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 <[email protected]>
  18  *   Hank Janssen  <[email protected]>
  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 <linux/rtnetlink.h>
  33 #include <linux/prefetch.h>
  34
  35 #include <asm/sync_bitops.h>
  36
  37 #include "hyperv_net.h"
  38 #include "netvsc_trace.h"
  39
  40 /*
  41  * Switch the data path from the synthetic interface to the VF
  42  * interface.
  43  */
  44 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  45 {
  46         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  47         struct hv_device *dev = net_device_ctx->device_ctx;
  48         struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
  49         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  50
  51         memset(init_pkt, 0, sizeof(struct nvsp_message));
  52         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  53         if (vf)
  54                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  55                         NVSP_DATAPATH_VF;
  56         else
  57                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  58                         NVSP_DATAPATH_SYNTHETIC;
  59
  60         trace_nvsp_send(ndev, init_pkt);
  61
  62         vmbus_sendpacket(dev->channel, init_pkt,
  63                                sizeof(struct nvsp_message),
  64                                (unsigned long)init_pkt,
  65                                VM_PKT_DATA_INBAND, 0);
  66 }
  67
  68 /* Worker to setup sub channels on initial setup
  69  * Initial hotplug event occurs in softirq context
  70  * and can't wait for channels.
  71  */
  72 static void netvsc_subchan_work(struct work_struct *w)
  73 {
  74         struct netvsc_device *nvdev =
  75                 container_of(w, struct netvsc_device, subchan_work);
  76         struct rndis_device *rdev;
  77         int i, ret;
  78
  79         /* Avoid deadlock with device removal already under RTNL */
  80         if (!rtnl_trylock()) {
  81                 schedule_work(w);
  82                 return;
  83         }
  84
  85         rdev = nvdev->extension;
  86         if (rdev) {
  87                 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
  88                 if (ret == 0) {
  89                         netif_device_attach(rdev->ndev);
  90                 } else {
  91                         /* fallback to only primary channel */
  92                         for (i = 1; i < nvdev->num_chn; i++)
  93                                 netif_napi_del(&nvdev->chan_table[i].napi);
  94
  95                         nvdev->max_chn = 1;
  96                         nvdev->num_chn = 1;
  97                 }
  98         }
  99
 100         rtnl_unlock();
 101 }
 102
 103 static struct netvsc_device *alloc_net_device(void)
 104 {
 105         struct netvsc_device *net_device;
 106
 107         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
 108         if (!net_device)
 109                 return NULL;
 110
 111         init_waitqueue_head(&net_device->wait_drain);
 112         net_device->destroy = false;
 113         net_device->tx_disable = false;
 114
 115         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
 116         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
 117
 118         init_completion(&net_device->channel_init_wait);
 119         init_waitqueue_head(&net_device->subchan_open);
 120         INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
 121
 122         return net_device;
 123 }
 124
 125 static void free_netvsc_device(struct rcu_head *head)
 126 {
 127         struct netvsc_device *nvdev
 128                 = container_of(head, struct netvsc_device, rcu);
 129         int i;
 130
 131         kfree(nvdev->extension);
 132         vfree(nvdev->recv_buf);
 133         vfree(nvdev->send_buf);
 134         kfree(nvdev->send_section_map);
 135
 136         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
 137                 vfree(nvdev->chan_table[i].mrc.slots);
 138
 139         kfree(nvdev);
 140 }
 141
 142 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
 143 {
 144         call_rcu(&nvdev->rcu, free_netvsc_device);
 145 }
 146
 147 static void netvsc_revoke_recv_buf(struct hv_device *device,
 148                                    struct netvsc_device *net_device,
 149                                    struct net_device *ndev)
 150 {
 151         struct nvsp_message *revoke_packet;
 152         int ret;
 153
 154         /*
 155          * If we got a section count, it means we received a
 156          * SendReceiveBufferComplete msg (ie sent
 157          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 158          * to send a revoke msg here
 159          */
 160         if (net_device->recv_section_cnt) {
 161                 /* Send the revoke receive buffer */
 162                 revoke_packet = &net_device->revoke_packet;
 163                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 164
 165                 revoke_packet->hdr.msg_type =
 166                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 167                 revoke_packet->msg.v1_msg.
 168                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 169
 170                 trace_nvsp_send(ndev, revoke_packet);
 171
 172                 ret = vmbus_sendpacket(device->channel,
 173                                        revoke_packet,
 174                                        sizeof(struct nvsp_message),
 175                                        (unsigned long)revoke_packet,
 176                                        VM_PKT_DATA_INBAND, 0);
 177                 /* If the failure is because the channel is rescinded;
 178                  * ignore the failure since we cannot send on a rescinded
 179                  * channel. This would allow us to properly cleanup
 180                  * even when the channel is rescinded.
 181                  */
 182                 if (device->channel->rescind)
 183                         ret = 0;
 184                 /*
 185                  * If we failed here, we might as well return and
 186                  * have a leak rather than continue and a bugchk
 187                  */
 188                 if (ret != 0) {
 189                         netdev_err(ndev, "unable to send "
 190                                 "revoke receive buffer to netvsp\n");
 191                         return;
 192                 }
 193                 net_device->recv_section_cnt = 0;
 194         }
 195 }
 196
 197 static void netvsc_revoke_send_buf(struct hv_device *device,
 198                                    struct netvsc_device *net_device,
 199                                    struct net_device *ndev)
 200 {
 201         struct nvsp_message *revoke_packet;
 202         int ret;
 203
 204         /* Deal with the send buffer we may have setup.
 205          * If we got a  send section size, it means we received a
 206          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 207          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 208          * to send a revoke msg here
 209          */
 210         if (net_device->send_section_cnt) {
 211                 /* Send the revoke receive buffer */
 212                 revoke_packet = &net_device->revoke_packet;
 213                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 214
 215                 revoke_packet->hdr.msg_type =
 216                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 217                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 218                         NETVSC_SEND_BUFFER_ID;
 219
 220                 trace_nvsp_send(ndev, revoke_packet);
 221
 222                 ret = vmbus_sendpacket(device->channel,
 223                                        revoke_packet,
 224                                        sizeof(struct nvsp_message),
 225                                        (unsigned long)revoke_packet,
 226                                        VM_PKT_DATA_INBAND, 0);
 227
 228                 /* If the failure is because the channel is rescinded;
 229                  * ignore the failure since we cannot send on a rescinded
 230                  * channel. This would allow us to properly cleanup
 231                  * even when the channel is rescinded.
 232                  */
 233                 if (device->channel->rescind)
 234                         ret = 0;
 235
 236                 /* If we failed here, we might as well return and
 237                  * have a leak rather than continue and a bugchk
 238                  */
 239                 if (ret != 0) {
 240                         netdev_err(ndev, "unable to send "
 241                                    "revoke send buffer to netvsp\n");
 242                         return;
 243                 }
 244                 net_device->send_section_cnt = 0;
 245         }
 246 }
 247
 248 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
 249                                        struct netvsc_device *net_device,
 250                                        struct net_device *ndev)
 251 {
 252         int ret;
 253
 254         if (net_device->recv_buf_gpadl_handle) {
 255                 ret = vmbus_teardown_gpadl(device->channel,
 256                                            net_device->recv_buf_gpadl_handle);
 257
 258                 /* If we failed here, we might as well return and have a leak
 259                  * rather than continue and a bugchk
 260                  */
 261                 if (ret != 0) {
 262                         netdev_err(ndev,
 263                                    "unable to teardown receive buffer's gpadl\n");
 264                         return;
 265                 }
 266                 net_device->recv_buf_gpadl_handle = 0;
 267         }
 268 }
 269
 270 static void netvsc_teardown_send_gpadl(struct hv_device *device,
 271                                        struct netvsc_device *net_device,
 272                                        struct net_device *ndev)
 273 {
 274         int ret;
 275
 276         if (net_device->send_buf_gpadl_handle) {
 277                 ret = vmbus_teardown_gpadl(device->channel,
 278                                            net_device->send_buf_gpadl_handle);
 279
 280                 /* If we failed here, we might as well return and have a leak
 281                  * rather than continue and a bugchk
 282                  */
 283                 if (ret != 0) {
 284                         netdev_err(ndev,
 285                                    "unable to teardown send buffer's gpadl\n");
 286                         return;
 287                 }
 288                 net_device->send_buf_gpadl_handle = 0;
 289         }
 290 }
 291
 292 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
 293 {
 294         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 295         int node = cpu_to_node(nvchan->channel->target_cpu);
 296         size_t size;
 297
 298         size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
 299         nvchan->mrc.slots = vzalloc_node(size, node);
 300         if (!nvchan->mrc.slots)
 301                 nvchan->mrc.slots = vzalloc(size);
 302
 303         return nvchan->mrc.slots ? 0 : -ENOMEM;
 304 }
 305
 306 static int netvsc_init_buf(struct hv_device *device,
 307                            struct netvsc_device *net_device,
 308                            const struct netvsc_device_info *device_info)
 309 {
 310         struct nvsp_1_message_send_receive_buffer_complete *resp;
 311         struct net_device *ndev = hv_get_drvdata(device);
 312         struct nvsp_message *init_packet;
 313         unsigned int buf_size;
 314         size_t map_words;
 315         int ret = 0;
 316
 317         /* Get receive buffer area. */
 318         buf_size = device_info->recv_sections * device_info->recv_section_size;
 319         buf_size = roundup(buf_size, PAGE_SIZE);
 320
 321         /* Legacy hosts only allow smaller receive buffer */
 322         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 323                 buf_size = min_t(unsigned int, buf_size,
 324                                  NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
 325
 326         net_device->recv_buf = vzalloc(buf_size);
 327         if (!net_device->recv_buf) {
 328                 netdev_err(ndev,
 329                            "unable to allocate receive buffer of size %u\n",
 330                            buf_size);
 331                 ret = -ENOMEM;
 332                 goto cleanup;
 333         }
 334
 335         net_device->recv_buf_size = buf_size;
 336
 337         /*
 338          * Establish the gpadl handle for this buffer on this
 339          * channel.  Note: This call uses the vmbus connection rather
 340          * than the channel to establish the gpadl handle.
 341          */
 342         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 343                                     buf_size,
 344                                     &net_device->recv_buf_gpadl_handle);
 345         if (ret != 0) {
 346                 netdev_err(ndev,
 347                         "unable to establish receive buffer's gpadl\n");
 348                 goto cleanup;
 349         }
 350
 351         /* Notify the NetVsp of the gpadl handle */
 352         init_packet = &net_device->channel_init_pkt;
 353         memset(init_packet, 0, sizeof(struct nvsp_message));
 354         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 355         init_packet->msg.v1_msg.send_recv_buf.
 356                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 357         init_packet->msg.v1_msg.
 358                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 359
 360         trace_nvsp_send(ndev, init_packet);
 361
 362         /* Send the gpadl notification request */
 363         ret = vmbus_sendpacket(device->channel, init_packet,
 364                                sizeof(struct nvsp_message),
 365                                (unsigned long)init_packet,
 366                                VM_PKT_DATA_INBAND,
 367                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 368         if (ret != 0) {
 369                 netdev_err(ndev,
 370                         "unable to send receive buffer's gpadl to netvsp\n");
 371                 goto cleanup;
 372         }
 373
 374         wait_for_completion(&net_device->channel_init_wait);
 375
 376         /* Check the response */
 377         resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
 378         if (resp->status != NVSP_STAT_SUCCESS) {
 379                 netdev_err(ndev,
 380                            "Unable to complete receive buffer initialization with NetVsp - status %d\n",
 381                            resp->status);
 382                 ret = -EINVAL;
 383                 goto cleanup;
 384         }
 385
 386         /* Parse the response */
 387         netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
 388                    resp->num_sections, resp->sections[0].sub_alloc_size,
 389                    resp->sections[0].num_sub_allocs);
 390
 391         /* There should only be one section for the entire receive buffer */
 392         if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
 393                 ret = -EINVAL;
 394                 goto cleanup;
 395         }
 396
 397         net_device->recv_section_size = resp->sections[0].sub_alloc_size;
 398         net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
 399
 400         /* Setup receive completion ring */
 401         net_device->recv_completion_cnt
 402                 = round_up(net_device->recv_section_cnt + 1,
 403                            PAGE_SIZE / sizeof(u64));
 404         ret = netvsc_alloc_recv_comp_ring(net_device, 0);
 405         if (ret)
 406                 goto cleanup;
 407
 408         /* Now setup the send buffer. */
 409         buf_size = device_info->send_sections * device_info->send_section_size;
 410         buf_size = round_up(buf_size, PAGE_SIZE);
 411
 412         net_device->send_buf = vzalloc(buf_size);
 413         if (!net_device->send_buf) {
 414                 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
 415                            buf_size);
 416                 ret = -ENOMEM;
 417                 goto cleanup;
 418         }
 419
 420         /* Establish the gpadl handle for this buffer on this
 421          * channel.  Note: This call uses the vmbus connection rather
 422          * than the channel to establish the gpadl handle.
 423          */
 424         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 425                                     buf_size,
 426                                     &net_device->send_buf_gpadl_handle);
 427         if (ret != 0) {
 428                 netdev_err(ndev,
 429                            "unable to establish send buffer's gpadl\n");
 430                 goto cleanup;
 431         }
 432
 433         /* Notify the NetVsp of the gpadl handle */
 434         init_packet = &net_device->channel_init_pkt;
 435         memset(init_packet, 0, sizeof(struct nvsp_message));
 436         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 437         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 438                 net_device->send_buf_gpadl_handle;
 439         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 440
 441         trace_nvsp_send(ndev, init_packet);
 442
 443         /* Send the gpadl notification request */
 444         ret = vmbus_sendpacket(device->channel, init_packet,
 445                                sizeof(struct nvsp_message),
 446                                (unsigned long)init_packet,
 447                                VM_PKT_DATA_INBAND,
 448                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 449         if (ret != 0) {
 450                 netdev_err(ndev,
 451                            "unable to send send buffer's gpadl to netvsp\n");
 452                 goto cleanup;
 453         }
 454
 455         wait_for_completion(&net_device->channel_init_wait);
 456
 457         /* Check the response */
 458         if (init_packet->msg.v1_msg.
 459             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 460                 netdev_err(ndev, "Unable to complete send buffer "
 461                            "initialization with NetVsp - status %d\n",
 462                            init_packet->msg.v1_msg.
 463                            send_send_buf_complete.status);
 464                 ret = -EINVAL;
 465                 goto cleanup;
 466         }
 467
 468         /* Parse the response */
 469         net_device->send_section_size = init_packet->msg.
 470                                 v1_msg.send_send_buf_complete.section_size;
 471
 472         /* Section count is simply the size divided by the section size. */
 473         net_device->send_section_cnt = buf_size / net_device->send_section_size;
 474
 475         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 476                    net_device->send_section_size, net_device->send_section_cnt);
 477
 478         /* Setup state for managing the send buffer. */
 479         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
 480
 481         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
 482         if (net_device->send_section_map == NULL) {
 483                 ret = -ENOMEM;
 484                 goto cleanup;
 485         }
 486
 487         goto exit;
 488
 489 cleanup:
 490         netvsc_revoke_recv_buf(device, net_device, ndev);
 491         netvsc_revoke_send_buf(device, net_device, ndev);
 492         netvsc_teardown_recv_gpadl(device, net_device, ndev);
 493         netvsc_teardown_send_gpadl(device, net_device, ndev);
 494
 495 exit:
 496         return ret;
 497 }
 498
 499 /* Negotiate NVSP protocol version */
 500 static int negotiate_nvsp_ver(struct hv_device *device,
 501                               struct netvsc_device *net_device,
 502                               struct nvsp_message *init_packet,
 503                               u32 nvsp_ver)
 504 {
 505         struct net_device *ndev = hv_get_drvdata(device);
 506         int ret;
 507
 508         memset(init_packet, 0, sizeof(struct nvsp_message));
 509         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 510         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 511         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 512         trace_nvsp_send(ndev, init_packet);
 513
 514         /* Send the init request */
 515         ret = vmbus_sendpacket(device->channel, init_packet,
 516                                sizeof(struct nvsp_message),
 517                                (unsigned long)init_packet,
 518                                VM_PKT_DATA_INBAND,
 519                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 520
 521         if (ret != 0)
 522                 return ret;
 523
 524         wait_for_completion(&net_device->channel_init_wait);
 525
 526         if (init_packet->msg.init_msg.init_complete.status !=
 527             NVSP_STAT_SUCCESS)
 528                 return -EINVAL;
 529
 530         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 531                 return 0;
 532
 533         /* NVSPv2 or later: Send NDIS config */
 534         memset(init_packet, 0, sizeof(struct nvsp_message));
 535         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 536         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 537         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 538
 539         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 540                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 541
 542                 /* Teaming bit is needed to receive link speed updates */
 543                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 544         }
 545
 546         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
 547                 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
 548
 549         trace_nvsp_send(ndev, init_packet);
 550
 551         ret = vmbus_sendpacket(device->channel, init_packet,
 552                                 sizeof(struct nvsp_message),
 553                                 (unsigned long)init_packet,
 554                                 VM_PKT_DATA_INBAND, 0);
 555
 556         return ret;
 557 }
 558
 559 static int netvsc_connect_vsp(struct hv_device *device,
 560                               struct netvsc_device *net_device,
 561                               const struct netvsc_device_info *device_info)
 562 {
 563         struct net_device *ndev = hv_get_drvdata(device);
 564         static const u32 ver_list[] = {
 565                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 566                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
 567                 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
 568         };
 569         struct nvsp_message *init_packet;
 570         int ndis_version, i, ret;
 571
 572         init_packet = &net_device->channel_init_pkt;
 573
 574         /* Negotiate the latest NVSP protocol supported */
 575         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 576                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 577                                        ver_list[i])  == 0) {
 578                         net_device->nvsp_version = ver_list[i];
 579                         break;
 580                 }
 581
 582         if (i < 0) {
 583                 ret = -EPROTO;
 584                 goto cleanup;
 585         }
 586
 587         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 588
 589         /* Send the ndis version */
 590         memset(init_packet, 0, sizeof(struct nvsp_message));
 591
 592         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 593                 ndis_version = 0x00060001;
 594         else
 595                 ndis_version = 0x0006001e;
 596
 597         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 598         init_packet->msg.v1_msg.
 599                 send_ndis_ver.ndis_major_ver =
 600                                 (ndis_version & 0xFFFF0000) >> 16;
 601         init_packet->msg.v1_msg.
 602                 send_ndis_ver.ndis_minor_ver =
 603                                 ndis_version & 0xFFFF;
 604
 605         trace_nvsp_send(ndev, init_packet);
 606
 607         /* Send the init request */
 608         ret = vmbus_sendpacket(device->channel, init_packet,
 609                                 sizeof(struct nvsp_message),
 610                                 (unsigned long)init_packet,
 611                                 VM_PKT_DATA_INBAND, 0);
 612         if (ret != 0)
 613                 goto cleanup;
 614
 615
 616         ret = netvsc_init_buf(device, net_device, device_info);
 617
 618 cleanup:
 619         return ret;
 620 }
 621
 622 /*
 623  * netvsc_device_remove - Callback when the root bus device is removed
 624  */
 625 void netvsc_device_remove(struct hv_device *device)
 626 {
 627         struct net_device *ndev = hv_get_drvdata(device);
 628         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 629         struct netvsc_device *net_device
 630                 = rtnl_dereference(net_device_ctx->nvdev);
 631         int i;
 632
 633         /*
 634          * Revoke receive buffer. If host is pre-Win2016 then tear down
 635          * receive buffer GPADL. Do the same for send buffer.
 636          */
 637         netvsc_revoke_recv_buf(device, net_device, ndev);
 638         if (vmbus_proto_version < VERSION_WIN10)
 639                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 640
 641         netvsc_revoke_send_buf(device, net_device, ndev);
 642         if (vmbus_proto_version < VERSION_WIN10)
 643                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 644
 645         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 646
 647         /* And disassociate NAPI context from device */
 648         for (i = 0; i < net_device->num_chn; i++)
 649                 netif_napi_del(&net_device->chan_table[i].napi);
 650
 651         /*
 652          * At this point, no one should be accessing net_device
 653          * except in here
 654          */
 655         netdev_dbg(ndev, "net device safe to remove\n");
 656
 657         /* Now, we can close the channel safely */
 658         vmbus_close(device->channel);
 659
 660         /*
 661          * If host is Win2016 or higher then we do the GPADL tear down
 662          * here after VMBus is closed.
 663         */
 664         if (vmbus_proto_version >= VERSION_WIN10) {
 665                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 666                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 667         }
 668
 669         /* Release all resources */
 670         free_netvsc_device_rcu(net_device);
 671 }
 672
 673 #define RING_AVAIL_PERCENT_HIWATER 20
 674 #define RING_AVAIL_PERCENT_LOWATER 10
 675
 676 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 677                                          u32 index)
 678 {
 679         sync_change_bit(index, net_device->send_section_map);
 680 }
 681
 682 static void netvsc_send_tx_complete(struct net_device *ndev,
 683                                     struct netvsc_device *net_device,
 684                                     struct vmbus_channel *channel,
 685                                     const struct vmpacket_descriptor *desc,
 686                                     int budget)
 687 {
 688         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 689         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 690         u16 q_idx = 0;
 691         int queue_sends;
 692
 693         /* Notify the layer above us */
 694         if (likely(skb)) {
 695                 const struct hv_netvsc_packet *packet
 696                         = (struct hv_netvsc_packet *)skb->cb;
 697                 u32 send_index = packet->send_buf_index;
 698                 struct netvsc_stats *tx_stats;
 699
 700                 if (send_index != NETVSC_INVALID_INDEX)
 701                         netvsc_free_send_slot(net_device, send_index);
 702                 q_idx = packet->q_idx;
 703
 704                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 705
 706                 u64_stats_update_begin(&tx_stats->syncp);
 707                 tx_stats->packets += packet->total_packets;
 708                 tx_stats->bytes += packet->total_bytes;
 709                 u64_stats_update_end(&tx_stats->syncp);
 710
 711                 napi_consume_skb(skb, budget);
 712         }
 713
 714         queue_sends =
 715                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 716
 717         if (unlikely(net_device->destroy)) {
 718                 if (queue_sends == 0)
 719                         wake_up(&net_device->wait_drain);
 720         } else {
 721                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
 722
 723                 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
 724                     (hv_get_avail_to_write_percent(&channel->outbound) >
 725                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
 726                         netif_tx_wake_queue(txq);
 727                         ndev_ctx->eth_stats.wake_queue++;
 728                 }
 729         }
 730 }
 731
 732 static void netvsc_send_completion(struct net_device *ndev,
 733                                    struct netvsc_device *net_device,
 734                                    struct vmbus_channel *incoming_channel,
 735                                    const struct vmpacket_descriptor *desc,
 736                                    int budget)
 737 {
 738         const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 739
 740         switch (nvsp_packet->hdr.msg_type) {
 741         case NVSP_MSG_TYPE_INIT_COMPLETE:
 742         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 743         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 744         case NVSP_MSG5_TYPE_SUBCHANNEL:
 745                 /* Copy the response back */
 746                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 747                        sizeof(struct nvsp_message));
 748                 complete(&net_device->channel_init_wait);
 749                 break;
 750
 751         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 752                 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
 753                                         desc, budget);
 754                 break;
 755
 756         default:
 757                 netdev_err(ndev,
 758                            "Unknown send completion type %d received!!\n",
 759                            nvsp_packet->hdr.msg_type);
 760         }
 761 }
 762
 763 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 764 {
 765         unsigned long *map_addr = net_device->send_section_map;
 766         unsigned int i;
 767
 768         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 769                 if (sync_test_and_set_bit(i, map_addr) == 0)
 770                         return i;
 771         }
 772
 773         return NETVSC_INVALID_INDEX;
 774 }
 775
 776 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 777                                     unsigned int section_index,
 778                                     u32 pend_size,
 779                                     struct hv_netvsc_packet *packet,
 780                                     struct rndis_message *rndis_msg,
 781                                     struct hv_page_buffer *pb,
 782                                     bool xmit_more)
 783 {
 784         char *start = net_device->send_buf;
 785         char *dest = start + (section_index * net_device->send_section_size)
 786                      + pend_size;
 787         int i;
 788         u32 padding = 0;
 789         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 790                 packet->page_buf_cnt;
 791         u32 remain;
 792
 793         /* Add padding */
 794         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
 795         if (xmit_more && remain) {
 796                 padding = net_device->pkt_align - remain;
 797                 rndis_msg->msg_len += padding;
 798                 packet->total_data_buflen += padding;
 799         }
 800
 801         for (i = 0; i < page_count; i++) {
 802                 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
 803                 u32 offset = pb[i].offset;
 804                 u32 len = pb[i].len;
 805
 806                 memcpy(dest, (src + offset), len);
 807                 dest += len;
 808         }
 809
 810         if (padding)
 811                 memset(dest, 0, padding);
 812 }
 813
 814 static inline int netvsc_send_pkt(
 815         struct hv_device *device,
 816         struct hv_netvsc_packet *packet,
 817         struct netvsc_device *net_device,
 818         struct hv_page_buffer *pb,
 819         struct sk_buff *skb)
 820 {
 821         struct nvsp_message nvmsg;
 822         struct nvsp_1_message_send_rndis_packet *rpkt =
 823                 &nvmsg.msg.v1_msg.send_rndis_pkt;
 824         struct netvsc_channel * const nvchan =
 825                 &net_device->chan_table[packet->q_idx];
 826         struct vmbus_channel *out_channel = nvchan->channel;
 827         struct net_device *ndev = hv_get_drvdata(device);
 828         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 829         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 830         u64 req_id;
 831         int ret;
 832         u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
 833
 834         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 835         if (skb)
 836                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
 837         else
 838                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
 839
 840         rpkt->send_buf_section_index = packet->send_buf_index;
 841         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 842                 rpkt->send_buf_section_size = 0;
 843         else
 844                 rpkt->send_buf_section_size = packet->total_data_buflen;
 845
 846         req_id = (ulong)skb;
 847
 848         if (out_channel->rescind)
 849                 return -ENODEV;
 850
 851         trace_nvsp_send_pkt(ndev, out_channel, rpkt);
 852
 853         if (packet->page_buf_cnt) {
 854                 if (packet->cp_partial)
 855                         pb += packet->rmsg_pgcnt;
 856
 857                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 858                                                   pb, packet->page_buf_cnt,
 859                                                   &nvmsg, sizeof(nvmsg),
 860                                                   req_id);
 861         } else {
 862                 ret = vmbus_sendpacket(out_channel,
 863                                        &nvmsg, sizeof(nvmsg),
 864                                        req_id, VM_PKT_DATA_INBAND,
 865                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 866         }
 867
 868         if (ret == 0) {
 869                 atomic_inc_return(&nvchan->queue_sends);
 870
 871                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 872                         netif_tx_stop_queue(txq);
 873                         ndev_ctx->eth_stats.stop_queue++;
 874                 }
 875         } else if (ret == -EAGAIN) {
 876                 netif_tx_stop_queue(txq);
 877                 ndev_ctx->eth_stats.stop_queue++;
 878         } else {
 879                 netdev_err(ndev,
 880                            "Unable to send packet pages %u len %u, ret %d\n",
 881                            packet->page_buf_cnt, packet->total_data_buflen,
 882                            ret);
 883         }
 884
 885         if (netif_tx_queue_stopped(txq) &&
 886             atomic_read(&nvchan->queue_sends) < 1 &&
 887             !net_device->tx_disable) {
 888                 netif_tx_wake_queue(txq);
 889                 ndev_ctx->eth_stats.wake_queue++;
 890                 if (ret == -EAGAIN)
 891                         ret = -ENOSPC;
 892         }
 893
 894         return ret;
 895 }
 896
 897 /* Move packet out of multi send data (msd), and clear msd */
 898 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 899                                 struct sk_buff **msd_skb,
 900                                 struct multi_send_data *msdp)
 901 {
 902         *msd_skb = msdp->skb;
 903         *msd_send = msdp->pkt;
 904         msdp->skb = NULL;
 905         msdp->pkt = NULL;
 906         msdp->count = 0;
 907 }
 908
 909 /* RCU already held by caller */
 910 int netvsc_send(struct net_device *ndev,
 911                 struct hv_netvsc_packet *packet,
 912                 struct rndis_message *rndis_msg,
 913                 struct hv_page_buffer *pb,
 914                 struct sk_buff *skb)
 915 {
 916         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 917         struct netvsc_device *net_device
 918                 = rcu_dereference_bh(ndev_ctx->nvdev);
 919         struct hv_device *device = ndev_ctx->device_ctx;
 920         int ret = 0;
 921         struct netvsc_channel *nvchan;
 922         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 923         unsigned int section_index = NETVSC_INVALID_INDEX;
 924         struct multi_send_data *msdp;
 925         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 926         struct sk_buff *msd_skb = NULL;
 927         bool try_batch, xmit_more;
 928
 929         /* If device is rescinded, return error and packet will get dropped. */
 930         if (unlikely(!net_device || net_device->destroy))
 931                 return -ENODEV;
 932
 933         nvchan = &net_device->chan_table[packet->q_idx];
 934         packet->send_buf_index = NETVSC_INVALID_INDEX;
 935         packet->cp_partial = false;
 936
 937         /* Send control message directly without accessing msd (Multi-Send
 938          * Data) field which may be changed during data packet processing.
 939          */
 940         if (!skb)
 941                 return netvsc_send_pkt(device, packet, net_device, pb, skb);
 942
 943         /* batch packets in send buffer if possible */
 944         msdp = &nvchan->msd;
 945         if (msdp->pkt)
 946                 msd_len = msdp->pkt->total_data_buflen;
 947
 948         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
 949         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 950             net_device->send_section_size) {
 951                 section_index = msdp->pkt->send_buf_index;
 952
 953         } else if (try_batch && msd_len + packet->rmsg_size <
 954                    net_device->send_section_size) {
 955                 section_index = msdp->pkt->send_buf_index;
 956                 packet->cp_partial = true;
 957
 958         } else if (pktlen + net_device->pkt_align <
 959                    net_device->send_section_size) {
 960                 section_index = netvsc_get_next_send_section(net_device);
 961                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
 962                         ++ndev_ctx->eth_stats.tx_send_full;
 963                 } else {
 964                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 965                         msd_len = 0;
 966                 }
 967         }
 968
 969         /* Keep aggregating only if stack says more data is coming
 970          * and not doing mixed modes send and not flow blocked
 971          */
 972         xmit_more = netdev_xmit_more() &&
 973                 !packet->cp_partial &&
 974                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
 975
 976         if (section_index != NETVSC_INVALID_INDEX) {
 977                 netvsc_copy_to_send_buf(net_device,
 978                                         section_index, msd_len,
 979                                         packet, rndis_msg, pb, xmit_more);
 980
 981                 packet->send_buf_index = section_index;
 982
 983                 if (packet->cp_partial) {
 984                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 985                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 986                 } else {
 987                         packet->page_buf_cnt = 0;
 988                         packet->total_data_buflen += msd_len;
 989                 }
 990
 991                 if (msdp->pkt) {
 992                         packet->total_packets += msdp->pkt->total_packets;
 993                         packet->total_bytes += msdp->pkt->total_bytes;
 994                 }
 995
 996                 if (msdp->skb)
 997                         dev_consume_skb_any(msdp->skb);
 998
 999                 if (xmit_more) {
1000                         msdp->skb = skb;
1001                         msdp->pkt = packet;
1002                         msdp->count++;
1003                 } else {
1004                         cur_send = packet;
1005                         msdp->skb = NULL;
1006                         msdp->pkt = NULL;
1007                         msdp->count = 0;
1008                 }
1009         } else {
1010                 move_pkt_msd(&msd_send, &msd_skb, msdp);
1011                 cur_send = packet;
1012         }
1013
1014         if (msd_send) {
1015                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1016                                             NULL, msd_skb);
1017
1018                 if (m_ret != 0) {
1019                         netvsc_free_send_slot(net_device,
1020                                               msd_send->send_buf_index);
1021                         dev_kfree_skb_any(msd_skb);
1022                 }
1023         }
1024
1025         if (cur_send)
1026                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1027
1028         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1029                 netvsc_free_send_slot(net_device, section_index);
1030
1031         return ret;
1032 }
1033
1034 /* Send pending recv completions */
1035 static int send_recv_completions(struct net_device *ndev,
1036                                  struct netvsc_device *nvdev,
1037                                  struct netvsc_channel *nvchan)
1038 {
1039         struct multi_recv_comp *mrc = &nvchan->mrc;
1040         struct recv_comp_msg {
1041                 struct nvsp_message_header hdr;
1042                 u32 status;
1043         }  __packed;
1044         struct recv_comp_msg msg = {
1045                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1046         };
1047         int ret;
1048
1049         while (mrc->first != mrc->next) {
1050                 const struct recv_comp_data *rcd
1051                         = mrc->slots + mrc->first;
1052
1053                 msg.status = rcd->status;
1054                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1055                                        rcd->tid, VM_PKT_COMP, 0);
1056                 if (unlikely(ret)) {
1057                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1058
1059                         ++ndev_ctx->eth_stats.rx_comp_busy;
1060                         return ret;
1061                 }
1062
1063                 if (++mrc->first == nvdev->recv_completion_cnt)
1064                         mrc->first = 0;
1065         }
1066
1067         /* receive completion ring has been emptied */
1068         if (unlikely(nvdev->destroy))
1069                 wake_up(&nvdev->wait_drain);
1070
1071         return 0;
1072 }
1073
1074 /* Count how many receive completions are outstanding */
1075 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1076                                  const struct multi_recv_comp *mrc,
1077                                  u32 *filled, u32 *avail)
1078 {
1079         u32 count = nvdev->recv_completion_cnt;
1080
1081         if (mrc->next >= mrc->first)
1082                 *filled = mrc->next - mrc->first;
1083         else
1084                 *filled = (count - mrc->first) + mrc->next;
1085
1086         *avail = count - *filled - 1;
1087 }
1088
1089 /* Add receive complete to ring to send to host. */
1090 static void enq_receive_complete(struct net_device *ndev,
1091                                  struct netvsc_device *nvdev, u16 q_idx,
1092                                  u64 tid, u32 status)
1093 {
1094         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1095         struct multi_recv_comp *mrc = &nvchan->mrc;
1096         struct recv_comp_data *rcd;
1097         u32 filled, avail;
1098
1099         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1100
1101         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1102                 send_recv_completions(ndev, nvdev, nvchan);
1103                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1104         }
1105
1106         if (unlikely(!avail)) {
1107                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1108                            q_idx, tid);
1109                 return;
1110         }
1111
1112         rcd = mrc->slots + mrc->next;
1113         rcd->tid = tid;
1114         rcd->status = status;
1115
1116         if (++mrc->next == nvdev->recv_completion_cnt)
1117                 mrc->next = 0;
1118 }
1119
1120 static int netvsc_receive(struct net_device *ndev,
1121                           struct netvsc_device *net_device,
1122                           struct netvsc_channel *nvchan,
1123                           const struct vmpacket_descriptor *desc,
1124                           const struct nvsp_message *nvsp)
1125 {
1126         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1127         struct vmbus_channel *channel = nvchan->channel;
1128         const struct vmtransfer_page_packet_header *vmxferpage_packet
1129                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1130         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1131         char *recv_buf = net_device->recv_buf;
1132         u32 status = NVSP_STAT_SUCCESS;
1133         int i;
1134         int count = 0;
1135
1136         /* Make sure this is a valid nvsp packet */
1137         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1138                 netif_err(net_device_ctx, rx_err, ndev,
1139                           "Unknown nvsp packet type received %u\n",
1140                           nvsp->hdr.msg_type);
1141                 return 0;
1142         }
1143
1144         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1145                 netif_err(net_device_ctx, rx_err, ndev,
1146                           "Invalid xfer page set id - expecting %x got %x\n",
1147                           NETVSC_RECEIVE_BUFFER_ID,
1148                           vmxferpage_packet->xfer_pageset_id);
1149                 return 0;
1150         }
1151
1152         count = vmxferpage_packet->range_cnt;
1153
1154         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1155         for (i = 0; i < count; i++) {
1156                 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1157                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1158                 void *data;
1159                 int ret;
1160
1161                 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1162                         nvchan->rsc.cnt = 0;
1163                         status = NVSP_STAT_FAIL;
1164                         netif_err(net_device_ctx, rx_err, ndev,
1165                                   "Packet offset:%u + len:%u too big\n",
1166                                   offset, buflen);
1167
1168                         continue;
1169                 }
1170
1171                 data = recv_buf + offset;
1172
1173                 nvchan->rsc.is_last = (i == count - 1);
1174
1175                 trace_rndis_recv(ndev, q_idx, data);
1176
1177                 /* Pass it to the upper layer */
1178                 ret = rndis_filter_receive(ndev, net_device,
1179                                            nvchan, data, buflen);
1180
1181                 if (unlikely(ret != NVSP_STAT_SUCCESS))
1182                         status = NVSP_STAT_FAIL;
1183         }
1184
1185         enq_receive_complete(ndev, net_device, q_idx,
1186                              vmxferpage_packet->d.trans_id, status);
1187
1188         return count;
1189 }
1190
1191 static void netvsc_send_table(struct net_device *ndev,
1192                               const struct nvsp_message *nvmsg)
1193 {
1194         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1195         u32 count, *tab;
1196         int i;
1197
1198         count = nvmsg->msg.v5_msg.send_table.count;
1199         if (count != VRSS_SEND_TAB_SIZE) {
1200                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1201                 return;
1202         }
1203
1204         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1205                       nvmsg->msg.v5_msg.send_table.offset);
1206
1207         for (i = 0; i < count; i++)
1208                 net_device_ctx->tx_table[i] = tab[i];
1209 }
1210
1211 static void netvsc_send_vf(struct net_device *ndev,
1212                            const struct nvsp_message *nvmsg)
1213 {
1214         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1215
1216         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1217         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1218         netdev_info(ndev, "VF slot %u %s\n",
1219                     net_device_ctx->vf_serial,
1220                     net_device_ctx->vf_alloc ? "added" : "removed");
1221 }
1222
1223 static  void netvsc_receive_inband(struct net_device *ndev,
1224                                    const struct nvsp_message *nvmsg)
1225 {
1226         switch (nvmsg->hdr.msg_type) {
1227         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1228                 netvsc_send_table(ndev, nvmsg);
1229                 break;
1230
1231         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1232                 netvsc_send_vf(ndev, nvmsg);
1233                 break;
1234         }
1235 }
1236
1237 static int netvsc_process_raw_pkt(struct hv_device *device,
1238                                   struct netvsc_channel *nvchan,
1239                                   struct netvsc_device *net_device,
1240                                   struct net_device *ndev,
1241                                   const struct vmpacket_descriptor *desc,
1242                                   int budget)
1243 {
1244         struct vmbus_channel *channel = nvchan->channel;
1245         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1246
1247         trace_nvsp_recv(ndev, channel, nvmsg);
1248
1249         switch (desc->type) {
1250         case VM_PKT_COMP:
1251                 netvsc_send_completion(ndev, net_device, channel,
1252                                        desc, budget);
1253                 break;
1254
1255         case VM_PKT_DATA_USING_XFER_PAGES:
1256                 return netvsc_receive(ndev, net_device, nvchan,
1257                                       desc, nvmsg);
1258                 break;
1259
1260         case VM_PKT_DATA_INBAND:
1261                 netvsc_receive_inband(ndev, nvmsg);
1262                 break;
1263
1264         default:
1265                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1266                            desc->type, desc->trans_id);
1267                 break;
1268         }
1269
1270         return 0;
1271 }
1272
1273 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1274 {
1275         struct vmbus_channel *primary = channel->primary_channel;
1276
1277         return primary ? primary->device_obj : channel->device_obj;
1278 }
1279
1280 /* Network processing softirq
1281  * Process data in incoming ring buffer from host
1282  * Stops when ring is empty or budget is met or exceeded.
1283  */
1284 int netvsc_poll(struct napi_struct *napi, int budget)
1285 {
1286         struct netvsc_channel *nvchan
1287                 = container_of(napi, struct netvsc_channel, napi);
1288         struct netvsc_device *net_device = nvchan->net_device;
1289         struct vmbus_channel *channel = nvchan->channel;
1290         struct hv_device *device = netvsc_channel_to_device(channel);
1291         struct net_device *ndev = hv_get_drvdata(device);
1292         int work_done = 0;
1293         int ret;
1294
1295         /* If starting a new interval */
1296         if (!nvchan->desc)
1297                 nvchan->desc = hv_pkt_iter_first(channel);
1298
1299         while (nvchan->desc && work_done < budget) {
1300                 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1301                                                     ndev, nvchan->desc, budget);
1302                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1303         }
1304
1305         /* Send any pending receive completions */
1306         ret = send_recv_completions(ndev, net_device, nvchan);
1307
1308         /* If it did not exhaust NAPI budget this time
1309          *  and not doing busy poll
1310          * then re-enable host interrupts
1311          *  and reschedule if ring is not empty
1312          *   or sending receive completion failed.
1313          */
1314         if (work_done < budget &&
1315             napi_complete_done(napi, work_done) &&
1316             (ret || hv_end_read(&channel->inbound)) &&
1317             napi_schedule_prep(napi)) {
1318                 hv_begin_read(&channel->inbound);
1319                 __napi_schedule(napi);
1320         }
1321
1322         /* Driver may overshoot since multiple packets per descriptor */
1323         return min(work_done, budget);
1324 }
1325
1326 /* Call back when data is available in host ring buffer.
1327  * Processing is deferred until network softirq (NAPI)
1328  */
1329 void netvsc_channel_cb(void *context)
1330 {
1331         struct netvsc_channel *nvchan = context;
1332         struct vmbus_channel *channel = nvchan->channel;
1333         struct hv_ring_buffer_info *rbi = &channel->inbound;
1334
1335         /* preload first vmpacket descriptor */
1336         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1337
1338         if (napi_schedule_prep(&nvchan->napi)) {
1339                 /* disable interrupts from host */
1340                 hv_begin_read(rbi);
1341
1342                 __napi_schedule_irqoff(&nvchan->napi);
1343         }
1344 }
1345
1346 /*
1347  * netvsc_device_add - Callback when the device belonging to this
1348  * driver is added
1349  */
1350 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1351                                 const struct netvsc_device_info *device_info)
1352 {
1353         int i, ret = 0;
1354         struct netvsc_device *net_device;
1355         struct net_device *ndev = hv_get_drvdata(device);
1356         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1357
1358         net_device = alloc_net_device();
1359         if (!net_device)
1360                 return ERR_PTR(-ENOMEM);
1361
1362         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1363                 net_device_ctx->tx_table[i] = 0;
1364
1365         /* Because the device uses NAPI, all the interrupt batching and
1366          * control is done via Net softirq, not the channel handling
1367          */
1368         set_channel_read_mode(device->channel, HV_CALL_ISR);
1369
1370         /* If we're reopening the device we may have multiple queues, fill the
1371          * chn_table with the default channel to use it before subchannels are
1372          * opened.
1373          * Initialize the channel state before we open;
1374          * we can be interrupted as soon as we open the channel.
1375          */
1376
1377         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1378                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1379
1380                 nvchan->channel = device->channel;
1381                 nvchan->net_device = net_device;
1382                 u64_stats_init(&nvchan->tx_stats.syncp);
1383                 u64_stats_init(&nvchan->rx_stats.syncp);
1384         }
1385
1386         /* Enable NAPI handler before init callbacks */
1387         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1388                        netvsc_poll, NAPI_POLL_WEIGHT);
1389
1390         /* Open the channel */
1391         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1392                          netvsc_ring_bytes,  NULL, 0,
1393                          netvsc_channel_cb, net_device->chan_table);
1394
1395         if (ret != 0) {
1396                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1397                 goto cleanup;
1398         }
1399
1400         /* Channel is opened */
1401         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1402
1403         napi_enable(&net_device->chan_table[0].napi);
1404
1405         /* Connect with the NetVsp */
1406         ret = netvsc_connect_vsp(device, net_device, device_info);
1407         if (ret != 0) {
1408                 netdev_err(ndev,
1409                         "unable to connect to NetVSP - %d\n", ret);
1410                 goto close;
1411         }
1412
1413         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1414          * populated.
1415          */
1416         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1417
1418         return net_device;
1419
1420 close:
1421         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1422         napi_disable(&net_device->chan_table[0].napi);
1423
1424         /* Now, we can close the channel safely */
1425         vmbus_close(device->channel);
1426
1427 cleanup:
1428         netif_napi_del(&net_device->chan_table[0].napi);
1429         free_netvsc_device(&net_device->rcu);
1430
1431         return ERR_PTR(ret);
1432 }