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  * Switch the data path from the synthetic interface to the VF
  38  * interface.
  39  */
  40 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  41 {
  42         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  43         struct hv_device *dev = net_device_ctx->device_ctx;
  44         struct netvsc_device *nv_dev = net_device_ctx->nvdev;
  45         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  46
  47         memset(init_pkt, 0, sizeof(struct nvsp_message));
  48         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  49         if (vf)
  50                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  51                         NVSP_DATAPATH_VF;
  52         else
  53                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  54                         NVSP_DATAPATH_SYNTHETIC;
  55
  56         vmbus_sendpacket(dev->channel, init_pkt,
  57                                sizeof(struct nvsp_message),
  58                                (unsigned long)init_pkt,
  59                                VM_PKT_DATA_INBAND, 0);
  60 }
  61
  62 static struct netvsc_device *alloc_net_device(void)
  63 {
  64         struct netvsc_device *net_device;
  65
  66         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  67         if (!net_device)
  68                 return NULL;
  69
  70         net_device->chan_table[0].mrc.buf
  71                 = vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
  72
  73         init_waitqueue_head(&net_device->wait_drain);
  74         net_device->destroy = false;
  75         atomic_set(&net_device->open_cnt, 0);
  76         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
  77         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
  78         init_completion(&net_device->channel_init_wait);
  79
  80         return net_device;
  81 }
  82
  83 static void free_netvsc_device(struct netvsc_device *nvdev)
  84 {
  85         int i;
  86
  87         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
  88                 vfree(nvdev->chan_table[i].mrc.buf);
  89
  90         kfree(nvdev);
  91 }
  92
  93
  94 static inline bool netvsc_channel_idle(const struct netvsc_device *net_device,
  95                                        u16 q_idx)
  96 {
  97         const struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
  98
  99         return atomic_read(&net_device->num_outstanding_recvs) == 0 &&
 100                 atomic_read(&nvchan->queue_sends) == 0;
 101 }
 102
 103 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
 104 {
 105         struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
 106
 107         if (net_device && net_device->destroy)
 108                 net_device = NULL;
 109
 110         return net_device;
 111 }
 112
 113 static void netvsc_destroy_buf(struct hv_device *device)
 114 {
 115         struct nvsp_message *revoke_packet;
 116         struct net_device *ndev = hv_get_drvdata(device);
 117         struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
 118         int ret;
 119
 120         /*
 121          * If we got a section count, it means we received a
 122          * SendReceiveBufferComplete msg (ie sent
 123          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 124          * to send a revoke msg here
 125          */
 126         if (net_device->recv_section_cnt) {
 127                 /* Send the revoke receive buffer */
 128                 revoke_packet = &net_device->revoke_packet;
 129                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 130
 131                 revoke_packet->hdr.msg_type =
 132                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 133                 revoke_packet->msg.v1_msg.
 134                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 135
 136                 ret = vmbus_sendpacket(device->channel,
 137                                        revoke_packet,
 138                                        sizeof(struct nvsp_message),
 139                                        (unsigned long)revoke_packet,
 140                                        VM_PKT_DATA_INBAND, 0);
 141                 /*
 142                  * If we failed here, we might as well return and
 143                  * have a leak rather than continue and a bugchk
 144                  */
 145                 if (ret != 0) {
 146                         netdev_err(ndev, "unable to send "
 147                                 "revoke receive buffer to netvsp\n");
 148                         return;
 149                 }
 150         }
 151
 152         /* Teardown the gpadl on the vsp end */
 153         if (net_device->recv_buf_gpadl_handle) {
 154                 ret = vmbus_teardown_gpadl(device->channel,
 155                                            net_device->recv_buf_gpadl_handle);
 156
 157                 /* If we failed here, we might as well return and have a leak
 158                  * rather than continue and a bugchk
 159                  */
 160                 if (ret != 0) {
 161                         netdev_err(ndev,
 162                                    "unable to teardown receive buffer's gpadl\n");
 163                         return;
 164                 }
 165                 net_device->recv_buf_gpadl_handle = 0;
 166         }
 167
 168         if (net_device->recv_buf) {
 169                 /* Free up the receive buffer */
 170                 vfree(net_device->recv_buf);
 171                 net_device->recv_buf = NULL;
 172         }
 173
 174         if (net_device->recv_section) {
 175                 net_device->recv_section_cnt = 0;
 176                 kfree(net_device->recv_section);
 177                 net_device->recv_section = NULL;
 178         }
 179
 180         /* Deal with the send buffer we may have setup.
 181          * If we got a  send section size, it means we received a
 182          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 183          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 184          * to send a revoke msg here
 185          */
 186         if (net_device->send_section_size) {
 187                 /* Send the revoke receive buffer */
 188                 revoke_packet = &net_device->revoke_packet;
 189                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 190
 191                 revoke_packet->hdr.msg_type =
 192                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 193                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 194                         NETVSC_SEND_BUFFER_ID;
 195
 196                 ret = vmbus_sendpacket(device->channel,
 197                                        revoke_packet,
 198                                        sizeof(struct nvsp_message),
 199                                        (unsigned long)revoke_packet,
 200                                        VM_PKT_DATA_INBAND, 0);
 201                 /* If we failed here, we might as well return and
 202                  * have a leak rather than continue and a bugchk
 203                  */
 204                 if (ret != 0) {
 205                         netdev_err(ndev, "unable to send "
 206                                    "revoke send buffer to netvsp\n");
 207                         return;
 208                 }
 209         }
 210         /* Teardown the gpadl on the vsp end */
 211         if (net_device->send_buf_gpadl_handle) {
 212                 ret = vmbus_teardown_gpadl(device->channel,
 213                                            net_device->send_buf_gpadl_handle);
 214
 215                 /* If we failed here, we might as well return and have a leak
 216                  * rather than continue and a bugchk
 217                  */
 218                 if (ret != 0) {
 219                         netdev_err(ndev,
 220                                    "unable to teardown send buffer's gpadl\n");
 221                         return;
 222                 }
 223                 net_device->send_buf_gpadl_handle = 0;
 224         }
 225         if (net_device->send_buf) {
 226                 /* Free up the send buffer */
 227                 vfree(net_device->send_buf);
 228                 net_device->send_buf = NULL;
 229         }
 230         kfree(net_device->send_section_map);
 231 }
 232
 233 static int netvsc_init_buf(struct hv_device *device)
 234 {
 235         int ret = 0;
 236         struct netvsc_device *net_device;
 237         struct nvsp_message *init_packet;
 238         struct net_device *ndev;
 239         int node;
 240
 241         net_device = get_outbound_net_device(device);
 242         if (!net_device)
 243                 return -ENODEV;
 244         ndev = hv_get_drvdata(device);
 245
 246         node = cpu_to_node(device->channel->target_cpu);
 247         net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
 248         if (!net_device->recv_buf)
 249                 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 250
 251         if (!net_device->recv_buf) {
 252                 netdev_err(ndev, "unable to allocate receive "
 253                         "buffer of size %d\n", net_device->recv_buf_size);
 254                 ret = -ENOMEM;
 255                 goto cleanup;
 256         }
 257
 258         /*
 259          * Establish the gpadl handle for this buffer on this
 260          * channel.  Note: This call uses the vmbus connection rather
 261          * than the channel to establish the gpadl handle.
 262          */
 263         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 264                                     net_device->recv_buf_size,
 265                                     &net_device->recv_buf_gpadl_handle);
 266         if (ret != 0) {
 267                 netdev_err(ndev,
 268                         "unable to establish receive buffer's gpadl\n");
 269                 goto cleanup;
 270         }
 271
 272         /* Notify the NetVsp of the gpadl handle */
 273         init_packet = &net_device->channel_init_pkt;
 274
 275         memset(init_packet, 0, sizeof(struct nvsp_message));
 276
 277         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 278         init_packet->msg.v1_msg.send_recv_buf.
 279                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 280         init_packet->msg.v1_msg.
 281                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 282
 283         /* Send the gpadl notification request */
 284         ret = vmbus_sendpacket(device->channel, init_packet,
 285                                sizeof(struct nvsp_message),
 286                                (unsigned long)init_packet,
 287                                VM_PKT_DATA_INBAND,
 288                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 289         if (ret != 0) {
 290                 netdev_err(ndev,
 291                         "unable to send receive buffer's gpadl to netvsp\n");
 292                 goto cleanup;
 293         }
 294
 295         wait_for_completion(&net_device->channel_init_wait);
 296
 297         /* Check the response */
 298         if (init_packet->msg.v1_msg.
 299             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 300                 netdev_err(ndev, "Unable to complete receive buffer "
 301                            "initialization with NetVsp - status %d\n",
 302                            init_packet->msg.v1_msg.
 303                            send_recv_buf_complete.status);
 304                 ret = -EINVAL;
 305                 goto cleanup;
 306         }
 307
 308         /* Parse the response */
 309
 310         net_device->recv_section_cnt = init_packet->msg.
 311                 v1_msg.send_recv_buf_complete.num_sections;
 312
 313         net_device->recv_section = kmemdup(
 314                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 315                 net_device->recv_section_cnt *
 316                 sizeof(struct nvsp_1_receive_buffer_section),
 317                 GFP_KERNEL);
 318         if (net_device->recv_section == NULL) {
 319                 ret = -EINVAL;
 320                 goto cleanup;
 321         }
 322
 323         /*
 324          * For 1st release, there should only be 1 section that represents the
 325          * entire receive buffer
 326          */
 327         if (net_device->recv_section_cnt != 1 ||
 328             net_device->recv_section->offset != 0) {
 329                 ret = -EINVAL;
 330                 goto cleanup;
 331         }
 332
 333         /* Now setup the send buffer.
 334          */
 335         net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
 336         if (!net_device->send_buf)
 337                 net_device->send_buf = vzalloc(net_device->send_buf_size);
 338         if (!net_device->send_buf) {
 339                 netdev_err(ndev, "unable to allocate send "
 340                            "buffer of size %d\n", net_device->send_buf_size);
 341                 ret = -ENOMEM;
 342                 goto cleanup;
 343         }
 344
 345         /* Establish the gpadl handle for this buffer on this
 346          * channel.  Note: This call uses the vmbus connection rather
 347          * than the channel to establish the gpadl handle.
 348          */
 349         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 350                                     net_device->send_buf_size,
 351                                     &net_device->send_buf_gpadl_handle);
 352         if (ret != 0) {
 353                 netdev_err(ndev,
 354                            "unable to establish send buffer's gpadl\n");
 355                 goto cleanup;
 356         }
 357
 358         /* Notify the NetVsp of the gpadl handle */
 359         init_packet = &net_device->channel_init_pkt;
 360         memset(init_packet, 0, sizeof(struct nvsp_message));
 361         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 362         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 363                 net_device->send_buf_gpadl_handle;
 364         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 365
 366         /* Send the gpadl notification request */
 367         ret = vmbus_sendpacket(device->channel, init_packet,
 368                                sizeof(struct nvsp_message),
 369                                (unsigned long)init_packet,
 370                                VM_PKT_DATA_INBAND,
 371                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 372         if (ret != 0) {
 373                 netdev_err(ndev,
 374                            "unable to send send buffer's gpadl to netvsp\n");
 375                 goto cleanup;
 376         }
 377
 378         wait_for_completion(&net_device->channel_init_wait);
 379
 380         /* Check the response */
 381         if (init_packet->msg.v1_msg.
 382             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 383                 netdev_err(ndev, "Unable to complete send buffer "
 384                            "initialization with NetVsp - status %d\n",
 385                            init_packet->msg.v1_msg.
 386                            send_send_buf_complete.status);
 387                 ret = -EINVAL;
 388                 goto cleanup;
 389         }
 390
 391         /* Parse the response */
 392         net_device->send_section_size = init_packet->msg.
 393                                 v1_msg.send_send_buf_complete.section_size;
 394
 395         /* Section count is simply the size divided by the section size.
 396          */
 397         net_device->send_section_cnt =
 398                 net_device->send_buf_size / net_device->send_section_size;
 399
 400         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 401                    net_device->send_section_size, net_device->send_section_cnt);
 402
 403         /* Setup state for managing the send buffer. */
 404         net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
 405                                              BITS_PER_LONG);
 406
 407         net_device->send_section_map = kcalloc(net_device->map_words,
 408                                                sizeof(ulong), GFP_KERNEL);
 409         if (net_device->send_section_map == NULL) {
 410                 ret = -ENOMEM;
 411                 goto cleanup;
 412         }
 413
 414         goto exit;
 415
 416 cleanup:
 417         netvsc_destroy_buf(device);
 418
 419 exit:
 420         return ret;
 421 }
 422
 423 /* Negotiate NVSP protocol version */
 424 static int negotiate_nvsp_ver(struct hv_device *device,
 425                               struct netvsc_device *net_device,
 426                               struct nvsp_message *init_packet,
 427                               u32 nvsp_ver)
 428 {
 429         struct net_device *ndev = hv_get_drvdata(device);
 430         int ret;
 431
 432         memset(init_packet, 0, sizeof(struct nvsp_message));
 433         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 434         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 435         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 436
 437         /* Send the init request */
 438         ret = vmbus_sendpacket(device->channel, init_packet,
 439                                sizeof(struct nvsp_message),
 440                                (unsigned long)init_packet,
 441                                VM_PKT_DATA_INBAND,
 442                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 443
 444         if (ret != 0)
 445                 return ret;
 446
 447         wait_for_completion(&net_device->channel_init_wait);
 448
 449         if (init_packet->msg.init_msg.init_complete.status !=
 450             NVSP_STAT_SUCCESS)
 451                 return -EINVAL;
 452
 453         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 454                 return 0;
 455
 456         /* NVSPv2 or later: Send NDIS config */
 457         memset(init_packet, 0, sizeof(struct nvsp_message));
 458         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 459         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 460         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 461
 462         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 463                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 464
 465                 /* Teaming bit is needed to receive link speed updates */
 466                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 467         }
 468
 469         ret = vmbus_sendpacket(device->channel, init_packet,
 470                                 sizeof(struct nvsp_message),
 471                                 (unsigned long)init_packet,
 472                                 VM_PKT_DATA_INBAND, 0);
 473
 474         return ret;
 475 }
 476
 477 static int netvsc_connect_vsp(struct hv_device *device)
 478 {
 479         int ret;
 480         struct netvsc_device *net_device;
 481         struct nvsp_message *init_packet;
 482         int ndis_version;
 483         const u32 ver_list[] = {
 484                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 485                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
 486         int i;
 487
 488         net_device = get_outbound_net_device(device);
 489         if (!net_device)
 490                 return -ENODEV;
 491
 492         init_packet = &net_device->channel_init_pkt;
 493
 494         /* Negotiate the latest NVSP protocol supported */
 495         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 496                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 497                                        ver_list[i])  == 0) {
 498                         net_device->nvsp_version = ver_list[i];
 499                         break;
 500                 }
 501
 502         if (i < 0) {
 503                 ret = -EPROTO;
 504                 goto cleanup;
 505         }
 506
 507         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 508
 509         /* Send the ndis version */
 510         memset(init_packet, 0, sizeof(struct nvsp_message));
 511
 512         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 513                 ndis_version = 0x00060001;
 514         else
 515                 ndis_version = 0x0006001e;
 516
 517         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 518         init_packet->msg.v1_msg.
 519                 send_ndis_ver.ndis_major_ver =
 520                                 (ndis_version & 0xFFFF0000) >> 16;
 521         init_packet->msg.v1_msg.
 522                 send_ndis_ver.ndis_minor_ver =
 523                                 ndis_version & 0xFFFF;
 524
 525         /* Send the init request */
 526         ret = vmbus_sendpacket(device->channel, init_packet,
 527                                 sizeof(struct nvsp_message),
 528                                 (unsigned long)init_packet,
 529                                 VM_PKT_DATA_INBAND, 0);
 530         if (ret != 0)
 531                 goto cleanup;
 532
 533         /* Post the big receive buffer to NetVSP */
 534         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 535                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 536         else
 537                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 538         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 539
 540         ret = netvsc_init_buf(device);
 541
 542 cleanup:
 543         return ret;
 544 }
 545
 546 static void netvsc_disconnect_vsp(struct hv_device *device)
 547 {
 548         netvsc_destroy_buf(device);
 549 }
 550
 551 /*
 552  * netvsc_device_remove - Callback when the root bus device is removed
 553  */
 554 void netvsc_device_remove(struct hv_device *device)
 555 {
 556         struct net_device *ndev = hv_get_drvdata(device);
 557         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 558         struct netvsc_device *net_device = net_device_ctx->nvdev;
 559         int i;
 560
 561         netvsc_disconnect_vsp(device);
 562
 563         net_device_ctx->nvdev = NULL;
 564
 565         /*
 566          * At this point, no one should be accessing net_device
 567          * except in here
 568          */
 569         netdev_dbg(ndev, "net device safe to remove\n");
 570
 571         /* Now, we can close the channel safely */
 572         vmbus_close(device->channel);
 573
 574         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
 575                 napi_disable(&net_device->chan_table[0].napi);
 576
 577         /* Release all resources */
 578         free_netvsc_device(net_device);
 579 }
 580
 581 #define RING_AVAIL_PERCENT_HIWATER 20
 582 #define RING_AVAIL_PERCENT_LOWATER 10
 583
 584 /*
 585  * Get the percentage of available bytes to write in the ring.
 586  * The return value is in range from 0 to 100.
 587  */
 588 static inline u32 hv_ringbuf_avail_percent(
 589                 struct hv_ring_buffer_info *ring_info)
 590 {
 591         u32 avail_read, avail_write;
 592
 593         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 594
 595         return avail_write * 100 / ring_info->ring_datasize;
 596 }
 597
 598 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 599                                          u32 index)
 600 {
 601         sync_change_bit(index, net_device->send_section_map);
 602 }
 603
 604 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
 605                                     struct vmbus_channel *incoming_channel,
 606                                     struct hv_device *device,
 607                                     const struct vmpacket_descriptor *desc)
 608 {
 609         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 610         struct net_device *ndev = hv_get_drvdata(device);
 611         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 612         struct vmbus_channel *channel = device->channel;
 613         u16 q_idx = 0;
 614         int queue_sends;
 615
 616         /* Notify the layer above us */
 617         if (likely(skb)) {
 618                 const struct hv_netvsc_packet *packet
 619                         = (struct hv_netvsc_packet *)skb->cb;
 620                 u32 send_index = packet->send_buf_index;
 621                 struct netvsc_stats *tx_stats;
 622
 623                 if (send_index != NETVSC_INVALID_INDEX)
 624                         netvsc_free_send_slot(net_device, send_index);
 625                 q_idx = packet->q_idx;
 626                 channel = incoming_channel;
 627
 628                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 629
 630                 u64_stats_update_begin(&tx_stats->syncp);
 631                 tx_stats->packets += packet->total_packets;
 632                 tx_stats->bytes += packet->total_bytes;
 633                 u64_stats_update_end(&tx_stats->syncp);
 634
 635                 dev_consume_skb_any(skb);
 636         }
 637
 638         queue_sends =
 639                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 640
 641         if (net_device->destroy && queue_sends == 0)
 642                 wake_up(&net_device->wait_drain);
 643
 644         if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 645             !net_device_ctx->start_remove &&
 646             (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
 647              queue_sends < 1))
 648                 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
 649 }
 650
 651 static void netvsc_send_completion(struct netvsc_device *net_device,
 652                                    struct vmbus_channel *incoming_channel,
 653                                    struct hv_device *device,
 654                                    const struct vmpacket_descriptor *desc)
 655 {
 656         struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 657         struct net_device *ndev = hv_get_drvdata(device);
 658
 659         switch (nvsp_packet->hdr.msg_type) {
 660         case NVSP_MSG_TYPE_INIT_COMPLETE:
 661         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 662         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 663         case NVSP_MSG5_TYPE_SUBCHANNEL:
 664                 /* Copy the response back */
 665                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 666                        sizeof(struct nvsp_message));
 667                 complete(&net_device->channel_init_wait);
 668                 break;
 669
 670         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 671                 netvsc_send_tx_complete(net_device, incoming_channel,
 672                                         device, desc);
 673                 break;
 674
 675         default:
 676                 netdev_err(ndev,
 677                            "Unknown send completion type %d received!!\n",
 678                            nvsp_packet->hdr.msg_type);
 679         }
 680 }
 681
 682 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 683 {
 684         unsigned long *map_addr = net_device->send_section_map;
 685         unsigned int i;
 686
 687         for_each_clear_bit(i, map_addr, net_device->map_words) {
 688                 if (sync_test_and_set_bit(i, map_addr) == 0)
 689                         return i;
 690         }
 691
 692         return NETVSC_INVALID_INDEX;
 693 }
 694
 695 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 696                                    unsigned int section_index,
 697                                    u32 pend_size,
 698                                    struct hv_netvsc_packet *packet,
 699                                    struct rndis_message *rndis_msg,
 700                                    struct hv_page_buffer **pb,
 701                                    struct sk_buff *skb)
 702 {
 703         char *start = net_device->send_buf;
 704         char *dest = start + (section_index * net_device->send_section_size)
 705                      + pend_size;
 706         int i;
 707         u32 msg_size = 0;
 708         u32 padding = 0;
 709         u32 remain = packet->total_data_buflen % net_device->pkt_align;
 710         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 711                 packet->page_buf_cnt;
 712
 713         /* Add padding */
 714         if (skb && skb->xmit_more && remain &&
 715             !packet->cp_partial) {
 716                 padding = net_device->pkt_align - remain;
 717                 rndis_msg->msg_len += padding;
 718                 packet->total_data_buflen += padding;
 719         }
 720
 721         for (i = 0; i < page_count; i++) {
 722                 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
 723                 u32 offset = (*pb)[i].offset;
 724                 u32 len = (*pb)[i].len;
 725
 726                 memcpy(dest, (src + offset), len);
 727                 msg_size += len;
 728                 dest += len;
 729         }
 730
 731         if (padding) {
 732                 memset(dest, 0, padding);
 733                 msg_size += padding;
 734         }
 735
 736         return msg_size;
 737 }
 738
 739 static inline int netvsc_send_pkt(
 740         struct hv_device *device,
 741         struct hv_netvsc_packet *packet,
 742         struct netvsc_device *net_device,
 743         struct hv_page_buffer **pb,
 744         struct sk_buff *skb)
 745 {
 746         struct nvsp_message nvmsg;
 747         struct netvsc_channel *nvchan
 748                 = &net_device->chan_table[packet->q_idx];
 749         struct vmbus_channel *out_channel = nvchan->channel;
 750         struct net_device *ndev = hv_get_drvdata(device);
 751         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 752         u64 req_id;
 753         int ret;
 754         struct hv_page_buffer *pgbuf;
 755         u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
 756
 757         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 758         if (skb != NULL) {
 759                 /* 0 is RMC_DATA; */
 760                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 761         } else {
 762                 /* 1 is RMC_CONTROL; */
 763                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 764         }
 765
 766         nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 767                 packet->send_buf_index;
 768         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 769                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 770         else
 771                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
 772                         packet->total_data_buflen;
 773
 774         req_id = (ulong)skb;
 775
 776         if (out_channel->rescind)
 777                 return -ENODEV;
 778
 779         if (packet->page_buf_cnt) {
 780                 pgbuf = packet->cp_partial ? (*pb) +
 781                         packet->rmsg_pgcnt : (*pb);
 782                 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 783                                                       pgbuf,
 784                                                       packet->page_buf_cnt,
 785                                                       &nvmsg,
 786                                                       sizeof(struct nvsp_message),
 787                                                       req_id,
 788                                                       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 789         } else {
 790                 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
 791                                            sizeof(struct nvsp_message),
 792                                            req_id,
 793                                            VM_PKT_DATA_INBAND,
 794                                            VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 795         }
 796
 797         if (ret == 0) {
 798                 atomic_inc_return(&nvchan->queue_sends);
 799
 800                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER)
 801                         netif_tx_stop_queue(txq);
 802         } else if (ret == -EAGAIN) {
 803                 netif_tx_stop_queue(txq);
 804                 if (atomic_read(&nvchan->queue_sends) < 1) {
 805                         netif_tx_wake_queue(txq);
 806                         ret = -ENOSPC;
 807                 }
 808         } else {
 809                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 810                            packet, ret);
 811         }
 812
 813         return ret;
 814 }
 815
 816 /* Move packet out of multi send data (msd), and clear msd */
 817 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 818                                 struct sk_buff **msd_skb,
 819                                 struct multi_send_data *msdp)
 820 {
 821         *msd_skb = msdp->skb;
 822         *msd_send = msdp->pkt;
 823         msdp->skb = NULL;
 824         msdp->pkt = NULL;
 825         msdp->count = 0;
 826 }
 827
 828 int netvsc_send(struct hv_device *device,
 829                 struct hv_netvsc_packet *packet,
 830                 struct rndis_message *rndis_msg,
 831                 struct hv_page_buffer **pb,
 832                 struct sk_buff *skb)
 833 {
 834         struct netvsc_device *net_device;
 835         int ret = 0;
 836         struct netvsc_channel *nvchan;
 837         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 838         unsigned int section_index = NETVSC_INVALID_INDEX;
 839         struct multi_send_data *msdp;
 840         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 841         struct sk_buff *msd_skb = NULL;
 842         bool try_batch;
 843         bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
 844
 845         net_device = get_outbound_net_device(device);
 846         if (!net_device)
 847                 return -ENODEV;
 848
 849         /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
 850          * here before the negotiation with the host is finished and
 851          * send_section_map may not be allocated yet.
 852          */
 853         if (!net_device->send_section_map)
 854                 return -EAGAIN;
 855
 856         nvchan = &net_device->chan_table[packet->q_idx];
 857         packet->send_buf_index = NETVSC_INVALID_INDEX;
 858         packet->cp_partial = false;
 859
 860         /* Send control message directly without accessing msd (Multi-Send
 861          * Data) field which may be changed during data packet processing.
 862          */
 863         if (!skb) {
 864                 cur_send = packet;
 865                 goto send_now;
 866         }
 867
 868         /* batch packets in send buffer if possible */
 869         msdp = &nvchan->msd;
 870         if (msdp->pkt)
 871                 msd_len = msdp->pkt->total_data_buflen;
 872
 873         try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
 874                     net_device->max_pkt;
 875
 876         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 877             net_device->send_section_size) {
 878                 section_index = msdp->pkt->send_buf_index;
 879
 880         } else if (try_batch && msd_len + packet->rmsg_size <
 881                    net_device->send_section_size) {
 882                 section_index = msdp->pkt->send_buf_index;
 883                 packet->cp_partial = true;
 884
 885         } else if ((skb != NULL) && pktlen + net_device->pkt_align <
 886                    net_device->send_section_size) {
 887                 section_index = netvsc_get_next_send_section(net_device);
 888                 if (section_index != NETVSC_INVALID_INDEX) {
 889                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 890                         msd_len = 0;
 891                 }
 892         }
 893
 894         if (section_index != NETVSC_INVALID_INDEX) {
 895                 netvsc_copy_to_send_buf(net_device,
 896                                         section_index, msd_len,
 897                                         packet, rndis_msg, pb, skb);
 898
 899                 packet->send_buf_index = section_index;
 900
 901                 if (packet->cp_partial) {
 902                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 903                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 904                 } else {
 905                         packet->page_buf_cnt = 0;
 906                         packet->total_data_buflen += msd_len;
 907                 }
 908
 909                 if (msdp->pkt) {
 910                         packet->total_packets += msdp->pkt->total_packets;
 911                         packet->total_bytes += msdp->pkt->total_bytes;
 912                 }
 913
 914                 if (msdp->skb)
 915                         dev_consume_skb_any(msdp->skb);
 916
 917                 if (xmit_more && !packet->cp_partial) {
 918                         msdp->skb = skb;
 919                         msdp->pkt = packet;
 920                         msdp->count++;
 921                 } else {
 922                         cur_send = packet;
 923                         msdp->skb = NULL;
 924                         msdp->pkt = NULL;
 925                         msdp->count = 0;
 926                 }
 927         } else {
 928                 move_pkt_msd(&msd_send, &msd_skb, msdp);
 929                 cur_send = packet;
 930         }
 931
 932         if (msd_send) {
 933                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
 934                                             NULL, msd_skb);
 935
 936                 if (m_ret != 0) {
 937                         netvsc_free_send_slot(net_device,
 938                                               msd_send->send_buf_index);
 939                         dev_kfree_skb_any(msd_skb);
 940                 }
 941         }
 942
 943 send_now:
 944         if (cur_send)
 945                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
 946
 947         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
 948                 netvsc_free_send_slot(net_device, section_index);
 949
 950         return ret;
 951 }
 952
 953 static int netvsc_send_recv_completion(struct vmbus_channel *channel,
 954                                        u64 transaction_id, u32 status)
 955 {
 956         struct nvsp_message recvcompMessage;
 957         int ret;
 958
 959         recvcompMessage.hdr.msg_type =
 960                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 961
 962         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 963
 964         /* Send the completion */
 965         ret = vmbus_sendpacket(channel, &recvcompMessage,
 966                                sizeof(struct nvsp_message_header) + sizeof(u32),
 967                                transaction_id, VM_PKT_COMP, 0);
 968
 969         return ret;
 970 }
 971
 972 static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
 973                                         u32 *filled, u32 *avail)
 974 {
 975         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
 976         u32 first = mrc->first;
 977         u32 next = mrc->next;
 978
 979         *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
 980                   next - first;
 981
 982         *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
 983 }
 984
 985 /* Read the first filled slot, no change to index */
 986 static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
 987                                                          *nvdev, u16 q_idx)
 988 {
 989         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
 990         u32 filled, avail;
 991
 992         if (unlikely(!mrc->buf))
 993                 return NULL;
 994
 995         count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
 996         if (!filled)
 997                 return NULL;
 998
 999         return mrc->buf + mrc->first * sizeof(struct recv_comp_data);
1000 }
1001
1002 /* Put the first filled slot back to available pool */
1003 static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
1004 {
1005         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1006         int num_recv;
1007
1008         mrc->first = (mrc->first + 1) % NETVSC_RECVSLOT_MAX;
1009
1010         num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1011
1012         if (nvdev->destroy && num_recv == 0)
1013                 wake_up(&nvdev->wait_drain);
1014 }
1015
1016 /* Check and send pending recv completions */
1017 static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1018                                  struct vmbus_channel *channel, u16 q_idx)
1019 {
1020         struct recv_comp_data *rcd;
1021         int ret;
1022
1023         while (true) {
1024                 rcd = read_recv_comp_slot(nvdev, q_idx);
1025                 if (!rcd)
1026                         break;
1027
1028                 ret = netvsc_send_recv_completion(channel, rcd->tid,
1029                                                   rcd->status);
1030                 if (ret)
1031                         break;
1032
1033                 put_recv_comp_slot(nvdev, q_idx);
1034         }
1035 }
1036
1037 #define NETVSC_RCD_WATERMARK 80
1038
1039 /* Get next available slot */
1040 static inline struct recv_comp_data *get_recv_comp_slot(
1041         struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1042 {
1043         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1044         u32 filled, avail, next;
1045         struct recv_comp_data *rcd;
1046
1047         if (unlikely(!nvdev->recv_section))
1048                 return NULL;
1049
1050         if (unlikely(!mrc->buf))
1051                 return NULL;
1052
1053         if (atomic_read(&nvdev->num_outstanding_recvs) >
1054             nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1055                 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1056
1057         count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1058         if (!avail)
1059                 return NULL;
1060
1061         next = mrc->next;
1062         rcd = mrc->buf + next * sizeof(struct recv_comp_data);
1063         mrc->next = (next + 1) % NETVSC_RECVSLOT_MAX;
1064
1065         atomic_inc(&nvdev->num_outstanding_recvs);
1066
1067         return rcd;
1068 }
1069
1070 static int netvsc_receive(struct net_device *ndev,
1071                    struct netvsc_device *net_device,
1072                    struct net_device_context *net_device_ctx,
1073                    struct hv_device *device,
1074                    struct vmbus_channel *channel,
1075                    const struct vmpacket_descriptor *desc,
1076                    struct nvsp_message *nvsp)
1077 {
1078         const struct vmtransfer_page_packet_header *vmxferpage_packet
1079                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1080         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1081         char *recv_buf = net_device->recv_buf;
1082         u32 status = NVSP_STAT_SUCCESS;
1083         int i;
1084         int count = 0;
1085         int ret;
1086
1087         /* Make sure this is a valid nvsp packet */
1088         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1089                 netif_err(net_device_ctx, rx_err, ndev,
1090                           "Unknown nvsp packet type received %u\n",
1091                           nvsp->hdr.msg_type);
1092                 return 0;
1093         }
1094
1095         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1096                 netif_err(net_device_ctx, rx_err, ndev,
1097                           "Invalid xfer page set id - expecting %x got %x\n",
1098                           NETVSC_RECEIVE_BUFFER_ID,
1099                           vmxferpage_packet->xfer_pageset_id);
1100                 return 0;
1101         }
1102
1103         count = vmxferpage_packet->range_cnt;
1104
1105         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1106         for (i = 0; i < count; i++) {
1107                 void *data = recv_buf
1108                         + vmxferpage_packet->ranges[i].byte_offset;
1109                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1110
1111                 /* Pass it to the upper layer */
1112                 status = rndis_filter_receive(ndev, net_device, device,
1113                                               channel, data, buflen);
1114         }
1115
1116         if (net_device->chan_table[q_idx].mrc.buf) {
1117                 struct recv_comp_data *rcd;
1118
1119                 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1120                 if (rcd) {
1121                         rcd->tid = vmxferpage_packet->d.trans_id;
1122                         rcd->status = status;
1123                 } else {
1124                         netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1125                                    q_idx, vmxferpage_packet->d.trans_id);
1126                 }
1127         } else {
1128                 ret = netvsc_send_recv_completion(channel,
1129                                                   vmxferpage_packet->d.trans_id,
1130                                                   status);
1131                 if (ret)
1132                         netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1133                                    q_idx, vmxferpage_packet->d.trans_id, ret);
1134         }
1135         return count;
1136 }
1137
1138 static void netvsc_send_table(struct hv_device *hdev,
1139                               struct nvsp_message *nvmsg)
1140 {
1141         struct netvsc_device *nvscdev;
1142         struct net_device *ndev = hv_get_drvdata(hdev);
1143         int i;
1144         u32 count, *tab;
1145
1146         nvscdev = get_outbound_net_device(hdev);
1147         if (!nvscdev)
1148                 return;
1149
1150         count = nvmsg->msg.v5_msg.send_table.count;
1151         if (count != VRSS_SEND_TAB_SIZE) {
1152                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1153                 return;
1154         }
1155
1156         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1157                       nvmsg->msg.v5_msg.send_table.offset);
1158
1159         for (i = 0; i < count; i++)
1160                 nvscdev->send_table[i] = tab[i];
1161 }
1162
1163 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1164                            struct nvsp_message *nvmsg)
1165 {
1166         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1167         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1168 }
1169
1170 static inline void netvsc_receive_inband(struct hv_device *hdev,
1171                                  struct net_device_context *net_device_ctx,
1172                                  struct nvsp_message *nvmsg)
1173 {
1174         switch (nvmsg->hdr.msg_type) {
1175         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1176                 netvsc_send_table(hdev, nvmsg);
1177                 break;
1178
1179         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1180                 netvsc_send_vf(net_device_ctx, nvmsg);
1181                 break;
1182         }
1183 }
1184
1185 static int netvsc_process_raw_pkt(struct hv_device *device,
1186                                   struct vmbus_channel *channel,
1187                                   struct netvsc_device *net_device,
1188                                   struct net_device *ndev,
1189                                   u64 request_id,
1190                                   const struct vmpacket_descriptor *desc)
1191 {
1192         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1193         struct nvsp_message *nvmsg = hv_pkt_data(desc);
1194
1195         switch (desc->type) {
1196         case VM_PKT_COMP:
1197                 netvsc_send_completion(net_device, channel, device, desc);
1198                 break;
1199
1200         case VM_PKT_DATA_USING_XFER_PAGES:
1201                 return netvsc_receive(ndev, net_device, net_device_ctx,
1202                                       device, channel, desc, nvmsg);
1203                 break;
1204
1205         case VM_PKT_DATA_INBAND:
1206                 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1207                 break;
1208
1209         default:
1210                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1211                            desc->type, request_id);
1212                 break;
1213         }
1214
1215         return 0;
1216 }
1217
1218 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1219 {
1220         struct vmbus_channel *primary = channel->primary_channel;
1221
1222         return primary ? primary->device_obj : channel->device_obj;
1223 }
1224
1225 int netvsc_poll(struct napi_struct *napi, int budget)
1226 {
1227         struct netvsc_channel *nvchan
1228                 = container_of(napi, struct netvsc_channel, napi);
1229         struct vmbus_channel *channel = nvchan->channel;
1230         struct hv_device *device = netvsc_channel_to_device(channel);
1231         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1232         struct net_device *ndev = hv_get_drvdata(device);
1233         struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
1234         const struct vmpacket_descriptor *desc;
1235         int work_done = 0;
1236
1237         desc = hv_pkt_iter_first(channel);
1238         while (desc) {
1239                 int count;
1240
1241                 count = netvsc_process_raw_pkt(device, channel, net_device,
1242                                                ndev, desc->trans_id, desc);
1243                 work_done += count;
1244                 desc = __hv_pkt_iter_next(channel, desc);
1245
1246                 /* If receive packet budget is exhausted, reschedule */
1247                 if (work_done >= budget) {
1248                         work_done = budget;
1249                         break;
1250                 }
1251         }
1252         hv_pkt_iter_close(channel);
1253
1254         /* If ring is empty and NAPI is not doing polling */
1255         if (work_done < budget &&
1256             napi_complete_done(napi, work_done) &&
1257             hv_end_read(&channel->inbound) != 0)
1258                 napi_reschedule(napi);
1259
1260         netvsc_chk_recv_comp(net_device, channel, q_idx);
1261         return work_done;
1262 }
1263
1264 void netvsc_channel_cb(void *context)
1265 {
1266         struct vmbus_channel *channel = context;
1267         struct hv_device *device = netvsc_channel_to_device(channel);
1268         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1269         struct netvsc_device *net_device;
1270         struct net_device *ndev;
1271
1272         ndev = hv_get_drvdata(device);
1273         if (unlikely(!ndev))
1274                 return;
1275
1276         net_device = net_device_to_netvsc_device(ndev);
1277         if (unlikely(net_device->destroy) &&
1278             netvsc_channel_idle(net_device, q_idx))
1279                 return;
1280
1281         /* disable interupts from host */
1282         hv_begin_read(&channel->inbound);
1283         napi_schedule(&net_device->chan_table[q_idx].napi);
1284 }
1285
1286 /*
1287  * netvsc_device_add - Callback when the device belonging to this
1288  * driver is added
1289  */
1290 int netvsc_device_add(struct hv_device *device,
1291                       const struct netvsc_device_info *device_info)
1292 {
1293         int i, ret = 0;
1294         int ring_size = device_info->ring_size;
1295         struct netvsc_device *net_device;
1296         struct net_device *ndev = hv_get_drvdata(device);
1297         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1298
1299         net_device = alloc_net_device();
1300         if (!net_device)
1301                 return -ENOMEM;
1302
1303         net_device->ring_size = ring_size;
1304
1305         /* Because the device uses NAPI, all the interrupt batching and
1306          * control is done via Net softirq, not the channel handling
1307          */
1308         set_channel_read_mode(device->channel, HV_CALL_ISR);
1309
1310         /* Open the channel */
1311         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1312                          ring_size * PAGE_SIZE, NULL, 0,
1313                          netvsc_channel_cb, device->channel);
1314
1315         if (ret != 0) {
1316                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1317                 goto cleanup;
1318         }
1319
1320         /* Channel is opened */
1321         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1322
1323         /* If we're reopening the device we may have multiple queues, fill the
1324          * chn_table with the default channel to use it before subchannels are
1325          * opened.
1326          */
1327         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1328                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1329
1330                 nvchan->channel = device->channel;
1331                 netif_napi_add(ndev, &nvchan->napi,
1332                                netvsc_poll, NAPI_POLL_WEIGHT);
1333         }
1334
1335         /* Enable NAPI handler for init callbacks */
1336         napi_enable(&net_device->chan_table[0].napi);
1337
1338         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1339          * populated.
1340          */
1341         wmb();
1342
1343         net_device_ctx->nvdev = net_device;
1344
1345         /* Connect with the NetVsp */
1346         ret = netvsc_connect_vsp(device);
1347         if (ret != 0) {
1348                 netdev_err(ndev,
1349                         "unable to connect to NetVSP - %d\n", ret);
1350                 goto close;
1351         }
1352
1353         return ret;
1354
1355 close:
1356         napi_disable(&net_device->chan_table[0].napi);
1357
1358         /* Now, we can close the channel safely */
1359         vmbus_close(device->channel);
1360
1361 cleanup:
1362         free_netvsc_device(net_device);
1363
1364         return ret;
1365 }