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