drivers/net/hyperv/netvsc.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/wait.h>
  25 #include <linux/mm.h>
  26 #include <linux/delay.h>
  27 #include <linux/io.h>
  28 #include <linux/slab.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/vmalloc.h>
  32 #include <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);
  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
 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
 124 static 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
 141 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
 142 {
 143         call_rcu(&nvdev->rcu, free_netvsc_device);
 144 }
 145
 146 static 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
 196 static 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
 247 static 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
 269 static 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
 291 int 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
 305 static 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
 488 cleanup:
 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
 494 exit:
 495         return ret;
 496 }
 497
 498 /* Negotiate NVSP protocol version */
 499 static 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         trace_nvsp_send(ndev, init_packet);
 546
 547         ret = vmbus_sendpacket(device->channel, init_packet,
 548                                 sizeof(struct nvsp_message),
 549                                 (unsigned long)init_packet,
 550                                 VM_PKT_DATA_INBAND, 0);
 551
 552         return ret;
 553 }
 554
 555 static int netvsc_connect_vsp(struct hv_device *device,
 556                               struct netvsc_device *net_device,
 557                               const struct netvsc_device_info *device_info)
 558 {
 559         struct net_device *ndev = hv_get_drvdata(device);
 560         static const u32 ver_list[] = {
 561                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 562                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
 563                 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
 564         };
 565         struct nvsp_message *init_packet;
 566         int ndis_version, i, ret;
 567
 568         init_packet = &net_device->channel_init_pkt;
 569
 570         /* Negotiate the latest NVSP protocol supported */
 571         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 572                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 573                                        ver_list[i])  == 0) {
 574                         net_device->nvsp_version = ver_list[i];
 575                         break;
 576                 }
 577
 578         if (i < 0) {
 579                 ret = -EPROTO;
 580                 goto cleanup;
 581         }
 582
 583         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 584
 585         /* Send the ndis version */
 586         memset(init_packet, 0, sizeof(struct nvsp_message));
 587
 588         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 589                 ndis_version = 0x00060001;
 590         else
 591                 ndis_version = 0x0006001e;
 592
 593         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 594         init_packet->msg.v1_msg.
 595                 send_ndis_ver.ndis_major_ver =
 596                                 (ndis_version & 0xFFFF0000) >> 16;
 597         init_packet->msg.v1_msg.
 598                 send_ndis_ver.ndis_minor_ver =
 599                                 ndis_version & 0xFFFF;
 600
 601         trace_nvsp_send(ndev, init_packet);
 602
 603         /* Send the init request */
 604         ret = vmbus_sendpacket(device->channel, init_packet,
 605                                 sizeof(struct nvsp_message),
 606                                 (unsigned long)init_packet,
 607                                 VM_PKT_DATA_INBAND, 0);
 608         if (ret != 0)
 609                 goto cleanup;
 610
 611
 612         ret = netvsc_init_buf(device, net_device, device_info);
 613
 614 cleanup:
 615         return ret;
 616 }
 617
 618 /*
 619  * netvsc_device_remove - Callback when the root bus device is removed
 620  */
 621 void netvsc_device_remove(struct hv_device *device)
 622 {
 623         struct net_device *ndev = hv_get_drvdata(device);
 624         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 625         struct netvsc_device *net_device
 626                 = rtnl_dereference(net_device_ctx->nvdev);
 627         int i;
 628
 629         /*
 630          * Revoke receive buffer. If host is pre-Win2016 then tear down
 631          * receive buffer GPADL. Do the same for send buffer.
 632          */
 633         netvsc_revoke_recv_buf(device, net_device, ndev);
 634         if (vmbus_proto_version < VERSION_WIN10)
 635                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 636
 637         netvsc_revoke_send_buf(device, net_device, ndev);
 638         if (vmbus_proto_version < VERSION_WIN10)
 639                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 640
 641         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 642
 643         /* And disassociate NAPI context from device */
 644         for (i = 0; i < net_device->num_chn; i++)
 645                 netif_napi_del(&net_device->chan_table[i].napi);
 646
 647         /*
 648          * At this point, no one should be accessing net_device
 649          * except in here
 650          */
 651         netdev_dbg(ndev, "net device safe to remove\n");
 652
 653         /* Now, we can close the channel safely */
 654         vmbus_close(device->channel);
 655
 656         /*
 657          * If host is Win2016 or higher then we do the GPADL tear down
 658          * here after VMBus is closed.
 659         */
 660         if (vmbus_proto_version >= VERSION_WIN10) {
 661                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 662                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 663         }
 664
 665         /* Release all resources */
 666         free_netvsc_device_rcu(net_device);
 667 }
 668
 669 #define RING_AVAIL_PERCENT_HIWATER 20
 670 #define RING_AVAIL_PERCENT_LOWATER 10
 671
 672 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 673                                          u32 index)
 674 {
 675         sync_change_bit(index, net_device->send_section_map);
 676 }
 677
 678 static void netvsc_send_tx_complete(struct net_device *ndev,
 679                                     struct netvsc_device *net_device,
 680                                     struct vmbus_channel *channel,
 681                                     const struct vmpacket_descriptor *desc,
 682                                     int budget)
 683 {
 684         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 685         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 686         u16 q_idx = 0;
 687         int queue_sends;
 688
 689         /* Notify the layer above us */
 690         if (likely(skb)) {
 691                 const struct hv_netvsc_packet *packet
 692                         = (struct hv_netvsc_packet *)skb->cb;
 693                 u32 send_index = packet->send_buf_index;
 694                 struct netvsc_stats *tx_stats;
 695
 696                 if (send_index != NETVSC_INVALID_INDEX)
 697                         netvsc_free_send_slot(net_device, send_index);
 698                 q_idx = packet->q_idx;
 699
 700                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 701
 702                 u64_stats_update_begin(&tx_stats->syncp);
 703                 tx_stats->packets += packet->total_packets;
 704                 tx_stats->bytes += packet->total_bytes;
 705                 u64_stats_update_end(&tx_stats->syncp);
 706
 707                 napi_consume_skb(skb, budget);
 708         }
 709
 710         queue_sends =
 711                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 712
 713         if (unlikely(net_device->destroy)) {
 714                 if (queue_sends == 0)
 715                         wake_up(&net_device->wait_drain);
 716         } else {
 717                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
 718
 719                 if (netif_tx_queue_stopped(txq) &&
 720                     (hv_get_avail_to_write_percent(&channel->outbound) >
 721                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
 722                         netif_tx_wake_queue(txq);
 723                         ndev_ctx->eth_stats.wake_queue++;
 724                 }
 725         }
 726 }
 727
 728 static void netvsc_send_completion(struct net_device *ndev,
 729                                    struct netvsc_device *net_device,
 730                                    struct vmbus_channel *incoming_channel,
 731                                    const struct vmpacket_descriptor *desc,
 732                                    int budget)
 733 {
 734         const struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 735
 736         switch (nvsp_packet->hdr.msg_type) {
 737         case NVSP_MSG_TYPE_INIT_COMPLETE:
 738         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 739         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 740         case NVSP_MSG5_TYPE_SUBCHANNEL:
 741                 /* Copy the response back */
 742                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 743                        sizeof(struct nvsp_message));
 744                 complete(&net_device->channel_init_wait);
 745                 break;
 746
 747         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 748                 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
 749                                         desc, budget);
 750                 break;
 751
 752         default:
 753                 netdev_err(ndev,
 754                            "Unknown send completion type %d received!!\n",
 755                            nvsp_packet->hdr.msg_type);
 756         }
 757 }
 758
 759 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 760 {
 761         unsigned long *map_addr = net_device->send_section_map;
 762         unsigned int i;
 763
 764         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 765                 if (sync_test_and_set_bit(i, map_addr) == 0)
 766                         return i;
 767         }
 768
 769         return NETVSC_INVALID_INDEX;
 770 }
 771
 772 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 773                                     unsigned int section_index,
 774                                     u32 pend_size,
 775                                     struct hv_netvsc_packet *packet,
 776                                     struct rndis_message *rndis_msg,
 777                                     struct hv_page_buffer *pb,
 778                                     bool xmit_more)
 779 {
 780         char *start = net_device->send_buf;
 781         char *dest = start + (section_index * net_device->send_section_size)
 782                      + pend_size;
 783         int i;
 784         u32 padding = 0;
 785         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 786                 packet->page_buf_cnt;
 787         u32 remain;
 788
 789         /* Add padding */
 790         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
 791         if (xmit_more && remain) {
 792                 padding = net_device->pkt_align - remain;
 793                 rndis_msg->msg_len += padding;
 794                 packet->total_data_buflen += padding;
 795         }
 796
 797         for (i = 0; i < page_count; i++) {
 798                 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
 799                 u32 offset = pb[i].offset;
 800                 u32 len = pb[i].len;
 801
 802                 memcpy(dest, (src + offset), len);
 803                 dest += len;
 804         }
 805
 806         if (padding)
 807                 memset(dest, 0, padding);
 808 }
 809
 810 static inline int netvsc_send_pkt(
 811         struct hv_device *device,
 812         struct hv_netvsc_packet *packet,
 813         struct netvsc_device *net_device,
 814         struct hv_page_buffer *pb,
 815         struct sk_buff *skb)
 816 {
 817         struct nvsp_message nvmsg;
 818         struct nvsp_1_message_send_rndis_packet *rpkt =
 819                 &nvmsg.msg.v1_msg.send_rndis_pkt;
 820         struct netvsc_channel * const nvchan =
 821                 &net_device->chan_table[packet->q_idx];
 822         struct vmbus_channel *out_channel = nvchan->channel;
 823         struct net_device *ndev = hv_get_drvdata(device);
 824         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 825         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 826         u64 req_id;
 827         int ret;
 828         u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
 829
 830         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 831         if (skb)
 832                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
 833         else
 834                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
 835
 836         rpkt->send_buf_section_index = packet->send_buf_index;
 837         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 838                 rpkt->send_buf_section_size = 0;
 839         else
 840                 rpkt->send_buf_section_size = packet->total_data_buflen;
 841
 842         req_id = (ulong)skb;
 843
 844         if (out_channel->rescind)
 845                 return -ENODEV;
 846
 847         trace_nvsp_send_pkt(ndev, out_channel, rpkt);
 848
 849         if (packet->page_buf_cnt) {
 850                 if (packet->cp_partial)
 851                         pb += packet->rmsg_pgcnt;
 852
 853                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 854                                                   pb, packet->page_buf_cnt,
 855                                                   &nvmsg, sizeof(nvmsg),
 856                                                   req_id);
 857         } else {
 858                 ret = vmbus_sendpacket(out_channel,
 859                                        &nvmsg, sizeof(nvmsg),
 860                                        req_id, VM_PKT_DATA_INBAND,
 861                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 862         }
 863
 864         if (ret == 0) {
 865                 atomic_inc_return(&nvchan->queue_sends);
 866
 867                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 868                         netif_tx_stop_queue(txq);
 869                         ndev_ctx->eth_stats.stop_queue++;
 870                 }
 871         } else if (ret == -EAGAIN) {
 872                 netif_tx_stop_queue(txq);
 873                 ndev_ctx->eth_stats.stop_queue++;
 874                 if (atomic_read(&nvchan->queue_sends) < 1) {
 875                         netif_tx_wake_queue(txq);
 876                         ndev_ctx->eth_stats.wake_queue++;
 877                         ret = -ENOSPC;
 878                 }
 879         } else {
 880                 netdev_err(ndev,
 881                            "Unable to send packet pages %u len %u, ret %d\n",
 882                            packet->page_buf_cnt, packet->total_data_buflen,
 883                            ret);
 884         }
 885
 886         return ret;
 887 }
 888
 889 /* Move packet out of multi send data (msd), and clear msd */
 890 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 891                                 struct sk_buff **msd_skb,
 892                                 struct multi_send_data *msdp)
 893 {
 894         *msd_skb = msdp->skb;
 895         *msd_send = msdp->pkt;
 896         msdp->skb = NULL;
 897         msdp->pkt = NULL;
 898         msdp->count = 0;
 899 }
 900
 901 /* RCU already held by caller */
 902 int netvsc_send(struct net_device *ndev,
 903                 struct hv_netvsc_packet *packet,
 904                 struct rndis_message *rndis_msg,
 905                 struct hv_page_buffer *pb,
 906                 struct sk_buff *skb)
 907 {
 908         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 909         struct netvsc_device *net_device
 910                 = rcu_dereference_bh(ndev_ctx->nvdev);
 911         struct hv_device *device = ndev_ctx->device_ctx;
 912         int ret = 0;
 913         struct netvsc_channel *nvchan;
 914         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 915         unsigned int section_index = NETVSC_INVALID_INDEX;
 916         struct multi_send_data *msdp;
 917         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 918         struct sk_buff *msd_skb = NULL;
 919         bool try_batch, xmit_more;
 920
 921         /* If device is rescinded, return error and packet will get dropped. */
 922         if (unlikely(!net_device || net_device->destroy))
 923                 return -ENODEV;
 924
 925         nvchan = &net_device->chan_table[packet->q_idx];
 926         packet->send_buf_index = NETVSC_INVALID_INDEX;
 927         packet->cp_partial = false;
 928
 929         /* Send control message directly without accessing msd (Multi-Send
 930          * Data) field which may be changed during data packet processing.
 931          */
 932         if (!skb)
 933                 return netvsc_send_pkt(device, packet, net_device, pb, skb);
 934
 935         /* batch packets in send buffer if possible */
 936         msdp = &nvchan->msd;
 937         if (msdp->pkt)
 938                 msd_len = msdp->pkt->total_data_buflen;
 939
 940         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
 941         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 942             net_device->send_section_size) {
 943                 section_index = msdp->pkt->send_buf_index;
 944
 945         } else if (try_batch && msd_len + packet->rmsg_size <
 946                    net_device->send_section_size) {
 947                 section_index = msdp->pkt->send_buf_index;
 948                 packet->cp_partial = true;
 949
 950         } else if (pktlen + net_device->pkt_align <
 951                    net_device->send_section_size) {
 952                 section_index = netvsc_get_next_send_section(net_device);
 953                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
 954                         ++ndev_ctx->eth_stats.tx_send_full;
 955                 } else {
 956                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 957                         msd_len = 0;
 958                 }
 959         }
 960
 961         /* Keep aggregating only if stack says more data is coming
 962          * and not doing mixed modes send and not flow blocked
 963          */
 964         xmit_more = skb->xmit_more &&
 965                 !packet->cp_partial &&
 966                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
 967
 968         if (section_index != NETVSC_INVALID_INDEX) {
 969                 netvsc_copy_to_send_buf(net_device,
 970                                         section_index, msd_len,
 971                                         packet, rndis_msg, pb, xmit_more);
 972
 973                 packet->send_buf_index = section_index;
 974
 975                 if (packet->cp_partial) {
 976                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 977                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 978                 } else {
 979                         packet->page_buf_cnt = 0;
 980                         packet->total_data_buflen += msd_len;
 981                 }
 982
 983                 if (msdp->pkt) {
 984                         packet->total_packets += msdp->pkt->total_packets;
 985                         packet->total_bytes += msdp->pkt->total_bytes;
 986                 }
 987
 988                 if (msdp->skb)
 989                         dev_consume_skb_any(msdp->skb);
 990
 991                 if (xmit_more) {
 992                         msdp->skb = skb;
 993                         msdp->pkt = packet;
 994                         msdp->count++;
 995                 } else {
 996                         cur_send = packet;
 997                         msdp->skb = NULL;
 998                         msdp->pkt = NULL;
 999                         msdp->count = 0;
1000                 }
1001         } else {
1002                 move_pkt_msd(&msd_send, &msd_skb, msdp);
1003                 cur_send = packet;
1004         }
1005
1006         if (msd_send) {
1007                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1008                                             NULL, msd_skb);
1009
1010                 if (m_ret != 0) {
1011                         netvsc_free_send_slot(net_device,
1012                                               msd_send->send_buf_index);
1013                         dev_kfree_skb_any(msd_skb);
1014                 }
1015         }
1016
1017         if (cur_send)
1018                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1019
1020         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1021                 netvsc_free_send_slot(net_device, section_index);
1022
1023         return ret;
1024 }
1025
1026 /* Send pending recv completions */
1027 static int send_recv_completions(struct net_device *ndev,
1028                                  struct netvsc_device *nvdev,
1029                                  struct netvsc_channel *nvchan)
1030 {
1031         struct multi_recv_comp *mrc = &nvchan->mrc;
1032         struct recv_comp_msg {
1033                 struct nvsp_message_header hdr;
1034                 u32 status;
1035         }  __packed;
1036         struct recv_comp_msg msg = {
1037                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1038         };
1039         int ret;
1040
1041         while (mrc->first != mrc->next) {
1042                 const struct recv_comp_data *rcd
1043                         = mrc->slots + mrc->first;
1044
1045                 msg.status = rcd->status;
1046                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1047                                        rcd->tid, VM_PKT_COMP, 0);
1048                 if (unlikely(ret)) {
1049                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1050
1051                         ++ndev_ctx->eth_stats.rx_comp_busy;
1052                         return ret;
1053                 }
1054
1055                 if (++mrc->first == nvdev->recv_completion_cnt)
1056                         mrc->first = 0;
1057         }
1058
1059         /* receive completion ring has been emptied */
1060         if (unlikely(nvdev->destroy))
1061                 wake_up(&nvdev->wait_drain);
1062
1063         return 0;
1064 }
1065
1066 /* Count how many receive completions are outstanding */
1067 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1068                                  const struct multi_recv_comp *mrc,
1069                                  u32 *filled, u32 *avail)
1070 {
1071         u32 count = nvdev->recv_completion_cnt;
1072
1073         if (mrc->next >= mrc->first)
1074                 *filled = mrc->next - mrc->first;
1075         else
1076                 *filled = (count - mrc->first) + mrc->next;
1077
1078         *avail = count - *filled - 1;
1079 }
1080
1081 /* Add receive complete to ring to send to host. */
1082 static void enq_receive_complete(struct net_device *ndev,
1083                                  struct netvsc_device *nvdev, u16 q_idx,
1084                                  u64 tid, u32 status)
1085 {
1086         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1087         struct multi_recv_comp *mrc = &nvchan->mrc;
1088         struct recv_comp_data *rcd;
1089         u32 filled, avail;
1090
1091         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1092
1093         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1094                 send_recv_completions(ndev, nvdev, nvchan);
1095                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1096         }
1097
1098         if (unlikely(!avail)) {
1099                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1100                            q_idx, tid);
1101                 return;
1102         }
1103
1104         rcd = mrc->slots + mrc->next;
1105         rcd->tid = tid;
1106         rcd->status = status;
1107
1108         if (++mrc->next == nvdev->recv_completion_cnt)
1109                 mrc->next = 0;
1110 }
1111
1112 static int netvsc_receive(struct net_device *ndev,
1113                           struct netvsc_device *net_device,
1114                           struct vmbus_channel *channel,
1115                           const struct vmpacket_descriptor *desc,
1116                           const struct nvsp_message *nvsp)
1117 {
1118         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1119         const struct vmtransfer_page_packet_header *vmxferpage_packet
1120                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1121         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1122         char *recv_buf = net_device->recv_buf;
1123         u32 status = NVSP_STAT_SUCCESS;
1124         int i;
1125         int count = 0;
1126
1127         /* Make sure this is a valid nvsp packet */
1128         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1129                 netif_err(net_device_ctx, rx_err, ndev,
1130                           "Unknown nvsp packet type received %u\n",
1131                           nvsp->hdr.msg_type);
1132                 return 0;
1133         }
1134
1135         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1136                 netif_err(net_device_ctx, rx_err, ndev,
1137                           "Invalid xfer page set id - expecting %x got %x\n",
1138                           NETVSC_RECEIVE_BUFFER_ID,
1139                           vmxferpage_packet->xfer_pageset_id);
1140                 return 0;
1141         }
1142
1143         count = vmxferpage_packet->range_cnt;
1144
1145         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1146         for (i = 0; i < count; i++) {
1147                 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1148                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1149                 void *data;
1150                 int ret;
1151
1152                 if (unlikely(offset + buflen > net_device->recv_buf_size)) {
1153                         status = NVSP_STAT_FAIL;
1154                         netif_err(net_device_ctx, rx_err, ndev,
1155                                   "Packet offset:%u + len:%u too big\n",
1156                                   offset, buflen);
1157
1158                         continue;
1159                 }
1160
1161                 data = recv_buf + offset;
1162
1163                 trace_rndis_recv(ndev, q_idx, data);
1164
1165                 /* Pass it to the upper layer */
1166                 ret = rndis_filter_receive(ndev, net_device,
1167                                            channel, data, buflen);
1168
1169                 if (unlikely(ret != NVSP_STAT_SUCCESS))
1170                         status = NVSP_STAT_FAIL;
1171         }
1172
1173         enq_receive_complete(ndev, net_device, q_idx,
1174                              vmxferpage_packet->d.trans_id, status);
1175
1176         return count;
1177 }
1178
1179 static void netvsc_send_table(struct net_device *ndev,
1180                               const struct nvsp_message *nvmsg)
1181 {
1182         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1183         u32 count, *tab;
1184         int i;
1185
1186         count = nvmsg->msg.v5_msg.send_table.count;
1187         if (count != VRSS_SEND_TAB_SIZE) {
1188                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1189                 return;
1190         }
1191
1192         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1193                       nvmsg->msg.v5_msg.send_table.offset);
1194
1195         for (i = 0; i < count; i++)
1196                 net_device_ctx->tx_table[i] = tab[i];
1197 }
1198
1199 static void netvsc_send_vf(struct net_device *ndev,
1200                            const struct nvsp_message *nvmsg)
1201 {
1202         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1203
1204         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1205         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1206         netdev_info(ndev, "VF slot %u %s\n",
1207                     net_device_ctx->vf_serial,
1208                     net_device_ctx->vf_alloc ? "added" : "removed");
1209 }
1210
1211 static  void netvsc_receive_inband(struct net_device *ndev,
1212                                    const struct nvsp_message *nvmsg)
1213 {
1214         switch (nvmsg->hdr.msg_type) {
1215         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1216                 netvsc_send_table(ndev, nvmsg);
1217                 break;
1218
1219         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1220                 netvsc_send_vf(ndev, nvmsg);
1221                 break;
1222         }
1223 }
1224
1225 static int netvsc_process_raw_pkt(struct hv_device *device,
1226                                   struct vmbus_channel *channel,
1227                                   struct netvsc_device *net_device,
1228                                   struct net_device *ndev,
1229                                   const struct vmpacket_descriptor *desc,
1230                                   int budget)
1231 {
1232         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1233
1234         trace_nvsp_recv(ndev, channel, nvmsg);
1235
1236         switch (desc->type) {
1237         case VM_PKT_COMP:
1238                 netvsc_send_completion(ndev, net_device, channel,
1239                                        desc, budget);
1240                 break;
1241
1242         case VM_PKT_DATA_USING_XFER_PAGES:
1243                 return netvsc_receive(ndev, net_device, channel,
1244                                       desc, nvmsg);
1245                 break;
1246
1247         case VM_PKT_DATA_INBAND:
1248                 netvsc_receive_inband(ndev, nvmsg);
1249                 break;
1250
1251         default:
1252                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1253                            desc->type, desc->trans_id);
1254                 break;
1255         }
1256
1257         return 0;
1258 }
1259
1260 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1261 {
1262         struct vmbus_channel *primary = channel->primary_channel;
1263
1264         return primary ? primary->device_obj : channel->device_obj;
1265 }
1266
1267 /* Network processing softirq
1268  * Process data in incoming ring buffer from host
1269  * Stops when ring is empty or budget is met or exceeded.
1270  */
1271 int netvsc_poll(struct napi_struct *napi, int budget)
1272 {
1273         struct netvsc_channel *nvchan
1274                 = container_of(napi, struct netvsc_channel, napi);
1275         struct netvsc_device *net_device = nvchan->net_device;
1276         struct vmbus_channel *channel = nvchan->channel;
1277         struct hv_device *device = netvsc_channel_to_device(channel);
1278         struct net_device *ndev = hv_get_drvdata(device);
1279         int work_done = 0;
1280         int ret;
1281
1282         /* If starting a new interval */
1283         if (!nvchan->desc)
1284                 nvchan->desc = hv_pkt_iter_first(channel);
1285
1286         while (nvchan->desc && work_done < budget) {
1287                 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1288                                                     ndev, nvchan->desc, budget);
1289                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1290         }
1291
1292         /* Send any pending receive completions */
1293         ret = send_recv_completions(ndev, net_device, nvchan);
1294
1295         /* If it did not exhaust NAPI budget this time
1296          *  and not doing busy poll
1297          * then re-enable host interrupts
1298          *  and reschedule if ring is not empty
1299          *   or sending receive completion failed.
1300          */
1301         if (work_done < budget &&
1302             napi_complete_done(napi, work_done) &&
1303             (ret || hv_end_read(&channel->inbound)) &&
1304             napi_schedule_prep(napi)) {
1305                 hv_begin_read(&channel->inbound);
1306                 __napi_schedule(napi);
1307         }
1308
1309         /* Driver may overshoot since multiple packets per descriptor */
1310         return min(work_done, budget);
1311 }
1312
1313 /* Call back when data is available in host ring buffer.
1314  * Processing is deferred until network softirq (NAPI)
1315  */
1316 void netvsc_channel_cb(void *context)
1317 {
1318         struct netvsc_channel *nvchan = context;
1319         struct vmbus_channel *channel = nvchan->channel;
1320         struct hv_ring_buffer_info *rbi = &channel->inbound;
1321
1322         /* preload first vmpacket descriptor */
1323         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1324
1325         if (napi_schedule_prep(&nvchan->napi)) {
1326                 /* disable interupts from host */
1327                 hv_begin_read(rbi);
1328
1329                 __napi_schedule_irqoff(&nvchan->napi);
1330         }
1331 }
1332
1333 /*
1334  * netvsc_device_add - Callback when the device belonging to this
1335  * driver is added
1336  */
1337 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1338                                 const struct netvsc_device_info *device_info)
1339 {
1340         int i, ret = 0;
1341         struct netvsc_device *net_device;
1342         struct net_device *ndev = hv_get_drvdata(device);
1343         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1344
1345         net_device = alloc_net_device();
1346         if (!net_device)
1347                 return ERR_PTR(-ENOMEM);
1348
1349         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1350                 net_device_ctx->tx_table[i] = 0;
1351
1352         /* Because the device uses NAPI, all the interrupt batching and
1353          * control is done via Net softirq, not the channel handling
1354          */
1355         set_channel_read_mode(device->channel, HV_CALL_ISR);
1356
1357         /* If we're reopening the device we may have multiple queues, fill the
1358          * chn_table with the default channel to use it before subchannels are
1359          * opened.
1360          * Initialize the channel state before we open;
1361          * we can be interrupted as soon as we open the channel.
1362          */
1363
1364         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1365                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1366
1367                 nvchan->channel = device->channel;
1368                 nvchan->net_device = net_device;
1369                 u64_stats_init(&nvchan->tx_stats.syncp);
1370                 u64_stats_init(&nvchan->rx_stats.syncp);
1371         }
1372
1373         /* Enable NAPI handler before init callbacks */
1374         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1375                        netvsc_poll, NAPI_POLL_WEIGHT);
1376
1377         /* Open the channel */
1378         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1379                          netvsc_ring_bytes,  NULL, 0,
1380                          netvsc_channel_cb, net_device->chan_table);
1381
1382         if (ret != 0) {
1383                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1384                 goto cleanup;
1385         }
1386
1387         /* Channel is opened */
1388         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1389
1390         napi_enable(&net_device->chan_table[0].napi);
1391
1392         /* Connect with the NetVsp */
1393         ret = netvsc_connect_vsp(device, net_device, device_info);
1394         if (ret != 0) {
1395                 netdev_err(ndev,
1396                         "unable to connect to NetVSP - %d\n", ret);
1397                 goto close;
1398         }
1399
1400         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1401          * populated.
1402          */
1403         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1404
1405         return net_device;
1406
1407 close:
1408         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1409         napi_disable(&net_device->chan_table[0].napi);
1410
1411         /* Now, we can close the channel safely */
1412         vmbus_close(device->channel);
1413
1414 cleanup:
1415         netif_napi_del(&net_device->chan_table[0].napi);
1416         free_netvsc_device(&net_device->rcu);
1417
1418         return ERR_PTR(ret);
1419 }