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
  42         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  43         if (!net_device)
  44                 return NULL;
  45
  46         net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
  47         if (!net_device->cb_buffer) {
  48                 kfree(net_device);
  49                 return NULL;
  50         }
  51
  52         init_waitqueue_head(&net_device->wait_drain);
  53         net_device->start_remove = false;
  54         net_device->destroy = false;
  55         net_device->dev = device;
  56         net_device->ndev = ndev;
  57         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
  58         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
  59
  60         hv_set_drvdata(device, net_device);
  61         return net_device;
  62 }
  63
  64 static void free_netvsc_device(struct netvsc_device *nvdev)
  65 {
  66         kfree(nvdev->cb_buffer);
  67         kfree(nvdev);
  68 }
  69
  70 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
  71 {
  72         struct netvsc_device *net_device;
  73
  74         net_device = hv_get_drvdata(device);
  75         if (net_device && net_device->destroy)
  76                 net_device = NULL;
  77
  78         return net_device;
  79 }
  80
  81 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
  82 {
  83         struct netvsc_device *net_device;
  84
  85         net_device = hv_get_drvdata(device);
  86
  87         if (!net_device)
  88                 goto get_in_err;
  89
  90         if (net_device->destroy &&
  91                 atomic_read(&net_device->num_outstanding_sends) == 0)
  92                 net_device = NULL;
  93
  94 get_in_err:
  95         return net_device;
  96 }
  97
  98
  99 static int netvsc_destroy_buf(struct netvsc_device *net_device)
 100 {
 101         struct nvsp_message *revoke_packet;
 102         int ret = 0;
 103         struct net_device *ndev = net_device->ndev;
 104
 105         /*
 106          * If we got a section count, it means we received a
 107          * SendReceiveBufferComplete msg (ie sent
 108          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 109          * to send a revoke msg here
 110          */
 111         if (net_device->recv_section_cnt) {
 112                 /* Send the revoke receive buffer */
 113                 revoke_packet = &net_device->revoke_packet;
 114                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 115
 116                 revoke_packet->hdr.msg_type =
 117                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 118                 revoke_packet->msg.v1_msg.
 119                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 120
 121                 ret = vmbus_sendpacket(net_device->dev->channel,
 122                                        revoke_packet,
 123                                        sizeof(struct nvsp_message),
 124                                        (unsigned long)revoke_packet,
 125                                        VM_PKT_DATA_INBAND, 0);
 126                 /*
 127                  * If we failed here, we might as well return and
 128                  * have a leak rather than continue and a bugchk
 129                  */
 130                 if (ret != 0) {
 131                         netdev_err(ndev, "unable to send "
 132                                 "revoke receive buffer to netvsp\n");
 133                         return ret;
 134                 }
 135         }
 136
 137         /* Teardown the gpadl on the vsp end */
 138         if (net_device->recv_buf_gpadl_handle) {
 139                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 140                            net_device->recv_buf_gpadl_handle);
 141
 142                 /* If we failed here, we might as well return and have a leak
 143                  * rather than continue and a bugchk
 144                  */
 145                 if (ret != 0) {
 146                         netdev_err(ndev,
 147                                    "unable to teardown receive buffer's gpadl\n");
 148                         return ret;
 149                 }
 150                 net_device->recv_buf_gpadl_handle = 0;
 151         }
 152
 153         if (net_device->recv_buf) {
 154                 /* Free up the receive buffer */
 155                 vfree(net_device->recv_buf);
 156                 net_device->recv_buf = NULL;
 157         }
 158
 159         if (net_device->recv_section) {
 160                 net_device->recv_section_cnt = 0;
 161                 kfree(net_device->recv_section);
 162                 net_device->recv_section = NULL;
 163         }
 164
 165         /* Deal with the send buffer we may have setup.
 166          * If we got a  send section size, it means we received a
 167          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 168          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 169          * to send a revoke msg here
 170          */
 171         if (net_device->send_section_size) {
 172                 /* Send the revoke receive buffer */
 173                 revoke_packet = &net_device->revoke_packet;
 174                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 175
 176                 revoke_packet->hdr.msg_type =
 177                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 178                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 179                         NETVSC_SEND_BUFFER_ID;
 180
 181                 ret = vmbus_sendpacket(net_device->dev->channel,
 182                                        revoke_packet,
 183                                        sizeof(struct nvsp_message),
 184                                        (unsigned long)revoke_packet,
 185                                        VM_PKT_DATA_INBAND, 0);
 186                 /* If we failed here, we might as well return and
 187                  * have a leak rather than continue and a bugchk
 188                  */
 189                 if (ret != 0) {
 190                         netdev_err(ndev, "unable to send "
 191                                    "revoke send buffer to netvsp\n");
 192                         return ret;
 193                 }
 194         }
 195         /* Teardown the gpadl on the vsp end */
 196         if (net_device->send_buf_gpadl_handle) {
 197                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
 198                                            net_device->send_buf_gpadl_handle);
 199
 200                 /* If we failed here, we might as well return and have a leak
 201                  * rather than continue and a bugchk
 202                  */
 203                 if (ret != 0) {
 204                         netdev_err(ndev,
 205                                    "unable to teardown send buffer's gpadl\n");
 206                         return ret;
 207                 }
 208                 net_device->send_buf_gpadl_handle = 0;
 209         }
 210         if (net_device->send_buf) {
 211                 /* Free up the send buffer */
 212                 vfree(net_device->send_buf);
 213                 net_device->send_buf = NULL;
 214         }
 215         kfree(net_device->send_section_map);
 216
 217         return ret;
 218 }
 219
 220 static int netvsc_init_buf(struct hv_device *device)
 221 {
 222         int ret = 0;
 223         unsigned long t;
 224         struct netvsc_device *net_device;
 225         struct nvsp_message *init_packet;
 226         struct net_device *ndev;
 227         int node;
 228
 229         net_device = get_outbound_net_device(device);
 230         if (!net_device)
 231                 return -ENODEV;
 232         ndev = net_device->ndev;
 233
 234         node = cpu_to_node(device->channel->target_cpu);
 235         net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
 236         if (!net_device->recv_buf)
 237                 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 238
 239         if (!net_device->recv_buf) {
 240                 netdev_err(ndev, "unable to allocate receive "
 241                         "buffer of size %d\n", net_device->recv_buf_size);
 242                 ret = -ENOMEM;
 243                 goto cleanup;
 244         }
 245
 246         /*
 247          * Establish the gpadl handle for this buffer on this
 248          * channel.  Note: This call uses the vmbus connection rather
 249          * than the channel to establish the gpadl handle.
 250          */
 251         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 252                                     net_device->recv_buf_size,
 253                                     &net_device->recv_buf_gpadl_handle);
 254         if (ret != 0) {
 255                 netdev_err(ndev,
 256                         "unable to establish receive buffer's gpadl\n");
 257                 goto cleanup;
 258         }
 259
 260
 261         /* Notify the NetVsp of the gpadl handle */
 262         init_packet = &net_device->channel_init_pkt;
 263
 264         memset(init_packet, 0, sizeof(struct nvsp_message));
 265
 266         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 267         init_packet->msg.v1_msg.send_recv_buf.
 268                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 269         init_packet->msg.v1_msg.
 270                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 271
 272         /* Send the gpadl notification request */
 273         ret = vmbus_sendpacket(device->channel, init_packet,
 274                                sizeof(struct nvsp_message),
 275                                (unsigned long)init_packet,
 276                                VM_PKT_DATA_INBAND,
 277                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 278         if (ret != 0) {
 279                 netdev_err(ndev,
 280                         "unable to send receive buffer's gpadl to netvsp\n");
 281                 goto cleanup;
 282         }
 283
 284         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 285         BUG_ON(t == 0);
 286
 287
 288         /* Check the response */
 289         if (init_packet->msg.v1_msg.
 290             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 291                 netdev_err(ndev, "Unable to complete receive buffer "
 292                            "initialization with NetVsp - status %d\n",
 293                            init_packet->msg.v1_msg.
 294                            send_recv_buf_complete.status);
 295                 ret = -EINVAL;
 296                 goto cleanup;
 297         }
 298
 299         /* Parse the response */
 300
 301         net_device->recv_section_cnt = init_packet->msg.
 302                 v1_msg.send_recv_buf_complete.num_sections;
 303
 304         net_device->recv_section = kmemdup(
 305                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 306                 net_device->recv_section_cnt *
 307                 sizeof(struct nvsp_1_receive_buffer_section),
 308                 GFP_KERNEL);
 309         if (net_device->recv_section == NULL) {
 310                 ret = -EINVAL;
 311                 goto cleanup;
 312         }
 313
 314         /*
 315          * For 1st release, there should only be 1 section that represents the
 316          * entire receive buffer
 317          */
 318         if (net_device->recv_section_cnt != 1 ||
 319             net_device->recv_section->offset != 0) {
 320                 ret = -EINVAL;
 321                 goto cleanup;
 322         }
 323
 324         /* Now setup the send buffer.
 325          */
 326         net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
 327         if (!net_device->send_buf)
 328                 net_device->send_buf = vzalloc(net_device->send_buf_size);
 329         if (!net_device->send_buf) {
 330                 netdev_err(ndev, "unable to allocate send "
 331                            "buffer of size %d\n", net_device->send_buf_size);
 332                 ret = -ENOMEM;
 333                 goto cleanup;
 334         }
 335
 336         /* Establish the gpadl handle for this buffer on this
 337          * channel.  Note: This call uses the vmbus connection rather
 338          * than the channel to establish the gpadl handle.
 339          */
 340         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 341                                     net_device->send_buf_size,
 342                                     &net_device->send_buf_gpadl_handle);
 343         if (ret != 0) {
 344                 netdev_err(ndev,
 345                            "unable to establish send buffer's gpadl\n");
 346                 goto cleanup;
 347         }
 348
 349         /* Notify the NetVsp of the gpadl handle */
 350         init_packet = &net_device->channel_init_pkt;
 351         memset(init_packet, 0, sizeof(struct nvsp_message));
 352         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 353         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 354                 net_device->send_buf_gpadl_handle;
 355         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 356
 357         /* Send the gpadl notification request */
 358         ret = vmbus_sendpacket(device->channel, init_packet,
 359                                sizeof(struct nvsp_message),
 360                                (unsigned long)init_packet,
 361                                VM_PKT_DATA_INBAND,
 362                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 363         if (ret != 0) {
 364                 netdev_err(ndev,
 365                            "unable to send send buffer's gpadl to netvsp\n");
 366                 goto cleanup;
 367         }
 368
 369         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 370         BUG_ON(t == 0);
 371
 372         /* Check the response */
 373         if (init_packet->msg.v1_msg.
 374             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 375                 netdev_err(ndev, "Unable to complete send buffer "
 376                            "initialization with NetVsp - status %d\n",
 377                            init_packet->msg.v1_msg.
 378                            send_send_buf_complete.status);
 379                 ret = -EINVAL;
 380                 goto cleanup;
 381         }
 382
 383         /* Parse the response */
 384         net_device->send_section_size = init_packet->msg.
 385                                 v1_msg.send_send_buf_complete.section_size;
 386
 387         /* Section count is simply the size divided by the section size.
 388          */
 389         net_device->send_section_cnt =
 390                 net_device->send_buf_size/net_device->send_section_size;
 391
 392         dev_info(&device->device, "Send section size: %d, Section count:%d\n",
 393                  net_device->send_section_size, net_device->send_section_cnt);
 394
 395         /* Setup state for managing the send buffer. */
 396         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 397                                              BITS_PER_LONG);
 398
 399         net_device->send_section_map =
 400                 kzalloc(net_device->map_words * sizeof(ulong), GFP_KERNEL);
 401         if (net_device->send_section_map == NULL) {
 402                 ret = -ENOMEM;
 403                 goto cleanup;
 404         }
 405
 406         goto exit;
 407
 408 cleanup:
 409         netvsc_destroy_buf(net_device);
 410
 411 exit:
 412         return ret;
 413 }
 414
 415
 416 /* Negotiate NVSP protocol version */
 417 static int negotiate_nvsp_ver(struct hv_device *device,
 418                               struct netvsc_device *net_device,
 419                               struct nvsp_message *init_packet,
 420                               u32 nvsp_ver)
 421 {
 422         int ret;
 423         unsigned long t;
 424
 425         memset(init_packet, 0, sizeof(struct nvsp_message));
 426         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 427         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 428         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 429
 430         /* Send the init request */
 431         ret = vmbus_sendpacket(device->channel, init_packet,
 432                                sizeof(struct nvsp_message),
 433                                (unsigned long)init_packet,
 434                                VM_PKT_DATA_INBAND,
 435                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 436
 437         if (ret != 0)
 438                 return ret;
 439
 440         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
 441
 442         if (t == 0)
 443                 return -ETIMEDOUT;
 444
 445         if (init_packet->msg.init_msg.init_complete.status !=
 446             NVSP_STAT_SUCCESS)
 447                 return -EINVAL;
 448
 449         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 450                 return 0;
 451
 452         /* NVSPv2 or later: Send NDIS config */
 453         memset(init_packet, 0, sizeof(struct nvsp_message));
 454         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 455         init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu +
 456                                                        ETH_HLEN;
 457         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 458
 459         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5)
 460                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 461
 462         ret = vmbus_sendpacket(device->channel, init_packet,
 463                                 sizeof(struct nvsp_message),
 464                                 (unsigned long)init_packet,
 465                                 VM_PKT_DATA_INBAND, 0);
 466
 467         return ret;
 468 }
 469
 470 static int netvsc_connect_vsp(struct hv_device *device)
 471 {
 472         int ret;
 473         struct netvsc_device *net_device;
 474         struct nvsp_message *init_packet;
 475         int ndis_version;
 476         struct net_device *ndev;
 477         u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 478                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 479         int i, num_ver = 4; /* number of different NVSP versions */
 480
 481         net_device = get_outbound_net_device(device);
 482         if (!net_device)
 483                 return -ENODEV;
 484         ndev = net_device->ndev;
 485
 486         init_packet = &net_device->channel_init_pkt;
 487
 488         /* Negotiate the latest NVSP protocol supported */
 489         for (i = num_ver - 1; i >= 0; i--)
 490                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 491                                        ver_list[i])  == 0) {
 492                         net_device->nvsp_version = ver_list[i];
 493                         break;
 494                 }
 495
 496         if (i < 0) {
 497                 ret = -EPROTO;
 498                 goto cleanup;
 499         }
 500
 501         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 502
 503         /* Send the ndis version */
 504         memset(init_packet, 0, sizeof(struct nvsp_message));
 505
 506         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 507                 ndis_version = 0x00060001;
 508         else
 509                 ndis_version = 0x0006001e;
 510
 511         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 512         init_packet->msg.v1_msg.
 513                 send_ndis_ver.ndis_major_ver =
 514                                 (ndis_version & 0xFFFF0000) >> 16;
 515         init_packet->msg.v1_msg.
 516                 send_ndis_ver.ndis_minor_ver =
 517                                 ndis_version & 0xFFFF;
 518
 519         /* Send the init request */
 520         ret = vmbus_sendpacket(device->channel, init_packet,
 521                                 sizeof(struct nvsp_message),
 522                                 (unsigned long)init_packet,
 523                                 VM_PKT_DATA_INBAND, 0);
 524         if (ret != 0)
 525                 goto cleanup;
 526
 527         /* Post the big receive buffer to NetVSP */
 528         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 529                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 530         else
 531                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 532         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 533
 534         ret = netvsc_init_buf(device);
 535
 536 cleanup:
 537         return ret;
 538 }
 539
 540 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
 541 {
 542         netvsc_destroy_buf(net_device);
 543 }
 544
 545 /*
 546  * netvsc_device_remove - Callback when the root bus device is removed
 547  */
 548 int netvsc_device_remove(struct hv_device *device)
 549 {
 550         struct netvsc_device *net_device;
 551         unsigned long flags;
 552
 553         net_device = hv_get_drvdata(device);
 554
 555         netvsc_disconnect_vsp(net_device);
 556
 557         /*
 558          * Since we have already drained, we don't need to busy wait
 559          * as was done in final_release_stor_device()
 560          * Note that we cannot set the ext pointer to NULL until
 561          * we have drained - to drain the outgoing packets, we need to
 562          * allow incoming packets.
 563          */
 564
 565         spin_lock_irqsave(&device->channel->inbound_lock, flags);
 566         hv_set_drvdata(device, NULL);
 567         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
 568
 569         /*
 570          * At this point, no one should be accessing net_device
 571          * except in here
 572          */
 573         dev_notice(&device->device, "net device safe to remove\n");
 574
 575         /* Now, we can close the channel safely */
 576         vmbus_close(device->channel);
 577
 578         /* Release all resources */
 579         vfree(net_device->sub_cb_buf);
 580         free_netvsc_device(net_device);
 581         return 0;
 582 }
 583
 584
 585 #define RING_AVAIL_PERCENT_HIWATER 20
 586 #define RING_AVAIL_PERCENT_LOWATER 10
 587
 588 /*
 589  * Get the percentage of available bytes to write in the ring.
 590  * The return value is in range from 0 to 100.
 591  */
 592 static inline u32 hv_ringbuf_avail_percent(
 593                 struct hv_ring_buffer_info *ring_info)
 594 {
 595         u32 avail_read, avail_write;
 596
 597         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 598
 599         return avail_write * 100 / ring_info->ring_datasize;
 600 }
 601
 602 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 603                                          u32 index)
 604 {
 605         sync_change_bit(index, net_device->send_section_map);
 606 }
 607
 608 static void netvsc_send_completion(struct netvsc_device *net_device,
 609                                    struct vmbus_channel *incoming_channel,
 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 = incoming_channel;
 652                         netvsc_xmit_completion(nvsc_packet);
 653                 }
 654
 655                 num_outstanding_sends =
 656                         atomic_dec_return(&net_device->num_outstanding_sends);
 657                 queue_sends = atomic_dec_return(&net_device->
 658                                                 queue_sends[q_idx]);
 659
 660                 if (net_device->destroy && num_outstanding_sends == 0)
 661                         wake_up(&net_device->wait_drain);
 662
 663                 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 664                     !net_device->start_remove &&
 665                     (hv_ringbuf_avail_percent(&channel->outbound) >
 666                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1))
 667                                 netif_tx_wake_queue(netdev_get_tx_queue(
 668                                                     ndev, q_idx));
 669         } else {
 670                 netdev_err(ndev, "Unknown send completion packet type- "
 671                            "%d received!!\n", nvsp_packet->hdr.msg_type);
 672         }
 673
 674 }
 675
 676 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 677 {
 678         unsigned long index;
 679         u32 max_words = net_device->map_words;
 680         unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
 681         u32 section_cnt = net_device->send_section_cnt;
 682         int ret_val = NETVSC_INVALID_INDEX;
 683         int i;
 684         int prev_val;
 685
 686         for (i = 0; i < max_words; i++) {
 687                 if (!~(map_addr[i]))
 688                         continue;
 689                 index = ffz(map_addr[i]);
 690                 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
 691                 if (prev_val)
 692                         continue;
 693                 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
 694                         break;
 695                 ret_val = (index + (i * BITS_PER_LONG));
 696                 break;
 697         }
 698         return ret_val;
 699 }
 700
 701 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 702                                    unsigned int section_index,
 703                                    u32 pend_size,
 704                                    struct hv_netvsc_packet *packet,
 705                                    struct rndis_message *rndis_msg)
 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                 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         u16 q_idx = packet->q_idx;
 749         struct vmbus_channel *out_channel = net_device->chn_table[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->completion_func)
 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                 struct rndis_message *rndis_msg)
 842 {
 843         struct netvsc_device *net_device;
 844         int ret = 0, m_ret = 0;
 845         struct vmbus_channel *out_channel;
 846         u16 q_idx = packet->q_idx;
 847         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 848         unsigned int section_index = NETVSC_INVALID_INDEX;
 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
 859         packet->send_buf_index = NETVSC_INVALID_INDEX;
 860         packet->cp_partial = false;
 861
 862         msdp = &net_device->msd[q_idx];
 863
 864         /* batch packets in send buffer if possible */
 865         if (msdp->pkt)
 866                 msd_len = msdp->pkt->total_data_buflen;
 867
 868         try_batch = packet->is_data_pkt && msd_len > 0 && msdp->count <
 869                     net_device->max_pkt;
 870
 871         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 872             net_device->send_section_size) {
 873                 section_index = msdp->pkt->send_buf_index;
 874
 875         } else if (try_batch && msd_len + packet->rmsg_size <
 876                    net_device->send_section_size) {
 877                 section_index = msdp->pkt->send_buf_index;
 878                 packet->cp_partial = true;
 879
 880         } else if (packet->is_data_pkt && pktlen + net_device->pkt_align <
 881                    net_device->send_section_size) {
 882                 section_index = netvsc_get_next_send_section(net_device);
 883                 if (section_index != NETVSC_INVALID_INDEX) {
 884                                 msd_send = msdp->pkt;
 885                                 msdp->pkt = NULL;
 886                                 msdp->count = 0;
 887                                 msd_len = 0;
 888                 }
 889         }
 890
 891         if (section_index != NETVSC_INVALID_INDEX) {
 892                 netvsc_copy_to_send_buf(net_device,
 893                                         section_index, msd_len,
 894                                         packet, rndis_msg);
 895
 896                 packet->send_buf_index = section_index;
 897
 898                 if (packet->cp_partial) {
 899                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 900                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 901                 } else {
 902                         packet->page_buf_cnt = 0;
 903                         packet->total_data_buflen += msd_len;
 904                 }
 905
 906                 if (msdp->pkt)
 907                         netvsc_xmit_completion(msdp->pkt);
 908
 909                 if (packet->xmit_more && !packet->cp_partial) {
 910                         msdp->pkt = packet;
 911                         msdp->count++;
 912                 } else {
 913                         cur_send = packet;
 914                         msdp->pkt = NULL;
 915                         msdp->count = 0;
 916                 }
 917         } else {
 918                 msd_send = msdp->pkt;
 919                 msdp->pkt = NULL;
 920                 msdp->count = 0;
 921                 cur_send = packet;
 922         }
 923
 924         if (msd_send) {
 925                 m_ret = netvsc_send_pkt(msd_send, net_device);
 926
 927                 if (m_ret != 0) {
 928                         netvsc_free_send_slot(net_device,
 929                                               msd_send->send_buf_index);
 930                         netvsc_xmit_completion(msd_send);
 931                 }
 932         }
 933
 934         if (cur_send)
 935                 ret = netvsc_send_pkt(cur_send, net_device);
 936
 937         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
 938                 netvsc_free_send_slot(net_device, section_index);
 939
 940         return ret;
 941 }
 942
 943 static void netvsc_send_recv_completion(struct hv_device *device,
 944                                         struct vmbus_channel *channel,
 945                                         struct netvsc_device *net_device,
 946                                         u64 transaction_id, u32 status)
 947 {
 948         struct nvsp_message recvcompMessage;
 949         int retries = 0;
 950         int ret;
 951         struct net_device *ndev;
 952
 953         ndev = net_device->ndev;
 954
 955         recvcompMessage.hdr.msg_type =
 956                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 957
 958         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 959
 960 retry_send_cmplt:
 961         /* Send the completion */
 962         ret = vmbus_sendpacket(channel, &recvcompMessage,
 963                                sizeof(struct nvsp_message), transaction_id,
 964                                VM_PKT_COMP, 0);
 965         if (ret == 0) {
 966                 /* success */
 967                 /* no-op */
 968         } else if (ret == -EAGAIN) {
 969                 /* no more room...wait a bit and attempt to retry 3 times */
 970                 retries++;
 971                 netdev_err(ndev, "unable to send receive completion pkt"
 972                         " (tid %llx)...retrying %d\n", transaction_id, retries);
 973
 974                 if (retries < 4) {
 975                         udelay(100);
 976                         goto retry_send_cmplt;
 977                 } else {
 978                         netdev_err(ndev, "unable to send receive "
 979                                 "completion pkt (tid %llx)...give up retrying\n",
 980                                 transaction_id);
 981                 }
 982         } else {
 983                 netdev_err(ndev, "unable to send receive "
 984                         "completion pkt - %llx\n", transaction_id);
 985         }
 986 }
 987
 988 static void netvsc_receive(struct netvsc_device *net_device,
 989                         struct vmbus_channel *channel,
 990                         struct hv_device *device,
 991                         struct vmpacket_descriptor *packet)
 992 {
 993         struct vmtransfer_page_packet_header *vmxferpage_packet;
 994         struct nvsp_message *nvsp_packet;
 995         struct hv_netvsc_packet nv_pkt;
 996         struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
 997         u32 status = NVSP_STAT_SUCCESS;
 998         int i;
 999         int count = 0;
1000         struct net_device *ndev;
1001         void *data;
1002
1003         ndev = net_device->ndev;
1004
1005         /*
1006          * All inbound packets other than send completion should be xfer page
1007          * packet
1008          */
1009         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1010                 netdev_err(ndev, "Unknown packet type received - %d\n",
1011                            packet->type);
1012                 return;
1013         }
1014
1015         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1016                         (packet->offset8 << 3));
1017
1018         /* Make sure this is a valid nvsp packet */
1019         if (nvsp_packet->hdr.msg_type !=
1020             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1021                 netdev_err(ndev, "Unknown nvsp packet type received-"
1022                         " %d\n", nvsp_packet->hdr.msg_type);
1023                 return;
1024         }
1025
1026         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1027
1028         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1029                 netdev_err(ndev, "Invalid xfer page set id - "
1030                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1031                            vmxferpage_packet->xfer_pageset_id);
1032                 return;
1033         }
1034
1035         count = vmxferpage_packet->range_cnt;
1036
1037         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1038         for (i = 0; i < count; i++) {
1039                 /* Initialize the netvsc packet */
1040                 netvsc_packet->status = NVSP_STAT_SUCCESS;
1041                 data = (void *)((unsigned long)net_device->
1042                         recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1043                 netvsc_packet->total_data_buflen =
1044                                         vmxferpage_packet->ranges[i].byte_count;
1045
1046                 /* Pass it to the upper layer */
1047                 rndis_filter_receive(device, netvsc_packet, &data, channel);
1048
1049                 if (netvsc_packet->status != NVSP_STAT_SUCCESS)
1050                         status = NVSP_STAT_FAIL;
1051         }
1052
1053         netvsc_send_recv_completion(device, channel, net_device,
1054                                     vmxferpage_packet->d.trans_id, status);
1055 }
1056
1057
1058 static void netvsc_send_table(struct hv_device *hdev,
1059                               struct nvsp_message *nvmsg)
1060 {
1061         struct netvsc_device *nvscdev;
1062         struct net_device *ndev;
1063         int i;
1064         u32 count, *tab;
1065
1066         nvscdev = get_outbound_net_device(hdev);
1067         if (!nvscdev)
1068                 return;
1069         ndev = nvscdev->ndev;
1070
1071         count = nvmsg->msg.v5_msg.send_table.count;
1072         if (count != VRSS_SEND_TAB_SIZE) {
1073                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1074                 return;
1075         }
1076
1077         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1078                       nvmsg->msg.v5_msg.send_table.offset);
1079
1080         for (i = 0; i < count; i++)
1081                 nvscdev->send_table[i] = tab[i];
1082 }
1083
1084 static void netvsc_send_vf(struct netvsc_device *nvdev,
1085                            struct nvsp_message *nvmsg)
1086 {
1087         nvdev->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1088         nvdev->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1089 }
1090
1091 static inline void netvsc_receive_inband(struct hv_device *hdev,
1092                                          struct netvsc_device *nvdev,
1093                                          struct nvsp_message *nvmsg)
1094 {
1095         switch (nvmsg->hdr.msg_type) {
1096         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1097                 netvsc_send_table(hdev, nvmsg);
1098                 break;
1099
1100         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1101                 netvsc_send_vf(nvdev, nvmsg);
1102                 break;
1103         }
1104 }
1105
1106 void netvsc_channel_cb(void *context)
1107 {
1108         int ret;
1109         struct vmbus_channel *channel = (struct vmbus_channel *)context;
1110         struct hv_device *device;
1111         struct netvsc_device *net_device;
1112         u32 bytes_recvd;
1113         u64 request_id;
1114         struct vmpacket_descriptor *desc;
1115         unsigned char *buffer;
1116         int bufferlen = NETVSC_PACKET_SIZE;
1117         struct net_device *ndev;
1118         struct nvsp_message *nvmsg;
1119
1120         if (channel->primary_channel != NULL)
1121                 device = channel->primary_channel->device_obj;
1122         else
1123                 device = channel->device_obj;
1124
1125         net_device = get_inbound_net_device(device);
1126         if (!net_device)
1127                 return;
1128         ndev = net_device->ndev;
1129         buffer = get_per_channel_state(channel);
1130
1131         do {
1132                 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1133                                            &bytes_recvd, &request_id);
1134                 if (ret == 0) {
1135                         if (bytes_recvd > 0) {
1136                                 desc = (struct vmpacket_descriptor *)buffer;
1137                                 nvmsg = (struct nvsp_message *)((unsigned long)
1138                                          desc + (desc->offset8 << 3));
1139                                 switch (desc->type) {
1140                                 case VM_PKT_COMP:
1141                                         netvsc_send_completion(net_device,
1142                                                                 channel,
1143                                                                 device, desc);
1144                                         break;
1145
1146                                 case VM_PKT_DATA_USING_XFER_PAGES:
1147                                         netvsc_receive(net_device, channel,
1148                                                        device, desc);
1149                                         break;
1150
1151                                 case VM_PKT_DATA_INBAND:
1152                                         netvsc_receive_inband(device,
1153                                                               net_device,
1154                                                               nvmsg);
1155                                         break;
1156
1157                                 default:
1158                                         netdev_err(ndev,
1159                                                    "unhandled packet type %d, "
1160                                                    "tid %llx len %d\n",
1161                                                    desc->type, request_id,
1162                                                    bytes_recvd);
1163                                         break;
1164                                 }
1165
1166                         } else {
1167                                 /*
1168                                  * We are done for this pass.
1169                                  */
1170                                 break;
1171                         }
1172
1173                 } else if (ret == -ENOBUFS) {
1174                         if (bufferlen > NETVSC_PACKET_SIZE)
1175                                 kfree(buffer);
1176                         /* Handle large packet */
1177                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1178                         if (buffer == NULL) {
1179                                 /* Try again next time around */
1180                                 netdev_err(ndev,
1181                                            "unable to allocate buffer of size "
1182                                            "(%d)!!\n", bytes_recvd);
1183                                 break;
1184                         }
1185
1186                         bufferlen = bytes_recvd;
1187                 }
1188         } while (1);
1189
1190         if (bufferlen > NETVSC_PACKET_SIZE)
1191                 kfree(buffer);
1192         return;
1193 }
1194
1195 /*
1196  * netvsc_device_add - Callback when the device belonging to this
1197  * driver is added
1198  */
1199 int netvsc_device_add(struct hv_device *device, void *additional_info)
1200 {
1201         int ret = 0;
1202         int ring_size =
1203         ((struct netvsc_device_info *)additional_info)->ring_size;
1204         struct netvsc_device *net_device;
1205         struct net_device *ndev;
1206
1207         net_device = alloc_net_device(device);
1208         if (!net_device)
1209                 return -ENOMEM;
1210
1211         net_device->ring_size = ring_size;
1212
1213         /*
1214          * Coming into this function, struct net_device * is
1215          * registered as the driver private data.
1216          * In alloc_net_device(), we register struct netvsc_device *
1217          * as the driver private data and stash away struct net_device *
1218          * in struct netvsc_device *.
1219          */
1220         ndev = net_device->ndev;
1221
1222         /* Add netvsc_device context to netvsc_device */
1223         net_device->nd_ctx = netdev_priv(ndev);
1224
1225         /* Initialize the NetVSC channel extension */
1226         init_completion(&net_device->channel_init_wait);
1227
1228         set_per_channel_state(device->channel, net_device->cb_buffer);
1229
1230         /* Open the channel */
1231         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1232                          ring_size * PAGE_SIZE, NULL, 0,
1233                          netvsc_channel_cb, device->channel);
1234
1235         if (ret != 0) {
1236                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1237                 goto cleanup;
1238         }
1239
1240         /* Channel is opened */
1241         pr_info("hv_netvsc channel opened successfully\n");
1242
1243         net_device->chn_table[0] = device->channel;
1244
1245         /* Connect with the NetVsp */
1246         ret = netvsc_connect_vsp(device);
1247         if (ret != 0) {
1248                 netdev_err(ndev,
1249                         "unable to connect to NetVSP - %d\n", ret);
1250                 goto close;
1251         }
1252
1253         return ret;
1254
1255 close:
1256         /* Now, we can close the channel safely */
1257         vmbus_close(device->channel);
1258
1259 cleanup:
1260         free_netvsc_device(net_device);
1261
1262         return ret;
1263 }