drivers/net/hyperv/netvsc.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <[email protected]>
  18  *   Hank Janssen  <[email protected]>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/wait.h>
  25 #include <linux/mm.h>
  26 #include <linux/delay.h>
  27 #include <linux/io.h>
  28 #include <linux/slab.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/rtnetlink.h>
  33 #include <linux/prefetch.h>
  34 #include <linux/reciprocal_div.h>
  35
  36 #include <asm/sync_bitops.h>
  37
  38 #include "hyperv_net.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         vmbus_sendpacket(dev->channel, init_pkt,
  61                                sizeof(struct nvsp_message),
  62                                (unsigned long)init_pkt,
  63                                VM_PKT_DATA_INBAND, 0);
  64 }
  65
  66 static struct netvsc_device *alloc_net_device(void)
  67 {
  68         struct netvsc_device *net_device;
  69
  70         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  71         if (!net_device)
  72                 return NULL;
  73
  74         init_waitqueue_head(&net_device->wait_drain);
  75         net_device->destroy = false;
  76
  77         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
  78         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
  79
  80         init_completion(&net_device->channel_init_wait);
  81         init_waitqueue_head(&net_device->subchan_open);
  82         INIT_WORK(&net_device->subchan_work, rndis_set_subchannel);
  83
  84         return net_device;
  85 }
  86
  87 static void free_netvsc_device(struct rcu_head *head)
  88 {
  89         struct netvsc_device *nvdev
  90                 = container_of(head, struct netvsc_device, rcu);
  91         int i;
  92
  93         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
  94                 vfree(nvdev->chan_table[i].mrc.slots);
  95
  96         kfree(nvdev);
  97 }
  98
  99 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
 100 {
 101         call_rcu(&nvdev->rcu, free_netvsc_device);
 102 }
 103
 104 static void netvsc_revoke_buf(struct hv_device *device,
 105                               struct netvsc_device *net_device)
 106 {
 107         struct nvsp_message *revoke_packet;
 108         struct net_device *ndev = hv_get_drvdata(device);
 109         int ret;
 110
 111         /*
 112          * If we got a section count, it means we received a
 113          * SendReceiveBufferComplete msg (ie sent
 114          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 115          * to send a revoke msg here
 116          */
 117         if (net_device->recv_section_cnt) {
 118                 /* Send the revoke receive buffer */
 119                 revoke_packet = &net_device->revoke_packet;
 120                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 121
 122                 revoke_packet->hdr.msg_type =
 123                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 124                 revoke_packet->msg.v1_msg.
 125                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 126
 127                 ret = vmbus_sendpacket(device->channel,
 128                                        revoke_packet,
 129                                        sizeof(struct nvsp_message),
 130                                        (unsigned long)revoke_packet,
 131                                        VM_PKT_DATA_INBAND, 0);
 132                 /* If the failure is because the channel is rescinded;
 133                  * ignore the failure since we cannot send on a rescinded
 134                  * channel. This would allow us to properly cleanup
 135                  * even when the channel is rescinded.
 136                  */
 137                 if (device->channel->rescind)
 138                         ret = 0;
 139                 /*
 140                  * If we failed here, we might as well return and
 141                  * have a leak rather than continue and a bugchk
 142                  */
 143                 if (ret != 0) {
 144                         netdev_err(ndev, "unable to send "
 145                                 "revoke receive buffer to netvsp\n");
 146                         return;
 147                 }
 148                 net_device->recv_section_cnt = 0;
 149         }
 150
 151         /* Deal with the send buffer we may have setup.
 152          * If we got a  send section size, it means we received a
 153          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 154          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 155          * to send a revoke msg here
 156          */
 157         if (net_device->send_section_cnt) {
 158                 /* Send the revoke receive buffer */
 159                 revoke_packet = &net_device->revoke_packet;
 160                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 161
 162                 revoke_packet->hdr.msg_type =
 163                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 164                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 165                         NETVSC_SEND_BUFFER_ID;
 166
 167                 ret = vmbus_sendpacket(device->channel,
 168                                        revoke_packet,
 169                                        sizeof(struct nvsp_message),
 170                                        (unsigned long)revoke_packet,
 171                                        VM_PKT_DATA_INBAND, 0);
 172
 173                 /* If the failure is because the channel is rescinded;
 174                  * ignore the failure since we cannot send on a rescinded
 175                  * channel. This would allow us to properly cleanup
 176                  * even when the channel is rescinded.
 177                  */
 178                 if (device->channel->rescind)
 179                         ret = 0;
 180
 181                 /* If we failed here, we might as well return and
 182                  * have a leak rather than continue and a bugchk
 183                  */
 184                 if (ret != 0) {
 185                         netdev_err(ndev, "unable to send "
 186                                    "revoke send buffer to netvsp\n");
 187                         return;
 188                 }
 189                 net_device->send_section_cnt = 0;
 190         }
 191 }
 192
 193 static void netvsc_teardown_gpadl(struct hv_device *device,
 194                                   struct netvsc_device *net_device)
 195 {
 196         struct net_device *ndev = hv_get_drvdata(device);
 197         int ret;
 198
 199         if (net_device->recv_buf_gpadl_handle) {
 200                 ret = vmbus_teardown_gpadl(device->channel,
 201                                            net_device->recv_buf_gpadl_handle);
 202
 203                 /* If we failed here, we might as well return and have a leak
 204                  * rather than continue and a bugchk
 205                  */
 206                 if (ret != 0) {
 207                         netdev_err(ndev,
 208                                    "unable to teardown receive buffer's gpadl\n");
 209                         return;
 210                 }
 211                 net_device->recv_buf_gpadl_handle = 0;
 212         }
 213
 214         if (net_device->recv_buf) {
 215                 /* Free up the receive buffer */
 216                 vfree(net_device->recv_buf);
 217                 net_device->recv_buf = NULL;
 218         }
 219
 220         if (net_device->send_buf_gpadl_handle) {
 221                 ret = vmbus_teardown_gpadl(device->channel,
 222                                            net_device->send_buf_gpadl_handle);
 223
 224                 /* If we failed here, we might as well return and have a leak
 225                  * rather than continue and a bugchk
 226                  */
 227                 if (ret != 0) {
 228                         netdev_err(ndev,
 229                                    "unable to teardown send buffer's gpadl\n");
 230                         return;
 231                 }
 232                 net_device->send_buf_gpadl_handle = 0;
 233         }
 234         if (net_device->send_buf) {
 235                 /* Free up the send buffer */
 236                 vfree(net_device->send_buf);
 237                 net_device->send_buf = NULL;
 238         }
 239         kfree(net_device->send_section_map);
 240 }
 241
 242 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
 243 {
 244         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 245         int node = cpu_to_node(nvchan->channel->target_cpu);
 246         size_t size;
 247
 248         size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
 249         nvchan->mrc.slots = vzalloc_node(size, node);
 250         if (!nvchan->mrc.slots)
 251                 nvchan->mrc.slots = vzalloc(size);
 252
 253         return nvchan->mrc.slots ? 0 : -ENOMEM;
 254 }
 255
 256 static int netvsc_init_buf(struct hv_device *device,
 257                            struct netvsc_device *net_device,
 258                            const struct netvsc_device_info *device_info)
 259 {
 260         struct nvsp_1_message_send_receive_buffer_complete *resp;
 261         struct net_device *ndev = hv_get_drvdata(device);
 262         struct nvsp_message *init_packet;
 263         unsigned int buf_size;
 264         size_t map_words;
 265         int ret = 0;
 266
 267         /* Get receive buffer area. */
 268         buf_size = device_info->recv_sections * device_info->recv_section_size;
 269         buf_size = roundup(buf_size, PAGE_SIZE);
 270
 271         /* Legacy hosts only allow smaller receive buffer */
 272         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 273                 buf_size = min_t(unsigned int, buf_size,
 274                                  NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
 275
 276         net_device->recv_buf = vzalloc(buf_size);
 277         if (!net_device->recv_buf) {
 278                 netdev_err(ndev,
 279                            "unable to allocate receive buffer of size %u\n",
 280                            buf_size);
 281                 ret = -ENOMEM;
 282                 goto cleanup;
 283         }
 284
 285         /*
 286          * Establish the gpadl handle for this buffer on this
 287          * channel.  Note: This call uses the vmbus connection rather
 288          * than the channel to establish the gpadl handle.
 289          */
 290         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 291                                     buf_size,
 292                                     &net_device->recv_buf_gpadl_handle);
 293         if (ret != 0) {
 294                 netdev_err(ndev,
 295                         "unable to establish receive buffer's gpadl\n");
 296                 goto cleanup;
 297         }
 298
 299         /* Notify the NetVsp of the gpadl handle */
 300         init_packet = &net_device->channel_init_pkt;
 301         memset(init_packet, 0, sizeof(struct nvsp_message));
 302         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 303         init_packet->msg.v1_msg.send_recv_buf.
 304                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 305         init_packet->msg.v1_msg.
 306                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 307
 308         /* Send the gpadl notification request */
 309         ret = vmbus_sendpacket(device->channel, init_packet,
 310                                sizeof(struct nvsp_message),
 311                                (unsigned long)init_packet,
 312                                VM_PKT_DATA_INBAND,
 313                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 314         if (ret != 0) {
 315                 netdev_err(ndev,
 316                         "unable to send receive buffer's gpadl to netvsp\n");
 317                 goto cleanup;
 318         }
 319
 320         wait_for_completion(&net_device->channel_init_wait);
 321
 322         /* Check the response */
 323         resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
 324         if (resp->status != NVSP_STAT_SUCCESS) {
 325                 netdev_err(ndev,
 326                            "Unable to complete receive buffer initialization with NetVsp - status %d\n",
 327                            resp->status);
 328                 ret = -EINVAL;
 329                 goto cleanup;
 330         }
 331
 332         /* Parse the response */
 333         netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
 334                    resp->num_sections, resp->sections[0].sub_alloc_size,
 335                    resp->sections[0].num_sub_allocs);
 336
 337         /* There should only be one section for the entire receive buffer */
 338         if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
 339                 ret = -EINVAL;
 340                 goto cleanup;
 341         }
 342
 343         net_device->recv_section_size = resp->sections[0].sub_alloc_size;
 344         net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
 345
 346         /* Setup receive completion ring */
 347         net_device->recv_completion_cnt
 348                 = round_up(net_device->recv_section_cnt + 1,
 349                            PAGE_SIZE / sizeof(u64));
 350         ret = netvsc_alloc_recv_comp_ring(net_device, 0);
 351         if (ret)
 352                 goto cleanup;
 353
 354         /* Now setup the send buffer. */
 355         buf_size = device_info->send_sections * device_info->send_section_size;
 356         buf_size = round_up(buf_size, PAGE_SIZE);
 357
 358         net_device->send_buf = vzalloc(buf_size);
 359         if (!net_device->send_buf) {
 360                 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
 361                            buf_size);
 362                 ret = -ENOMEM;
 363                 goto cleanup;
 364         }
 365
 366         /* Establish the gpadl handle for this buffer on this
 367          * channel.  Note: This call uses the vmbus connection rather
 368          * than the channel to establish the gpadl handle.
 369          */
 370         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 371                                     buf_size,
 372                                     &net_device->send_buf_gpadl_handle);
 373         if (ret != 0) {
 374                 netdev_err(ndev,
 375                            "unable to establish send buffer's gpadl\n");
 376                 goto cleanup;
 377         }
 378
 379         /* Notify the NetVsp of the gpadl handle */
 380         init_packet = &net_device->channel_init_pkt;
 381         memset(init_packet, 0, sizeof(struct nvsp_message));
 382         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 383         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 384                 net_device->send_buf_gpadl_handle;
 385         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 386
 387         /* Send the gpadl notification request */
 388         ret = vmbus_sendpacket(device->channel, init_packet,
 389                                sizeof(struct nvsp_message),
 390                                (unsigned long)init_packet,
 391                                VM_PKT_DATA_INBAND,
 392                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 393         if (ret != 0) {
 394                 netdev_err(ndev,
 395                            "unable to send send buffer's gpadl to netvsp\n");
 396                 goto cleanup;
 397         }
 398
 399         wait_for_completion(&net_device->channel_init_wait);
 400
 401         /* Check the response */
 402         if (init_packet->msg.v1_msg.
 403             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 404                 netdev_err(ndev, "Unable to complete send buffer "
 405                            "initialization with NetVsp - status %d\n",
 406                            init_packet->msg.v1_msg.
 407                            send_send_buf_complete.status);
 408                 ret = -EINVAL;
 409                 goto cleanup;
 410         }
 411
 412         /* Parse the response */
 413         net_device->send_section_size = init_packet->msg.
 414                                 v1_msg.send_send_buf_complete.section_size;
 415
 416         /* Section count is simply the size divided by the section size. */
 417         net_device->send_section_cnt = buf_size / net_device->send_section_size;
 418
 419         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 420                    net_device->send_section_size, net_device->send_section_cnt);
 421
 422         /* Setup state for managing the send buffer. */
 423         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
 424
 425         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
 426         if (net_device->send_section_map == NULL) {
 427                 ret = -ENOMEM;
 428                 goto cleanup;
 429         }
 430
 431         goto exit;
 432
 433 cleanup:
 434         netvsc_revoke_buf(device, net_device);
 435         netvsc_teardown_gpadl(device, net_device);
 436
 437 exit:
 438         return ret;
 439 }
 440
 441 /* Negotiate NVSP protocol version */
 442 static int negotiate_nvsp_ver(struct hv_device *device,
 443                               struct netvsc_device *net_device,
 444                               struct nvsp_message *init_packet,
 445                               u32 nvsp_ver)
 446 {
 447         struct net_device *ndev = hv_get_drvdata(device);
 448         int ret;
 449
 450         memset(init_packet, 0, sizeof(struct nvsp_message));
 451         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 452         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 453         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 454
 455         /* Send the init request */
 456         ret = vmbus_sendpacket(device->channel, init_packet,
 457                                sizeof(struct nvsp_message),
 458                                (unsigned long)init_packet,
 459                                VM_PKT_DATA_INBAND,
 460                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 461
 462         if (ret != 0)
 463                 return ret;
 464
 465         wait_for_completion(&net_device->channel_init_wait);
 466
 467         if (init_packet->msg.init_msg.init_complete.status !=
 468             NVSP_STAT_SUCCESS)
 469                 return -EINVAL;
 470
 471         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 472                 return 0;
 473
 474         /* NVSPv2 or later: Send NDIS config */
 475         memset(init_packet, 0, sizeof(struct nvsp_message));
 476         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 477         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 478         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 479
 480         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 481                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 482
 483                 /* Teaming bit is needed to receive link speed updates */
 484                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 485         }
 486
 487         ret = vmbus_sendpacket(device->channel, init_packet,
 488                                 sizeof(struct nvsp_message),
 489                                 (unsigned long)init_packet,
 490                                 VM_PKT_DATA_INBAND, 0);
 491
 492         return ret;
 493 }
 494
 495 static int netvsc_connect_vsp(struct hv_device *device,
 496                               struct netvsc_device *net_device,
 497                               const struct netvsc_device_info *device_info)
 498 {
 499         static const u32 ver_list[] = {
 500                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 501                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
 502         };
 503         struct nvsp_message *init_packet;
 504         int ndis_version, i, ret;
 505
 506         init_packet = &net_device->channel_init_pkt;
 507
 508         /* Negotiate the latest NVSP protocol supported */
 509         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 510                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 511                                        ver_list[i])  == 0) {
 512                         net_device->nvsp_version = ver_list[i];
 513                         break;
 514                 }
 515
 516         if (i < 0) {
 517                 ret = -EPROTO;
 518                 goto cleanup;
 519         }
 520
 521         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 522
 523         /* Send the ndis version */
 524         memset(init_packet, 0, sizeof(struct nvsp_message));
 525
 526         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 527                 ndis_version = 0x00060001;
 528         else
 529                 ndis_version = 0x0006001e;
 530
 531         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 532         init_packet->msg.v1_msg.
 533                 send_ndis_ver.ndis_major_ver =
 534                                 (ndis_version & 0xFFFF0000) >> 16;
 535         init_packet->msg.v1_msg.
 536                 send_ndis_ver.ndis_minor_ver =
 537                                 ndis_version & 0xFFFF;
 538
 539         /* Send the init request */
 540         ret = vmbus_sendpacket(device->channel, init_packet,
 541                                 sizeof(struct nvsp_message),
 542                                 (unsigned long)init_packet,
 543                                 VM_PKT_DATA_INBAND, 0);
 544         if (ret != 0)
 545                 goto cleanup;
 546
 547
 548         ret = netvsc_init_buf(device, net_device, device_info);
 549
 550 cleanup:
 551         return ret;
 552 }
 553
 554 /*
 555  * netvsc_device_remove - Callback when the root bus device is removed
 556  */
 557 void netvsc_device_remove(struct hv_device *device)
 558 {
 559         struct net_device *ndev = hv_get_drvdata(device);
 560         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 561         struct netvsc_device *net_device
 562                 = rtnl_dereference(net_device_ctx->nvdev);
 563         int i;
 564
 565         cancel_work_sync(&net_device->subchan_work);
 566
 567         netvsc_revoke_buf(device, net_device);
 568
 569         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 570
 571         /*
 572          * At this point, no one should be accessing net_device
 573          * except in here
 574          */
 575         netdev_dbg(ndev, "net device safe to remove\n");
 576
 577         /* Now, we can close the channel safely */
 578         vmbus_close(device->channel);
 579
 580         netvsc_teardown_gpadl(device, net_device);
 581
 582         /* And dissassociate NAPI context from device */
 583         for (i = 0; i < net_device->num_chn; i++)
 584                 netif_napi_del(&net_device->chan_table[i].napi);
 585
 586         /* Release all resources */
 587         free_netvsc_device_rcu(net_device);
 588 }
 589
 590 #define RING_AVAIL_PERCENT_HIWATER 20
 591 #define RING_AVAIL_PERCENT_LOWATER 10
 592
 593 /*
 594  * Get the percentage of available bytes to write in the ring.
 595  * The return value is in range from 0 to 100.
 596  */
 597 static u32 hv_ringbuf_avail_percent(const struct hv_ring_buffer_info *ring_info)
 598 {
 599         u32 avail_write = hv_get_bytes_to_write(ring_info);
 600
 601         return reciprocal_divide(avail_write  * 100, netvsc_ring_reciprocal);
 602 }
 603
 604 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 605                                          u32 index)
 606 {
 607         sync_change_bit(index, net_device->send_section_map);
 608 }
 609
 610 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
 611                                     struct vmbus_channel *incoming_channel,
 612                                     struct hv_device *device,
 613                                     const struct vmpacket_descriptor *desc,
 614                                     int budget)
 615 {
 616         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 617         struct net_device *ndev = hv_get_drvdata(device);
 618         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 619         struct vmbus_channel *channel = device->channel;
 620         u16 q_idx = 0;
 621         int queue_sends;
 622
 623         /* Notify the layer above us */
 624         if (likely(skb)) {
 625                 const struct hv_netvsc_packet *packet
 626                         = (struct hv_netvsc_packet *)skb->cb;
 627                 u32 send_index = packet->send_buf_index;
 628                 struct netvsc_stats *tx_stats;
 629
 630                 if (send_index != NETVSC_INVALID_INDEX)
 631                         netvsc_free_send_slot(net_device, send_index);
 632                 q_idx = packet->q_idx;
 633                 channel = incoming_channel;
 634
 635                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 636
 637                 u64_stats_update_begin(&tx_stats->syncp);
 638                 tx_stats->packets += packet->total_packets;
 639                 tx_stats->bytes += packet->total_bytes;
 640                 u64_stats_update_end(&tx_stats->syncp);
 641
 642                 napi_consume_skb(skb, budget);
 643         }
 644
 645         queue_sends =
 646                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 647
 648         if (net_device->destroy && queue_sends == 0)
 649                 wake_up(&net_device->wait_drain);
 650
 651         if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 652             (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
 653              queue_sends < 1)) {
 654                 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
 655                 ndev_ctx->eth_stats.wake_queue++;
 656         }
 657 }
 658
 659 static void netvsc_send_completion(struct netvsc_device *net_device,
 660                                    struct vmbus_channel *incoming_channel,
 661                                    struct hv_device *device,
 662                                    const struct vmpacket_descriptor *desc,
 663                                    int budget)
 664 {
 665         struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 666         struct net_device *ndev = hv_get_drvdata(device);
 667
 668         switch (nvsp_packet->hdr.msg_type) {
 669         case NVSP_MSG_TYPE_INIT_COMPLETE:
 670         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 671         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 672         case NVSP_MSG5_TYPE_SUBCHANNEL:
 673                 /* Copy the response back */
 674                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 675                        sizeof(struct nvsp_message));
 676                 complete(&net_device->channel_init_wait);
 677                 break;
 678
 679         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 680                 netvsc_send_tx_complete(net_device, incoming_channel,
 681                                         device, desc, budget);
 682                 break;
 683
 684         default:
 685                 netdev_err(ndev,
 686                            "Unknown send completion type %d received!!\n",
 687                            nvsp_packet->hdr.msg_type);
 688         }
 689 }
 690
 691 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 692 {
 693         unsigned long *map_addr = net_device->send_section_map;
 694         unsigned int i;
 695
 696         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 697                 if (sync_test_and_set_bit(i, map_addr) == 0)
 698                         return i;
 699         }
 700
 701         return NETVSC_INVALID_INDEX;
 702 }
 703
 704 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 705                                     unsigned int section_index,
 706                                     u32 pend_size,
 707                                     struct hv_netvsc_packet *packet,
 708                                     struct rndis_message *rndis_msg,
 709                                     struct hv_page_buffer *pb,
 710                                     bool xmit_more)
 711 {
 712         char *start = net_device->send_buf;
 713         char *dest = start + (section_index * net_device->send_section_size)
 714                      + pend_size;
 715         int i;
 716         u32 padding = 0;
 717         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 718                 packet->page_buf_cnt;
 719         u32 remain;
 720
 721         /* Add padding */
 722         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
 723         if (xmit_more && remain) {
 724                 padding = net_device->pkt_align - remain;
 725                 rndis_msg->msg_len += padding;
 726                 packet->total_data_buflen += padding;
 727         }
 728
 729         for (i = 0; i < page_count; i++) {
 730                 char *src = phys_to_virt(pb[i].pfn << PAGE_SHIFT);
 731                 u32 offset = pb[i].offset;
 732                 u32 len = pb[i].len;
 733
 734                 memcpy(dest, (src + offset), len);
 735                 dest += len;
 736         }
 737
 738         if (padding)
 739                 memset(dest, 0, padding);
 740 }
 741
 742 static inline int netvsc_send_pkt(
 743         struct hv_device *device,
 744         struct hv_netvsc_packet *packet,
 745         struct netvsc_device *net_device,
 746         struct hv_page_buffer *pb,
 747         struct sk_buff *skb)
 748 {
 749         struct nvsp_message nvmsg;
 750         struct nvsp_1_message_send_rndis_packet * const rpkt =
 751                 &nvmsg.msg.v1_msg.send_rndis_pkt;
 752         struct netvsc_channel * const nvchan =
 753                 &net_device->chan_table[packet->q_idx];
 754         struct vmbus_channel *out_channel = nvchan->channel;
 755         struct net_device *ndev = hv_get_drvdata(device);
 756         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 757         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 758         u64 req_id;
 759         int ret;
 760         u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
 761
 762         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 763         if (skb)
 764                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
 765         else
 766                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
 767
 768         rpkt->send_buf_section_index = packet->send_buf_index;
 769         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 770                 rpkt->send_buf_section_size = 0;
 771         else
 772                 rpkt->send_buf_section_size = packet->total_data_buflen;
 773
 774         req_id = (ulong)skb;
 775
 776         if (out_channel->rescind)
 777                 return -ENODEV;
 778
 779         if (packet->page_buf_cnt) {
 780                 if (packet->cp_partial)
 781                         pb += packet->rmsg_pgcnt;
 782
 783                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 784                                                   pb, packet->page_buf_cnt,
 785                                                   &nvmsg, sizeof(nvmsg),
 786                                                   req_id);
 787         } else {
 788                 ret = vmbus_sendpacket(out_channel,
 789                                        &nvmsg, sizeof(nvmsg),
 790                                        req_id, VM_PKT_DATA_INBAND,
 791                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 792         }
 793
 794         if (ret == 0) {
 795                 atomic_inc_return(&nvchan->queue_sends);
 796
 797                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 798                         netif_tx_stop_queue(txq);
 799                         ndev_ctx->eth_stats.stop_queue++;
 800                 }
 801         } else if (ret == -EAGAIN) {
 802                 netif_tx_stop_queue(txq);
 803                 ndev_ctx->eth_stats.stop_queue++;
 804                 if (atomic_read(&nvchan->queue_sends) < 1) {
 805                         netif_tx_wake_queue(txq);
 806                         ndev_ctx->eth_stats.wake_queue++;
 807                         ret = -ENOSPC;
 808                 }
 809         } else {
 810                 netdev_err(ndev,
 811                            "Unable to send packet pages %u len %u, ret %d\n",
 812                            packet->page_buf_cnt, packet->total_data_buflen,
 813                            ret);
 814         }
 815
 816         return ret;
 817 }
 818
 819 /* Move packet out of multi send data (msd), and clear msd */
 820 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 821                                 struct sk_buff **msd_skb,
 822                                 struct multi_send_data *msdp)
 823 {
 824         *msd_skb = msdp->skb;
 825         *msd_send = msdp->pkt;
 826         msdp->skb = NULL;
 827         msdp->pkt = NULL;
 828         msdp->count = 0;
 829 }
 830
 831 /* RCU already held by caller */
 832 int netvsc_send(struct net_device *ndev,
 833                 struct hv_netvsc_packet *packet,
 834                 struct rndis_message *rndis_msg,
 835                 struct hv_page_buffer *pb,
 836                 struct sk_buff *skb)
 837 {
 838         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 839         struct netvsc_device *net_device
 840                 = rcu_dereference_bh(ndev_ctx->nvdev);
 841         struct hv_device *device = ndev_ctx->device_ctx;
 842         int ret = 0;
 843         struct netvsc_channel *nvchan;
 844         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 845         unsigned int section_index = NETVSC_INVALID_INDEX;
 846         struct multi_send_data *msdp;
 847         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 848         struct sk_buff *msd_skb = NULL;
 849         bool try_batch, xmit_more;
 850
 851         /* If device is rescinded, return error and packet will get dropped. */
 852         if (unlikely(!net_device || net_device->destroy))
 853                 return -ENODEV;
 854
 855         /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
 856          * here before the negotiation with the host is finished and
 857          * send_section_map may not be allocated yet.
 858          */
 859         if (unlikely(!net_device->send_section_map))
 860                 return -EAGAIN;
 861
 862         nvchan = &net_device->chan_table[packet->q_idx];
 863         packet->send_buf_index = NETVSC_INVALID_INDEX;
 864         packet->cp_partial = false;
 865
 866         /* Send control message directly without accessing msd (Multi-Send
 867          * Data) field which may be changed during data packet processing.
 868          */
 869         if (!skb) {
 870                 cur_send = packet;
 871                 goto send_now;
 872         }
 873
 874         /* batch packets in send buffer if possible */
 875         msdp = &nvchan->msd;
 876         if (msdp->pkt)
 877                 msd_len = msdp->pkt->total_data_buflen;
 878
 879         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
 880         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 881             net_device->send_section_size) {
 882                 section_index = msdp->pkt->send_buf_index;
 883
 884         } else if (try_batch && msd_len + packet->rmsg_size <
 885                    net_device->send_section_size) {
 886                 section_index = msdp->pkt->send_buf_index;
 887                 packet->cp_partial = true;
 888
 889         } else if (pktlen + net_device->pkt_align <
 890                    net_device->send_section_size) {
 891                 section_index = netvsc_get_next_send_section(net_device);
 892                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
 893                         ++ndev_ctx->eth_stats.tx_send_full;
 894                 } else {
 895                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 896                         msd_len = 0;
 897                 }
 898         }
 899
 900         /* Keep aggregating only if stack says more data is coming
 901          * and not doing mixed modes send and not flow blocked
 902          */
 903         xmit_more = skb->xmit_more &&
 904                 !packet->cp_partial &&
 905                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
 906
 907         if (section_index != NETVSC_INVALID_INDEX) {
 908                 netvsc_copy_to_send_buf(net_device,
 909                                         section_index, msd_len,
 910                                         packet, rndis_msg, pb, xmit_more);
 911
 912                 packet->send_buf_index = section_index;
 913
 914                 if (packet->cp_partial) {
 915                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 916                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 917                 } else {
 918                         packet->page_buf_cnt = 0;
 919                         packet->total_data_buflen += msd_len;
 920                 }
 921
 922                 if (msdp->pkt) {
 923                         packet->total_packets += msdp->pkt->total_packets;
 924                         packet->total_bytes += msdp->pkt->total_bytes;
 925                 }
 926
 927                 if (msdp->skb)
 928                         dev_consume_skb_any(msdp->skb);
 929
 930                 if (xmit_more) {
 931                         msdp->skb = skb;
 932                         msdp->pkt = packet;
 933                         msdp->count++;
 934                 } else {
 935                         cur_send = packet;
 936                         msdp->skb = NULL;
 937                         msdp->pkt = NULL;
 938                         msdp->count = 0;
 939                 }
 940         } else {
 941                 move_pkt_msd(&msd_send, &msd_skb, msdp);
 942                 cur_send = packet;
 943         }
 944
 945         if (msd_send) {
 946                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
 947                                             NULL, msd_skb);
 948
 949                 if (m_ret != 0) {
 950                         netvsc_free_send_slot(net_device,
 951                                               msd_send->send_buf_index);
 952                         dev_kfree_skb_any(msd_skb);
 953                 }
 954         }
 955
 956 send_now:
 957         if (cur_send)
 958                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
 959
 960         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
 961                 netvsc_free_send_slot(net_device, section_index);
 962
 963         return ret;
 964 }
 965
 966 /* Send pending recv completions */
 967 static int send_recv_completions(struct net_device *ndev,
 968                                  struct netvsc_device *nvdev,
 969                                  struct netvsc_channel *nvchan)
 970 {
 971         struct multi_recv_comp *mrc = &nvchan->mrc;
 972         struct recv_comp_msg {
 973                 struct nvsp_message_header hdr;
 974                 u32 status;
 975         }  __packed;
 976         struct recv_comp_msg msg = {
 977                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
 978         };
 979         int ret;
 980
 981         while (mrc->first != mrc->next) {
 982                 const struct recv_comp_data *rcd
 983                         = mrc->slots + mrc->first;
 984
 985                 msg.status = rcd->status;
 986                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
 987                                        rcd->tid, VM_PKT_COMP, 0);
 988                 if (unlikely(ret)) {
 989                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 990
 991                         ++ndev_ctx->eth_stats.rx_comp_busy;
 992                         return ret;
 993                 }
 994
 995                 if (++mrc->first == nvdev->recv_completion_cnt)
 996                         mrc->first = 0;
 997         }
 998
 999         /* receive completion ring has been emptied */
1000         if (unlikely(nvdev->destroy))
1001                 wake_up(&nvdev->wait_drain);
1002
1003         return 0;
1004 }
1005
1006 /* Count how many receive completions are outstanding */
1007 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1008                                  const struct multi_recv_comp *mrc,
1009                                  u32 *filled, u32 *avail)
1010 {
1011         u32 count = nvdev->recv_completion_cnt;
1012
1013         if (mrc->next >= mrc->first)
1014                 *filled = mrc->next - mrc->first;
1015         else
1016                 *filled = (count - mrc->first) + mrc->next;
1017
1018         *avail = count - *filled - 1;
1019 }
1020
1021 /* Add receive complete to ring to send to host. */
1022 static void enq_receive_complete(struct net_device *ndev,
1023                                  struct netvsc_device *nvdev, u16 q_idx,
1024                                  u64 tid, u32 status)
1025 {
1026         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1027         struct multi_recv_comp *mrc = &nvchan->mrc;
1028         struct recv_comp_data *rcd;
1029         u32 filled, avail;
1030
1031         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1032
1033         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1034                 send_recv_completions(ndev, nvdev, nvchan);
1035                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1036         }
1037
1038         if (unlikely(!avail)) {
1039                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1040                            q_idx, tid);
1041                 return;
1042         }
1043
1044         rcd = mrc->slots + mrc->next;
1045         rcd->tid = tid;
1046         rcd->status = status;
1047
1048         if (++mrc->next == nvdev->recv_completion_cnt)
1049                 mrc->next = 0;
1050 }
1051
1052 static int netvsc_receive(struct net_device *ndev,
1053                           struct netvsc_device *net_device,
1054                           struct net_device_context *net_device_ctx,
1055                           struct hv_device *device,
1056                           struct vmbus_channel *channel,
1057                           const struct vmpacket_descriptor *desc,
1058                           struct nvsp_message *nvsp)
1059 {
1060         const struct vmtransfer_page_packet_header *vmxferpage_packet
1061                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1062         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1063         char *recv_buf = net_device->recv_buf;
1064         u32 status = NVSP_STAT_SUCCESS;
1065         int i;
1066         int count = 0;
1067
1068         /* Make sure this is a valid nvsp packet */
1069         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1070                 netif_err(net_device_ctx, rx_err, ndev,
1071                           "Unknown nvsp packet type received %u\n",
1072                           nvsp->hdr.msg_type);
1073                 return 0;
1074         }
1075
1076         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1077                 netif_err(net_device_ctx, rx_err, ndev,
1078                           "Invalid xfer page set id - expecting %x got %x\n",
1079                           NETVSC_RECEIVE_BUFFER_ID,
1080                           vmxferpage_packet->xfer_pageset_id);
1081                 return 0;
1082         }
1083
1084         count = vmxferpage_packet->range_cnt;
1085
1086         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1087         for (i = 0; i < count; i++) {
1088                 void *data = recv_buf
1089                         + vmxferpage_packet->ranges[i].byte_offset;
1090                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1091
1092                 /* Pass it to the upper layer */
1093                 status = rndis_filter_receive(ndev, net_device,
1094                                               channel, data, buflen);
1095         }
1096
1097         enq_receive_complete(ndev, net_device, q_idx,
1098                              vmxferpage_packet->d.trans_id, status);
1099
1100         return count;
1101 }
1102
1103 static void netvsc_send_table(struct hv_device *hdev,
1104                               struct nvsp_message *nvmsg)
1105 {
1106         struct net_device *ndev = hv_get_drvdata(hdev);
1107         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1108         int i;
1109         u32 count, *tab;
1110
1111         count = nvmsg->msg.v5_msg.send_table.count;
1112         if (count != VRSS_SEND_TAB_SIZE) {
1113                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1114                 return;
1115         }
1116
1117         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1118                       nvmsg->msg.v5_msg.send_table.offset);
1119
1120         for (i = 0; i < count; i++)
1121                 net_device_ctx->tx_table[i] = tab[i];
1122 }
1123
1124 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1125                            struct nvsp_message *nvmsg)
1126 {
1127         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1128         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1129 }
1130
1131 static inline void netvsc_receive_inband(struct hv_device *hdev,
1132                                  struct net_device_context *net_device_ctx,
1133                                  struct nvsp_message *nvmsg)
1134 {
1135         switch (nvmsg->hdr.msg_type) {
1136         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1137                 netvsc_send_table(hdev, nvmsg);
1138                 break;
1139
1140         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1141                 netvsc_send_vf(net_device_ctx, nvmsg);
1142                 break;
1143         }
1144 }
1145
1146 static int netvsc_process_raw_pkt(struct hv_device *device,
1147                                   struct vmbus_channel *channel,
1148                                   struct netvsc_device *net_device,
1149                                   struct net_device *ndev,
1150                                   const struct vmpacket_descriptor *desc,
1151                                   int budget)
1152 {
1153         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1154         struct nvsp_message *nvmsg = hv_pkt_data(desc);
1155
1156         switch (desc->type) {
1157         case VM_PKT_COMP:
1158                 netvsc_send_completion(net_device, channel, device,
1159                                        desc, budget);
1160                 break;
1161
1162         case VM_PKT_DATA_USING_XFER_PAGES:
1163                 return netvsc_receive(ndev, net_device, net_device_ctx,
1164                                       device, channel, desc, nvmsg);
1165                 break;
1166
1167         case VM_PKT_DATA_INBAND:
1168                 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1169                 break;
1170
1171         default:
1172                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1173                            desc->type, desc->trans_id);
1174                 break;
1175         }
1176
1177         return 0;
1178 }
1179
1180 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1181 {
1182         struct vmbus_channel *primary = channel->primary_channel;
1183
1184         return primary ? primary->device_obj : channel->device_obj;
1185 }
1186
1187 /* Network processing softirq
1188  * Process data in incoming ring buffer from host
1189  * Stops when ring is empty or budget is met or exceeded.
1190  */
1191 int netvsc_poll(struct napi_struct *napi, int budget)
1192 {
1193         struct netvsc_channel *nvchan
1194                 = container_of(napi, struct netvsc_channel, napi);
1195         struct netvsc_device *net_device = nvchan->net_device;
1196         struct vmbus_channel *channel = nvchan->channel;
1197         struct hv_device *device = netvsc_channel_to_device(channel);
1198         struct net_device *ndev = hv_get_drvdata(device);
1199         int work_done = 0;
1200
1201         /* If starting a new interval */
1202         if (!nvchan->desc)
1203                 nvchan->desc = hv_pkt_iter_first(channel);
1204
1205         while (nvchan->desc && work_done < budget) {
1206                 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1207                                                     ndev, nvchan->desc, budget);
1208                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1209         }
1210
1211         /* If send of pending receive completions suceeded
1212          *   and did not exhaust NAPI budget this time
1213          *   and not doing busy poll
1214          * then re-enable host interrupts
1215          *     and reschedule if ring is not empty.
1216          */
1217         if (send_recv_completions(ndev, net_device, nvchan) == 0 &&
1218             work_done < budget &&
1219             napi_complete_done(napi, work_done) &&
1220             hv_end_read(&channel->inbound)) {
1221                 hv_begin_read(&channel->inbound);
1222                 napi_reschedule(napi);
1223         }
1224
1225         /* Driver may overshoot since multiple packets per descriptor */
1226         return min(work_done, budget);
1227 }
1228
1229 /* Call back when data is available in host ring buffer.
1230  * Processing is deferred until network softirq (NAPI)
1231  */
1232 void netvsc_channel_cb(void *context)
1233 {
1234         struct netvsc_channel *nvchan = context;
1235         struct vmbus_channel *channel = nvchan->channel;
1236         struct hv_ring_buffer_info *rbi = &channel->inbound;
1237
1238         /* preload first vmpacket descriptor */
1239         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1240
1241         if (napi_schedule_prep(&nvchan->napi)) {
1242                 /* disable interupts from host */
1243                 hv_begin_read(rbi);
1244
1245                 __napi_schedule(&nvchan->napi);
1246         }
1247 }
1248
1249 /*
1250  * netvsc_device_add - Callback when the device belonging to this
1251  * driver is added
1252  */
1253 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1254                                 const struct netvsc_device_info *device_info)
1255 {
1256         int i, ret = 0;
1257         struct netvsc_device *net_device;
1258         struct net_device *ndev = hv_get_drvdata(device);
1259         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1260
1261         net_device = alloc_net_device();
1262         if (!net_device)
1263                 return ERR_PTR(-ENOMEM);
1264
1265         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1266                 net_device_ctx->tx_table[i] = 0;
1267
1268         /* Because the device uses NAPI, all the interrupt batching and
1269          * control is done via Net softirq, not the channel handling
1270          */
1271         set_channel_read_mode(device->channel, HV_CALL_ISR);
1272
1273         /* If we're reopening the device we may have multiple queues, fill the
1274          * chn_table with the default channel to use it before subchannels are
1275          * opened.
1276          * Initialize the channel state before we open;
1277          * we can be interrupted as soon as we open the channel.
1278          */
1279
1280         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1281                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1282
1283                 nvchan->channel = device->channel;
1284                 nvchan->net_device = net_device;
1285                 u64_stats_init(&nvchan->tx_stats.syncp);
1286                 u64_stats_init(&nvchan->rx_stats.syncp);
1287         }
1288
1289         /* Enable NAPI handler before init callbacks */
1290         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1291                        netvsc_poll, NAPI_POLL_WEIGHT);
1292
1293         /* Open the channel */
1294         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1295                          netvsc_ring_bytes,  NULL, 0,
1296                          netvsc_channel_cb, net_device->chan_table);
1297
1298         if (ret != 0) {
1299                 netif_napi_del(&net_device->chan_table[0].napi);
1300                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1301                 goto cleanup;
1302         }
1303
1304         /* Channel is opened */
1305         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1306
1307         napi_enable(&net_device->chan_table[0].napi);
1308
1309         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1310          * populated.
1311          */
1312         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1313
1314         /* Connect with the NetVsp */
1315         ret = netvsc_connect_vsp(device, net_device, device_info);
1316         if (ret != 0) {
1317                 netdev_err(ndev,
1318                         "unable to connect to NetVSP - %d\n", ret);
1319                 goto close;
1320         }
1321
1322         return net_device;
1323
1324 close:
1325         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1326         napi_disable(&net_device->chan_table[0].napi);
1327
1328         /* Now, we can close the channel safely */
1329         vmbus_close(device->channel);
1330
1331 cleanup:
1332         free_netvsc_device(&net_device->rcu);
1333
1334         return ERR_PTR(ret);
1335 }