drivers/net/hyperv/netvsc.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/wait.h>
  25 #include <linux/mm.h>
  26 #include <linux/delay.h>
  27 #include <linux/io.h>
  28 #include <linux/slab.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <asm/sync_bitops.h>
  32
  33 #include "hyperv_net.h"
  34
  35
  36 static struct netvsc_device *alloc_net_device(struct hv_device *device)
  37 {
  38         struct netvsc_device *net_device;
  39         struct net_device *ndev = hv_get_drvdata(device);
  40
  41         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  42         if (!net_device)
  43                 return NULL;
  44
  45         net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
  46         if (!net_device->cb_buffer) {
  47                 kfree(net_device);
  48                 return NULL;
  49         }
  50
  51         init_waitqueue_head(&net_device->wait_drain);
  52         net_device->start_remove = false;
  53         net_device->destroy = false;
  54         net_device->dev = device;
  55         net_device->ndev = ndev;
  56
  57         hv_set_drvdata(device, net_device);
  58         return net_device;
  59 }
  60
  61 static void free_netvsc_device(struct netvsc_device *nvdev)
  62 {
  63         kfree(nvdev->cb_buffer);
  64         kfree(nvdev);
  65 }
  66
  67 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  68 {
  69         struct netvsc_device *net_device;
  70
  71         net_device = hv_get_drvdata(device);
  72         if (net_device && net_device->destroy)
  73                 net_device = NULL;
  74
  75         return net_device;
  76 }
  77
  78 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  79 {
  80         struct netvsc_device *net_device;
  81
  82         net_device = hv_get_drvdata(device);
  83
  84         if (!net_device)
  85                 goto get_in_err;
  86
  87         if (net_device->destroy &&
  88                 atomic_read(&net_device->num_outstanding_sends) == 0)
  89                 net_device = NULL;
  90
  91 get_in_err:
  92         return net_device;
  93 }
  94
  95
  96 static int netvsc_destroy_buf(struct netvsc_device *net_device)
  97 {
  98         struct nvsp_message *revoke_packet;
  99         int ret = 0;
 100         struct net_device *ndev = net_device->ndev;
 101
 102         /*
 103          * If we got a section count, it means we received a
 104          * SendReceiveBufferComplete msg (ie sent
 105          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 106          * to send a revoke msg here
 107          */
 108         if (net_device->recv_section_cnt) {
 109                 /* Send the revoke receive buffer */
 110                 revoke_packet = &net_device->revoke_packet;
 111                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 112
 113                 revoke_packet->hdr.msg_type =
 114                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 115                 revoke_packet->msg.v1_msg.
 116                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 117
 118                 ret = vmbus_sendpacket(net_device->dev->channel,
 119                                        revoke_packet,
 120                                        sizeof(struct nvsp_message),
 121                                        (unsigned long)revoke_packet,
 122                                        VM_PKT_DATA_INBAND, 0);
 123                 /*
 124                  * If we failed here, we might as well return and
 125                  * have a leak rather than continue and a bugchk
 126                  */
 127                 if (ret != 0) {
 128                         netdev_err(ndev, "unable to send "
 129                                 "revoke receive buffer to netvsp\n");
 130                         return ret;
 131                 }
 132         }
 133
 134         /* Teardown the gpadl on the vsp end */
 135         if (net_device->recv_buf_gpadl_handle) {
 136                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 137                            net_device->recv_buf_gpadl_handle);
 138
 139                 /* If we failed here, we might as well return and have a leak
 140                  * rather than continue and a bugchk
 141                  */
 142                 if (ret != 0) {
 143                         netdev_err(ndev,
 144                                    "unable to teardown receive buffer's gpadl\n");
 145                         return ret;
 146                 }
 147                 net_device->recv_buf_gpadl_handle = 0;
 148         }
 149
 150         if (net_device->recv_buf) {
 151                 /* Free up the receive buffer */
 152                 vfree(net_device->recv_buf);
 153                 net_device->recv_buf = NULL;
 154         }
 155
 156         if (net_device->recv_section) {
 157                 net_device->recv_section_cnt = 0;
 158                 kfree(net_device->recv_section);
 159                 net_device->recv_section = NULL;
 160         }
 161
 162         /* Deal with the send buffer we may have setup.
 163          * If we got a  send section size, it means we received a
 164          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 165          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 166          * to send a revoke msg here
 167          */
 168         if (net_device->send_section_size) {
 169                 /* Send the revoke receive buffer */
 170                 revoke_packet = &net_device->revoke_packet;
 171                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 172
 173                 revoke_packet->hdr.msg_type =
 174                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 175                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 176                         NETVSC_SEND_BUFFER_ID;
 177
 178                 ret = vmbus_sendpacket(net_device->dev->channel,
 179                                        revoke_packet,
 180                                        sizeof(struct nvsp_message),
 181                                        (unsigned long)revoke_packet,
 182                                        VM_PKT_DATA_INBAND, 0);
 183                 /* If we failed here, we might as well return and
 184                  * have a leak rather than continue and a bugchk
 185                  */
 186                 if (ret != 0) {
 187                         netdev_err(ndev, "unable to send "
 188                                    "revoke send buffer to netvsp\n");
 189                         return ret;
 190                 }
 191         }
 192         /* Teardown the gpadl on the vsp end */
 193         if (net_device->send_buf_gpadl_handle) {
 194                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 195                                            net_device->send_buf_gpadl_handle);
 196
 197                 /* If we failed here, we might as well return and have a leak
 198                  * rather than continue and a bugchk
 199                  */
 200                 if (ret != 0) {
 201                         netdev_err(ndev,
 202                                    "unable to teardown send buffer's gpadl\n");
 203                         return ret;
 204                 }
 205                 net_device->send_buf_gpadl_handle = 0;
 206         }
 207         if (net_device->send_buf) {
 208                 /* Free up the send buffer */
 209                 vfree(net_device->send_buf);
 210                 net_device->send_buf = NULL;
 211         }
 212         kfree(net_device->send_section_map);
 213
 214         return ret;
 215 }
 216
 217 static int netvsc_init_buf(struct hv_device *device)
 218 {
 219         int ret = 0;
 220         unsigned long t;
 221         struct netvsc_device *net_device;
 222         struct nvsp_message *init_packet;
 223         struct net_device *ndev;
 224
 225         net_device = get_outbound_net_device(device);
 226         if (!net_device)
 227                 return -ENODEV;
 228         ndev = net_device->ndev;
 229
 230         net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 231         if (!net_device->recv_buf) {
 232                 netdev_err(ndev, "unable to allocate receive "
 233                         "buffer of size %d\n", net_device->recv_buf_size);
 234                 ret = -ENOMEM;
 235                 goto cleanup;
 236         }
 237
 238         /*
 239          * Establish the gpadl handle for this buffer on this
 240          * channel.  Note: This call uses the vmbus connection rather
 241          * than the channel to establish the gpadl handle.
 242          */
 243         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 244                                     net_device->recv_buf_size,
 245                                     &net_device->recv_buf_gpadl_handle);
 246         if (ret != 0) {
 247                 netdev_err(ndev,
 248                         "unable to establish receive buffer's gpadl\n");
 249                 goto cleanup;
 250         }
 251
 252
 253         /* Notify the NetVsp of the gpadl handle */
 254         init_packet = &net_device->channel_init_pkt;
 255
 256         memset(init_packet, 0, sizeof(struct nvsp_message));
 257
 258         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 259         init_packet->msg.v1_msg.send_recv_buf.
 260                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 261         init_packet->msg.v1_msg.
 262                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 263
 264         /* Send the gpadl notification request */
 265         ret = vmbus_sendpacket(device->channel, init_packet,
 266                                sizeof(struct nvsp_message),
 267                                (unsigned long)init_packet,
 268                                VM_PKT_DATA_INBAND,
 269                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 270         if (ret != 0) {
 271                 netdev_err(ndev,
 272                         "unable to send receive buffer's gpadl to netvsp\n");
 273                 goto cleanup;
 274         }
 275
 276         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 277         BUG_ON(t == 0);
 278
 279
 280         /* Check the response */
 281         if (init_packet->msg.v1_msg.
 282             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 283                 netdev_err(ndev, "Unable to complete receive buffer "
 284                            "initialization with NetVsp - status %d\n",
 285                            init_packet->msg.v1_msg.
 286                            send_recv_buf_complete.status);
 287                 ret = -EINVAL;
 288                 goto cleanup;
 289         }
 290
 291         /* Parse the response */
 292
 293         net_device->recv_section_cnt = init_packet->msg.
 294                 v1_msg.send_recv_buf_complete.num_sections;
 295
 296         net_device->recv_section = kmemdup(
 297                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 298                 net_device->recv_section_cnt *
 299                 sizeof(struct nvsp_1_receive_buffer_section),
 300                 GFP_KERNEL);
 301         if (net_device->recv_section == NULL) {
 302                 ret = -EINVAL;
 303                 goto cleanup;
 304         }
 305
 306         /*
 307          * For 1st release, there should only be 1 section that represents the
 308          * entire receive buffer
 309          */
 310         if (net_device->recv_section_cnt != 1 ||
 311             net_device->recv_section->offset != 0) {
 312                 ret = -EINVAL;
 313                 goto cleanup;
 314         }
 315
 316         /* Now setup the send buffer.
 317          */
 318         net_device->send_buf = vzalloc(net_device->send_buf_size);
 319         if (!net_device->send_buf) {
 320                 netdev_err(ndev, "unable to allocate send "
 321                            "buffer of size %d\n", net_device->send_buf_size);
 322                 ret = -ENOMEM;
 323                 goto cleanup;
 324         }
 325
 326         /* Establish the gpadl handle for this buffer on this
 327          * channel.  Note: This call uses the vmbus connection rather
 328          * than the channel to establish the gpadl handle.
 329          */
 330         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 331                                     net_device->send_buf_size,
 332                                     &net_device->send_buf_gpadl_handle);
 333         if (ret != 0) {
 334                 netdev_err(ndev,
 335                            "unable to establish send buffer's gpadl\n");
 336                 goto cleanup;
 337         }
 338
 339         /* Notify the NetVsp of the gpadl handle */
 340         init_packet = &net_device->channel_init_pkt;
 341         memset(init_packet, 0, sizeof(struct nvsp_message));
 342         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 343         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 344                 net_device->send_buf_gpadl_handle;
 345         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 346
 347         /* Send the gpadl notification request */
 348         ret = vmbus_sendpacket(device->channel, init_packet,
 349                                sizeof(struct nvsp_message),
 350                                (unsigned long)init_packet,
 351                                VM_PKT_DATA_INBAND,
 352                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 353         if (ret != 0) {
 354                 netdev_err(ndev,
 355                            "unable to send send buffer's gpadl to netvsp\n");
 356                 goto cleanup;
 357         }
 358
 359         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 360         BUG_ON(t == 0);
 361
 362         /* Check the response */
 363         if (init_packet->msg.v1_msg.
 364             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 365                 netdev_err(ndev, "Unable to complete send buffer "
 366                            "initialization with NetVsp - status %d\n",
 367                            init_packet->msg.v1_msg.
 368                            send_send_buf_complete.status);
 369                 ret = -EINVAL;
 370                 goto cleanup;
 371         }
 372
 373         /* Parse the response */
 374         net_device->send_section_size = init_packet->msg.
 375                                 v1_msg.send_send_buf_complete.section_size;
 376
 377         /* Section count is simply the size divided by the section size.
 378          */
 379         net_device->send_section_cnt =
 380                 net_device->send_buf_size/net_device->send_section_size;
 381
 382         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 383                  net_device->send_section_size, net_device->send_section_cnt);
 384
 385         /* Setup state for managing the send buffer. */
 386         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 387                                              BITS_PER_LONG);
 388
 389         net_device->send_section_map =
 390                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 391         if (net_device->send_section_map == NULL) {
 392                 ret = -ENOMEM;
 393                 goto cleanup;
 394         }
 395
 396         goto exit;
 397
 398 cleanup:
 399         netvsc_destroy_buf(net_device);
 400
 401 exit:
 402         return ret;
 403 }
 404
 405
 406 /* Negotiate NVSP protocol version */
 407 static int negotiate_nvsp_ver(struct hv_device *device,
 408                               struct netvsc_device *net_device,
 409                               struct nvsp_message *init_packet,
 410                               u32 nvsp_ver)
 411 {
 412         int ret;
 413         unsigned long t;
 414
 415         memset(init_packet, 0, sizeof(struct nvsp_message));
 416         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 417         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 418         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 419
 420         /* Send the init request */
 421         ret = vmbus_sendpacket(device->channel, init_packet,
 422                                sizeof(struct nvsp_message),
 423                                (unsigned long)init_packet,
 424                                VM_PKT_DATA_INBAND,
 425                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 426
 427         if (ret != 0)
 428                 return ret;
 429
 430         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 431
 432         if (t == 0)
 433                 return -ETIMEDOUT;
 434
 435         if (init_packet->msg.init_msg.init_complete.status !=
 436             NVSP_STAT_SUCCESS)
 437                 return -EINVAL;
 438
 439         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 440                 return 0;
 441
 442         /* NVSPv2 only: Send NDIS config */
 443         memset(init_packet, 0, sizeof(struct nvsp_message));
 444         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 445         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
 446                                                        ETH_HLEN;
 447         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 448
 449         ret = vmbus_sendpacket(device->channel, init_packet,
 450                                 sizeof(struct nvsp_message),
 451                                 (unsigned long)init_packet,
 452                                 VM_PKT_DATA_INBAND, 0);
 453
 454         return ret;
 455 }
 456
 457 static int netvsc_connect_vsp(struct hv_device *device)
 458 {
 459         int ret;
 460         struct netvsc_device *net_device;
 461         struct nvsp_message *init_packet;
 462         int ndis_version;
 463         struct net_device *ndev;
 464         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 465                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 466         int i, num_ver = 4; /* number of different NVSP versions */
 467
 468         net_device = get_outbound_net_device(device);
 469         if (!net_device)
 470                 return -ENODEV;
 471         ndev = net_device->ndev;
 472
 473         init_packet = &net_device->channel_init_pkt;
 474
 475         /* Negotiate the latest NVSP protocol supported */
 476         for (i = num_ver - 1; i >= 0; i--)
 477                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 478                                        ver_list[i])  == 0) {
 479                         net_device->nvsp_version = ver_list[i];
 480                         break;
 481                 }
 482
 483         if (i < 0) {
 484                 ret = -EPROTO;
 485                 goto cleanup;
 486         }
 487
 488         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 489
 490         /* Send the ndis version */
 491         memset(init_packet, 0, sizeof(struct nvsp_message));
 492
 493         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 494                 ndis_version = 0x00060001;
 495         else
 496                 ndis_version = 0x0006001e;
 497
 498         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 499         init_packet->msg.v1_msg.
 500                 send_ndis_ver.ndis_major_ver =
 501                                 (ndis_version & 0xFFFF0000) >> 16;
 502         init_packet->msg.v1_msg.
 503                 send_ndis_ver.ndis_minor_ver =
 504                                 ndis_version & 0xFFFF;
 505
 506         /* Send the init request */
 507         ret = vmbus_sendpacket(device->channel, init_packet,
 508                                 sizeof(struct nvsp_message),
 509                                 (unsigned long)init_packet,
 510                                 VM_PKT_DATA_INBAND, 0);
 511         if (ret != 0)
 512                 goto cleanup;
 513
 514         /* Post the big receive buffer to NetVSP */
 515         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 516                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 517         else
 518                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 519         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 520
 521         ret = netvsc_init_buf(device);
 522
 523 cleanup:
 524         return ret;
 525 }
 526
 527 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 528 {
 529         netvsc_destroy_buf(net_device);
 530 }
 531
 532 /*
 533  * netvsc_device_remove - Callback when the root bus device is removed
 534  */
 535 int netvsc_device_remove(struct hv_device *device)
 536 {
 537         struct netvsc_device *net_device;
 538         unsigned long flags;
 539
 540         net_device = hv_get_drvdata(device);
 541
 542         netvsc_disconnect_vsp(net_device);
 543
 544         /*
 545          * Since we have already drained, we don't need to busy wait
 546          * as was done in final_release_stor_device()
 547          * Note that we cannot set the ext pointer to NULL until
 548          * we have drained - to drain the outgoing packets, we need to
 549          * allow incoming packets.
 550          */
 551
 552         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 553         hv_set_drvdata(device, NULL);
 554         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 555
 556         /*
 557          * At this point, no one should be accessing net_device
 558          * except in here
 559          */
 560         dev_notice(&device->device, "net device safe to remove\n");
 561
 562         /* Now, we can close the channel safely */
 563         vmbus_close(device->channel);
 564
 565         /* Release all resources */
 566         vfree(net_device->sub_cb_buf);
 567         free_netvsc_device(net_device);
 568         return 0;
 569 }
 570
 571
 572 #define RING_AVAIL_PERCENT_HIWATER 20
 573 #define RING_AVAIL_PERCENT_LOWATER 10
 574
 575 /*
 576  * Get the percentage of available bytes to write in the ring.
 577  * The return value is in range from 0 to 100.
 578  */
 579 static inline u32 hv_ringbuf_avail_percent(
 580                 struct hv_ring_buffer_info *ring_info)
 581 {
 582         u32 avail_read, avail_write;
 583
 584         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 585
 586         return avail_write * 100 / ring_info->ring_datasize;
 587 }
 588
 589 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 590                                          u32 index)
 591 {
 592         sync_change_bit(index, net_device->send_section_map);
 593 }
 594
 595 static void netvsc_send_completion(struct netvsc_device *net_device,
 596                                    struct hv_device *device,
 597                                    struct vmpacket_descriptor *packet)
 598 {
 599         struct nvsp_message *nvsp_packet;
 600         struct hv_netvsc_packet *nvsc_packet;
 601         struct net_device *ndev;
 602         u32 send_index;
 603
 604         ndev = net_device->ndev;
 605
 606         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 607                         (packet->offset8 << 3));
 608
 609         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 610             (nvsp_packet->hdr.msg_type ==
 611              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 612             (nvsp_packet->hdr.msg_type ==
 613              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
 614             (nvsp_packet->hdr.msg_type ==
 615              NVSP_MSG5_TYPE_SUBCHANNEL)) {
 616                 /* Copy the response back */
 617                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 618                        sizeof(struct nvsp_message));
 619                 complete(&net_device->channel_init_wait);
 620         } else if (nvsp_packet->hdr.msg_type ==
 621                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 622                 int num_outstanding_sends;
 623                 u16 q_idx = 0;
 624                 struct vmbus_channel *channel = device->channel;
 625                 int queue_sends;
 626
 627                 /* Get the send context */
 628                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 629                         packet->trans_id;
 630
 631                 /* Notify the layer above us */
 632                 if (nvsc_packet) {
 633                         send_index = nvsc_packet->send_buf_index;
 634                         if (send_index != NETVSC_INVALID_INDEX)
 635                                 netvsc_free_send_slot(net_device, send_index);
 636                         q_idx = nvsc_packet->q_idx;
 637                         channel = nvsc_packet->channel;
 638                         nvsc_packet->send_completion(nvsc_packet->
 639                                                      send_completion_ctx);
 640                 }
 641
 642                 num_outstanding_sends =
 643                         atomic_dec_return(&net_device->num_outstanding_sends);
 644                 queue_sends = atomic_dec_return(&net_device->
 645                                                 queue_sends[q_idx]);
 646
 647                 if (net_device->destroy && num_outstanding_sends == 0)
 648                         wake_up(&net_device->wait_drain);
 649
 650                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 651                     !net_device->start_remove &&
 652                     (hv_ringbuf_avail_percent(&channel->outbound) >
 653                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 654                                 netif_tx_wake_queue(netdev_get_tx_queue(
 655                                                     ndev, q_idx));
 656         } else {
 657                 netdev_err(ndev, "Unknown send completion packet type- "
 658                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 659         }
 660
 661 }
 662
 663 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 664 {
 665         unsigned long index;
 666         u32 max_words = net_device->map_words;
 667         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 668         u32 section_cnt = net_device->send_section_cnt;
 669         int ret_val = NETVSC_INVALID_INDEX;
 670         int i;
 671         int prev_val;
 672
 673         for (i = 0; i < max_words; i++) {
 674                 if (!~(map_addr[i]))
 675                         continue;
 676                 index = ffz(map_addr[i]);
 677                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 678                 if (prev_val)
 679                         continue;
 680                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 681                         break;
 682                 ret_val = (index + (i * BITS_PER_LONG));
 683                 break;
 684         }
 685         return ret_val;
 686 }
 687
 688 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 689                                    unsigned int section_index,
 690                                    struct hv_netvsc_packet *packet)
 691 {
 692         char *start = net_device->send_buf;
 693         char *dest = (start + (section_index * net_device->send_section_size));
 694         int i;
 695         u32 msg_size = 0;
 696
 697         for (i = 0; i < packet->page_buf_cnt; i++) {
 698                 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
 699                 u32 offset = packet->page_buf[i].offset;
 700                 u32 len = packet->page_buf[i].len;
 701
 702                 memcpy(dest, (src + offset), len);
 703                 msg_size += len;
 704                 dest += len;
 705         }
 706         return msg_size;
 707 }
 708
 709 int netvsc_send(struct hv_device *device,
 710                         struct hv_netvsc_packet *packet)
 711 {
 712         struct netvsc_device *net_device;
 713         int ret = 0;
 714         struct nvsp_message sendMessage;
 715         struct net_device *ndev;
 716         struct vmbus_channel *out_channel = NULL;
 717         u64 req_id;
 718         unsigned int section_index = NETVSC_INVALID_INDEX;
 719         u32 msg_size = 0;
 720         struct sk_buff *skb = NULL;
 721         u16 q_idx = packet->q_idx;
 722
 723
 724         net_device = get_outbound_net_device(device);
 725         if (!net_device)
 726                 return -ENODEV;
 727         ndev = net_device->ndev;
 728
 729         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 730         if (packet->is_data_pkt) {
 731                 /* 0 is RMC_DATA; */
 732                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 733         } else {
 734                 /* 1 is RMC_CONTROL; */
 735                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 736         }
 737
 738         /* Attempt to send via sendbuf */
 739         if (packet->total_data_buflen < net_device->send_section_size) {
 740                 section_index = netvsc_get_next_send_section(net_device);
 741                 if (section_index != NETVSC_INVALID_INDEX) {
 742                         msg_size = netvsc_copy_to_send_buf(net_device,
 743                                                            section_index,
 744                                                            packet);
 745                         skb = (struct sk_buff *)
 746                               (unsigned long)packet->send_completion_tid;
 747                         packet->page_buf_cnt = 0;
 748                 }
 749         }
 750         packet->send_buf_index = section_index;
 751
 752
 753         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 754                 section_index;
 755         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = msg_size;
 756
 757         if (packet->send_completion)
 758                 req_id = (ulong)packet;
 759         else
 760                 req_id = 0;
 761
 762         out_channel = net_device->chn_table[packet->q_idx];
 763         if (out_channel == NULL)
 764                 out_channel = device->channel;
 765         packet->channel = out_channel;
 766
 767         if (out_channel->rescind)
 768                 return -ENODEV;
 769
 770         if (packet->page_buf_cnt) {
 771                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 772                                                   packet->page_buf,
 773                                                   packet->page_buf_cnt,
 774                                                   &sendMessage,
 775                                                   sizeof(struct nvsp_message),
 776                                                   req_id);
 777         } else {
 778                 ret = vmbus_sendpacket(out_channel, &sendMessage,
 779                                 sizeof(struct nvsp_message),
 780                                 req_id,
 781                                 VM_PKT_DATA_INBAND,
 782                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 783         }
 784
 785         if (ret == 0) {
 786                 atomic_inc(&net_device->num_outstanding_sends);
 787                 atomic_inc(&net_device->queue_sends[q_idx]);
 788
 789                 if (hv_ringbuf_avail_percent(&out_channel->outbound) <
 790                         RING_AVAIL_PERCENT_LOWATER) {
 791                         netif_tx_stop_queue(netdev_get_tx_queue(
 792                                             ndev, q_idx));
 793
 794                         if (atomic_read(&net_device->
 795                                 queue_sends[q_idx]) < 1)
 796                                 netif_tx_wake_queue(netdev_get_tx_queue(
 797                                                     ndev, q_idx));
 798                 }
 799         } else if (ret == -EAGAIN) {
 800                 netif_tx_stop_queue(netdev_get_tx_queue(
 801                                     ndev, q_idx));
 802                 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
 803                         netif_tx_wake_queue(netdev_get_tx_queue(
 804                                             ndev, q_idx));
 805                         ret = -ENOSPC;
 806                 }
 807         } else {
 808                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 809                            packet, ret);
 810         }
 811
 812         if (ret != 0) {
 813                 if (section_index != NETVSC_INVALID_INDEX)
 814                         netvsc_free_send_slot(net_device, section_index);
 815         } else if (skb) {
 816                 dev_kfree_skb_any(skb);
 817         }
 818
 819         return ret;
 820 }
 821
 822 static void netvsc_send_recv_completion(struct hv_device *device,
 823                                         struct vmbus_channel *channel,
 824                                         struct netvsc_device *net_device,
 825                                         u64 transaction_id, u32 status)
 826 {
 827         struct nvsp_message recvcompMessage;
 828         int retries = 0;
 829         int ret;
 830         struct net_device *ndev;
 831
 832         ndev = net_device->ndev;
 833
 834         recvcompMessage.hdr.msg_type =
 835                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 836
 837         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 838
 839 retry_send_cmplt:
 840         /* Send the completion */
 841         ret = vmbus_sendpacket(channel, &recvcompMessage,
 842                                sizeof(struct nvsp_message), transaction_id,
 843                                VM_PKT_COMP, 0);
 844         if (ret == 0) {
 845                 /* success */
 846                 /* no-op */
 847         } else if (ret == -EAGAIN) {
 848                 /* no more room...wait a bit and attempt to retry 3 times */
 849                 retries++;
 850                 netdev_err(ndev, "unable to send receive completion pkt"
 851                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 852
 853                 if (retries < 4) {
 854                         udelay(100);
 855                         goto retry_send_cmplt;
 856                 } else {
 857                         netdev_err(ndev, "unable to send receive "
 858                                 "completion pkt (tid %llx)...give up retrying\n",
 859                                 transaction_id);
 860                 }
 861         } else {
 862                 netdev_err(ndev, "unable to send receive "
 863                         "completion pkt - %llx\n", transaction_id);
 864         }
 865 }
 866
 867 static void netvsc_receive(struct netvsc_device *net_device,
 868                         struct vmbus_channel *channel,
 869                         struct hv_device *device,
 870                         struct vmpacket_descriptor *packet)
 871 {
 872         struct vmtransfer_page_packet_header *vmxferpage_packet;
 873         struct nvsp_message *nvsp_packet;
 874         struct hv_netvsc_packet nv_pkt;
 875         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
 876         u32 status = NVSP_STAT_SUCCESS;
 877         int i;
 878         int count = 0;
 879         struct net_device *ndev;
 880
 881         ndev = net_device->ndev;
 882
 883         /*
 884          * All inbound packets other than send completion should be xfer page
 885          * packet
 886          */
 887         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
 888                 netdev_err(ndev, "Unknown packet type received - %d\n",
 889                            packet->type);
 890                 return;
 891         }
 892
 893         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 894                         (packet->offset8 << 3));
 895
 896         /* Make sure this is a valid nvsp packet */
 897         if (nvsp_packet->hdr.msg_type !=
 898             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
 899                 netdev_err(ndev, "Unknown nvsp packet type received-"
 900                         " %d\n", nvsp_packet->hdr.msg_type);
 901                 return;
 902         }
 903
 904         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
 905
 906         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
 907                 netdev_err(ndev, "Invalid xfer page set id - "
 908                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
 909                            vmxferpage_packet->xfer_pageset_id);
 910                 return;
 911         }
 912
 913         count = vmxferpage_packet->range_cnt;
 914         netvsc_packet->device = device;
 915         netvsc_packet->channel = channel;
 916
 917         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
 918         for (i = 0; i < count; i++) {
 919                 /* Initialize the netvsc packet */
 920                 netvsc_packet->status = NVSP_STAT_SUCCESS;
 921                 netvsc_packet->data = (void *)((unsigned long)net_device->
 922                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
 923                 netvsc_packet->total_data_buflen =
 924                                         vmxferpage_packet->ranges[i].byte_count;
 925
 926                 /* Pass it to the upper layer */
 927                 rndis_filter_receive(device, netvsc_packet);
 928
 929                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
 930                         status = NVSP_STAT_FAIL;
 931         }
 932
 933         netvsc_send_recv_completion(device, channel, net_device,
 934                                     vmxferpage_packet->d.trans_id, status);
 935 }
 936
 937
 938 static void netvsc_send_table(struct hv_device *hdev,
 939                               struct vmpacket_descriptor *vmpkt)
 940 {
 941         struct netvsc_device *nvscdev;
 942         struct net_device *ndev;
 943         struct nvsp_message *nvmsg;
 944         int i;
 945         u32 count, *tab;
 946
 947         nvscdev = get_outbound_net_device(hdev);
 948         if (!nvscdev)
 949                 return;
 950         ndev = nvscdev->ndev;
 951
 952         nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
 953                                         (vmpkt->offset8 << 3));
 954
 955         if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
 956                 return;
 957
 958         count = nvmsg->msg.v5_msg.send_table.count;
 959         if (count != VRSS_SEND_TAB_SIZE) {
 960                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
 961                 return;
 962         }
 963
 964         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
 965                       nvmsg->msg.v5_msg.send_table.offset);
 966
 967         for (i = 0; i < count; i++)
 968                 nvscdev->send_table[i] = tab[i];
 969 }
 970
 971 void netvsc_channel_cb(void *context)
 972 {
 973         int ret;
 974         struct vmbus_channel *channel = (struct vmbus_channel *)context;
 975         struct hv_device *device;
 976         struct netvsc_device *net_device;
 977         u32 bytes_recvd;
 978         u64 request_id;
 979         struct vmpacket_descriptor *desc;
 980         unsigned char *buffer;
 981         int bufferlen = NETVSC_PACKET_SIZE;
 982         struct net_device *ndev;
 983
 984         if (channel->primary_channel != NULL)
 985                 device = channel->primary_channel->device_obj;
 986         else
 987                 device = channel->device_obj;
 988
 989         net_device = get_inbound_net_device(device);
 990         if (!net_device)
 991                 return;
 992         ndev = net_device->ndev;
 993         buffer = get_per_channel_state(channel);
 994
 995         do {
 996                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
 997                                            &bytes_recvd, &request_id);
 998                 if (ret == 0) {
 999                         if (bytes_recvd > 0) {
1000                                 desc = (struct vmpacket_descriptor *)buffer;
1001                                 switch (desc->type) {
1002                                 case VM_PKT_COMP:
1003                                         netvsc_send_completion(net_device,
1004                                                                 device, desc);
1005                                         break;
1006
1007                                 case VM_PKT_DATA_USING_XFER_PAGES:
1008                                         netvsc_receive(net_device, channel,
1009                                                        device, desc);
1010                                         break;
1011
1012                                 case VM_PKT_DATA_INBAND:
1013                                         netvsc_send_table(device, desc);
1014                                         break;
1015
1016                                 default:
1017                                         netdev_err(ndev,
1018                                                    "unhandled packet type %d, "
1019                                                    "tid %llx len %d\n",
1020                                                    desc->type, request_id,
1021                                                    bytes_recvd);
1022                                         break;
1023                                 }
1024
1025                         } else {
1026                                 /*
1027                                  * We are done for this pass.
1028                                  */
1029                                 break;
1030                         }
1031
1032                 } else if (ret == -ENOBUFS) {
1033                         if (bufferlen > NETVSC_PACKET_SIZE)
1034                                 kfree(buffer);
1035                         /* Handle large packet */
1036                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1037                         if (buffer == NULL) {
1038                                 /* Try again next time around */
1039                                 netdev_err(ndev,
1040                                            "unable to allocate buffer of size "
1041                                            "(%d)!!\n", bytes_recvd);
1042                                 break;
1043                         }
1044
1045                         bufferlen = bytes_recvd;
1046                 }
1047         } while (1);
1048
1049         if (bufferlen > NETVSC_PACKET_SIZE)
1050                 kfree(buffer);
1051         return;
1052 }
1053
1054 /*
1055  * netvsc_device_add - Callback when the device belonging to this
1056  * driver is added
1057  */
1058 int netvsc_device_add(struct hv_device *device, void *additional_info)
1059 {
1060         int ret = 0;
1061         int ring_size =
1062         ((struct netvsc_device_info *)additional_info)->ring_size;
1063         struct netvsc_device *net_device;
1064         struct net_device *ndev;
1065
1066         net_device = alloc_net_device(device);
1067         if (!net_device)
1068                 return -ENOMEM;
1069
1070         net_device->ring_size = ring_size;
1071
1072         /*
1073          * Coming into this function, struct net_device * is
1074          * registered as the driver private data.
1075          * In alloc_net_device(), we register struct netvsc_device *
1076          * as the driver private data and stash away struct net_device *
1077          * in struct netvsc_device *.
1078          */
1079         ndev = net_device->ndev;
1080
1081         /* Initialize the NetVSC channel extension */
1082         init_completion(&net_device->channel_init_wait);
1083
1084         set_per_channel_state(device->channel, net_device->cb_buffer);
1085
1086         /* Open the channel */
1087         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1088                          ring_size * PAGE_SIZE, NULL, 0,
1089                          netvsc_channel_cb, device->channel);
1090
1091         if (ret != 0) {
1092                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1093                 goto cleanup;
1094         }
1095
1096         /* Channel is opened */
1097         pr_info("hv_netvsc channel opened successfully\n");
1098
1099         net_device->chn_table[0] = device->channel;
1100
1101         /* Connect with the NetVsp */
1102         ret = netvsc_connect_vsp(device);
1103         if (ret != 0) {
1104                 netdev_err(ndev,
1105                         "unable to connect to NetVSP - %d\n", ret);
1106                 goto close;
1107         }
1108
1109         return ret;
1110
1111 close:
1112         /* Now, we can close the channel safely */
1113         vmbus_close(device->channel);
1114
1115 cleanup:
1116         free_netvsc_device(net_device);
1117
1118         return ret;
1119 }