drivers/net/hyperv/netvsc.c

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