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/hyperv.h>
  32
  33 #include "hyperv_vmbus.h"
  34
  35 static void init_vp_index(struct vmbus_channel *channel,
  36                           const uuid_le *type_guid);
  37
  38 /**
  39  * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
  40  * @icmsghdrp: Pointer to msg header structure
  41  * @icmsg_negotiate: Pointer to negotiate message structure
  42  * @buf: Raw buffer channel data
  43  *
  44  * @icmsghdrp is of type &struct icmsg_hdr.
  45  * @negop is of type &struct icmsg_negotiate.
  46  * Set up and fill in default negotiate response message.
  47  *
  48  * The fw_version specifies the  framework version that
  49  * we can support and srv_version specifies the service
  50  * version we can support.
  51  *
  52  * Mainly used by Hyper-V drivers.
  53  */
  54 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
  55                                 struct icmsg_negotiate *negop, u8 *buf,
  56                                 int fw_version, int srv_version)
  57 {
  58         int icframe_major, icframe_minor;
  59         int icmsg_major, icmsg_minor;
  60         int fw_major, fw_minor;
  61         int srv_major, srv_minor;
  62         int i;
  63         bool found_match = false;
  64
  65         icmsghdrp->icmsgsize = 0x10;
  66         fw_major = (fw_version >> 16);
  67         fw_minor = (fw_version & 0xFFFF);
  68
  69         srv_major = (srv_version >> 16);
  70         srv_minor = (srv_version & 0xFFFF);
  71
  72         negop = (struct icmsg_negotiate *)&buf[
  73                 sizeof(struct vmbuspipe_hdr) +
  74                 sizeof(struct icmsg_hdr)];
  75
  76         icframe_major = negop->icframe_vercnt;
  77         icframe_minor = 0;
  78
  79         icmsg_major = negop->icmsg_vercnt;
  80         icmsg_minor = 0;
  81
  82         /*
  83          * Select the framework version number we will
  84          * support.
  85          */
  86
  87         for (i = 0; i < negop->icframe_vercnt; i++) {
  88                 if ((negop->icversion_data[i].major == fw_major) &&
  89                    (negop->icversion_data[i].minor == fw_minor)) {
  90                         icframe_major = negop->icversion_data[i].major;
  91                         icframe_minor = negop->icversion_data[i].minor;
  92                         found_match = true;
  93                 }
  94         }
  95
  96         if (!found_match)
  97                 goto fw_error;
  98
  99         found_match = false;
 100
 101         for (i = negop->icframe_vercnt;
 102                  (i < negop->icframe_vercnt + negop->icmsg_vercnt); i++) {
 103                 if ((negop->icversion_data[i].major == srv_major) &&
 104                    (negop->icversion_data[i].minor == srv_minor)) {
 105                         icmsg_major = negop->icversion_data[i].major;
 106                         icmsg_minor = negop->icversion_data[i].minor;
 107                         found_match = true;
 108                 }
 109         }
 110
 111         /*
 112          * Respond with the framework and service
 113          * version numbers we can support.
 114          */
 115
 116 fw_error:
 117         if (!found_match) {
 118                 negop->icframe_vercnt = 0;
 119                 negop->icmsg_vercnt = 0;
 120         } else {
 121                 negop->icframe_vercnt = 1;
 122                 negop->icmsg_vercnt = 1;
 123         }
 124
 125         negop->icversion_data[0].major = icframe_major;
 126         negop->icversion_data[0].minor = icframe_minor;
 127         negop->icversion_data[1].major = icmsg_major;
 128         negop->icversion_data[1].minor = icmsg_minor;
 129         return found_match;
 130 }
 131
 132 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
 133
 134 /*
 135  * alloc_channel - Allocate and initialize a vmbus channel object
 136  */
 137 static struct vmbus_channel *alloc_channel(void)
 138 {
 139         static atomic_t chan_num = ATOMIC_INIT(0);
 140         struct vmbus_channel *channel;
 141
 142         channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
 143         if (!channel)
 144                 return NULL;
 145
 146         channel->id = atomic_inc_return(&chan_num);
 147         spin_lock_init(&channel->inbound_lock);
 148         spin_lock_init(&channel->lock);
 149
 150         INIT_LIST_HEAD(&channel->sc_list);
 151         INIT_LIST_HEAD(&channel->percpu_list);
 152
 153         return channel;
 154 }
 155
 156 /*
 157  * free_channel - Release the resources used by the vmbus channel object
 158  */
 159 static void free_channel(struct vmbus_channel *channel)
 160 {
 161         kfree(channel);
 162 }
 163
 164 static void percpu_channel_enq(void *arg)
 165 {
 166         struct vmbus_channel *channel = arg;
 167         int cpu = smp_processor_id();
 168
 169         list_add_tail(&channel->percpu_list, &hv_context.percpu_list[cpu]);
 170 }
 171
 172 static void percpu_channel_deq(void *arg)
 173 {
 174         struct vmbus_channel *channel = arg;
 175
 176         list_del(&channel->percpu_list);
 177 }
 178
 179
 180 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
 181 {
 182         struct vmbus_channel_relid_released msg;
 183         unsigned long flags;
 184         struct vmbus_channel *primary_channel;
 185
 186         memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
 187         msg.child_relid = relid;
 188         msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
 189         vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released));
 190
 191         if (channel == NULL)
 192                 return;
 193
 194         if (channel->target_cpu != get_cpu()) {
 195                 put_cpu();
 196                 smp_call_function_single(channel->target_cpu,
 197                                          percpu_channel_deq, channel, true);
 198         } else {
 199                 percpu_channel_deq(channel);
 200                 put_cpu();
 201         }
 202
 203         if (channel->primary_channel == NULL) {
 204                 spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
 205                 list_del(&channel->listentry);
 206                 spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
 207
 208                 primary_channel = channel;
 209         } else {
 210                 primary_channel = channel->primary_channel;
 211                 spin_lock_irqsave(&primary_channel->lock, flags);
 212                 list_del(&channel->sc_list);
 213                 primary_channel->num_sc--;
 214                 spin_unlock_irqrestore(&primary_channel->lock, flags);
 215         }
 216
 217         /*
 218          * We need to free the bit for init_vp_index() to work in the case
 219          * of sub-channel, when we reload drivers like hv_netvsc.
 220          */
 221         cpumask_clear_cpu(channel->target_cpu,
 222                           &primary_channel->alloced_cpus_in_node);
 223
 224         free_channel(channel);
 225 }
 226
 227 void vmbus_free_channels(void)
 228 {
 229         struct vmbus_channel *channel, *tmp;
 230
 231         list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
 232                 listentry) {
 233                 /* if we don't set rescind to true, vmbus_close_internal()
 234                  * won't invoke hv_process_channel_removal().
 235                  */
 236                 channel->rescind = true;
 237
 238                 vmbus_device_unregister(channel->device_obj);
 239         }
 240 }
 241
 242 /*
 243  * vmbus_process_offer - Process the offer by creating a channel/device
 244  * associated with this offer
 245  */
 246 static void vmbus_process_offer(struct vmbus_channel *newchannel)
 247 {
 248         struct vmbus_channel *channel;
 249         bool fnew = true;
 250         unsigned long flags;
 251
 252         /* Make sure this is a new offer */
 253         spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
 254
 255         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 256                 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
 257                         newchannel->offermsg.offer.if_type) &&
 258                         !uuid_le_cmp(channel->offermsg.offer.if_instance,
 259                                 newchannel->offermsg.offer.if_instance)) {
 260                         fnew = false;
 261                         break;
 262                 }
 263         }
 264
 265         if (fnew)
 266                 list_add_tail(&newchannel->listentry,
 267                               &vmbus_connection.chn_list);
 268
 269         spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
 270
 271         if (!fnew) {
 272                 /*
 273                  * Check to see if this is a sub-channel.
 274                  */
 275                 if (newchannel->offermsg.offer.sub_channel_index != 0) {
 276                         /*
 277                          * Process the sub-channel.
 278                          */
 279                         newchannel->primary_channel = channel;
 280                         spin_lock_irqsave(&channel->lock, flags);
 281                         list_add_tail(&newchannel->sc_list, &channel->sc_list);
 282                         channel->num_sc++;
 283                         spin_unlock_irqrestore(&channel->lock, flags);
 284                 } else
 285                         goto err_free_chan;
 286         }
 287
 288         init_vp_index(newchannel, &newchannel->offermsg.offer.if_type);
 289
 290         if (newchannel->target_cpu != get_cpu()) {
 291                 put_cpu();
 292                 smp_call_function_single(newchannel->target_cpu,
 293                                          percpu_channel_enq,
 294                                          newchannel, true);
 295         } else {
 296                 percpu_channel_enq(newchannel);
 297                 put_cpu();
 298         }
 299
 300         /*
 301          * This state is used to indicate a successful open
 302          * so that when we do close the channel normally, we
 303          * can cleanup properly
 304          */
 305         newchannel->state = CHANNEL_OPEN_STATE;
 306
 307         if (!fnew) {
 308                 if (channel->sc_creation_callback != NULL)
 309                         channel->sc_creation_callback(newchannel);
 310                 return;
 311         }
 312
 313         /*
 314          * Start the process of binding this offer to the driver
 315          * We need to set the DeviceObject field before calling
 316          * vmbus_child_dev_add()
 317          */
 318         newchannel->device_obj = vmbus_device_create(
 319                 &newchannel->offermsg.offer.if_type,
 320                 &newchannel->offermsg.offer.if_instance,
 321                 newchannel);
 322         if (!newchannel->device_obj)
 323                 goto err_deq_chan;
 324
 325         /*
 326          * Add the new device to the bus. This will kick off device-driver
 327          * binding which eventually invokes the device driver's AddDevice()
 328          * method.
 329          */
 330         if (vmbus_device_register(newchannel->device_obj) != 0) {
 331                 pr_err("unable to add child device object (relid %d)\n",
 332                         newchannel->offermsg.child_relid);
 333                 kfree(newchannel->device_obj);
 334                 goto err_deq_chan;
 335         }
 336         return;
 337
 338 err_deq_chan:
 339         spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
 340         list_del(&newchannel->listentry);
 341         spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
 342
 343         if (newchannel->target_cpu != get_cpu()) {
 344                 put_cpu();
 345                 smp_call_function_single(newchannel->target_cpu,
 346                                          percpu_channel_deq, newchannel, true);
 347         } else {
 348                 percpu_channel_deq(newchannel);
 349                 put_cpu();
 350         }
 351
 352 err_free_chan:
 353         free_channel(newchannel);
 354 }
 355
 356 enum {
 357         IDE = 0,
 358         SCSI,
 359         NIC,
 360         ND_NIC,
 361         MAX_PERF_CHN,
 362 };
 363
 364 /*
 365  * This is an array of device_ids (device types) that are performance critical.
 366  * We attempt to distribute the interrupt load for these devices across
 367  * all available CPUs.
 368  */
 369 static const struct hv_vmbus_device_id hp_devs[] = {
 370         /* IDE */
 371         { HV_IDE_GUID, },
 372         /* Storage - SCSI */
 373         { HV_SCSI_GUID, },
 374         /* Network */
 375         { HV_NIC_GUID, },
 376         /* NetworkDirect Guest RDMA */
 377         { HV_ND_GUID, },
 378 };
 379
 380
 381 /*
 382  * We use this state to statically distribute the channel interrupt load.
 383  */
 384 static int next_numa_node_id;
 385
 386 /*
 387  * Starting with Win8, we can statically distribute the incoming
 388  * channel interrupt load by binding a channel to VCPU.
 389  * We do this in a hierarchical fashion:
 390  * First distribute the primary channels across available NUMA nodes
 391  * and then distribute the subchannels amongst the CPUs in the NUMA
 392  * node assigned to the primary channel.
 393  *
 394  * For pre-win8 hosts or non-performance critical channels we assign the
 395  * first CPU in the first NUMA node.
 396  */
 397 static void init_vp_index(struct vmbus_channel *channel, const uuid_le *type_guid)
 398 {
 399         u32 cur_cpu;
 400         int i;
 401         bool perf_chn = false;
 402         struct vmbus_channel *primary = channel->primary_channel;
 403         int next_node;
 404         struct cpumask available_mask;
 405         struct cpumask *alloced_mask;
 406
 407         for (i = IDE; i < MAX_PERF_CHN; i++) {
 408                 if (!memcmp(type_guid->b, hp_devs[i].guid,
 409                                  sizeof(uuid_le))) {
 410                         perf_chn = true;
 411                         break;
 412                 }
 413         }
 414         if ((vmbus_proto_version == VERSION_WS2008) ||
 415             (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
 416                 /*
 417                  * Prior to win8, all channel interrupts are
 418                  * delivered on cpu 0.
 419                  * Also if the channel is not a performance critical
 420                  * channel, bind it to cpu 0.
 421                  */
 422                 channel->numa_node = 0;
 423                 channel->target_cpu = 0;
 424                 channel->target_vp = hv_context.vp_index[0];
 425                 return;
 426         }
 427
 428         /*
 429          * We distribute primary channels evenly across all the available
 430          * NUMA nodes and within the assigned NUMA node we will assign the
 431          * first available CPU to the primary channel.
 432          * The sub-channels will be assigned to the CPUs available in the
 433          * NUMA node evenly.
 434          */
 435         if (!primary) {
 436                 while (true) {
 437                         next_node = next_numa_node_id++;
 438                         if (next_node == nr_node_ids)
 439                                 next_node = next_numa_node_id = 0;
 440                         if (cpumask_empty(cpumask_of_node(next_node)))
 441                                 continue;
 442                         break;
 443                 }
 444                 channel->numa_node = next_node;
 445                 primary = channel;
 446         }
 447         alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
 448
 449         if (cpumask_weight(alloced_mask) ==
 450             cpumask_weight(cpumask_of_node(primary->numa_node))) {
 451                 /*
 452                  * We have cycled through all the CPUs in the node;
 453                  * reset the alloced map.
 454                  */
 455                 cpumask_clear(alloced_mask);
 456         }
 457
 458         cpumask_xor(&available_mask, alloced_mask,
 459                     cpumask_of_node(primary->numa_node));
 460
 461         cur_cpu = -1;
 462         while (true) {
 463                 cur_cpu = cpumask_next(cur_cpu, &available_mask);
 464                 if (cur_cpu >= nr_cpu_ids) {
 465                         cur_cpu = -1;
 466                         cpumask_copy(&available_mask,
 467                                      cpumask_of_node(primary->numa_node));
 468                         continue;
 469                 }
 470
 471                 /*
 472                  * NOTE: in the case of sub-channel, we clear the sub-channel
 473                  * related bit(s) in primary->alloced_cpus_in_node in
 474                  * hv_process_channel_removal(), so when we reload drivers
 475                  * like hv_netvsc in SMP guest, here we're able to re-allocate
 476                  * bit from primary->alloced_cpus_in_node.
 477                  */
 478                 if (!cpumask_test_cpu(cur_cpu,
 479                                 &primary->alloced_cpus_in_node)) {
 480                         cpumask_set_cpu(cur_cpu,
 481                                         &primary->alloced_cpus_in_node);
 482                         cpumask_set_cpu(cur_cpu, alloced_mask);
 483                         break;
 484                 }
 485         }
 486
 487         channel->target_cpu = cur_cpu;
 488         channel->target_vp = hv_context.vp_index[cur_cpu];
 489 }
 490
 491 /*
 492  * vmbus_unload_response - Handler for the unload response.
 493  */
 494 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
 495 {
 496         /*
 497          * This is a global event; just wakeup the waiting thread.
 498          * Once we successfully unload, we can cleanup the monitor state.
 499          */
 500         complete(&vmbus_connection.unload_event);
 501 }
 502
 503 void vmbus_initiate_unload(void)
 504 {
 505         struct vmbus_channel_message_header hdr;
 506
 507         /* Pre-Win2012R2 hosts don't support reconnect */
 508         if (vmbus_proto_version < VERSION_WIN8_1)
 509                 return;
 510
 511         init_completion(&vmbus_connection.unload_event);
 512         memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 513         hdr.msgtype = CHANNELMSG_UNLOAD;
 514         vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header));
 515
 516         wait_for_completion(&vmbus_connection.unload_event);
 517 }
 518
 519 /*
 520  * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
 521  *
 522  */
 523 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
 524 {
 525         struct vmbus_channel_offer_channel *offer;
 526         struct vmbus_channel *newchannel;
 527
 528         offer = (struct vmbus_channel_offer_channel *)hdr;
 529
 530         /* Allocate the channel object and save this offer. */
 531         newchannel = alloc_channel();
 532         if (!newchannel) {
 533                 pr_err("Unable to allocate channel object\n");
 534                 return;
 535         }
 536
 537         /*
 538          * By default we setup state to enable batched
 539          * reading. A specific service can choose to
 540          * disable this prior to opening the channel.
 541          */
 542         newchannel->batched_reading = true;
 543
 544         /*
 545          * Setup state for signalling the host.
 546          */
 547         newchannel->sig_event = (struct hv_input_signal_event *)
 548                                 (ALIGN((unsigned long)
 549                                 &newchannel->sig_buf,
 550                                 HV_HYPERCALL_PARAM_ALIGN));
 551
 552         newchannel->sig_event->connectionid.asu32 = 0;
 553         newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
 554         newchannel->sig_event->flag_number = 0;
 555         newchannel->sig_event->rsvdz = 0;
 556
 557         if (vmbus_proto_version != VERSION_WS2008) {
 558                 newchannel->is_dedicated_interrupt =
 559                                 (offer->is_dedicated_interrupt != 0);
 560                 newchannel->sig_event->connectionid.u.id =
 561                                 offer->connection_id;
 562         }
 563
 564         memcpy(&newchannel->offermsg, offer,
 565                sizeof(struct vmbus_channel_offer_channel));
 566         newchannel->monitor_grp = (u8)offer->monitorid / 32;
 567         newchannel->monitor_bit = (u8)offer->monitorid % 32;
 568
 569         vmbus_process_offer(newchannel);
 570 }
 571
 572 /*
 573  * vmbus_onoffer_rescind - Rescind offer handler.
 574  *
 575  * We queue a work item to process this offer synchronously
 576  */
 577 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 578 {
 579         struct vmbus_channel_rescind_offer *rescind;
 580         struct vmbus_channel *channel;
 581         unsigned long flags;
 582         struct device *dev;
 583
 584         rescind = (struct vmbus_channel_rescind_offer *)hdr;
 585         channel = relid2channel(rescind->child_relid);
 586
 587         if (channel == NULL) {
 588                 hv_process_channel_removal(NULL, rescind->child_relid);
 589                 return;
 590         }
 591
 592         spin_lock_irqsave(&channel->lock, flags);
 593         channel->rescind = true;
 594         spin_unlock_irqrestore(&channel->lock, flags);
 595
 596         if (channel->device_obj) {
 597                 /*
 598                  * We will have to unregister this device from the
 599                  * driver core.
 600                  */
 601                 dev = get_device(&channel->device_obj->device);
 602                 if (dev) {
 603                         vmbus_device_unregister(channel->device_obj);
 604                         put_device(dev);
 605                 }
 606         } else {
 607                 hv_process_channel_removal(channel,
 608                         channel->offermsg.child_relid);
 609         }
 610 }
 611
 612 /*
 613  * vmbus_onoffers_delivered -
 614  * This is invoked when all offers have been delivered.
 615  *
 616  * Nothing to do here.
 617  */
 618 static void vmbus_onoffers_delivered(
 619                         struct vmbus_channel_message_header *hdr)
 620 {
 621 }
 622
 623 /*
 624  * vmbus_onopen_result - Open result handler.
 625  *
 626  * This is invoked when we received a response to our channel open request.
 627  * Find the matching request, copy the response and signal the requesting
 628  * thread.
 629  */
 630 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
 631 {
 632         struct vmbus_channel_open_result *result;
 633         struct vmbus_channel_msginfo *msginfo;
 634         struct vmbus_channel_message_header *requestheader;
 635         struct vmbus_channel_open_channel *openmsg;
 636         unsigned long flags;
 637
 638         result = (struct vmbus_channel_open_result *)hdr;
 639
 640         /*
 641          * Find the open msg, copy the result and signal/unblock the wait event
 642          */
 643         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 644
 645         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 646                                 msglistentry) {
 647                 requestheader =
 648                         (struct vmbus_channel_message_header *)msginfo->msg;
 649
 650                 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
 651                         openmsg =
 652                         (struct vmbus_channel_open_channel *)msginfo->msg;
 653                         if (openmsg->child_relid == result->child_relid &&
 654                             openmsg->openid == result->openid) {
 655                                 memcpy(&msginfo->response.open_result,
 656                                        result,
 657                                        sizeof(
 658                                         struct vmbus_channel_open_result));
 659                                 complete(&msginfo->waitevent);
 660                                 break;
 661                         }
 662                 }
 663         }
 664         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 665 }
 666
 667 /*
 668  * vmbus_ongpadl_created - GPADL created handler.
 669  *
 670  * This is invoked when we received a response to our gpadl create request.
 671  * Find the matching request, copy the response and signal the requesting
 672  * thread.
 673  */
 674 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
 675 {
 676         struct vmbus_channel_gpadl_created *gpadlcreated;
 677         struct vmbus_channel_msginfo *msginfo;
 678         struct vmbus_channel_message_header *requestheader;
 679         struct vmbus_channel_gpadl_header *gpadlheader;
 680         unsigned long flags;
 681
 682         gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
 683
 684         /*
 685          * Find the establish msg, copy the result and signal/unblock the wait
 686          * event
 687          */
 688         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 689
 690         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 691                                 msglistentry) {
 692                 requestheader =
 693                         (struct vmbus_channel_message_header *)msginfo->msg;
 694
 695                 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
 696                         gpadlheader =
 697                         (struct vmbus_channel_gpadl_header *)requestheader;
 698
 699                         if ((gpadlcreated->child_relid ==
 700                              gpadlheader->child_relid) &&
 701                             (gpadlcreated->gpadl == gpadlheader->gpadl)) {
 702                                 memcpy(&msginfo->response.gpadl_created,
 703                                        gpadlcreated,
 704                                        sizeof(
 705                                         struct vmbus_channel_gpadl_created));
 706                                 complete(&msginfo->waitevent);
 707                                 break;
 708                         }
 709                 }
 710         }
 711         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 712 }
 713
 714 /*
 715  * vmbus_ongpadl_torndown - GPADL torndown handler.
 716  *
 717  * This is invoked when we received a response to our gpadl teardown request.
 718  * Find the matching request, copy the response and signal the requesting
 719  * thread.
 720  */
 721 static void vmbus_ongpadl_torndown(
 722                         struct vmbus_channel_message_header *hdr)
 723 {
 724         struct vmbus_channel_gpadl_torndown *gpadl_torndown;
 725         struct vmbus_channel_msginfo *msginfo;
 726         struct vmbus_channel_message_header *requestheader;
 727         struct vmbus_channel_gpadl_teardown *gpadl_teardown;
 728         unsigned long flags;
 729
 730         gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
 731
 732         /*
 733          * Find the open msg, copy the result and signal/unblock the wait event
 734          */
 735         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 736
 737         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 738                                 msglistentry) {
 739                 requestheader =
 740                         (struct vmbus_channel_message_header *)msginfo->msg;
 741
 742                 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
 743                         gpadl_teardown =
 744                         (struct vmbus_channel_gpadl_teardown *)requestheader;
 745
 746                         if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
 747                                 memcpy(&msginfo->response.gpadl_torndown,
 748                                        gpadl_torndown,
 749                                        sizeof(
 750                                         struct vmbus_channel_gpadl_torndown));
 751                                 complete(&msginfo->waitevent);
 752                                 break;
 753                         }
 754                 }
 755         }
 756         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 757 }
 758
 759 /*
 760  * vmbus_onversion_response - Version response handler
 761  *
 762  * This is invoked when we received a response to our initiate contact request.
 763  * Find the matching request, copy the response and signal the requesting
 764  * thread.
 765  */
 766 static void vmbus_onversion_response(
 767                 struct vmbus_channel_message_header *hdr)
 768 {
 769         struct vmbus_channel_msginfo *msginfo;
 770         struct vmbus_channel_message_header *requestheader;
 771         struct vmbus_channel_version_response *version_response;
 772         unsigned long flags;
 773
 774         version_response = (struct vmbus_channel_version_response *)hdr;
 775         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 776
 777         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 778                                 msglistentry) {
 779                 requestheader =
 780                         (struct vmbus_channel_message_header *)msginfo->msg;
 781
 782                 if (requestheader->msgtype ==
 783                     CHANNELMSG_INITIATE_CONTACT) {
 784                         memcpy(&msginfo->response.version_response,
 785                               version_response,
 786                               sizeof(struct vmbus_channel_version_response));
 787                         complete(&msginfo->waitevent);
 788                 }
 789         }
 790         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 791 }
 792
 793 /* Channel message dispatch table */
 794 struct vmbus_channel_message_table_entry
 795         channel_message_table[CHANNELMSG_COUNT] = {
 796         {CHANNELMSG_INVALID,                    0, NULL},
 797         {CHANNELMSG_OFFERCHANNEL,               0, vmbus_onoffer},
 798         {CHANNELMSG_RESCIND_CHANNELOFFER,       0, vmbus_onoffer_rescind},
 799         {CHANNELMSG_REQUESTOFFERS,              0, NULL},
 800         {CHANNELMSG_ALLOFFERS_DELIVERED,        1, vmbus_onoffers_delivered},
 801         {CHANNELMSG_OPENCHANNEL,                0, NULL},
 802         {CHANNELMSG_OPENCHANNEL_RESULT,         1, vmbus_onopen_result},
 803         {CHANNELMSG_CLOSECHANNEL,               0, NULL},
 804         {CHANNELMSG_GPADL_HEADER,               0, NULL},
 805         {CHANNELMSG_GPADL_BODY,                 0, NULL},
 806         {CHANNELMSG_GPADL_CREATED,              1, vmbus_ongpadl_created},
 807         {CHANNELMSG_GPADL_TEARDOWN,             0, NULL},
 808         {CHANNELMSG_GPADL_TORNDOWN,             1, vmbus_ongpadl_torndown},
 809         {CHANNELMSG_RELID_RELEASED,             0, NULL},
 810         {CHANNELMSG_INITIATE_CONTACT,           0, NULL},
 811         {CHANNELMSG_VERSION_RESPONSE,           1, vmbus_onversion_response},
 812         {CHANNELMSG_UNLOAD,                     0, NULL},
 813         {CHANNELMSG_UNLOAD_RESPONSE,            1, vmbus_unload_response},
 814 };
 815
 816 /*
 817  * vmbus_onmessage - Handler for channel protocol messages.
 818  *
 819  * This is invoked in the vmbus worker thread context.
 820  */
 821 void vmbus_onmessage(void *context)
 822 {
 823         struct hv_message *msg = context;
 824         struct vmbus_channel_message_header *hdr;
 825         int size;
 826
 827         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 828         size = msg->header.payload_size;
 829
 830         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 831                 pr_err("Received invalid channel message type %d size %d\n",
 832                            hdr->msgtype, size);
 833                 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
 834                                      (unsigned char *)msg->u.payload, size);
 835                 return;
 836         }
 837
 838         if (channel_message_table[hdr->msgtype].message_handler)
 839                 channel_message_table[hdr->msgtype].message_handler(hdr);
 840         else
 841                 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
 842 }
 843
 844 /*
 845  * vmbus_request_offers - Send a request to get all our pending offers.
 846  */
 847 int vmbus_request_offers(void)
 848 {
 849         struct vmbus_channel_message_header *msg;
 850         struct vmbus_channel_msginfo *msginfo;
 851         int ret;
 852
 853         msginfo = kmalloc(sizeof(*msginfo) +
 854                           sizeof(struct vmbus_channel_message_header),
 855                           GFP_KERNEL);
 856         if (!msginfo)
 857                 return -ENOMEM;
 858
 859         msg = (struct vmbus_channel_message_header *)msginfo->msg;
 860
 861         msg->msgtype = CHANNELMSG_REQUESTOFFERS;
 862
 863
 864         ret = vmbus_post_msg(msg,
 865                                sizeof(struct vmbus_channel_message_header));
 866         if (ret != 0) {
 867                 pr_err("Unable to request offers - %d\n", ret);
 868
 869                 goto cleanup;
 870         }
 871
 872 cleanup:
 873         kfree(msginfo);
 874
 875         return ret;
 876 }
 877
 878 /*
 879  * Retrieve the (sub) channel on which to send an outgoing request.
 880  * When a primary channel has multiple sub-channels, we try to
 881  * distribute the load equally amongst all available channels.
 882  */
 883 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
 884 {
 885         struct list_head *cur, *tmp;
 886         int cur_cpu;
 887         struct vmbus_channel *cur_channel;
 888         struct vmbus_channel *outgoing_channel = primary;
 889         int next_channel;
 890         int i = 1;
 891
 892         if (list_empty(&primary->sc_list))
 893                 return outgoing_channel;
 894
 895         next_channel = primary->next_oc++;
 896
 897         if (next_channel > (primary->num_sc)) {
 898                 primary->next_oc = 0;
 899                 return outgoing_channel;
 900         }
 901
 902         cur_cpu = hv_context.vp_index[get_cpu()];
 903         put_cpu();
 904         list_for_each_safe(cur, tmp, &primary->sc_list) {
 905                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
 906                 if (cur_channel->state != CHANNEL_OPENED_STATE)
 907                         continue;
 908
 909                 if (cur_channel->target_vp == cur_cpu)
 910                         return cur_channel;
 911
 912                 if (i == next_channel)
 913                         return cur_channel;
 914
 915                 i++;
 916         }
 917
 918         return outgoing_channel;
 919 }
 920 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
 921
 922 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
 923 {
 924         struct list_head *cur, *tmp;
 925         struct vmbus_channel *cur_channel;
 926
 927         if (primary_channel->sc_creation_callback == NULL)
 928                 return;
 929
 930         list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
 931                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
 932
 933                 primary_channel->sc_creation_callback(cur_channel);
 934         }
 935 }
 936
 937 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
 938                                 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
 939 {
 940         primary_channel->sc_creation_callback = sc_cr_cb;
 941 }
 942 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
 943
 944 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
 945 {
 946         bool ret;
 947
 948         ret = !list_empty(&primary->sc_list);
 949
 950         if (ret) {
 951                 /*
 952                  * Invoke the callback on sub-channel creation.
 953                  * This will present a uniform interface to the
 954                  * clients.
 955                  */
 956                 invoke_sc_cb(primary);
 957         }
 958
 959         return ret;
 960 }
 961 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);