drivers/net/hyperv/netvsc.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2009, Microsoft Corporation.
   4  *
   5  * Authors:
   6  *   Haiyang Zhang <haiyangz@microsoft.com>
   7  *   Hank Janssen  <hjanssen@microsoft.com>
   8  */
   9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/kernel.h>
  12 #include <linux/sched.h>
  13 #include <linux/wait.h>
  14 #include <linux/mm.h>
  15 #include <linux/delay.h>
  16 #include <linux/io.h>
  17 #include <linux/slab.h>
  18 #include <linux/netdevice.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/vmalloc.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/prefetch.h>
  23
  24 #include <asm/sync_bitops.h>
  25
  26 #include "hyperv_net.h"
  27 #include "netvsc_trace.h"
  28
  29 /*
  30  * Switch the data path from the synthetic interface to the VF
  31  * interface.
  32  */
  33 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  34 {
  35         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  36         struct hv_device *dev = net_device_ctx->device_ctx;
  37         struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
  38         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  39
  40         /* Block sending traffic to VF if it's about to be gone */
  41         if (!vf)
  42                 net_device_ctx->data_path_is_vf = vf;
  43
  44         memset(init_pkt, 0, sizeof(struct nvsp_message));
  45         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  46         if (vf)
  47                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  48                         NVSP_DATAPATH_VF;
  49         else
  50                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  51                         NVSP_DATAPATH_SYNTHETIC;
  52
  53         trace_nvsp_send(ndev, init_pkt);
  54
  55         vmbus_sendpacket(dev->channel, init_pkt,
  56                                sizeof(struct nvsp_message),
  57                                (unsigned long)init_pkt,
  58                                VM_PKT_DATA_INBAND,
  59                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
  60         wait_for_completion(&nv_dev->channel_init_wait);
  61         net_device_ctx->data_path_is_vf = vf;
  62 }
  63
  64 /* Worker to setup sub channels on initial setup
  65  * Initial hotplug event occurs in softirq context
  66  * and can't wait for channels.
  67  */
  68 static void netvsc_subchan_work(struct work_struct *w)
  69 {
  70         struct netvsc_device *nvdev =
  71                 container_of(w, struct netvsc_device, subchan_work);
  72         struct rndis_device *rdev;
  73         int i, ret;
  74
  75         /* Avoid deadlock with device removal already under RTNL */
  76         if (!rtnl_trylock()) {
  77                 schedule_work(w);
  78                 return;
  79         }
  80
  81         rdev = nvdev->extension;
  82         if (rdev) {
  83                 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
  84                 if (ret == 0) {
  85                         netif_device_attach(rdev->ndev);
  86                 } else {
  87                         /* fallback to only primary channel */
  88                         for (i = 1; i < nvdev->num_chn; i++)
  89                                 netif_napi_del(&nvdev->chan_table[i].napi);
  90
  91                         nvdev->max_chn = 1;
  92                         nvdev->num_chn = 1;
  93                 }
  94         }
  95
  96         rtnl_unlock();
  97 }
  98
  99 static struct netvsc_device *alloc_net_device(void)
 100 {
 101         struct netvsc_device *net_device;
 102
 103         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
 104         if (!net_device)
 105                 return NULL;
 106
 107         init_waitqueue_head(&net_device->wait_drain);
 108         net_device->destroy = false;
 109         net_device->tx_disable = true;
 110
 111         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
 112         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
 113
 114         init_completion(&net_device->channel_init_wait);
 115         init_waitqueue_head(&net_device->subchan_open);
 116         INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
 117
 118         return net_device;
 119 }
 120
 121 static void free_netvsc_device(struct rcu_head *head)
 122 {
 123         struct netvsc_device *nvdev
 124                 = container_of(head, struct netvsc_device, rcu);
 125         int i;
 126
 127         kfree(nvdev->extension);
 128         vfree(nvdev->recv_buf);
 129         vfree(nvdev->send_buf);
 130         kfree(nvdev->send_section_map);
 131
 132         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
 133                 xdp_rxq_info_unreg(&nvdev->chan_table[i].xdp_rxq);
 134                 vfree(nvdev->chan_table[i].mrc.slots);
 135         }
 136
 137         kfree(nvdev);
 138 }
 139
 140 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
 141 {
 142         call_rcu(&nvdev->rcu, free_netvsc_device);
 143 }
 144
 145 static void netvsc_revoke_recv_buf(struct hv_device *device,
 146                                    struct netvsc_device *net_device,
 147                                    struct net_device *ndev)
 148 {
 149         struct nvsp_message *revoke_packet;
 150         int ret;
 151
 152         /*
 153          * If we got a section count, it means we received a
 154          * SendReceiveBufferComplete msg (ie sent
 155          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 156          * to send a revoke msg here
 157          */
 158         if (net_device->recv_section_cnt) {
 159                 /* Send the revoke receive buffer */
 160                 revoke_packet = &net_device->revoke_packet;
 161                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 162
 163                 revoke_packet->hdr.msg_type =
 164                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 165                 revoke_packet->msg.v1_msg.
 166                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 167
 168                 trace_nvsp_send(ndev, revoke_packet);
 169
 170                 ret = vmbus_sendpacket(device->channel,
 171                                        revoke_packet,
 172                                        sizeof(struct nvsp_message),
 173                                        VMBUS_RQST_ID_NO_RESPONSE,
 174                                        VM_PKT_DATA_INBAND, 0);
 175                 /* If the failure is because the channel is rescinded;
 176                  * ignore the failure since we cannot send on a rescinded
 177                  * channel. This would allow us to properly cleanup
 178                  * even when the channel is rescinded.
 179                  */
 180                 if (device->channel->rescind)
 181                         ret = 0;
 182                 /*
 183                  * If we failed here, we might as well return and
 184                  * have a leak rather than continue and a bugchk
 185                  */
 186                 if (ret != 0) {
 187                         netdev_err(ndev, "unable to send "
 188                                 "revoke receive buffer to netvsp\n");
 189                         return;
 190                 }
 191                 net_device->recv_section_cnt = 0;
 192         }
 193 }
 194
 195 static void netvsc_revoke_send_buf(struct hv_device *device,
 196                                    struct netvsc_device *net_device,
 197                                    struct net_device *ndev)
 198 {
 199         struct nvsp_message *revoke_packet;
 200         int ret;
 201
 202         /* Deal with the send buffer we may have setup.
 203          * If we got a  send section size, it means we received a
 204          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 205          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 206          * to send a revoke msg here
 207          */
 208         if (net_device->send_section_cnt) {
 209                 /* Send the revoke receive buffer */
 210                 revoke_packet = &net_device->revoke_packet;
 211                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 212
 213                 revoke_packet->hdr.msg_type =
 214                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 215                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 216                         NETVSC_SEND_BUFFER_ID;
 217
 218                 trace_nvsp_send(ndev, revoke_packet);
 219
 220                 ret = vmbus_sendpacket(device->channel,
 221                                        revoke_packet,
 222                                        sizeof(struct nvsp_message),
 223                                        VMBUS_RQST_ID_NO_RESPONSE,
 224                                        VM_PKT_DATA_INBAND, 0);
 225
 226                 /* If the failure is because the channel is rescinded;
 227                  * ignore the failure since we cannot send on a rescinded
 228                  * channel. This would allow us to properly cleanup
 229                  * even when the channel is rescinded.
 230                  */
 231                 if (device->channel->rescind)
 232                         ret = 0;
 233
 234                 /* If we failed here, we might as well return and
 235                  * have a leak rather than continue and a bugchk
 236                  */
 237                 if (ret != 0) {
 238                         netdev_err(ndev, "unable to send "
 239                                    "revoke send buffer to netvsp\n");
 240                         return;
 241                 }
 242                 net_device->send_section_cnt = 0;
 243         }
 244 }
 245
 246 static void netvsc_teardown_recv_gpadl(struct hv_device *device,
 247                                        struct netvsc_device *net_device,
 248                                        struct net_device *ndev)
 249 {
 250         int ret;
 251
 252         if (net_device->recv_buf_gpadl_handle) {
 253                 ret = vmbus_teardown_gpadl(device->channel,
 254                                            net_device->recv_buf_gpadl_handle);
 255
 256                 /* If we failed here, we might as well return and have a leak
 257                  * rather than continue and a bugchk
 258                  */
 259                 if (ret != 0) {
 260                         netdev_err(ndev,
 261                                    "unable to teardown receive buffer's gpadl\n");
 262                         return;
 263                 }
 264                 net_device->recv_buf_gpadl_handle = 0;
 265         }
 266 }
 267
 268 static void netvsc_teardown_send_gpadl(struct hv_device *device,
 269                                        struct netvsc_device *net_device,
 270                                        struct net_device *ndev)
 271 {
 272         int ret;
 273
 274         if (net_device->send_buf_gpadl_handle) {
 275                 ret = vmbus_teardown_gpadl(device->channel,
 276                                            net_device->send_buf_gpadl_handle);
 277
 278                 /* If we failed here, we might as well return and have a leak
 279                  * rather than continue and a bugchk
 280                  */
 281                 if (ret != 0) {
 282                         netdev_err(ndev,
 283                                    "unable to teardown send buffer's gpadl\n");
 284                         return;
 285                 }
 286                 net_device->send_buf_gpadl_handle = 0;
 287         }
 288 }
 289
 290 int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
 291 {
 292         struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
 293         int node = cpu_to_node(nvchan->channel->target_cpu);
 294         size_t size;
 295
 296         size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
 297         nvchan->mrc.slots = vzalloc_node(size, node);
 298         if (!nvchan->mrc.slots)
 299                 nvchan->mrc.slots = vzalloc(size);
 300
 301         return nvchan->mrc.slots ? 0 : -ENOMEM;
 302 }
 303
 304 static int netvsc_init_buf(struct hv_device *device,
 305                            struct netvsc_device *net_device,
 306                            const struct netvsc_device_info *device_info)
 307 {
 308         struct nvsp_1_message_send_receive_buffer_complete *resp;
 309         struct net_device *ndev = hv_get_drvdata(device);
 310         struct nvsp_message *init_packet;
 311         unsigned int buf_size;
 312         size_t map_words;
 313         int ret = 0;
 314
 315         /* Get receive buffer area. */
 316         buf_size = device_info->recv_sections * device_info->recv_section_size;
 317         buf_size = roundup(buf_size, PAGE_SIZE);
 318
 319         /* Legacy hosts only allow smaller receive buffer */
 320         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 321                 buf_size = min_t(unsigned int, buf_size,
 322                                  NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
 323
 324         net_device->recv_buf = vzalloc(buf_size);
 325         if (!net_device->recv_buf) {
 326                 netdev_err(ndev,
 327                            "unable to allocate receive buffer of size %u\n",
 328                            buf_size);
 329                 ret = -ENOMEM;
 330                 goto cleanup;
 331         }
 332
 333         net_device->recv_buf_size = buf_size;
 334
 335         /*
 336          * Establish the gpadl handle for this buffer on this
 337          * channel.  Note: This call uses the vmbus connection rather
 338          * than the channel to establish the gpadl handle.
 339          */
 340         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 341                                     buf_size,
 342                                     &net_device->recv_buf_gpadl_handle);
 343         if (ret != 0) {
 344                 netdev_err(ndev,
 345                         "unable to establish receive buffer's gpadl\n");
 346                 goto cleanup;
 347         }
 348
 349         /* Notify the NetVsp of the gpadl handle */
 350         init_packet = &net_device->channel_init_pkt;
 351         memset(init_packet, 0, sizeof(struct nvsp_message));
 352         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 353         init_packet->msg.v1_msg.send_recv_buf.
 354                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 355         init_packet->msg.v1_msg.
 356                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 357
 358         trace_nvsp_send(ndev, init_packet);
 359
 360         /* Send the gpadl notification request */
 361         ret = vmbus_sendpacket(device->channel, init_packet,
 362                                sizeof(struct nvsp_message),
 363                                (unsigned long)init_packet,
 364                                VM_PKT_DATA_INBAND,
 365                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 366         if (ret != 0) {
 367                 netdev_err(ndev,
 368                         "unable to send receive buffer's gpadl to netvsp\n");
 369                 goto cleanup;
 370         }
 371
 372         wait_for_completion(&net_device->channel_init_wait);
 373
 374         /* Check the response */
 375         resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
 376         if (resp->status != NVSP_STAT_SUCCESS) {
 377                 netdev_err(ndev,
 378                            "Unable to complete receive buffer initialization with NetVsp - status %d\n",
 379                            resp->status);
 380                 ret = -EINVAL;
 381                 goto cleanup;
 382         }
 383
 384         /* Parse the response */
 385         netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
 386                    resp->num_sections, resp->sections[0].sub_alloc_size,
 387                    resp->sections[0].num_sub_allocs);
 388
 389         /* There should only be one section for the entire receive buffer */
 390         if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
 391                 ret = -EINVAL;
 392                 goto cleanup;
 393         }
 394
 395         net_device->recv_section_size = resp->sections[0].sub_alloc_size;
 396         net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
 397
 398         /* Ensure buffer will not overflow */
 399         if (net_device->recv_section_size < NETVSC_MTU_MIN || (u64)net_device->recv_section_size *
 400             (u64)net_device->recv_section_cnt > (u64)buf_size) {
 401                 netdev_err(ndev, "invalid recv_section_size %u\n",
 402                            net_device->recv_section_size);
 403                 ret = -EINVAL;
 404                 goto cleanup;
 405         }
 406
 407         /* Setup receive completion ring.
 408          * Add 1 to the recv_section_cnt because at least one entry in a
 409          * ring buffer has to be empty.
 410          */
 411         net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
 412         ret = netvsc_alloc_recv_comp_ring(net_device, 0);
 413         if (ret)
 414                 goto cleanup;
 415
 416         /* Now setup the send buffer. */
 417         buf_size = device_info->send_sections * device_info->send_section_size;
 418         buf_size = round_up(buf_size, PAGE_SIZE);
 419
 420         net_device->send_buf = vzalloc(buf_size);
 421         if (!net_device->send_buf) {
 422                 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
 423                            buf_size);
 424                 ret = -ENOMEM;
 425                 goto cleanup;
 426         }
 427
 428         /* Establish the gpadl handle for this buffer on this
 429          * channel.  Note: This call uses the vmbus connection rather
 430          * than the channel to establish the gpadl handle.
 431          */
 432         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 433                                     buf_size,
 434                                     &net_device->send_buf_gpadl_handle);
 435         if (ret != 0) {
 436                 netdev_err(ndev,
 437                            "unable to establish send buffer's gpadl\n");
 438                 goto cleanup;
 439         }
 440
 441         /* Notify the NetVsp of the gpadl handle */
 442         init_packet = &net_device->channel_init_pkt;
 443         memset(init_packet, 0, sizeof(struct nvsp_message));
 444         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 445         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 446                 net_device->send_buf_gpadl_handle;
 447         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 448
 449         trace_nvsp_send(ndev, init_packet);
 450
 451         /* Send the gpadl notification request */
 452         ret = vmbus_sendpacket(device->channel, init_packet,
 453                                sizeof(struct nvsp_message),
 454                                (unsigned long)init_packet,
 455                                VM_PKT_DATA_INBAND,
 456                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 457         if (ret != 0) {
 458                 netdev_err(ndev,
 459                            "unable to send send buffer's gpadl to netvsp\n");
 460                 goto cleanup;
 461         }
 462
 463         wait_for_completion(&net_device->channel_init_wait);
 464
 465         /* Check the response */
 466         if (init_packet->msg.v1_msg.
 467             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 468                 netdev_err(ndev, "Unable to complete send buffer "
 469                            "initialization with NetVsp - status %d\n",
 470                            init_packet->msg.v1_msg.
 471                            send_send_buf_complete.status);
 472                 ret = -EINVAL;
 473                 goto cleanup;
 474         }
 475
 476         /* Parse the response */
 477         net_device->send_section_size = init_packet->msg.
 478                                 v1_msg.send_send_buf_complete.section_size;
 479         if (net_device->send_section_size < NETVSC_MTU_MIN) {
 480                 netdev_err(ndev, "invalid send_section_size %u\n",
 481                            net_device->send_section_size);
 482                 ret = -EINVAL;
 483                 goto cleanup;
 484         }
 485
 486         /* Section count is simply the size divided by the section size. */
 487         net_device->send_section_cnt = buf_size / net_device->send_section_size;
 488
 489         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 490                    net_device->send_section_size, net_device->send_section_cnt);
 491
 492         /* Setup state for managing the send buffer. */
 493         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
 494
 495         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
 496         if (net_device->send_section_map == NULL) {
 497                 ret = -ENOMEM;
 498                 goto cleanup;
 499         }
 500
 501         goto exit;
 502
 503 cleanup:
 504         netvsc_revoke_recv_buf(device, net_device, ndev);
 505         netvsc_revoke_send_buf(device, net_device, ndev);
 506         netvsc_teardown_recv_gpadl(device, net_device, ndev);
 507         netvsc_teardown_send_gpadl(device, net_device, ndev);
 508
 509 exit:
 510         return ret;
 511 }
 512
 513 /* Negotiate NVSP protocol version */
 514 static int negotiate_nvsp_ver(struct hv_device *device,
 515                               struct netvsc_device *net_device,
 516                               struct nvsp_message *init_packet,
 517                               u32 nvsp_ver)
 518 {
 519         struct net_device *ndev = hv_get_drvdata(device);
 520         int ret;
 521
 522         memset(init_packet, 0, sizeof(struct nvsp_message));
 523         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 524         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 525         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 526         trace_nvsp_send(ndev, init_packet);
 527
 528         /* Send the init request */
 529         ret = vmbus_sendpacket(device->channel, init_packet,
 530                                sizeof(struct nvsp_message),
 531                                (unsigned long)init_packet,
 532                                VM_PKT_DATA_INBAND,
 533                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 534
 535         if (ret != 0)
 536                 return ret;
 537
 538         wait_for_completion(&net_device->channel_init_wait);
 539
 540         if (init_packet->msg.init_msg.init_complete.status !=
 541             NVSP_STAT_SUCCESS)
 542                 return -EINVAL;
 543
 544         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 545                 return 0;
 546
 547         /* NVSPv2 or later: Send NDIS config */
 548         memset(init_packet, 0, sizeof(struct nvsp_message));
 549         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 550         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 551         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 552
 553         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 554                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 555
 556                 /* Teaming bit is needed to receive link speed updates */
 557                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 558         }
 559
 560         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
 561                 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
 562
 563         trace_nvsp_send(ndev, init_packet);
 564
 565         ret = vmbus_sendpacket(device->channel, init_packet,
 566                                 sizeof(struct nvsp_message),
 567                                 VMBUS_RQST_ID_NO_RESPONSE,
 568                                 VM_PKT_DATA_INBAND, 0);
 569
 570         return ret;
 571 }
 572
 573 static int netvsc_connect_vsp(struct hv_device *device,
 574                               struct netvsc_device *net_device,
 575                               const struct netvsc_device_info *device_info)
 576 {
 577         struct net_device *ndev = hv_get_drvdata(device);
 578         static const u32 ver_list[] = {
 579                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 580                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
 581                 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
 582         };
 583         struct nvsp_message *init_packet;
 584         int ndis_version, i, ret;
 585
 586         init_packet = &net_device->channel_init_pkt;
 587
 588         /* Negotiate the latest NVSP protocol supported */
 589         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 590                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 591                                        ver_list[i])  == 0) {
 592                         net_device->nvsp_version = ver_list[i];
 593                         break;
 594                 }
 595
 596         if (i < 0) {
 597                 ret = -EPROTO;
 598                 goto cleanup;
 599         }
 600
 601         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 602
 603         /* Send the ndis version */
 604         memset(init_packet, 0, sizeof(struct nvsp_message));
 605
 606         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 607                 ndis_version = 0x00060001;
 608         else
 609                 ndis_version = 0x0006001e;
 610
 611         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 612         init_packet->msg.v1_msg.
 613                 send_ndis_ver.ndis_major_ver =
 614                                 (ndis_version & 0xFFFF0000) >> 16;
 615         init_packet->msg.v1_msg.
 616                 send_ndis_ver.ndis_minor_ver =
 617                                 ndis_version & 0xFFFF;
 618
 619         trace_nvsp_send(ndev, init_packet);
 620
 621         /* Send the init request */
 622         ret = vmbus_sendpacket(device->channel, init_packet,
 623                                 sizeof(struct nvsp_message),
 624                                 VMBUS_RQST_ID_NO_RESPONSE,
 625                                 VM_PKT_DATA_INBAND, 0);
 626         if (ret != 0)
 627                 goto cleanup;
 628
 629
 630         ret = netvsc_init_buf(device, net_device, device_info);
 631
 632 cleanup:
 633         return ret;
 634 }
 635
 636 /*
 637  * netvsc_device_remove - Callback when the root bus device is removed
 638  */
 639 void netvsc_device_remove(struct hv_device *device)
 640 {
 641         struct net_device *ndev = hv_get_drvdata(device);
 642         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 643         struct netvsc_device *net_device
 644                 = rtnl_dereference(net_device_ctx->nvdev);
 645         int i;
 646
 647         /*
 648          * Revoke receive buffer. If host is pre-Win2016 then tear down
 649          * receive buffer GPADL. Do the same for send buffer.
 650          */
 651         netvsc_revoke_recv_buf(device, net_device, ndev);
 652         if (vmbus_proto_version < VERSION_WIN10)
 653                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 654
 655         netvsc_revoke_send_buf(device, net_device, ndev);
 656         if (vmbus_proto_version < VERSION_WIN10)
 657                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 658
 659         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 660
 661         /* Disable NAPI and disassociate its context from the device. */
 662         for (i = 0; i < net_device->num_chn; i++) {
 663                 /* See also vmbus_reset_channel_cb(). */
 664                 napi_disable(&net_device->chan_table[i].napi);
 665                 netif_napi_del(&net_device->chan_table[i].napi);
 666         }
 667
 668         /*
 669          * At this point, no one should be accessing net_device
 670          * except in here
 671          */
 672         netdev_dbg(ndev, "net device safe to remove\n");
 673
 674         /* Now, we can close the channel safely */
 675         vmbus_close(device->channel);
 676
 677         /*
 678          * If host is Win2016 or higher then we do the GPADL tear down
 679          * here after VMBus is closed.
 680         */
 681         if (vmbus_proto_version >= VERSION_WIN10) {
 682                 netvsc_teardown_recv_gpadl(device, net_device, ndev);
 683                 netvsc_teardown_send_gpadl(device, net_device, ndev);
 684         }
 685
 686         /* Release all resources */
 687         free_netvsc_device_rcu(net_device);
 688 }
 689
 690 #define RING_AVAIL_PERCENT_HIWATER 20
 691 #define RING_AVAIL_PERCENT_LOWATER 10
 692
 693 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 694                                          u32 index)
 695 {
 696         sync_change_bit(index, net_device->send_section_map);
 697 }
 698
 699 static void netvsc_send_tx_complete(struct net_device *ndev,
 700                                     struct netvsc_device *net_device,
 701                                     struct vmbus_channel *channel,
 702                                     const struct vmpacket_descriptor *desc,
 703                                     int budget)
 704 {
 705         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 706         struct sk_buff *skb;
 707         u16 q_idx = 0;
 708         int queue_sends;
 709         u64 cmd_rqst;
 710
 711         cmd_rqst = vmbus_request_addr(&channel->requestor, (u64)desc->trans_id);
 712         if (cmd_rqst == VMBUS_RQST_ERROR) {
 713                 netdev_err(ndev, "Incorrect transaction id\n");
 714                 return;
 715         }
 716
 717         skb = (struct sk_buff *)(unsigned long)cmd_rqst;
 718
 719         /* Notify the layer above us */
 720         if (likely(skb)) {
 721                 const struct hv_netvsc_packet *packet
 722                         = (struct hv_netvsc_packet *)skb->cb;
 723                 u32 send_index = packet->send_buf_index;
 724                 struct netvsc_stats *tx_stats;
 725
 726                 if (send_index != NETVSC_INVALID_INDEX)
 727                         netvsc_free_send_slot(net_device, send_index);
 728                 q_idx = packet->q_idx;
 729
 730                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 731
 732                 u64_stats_update_begin(&tx_stats->syncp);
 733                 tx_stats->packets += packet->total_packets;
 734                 tx_stats->bytes += packet->total_bytes;
 735                 u64_stats_update_end(&tx_stats->syncp);
 736
 737                 napi_consume_skb(skb, budget);
 738         }
 739
 740         queue_sends =
 741                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 742
 743         if (unlikely(net_device->destroy)) {
 744                 if (queue_sends == 0)
 745                         wake_up(&net_device->wait_drain);
 746         } else {
 747                 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
 748
 749                 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
 750                     (hv_get_avail_to_write_percent(&channel->outbound) >
 751                      RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
 752                         netif_tx_wake_queue(txq);
 753                         ndev_ctx->eth_stats.wake_queue++;
 754                 }
 755         }
 756 }
 757
 758 static void netvsc_send_completion(struct net_device *ndev,
 759                                    struct netvsc_device *net_device,
 760                                    struct vmbus_channel *incoming_channel,
 761                                    const struct vmpacket_descriptor *desc,
 762                                    int budget)
 763 {
 764         const struct nvsp_message *nvsp_packet;
 765         u32 msglen = hv_pkt_datalen(desc);
 766         struct nvsp_message *pkt_rqst;
 767         u64 cmd_rqst;
 768
 769         /* First check if this is a VMBUS completion without data payload */
 770         if (!msglen) {
 771                 cmd_rqst = vmbus_request_addr(&incoming_channel->requestor,
 772                                               (u64)desc->trans_id);
 773                 if (cmd_rqst == VMBUS_RQST_ERROR) {
 774                         netdev_err(ndev, "Invalid transaction id\n");
 775                         return;
 776                 }
 777
 778                 pkt_rqst = (struct nvsp_message *)(uintptr_t)cmd_rqst;
 779                 switch (pkt_rqst->hdr.msg_type) {
 780                 case NVSP_MSG4_TYPE_SWITCH_DATA_PATH:
 781                         complete(&net_device->channel_init_wait);
 782                         break;
 783
 784                 default:
 785                         netdev_err(ndev, "Unexpected VMBUS completion!!\n");
 786                 }
 787                 return;
 788         }
 789
 790         /* Ensure packet is big enough to read header fields */
 791         if (msglen < sizeof(struct nvsp_message_header)) {
 792                 netdev_err(ndev, "nvsp_message length too small: %u\n", msglen);
 793                 return;
 794         }
 795
 796         nvsp_packet = hv_pkt_data(desc);
 797         switch (nvsp_packet->hdr.msg_type) {
 798         case NVSP_MSG_TYPE_INIT_COMPLETE:
 799                 if (msglen < sizeof(struct nvsp_message_header) +
 800                                 sizeof(struct nvsp_message_init_complete)) {
 801                         netdev_err(ndev, "nvsp_msg length too small: %u\n",
 802                                    msglen);
 803                         return;
 804                 }
 805                 fallthrough;
 806
 807         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 808                 if (msglen < sizeof(struct nvsp_message_header) +
 809                                 sizeof(struct nvsp_1_message_send_receive_buffer_complete)) {
 810                         netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
 811                                    msglen);
 812                         return;
 813                 }
 814                 fallthrough;
 815
 816         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 817                 if (msglen < sizeof(struct nvsp_message_header) +
 818                                 sizeof(struct nvsp_1_message_send_send_buffer_complete)) {
 819                         netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
 820                                    msglen);
 821                         return;
 822                 }
 823                 fallthrough;
 824
 825         case NVSP_MSG5_TYPE_SUBCHANNEL:
 826                 if (msglen < sizeof(struct nvsp_message_header) +
 827                                 sizeof(struct nvsp_5_subchannel_complete)) {
 828                         netdev_err(ndev, "nvsp_msg5 length too small: %u\n",
 829                                    msglen);
 830                         return;
 831                 }
 832                 /* Copy the response back */
 833                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 834                        sizeof(struct nvsp_message));
 835                 complete(&net_device->channel_init_wait);
 836                 break;
 837
 838         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 839                 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
 840                                         desc, budget);
 841                 break;
 842
 843         default:
 844                 netdev_err(ndev,
 845                            "Unknown send completion type %d received!!\n",
 846                            nvsp_packet->hdr.msg_type);
 847         }
 848 }
 849
 850 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 851 {
 852         unsigned long *map_addr = net_device->send_section_map;
 853         unsigned int i;
 854
 855         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 856                 if (sync_test_and_set_bit(i, map_addr) == 0)
 857                         return i;
 858         }
 859
 860         return NETVSC_INVALID_INDEX;
 861 }
 862
 863 static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 864                                     unsigned int section_index,
 865                                     u32 pend_size,
 866                                     struct hv_netvsc_packet *packet,
 867                                     struct rndis_message *rndis_msg,
 868                                     struct hv_page_buffer *pb,
 869                                     bool xmit_more)
 870 {
 871         char *start = net_device->send_buf;
 872         char *dest = start + (section_index * net_device->send_section_size)
 873                      + pend_size;
 874         int i;
 875         u32 padding = 0;
 876         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 877                 packet->page_buf_cnt;
 878         u32 remain;
 879
 880         /* Add padding */
 881         remain = packet->total_data_buflen & (net_device->pkt_align - 1);
 882         if (xmit_more && remain) {
 883                 padding = net_device->pkt_align - remain;
 884                 rndis_msg->msg_len += padding;
 885                 packet->total_data_buflen += padding;
 886         }
 887
 888         for (i = 0; i < page_count; i++) {
 889                 char *src = phys_to_virt(pb[i].pfn << HV_HYP_PAGE_SHIFT);
 890                 u32 offset = pb[i].offset;
 891                 u32 len = pb[i].len;
 892
 893                 memcpy(dest, (src + offset), len);
 894                 dest += len;
 895         }
 896
 897         if (padding)
 898                 memset(dest, 0, padding);
 899 }
 900
 901 static inline int netvsc_send_pkt(
 902         struct hv_device *device,
 903         struct hv_netvsc_packet *packet,
 904         struct netvsc_device *net_device,
 905         struct hv_page_buffer *pb,
 906         struct sk_buff *skb)
 907 {
 908         struct nvsp_message nvmsg;
 909         struct nvsp_1_message_send_rndis_packet *rpkt =
 910                 &nvmsg.msg.v1_msg.send_rndis_pkt;
 911         struct netvsc_channel * const nvchan =
 912                 &net_device->chan_table[packet->q_idx];
 913         struct vmbus_channel *out_channel = nvchan->channel;
 914         struct net_device *ndev = hv_get_drvdata(device);
 915         struct net_device_context *ndev_ctx = netdev_priv(ndev);
 916         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 917         u64 req_id;
 918         int ret;
 919         u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
 920
 921         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 922         if (skb)
 923                 rpkt->channel_type = 0;         /* 0 is RMC_DATA */
 924         else
 925                 rpkt->channel_type = 1;         /* 1 is RMC_CONTROL */
 926
 927         rpkt->send_buf_section_index = packet->send_buf_index;
 928         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 929                 rpkt->send_buf_section_size = 0;
 930         else
 931                 rpkt->send_buf_section_size = packet->total_data_buflen;
 932
 933         req_id = (ulong)skb;
 934
 935         if (out_channel->rescind)
 936                 return -ENODEV;
 937
 938         trace_nvsp_send_pkt(ndev, out_channel, rpkt);
 939
 940         if (packet->page_buf_cnt) {
 941                 if (packet->cp_partial)
 942                         pb += packet->rmsg_pgcnt;
 943
 944                 ret = vmbus_sendpacket_pagebuffer(out_channel,
 945                                                   pb, packet->page_buf_cnt,
 946                                                   &nvmsg, sizeof(nvmsg),
 947                                                   req_id);
 948         } else {
 949                 ret = vmbus_sendpacket(out_channel,
 950                                        &nvmsg, sizeof(nvmsg),
 951                                        req_id, VM_PKT_DATA_INBAND,
 952                                        VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 953         }
 954
 955         if (ret == 0) {
 956                 atomic_inc_return(&nvchan->queue_sends);
 957
 958                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
 959                         netif_tx_stop_queue(txq);
 960                         ndev_ctx->eth_stats.stop_queue++;
 961                 }
 962         } else if (ret == -EAGAIN) {
 963                 netif_tx_stop_queue(txq);
 964                 ndev_ctx->eth_stats.stop_queue++;
 965         } else {
 966                 netdev_err(ndev,
 967                            "Unable to send packet pages %u len %u, ret %d\n",
 968                            packet->page_buf_cnt, packet->total_data_buflen,
 969                            ret);
 970         }
 971
 972         if (netif_tx_queue_stopped(txq) &&
 973             atomic_read(&nvchan->queue_sends) < 1 &&
 974             !net_device->tx_disable) {
 975                 netif_tx_wake_queue(txq);
 976                 ndev_ctx->eth_stats.wake_queue++;
 977                 if (ret == -EAGAIN)
 978                         ret = -ENOSPC;
 979         }
 980
 981         return ret;
 982 }
 983
 984 /* Move packet out of multi send data (msd), and clear msd */
 985 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 986                                 struct sk_buff **msd_skb,
 987                                 struct multi_send_data *msdp)
 988 {
 989         *msd_skb = msdp->skb;
 990         *msd_send = msdp->pkt;
 991         msdp->skb = NULL;
 992         msdp->pkt = NULL;
 993         msdp->count = 0;
 994 }
 995
 996 /* RCU already held by caller */
 997 int netvsc_send(struct net_device *ndev,
 998                 struct hv_netvsc_packet *packet,
 999                 struct rndis_message *rndis_msg,
1000                 struct hv_page_buffer *pb,
1001                 struct sk_buff *skb,
1002                 bool xdp_tx)
1003 {
1004         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1005         struct netvsc_device *net_device
1006                 = rcu_dereference_bh(ndev_ctx->nvdev);
1007         struct hv_device *device = ndev_ctx->device_ctx;
1008         int ret = 0;
1009         struct netvsc_channel *nvchan;
1010         u32 pktlen = packet->total_data_buflen, msd_len = 0;
1011         unsigned int section_index = NETVSC_INVALID_INDEX;
1012         struct multi_send_data *msdp;
1013         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
1014         struct sk_buff *msd_skb = NULL;
1015         bool try_batch, xmit_more;
1016
1017         /* If device is rescinded, return error and packet will get dropped. */
1018         if (unlikely(!net_device || net_device->destroy))
1019                 return -ENODEV;
1020
1021         nvchan = &net_device->chan_table[packet->q_idx];
1022         packet->send_buf_index = NETVSC_INVALID_INDEX;
1023         packet->cp_partial = false;
1024
1025         /* Send a control message or XDP packet directly without accessing
1026          * msd (Multi-Send Data) field which may be changed during data packet
1027          * processing.
1028          */
1029         if (!skb || xdp_tx)
1030                 return netvsc_send_pkt(device, packet, net_device, pb, skb);
1031
1032         /* batch packets in send buffer if possible */
1033         msdp = &nvchan->msd;
1034         if (msdp->pkt)
1035                 msd_len = msdp->pkt->total_data_buflen;
1036
1037         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
1038         if (try_batch && msd_len + pktlen + net_device->pkt_align <
1039             net_device->send_section_size) {
1040                 section_index = msdp->pkt->send_buf_index;
1041
1042         } else if (try_batch && msd_len + packet->rmsg_size <
1043                    net_device->send_section_size) {
1044                 section_index = msdp->pkt->send_buf_index;
1045                 packet->cp_partial = true;
1046
1047         } else if (pktlen + net_device->pkt_align <
1048                    net_device->send_section_size) {
1049                 section_index = netvsc_get_next_send_section(net_device);
1050                 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
1051                         ++ndev_ctx->eth_stats.tx_send_full;
1052                 } else {
1053                         move_pkt_msd(&msd_send, &msd_skb, msdp);
1054                         msd_len = 0;
1055                 }
1056         }
1057
1058         /* Keep aggregating only if stack says more data is coming
1059          * and not doing mixed modes send and not flow blocked
1060          */
1061         xmit_more = netdev_xmit_more() &&
1062                 !packet->cp_partial &&
1063                 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
1064
1065         if (section_index != NETVSC_INVALID_INDEX) {
1066                 netvsc_copy_to_send_buf(net_device,
1067                                         section_index, msd_len,
1068                                         packet, rndis_msg, pb, xmit_more);
1069
1070                 packet->send_buf_index = section_index;
1071
1072                 if (packet->cp_partial) {
1073                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
1074                         packet->total_data_buflen = msd_len + packet->rmsg_size;
1075                 } else {
1076                         packet->page_buf_cnt = 0;
1077                         packet->total_data_buflen += msd_len;
1078                 }
1079
1080                 if (msdp->pkt) {
1081                         packet->total_packets += msdp->pkt->total_packets;
1082                         packet->total_bytes += msdp->pkt->total_bytes;
1083                 }
1084
1085                 if (msdp->skb)
1086                         dev_consume_skb_any(msdp->skb);
1087
1088                 if (xmit_more) {
1089                         msdp->skb = skb;
1090                         msdp->pkt = packet;
1091                         msdp->count++;
1092                 } else {
1093                         cur_send = packet;
1094                         msdp->skb = NULL;
1095                         msdp->pkt = NULL;
1096                         msdp->count = 0;
1097                 }
1098         } else {
1099                 move_pkt_msd(&msd_send, &msd_skb, msdp);
1100                 cur_send = packet;
1101         }
1102
1103         if (msd_send) {
1104                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1105                                             NULL, msd_skb);
1106
1107                 if (m_ret != 0) {
1108                         netvsc_free_send_slot(net_device,
1109                                               msd_send->send_buf_index);
1110                         dev_kfree_skb_any(msd_skb);
1111                 }
1112         }
1113
1114         if (cur_send)
1115                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1116
1117         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1118                 netvsc_free_send_slot(net_device, section_index);
1119
1120         return ret;
1121 }
1122
1123 /* Send pending recv completions */
1124 static int send_recv_completions(struct net_device *ndev,
1125                                  struct netvsc_device *nvdev,
1126                                  struct netvsc_channel *nvchan)
1127 {
1128         struct multi_recv_comp *mrc = &nvchan->mrc;
1129         struct recv_comp_msg {
1130                 struct nvsp_message_header hdr;
1131                 u32 status;
1132         }  __packed;
1133         struct recv_comp_msg msg = {
1134                 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1135         };
1136         int ret;
1137
1138         while (mrc->first != mrc->next) {
1139                 const struct recv_comp_data *rcd
1140                         = mrc->slots + mrc->first;
1141
1142                 msg.status = rcd->status;
1143                 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1144                                        rcd->tid, VM_PKT_COMP, 0);
1145                 if (unlikely(ret)) {
1146                         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1147
1148                         ++ndev_ctx->eth_stats.rx_comp_busy;
1149                         return ret;
1150                 }
1151
1152                 if (++mrc->first == nvdev->recv_completion_cnt)
1153                         mrc->first = 0;
1154         }
1155
1156         /* receive completion ring has been emptied */
1157         if (unlikely(nvdev->destroy))
1158                 wake_up(&nvdev->wait_drain);
1159
1160         return 0;
1161 }
1162
1163 /* Count how many receive completions are outstanding */
1164 static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1165                                  const struct multi_recv_comp *mrc,
1166                                  u32 *filled, u32 *avail)
1167 {
1168         u32 count = nvdev->recv_completion_cnt;
1169
1170         if (mrc->next >= mrc->first)
1171                 *filled = mrc->next - mrc->first;
1172         else
1173                 *filled = (count - mrc->first) + mrc->next;
1174
1175         *avail = count - *filled - 1;
1176 }
1177
1178 /* Add receive complete to ring to send to host. */
1179 static void enq_receive_complete(struct net_device *ndev,
1180                                  struct netvsc_device *nvdev, u16 q_idx,
1181                                  u64 tid, u32 status)
1182 {
1183         struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1184         struct multi_recv_comp *mrc = &nvchan->mrc;
1185         struct recv_comp_data *rcd;
1186         u32 filled, avail;
1187
1188         recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1189
1190         if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1191                 send_recv_completions(ndev, nvdev, nvchan);
1192                 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1193         }
1194
1195         if (unlikely(!avail)) {
1196                 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1197                            q_idx, tid);
1198                 return;
1199         }
1200
1201         rcd = mrc->slots + mrc->next;
1202         rcd->tid = tid;
1203         rcd->status = status;
1204
1205         if (++mrc->next == nvdev->recv_completion_cnt)
1206                 mrc->next = 0;
1207 }
1208
1209 static int netvsc_receive(struct net_device *ndev,
1210                           struct netvsc_device *net_device,
1211                           struct netvsc_channel *nvchan,
1212                           const struct vmpacket_descriptor *desc)
1213 {
1214         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1215         struct vmbus_channel *channel = nvchan->channel;
1216         const struct vmtransfer_page_packet_header *vmxferpage_packet
1217                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1218         const struct nvsp_message *nvsp = hv_pkt_data(desc);
1219         u32 msglen = hv_pkt_datalen(desc);
1220         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1221         char *recv_buf = net_device->recv_buf;
1222         u32 status = NVSP_STAT_SUCCESS;
1223         int i;
1224         int count = 0;
1225
1226         /* Ensure packet is big enough to read header fields */
1227         if (msglen < sizeof(struct nvsp_message_header)) {
1228                 netif_err(net_device_ctx, rx_err, ndev,
1229                           "invalid nvsp header, length too small: %u\n",
1230                           msglen);
1231                 return 0;
1232         }
1233
1234         /* Make sure this is a valid nvsp packet */
1235         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1236                 netif_err(net_device_ctx, rx_err, ndev,
1237                           "Unknown nvsp packet type received %u\n",
1238                           nvsp->hdr.msg_type);
1239                 return 0;
1240         }
1241
1242         /* Validate xfer page pkt header */
1243         if ((desc->offset8 << 3) < sizeof(struct vmtransfer_page_packet_header)) {
1244                 netif_err(net_device_ctx, rx_err, ndev,
1245                           "Invalid xfer page pkt, offset too small: %u\n",
1246                           desc->offset8 << 3);
1247                 return 0;
1248         }
1249
1250         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1251                 netif_err(net_device_ctx, rx_err, ndev,
1252                           "Invalid xfer page set id - expecting %x got %x\n",
1253                           NETVSC_RECEIVE_BUFFER_ID,
1254                           vmxferpage_packet->xfer_pageset_id);
1255                 return 0;
1256         }
1257
1258         count = vmxferpage_packet->range_cnt;
1259
1260         /* Check count for a valid value */
1261         if (NETVSC_XFER_HEADER_SIZE(count) > desc->offset8 << 3) {
1262                 netif_err(net_device_ctx, rx_err, ndev,
1263                           "Range count is not valid: %d\n",
1264                           count);
1265                 return 0;
1266         }
1267
1268         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1269         for (i = 0; i < count; i++) {
1270                 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1271                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1272                 void *data;
1273                 int ret;
1274
1275                 if (unlikely(offset > net_device->recv_buf_size ||
1276                              buflen > net_device->recv_buf_size - offset)) {
1277                         nvchan->rsc.cnt = 0;
1278                         status = NVSP_STAT_FAIL;
1279                         netif_err(net_device_ctx, rx_err, ndev,
1280                                   "Packet offset:%u + len:%u too big\n",
1281                                   offset, buflen);
1282
1283                         continue;
1284                 }
1285
1286                 data = recv_buf + offset;
1287
1288                 nvchan->rsc.is_last = (i == count - 1);
1289
1290                 trace_rndis_recv(ndev, q_idx, data);
1291
1292                 /* Pass it to the upper layer */
1293                 ret = rndis_filter_receive(ndev, net_device,
1294                                            nvchan, data, buflen);
1295
1296                 if (unlikely(ret != NVSP_STAT_SUCCESS))
1297                         status = NVSP_STAT_FAIL;
1298         }
1299
1300         enq_receive_complete(ndev, net_device, q_idx,
1301                              vmxferpage_packet->d.trans_id, status);
1302
1303         return count;
1304 }
1305
1306 static void netvsc_send_table(struct net_device *ndev,
1307                               struct netvsc_device *nvscdev,
1308                               const struct nvsp_message *nvmsg,
1309                               u32 msglen)
1310 {
1311         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1312         u32 count, offset, *tab;
1313         int i;
1314
1315         /* Ensure packet is big enough to read send_table fields */
1316         if (msglen < sizeof(struct nvsp_message_header) +
1317                      sizeof(struct nvsp_5_send_indirect_table)) {
1318                 netdev_err(ndev, "nvsp_v5_msg length too small: %u\n", msglen);
1319                 return;
1320         }
1321
1322         count = nvmsg->msg.v5_msg.send_table.count;
1323         offset = nvmsg->msg.v5_msg.send_table.offset;
1324
1325         if (count != VRSS_SEND_TAB_SIZE) {
1326                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1327                 return;
1328         }
1329
1330         /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1331          * wrong due to a host bug. So fix the offset here.
1332          */
1333         if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1334             msglen >= sizeof(struct nvsp_message_header) +
1335             sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1336                 offset = sizeof(struct nvsp_message_header) +
1337                          sizeof(union nvsp_6_message_uber);
1338
1339         /* Boundary check for all versions */
1340         if (offset > msglen - count * sizeof(u32)) {
1341                 netdev_err(ndev, "Received send-table offset too big:%u\n",
1342                            offset);
1343                 return;
1344         }
1345
1346         tab = (void *)nvmsg + offset;
1347
1348         for (i = 0; i < count; i++)
1349                 net_device_ctx->tx_table[i] = tab[i];
1350 }
1351
1352 static void netvsc_send_vf(struct net_device *ndev,
1353                            const struct nvsp_message *nvmsg,
1354                            u32 msglen)
1355 {
1356         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1357
1358         /* Ensure packet is big enough to read its fields */
1359         if (msglen < sizeof(struct nvsp_message_header) +
1360                      sizeof(struct nvsp_4_send_vf_association)) {
1361                 netdev_err(ndev, "nvsp_v4_msg length too small: %u\n", msglen);
1362                 return;
1363         }
1364
1365         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1366         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1367         netdev_info(ndev, "VF slot %u %s\n",
1368                     net_device_ctx->vf_serial,
1369                     net_device_ctx->vf_alloc ? "added" : "removed");
1370 }
1371
1372 static void netvsc_receive_inband(struct net_device *ndev,
1373                                   struct netvsc_device *nvscdev,
1374                                   const struct vmpacket_descriptor *desc)
1375 {
1376         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1377         u32 msglen = hv_pkt_datalen(desc);
1378
1379         /* Ensure packet is big enough to read header fields */
1380         if (msglen < sizeof(struct nvsp_message_header)) {
1381                 netdev_err(ndev, "inband nvsp_message length too small: %u\n", msglen);
1382                 return;
1383         }
1384
1385         switch (nvmsg->hdr.msg_type) {
1386         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1387                 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1388                 break;
1389
1390         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1391                 netvsc_send_vf(ndev, nvmsg, msglen);
1392                 break;
1393         }
1394 }
1395
1396 static int netvsc_process_raw_pkt(struct hv_device *device,
1397                                   struct netvsc_channel *nvchan,
1398                                   struct netvsc_device *net_device,
1399                                   struct net_device *ndev,
1400                                   const struct vmpacket_descriptor *desc,
1401                                   int budget)
1402 {
1403         struct vmbus_channel *channel = nvchan->channel;
1404         const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1405
1406         trace_nvsp_recv(ndev, channel, nvmsg);
1407
1408         switch (desc->type) {
1409         case VM_PKT_COMP:
1410                 netvsc_send_completion(ndev, net_device, channel, desc, budget);
1411                 break;
1412
1413         case VM_PKT_DATA_USING_XFER_PAGES:
1414                 return netvsc_receive(ndev, net_device, nvchan, desc);
1415                 break;
1416
1417         case VM_PKT_DATA_INBAND:
1418                 netvsc_receive_inband(ndev, net_device, desc);
1419                 break;
1420
1421         default:
1422                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1423                            desc->type, desc->trans_id);
1424                 break;
1425         }
1426
1427         return 0;
1428 }
1429
1430 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1431 {
1432         struct vmbus_channel *primary = channel->primary_channel;
1433
1434         return primary ? primary->device_obj : channel->device_obj;
1435 }
1436
1437 /* Network processing softirq
1438  * Process data in incoming ring buffer from host
1439  * Stops when ring is empty or budget is met or exceeded.
1440  */
1441 int netvsc_poll(struct napi_struct *napi, int budget)
1442 {
1443         struct netvsc_channel *nvchan
1444                 = container_of(napi, struct netvsc_channel, napi);
1445         struct netvsc_device *net_device = nvchan->net_device;
1446         struct vmbus_channel *channel = nvchan->channel;
1447         struct hv_device *device = netvsc_channel_to_device(channel);
1448         struct net_device *ndev = hv_get_drvdata(device);
1449         int work_done = 0;
1450         int ret;
1451
1452         /* If starting a new interval */
1453         if (!nvchan->desc)
1454                 nvchan->desc = hv_pkt_iter_first(channel);
1455
1456         while (nvchan->desc && work_done < budget) {
1457                 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1458                                                     ndev, nvchan->desc, budget);
1459                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1460         }
1461
1462         /* Send any pending receive completions */
1463         ret = send_recv_completions(ndev, net_device, nvchan);
1464
1465         /* If it did not exhaust NAPI budget this time
1466          *  and not doing busy poll
1467          * then re-enable host interrupts
1468          *  and reschedule if ring is not empty
1469          *   or sending receive completion failed.
1470          */
1471         if (work_done < budget &&
1472             napi_complete_done(napi, work_done) &&
1473             (ret || hv_end_read(&channel->inbound)) &&
1474             napi_schedule_prep(napi)) {
1475                 hv_begin_read(&channel->inbound);
1476                 __napi_schedule(napi);
1477         }
1478
1479         /* Driver may overshoot since multiple packets per descriptor */
1480         return min(work_done, budget);
1481 }
1482
1483 /* Call back when data is available in host ring buffer.
1484  * Processing is deferred until network softirq (NAPI)
1485  */
1486 void netvsc_channel_cb(void *context)
1487 {
1488         struct netvsc_channel *nvchan = context;
1489         struct vmbus_channel *channel = nvchan->channel;
1490         struct hv_ring_buffer_info *rbi = &channel->inbound;
1491
1492         /* preload first vmpacket descriptor */
1493         prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1494
1495         if (napi_schedule_prep(&nvchan->napi)) {
1496                 /* disable interrupts from host */
1497                 hv_begin_read(rbi);
1498
1499                 __napi_schedule_irqoff(&nvchan->napi);
1500         }
1501 }
1502
1503 /*
1504  * netvsc_device_add - Callback when the device belonging to this
1505  * driver is added
1506  */
1507 struct netvsc_device *netvsc_device_add(struct hv_device *device,
1508                                 const struct netvsc_device_info *device_info)
1509 {
1510         int i, ret = 0;
1511         struct netvsc_device *net_device;
1512         struct net_device *ndev = hv_get_drvdata(device);
1513         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1514
1515         net_device = alloc_net_device();
1516         if (!net_device)
1517                 return ERR_PTR(-ENOMEM);
1518
1519         for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1520                 net_device_ctx->tx_table[i] = 0;
1521
1522         /* Because the device uses NAPI, all the interrupt batching and
1523          * control is done via Net softirq, not the channel handling
1524          */
1525         set_channel_read_mode(device->channel, HV_CALL_ISR);
1526
1527         /* If we're reopening the device we may have multiple queues, fill the
1528          * chn_table with the default channel to use it before subchannels are
1529          * opened.
1530          * Initialize the channel state before we open;
1531          * we can be interrupted as soon as we open the channel.
1532          */
1533
1534         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1535                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1536
1537                 nvchan->channel = device->channel;
1538                 nvchan->net_device = net_device;
1539                 u64_stats_init(&nvchan->tx_stats.syncp);
1540                 u64_stats_init(&nvchan->rx_stats.syncp);
1541
1542                 ret = xdp_rxq_info_reg(&nvchan->xdp_rxq, ndev, i, 0);
1543
1544                 if (ret) {
1545                         netdev_err(ndev, "xdp_rxq_info_reg fail: %d\n", ret);
1546                         goto cleanup2;
1547                 }
1548
1549                 ret = xdp_rxq_info_reg_mem_model(&nvchan->xdp_rxq,
1550                                                  MEM_TYPE_PAGE_SHARED, NULL);
1551
1552                 if (ret) {
1553                         netdev_err(ndev, "xdp reg_mem_model fail: %d\n", ret);
1554                         goto cleanup2;
1555                 }
1556         }
1557
1558         /* Enable NAPI handler before init callbacks */
1559         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1560                        netvsc_poll, NAPI_POLL_WEIGHT);
1561
1562         /* Open the channel */
1563         device->channel->rqstor_size = netvsc_rqstor_size(netvsc_ring_bytes);
1564         ret = vmbus_open(device->channel, netvsc_ring_bytes,
1565                          netvsc_ring_bytes,  NULL, 0,
1566                          netvsc_channel_cb, net_device->chan_table);
1567
1568         if (ret != 0) {
1569                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1570                 goto cleanup;
1571         }
1572
1573         /* Channel is opened */
1574         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1575
1576         napi_enable(&net_device->chan_table[0].napi);
1577
1578         /* Connect with the NetVsp */
1579         ret = netvsc_connect_vsp(device, net_device, device_info);
1580         if (ret != 0) {
1581                 netdev_err(ndev,
1582                         "unable to connect to NetVSP - %d\n", ret);
1583                 goto close;
1584         }
1585
1586         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1587          * populated.
1588          */
1589         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1590
1591         return net_device;
1592
1593 close:
1594         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1595         napi_disable(&net_device->chan_table[0].napi);
1596
1597         /* Now, we can close the channel safely */
1598         vmbus_close(device->channel);
1599
1600 cleanup:
1601         netif_napi_del(&net_device->chan_table[0].napi);
1602
1603 cleanup2:
1604         free_netvsc_device(&net_device->rcu);
1605
1606         return ERR_PTR(ret);
1607 }