdev)
 930                 return;
 931         ndev = nvscdev->ndev;
 932
 933         nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
 934                                         (vmpkt->offset8 << 3));
 935
 936         if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
 937                 return;
 938
 939         count = nvmsg->msg.v5_msg.send_table.count;
 940         if (count != VRSS_SEND_TAB_SIZE) {
 941                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
 942                 return;
 943         }
 944
 945         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
 946                       nvmsg->msg.v5_msg.send_table.offset);
 947
 948         for (i = 0; i < count; i++)
 949                 nvscdev->send_table[i] = tab[i];
 950 }
 951
 952 void netvsc_channel_cb(void *context)
 953 {
 954         int ret;
 955         struct vmbus_channel *channel = (struct vmbus_channel *)context;
 956         struct hv_device *device;
 957         struct netvsc_device *net_device;
 958         u32 bytes_recvd;
 959         u64 request_id;
 960         struct vmpacket_descriptor *desc;
 961         unsigned char *buffer;
 962         int bufferlen = NETVSC_PACKET_SIZE;
 963         struct net_device *ndev;
 964
 965         if (channel->primary_channel != NULL)
 966                 device = channel->primary_channel->device_obj;
 967         else
 968                 device = channel->device_obj;
 969
 970         net_device = get_inbound_net_device(device);
 971         if (!net_device)
 972                 return;
 973         ndev = net_device->ndev;
 974         buffer = get_per_channel_state(channel);
 975
 976         do {
 977                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
 978                                            &bytes_recvd, &request_id);
 979                 if (ret == 0) {
 980                         if (bytes_recvd > 0) {
 981                                 desc = (struct vmpacket_descriptor *)buffer;
 982                                 switch (desc->type) {
 983                                 case VM_PKT_COMP:
 984                                         netvsc_send_completion(net_device,
 985                                                                 device, desc);
 986                                         break;
 987
 988                                 case VM_PKT_DATA_USING_XFER_PAGES:
 989                                         netvsc_receive(net_device, channel,
 990                                                        device, desc);
 991                                         break;
 992
 993                                 case VM_PKT_DATA_INBAND:
 994                                         netvsc_send_table(device, desc);
 995                                         break;
 996
 997                                 default:
 998                                         netdev_err(ndev,
 999                                                    "unhandled packet type %d, "
1000                                                    "tid %llx len %d\n",
1001                                                    desc->type, request_id,
1002                                                    bytes_recvd);
1003                                         break;
1004                                 }
1005
1006                         } else {
1007                                 /*
1008                                  * We are done for this pass.
1009                                  */
1010                                 break;
1011                         }
1012
1013                 } else if (ret == -ENOBUFS) {
1014                         if (bufferlen > NETVSC_PACKET_SIZE)
1015                                 kfree(buffer);
1016                         /* Handle large packet */
1017                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1018                         if (buffer == NULL) {
1019                                 /* Try again next time around */
1020                                 netdev_err(ndev,
1021                                            "unable to allocate buffer of size "
1022                                            "(%d)!!\n", bytes_recvd);
1023                                 break;
1024                         }
1025
1026                         bufferlen = bytes_recvd;
1027                 }
1028         } while (1);
1029
1030         if (bufferlen > NETVSC_PACKET_SIZE)
1031                 kfree(buffer);
1032         return;
1033 }
1034
1035 /*
1036  * netvsc_device_add - Callback when the device belonging to this
1037  * driver is added
1038  */
1039 int netvsc_device_add(struct hv_device *device, void *additional_info)
1040 {
1041         int ret = 0;
1042         int ring_size =
1043         ((struct netvsc_device_info *)additional_info)->ring_size;
1044         struct netvsc_device *net_device;
1045         struct net_device *ndev;
1046
1047         net_device = alloc_net_device(device);
1048         if (!net_device) {
1049                 ret = -ENOMEM;
1050                 goto cleanup;
1051         }
1052
1053         net_device->ring_size = ring_size;
1054
1055         /*
1056          * Coming into this function, struct net_device * is
1057          * registered as the driver private data.
1058          * In alloc_net_device(), we register struct netvsc_device *
1059          * as the driver private data and stash away struct net_device *
1060          * in struct netvsc_device *.
1061          */
1062         ndev = net_device->ndev;
1063
1064         /* Initialize the NetVSC channel extension */
1065         init_completion(&net_device->channel_init_wait);
1066
1067         set_per_channel_state(device->channel, net_device->cb_buffer);
1068
1069         /* Open the channel */
1070         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1071                          ring_size * PAGE_SIZE, NULL, 0,
1072                          netvsc_channel_cb, device->channel);
1073
1074         if (ret != 0) {
1075                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1076                 goto cleanup;
1077         }
1078
1079         /* Channel is opened */
1080         pr_info("hv_netvsc channel opened successfully\n");
1081
1082         net_device->chn_table[0] = device->channel;
1083
1084         /* Connect with the NetVsp */
1085         ret = netvsc_connect_vsp(device);
1086         if (ret != 0) {
1087                 netdev_err(ndev,
1088                         "unable to connect to NetVSP - %d\n", ret);
1089                 goto close;
1090         }
1091
1092         return ret;
1093
1094 close:
1095         /* Now, we can close the channel safely */
1096         vmbus_close(device->channel);
1097
1098 cleanup:
1099
1100         if (net_device)
1101                 kfree(net_device);
1102
1103         return ret;
1104 }