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 <linux/vmalloc.h>
  32 #include <asm/sync_bitops.h>
  33
  34 #include "hyperv_net.h"
  35
  36
  37 static struct netvsc_device *alloc_net_device(struct hv_device *device)
  38 {
  39         struct netvsc_device *net_device;
  40         struct net_device *ndev = hv_get_drvdata(device);
  41         int i;
  42
  43         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  44         if (!net_device)
  45                 return NULL;
  46
  47         net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
  48         if (!net_device->cb_buffer) {
  49                 kfree(net_device);
  50                 return NULL;
  51         }
  52
  53         init_waitqueue_head(&net_device->wait_drain);
  54         net_device->start_remove = false;
  55         net_device->destroy = false;
  56         net_device->dev = device;
  57         net_device->ndev = ndev;
  58         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
  59         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
  60
  61         for (i = 0; i < num_online_cpus(); i++)
  62                 spin_lock_init(&net_device->msd[i].lock);
  63
  64         hv_set_drvdata(device, net_device);
  65         return net_device;
  66 }
  67
  68 static void free_netvsc_device(struct netvsc_device *nvdev)
  69 {
  70         kfree(nvdev->cb_buffer);
  71         kfree(nvdev);
  72 }
  73
  74 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  75 {
  76         struct netvsc_device *net_device;
  77
  78         net_device = hv_get_drvdata(device);
  79         if (net_device && net_device->destroy)
  80                 net_device = NULL;
  81
  82         return net_device;
  83 }
  84
  85 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  86 {
  87         struct netvsc_device *net_device;
  88
  89         net_device = hv_get_drvdata(device);
  90
  91         if (!net_device)
  92                 goto get_in_err;
  93
  94         if (net_device->destroy &&
  95                 atomic_read(&net_device->num_outstanding_sends) == 0)
  96                 net_device = NULL;
  97
  98 get_in_err:
  99         return net_device;
 100 }
 101
 102
 103 static int netvsc_destroy_buf(struct netvsc_device *net_device)
 104 {
 105         struct nvsp_message *revoke_packet;
 106         int ret = 0;
 107         struct net_device *ndev = net_device->ndev;
 108
 109         /*
 110          * If we got a section count, it means we received a
 111          * SendReceiveBufferComplete msg (ie sent
 112          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 113          * to send a revoke msg here
 114          */
 115         if (net_device->recv_section_cnt) {
 116                 /* Send the revoke receive buffer */
 117                 revoke_packet = &net_device->revoke_packet;
 118                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 119
 120                 revoke_packet->hdr.msg_type =
 121                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 122                 revoke_packet->msg.v1_msg.
 123                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 124
 125                 ret = vmbus_sendpacket(net_device->dev->channel,
 126                                        revoke_packet,
 127                                        sizeof(struct nvsp_message),
 128                                        (unsigned long)revoke_packet,
 129                                        VM_PKT_DATA_INBAND, 0);
 130                 /*
 131                  * If we failed here, we might as well return and
 132                  * have a leak rather than continue and a bugchk
 133                  */
 134                 if (ret != 0) {
 135                         netdev_err(ndev, "unable to send "
 136                                 "revoke receive buffer to netvsp\n");
 137                         return ret;
 138                 }
 139         }
 140
 141         /* Teardown the gpadl on the vsp end */
 142         if (net_device->recv_buf_gpadl_handle) {
 143                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 144                            net_device->recv_buf_gpadl_handle);
 145
 146                 /* If we failed here, we might as well return and have a leak
 147                  * rather than continue and a bugchk
 148                  */
 149                 if (ret != 0) {
 150                         netdev_err(ndev,
 151                                    "unable to teardown receive buffer's gpadl\n");
 152                         return ret;
 153                 }
 154                 net_device->recv_buf_gpadl_handle = 0;
 155         }
 156
 157         if (net_device->recv_buf) {
 158                 /* Free up the receive buffer */
 159                 vfree(net_device->recv_buf);
 160                 net_device->recv_buf = NULL;
 161         }
 162
 163         if (net_device->recv_section) {
 164                 net_device->recv_section_cnt = 0;
 165                 kfree(net_device->recv_section);
 166                 net_device->recv_section = NULL;
 167         }
 168
 169         /* Deal with the send buffer we may have setup.
 170          * If we got a  send section size, it means we received a
 171          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 172          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 173          * to send a revoke msg here
 174          */
 175         if (net_device->send_section_size) {
 176                 /* Send the revoke receive buffer */
 177                 revoke_packet = &net_device->revoke_packet;
 178                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 179
 180                 revoke_packet->hdr.msg_type =
 181                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 182                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 183                         NETVSC_SEND_BUFFER_ID;
 184
 185                 ret = vmbus_sendpacket(net_device->dev->channel,
 186                                        revoke_packet,
 187                                        sizeof(struct nvsp_message),
 188                                        (unsigned long)revoke_packet,
 189                                        VM_PKT_DATA_INBAND, 0);
 190                 /* If we failed here, we might as well return and
 191                  * have a leak rather than continue and a bugchk
 192                  */
 193                 if (ret != 0) {
 194                         netdev_err(ndev, "unable to send "
 195                                    "revoke send buffer to netvsp\n");
 196                         return ret;
 197                 }
 198         }
 199         /* Teardown the gpadl on the vsp end */
 200         if (net_device->send_buf_gpadl_handle) {
 201                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 202                                            net_device->send_buf_gpadl_handle);
 203
 204                 /* If we failed here, we might as well return and have a leak
 205                  * rather than continue and a bugchk
 206                  */
 207                 if (ret != 0) {
 208                         netdev_err(ndev,
 209                                    "unable to teardown send buffer's gpadl\n");
 210                         return ret;
 211                 }
 212                 net_device->send_buf_gpadl_handle = 0;
 213         }
 214         if (net_device->send_buf) {
 215                 /* Free up the send buffer */
 216                 vfree(net_device->send_buf);
 217                 net_device->send_buf = NULL;
 218         }
 219         kfree(net_device->send_section_map);
 220
 221         return ret;
 222 }
 223
 224 static int netvsc_init_buf(struct hv_device *device)
 225 {
 226         int ret = 0;
 227         unsigned long t;
 228         struct netvsc_device *net_device;
 229         struct nvsp_message *init_packet;
 230         struct net_device *ndev;
 231         int node;
 232
 233         net_device = get_outbound_net_device(device);
 234         if (!net_device)
 235                 return -ENODEV;
 236         ndev = net_device->ndev;
 237
 238         node = cpu_to_node(device->channel->target_cpu);
 239         net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
 240         if (!net_device->recv_buf)
 241                 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 242
 243         if (!net_device->recv_buf) {
 244                 netdev_err(ndev, "unable to allocate receive "
 245                         "buffer of size %d\n", net_device->recv_buf_size);
 246                 ret = -ENOMEM;
 247                 goto cleanup;
 248         }
 249
 250         /*
 251          * Establish the gpadl handle for this buffer on this
 252          * channel.  Note: This call uses the vmbus connection rather
 253          * than the channel to establish the gpadl handle.
 254          */
 255         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 256                                     net_device->recv_buf_size,
 257                                     &net_device->recv_buf_gpadl_handle);
 258         if (ret != 0) {
 259                 netdev_err(ndev,
 260                         "unable to establish receive buffer's gpadl\n");
 261                 goto cleanup;
 262         }
 263
 264
 265         /* Notify the NetVsp of the gpadl handle */
 266         init_packet = &net_device->channel_init_pkt;
 267
 268         memset(init_packet, 0, sizeof(struct nvsp_message));
 269
 270         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 271         init_packet->msg.v1_msg.send_recv_buf.
 272                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 273         init_packet->msg.v1_msg.
 274                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 275
 276         /* Send the gpadl notification request */
 277         ret = vmbus_sendpacket(device->channel, init_packet,
 278                                sizeof(struct nvsp_message),
 279                                (unsigned long)init_packet,
 280                                VM_PKT_DATA_INBAND,
 281                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 282         if (ret != 0) {
 283                 netdev_err(ndev,
 284                         "unable to send receive buffer's gpadl to netvsp\n");
 285                 goto cleanup;
 286         }
 287
 288         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 289         BUG_ON(t == 0);
 290
 291
 292         /* Check the response */
 293         if (init_packet->msg.v1_msg.
 294             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 295                 netdev_err(ndev, "Unable to complete receive buffer "
 296                            "initialization with NetVsp - status %d\n",
 297                            init_packet->msg.v1_msg.
 298                            send_recv_buf_complete.status);
 299                 ret = -EINVAL;
 300                 goto cleanup;
 301         }
 302
 303         /* Parse the response */
 304
 305         net_device->recv_section_cnt = init_packet->msg.
 306                 v1_msg.send_recv_buf_complete.num_sections;
 307
 308         net_device->recv_section = kmemdup(
 309                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 310                 net_device->recv_section_cnt *
 311                 sizeof(struct nvsp_1_receive_buffer_section),
 312                 GFP_KERNEL);
 313         if (net_device->recv_section == NULL) {
 314                 ret = -EINVAL;
 315                 goto cleanup;
 316         }
 317
 318         /*
 319          * For 1st release, there should only be 1 section that represents the
 320          * entire receive buffer
 321          */
 322         if (net_device->recv_section_cnt != 1 ||
 323             net_device->recv_section->offset != 0) {
 324                 ret = -EINVAL;
 325                 goto cleanup;
 326         }
 327
 328         /* Now setup the send buffer.
 329          */
 330         net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
 331         if (!net_device->send_buf)
 332                 net_device->send_buf = vzalloc(net_device->send_buf_size);
 333         if (!net_device->send_buf) {
 334                 netdev_err(ndev, "unable to allocate send "
 335                            "buffer of size %d\n", net_device->send_buf_size);
 336                 ret = -ENOMEM;
 337                 goto cleanup;
 338         }
 339
 340         /* Establish the gpadl handle for this buffer on this
 341          * channel.  Note: This call uses the vmbus connection rather
 342          * than the channel to establish the gpadl handle.
 343          */
 344         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 345                                     net_device->send_buf_size,
 346                                     &net_device->send_buf_gpadl_handle);
 347         if (ret != 0) {
 348                 netdev_err(ndev,
 349                            "unable to establish send buffer's gpadl\n");
 350                 goto cleanup;
 351         }
 352
 353         /* Notify the NetVsp of the gpadl handle */
 354         init_packet = &net_device->channel_init_pkt;
 355         memset(init_packet, 0, sizeof(struct nvsp_message));
 356         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 357         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 358                 net_device->send_buf_gpadl_handle;
 359         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 360
 361         /* Send the gpadl notification request */
 362         ret = vmbus_sendpacket(device->channel, init_packet,
 363                                sizeof(struct nvsp_message),
 364                                (unsigned long)init_packet,
 365                                VM_PKT_DATA_INBAND,
 366                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 367         if (ret != 0) {
 368                 netdev_err(ndev,
 369                            "unable to send send buffer's gpadl to netvsp\n");
 370                 goto cleanup;
 371         }
 372
 373         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 374         BUG_ON(t == 0);
 375
 376         /* Check the response */
 377         if (init_packet->msg.v1_msg.
 378             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 379                 netdev_err(ndev, "Unable to complete send buffer "
 380                            "initialization with NetVsp - status %d\n",
 381                            init_packet->msg.v1_msg.
 382                            send_send_buf_complete.status);
 383                 ret = -EINVAL;
 384                 goto cleanup;
 385         }
 386
 387         /* Parse the response */
 388         net_device->send_section_size = init_packet->msg.
 389                                 v1_msg.send_send_buf_complete.section_size;
 390
 391         /* Section count is simply the size divided by the section size.
 392          */
 393         net_device->send_section_cnt =
 394                 net_device->send_buf_size/net_device->send_section_size;
 395
 396         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 397                  net_device->send_section_size, net_device->send_section_cnt);
 398
 399         /* Setup state for managing the send buffer. */
 400         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 401                                              BITS_PER_LONG);
 402
 403         net_device->send_section_map =
 404                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 405         if (net_device->send_section_map == NULL) {
 406                 ret = -ENOMEM;
 407                 goto cleanup;
 408         }
 409
 410         goto exit;
 411
 412 cleanup:
 413         netvsc_destroy_buf(net_device);
 414
 415 exit:
 416         return ret;
 417 }
 418
 419
 420 /* Negotiate NVSP protocol version */
 421 static int negotiate_nvsp_ver(struct hv_device *device,
 422                               struct netvsc_device *net_device,
 423                               struct nvsp_message *init_packet,
 424                               u32 nvsp_ver)
 425 {
 426         int ret;
 427         unsigned long t;
 428
 429         memset(init_packet, 0, sizeof(struct nvsp_message));
 430         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 431         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 432         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 433
 434         /* Send the init request */
 435         ret = vmbus_sendpacket(device->channel, init_packet,
 436                                sizeof(struct nvsp_message),
 437                                (unsigned long)init_packet,
 438                                VM_PKT_DATA_INBAND,
 439                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 440
 441         if (ret != 0)
 442                 return ret;
 443
 444         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 445
 446         if (t == 0)
 447                 return -ETIMEDOUT;
 448
 449         if (init_packet->msg.init_msg.init_complete.status !=
 450             NVSP_STAT_SUCCESS)
 451                 return -EINVAL;
 452
 453         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 454                 return 0;
 455
 456         /* NVSPv2 only: Send NDIS config */
 457         memset(init_packet, 0, sizeof(struct nvsp_message));
 458         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 459         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
 460                                                        ETH_HLEN;
 461         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 462
 463         ret = vmbus_sendpacket(device->channel, init_packet,
 464                                 sizeof(struct nvsp_message),
 465                                 (unsigned long)init_packet,
 466                                 VM_PKT_DATA_INBAND, 0);
 467
 468         return ret;
 469 }
 470
 471 static int netvsc_connect_vsp(struct hv_device *device)
 472 {
 473         int ret;
 474         struct netvsc_device *net_device;
 475         struct nvsp_message *init_packet;
 476         int ndis_version;
 477         struct net_device *ndev;
 478         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 479                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 480         int i, num_ver = 4; /* number of different NVSP versions */
 481
 482         net_device = get_outbound_net_device(device);
 483         if (!net_device)
 484                 return -ENODEV;
 485         ndev = net_device->ndev;
 486
 487         init_packet = &net_device->channel_init_pkt;
 488
 489         /* Negotiate the latest NVSP protocol supported */
 490         for (i = num_ver - 1; i >= 0; i--)
 491                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 492                                        ver_list[i])  == 0) {
 493                         net_device->nvsp_version = ver_list[i];
 494                         break;
 495                 }
 496
 497         if (i < 0) {
 498                 ret = -EPROTO;
 499                 goto cleanup;
 500         }
 501
 502         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 503
 504         /* Send the ndis version */
 505         memset(init_packet, 0, sizeof(struct nvsp_message));
 506
 507         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 508                 ndis_version = 0x00060001;
 509         else
 510                 ndis_version = 0x0006001e;
 511
 512         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 513         init_packet->msg.v1_msg.
 514                 send_ndis_ver.ndis_major_ver =
 515                                 (ndis_version & 0xFFFF0000) >> 16;
 516         init_packet->msg.v1_msg.
 517                 send_ndis_ver.ndis_minor_ver =
 518                                 ndis_version & 0xFFFF;
 519
 520         /* Send the init request */
 521         ret = vmbus_sendpacket(device->channel, init_packet,
 522                                 sizeof(struct nvsp_message),
 523                                 (unsigned long)init_packet,
 524                                 VM_PKT_DATA_INBAND, 0);
 525         if (ret != 0)
 526                 goto cleanup;
 527
 528         /* Post the big receive buffer to NetVSP */
 529         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 530                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 531         else
 532                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 533         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 534
 535         ret = netvsc_init_buf(device);
 536
 537 cleanup:
 538         return ret;
 539 }
 540
 541 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 542 {
 543         netvsc_destroy_buf(net_device);
 544 }
 545
 546 /*
 547  * netvsc_device_remove - Callback when the root bus device is removed
 548  */
 549 int netvsc_device_remove(struct hv_device *device)
 550 {
 551         struct netvsc_device *net_device;
 552         unsigned long flags;
 553
 554         net_device = hv_get_drvdata(device);
 555
 556         netvsc_disconnect_vsp(net_device);
 557
 558         /*
 559          * Since we have already drained, we don't need to busy wait
 560          * as was done in final_release_stor_device()
 561          * Note that we cannot set the ext pointer to NULL until
 562          * we have drained - to drain the outgoing packets, we need to
 563          * allow incoming packets.
 564          */
 565
 566         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 567         hv_set_drvdata(device, NULL);
 568         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 569
 570         /*
 571          * At this point, no one should be accessing net_device
 572          * except in here
 573          */
 574         dev_notice(&device->device, "net device safe to remove\n");
 575
 576         /* Now, we can close the channel safely */
 577         vmbus_close(device->channel);
 578
 579         /* Release all resources */
 580         vfree(net_device->sub_cb_buf);
 581         free_netvsc_device(net_device);
 582         return 0;
 583 }
 584
 585
 586 #define RING_AVAIL_PERCENT_HIWATER 20
 587 #define RING_AVAIL_PERCENT_LOWATER 10
 588
 589 /*
 590  * Get the percentage of available bytes to write in the ring.
 591  * The return value is in range from 0 to 100.
 592  */
 593 static inline u32 hv_ringbuf_avail_percent(
 594                 struct hv_ring_buffer_info *ring_info)
 595 {
 596         u32 avail_read, avail_write;
 597
 598         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 599
 600         return avail_write * 100 / ring_info->ring_datasize;
 601 }
 602
 603 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 604                                          u32 index)
 605 {
 606         sync_change_bit(index, net_device->send_section_map);
 607 }
 608
 609 static void netvsc_send_completion(struct netvsc_device *net_device,
 610                                    struct hv_device *device,
 611                                    struct vmpacket_descriptor *packet)
 612 {
 613         struct nvsp_message *nvsp_packet;
 614         struct hv_netvsc_packet *nvsc_packet;
 615         struct net_device *ndev;
 616         u32 send_index;
 617
 618         ndev = net_device->ndev;
 619
 620         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
 621                         (packet->offset8 << 3));
 622
 623         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
 624             (nvsp_packet->hdr.msg_type ==
 625              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
 626             (nvsp_packet->hdr.msg_type ==
 627              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE) ||
 628             (nvsp_packet->hdr.msg_type ==
 629              NVSP_MSG5_TYPE_SUBCHANNEL)) {
 630                 /* Copy the response back */
 631                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 632                        sizeof(struct nvsp_message));
 633                 complete(&net_device->channel_init_wait);
 634         } else if (nvsp_packet->hdr.msg_type ==
 635                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
 636                 int num_outstanding_sends;
 637                 u16 q_idx = 0;
 638                 struct vmbus_channel *channel = device->channel;
 639                 int queue_sends;
 640
 641                 /* Get the send context */
 642                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
 643                         packet->trans_id;
 644
 645                 /* Notify the layer above us */
 646                 if (nvsc_packet) {
 647                         send_index = nvsc_packet->send_buf_index;
 648                         if (send_index != NETVSC_INVALID_INDEX)
 649                                 netvsc_free_send_slot(net_device, send_index);
 650                         q_idx = nvsc_packet->q_idx;
 651                         channel = nvsc_packet->channel;
 652                         nvsc_packet->send_completion(nvsc_packet->
 653                                                      send_completion_ctx);
 654                 }
 655
 656                 num_outstanding_sends =
 657                         atomic_dec_return(&net_device->num_outstanding_sends);
 658                 queue_sends = atomic_dec_return(&net_device->
 659                                                 queue_sends[q_idx]);
 660
 661                 if (net_device->destroy && num_outstanding_sends == 0)
 662                         wake_up(&net_device->wait_drain);
 663
 664                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 665                     !net_device->start_remove &&
 666                     (hv_ringbuf_avail_percent(&channel->outbound) >
 667                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 668                                 netif_tx_wake_queue(netdev_get_tx_queue(
 669                                                     ndev, q_idx));
 670         } else {
 671                 netdev_err(ndev, "Unknown send completion packet type- "
 672                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 673         }
 674
 675 }
 676
 677 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 678 {
 679         unsigned long index;
 680         u32 max_words = net_device->map_words;
 681         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 682         u32 section_cnt = net_device->send_section_cnt;
 683         int ret_val = NETVSC_INVALID_INDEX;
 684         int i;
 685         int prev_val;
 686
 687         for (i = 0; i < max_words; i++) {
 688                 if (!~(map_addr[i]))
 689                         continue;
 690                 index = ffz(map_addr[i]);
 691                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 692                 if (prev_val)
 693                         continue;
 694                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 695                         break;
 696                 ret_val = (index + (i * BITS_PER_LONG));
 697                 break;
 698         }
 699         return ret_val;
 700 }
 701
 702 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 703                                    unsigned int section_index,
 704                                    u32 pend_size,
 705                                    struct hv_netvsc_packet *packet)
 706 {
 707         char *start = net_device->send_buf;
 708         char *dest = start + (section_index * net_device->send_section_size)
 709                      + pend_size;
 710         int i;
 711         u32 msg_size = 0;
 712         u32 padding = 0;
 713         u32 remain = packet->total_data_buflen % net_device->pkt_align;
 714         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 715                 packet->page_buf_cnt;
 716
 717         /* Add padding */
 718         if (packet->is_data_pkt && packet->xmit_more && remain &&
 719             !packet->cp_partial) {
 720                 padding = net_device->pkt_align - remain;
 721                 packet->rndis_msg->msg_len += padding;
 722                 packet->total_data_buflen += padding;
 723         }
 724
 725         for (i = 0; i < page_count; i++) {
 726                 char *src = phys_to_virt(packet->page_buf[i].pfn << PAGE_SHIFT);
 727                 u32 offset = packet->page_buf[i].offset;
 728                 u32 len = packet->page_buf[i].len;
 729
 730                 memcpy(dest, (src + offset), len);
 731                 msg_size += len;
 732                 dest += len;
 733         }
 734
 735         if (padding) {
 736                 memset(dest, 0, padding);
 737                 msg_size += padding;
 738         }
 739
 740         return msg_size;
 741 }
 742
 743 static inline int netvsc_send_pkt(
 744         struct hv_netvsc_packet *packet,
 745         struct netvsc_device *net_device)
 746 {
 747         struct nvsp_message nvmsg;
 748         struct vmbus_channel *out_channel = packet->channel;
 749         u16 q_idx = packet->q_idx;
 750         struct net_device *ndev = net_device->ndev;
 751         u64 req_id;
 752         int ret;
 753         struct hv_page_buffer *pgbuf;
 754         u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
 755
 756         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 757         if (packet->is_data_pkt) {
 758                 /* 0 is RMC_DATA; */
 759                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 760         } else {
 761                 /* 1 is RMC_CONTROL; */
 762                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 763         }
 764
 765         nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 766                 packet->send_buf_index;
 767         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 768                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 769         else
 770                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
 771                         packet->total_data_buflen;
 772
 773         if (packet->send_completion)
 774                 req_id = (ulong)packet;
 775         else
 776                 req_id = 0;
 777
 778         if (out_channel->rescind)
 779                 return -ENODEV;
 780
 781         /*
 782          * It is possible that once we successfully place this packet
 783          * on the ringbuffer, we may stop the queue. In that case, we want
 784          * to notify the host independent of the xmit_more flag. We don't
 785          * need to be precise here; in the worst case we may signal the host
 786          * unnecessarily.
 787          */
 788         if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
 789                 packet->xmit_more = false;
 790
 791         if (packet->page_buf_cnt) {
 792                 pgbuf = packet->cp_partial ? packet->page_buf +
 793                         packet->rmsg_pgcnt : packet->page_buf;
 794                 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 795                                                       pgbuf,
 796                                                       packet->page_buf_cnt,
 797                                                       &nvmsg,
 798                                                       sizeof(struct nvsp_message),
 799                                                       req_id,
 800                                                       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
 801                                                       !packet->xmit_more);
 802         } else {
 803                 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
 804                                            sizeof(struct nvsp_message),
 805                                            req_id,
 806                                            VM_PKT_DATA_INBAND,
 807                                            VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
 808                                            !packet->xmit_more);
 809         }
 810
 811         if (ret == 0) {
 812                 atomic_inc(&net_device->num_outstanding_sends);
 813                 atomic_inc(&net_device->queue_sends[q_idx]);
 814
 815                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 816                         netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
 817
 818                         if (atomic_read(&net_device->
 819                                 queue_sends[q_idx]) < 1)
 820                                 netif_tx_wake_queue(netdev_get_tx_queue(
 821                                                     ndev, q_idx));
 822                 }
 823         } else if (ret == -EAGAIN) {
 824                 netif_tx_stop_queue(netdev_get_tx_queue(
 825                                     ndev, q_idx));
 826                 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
 827                         netif_tx_wake_queue(netdev_get_tx_queue(
 828                                             ndev, q_idx));
 829                         ret = -ENOSPC;
 830                 }
 831         } else {
 832                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 833                            packet, ret);
 834         }
 835
 836         return ret;
 837 }
 838
 839 int netvsc_send(struct hv_device *device,
 840                 struct hv_netvsc_packet *packet)
 841 {
 842         struct netvsc_device *net_device;
 843         int ret = 0, m_ret = 0;
 844         struct vmbus_channel *out_channel;
 845         u16 q_idx = packet->q_idx;
 846         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 847         unsigned int section_index = NETVSC_INVALID_INDEX;
 848         unsigned long flag;
 849         struct multi_send_data *msdp;
 850         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 851         bool try_batch;
 852
 853         net_device = get_outbound_net_device(device);
 854         if (!net_device)
 855                 return -ENODEV;
 856
 857         out_channel = net_device->chn_table[q_idx];
 858         if (!out_channel) {
 859                 out_channel = device->channel;
 860                 q_idx = 0;
 861                 packet->q_idx = 0;
 862         }
 863         packet->channel = out_channel;
 864         packet->send_buf_index = NETVSC_INVALID_INDEX;
 865         packet->cp_partial = false;
 866
 867         msdp = &net_device->msd[q_idx];
 868
 869         /* batch packets in send buffer if possible */
 870         spin_lock_irqsave(&msdp->lock, flag);
 871         if (msdp->pkt)
 872                 msd_len = msdp->pkt->total_data_buflen;
 873
 874         try_batch = packet->is_data_pkt && msd_len > 0 && msdp->count <
 875                     net_device->max_pkt;
 876
 877         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 878             net_device->send_section_size) {
 879                 section_index = msdp->pkt->send_buf_index;
 880
 881         } else if (try_batch && msd_len + packet->rmsg_size <
 882                    net_device->send_section_size) {
 883                 section_index = msdp->pkt->send_buf_index;
 884                 packet->cp_partial = true;
 885
 886         } else if (packet->is_data_pkt && pktlen + net_device->pkt_align <
 887                    net_device->send_section_size) {
 888                 section_index = netvsc_get_next_send_section(net_device);
 889                 if (section_index != NETVSC_INVALID_INDEX) {
 890                                 msd_send = msdp->pkt;
 891                                 msdp->pkt = NULL;
 892                                 msdp->count = 0;
 893                                 msd_len = 0;
 894                 }
 895         }
 896
 897         if (section_index != NETVSC_INVALID_INDEX) {
 898                 netvsc_copy_to_send_buf(net_device,
 899                                         section_index, msd_len,
 900                                         packet);
 901
 902                 packet->send_buf_index = section_index;
 903
 904                 if (packet->cp_partial) {
 905                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 906                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 907                 } else {
 908                         packet->page_buf_cnt = 0;
 909                         packet->total_data_buflen += msd_len;
 910                 }
 911
 912                 if (msdp->pkt)
 913                         netvsc_xmit_completion(msdp->pkt);
 914
 915                 if (packet->xmit_more && !packet->cp_partial) {
 916                         msdp->pkt = packet;
 917                         msdp->count++;
 918                 } else {
 919                         cur_send = packet;
 920                         msdp->pkt = NULL;
 921                         msdp->count = 0;
 922                 }
 923         } else {
 924                 msd_send = msdp->pkt;
 925                 msdp->pkt = NULL;
 926                 msdp->count = 0;
 927                 cur_send = packet;
 928         }
 929
 930         spin_unlock_irqrestore(&msdp->lock, flag);
 931
 932         if (msd_send) {
 933                 m_ret = netvsc_send_pkt(msd_send, net_device);
 934
 935                 if (m_ret != 0) {
 936                         netvsc_free_send_slot(net_device,
 937                                               msd_send->send_buf_index);
 938                         netvsc_xmit_completion(msd_send);
 939                 }
 940         }
 941
 942         if (cur_send)
 943                 ret = netvsc_send_pkt(cur_send, net_device);
 944
 945         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
 946                 netvsc_free_send_slot(net_device, section_index);
 947
 948         return ret;
 949 }
 950
 951 static void netvsc_send_recv_completion(struct hv_device *device,
 952                                         struct vmbus_channel *channel,
 953                                         struct netvsc_device *net_device,
 954                                         u64 transaction_id, u32 status)
 955 {
 956         struct nvsp_message recvcompMessage;
 957         int retries = 0;
 958         int ret;
 959         struct net_device *ndev;
 960
 961         ndev = net_device->ndev;
 962
 963         recvcompMessage.hdr.msg_type =
 964                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 965
 966         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 967
 968 retry_send_cmplt:
 969         /* Send the completion */
 970         ret = vmbus_sendpacket(channel, &recvcompMessage,
 971                                sizeof(struct nvsp_message), transaction_id,
 972                                VM_PKT_COMP, 0);
 973         if (ret == 0) {
 974                 /* success */
 975                 /* no-op */
 976         } else if (ret == -EAGAIN) {
 977                 /* no more room...wait a bit and attempt to retry 3 times */
 978                 retries++;
 979                 netdev_err(ndev, "unable to send receive completion pkt"
 980                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 981
 982                 if (retries < 4) {
 983                         udelay(100);
 984                         goto retry_send_cmplt;
 985                 } else {
 986                         netdev_err(ndev, "unable to send receive "
 987                                 "completion pkt (tid %llx)...give up retrying\n",
 988                                 transaction_id);
 989                 }
 990         } else {
 991                 netdev_err(ndev, "unable to send receive "
 992                         "completion pkt - %llx\n", transaction_id);
 993         }
 994 }
 995
 996 static void netvsc_receive(struct netvsc_device *net_device,
 997                         struct vmbus_channel *channel,
 998                         struct hv_device *device,
 999                         struct vmpacket_descriptor *packet)
1000 {
1001         struct vmtransfer_page_packet_header *vmxferpage_packet;
1002         struct nvsp_message *nvsp_packet;
1003         struct hv_netvsc_packet nv_pkt;
1004         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1005         u32 status = NVSP_STAT_SUCCESS;
1006         int i;
1007         int count = 0;
1008         struct net_device *ndev;
1009
1010         ndev = net_device->ndev;
1011
1012         /*
1013          * All inbound packets other than send completion should be xfer page
1014          * packet
1015          */
1016         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1017                 netdev_err(ndev, "Unknown packet type received - %d\n",
1018                            packet->type);
1019                 return;
1020         }
1021
1022         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1023                         (packet->offset8 << 3));
1024
1025         /* Make sure this is a valid nvsp packet */
1026         if (nvsp_packet->hdr.msg_type !=
1027             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1028                 netdev_err(ndev, "Unknown nvsp packet type received-"
1029                         " %d\n", nvsp_packet->hdr.msg_type);
1030                 return;
1031         }
1032
1033         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1034
1035         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1036                 netdev_err(ndev, "Invalid xfer page set id - "
1037                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1038                            vmxferpage_packet->xfer_pageset_id);
1039                 return;
1040         }
1041
1042         count = vmxferpage_packet->range_cnt;
1043         netvsc_packet->channel = channel;
1044
1045         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1046         for (i = 0; i < count; i++) {
1047                 /* Initialize the netvsc packet */
1048                 netvsc_packet->status = NVSP_STAT_SUCCESS;
1049                 netvsc_packet->data = (void *)((unsigned long)net_device->
1050                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1051                 netvsc_packet->total_data_buflen =
1052                                         vmxferpage_packet->ranges[i].byte_count;
1053
1054                 /* Pass it to the upper layer */
1055                 rndis_filter_receive(device, netvsc_packet);
1056
1057                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
1058                         status = NVSP_STAT_FAIL;
1059         }
1060
1061         netvsc_send_recv_completion(device, channel, net_device,
1062                                     vmxferpage_packet->d.trans_id, status);
1063 }
1064
1065
1066 static void netvsc_send_table(struct hv_device *hdev,
1067                               struct vmpacket_descriptor *vmpkt)
1068 {
1069         struct netvsc_device *nvscdev;
1070         struct net_device *ndev;
1071         struct nvsp_message *nvmsg;
1072         int i;
1073         u32 count, *tab;
1074
1075         nvscdev = get_outbound_net_device(hdev);
1076         if (!nvscdev)
1077                 return;
1078         ndev = nvscdev->ndev;
1079
1080         nvmsg = (struct nvsp_message *)((unsigned long)vmpkt +
1081                                         (vmpkt->offset8 << 3));
1082
1083         if (nvmsg->hdr.msg_type != NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE)
1084                 return;
1085
1086         count = nvmsg->msg.v5_msg.send_table.count;
1087         if (count != VRSS_SEND_TAB_SIZE) {
1088                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1089                 return;
1090         }
1091
1092         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1093                       nvmsg->msg.v5_msg.send_table.offset);
1094
1095         for (i = 0; i < count; i++)
1096                 nvscdev->send_table[i] = tab[i];
1097 }
1098
1099 void netvsc_channel_cb(void *context)
1100 {
1101         int ret;
1102         struct vmbus_channel *channel = (struct vmbus_channel *)context;
1103         struct hv_device *device;
1104         struct netvsc_device *net_device;
1105         u32 bytes_recvd;
1106         u64 request_id;
1107         struct vmpacket_descriptor *desc;
1108         unsigned char *buffer;
1109         int bufferlen = NETVSC_PACKET_SIZE;
1110         struct net_device *ndev;
1111
1112         if (channel->primary_channel != NULL)
1113                 device = channel->primary_channel->device_obj;
1114         else
1115                 device = channel->device_obj;
1116
1117         net_device = get_inbound_net_device(device);
1118         if (!net_device)
1119                 return;
1120         ndev = net_device->ndev;
1121         buffer = get_per_channel_state(channel);
1122
1123         do {
1124                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1125                                            &bytes_recvd, &request_id);
1126                 if (ret == 0) {
1127                         if (bytes_recvd > 0) {
1128                                 desc = (struct vmpacket_descriptor *)buffer;
1129                                 switch (desc->type) {
1130                                 case VM_PKT_COMP:
1131                                         netvsc_send_completion(net_device,
1132                                                                 device, desc);
1133                                         break;
1134
1135                                 case VM_PKT_DATA_USING_XFER_PAGES:
1136                                         netvsc_receive(net_device, channel,
1137                                                        device, desc);
1138                                         break;
1139
1140                                 case VM_PKT_DATA_INBAND:
1141                                         netvsc_send_table(device, desc);
1142                                         break;
1143
1144                                 default:
1145                                         netdev_err(ndev,
1146                                                    "unhandled packet type %d, "
1147                                                    "tid %llx len %d\n",
1148                                                    desc->type, request_id,
1149                                                    bytes_recvd);
1150                                         break;
1151                                 }
1152
1153                         } else {
1154                                 /*
1155                                  * We are done for this pass.
1156                                  */
1157                                 break;
1158                         }
1159
1160                 } else if (ret == -ENOBUFS) {
1161                         if (bufferlen > NETVSC_PACKET_SIZE)
1162                                 kfree(buffer);
1163                         /* Handle large packet */
1164                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1165                         if (buffer == NULL) {
1166                                 /* Try again next time around */
1167                                 netdev_err(ndev,
1168                                            "unable to allocate buffer of size "
1169                                            "(%d)!!\n", bytes_recvd);
1170                                 break;
1171                         }
1172
1173                         bufferlen = bytes_recvd;
1174                 }
1175         } while (1);
1176
1177         if (bufferlen > NETVSC_PACKET_SIZE)
1178                 kfree(buffer);
1179         return;
1180 }
1181
1182 /*
1183  * netvsc_device_add - Callback when the device belonging to this
1184  * driver is added
1185  */
1186 int netvsc_device_add(struct hv_device *device, void *additional_info)
1187 {
1188         int ret = 0;
1189         int ring_size =
1190         ((struct netvsc_device_info *)additional_info)->ring_size;
1191         struct netvsc_device *net_device;
1192         struct net_device *ndev;
1193
1194         net_device = alloc_net_device(device);
1195         if (!net_device)
1196                 return -ENOMEM;
1197
1198         net_device->ring_size = ring_size;
1199
1200         /*
1201          * Coming into this function, struct net_device * is
1202          * registered as the driver private data.
1203          * In alloc_net_device(), we register struct netvsc_device *
1204          * as the driver private data and stash away struct net_device *
1205          * in struct netvsc_device *.
1206          */
1207         ndev = net_device->ndev;
1208
1209         /* Add netvsc_device context to netvsc_device */
1210         net_device->nd_ctx = netdev_priv(ndev);
1211
1212         /* Initialize the NetVSC channel extension */
1213         init_completion(&net_device->channel_init_wait);
1214
1215         set_per_channel_state(device->channel, net_device->cb_buffer);
1216
1217         /* Open the channel */
1218         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1219                          ring_size * PAGE_SIZE, NULL, 0,
1220                          netvsc_channel_cb, device->channel);
1221
1222         if (ret != 0) {
1223                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1224                 goto cleanup;
1225         }
1226
1227         /* Channel is opened */
1228         pr_info("hv_netvsc channel opened successfully\n");
1229
1230         net_device->chn_table[0] = device->channel;
1231
1232         /* Connect with the NetVsp */
1233         ret = netvsc_connect_vsp(device);
1234         if (ret != 0) {
1235                 netdev_err(ndev,
1236                         "unable to connect to NetVSP - %d\n", ret);
1237                 goto close;
1238         }
1239
1240         return ret;
1241
1242 close:
1243         /* Now, we can close the channel safely */
1244         vmbus_close(device->channel);
1245
1246 cleanup:
1247         free_netvsc_device(net_device);
1248
1249         return ret;
1250 }