drivers/net/hyperv/netvsc.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * Authors:
  17  *   Haiyang Zhang <haiyangz@microsoft.com>
  18  *   Hank Janssen  <hjanssen@microsoft.com>
  19  */
  20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22 #include <linux/kernel.h>
  23 #include <linux/sched.h>
  24 #include <linux/wait.h>
  25 #include <linux/mm.h>
  26 #include <linux/delay.h>
  27 #include <linux/io.h>
  28 #include <linux/slab.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/vmalloc.h>
  32 #include <asm/sync_bitops.h>
  33
  34 #include "hyperv_net.h"
  35
  36 /*
  37  * Switch the data path from the synthetic interface to the VF
  38  * interface.
  39  */
  40 void netvsc_switch_datapath(struct net_device *ndev, bool vf)
  41 {
  42         struct net_device_context *net_device_ctx = netdev_priv(ndev);
  43         struct hv_device *dev = net_device_ctx->device_ctx;
  44         struct netvsc_device *nv_dev = net_device_ctx->nvdev;
  45         struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
  46
  47         memset(init_pkt, 0, sizeof(struct nvsp_message));
  48         init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
  49         if (vf)
  50                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  51                         NVSP_DATAPATH_VF;
  52         else
  53                 init_pkt->msg.v4_msg.active_dp.active_datapath =
  54                         NVSP_DATAPATH_SYNTHETIC;
  55
  56         vmbus_sendpacket(dev->channel, init_pkt,
  57                                sizeof(struct nvsp_message),
  58                                (unsigned long)init_pkt,
  59                                VM_PKT_DATA_INBAND, 0);
  60
  61         net_device_ctx->datapath = vf;
  62 }
  63
  64 static struct netvsc_device *alloc_net_device(void)
  65 {
  66         struct netvsc_device *net_device;
  67
  68         net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
  69         if (!net_device)
  70                 return NULL;
  71
  72         net_device->chan_table[0].mrc.buf
  73                 = vzalloc(NETVSC_RECVSLOT_MAX * sizeof(struct recv_comp_data));
  74
  75         init_waitqueue_head(&net_device->wait_drain);
  76         net_device->destroy = false;
  77         atomic_set(&net_device->open_cnt, 0);
  78         net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
  79         net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
  80         init_completion(&net_device->channel_init_wait);
  81
  82         return net_device;
  83 }
  84
  85 static void free_netvsc_device(struct rcu_head *head)
  86 {
  87         struct netvsc_device *nvdev
  88                 = container_of(head, struct netvsc_device, rcu);
  89         int i;
  90
  91         for (i = 0; i < VRSS_CHANNEL_MAX; i++)
  92                 vfree(nvdev->chan_table[i].mrc.buf);
  93
  94         kfree(nvdev);
  95 }
  96
  97 static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
  98 {
  99         call_rcu(&nvdev->rcu, free_netvsc_device);
 100 }
 101
 102 static void netvsc_destroy_buf(struct hv_device *device)
 103 {
 104         struct nvsp_message *revoke_packet;
 105         struct net_device *ndev = hv_get_drvdata(device);
 106         struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
 107         int ret;
 108
 109         /*
 110          * If we got a section count, it means we received a
 111          * SendReceiveBufferComplete msg (ie sent
 112          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
 113          * to send a revoke msg here
 114          */
 115         if (net_device->recv_section_cnt) {
 116                 /* Send the revoke receive buffer */
 117                 revoke_packet = &net_device->revoke_packet;
 118                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 119
 120                 revoke_packet->hdr.msg_type =
 121                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
 122                 revoke_packet->msg.v1_msg.
 123                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 124
 125                 ret = vmbus_sendpacket(device->channel,
 126                                        revoke_packet,
 127                                        sizeof(struct nvsp_message),
 128                                        (unsigned long)revoke_packet,
 129                                        VM_PKT_DATA_INBAND, 0);
 130                 /* If the failure is because the channel is rescinded;
 131                  * ignore the failure since we cannot send on a rescinded
 132                  * channel. This would allow us to properly cleanup
 133                  * even when the channel is rescinded.
 134                  */
 135                 if (device->channel->rescind)
 136                         ret = 0;
 137                 /*
 138                  * If we failed here, we might as well return and
 139                  * have a leak rather than continue and a bugchk
 140                  */
 141                 if (ret != 0) {
 142                         netdev_err(ndev, "unable to send "
 143                                 "revoke receive buffer to netvsp\n");
 144                         return;
 145                 }
 146         }
 147
 148         /* Teardown the gpadl on the vsp end */
 149         if (net_device->recv_buf_gpadl_handle) {
 150                 ret = vmbus_teardown_gpadl(device->channel,
 151                                            net_device->recv_buf_gpadl_handle);
 152
 153                 /* If we failed here, we might as well return and have a leak
 154                  * rather than continue and a bugchk
 155                  */
 156                 if (ret != 0) {
 157                         netdev_err(ndev,
 158                                    "unable to teardown receive buffer's gpadl\n");
 159                         return;
 160                 }
 161                 net_device->recv_buf_gpadl_handle = 0;
 162         }
 163
 164         if (net_device->recv_buf) {
 165                 /* Free up the receive buffer */
 166                 vfree(net_device->recv_buf);
 167                 net_device->recv_buf = NULL;
 168         }
 169
 170         if (net_device->recv_section) {
 171                 net_device->recv_section_cnt = 0;
 172                 kfree(net_device->recv_section);
 173                 net_device->recv_section = NULL;
 174         }
 175
 176         /* Deal with the send buffer we may have setup.
 177          * If we got a  send section size, it means we received a
 178          * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
 179          * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 180          * to send a revoke msg here
 181          */
 182         if (net_device->send_section_size) {
 183                 /* Send the revoke receive buffer */
 184                 revoke_packet = &net_device->revoke_packet;
 185                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
 186
 187                 revoke_packet->hdr.msg_type =
 188                         NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
 189                 revoke_packet->msg.v1_msg.revoke_send_buf.id =
 190                         NETVSC_SEND_BUFFER_ID;
 191
 192                 ret = vmbus_sendpacket(device->channel,
 193                                        revoke_packet,
 194                                        sizeof(struct nvsp_message),
 195                                        (unsigned long)revoke_packet,
 196                                        VM_PKT_DATA_INBAND, 0);
 197
 198                 /* If the failure is because the channel is rescinded;
 199                  * ignore the failure since we cannot send on a rescinded
 200                  * channel. This would allow us to properly cleanup
 201                  * even when the channel is rescinded.
 202                  */
 203                 if (device->channel->rescind)
 204                         ret = 0;
 205
 206                 /* If we failed here, we might as well return and
 207                  * have a leak rather than continue and a bugchk
 208                  */
 209                 if (ret != 0) {
 210                         netdev_err(ndev, "unable to send "
 211                                    "revoke send buffer to netvsp\n");
 212                         return;
 213                 }
 214         }
 215         /* Teardown the gpadl on the vsp end */
 216         if (net_device->send_buf_gpadl_handle) {
 217                 ret = vmbus_teardown_gpadl(device->channel,
 218                                            net_device->send_buf_gpadl_handle);
 219
 220                 /* If we failed here, we might as well return and have a leak
 221                  * rather than continue and a bugchk
 222                  */
 223                 if (ret != 0) {
 224                         netdev_err(ndev,
 225                                    "unable to teardown send buffer's gpadl\n");
 226                         return;
 227                 }
 228                 net_device->send_buf_gpadl_handle = 0;
 229         }
 230         if (net_device->send_buf) {
 231                 /* Free up the send buffer */
 232                 vfree(net_device->send_buf);
 233                 net_device->send_buf = NULL;
 234         }
 235         kfree(net_device->send_section_map);
 236 }
 237
 238 static int netvsc_init_buf(struct hv_device *device,
 239                            struct netvsc_device *net_device)
 240 {
 241         int ret = 0;
 242         struct nvsp_message *init_packet;
 243         struct net_device *ndev;
 244         size_t map_words;
 245         int node;
 246
 247         ndev = hv_get_drvdata(device);
 248
 249         node = cpu_to_node(device->channel->target_cpu);
 250         net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
 251         if (!net_device->recv_buf)
 252                 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
 253
 254         if (!net_device->recv_buf) {
 255                 netdev_err(ndev, "unable to allocate receive "
 256                         "buffer of size %d\n", net_device->recv_buf_size);
 257                 ret = -ENOMEM;
 258                 goto cleanup;
 259         }
 260
 261         /*
 262          * Establish the gpadl handle for this buffer on this
 263          * channel.  Note: This call uses the vmbus connection rather
 264          * than the channel to establish the gpadl handle.
 265          */
 266         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
 267                                     net_device->recv_buf_size,
 268                                     &net_device->recv_buf_gpadl_handle);
 269         if (ret != 0) {
 270                 netdev_err(ndev,
 271                         "unable to establish receive buffer's gpadl\n");
 272                 goto cleanup;
 273         }
 274
 275         /* Notify the NetVsp of the gpadl handle */
 276         init_packet = &net_device->channel_init_pkt;
 277         memset(init_packet, 0, sizeof(struct nvsp_message));
 278         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
 279         init_packet->msg.v1_msg.send_recv_buf.
 280                 gpadl_handle = net_device->recv_buf_gpadl_handle;
 281         init_packet->msg.v1_msg.
 282                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
 283
 284         /* Send the gpadl notification request */
 285         ret = vmbus_sendpacket(device->channel, init_packet,
 286                                sizeof(struct nvsp_message),
 287                                (unsigned long)init_packet,
 288                                VM_PKT_DATA_INBAND,
 289                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 290         if (ret != 0) {
 291                 netdev_err(ndev,
 292                         "unable to send receive buffer's gpadl to netvsp\n");
 293                 goto cleanup;
 294         }
 295
 296         wait_for_completion(&net_device->channel_init_wait);
 297
 298         /* Check the response */
 299         if (init_packet->msg.v1_msg.
 300             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
 301                 netdev_err(ndev, "Unable to complete receive buffer "
 302                            "initialization with NetVsp - status %d\n",
 303                            init_packet->msg.v1_msg.
 304                            send_recv_buf_complete.status);
 305                 ret = -EINVAL;
 306                 goto cleanup;
 307         }
 308
 309         /* Parse the response */
 310
 311         net_device->recv_section_cnt = init_packet->msg.
 312                 v1_msg.send_recv_buf_complete.num_sections;
 313
 314         net_device->recv_section = kmemdup(
 315                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
 316                 net_device->recv_section_cnt *
 317                 sizeof(struct nvsp_1_receive_buffer_section),
 318                 GFP_KERNEL);
 319         if (net_device->recv_section == NULL) {
 320                 ret = -EINVAL;
 321                 goto cleanup;
 322         }
 323
 324         /*
 325          * For 1st release, there should only be 1 section that represents the
 326          * entire receive buffer
 327          */
 328         if (net_device->recv_section_cnt != 1 ||
 329             net_device->recv_section->offset != 0) {
 330                 ret = -EINVAL;
 331                 goto cleanup;
 332         }
 333
 334         /* Now setup the send buffer.
 335          */
 336         net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
 337         if (!net_device->send_buf)
 338                 net_device->send_buf = vzalloc(net_device->send_buf_size);
 339         if (!net_device->send_buf) {
 340                 netdev_err(ndev, "unable to allocate send "
 341                            "buffer of size %d\n", net_device->send_buf_size);
 342                 ret = -ENOMEM;
 343                 goto cleanup;
 344         }
 345
 346         /* Establish the gpadl handle for this buffer on this
 347          * channel.  Note: This call uses the vmbus connection rather
 348          * than the channel to establish the gpadl handle.
 349          */
 350         ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
 351                                     net_device->send_buf_size,
 352                                     &net_device->send_buf_gpadl_handle);
 353         if (ret != 0) {
 354                 netdev_err(ndev,
 355                            "unable to establish send buffer's gpadl\n");
 356                 goto cleanup;
 357         }
 358
 359         /* Notify the NetVsp of the gpadl handle */
 360         init_packet = &net_device->channel_init_pkt;
 361         memset(init_packet, 0, sizeof(struct nvsp_message));
 362         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
 363         init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
 364                 net_device->send_buf_gpadl_handle;
 365         init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
 366
 367         /* Send the gpadl notification request */
 368         ret = vmbus_sendpacket(device->channel, init_packet,
 369                                sizeof(struct nvsp_message),
 370                                (unsigned long)init_packet,
 371                                VM_PKT_DATA_INBAND,
 372                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 373         if (ret != 0) {
 374                 netdev_err(ndev,
 375                            "unable to send send buffer's gpadl to netvsp\n");
 376                 goto cleanup;
 377         }
 378
 379         wait_for_completion(&net_device->channel_init_wait);
 380
 381         /* Check the response */
 382         if (init_packet->msg.v1_msg.
 383             send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
 384                 netdev_err(ndev, "Unable to complete send buffer "
 385                            "initialization with NetVsp - status %d\n",
 386                            init_packet->msg.v1_msg.
 387                            send_send_buf_complete.status);
 388                 ret = -EINVAL;
 389                 goto cleanup;
 390         }
 391
 392         /* Parse the response */
 393         net_device->send_section_size = init_packet->msg.
 394                                 v1_msg.send_send_buf_complete.section_size;
 395
 396         /* Section count is simply the size divided by the section size.
 397          */
 398         net_device->send_section_cnt =
 399                 net_device->send_buf_size / net_device->send_section_size;
 400
 401         netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 402                    net_device->send_section_size, net_device->send_section_cnt);
 403
 404         /* Setup state for managing the send buffer. */
 405         map_words = DIV_ROUND_UP(net_device->send_section_cnt, BITS_PER_LONG);
 406
 407         net_device->send_section_map = kcalloc(map_words, sizeof(ulong), GFP_KERNEL);
 408         if (net_device->send_section_map == NULL) {
 409                 ret = -ENOMEM;
 410                 goto cleanup;
 411         }
 412
 413         goto exit;
 414
 415 cleanup:
 416         netvsc_destroy_buf(device);
 417
 418 exit:
 419         return ret;
 420 }
 421
 422 /* Negotiate NVSP protocol version */
 423 static int negotiate_nvsp_ver(struct hv_device *device,
 424                               struct netvsc_device *net_device,
 425                               struct nvsp_message *init_packet,
 426                               u32 nvsp_ver)
 427 {
 428         struct net_device *ndev = hv_get_drvdata(device);
 429         int ret;
 430
 431         memset(init_packet, 0, sizeof(struct nvsp_message));
 432         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
 433         init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
 434         init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
 435
 436         /* Send the init request */
 437         ret = vmbus_sendpacket(device->channel, init_packet,
 438                                sizeof(struct nvsp_message),
 439                                (unsigned long)init_packet,
 440                                VM_PKT_DATA_INBAND,
 441                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 442
 443         if (ret != 0)
 444                 return ret;
 445
 446         wait_for_completion(&net_device->channel_init_wait);
 447
 448         if (init_packet->msg.init_msg.init_complete.status !=
 449             NVSP_STAT_SUCCESS)
 450                 return -EINVAL;
 451
 452         if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
 453                 return 0;
 454
 455         /* NVSPv2 or later: Send NDIS config */
 456         memset(init_packet, 0, sizeof(struct nvsp_message));
 457         init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
 458         init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
 459         init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
 460
 461         if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
 462                 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
 463
 464                 /* Teaming bit is needed to receive link speed updates */
 465                 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
 466         }
 467
 468         ret = vmbus_sendpacket(device->channel, init_packet,
 469                                 sizeof(struct nvsp_message),
 470                                 (unsigned long)init_packet,
 471                                 VM_PKT_DATA_INBAND, 0);
 472
 473         return ret;
 474 }
 475
 476 static int netvsc_connect_vsp(struct hv_device *device,
 477                               struct netvsc_device *net_device)
 478 {
 479         const u32 ver_list[] = {
 480                 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
 481                 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5
 482         };
 483         struct nvsp_message *init_packet;
 484         int ndis_version, i, ret;
 485
 486         init_packet = &net_device->channel_init_pkt;
 487
 488         /* Negotiate the latest NVSP protocol supported */
 489         for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
 490                 if (negotiate_nvsp_ver(device, net_device, init_packet,
 491                                        ver_list[i])  == 0) {
 492                         net_device->nvsp_version = ver_list[i];
 493                         break;
 494                 }
 495
 496         if (i < 0) {
 497                 ret = -EPROTO;
 498                 goto cleanup;
 499         }
 500
 501         pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
 502
 503         /* Send the ndis version */
 504         memset(init_packet, 0, sizeof(struct nvsp_message));
 505
 506         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
 507                 ndis_version = 0x00060001;
 508         else
 509                 ndis_version = 0x0006001e;
 510
 511         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
 512         init_packet->msg.v1_msg.
 513                 send_ndis_ver.ndis_major_ver =
 514                                 (ndis_version & 0xFFFF0000) >> 16;
 515         init_packet->msg.v1_msg.
 516                 send_ndis_ver.ndis_minor_ver =
 517                                 ndis_version & 0xFFFF;
 518
 519         /* Send the init request */
 520         ret = vmbus_sendpacket(device->channel, init_packet,
 521                                 sizeof(struct nvsp_message),
 522                                 (unsigned long)init_packet,
 523                                 VM_PKT_DATA_INBAND, 0);
 524         if (ret != 0)
 525                 goto cleanup;
 526
 527         /* Post the big receive buffer to NetVSP */
 528         if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
 529                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
 530         else
 531                 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
 532         net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 533
 534         ret = netvsc_init_buf(device, net_device);
 535
 536 cleanup:
 537         return ret;
 538 }
 539
 540 static void netvsc_disconnect_vsp(struct hv_device *device)
 541 {
 542         netvsc_destroy_buf(device);
 543 }
 544
 545 /*
 546  * netvsc_device_remove - Callback when the root bus device is removed
 547  */
 548 void netvsc_device_remove(struct hv_device *device)
 549 {
 550         struct net_device *ndev = hv_get_drvdata(device);
 551         struct net_device_context *net_device_ctx = netdev_priv(ndev);
 552         struct netvsc_device *net_device = net_device_ctx->nvdev;
 553         int i;
 554
 555         netvsc_disconnect_vsp(device);
 556
 557         RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
 558
 559         /*
 560          * At this point, no one should be accessing net_device
 561          * except in here
 562          */
 563         netdev_dbg(ndev, "net device safe to remove\n");
 564
 565         /* Now, we can close the channel safely */
 566         vmbus_close(device->channel);
 567
 568         /* And dissassociate NAPI context from device */
 569         for (i = 0; i < net_device->num_chn; i++)
 570                 netif_napi_del(&net_device->chan_table[i].napi);
 571
 572         /* Release all resources */
 573         free_netvsc_device_rcu(net_device);
 574 }
 575
 576 #define RING_AVAIL_PERCENT_HIWATER 20
 577 #define RING_AVAIL_PERCENT_LOWATER 10
 578
 579 /*
 580  * Get the percentage of available bytes to write in the ring.
 581  * The return value is in range from 0 to 100.
 582  */
 583 static inline u32 hv_ringbuf_avail_percent(
 584                 struct hv_ring_buffer_info *ring_info)
 585 {
 586         u32 avail_read, avail_write;
 587
 588         hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
 589
 590         return avail_write * 100 / ring_info->ring_datasize;
 591 }
 592
 593 static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
 594                                          u32 index)
 595 {
 596         sync_change_bit(index, net_device->send_section_map);
 597 }
 598
 599 static void netvsc_send_tx_complete(struct netvsc_device *net_device,
 600                                     struct vmbus_channel *incoming_channel,
 601                                     struct hv_device *device,
 602                                     const struct vmpacket_descriptor *desc,
 603                                     int budget)
 604 {
 605         struct sk_buff *skb = (struct sk_buff *)(unsigned long)desc->trans_id;
 606         struct net_device *ndev = hv_get_drvdata(device);
 607         struct vmbus_channel *channel = device->channel;
 608         u16 q_idx = 0;
 609         int queue_sends;
 610
 611         /* Notify the layer above us */
 612         if (likely(skb)) {
 613                 const struct hv_netvsc_packet *packet
 614                         = (struct hv_netvsc_packet *)skb->cb;
 615                 u32 send_index = packet->send_buf_index;
 616                 struct netvsc_stats *tx_stats;
 617
 618                 if (send_index != NETVSC_INVALID_INDEX)
 619                         netvsc_free_send_slot(net_device, send_index);
 620                 q_idx = packet->q_idx;
 621                 channel = incoming_channel;
 622
 623                 tx_stats = &net_device->chan_table[q_idx].tx_stats;
 624
 625                 u64_stats_update_begin(&tx_stats->syncp);
 626                 tx_stats->packets += packet->total_packets;
 627                 tx_stats->bytes += packet->total_bytes;
 628                 u64_stats_update_end(&tx_stats->syncp);
 629
 630                 napi_consume_skb(skb, budget);
 631         }
 632
 633         queue_sends =
 634                 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
 635
 636         if (net_device->destroy && queue_sends == 0)
 637                 wake_up(&net_device->wait_drain);
 638
 639         if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
 640             (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
 641              queue_sends < 1))
 642                 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
 643 }
 644
 645 static void netvsc_send_completion(struct netvsc_device *net_device,
 646                                    struct vmbus_channel *incoming_channel,
 647                                    struct hv_device *device,
 648                                    const struct vmpacket_descriptor *desc,
 649                                    int budget)
 650 {
 651         struct nvsp_message *nvsp_packet = hv_pkt_data(desc);
 652         struct net_device *ndev = hv_get_drvdata(device);
 653
 654         switch (nvsp_packet->hdr.msg_type) {
 655         case NVSP_MSG_TYPE_INIT_COMPLETE:
 656         case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
 657         case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
 658         case NVSP_MSG5_TYPE_SUBCHANNEL:
 659                 /* Copy the response back */
 660                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
 661                        sizeof(struct nvsp_message));
 662                 complete(&net_device->channel_init_wait);
 663                 break;
 664
 665         case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
 666                 netvsc_send_tx_complete(net_device, incoming_channel,
 667                                         device, desc, budget);
 668                 break;
 669
 670         default:
 671                 netdev_err(ndev,
 672                            "Unknown send completion type %d received!!\n",
 673                            nvsp_packet->hdr.msg_type);
 674         }
 675 }
 676
 677 static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
 678 {
 679         unsigned long *map_addr = net_device->send_section_map;
 680         unsigned int i;
 681
 682         for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
 683                 if (sync_test_and_set_bit(i, map_addr) == 0)
 684                         return i;
 685         }
 686
 687         return NETVSC_INVALID_INDEX;
 688 }
 689
 690 static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 691                                    unsigned int section_index,
 692                                    u32 pend_size,
 693                                    struct hv_netvsc_packet *packet,
 694                                    struct rndis_message *rndis_msg,
 695                                    struct hv_page_buffer **pb,
 696                                    struct sk_buff *skb)
 697 {
 698         char *start = net_device->send_buf;
 699         char *dest = start + (section_index * net_device->send_section_size)
 700                      + pend_size;
 701         int i;
 702         u32 msg_size = 0;
 703         u32 padding = 0;
 704         u32 remain = packet->total_data_buflen % net_device->pkt_align;
 705         u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
 706                 packet->page_buf_cnt;
 707
 708         /* Add padding */
 709         if (skb->xmit_more && remain && !packet->cp_partial) {
 710                 padding = net_device->pkt_align - remain;
 711                 rndis_msg->msg_len += padding;
 712                 packet->total_data_buflen += padding;
 713         }
 714
 715         for (i = 0; i < page_count; i++) {
 716                 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
 717                 u32 offset = (*pb)[i].offset;
 718                 u32 len = (*pb)[i].len;
 719
 720                 memcpy(dest, (src + offset), len);
 721                 msg_size += len;
 722                 dest += len;
 723         }
 724
 725         if (padding) {
 726                 memset(dest, 0, padding);
 727                 msg_size += padding;
 728         }
 729
 730         return msg_size;
 731 }
 732
 733 static inline int netvsc_send_pkt(
 734         struct hv_device *device,
 735         struct hv_netvsc_packet *packet,
 736         struct netvsc_device *net_device,
 737         struct hv_page_buffer **pb,
 738         struct sk_buff *skb)
 739 {
 740         struct nvsp_message nvmsg;
 741         struct netvsc_channel *nvchan
 742                 = &net_device->chan_table[packet->q_idx];
 743         struct vmbus_channel *out_channel = nvchan->channel;
 744         struct net_device *ndev = hv_get_drvdata(device);
 745         struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
 746         u64 req_id;
 747         int ret;
 748         struct hv_page_buffer *pgbuf;
 749         u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
 750
 751         nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
 752         if (skb != NULL) {
 753                 /* 0 is RMC_DATA; */
 754                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
 755         } else {
 756                 /* 1 is RMC_CONTROL; */
 757                 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
 758         }
 759
 760         nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
 761                 packet->send_buf_index;
 762         if (packet->send_buf_index == NETVSC_INVALID_INDEX)
 763                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
 764         else
 765                 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
 766                         packet->total_data_buflen;
 767
 768         req_id = (ulong)skb;
 769
 770         if (out_channel->rescind)
 771                 return -ENODEV;
 772
 773         if (packet->page_buf_cnt) {
 774                 pgbuf = packet->cp_partial ? (*pb) +
 775                         packet->rmsg_pgcnt : (*pb);
 776                 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 777                                                       pgbuf,
 778                                                       packet->page_buf_cnt,
 779                                                       &nvmsg,
 780                                                       sizeof(struct nvsp_message),
 781                                                       req_id,
 782                                                       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 783         } else {
 784                 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
 785                                            sizeof(struct nvsp_message),
 786                                            req_id,
 787                                            VM_PKT_DATA_INBAND,
 788                                            VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
 789         }
 790
 791         if (ret == 0) {
 792                 atomic_inc_return(&nvchan->queue_sends);
 793
 794                 if (ring_avail < RING_AVAIL_PERCENT_LOWATER)
 795                         netif_tx_stop_queue(txq);
 796         } else if (ret == -EAGAIN) {
 797                 netif_tx_stop_queue(txq);
 798                 if (atomic_read(&nvchan->queue_sends) < 1) {
 799                         netif_tx_wake_queue(txq);
 800                         ret = -ENOSPC;
 801                 }
 802         } else {
 803                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
 804                            packet, ret);
 805         }
 806
 807         return ret;
 808 }
 809
 810 /* Move packet out of multi send data (msd), and clear msd */
 811 static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
 812                                 struct sk_buff **msd_skb,
 813                                 struct multi_send_data *msdp)
 814 {
 815         *msd_skb = msdp->skb;
 816         *msd_send = msdp->pkt;
 817         msdp->skb = NULL;
 818         msdp->pkt = NULL;
 819         msdp->count = 0;
 820 }
 821
 822 int netvsc_send(struct hv_device *device,
 823                 struct hv_netvsc_packet *packet,
 824                 struct rndis_message *rndis_msg,
 825                 struct hv_page_buffer **pb,
 826                 struct sk_buff *skb)
 827 {
 828         struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
 829         int ret = 0;
 830         struct netvsc_channel *nvchan;
 831         u32 pktlen = packet->total_data_buflen, msd_len = 0;
 832         unsigned int section_index = NETVSC_INVALID_INDEX;
 833         struct multi_send_data *msdp;
 834         struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
 835         struct sk_buff *msd_skb = NULL;
 836         bool try_batch;
 837         bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
 838
 839         /* If device is rescinded, return error and packet will get dropped. */
 840         if (unlikely(net_device->destroy))
 841                 return -ENODEV;
 842
 843         /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
 844          * here before the negotiation with the host is finished and
 845          * send_section_map may not be allocated yet.
 846          */
 847         if (unlikely(!net_device->send_section_map))
 848                 return -EAGAIN;
 849
 850         nvchan = &net_device->chan_table[packet->q_idx];
 851         packet->send_buf_index = NETVSC_INVALID_INDEX;
 852         packet->cp_partial = false;
 853
 854         /* Send control message directly without accessing msd (Multi-Send
 855          * Data) field which may be changed during data packet processing.
 856          */
 857         if (!skb) {
 858                 cur_send = packet;
 859                 goto send_now;
 860         }
 861
 862         /* batch packets in send buffer if possible */
 863         msdp = &nvchan->msd;
 864         if (msdp->pkt)
 865                 msd_len = msdp->pkt->total_data_buflen;
 866
 867         try_batch =  msd_len > 0 && msdp->count < net_device->max_pkt;
 868         if (try_batch && msd_len + pktlen + net_device->pkt_align <
 869             net_device->send_section_size) {
 870                 section_index = msdp->pkt->send_buf_index;
 871
 872         } else if (try_batch && msd_len + packet->rmsg_size <
 873                    net_device->send_section_size) {
 874                 section_index = msdp->pkt->send_buf_index;
 875                 packet->cp_partial = true;
 876
 877         } else if (pktlen + net_device->pkt_align <
 878                    net_device->send_section_size) {
 879                 section_index = netvsc_get_next_send_section(net_device);
 880                 if (section_index != NETVSC_INVALID_INDEX) {
 881                         move_pkt_msd(&msd_send, &msd_skb, msdp);
 882                         msd_len = 0;
 883                 }
 884         }
 885
 886         if (section_index != NETVSC_INVALID_INDEX) {
 887                 netvsc_copy_to_send_buf(net_device,
 888                                         section_index, msd_len,
 889                                         packet, rndis_msg, pb, skb);
 890
 891                 packet->send_buf_index = section_index;
 892
 893                 if (packet->cp_partial) {
 894                         packet->page_buf_cnt -= packet->rmsg_pgcnt;
 895                         packet->total_data_buflen = msd_len + packet->rmsg_size;
 896                 } else {
 897                         packet->page_buf_cnt = 0;
 898                         packet->total_data_buflen += msd_len;
 899                 }
 900
 901                 if (msdp->pkt) {
 902                         packet->total_packets += msdp->pkt->total_packets;
 903                         packet->total_bytes += msdp->pkt->total_bytes;
 904                 }
 905
 906                 if (msdp->skb)
 907                         dev_consume_skb_any(msdp->skb);
 908
 909                 if (xmit_more && !packet->cp_partial) {
 910                         msdp->skb = skb;
 911                         msdp->pkt = packet;
 912                         msdp->count++;
 913                 } else {
 914                         cur_send = packet;
 915                         msdp->skb = NULL;
 916                         msdp->pkt = NULL;
 917                         msdp->count = 0;
 918                 }
 919         } else {
 920                 move_pkt_msd(&msd_send, &msd_skb, msdp);
 921                 cur_send = packet;
 922         }
 923
 924         if (msd_send) {
 925                 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
 926                                             NULL, msd_skb);
 927
 928                 if (m_ret != 0) {
 929                         netvsc_free_send_slot(net_device,
 930                                               msd_send->send_buf_index);
 931                         dev_kfree_skb_any(msd_skb);
 932                 }
 933         }
 934
 935 send_now:
 936         if (cur_send)
 937                 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
 938
 939         if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
 940                 netvsc_free_send_slot(net_device, section_index);
 941
 942         return ret;
 943 }
 944
 945 static int netvsc_send_recv_completion(struct vmbus_channel *channel,
 946                                        u64 transaction_id, u32 status)
 947 {
 948         struct nvsp_message recvcompMessage;
 949         int ret;
 950
 951         recvcompMessage.hdr.msg_type =
 952                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
 953
 954         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
 955
 956         /* Send the completion */
 957         ret = vmbus_sendpacket(channel, &recvcompMessage,
 958                                sizeof(struct nvsp_message_header) + sizeof(u32),
 959                                transaction_id, VM_PKT_COMP, 0);
 960
 961         return ret;
 962 }
 963
 964 static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
 965                                         u32 *filled, u32 *avail)
 966 {
 967         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
 968         u32 first = mrc->first;
 969         u32 next = mrc->next;
 970
 971         *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
 972                   next - first;
 973
 974         *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
 975 }
 976
 977 /* Read the first filled slot, no change to index */
 978 static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
 979                                                          *nvdev, u16 q_idx)
 980 {
 981         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
 982         u32 filled, avail;
 983
 984         if (unlikely(!mrc->buf))
 985                 return NULL;
 986
 987         count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
 988         if (!filled)
 989                 return NULL;
 990
 991         return mrc->buf + mrc->first * sizeof(struct recv_comp_data);
 992 }
 993
 994 /* Put the first filled slot back to available pool */
 995 static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
 996 {
 997         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
 998         int num_recv;
 999
1000         mrc->first = (mrc->first + 1) % NETVSC_RECVSLOT_MAX;
1001
1002         num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1003
1004         if (nvdev->destroy && num_recv == 0)
1005                 wake_up(&nvdev->wait_drain);
1006 }
1007
1008 /* Check and send pending recv completions */
1009 static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1010                                  struct vmbus_channel *channel, u16 q_idx)
1011 {
1012         struct recv_comp_data *rcd;
1013         int ret;
1014
1015         while (true) {
1016                 rcd = read_recv_comp_slot(nvdev, q_idx);
1017                 if (!rcd)
1018                         break;
1019
1020                 ret = netvsc_send_recv_completion(channel, rcd->tid,
1021                                                   rcd->status);
1022                 if (ret)
1023                         break;
1024
1025                 put_recv_comp_slot(nvdev, q_idx);
1026         }
1027 }
1028
1029 #define NETVSC_RCD_WATERMARK 80
1030
1031 /* Get next available slot */
1032 static inline struct recv_comp_data *get_recv_comp_slot(
1033         struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1034 {
1035         struct multi_recv_comp *mrc = &nvdev->chan_table[q_idx].mrc;
1036         u32 filled, avail, next;
1037         struct recv_comp_data *rcd;
1038
1039         if (unlikely(!nvdev->recv_section))
1040                 return NULL;
1041
1042         if (unlikely(!mrc->buf))
1043                 return NULL;
1044
1045         if (atomic_read(&nvdev->num_outstanding_recvs) >
1046             nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1047                 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1048
1049         count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1050         if (!avail)
1051                 return NULL;
1052
1053         next = mrc->next;
1054         rcd = mrc->buf + next * sizeof(struct recv_comp_data);
1055         mrc->next = (next + 1) % NETVSC_RECVSLOT_MAX;
1056
1057         atomic_inc(&nvdev->num_outstanding_recvs);
1058
1059         return rcd;
1060 }
1061
1062 static int netvsc_receive(struct net_device *ndev,
1063                    struct netvsc_device *net_device,
1064                    struct net_device_context *net_device_ctx,
1065                    struct hv_device *device,
1066                    struct vmbus_channel *channel,
1067                    const struct vmpacket_descriptor *desc,
1068                    struct nvsp_message *nvsp)
1069 {
1070         const struct vmtransfer_page_packet_header *vmxferpage_packet
1071                 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1072         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1073         char *recv_buf = net_device->recv_buf;
1074         u32 status = NVSP_STAT_SUCCESS;
1075         int i;
1076         int count = 0;
1077         int ret;
1078
1079         /* Make sure this is a valid nvsp packet */
1080         if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1081                 netif_err(net_device_ctx, rx_err, ndev,
1082                           "Unknown nvsp packet type received %u\n",
1083                           nvsp->hdr.msg_type);
1084                 return 0;
1085         }
1086
1087         if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1088                 netif_err(net_device_ctx, rx_err, ndev,
1089                           "Invalid xfer page set id - expecting %x got %x\n",
1090                           NETVSC_RECEIVE_BUFFER_ID,
1091                           vmxferpage_packet->xfer_pageset_id);
1092                 return 0;
1093         }
1094
1095         count = vmxferpage_packet->range_cnt;
1096
1097         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1098         for (i = 0; i < count; i++) {
1099                 void *data = recv_buf
1100                         + vmxferpage_packet->ranges[i].byte_offset;
1101                 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1102
1103                 /* Pass it to the upper layer */
1104                 status = rndis_filter_receive(ndev, net_device, device,
1105                                               channel, data, buflen);
1106         }
1107
1108         if (net_device->chan_table[q_idx].mrc.buf) {
1109                 struct recv_comp_data *rcd;
1110
1111                 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1112                 if (rcd) {
1113                         rcd->tid = vmxferpage_packet->d.trans_id;
1114                         rcd->status = status;
1115                 } else {
1116                         netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1117                                    q_idx, vmxferpage_packet->d.trans_id);
1118                 }
1119         } else {
1120                 ret = netvsc_send_recv_completion(channel,
1121                                                   vmxferpage_packet->d.trans_id,
1122                                                   status);
1123                 if (ret)
1124                         netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1125                                    q_idx, vmxferpage_packet->d.trans_id, ret);
1126         }
1127         return count;
1128 }
1129
1130 static void netvsc_send_table(struct hv_device *hdev,
1131                               struct nvsp_message *nvmsg)
1132 {
1133         struct net_device *ndev = hv_get_drvdata(hdev);
1134         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1135         int i;
1136         u32 count, *tab;
1137
1138         count = nvmsg->msg.v5_msg.send_table.count;
1139         if (count != VRSS_SEND_TAB_SIZE) {
1140                 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1141                 return;
1142         }
1143
1144         tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1145                       nvmsg->msg.v5_msg.send_table.offset);
1146
1147         for (i = 0; i < count; i++)
1148                 net_device_ctx->tx_send_table[i] = tab[i];
1149 }
1150
1151 static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1152                            struct nvsp_message *nvmsg)
1153 {
1154         net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1155         net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1156 }
1157
1158 static inline void netvsc_receive_inband(struct hv_device *hdev,
1159                                  struct net_device_context *net_device_ctx,
1160                                  struct nvsp_message *nvmsg)
1161 {
1162         switch (nvmsg->hdr.msg_type) {
1163         case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1164                 netvsc_send_table(hdev, nvmsg);
1165                 break;
1166
1167         case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1168                 netvsc_send_vf(net_device_ctx, nvmsg);
1169                 break;
1170         }
1171 }
1172
1173 static int netvsc_process_raw_pkt(struct hv_device *device,
1174                                   struct vmbus_channel *channel,
1175                                   struct netvsc_device *net_device,
1176                                   struct net_device *ndev,
1177                                   const struct vmpacket_descriptor *desc,
1178                                   int budget)
1179 {
1180         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1181         struct nvsp_message *nvmsg = hv_pkt_data(desc);
1182
1183         switch (desc->type) {
1184         case VM_PKT_COMP:
1185                 netvsc_send_completion(net_device, channel, device,
1186                                        desc, budget);
1187                 break;
1188
1189         case VM_PKT_DATA_USING_XFER_PAGES:
1190                 return netvsc_receive(ndev, net_device, net_device_ctx,
1191                                       device, channel, desc, nvmsg);
1192                 break;
1193
1194         case VM_PKT_DATA_INBAND:
1195                 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1196                 break;
1197
1198         default:
1199                 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1200                            desc->type, desc->trans_id);
1201                 break;
1202         }
1203
1204         return 0;
1205 }
1206
1207 static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1208 {
1209         struct vmbus_channel *primary = channel->primary_channel;
1210
1211         return primary ? primary->device_obj : channel->device_obj;
1212 }
1213
1214 /* Network processing softirq
1215  * Process data in incoming ring buffer from host
1216  * Stops when ring is empty or budget is met or exceeded.
1217  */
1218 int netvsc_poll(struct napi_struct *napi, int budget)
1219 {
1220         struct netvsc_channel *nvchan
1221                 = container_of(napi, struct netvsc_channel, napi);
1222         struct vmbus_channel *channel = nvchan->channel;
1223         struct hv_device *device = netvsc_channel_to_device(channel);
1224         u16 q_idx = channel->offermsg.offer.sub_channel_index;
1225         struct net_device *ndev = hv_get_drvdata(device);
1226         struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
1227         int work_done = 0;
1228
1229         /* If starting a new interval */
1230         if (!nvchan->desc)
1231                 nvchan->desc = hv_pkt_iter_first(channel);
1232
1233         while (nvchan->desc && work_done < budget) {
1234                 work_done += netvsc_process_raw_pkt(device, channel, net_device,
1235                                                     ndev, nvchan->desc, budget);
1236                 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1237         }
1238
1239         /* If receive ring was exhausted
1240          * and not doing busy poll
1241          * then re-enable host interrupts
1242          *  and reschedule if ring is not empty.
1243          */
1244         if (work_done < budget &&
1245             napi_complete_done(napi, work_done) &&
1246             hv_end_read(&channel->inbound) != 0)
1247                 napi_reschedule(napi);
1248
1249         netvsc_chk_recv_comp(net_device, channel, q_idx);
1250
1251         /* Driver may overshoot since multiple packets per descriptor */
1252         return min(work_done, budget);
1253 }
1254
1255 /* Call back when data is available in host ring buffer.
1256  * Processing is deferred until network softirq (NAPI)
1257  */
1258 void netvsc_channel_cb(void *context)
1259 {
1260         struct netvsc_channel *nvchan = context;
1261
1262         if (napi_schedule_prep(&nvchan->napi)) {
1263                 /* disable interupts from host */
1264                 hv_begin_read(&nvchan->channel->inbound);
1265
1266                 __napi_schedule(&nvchan->napi);
1267         }
1268 }
1269
1270 /*
1271  * netvsc_device_add - Callback when the device belonging to this
1272  * driver is added
1273  */
1274 int netvsc_device_add(struct hv_device *device,
1275                       const struct netvsc_device_info *device_info)
1276 {
1277         int i, ret = 0;
1278         int ring_size = device_info->ring_size;
1279         struct netvsc_device *net_device;
1280         struct net_device *ndev = hv_get_drvdata(device);
1281         struct net_device_context *net_device_ctx = netdev_priv(ndev);
1282
1283         net_device = alloc_net_device();
1284         if (!net_device)
1285                 return -ENOMEM;
1286
1287         net_device->ring_size = ring_size;
1288
1289         /* Because the device uses NAPI, all the interrupt batching and
1290          * control is done via Net softirq, not the channel handling
1291          */
1292         set_channel_read_mode(device->channel, HV_CALL_ISR);
1293
1294         /* If we're reopening the device we may have multiple queues, fill the
1295          * chn_table with the default channel to use it before subchannels are
1296          * opened.
1297          * Initialize the channel state before we open;
1298          * we can be interrupted as soon as we open the channel.
1299          */
1300
1301         for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1302                 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1303
1304                 nvchan->channel = device->channel;
1305                 u64_stats_init(&nvchan->tx_stats.syncp);
1306                 u64_stats_init(&nvchan->rx_stats.syncp);
1307         }
1308
1309         /* Enable NAPI handler before init callbacks */
1310         netif_napi_add(ndev, &net_device->chan_table[0].napi,
1311                        netvsc_poll, NAPI_POLL_WEIGHT);
1312
1313         /* Open the channel */
1314         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1315                          ring_size * PAGE_SIZE, NULL, 0,
1316                          netvsc_channel_cb,
1317                          net_device->chan_table);
1318
1319         if (ret != 0) {
1320                 netif_napi_del(&net_device->chan_table[0].napi);
1321                 netdev_err(ndev, "unable to open channel: %d\n", ret);
1322                 goto cleanup;
1323         }
1324
1325         /* Channel is opened */
1326         netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1327
1328         napi_enable(&net_device->chan_table[0].napi);
1329
1330         /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1331          * populated.
1332          */
1333         rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1334
1335         /* Connect with the NetVsp */
1336         ret = netvsc_connect_vsp(device, net_device);
1337         if (ret != 0) {
1338                 netdev_err(ndev,
1339                         "unable to connect to NetVSP - %d\n", ret);
1340                 goto close;
1341         }
1342
1343         return ret;
1344
1345 close:
1346         netif_napi_del(&net_device->chan_table[0].napi);
1347
1348         /* Now, we can close the channel safely */
1349         vmbus_close(device->channel);
1350
1351 cleanup:
1352         free_netvsc_device(&net_device->rcu);
1353
1354         return ret;
1355
1356 }