drivers/hv/channel_mgmt.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, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/kernel.h>
  24 #include <linux/sched.h>
  25 #include <linux/wait.h>
  26 #include <linux/mm.h>
  27 #include <linux/slab.h>
  28 #include <linux/list.h>
  29 #include <linux/module.h>
  30 #include <linux/completion.h>
  31 #include <linux/delay.h>
  32 #include <linux/hyperv.h>
  33
  34 #include "hyperv_vmbus.h"
  35
  36 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
  37
  38 static const struct vmbus_device vmbus_devs[] = {
  39         /* IDE */
  40         { .dev_type = HV_IDE,
  41           HV_IDE_GUID,
  42           .perf_device = true,
  43         },
  44
  45         /* SCSI */
  46         { .dev_type = HV_SCSI,
  47           HV_SCSI_GUID,
  48           .perf_device = true,
  49         },
  50
  51         /* Fibre Channel */
  52         { .dev_type = HV_FC,
  53           HV_SYNTHFC_GUID,
  54           .perf_device = true,
  55         },
  56
  57         /* Synthetic NIC */
  58         { .dev_type = HV_NIC,
  59           HV_NIC_GUID,
  60           .perf_device = true,
  61         },
  62
  63         /* Network Direct */
  64         { .dev_type = HV_ND,
  65           HV_ND_GUID,
  66           .perf_device = true,
  67         },
  68
  69         /* PCIE */
  70         { .dev_type = HV_PCIE,
  71           HV_PCIE_GUID,
  72           .perf_device = true,
  73         },
  74
  75         /* Synthetic Frame Buffer */
  76         { .dev_type = HV_FB,
  77           HV_SYNTHVID_GUID,
  78           .perf_device = false,
  79         },
  80
  81         /* Synthetic Keyboard */
  82         { .dev_type = HV_KBD,
  83           HV_KBD_GUID,
  84           .perf_device = false,
  85         },
  86
  87         /* Synthetic MOUSE */
  88         { .dev_type = HV_MOUSE,
  89           HV_MOUSE_GUID,
  90           .perf_device = false,
  91         },
  92
  93         /* KVP */
  94         { .dev_type = HV_KVP,
  95           HV_KVP_GUID,
  96           .perf_device = false,
  97         },
  98
  99         /* Time Synch */
 100         { .dev_type = HV_TS,
 101           HV_TS_GUID,
 102           .perf_device = false,
 103         },
 104
 105         /* Heartbeat */
 106         { .dev_type = HV_HB,
 107           HV_HEART_BEAT_GUID,
 108           .perf_device = false,
 109         },
 110
 111         /* Shutdown */
 112         { .dev_type = HV_SHUTDOWN,
 113           HV_SHUTDOWN_GUID,
 114           .perf_device = false,
 115         },
 116
 117         /* File copy */
 118         { .dev_type = HV_FCOPY,
 119           HV_FCOPY_GUID,
 120           .perf_device = false,
 121         },
 122
 123         /* Backup */
 124         { .dev_type = HV_BACKUP,
 125           HV_VSS_GUID,
 126           .perf_device = false,
 127         },
 128
 129         /* Dynamic Memory */
 130         { .dev_type = HV_DM,
 131           HV_DM_GUID,
 132           .perf_device = false,
 133         },
 134
 135         /* Unknown GUID */
 136         { .dev_type = HV_UNKOWN,
 137           .perf_device = false,
 138         },
 139 };
 140
 141 static u16 hv_get_dev_type(const uuid_le *guid)
 142 {
 143         u16 i;
 144
 145         for (i = HV_IDE; i < HV_UNKOWN; i++) {
 146                 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid))
 147                         return i;
 148         }
 149         pr_info("Unknown GUID: %pUl\n", guid);
 150         return i;
 151 }
 152
 153 /**
 154  * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
 155  * @icmsghdrp: Pointer to msg header structure
 156  * @icmsg_negotiate: Pointer to negotiate message structure
 157  * @buf: Raw buffer channel data
 158  *
 159  * @icmsghdrp is of type &struct icmsg_hdr.
 160  * @negop is of type &struct icmsg_negotiate.
 161  * Set up and fill in default negotiate response message.
 162  *
 163  * The fw_version specifies the  framework version that
 164  * we can support and srv_version specifies the service
 165  * version we can support.
 166  *
 167  * Mainly used by Hyper-V drivers.
 168  */
 169 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
 170                                 struct icmsg_negotiate *negop, u8 *buf,
 171                                 int fw_version, int srv_version)
 172 {
 173         int icframe_major, icframe_minor;
 174         int icmsg_major, icmsg_minor;
 175         int fw_major, fw_minor;
 176         int srv_major, srv_minor;
 177         int i;
 178         bool found_match = false;
 179
 180         icmsghdrp->icmsgsize = 0x10;
 181         fw_major = (fw_version >> 16);
 182         fw_minor = (fw_version & 0xFFFF);
 183
 184         srv_major = (srv_version >> 16);
 185         srv_minor = (srv_version & 0xFFFF);
 186
 187         negop = (struct icmsg_negotiate *)&buf[
 188                 sizeof(struct vmbuspipe_hdr) +
 189                 sizeof(struct icmsg_hdr)];
 190
 191         icframe_major = negop->icframe_vercnt;
 192         icframe_minor = 0;
 193
 194         icmsg_major = negop->icmsg_vercnt;
 195         icmsg_minor = 0;
 196
 197         /*
 198          * Select the framework version number we will
 199          * support.
 200          */
 201
 202         for (i = 0; i < negop->icframe_vercnt; i++) {
 203                 if ((negop->icversion_data[i].major == fw_major) &&
 204                    (negop->icversion_data[i].minor == fw_minor)) {
 205                         icframe_major = negop->icversion_data[i].major;
 206                         icframe_minor = negop->icversion_data[i].minor;
 207                         found_match = true;
 208                 }
 209         }
 210
 211         if (!found_match)
 212                 goto fw_error;
 213
 214         found_match = false;
 215
 216         for (i = negop->icframe_vercnt;
 217                  (i < negop->icframe_vercnt + negop->icmsg_vercnt); i++) {
 218                 if ((negop->icversion_data[i].major == srv_major) &&
 219                    (negop->icversion_data[i].minor == srv_minor)) {
 220                         icmsg_major = negop->icversion_data[i].major;
 221                         icmsg_minor = negop->icversion_data[i].minor;
 222                         found_match = true;
 223                 }
 224         }
 225
 226         /*
 227          * Respond with the framework and service
 228          * version numbers we can support.
 229          */
 230
 231 fw_error:
 232         if (!found_match) {
 233                 negop->icframe_vercnt = 0;
 234                 negop->icmsg_vercnt = 0;
 235         } else {
 236                 negop->icframe_vercnt = 1;
 237                 negop->icmsg_vercnt = 1;
 238         }
 239
 240         negop->icversion_data[0].major = icframe_major;
 241         negop->icversion_data[0].minor = icframe_minor;
 242         negop->icversion_data[1].major = icmsg_major;
 243         negop->icversion_data[1].minor = icmsg_minor;
 244         return found_match;
 245 }
 246
 247 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
 248
 249 /*
 250  * alloc_channel - Allocate and initialize a vmbus channel object
 251  */
 252 static struct vmbus_channel *alloc_channel(void)
 253 {
 254         static atomic_t chan_num = ATOMIC_INIT(0);
 255         struct vmbus_channel *channel;
 256
 257         channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
 258         if (!channel)
 259                 return NULL;
 260
 261         channel->id = atomic_inc_return(&chan_num);
 262         channel->acquire_ring_lock = true;
 263         spin_lock_init(&channel->inbound_lock);
 264         spin_lock_init(&channel->lock);
 265
 266         INIT_LIST_HEAD(&channel->sc_list);
 267         INIT_LIST_HEAD(&channel->percpu_list);
 268
 269         return channel;
 270 }
 271
 272 /*
 273  * free_channel - Release the resources used by the vmbus channel object
 274  */
 275 static void free_channel(struct vmbus_channel *channel)
 276 {
 277         kfree(channel);
 278 }
 279
 280 static void percpu_channel_enq(void *arg)
 281 {
 282         struct vmbus_channel *channel = arg;
 283         int cpu = smp_processor_id();
 284
 285         list_add_tail(&channel->percpu_list, &hv_context.percpu_list[cpu]);
 286 }
 287
 288 static void percpu_channel_deq(void *arg)
 289 {
 290         struct vmbus_channel *channel = arg;
 291
 292         list_del(&channel->percpu_list);
 293 }
 294
 295
 296 static void vmbus_release_relid(u32 relid)
 297 {
 298         struct vmbus_channel_relid_released msg;
 299
 300         memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
 301         msg.child_relid = relid;
 302         msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
 303         vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released));
 304 }
 305
 306 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
 307 {
 308         unsigned long flags;
 309         struct vmbus_channel *primary_channel;
 310
 311         vmbus_release_relid(relid);
 312
 313         BUG_ON(!channel->rescind);
 314         BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
 315
 316         if (channel->target_cpu != get_cpu()) {
 317                 put_cpu();
 318                 smp_call_function_single(channel->target_cpu,
 319                                          percpu_channel_deq, channel, true);
 320         } else {
 321                 percpu_channel_deq(channel);
 322                 put_cpu();
 323         }
 324
 325         if (channel->primary_channel == NULL) {
 326                 list_del(&channel->listentry);
 327
 328                 primary_channel = channel;
 329         } else {
 330                 primary_channel = channel->primary_channel;
 331                 spin_lock_irqsave(&primary_channel->lock, flags);
 332                 list_del(&channel->sc_list);
 333                 primary_channel->num_sc--;
 334                 spin_unlock_irqrestore(&primary_channel->lock, flags);
 335         }
 336
 337         /*
 338          * We need to free the bit for init_vp_index() to work in the case
 339          * of sub-channel, when we reload drivers like hv_netvsc.
 340          */
 341         cpumask_clear_cpu(channel->target_cpu,
 342                           &primary_channel->alloced_cpus_in_node);
 343
 344         free_channel(channel);
 345 }
 346
 347 void vmbus_free_channels(void)
 348 {
 349         struct vmbus_channel *channel, *tmp;
 350
 351         list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
 352                 listentry) {
 353                 /* hv_process_channel_removal() needs this */
 354                 channel->rescind = true;
 355
 356                 vmbus_device_unregister(channel->device_obj);
 357         }
 358 }
 359
 360 /*
 361  * vmbus_process_offer - Process the offer by creating a channel/device
 362  * associated with this offer
 363  */
 364 static void vmbus_process_offer(struct vmbus_channel *newchannel)
 365 {
 366         struct vmbus_channel *channel;
 367         bool fnew = true;
 368         unsigned long flags;
 369         u16 dev_type;
 370         int ret;
 371
 372         /* Make sure this is a new offer */
 373         mutex_lock(&vmbus_connection.channel_mutex);
 374
 375         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 376                 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
 377                         newchannel->offermsg.offer.if_type) &&
 378                         !uuid_le_cmp(channel->offermsg.offer.if_instance,
 379                                 newchannel->offermsg.offer.if_instance)) {
 380                         fnew = false;
 381                         break;
 382                 }
 383         }
 384
 385         if (fnew)
 386                 list_add_tail(&newchannel->listentry,
 387                               &vmbus_connection.chn_list);
 388
 389         mutex_unlock(&vmbus_connection.channel_mutex);
 390
 391         if (!fnew) {
 392                 /*
 393                  * Check to see if this is a sub-channel.
 394                  */
 395                 if (newchannel->offermsg.offer.sub_channel_index != 0) {
 396                         /*
 397                          * Process the sub-channel.
 398                          */
 399                         newchannel->primary_channel = channel;
 400                         spin_lock_irqsave(&channel->lock, flags);
 401                         list_add_tail(&newchannel->sc_list, &channel->sc_list);
 402                         channel->num_sc++;
 403                         spin_unlock_irqrestore(&channel->lock, flags);
 404                 } else
 405                         goto err_free_chan;
 406         }
 407
 408         dev_type = hv_get_dev_type(&newchannel->offermsg.offer.if_type);
 409
 410         init_vp_index(newchannel, dev_type);
 411
 412         if (newchannel->target_cpu != get_cpu()) {
 413                 put_cpu();
 414                 smp_call_function_single(newchannel->target_cpu,
 415                                          percpu_channel_enq,
 416                                          newchannel, true);
 417         } else {
 418                 percpu_channel_enq(newchannel);
 419                 put_cpu();
 420         }
 421
 422         /*
 423          * This state is used to indicate a successful open
 424          * so that when we do close the channel normally, we
 425          * can cleanup properly
 426          */
 427         newchannel->state = CHANNEL_OPEN_STATE;
 428
 429         if (!fnew) {
 430                 if (channel->sc_creation_callback != NULL)
 431                         channel->sc_creation_callback(newchannel);
 432                 return;
 433         }
 434
 435         /*
 436          * Start the process of binding this offer to the driver
 437          * We need to set the DeviceObject field before calling
 438          * vmbus_child_dev_add()
 439          */
 440         newchannel->device_obj = vmbus_device_create(
 441                 &newchannel->offermsg.offer.if_type,
 442                 &newchannel->offermsg.offer.if_instance,
 443                 newchannel);
 444         if (!newchannel->device_obj)
 445                 goto err_deq_chan;
 446
 447         newchannel->device_obj->device_id = dev_type;
 448         /*
 449          * Add the new device to the bus. This will kick off device-driver
 450          * binding which eventually invokes the device driver's AddDevice()
 451          * method.
 452          */
 453         mutex_lock(&vmbus_connection.channel_mutex);
 454         ret = vmbus_device_register(newchannel->device_obj);
 455         mutex_unlock(&vmbus_connection.channel_mutex);
 456
 457         if (ret != 0) {
 458                 pr_err("unable to add child device object (relid %d)\n",
 459                         newchannel->offermsg.child_relid);
 460                 kfree(newchannel->device_obj);
 461                 goto err_deq_chan;
 462         }
 463         return;
 464
 465 err_deq_chan:
 466         vmbus_release_relid(newchannel->offermsg.child_relid);
 467
 468         mutex_lock(&vmbus_connection.channel_mutex);
 469         list_del(&newchannel->listentry);
 470         mutex_unlock(&vmbus_connection.channel_mutex);
 471
 472         if (newchannel->target_cpu != get_cpu()) {
 473                 put_cpu();
 474                 smp_call_function_single(newchannel->target_cpu,
 475                                          percpu_channel_deq, newchannel, true);
 476         } else {
 477                 percpu_channel_deq(newchannel);
 478                 put_cpu();
 479         }
 480
 481 err_free_chan:
 482         free_channel(newchannel);
 483 }
 484
 485 /*
 486  * We use this state to statically distribute the channel interrupt load.
 487  */
 488 static int next_numa_node_id;
 489
 490 /*
 491  * Starting with Win8, we can statically distribute the incoming
 492  * channel interrupt load by binding a channel to VCPU.
 493  * We do this in a hierarchical fashion:
 494  * First distribute the primary channels across available NUMA nodes
 495  * and then distribute the subchannels amongst the CPUs in the NUMA
 496  * node assigned to the primary channel.
 497  *
 498  * For pre-win8 hosts or non-performance critical channels we assign the
 499  * first CPU in the first NUMA node.
 500  */
 501 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
 502 {
 503         u32 cur_cpu;
 504         bool perf_chn = vmbus_devs[dev_type].perf_device;
 505         struct vmbus_channel *primary = channel->primary_channel;
 506         int next_node;
 507         struct cpumask available_mask;
 508         struct cpumask *alloced_mask;
 509
 510         if ((vmbus_proto_version == VERSION_WS2008) ||
 511             (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
 512                 /*
 513                  * Prior to win8, all channel interrupts are
 514                  * delivered on cpu 0.
 515                  * Also if the channel is not a performance critical
 516                  * channel, bind it to cpu 0.
 517                  */
 518                 channel->numa_node = 0;
 519                 channel->target_cpu = 0;
 520                 channel->target_vp = hv_context.vp_index[0];
 521                 return;
 522         }
 523
 524         /*
 525          * We distribute primary channels evenly across all the available
 526          * NUMA nodes and within the assigned NUMA node we will assign the
 527          * first available CPU to the primary channel.
 528          * The sub-channels will be assigned to the CPUs available in the
 529          * NUMA node evenly.
 530          */
 531         if (!primary) {
 532                 while (true) {
 533                         next_node = next_numa_node_id++;
 534                         if (next_node == nr_node_ids)
 535                                 next_node = next_numa_node_id = 0;
 536                         if (cpumask_empty(cpumask_of_node(next_node)))
 537                                 continue;
 538                         break;
 539                 }
 540                 channel->numa_node = next_node;
 541                 primary = channel;
 542         }
 543         alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
 544
 545         if (cpumask_weight(alloced_mask) ==
 546             cpumask_weight(cpumask_of_node(primary->numa_node))) {
 547                 /*
 548                  * We have cycled through all the CPUs in the node;
 549                  * reset the alloced map.
 550                  */
 551                 cpumask_clear(alloced_mask);
 552         }
 553
 554         cpumask_xor(&available_mask, alloced_mask,
 555                     cpumask_of_node(primary->numa_node));
 556
 557         cur_cpu = -1;
 558
 559         /*
 560          * Normally Hyper-V host doesn't create more subchannels than there
 561          * are VCPUs on the node but it is possible when not all present VCPUs
 562          * on the node are initialized by guest. Clear the alloced_cpus_in_node
 563          * to start over.
 564          */
 565         if (cpumask_equal(&primary->alloced_cpus_in_node,
 566                           cpumask_of_node(primary->numa_node)))
 567                 cpumask_clear(&primary->alloced_cpus_in_node);
 568
 569         while (true) {
 570                 cur_cpu = cpumask_next(cur_cpu, &available_mask);
 571                 if (cur_cpu >= nr_cpu_ids) {
 572                         cur_cpu = -1;
 573                         cpumask_copy(&available_mask,
 574                                      cpumask_of_node(primary->numa_node));
 575                         continue;
 576                 }
 577
 578                 /*
 579                  * NOTE: in the case of sub-channel, we clear the sub-channel
 580                  * related bit(s) in primary->alloced_cpus_in_node in
 581                  * hv_process_channel_removal(), so when we reload drivers
 582                  * like hv_netvsc in SMP guest, here we're able to re-allocate
 583                  * bit from primary->alloced_cpus_in_node.
 584                  */
 585                 if (!cpumask_test_cpu(cur_cpu,
 586                                 &primary->alloced_cpus_in_node)) {
 587                         cpumask_set_cpu(cur_cpu,
 588                                         &primary->alloced_cpus_in_node);
 589                         cpumask_set_cpu(cur_cpu, alloced_mask);
 590                         break;
 591                 }
 592         }
 593
 594         channel->target_cpu = cur_cpu;
 595         channel->target_vp = hv_context.vp_index[cur_cpu];
 596 }
 597
 598 static void vmbus_wait_for_unload(void)
 599 {
 600         int cpu = smp_processor_id();
 601         void *page_addr = hv_context.synic_message_page[cpu];
 602         struct hv_message *msg = (struct hv_message *)page_addr +
 603                                   VMBUS_MESSAGE_SINT;
 604         struct vmbus_channel_message_header *hdr;
 605         bool unloaded = false;
 606
 607         while (1) {
 608                 if (msg->header.message_type == HVMSG_NONE) {
 609                         mdelay(10);
 610                         continue;
 611                 }
 612
 613                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 614                 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
 615                         unloaded = true;
 616
 617                 msg->header.message_type = HVMSG_NONE;
 618                 /*
 619                  * header.message_type needs to be written before we do
 620                  * wrmsrl() below.
 621                  */
 622                 mb();
 623
 624                 if (msg->header.message_flags.msg_pending)
 625                         wrmsrl(HV_X64_MSR_EOM, 0);
 626
 627                 if (unloaded)
 628                         break;
 629         }
 630 }
 631
 632 /*
 633  * vmbus_unload_response - Handler for the unload response.
 634  */
 635 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
 636 {
 637         /*
 638          * This is a global event; just wakeup the waiting thread.
 639          * Once we successfully unload, we can cleanup the monitor state.
 640          */
 641         complete(&vmbus_connection.unload_event);
 642 }
 643
 644 void vmbus_initiate_unload(bool crash)
 645 {
 646         struct vmbus_channel_message_header hdr;
 647
 648         /* Pre-Win2012R2 hosts don't support reconnect */
 649         if (vmbus_proto_version < VERSION_WIN8_1)
 650                 return;
 651
 652         init_completion(&vmbus_connection.unload_event);
 653         memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 654         hdr.msgtype = CHANNELMSG_UNLOAD;
 655         vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header));
 656
 657         /*
 658          * vmbus_initiate_unload() is also called on crash and the crash can be
 659          * happening in an interrupt context, where scheduling is impossible.
 660          */
 661         if (!crash)
 662                 wait_for_completion(&vmbus_connection.unload_event);
 663         else
 664                 vmbus_wait_for_unload();
 665 }
 666
 667 /*
 668  * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
 669  *
 670  */
 671 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
 672 {
 673         struct vmbus_channel_offer_channel *offer;
 674         struct vmbus_channel *newchannel;
 675
 676         offer = (struct vmbus_channel_offer_channel *)hdr;
 677
 678         /* Allocate the channel object and save this offer. */
 679         newchannel = alloc_channel();
 680         if (!newchannel) {
 681                 pr_err("Unable to allocate channel object\n");
 682                 return;
 683         }
 684
 685         /*
 686          * By default we setup state to enable batched
 687          * reading. A specific service can choose to
 688          * disable this prior to opening the channel.
 689          */
 690         newchannel->batched_reading = true;
 691
 692         /*
 693          * Setup state for signalling the host.
 694          */
 695         newchannel->sig_event = (struct hv_input_signal_event *)
 696                                 (ALIGN((unsigned long)
 697                                 &newchannel->sig_buf,
 698                                 HV_HYPERCALL_PARAM_ALIGN));
 699
 700         newchannel->sig_event->connectionid.asu32 = 0;
 701         newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
 702         newchannel->sig_event->flag_number = 0;
 703         newchannel->sig_event->rsvdz = 0;
 704
 705         if (vmbus_proto_version != VERSION_WS2008) {
 706                 newchannel->is_dedicated_interrupt =
 707                                 (offer->is_dedicated_interrupt != 0);
 708                 newchannel->sig_event->connectionid.u.id =
 709                                 offer->connection_id;
 710         }
 711
 712         memcpy(&newchannel->offermsg, offer,
 713                sizeof(struct vmbus_channel_offer_channel));
 714         newchannel->monitor_grp = (u8)offer->monitorid / 32;
 715         newchannel->monitor_bit = (u8)offer->monitorid % 32;
 716
 717         vmbus_process_offer(newchannel);
 718 }
 719
 720 /*
 721  * vmbus_onoffer_rescind - Rescind offer handler.
 722  *
 723  * We queue a work item to process this offer synchronously
 724  */
 725 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 726 {
 727         struct vmbus_channel_rescind_offer *rescind;
 728         struct vmbus_channel *channel;
 729         unsigned long flags;
 730         struct device *dev;
 731
 732         rescind = (struct vmbus_channel_rescind_offer *)hdr;
 733
 734         mutex_lock(&vmbus_connection.channel_mutex);
 735         channel = relid2channel(rescind->child_relid);
 736
 737         if (channel == NULL) {
 738                 /*
 739                  * This is very impossible, because in
 740                  * vmbus_process_offer(), we have already invoked
 741                  * vmbus_release_relid() on error.
 742                  */
 743                 goto out;
 744         }
 745
 746         spin_lock_irqsave(&channel->lock, flags);
 747         channel->rescind = true;
 748         spin_unlock_irqrestore(&channel->lock, flags);
 749
 750         if (channel->device_obj) {
 751                 if (channel->chn_rescind_callback) {
 752                         channel->chn_rescind_callback(channel);
 753                         goto out;
 754                 }
 755                 /*
 756                  * We will have to unregister this device from the
 757                  * driver core.
 758                  */
 759                 dev = get_device(&channel->device_obj->device);
 760                 if (dev) {
 761                         vmbus_device_unregister(channel->device_obj);
 762                         put_device(dev);
 763                 }
 764         } else {
 765                 hv_process_channel_removal(channel,
 766                         channel->offermsg.child_relid);
 767         }
 768
 769 out:
 770         mutex_unlock(&vmbus_connection.channel_mutex);
 771 }
 772
 773 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
 774 {
 775         mutex_lock(&vmbus_connection.channel_mutex);
 776
 777         BUG_ON(!is_hvsock_channel(channel));
 778
 779         channel->rescind = true;
 780         vmbus_device_unregister(channel->device_obj);
 781
 782         mutex_unlock(&vmbus_connection.channel_mutex);
 783 }
 784 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
 785
 786
 787 /*
 788  * vmbus_onoffers_delivered -
 789  * This is invoked when all offers have been delivered.
 790  *
 791  * Nothing to do here.
 792  */
 793 static void vmbus_onoffers_delivered(
 794                         struct vmbus_channel_message_header *hdr)
 795 {
 796 }
 797
 798 /*
 799  * vmbus_onopen_result - Open result handler.
 800  *
 801  * This is invoked when we received a response to our channel open request.
 802  * Find the matching request, copy the response and signal the requesting
 803  * thread.
 804  */
 805 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
 806 {
 807         struct vmbus_channel_open_result *result;
 808         struct vmbus_channel_msginfo *msginfo;
 809         struct vmbus_channel_message_header *requestheader;
 810         struct vmbus_channel_open_channel *openmsg;
 811         unsigned long flags;
 812
 813         result = (struct vmbus_channel_open_result *)hdr;
 814
 815         /*
 816          * Find the open msg, copy the result and signal/unblock the wait event
 817          */
 818         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 819
 820         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 821                                 msglistentry) {
 822                 requestheader =
 823                         (struct vmbus_channel_message_header *)msginfo->msg;
 824
 825                 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
 826                         openmsg =
 827                         (struct vmbus_channel_open_channel *)msginfo->msg;
 828                         if (openmsg->child_relid == result->child_relid &&
 829                             openmsg->openid == result->openid) {
 830                                 memcpy(&msginfo->response.open_result,
 831                                        result,
 832                                        sizeof(
 833                                         struct vmbus_channel_open_result));
 834                                 complete(&msginfo->waitevent);
 835                                 break;
 836                         }
 837                 }
 838         }
 839         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 840 }
 841
 842 /*
 843  * vmbus_ongpadl_created - GPADL created handler.
 844  *
 845  * This is invoked when we received a response to our gpadl create request.
 846  * Find the matching request, copy the response and signal the requesting
 847  * thread.
 848  */
 849 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
 850 {
 851         struct vmbus_channel_gpadl_created *gpadlcreated;
 852         struct vmbus_channel_msginfo *msginfo;
 853         struct vmbus_channel_message_header *requestheader;
 854         struct vmbus_channel_gpadl_header *gpadlheader;
 855         unsigned long flags;
 856
 857         gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
 858
 859         /*
 860          * Find the establish msg, copy the result and signal/unblock the wait
 861          * event
 862          */
 863         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 864
 865         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 866                                 msglistentry) {
 867                 requestheader =
 868                         (struct vmbus_channel_message_header *)msginfo->msg;
 869
 870                 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
 871                         gpadlheader =
 872                         (struct vmbus_channel_gpadl_header *)requestheader;
 873
 874                         if ((gpadlcreated->child_relid ==
 875                              gpadlheader->child_relid) &&
 876                             (gpadlcreated->gpadl == gpadlheader->gpadl)) {
 877                                 memcpy(&msginfo->response.gpadl_created,
 878                                        gpadlcreated,
 879                                        sizeof(
 880                                         struct vmbus_channel_gpadl_created));
 881                                 complete(&msginfo->waitevent);
 882                                 break;
 883                         }
 884                 }
 885         }
 886         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 887 }
 888
 889 /*
 890  * vmbus_ongpadl_torndown - GPADL torndown handler.
 891  *
 892  * This is invoked when we received a response to our gpadl teardown request.
 893  * Find the matching request, copy the response and signal the requesting
 894  * thread.
 895  */
 896 static void vmbus_ongpadl_torndown(
 897                         struct vmbus_channel_message_header *hdr)
 898 {
 899         struct vmbus_channel_gpadl_torndown *gpadl_torndown;
 900         struct vmbus_channel_msginfo *msginfo;
 901         struct vmbus_channel_message_header *requestheader;
 902         struct vmbus_channel_gpadl_teardown *gpadl_teardown;
 903         unsigned long flags;
 904
 905         gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
 906
 907         /*
 908          * Find the open msg, copy the result and signal/unblock the wait event
 909          */
 910         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 911
 912         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 913                                 msglistentry) {
 914                 requestheader =
 915                         (struct vmbus_channel_message_header *)msginfo->msg;
 916
 917                 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
 918                         gpadl_teardown =
 919                         (struct vmbus_channel_gpadl_teardown *)requestheader;
 920
 921                         if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
 922                                 memcpy(&msginfo->response.gpadl_torndown,
 923                                        gpadl_torndown,
 924                                        sizeof(
 925                                         struct vmbus_channel_gpadl_torndown));
 926                                 complete(&msginfo->waitevent);
 927                                 break;
 928                         }
 929                 }
 930         }
 931         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 932 }
 933
 934 /*
 935  * vmbus_onversion_response - Version response handler
 936  *
 937  * This is invoked when we received a response to our initiate contact request.
 938  * Find the matching request, copy the response and signal the requesting
 939  * thread.
 940  */
 941 static void vmbus_onversion_response(
 942                 struct vmbus_channel_message_header *hdr)
 943 {
 944         struct vmbus_channel_msginfo *msginfo;
 945         struct vmbus_channel_message_header *requestheader;
 946         struct vmbus_channel_version_response *version_response;
 947         unsigned long flags;
 948
 949         version_response = (struct vmbus_channel_version_response *)hdr;
 950         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 951
 952         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 953                                 msglistentry) {
 954                 requestheader =
 955                         (struct vmbus_channel_message_header *)msginfo->msg;
 956
 957                 if (requestheader->msgtype ==
 958                     CHANNELMSG_INITIATE_CONTACT) {
 959                         memcpy(&msginfo->response.version_response,
 960                               version_response,
 961                               sizeof(struct vmbus_channel_version_response));
 962                         complete(&msginfo->waitevent);
 963                 }
 964         }
 965         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 966 }
 967
 968 /* Channel message dispatch table */
 969 struct vmbus_channel_message_table_entry
 970         channel_message_table[CHANNELMSG_COUNT] = {
 971         {CHANNELMSG_INVALID,                    0, NULL},
 972         {CHANNELMSG_OFFERCHANNEL,               0, vmbus_onoffer},
 973         {CHANNELMSG_RESCIND_CHANNELOFFER,       0, vmbus_onoffer_rescind},
 974         {CHANNELMSG_REQUESTOFFERS,              0, NULL},
 975         {CHANNELMSG_ALLOFFERS_DELIVERED,        1, vmbus_onoffers_delivered},
 976         {CHANNELMSG_OPENCHANNEL,                0, NULL},
 977         {CHANNELMSG_OPENCHANNEL_RESULT,         1, vmbus_onopen_result},
 978         {CHANNELMSG_CLOSECHANNEL,               0, NULL},
 979         {CHANNELMSG_GPADL_HEADER,               0, NULL},
 980         {CHANNELMSG_GPADL_BODY,                 0, NULL},
 981         {CHANNELMSG_GPADL_CREATED,              1, vmbus_ongpadl_created},
 982         {CHANNELMSG_GPADL_TEARDOWN,             0, NULL},
 983         {CHANNELMSG_GPADL_TORNDOWN,             1, vmbus_ongpadl_torndown},
 984         {CHANNELMSG_RELID_RELEASED,             0, NULL},
 985         {CHANNELMSG_INITIATE_CONTACT,           0, NULL},
 986         {CHANNELMSG_VERSION_RESPONSE,           1, vmbus_onversion_response},
 987         {CHANNELMSG_UNLOAD,                     0, NULL},
 988         {CHANNELMSG_UNLOAD_RESPONSE,            1, vmbus_unload_response},
 989         {CHANNELMSG_18,                         0, NULL},
 990         {CHANNELMSG_19,                         0, NULL},
 991         {CHANNELMSG_20,                         0, NULL},
 992         {CHANNELMSG_TL_CONNECT_REQUEST,         0, NULL},
 993 };
 994
 995 /*
 996  * vmbus_onmessage - Handler for channel protocol messages.
 997  *
 998  * This is invoked in the vmbus worker thread context.
 999  */
1000 void vmbus_onmessage(void *context)
1001 {
1002         struct hv_message *msg = context;
1003         struct vmbus_channel_message_header *hdr;
1004         int size;
1005
1006         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1007         size = msg->header.payload_size;
1008
1009         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1010                 pr_err("Received invalid channel message type %d size %d\n",
1011                            hdr->msgtype, size);
1012                 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1013                                      (unsigned char *)msg->u.payload, size);
1014                 return;
1015         }
1016
1017         if (channel_message_table[hdr->msgtype].message_handler)
1018                 channel_message_table[hdr->msgtype].message_handler(hdr);
1019         else
1020                 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1021 }
1022
1023 /*
1024  * vmbus_request_offers - Send a request to get all our pending offers.
1025  */
1026 int vmbus_request_offers(void)
1027 {
1028         struct vmbus_channel_message_header *msg;
1029         struct vmbus_channel_msginfo *msginfo;
1030         int ret;
1031
1032         msginfo = kmalloc(sizeof(*msginfo) +
1033                           sizeof(struct vmbus_channel_message_header),
1034                           GFP_KERNEL);
1035         if (!msginfo)
1036                 return -ENOMEM;
1037
1038         msg = (struct vmbus_channel_message_header *)msginfo->msg;
1039
1040         msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1041
1042
1043         ret = vmbus_post_msg(msg,
1044                                sizeof(struct vmbus_channel_message_header));
1045         if (ret != 0) {
1046                 pr_err("Unable to request offers - %d\n", ret);
1047
1048                 goto cleanup;
1049         }
1050
1051 cleanup:
1052         kfree(msginfo);
1053
1054         return ret;
1055 }
1056
1057 /*
1058  * Retrieve the (sub) channel on which to send an outgoing request.
1059  * When a primary channel has multiple sub-channels, we try to
1060  * distribute the load equally amongst all available channels.
1061  */
1062 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1063 {
1064         struct list_head *cur, *tmp;
1065         int cur_cpu;
1066         struct vmbus_channel *cur_channel;
1067         struct vmbus_channel *outgoing_channel = primary;
1068         int next_channel;
1069         int i = 1;
1070
1071         if (list_empty(&primary->sc_list))
1072                 return outgoing_channel;
1073
1074         next_channel = primary->next_oc++;
1075
1076         if (next_channel > (primary->num_sc)) {
1077                 primary->next_oc = 0;
1078                 return outgoing_channel;
1079         }
1080
1081         cur_cpu = hv_context.vp_index[get_cpu()];
1082         put_cpu();
1083         list_for_each_safe(cur, tmp, &primary->sc_list) {
1084                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1085                 if (cur_channel->state != CHANNEL_OPENED_STATE)
1086                         continue;
1087
1088                 if (cur_channel->target_vp == cur_cpu)
1089                         return cur_channel;
1090
1091                 if (i == next_channel)
1092                         return cur_channel;
1093
1094                 i++;
1095         }
1096
1097         return outgoing_channel;
1098 }
1099 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1100
1101 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1102 {
1103         struct list_head *cur, *tmp;
1104         struct vmbus_channel *cur_channel;
1105
1106         if (primary_channel->sc_creation_callback == NULL)
1107                 return;
1108
1109         list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1110                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1111
1112                 primary_channel->sc_creation_callback(cur_channel);
1113         }
1114 }
1115
1116 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1117                                 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1118 {
1119         primary_channel->sc_creation_callback = sc_cr_cb;
1120 }
1121 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1122
1123 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1124 {
1125         bool ret;
1126
1127         ret = !list_empty(&primary->sc_list);
1128
1129         if (ret) {
1130                 /*
1131                  * Invoke the callback on sub-channel creation.
1132                  * This will present a uniform interface to the
1133                  * clients.
1134                  */
1135                 invoke_sc_cb(primary);
1136         }
1137
1138         return ret;
1139 }
1140 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1141
1142 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1143                 void (*chn_rescind_cb)(struct vmbus_channel *))
1144 {
1145         channel->chn_rescind_callback = chn_rescind_cb;
1146 }
1147 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);