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 static void vmbus_release_relid(u32 relid)
 181 {
 182         struct vmbus_channel_relid_released msg;
 183
 184         memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
 185         msg.child_relid = relid;
 186         msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
 187         vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released));
 188 }
 189
 190 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
 191 {
 192         unsigned long flags;
 193         struct vmbus_channel *primary_channel;
 194
 195         vmbus_release_relid(relid);
 196
 197         BUG_ON(!channel->rescind);
 198
 199         if (channel->target_cpu != get_cpu()) {
 200                 put_cpu();
 201                 smp_call_function_single(channel->target_cpu,
 202                                          percpu_channel_deq, channel, true);
 203         } else {
 204                 percpu_channel_deq(channel);
 205                 put_cpu();
 206         }
 207
 208         if (channel->primary_channel == NULL) {
 209                 spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
 210                 list_del(&channel->listentry);
 211                 spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
 212
 213                 primary_channel = channel;
 214         } else {
 215                 primary_channel = channel->primary_channel;
 216                 spin_lock_irqsave(&primary_channel->lock, flags);
 217                 list_del(&channel->sc_list);
 218                 primary_channel->num_sc--;
 219                 spin_unlock_irqrestore(&primary_channel->lock, flags);
 220         }
 221
 222         /*
 223          * We need to free the bit for init_vp_index() to work in the case
 224          * of sub-channel, when we reload drivers like hv_netvsc.
 225          */
 226         cpumask_clear_cpu(channel->target_cpu,
 227                           &primary_channel->alloced_cpus_in_node);
 228
 229         free_channel(channel);
 230 }
 231
 232 void vmbus_free_channels(void)
 233 {
 234         struct vmbus_channel *channel, *tmp;
 235
 236         list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
 237                 listentry) {
 238                 /* hv_process_channel_removal() needs this */
 239                 channel->rescind = true;
 240
 241                 vmbus_device_unregister(channel->device_obj);
 242         }
 243 }
 244
 245 /*
 246  * vmbus_process_offer - Process the offer by creating a channel/device
 247  * associated with this offer
 248  */
 249 static void vmbus_process_offer(struct vmbus_channel *newchannel)
 250 {
 251         struct vmbus_channel *channel;
 252         bool fnew = true;
 253         unsigned long flags;
 254
 255         /* Make sure this is a new offer */
 256         spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
 257
 258         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 259                 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
 260                         newchannel->offermsg.offer.if_type) &&
 261                         !uuid_le_cmp(channel->offermsg.offer.if_instance,
 262                                 newchannel->offermsg.offer.if_instance)) {
 263                         fnew = false;
 264                         break;
 265                 }
 266         }
 267
 268         if (fnew)
 269                 list_add_tail(&newchannel->listentry,
 270                               &vmbus_connection.chn_list);
 271
 272         spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
 273
 274         if (!fnew) {
 275                 /*
 276                  * Check to see if this is a sub-channel.
 277                  */
 278                 if (newchannel->offermsg.offer.sub_channel_index != 0) {
 279                         /*
 280                          * Process the sub-channel.
 281                          */
 282                         newchannel->primary_channel = channel;
 283                         spin_lock_irqsave(&channel->lock, flags);
 284                         list_add_tail(&newchannel->sc_list, &channel->sc_list);
 285                         channel->num_sc++;
 286                         spin_unlock_irqrestore(&channel->lock, flags);
 287                 } else
 288                         goto err_free_chan;
 289         }
 290
 291         init_vp_index(newchannel, &newchannel->offermsg.offer.if_type);
 292
 293         if (newchannel->target_cpu != get_cpu()) {
 294                 put_cpu();
 295                 smp_call_function_single(newchannel->target_cpu,
 296                                          percpu_channel_enq,
 297                                          newchannel, true);
 298         } else {
 299                 percpu_channel_enq(newchannel);
 300                 put_cpu();
 301         }
 302
 303         /*
 304          * This state is used to indicate a successful open
 305          * so that when we do close the channel normally, we
 306          * can cleanup properly
 307          */
 308         newchannel->state = CHANNEL_OPEN_STATE;
 309
 310         if (!fnew) {
 311                 if (channel->sc_creation_callback != NULL)
 312                         channel->sc_creation_callback(newchannel);
 313                 return;
 314         }
 315
 316         /*
 317          * Start the process of binding this offer to the driver
 318          * We need to set the DeviceObject field before calling
 319          * vmbus_child_dev_add()
 320          */
 321         newchannel->device_obj = vmbus_device_create(
 322                 &newchannel->offermsg.offer.if_type,
 323                 &newchannel->offermsg.offer.if_instance,
 324                 newchannel);
 325         if (!newchannel->device_obj)
 326                 goto err_deq_chan;
 327
 328         /*
 329          * Add the new device to the bus. This will kick off device-driver
 330          * binding which eventually invokes the device driver's AddDevice()
 331          * method.
 332          */
 333         if (vmbus_device_register(newchannel->device_obj) != 0) {
 334                 pr_err("unable to add child device object (relid %d)\n",
 335                         newchannel->offermsg.child_relid);
 336                 kfree(newchannel->device_obj);
 337                 goto err_deq_chan;
 338         }
 339         return;
 340
 341 err_deq_chan:
 342         vmbus_release_relid(newchannel->offermsg.child_relid);
 343
 344         spin_lock_irqsave(&vmbus_connection.channel_lock, flags);
 345         list_del(&newchannel->listentry);
 346         spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags);
 347
 348         if (newchannel->target_cpu != get_cpu()) {
 349                 put_cpu();
 350                 smp_call_function_single(newchannel->target_cpu,
 351                                          percpu_channel_deq, newchannel, true);
 352         } else {
 353                 percpu_channel_deq(newchannel);
 354                 put_cpu();
 355         }
 356
 357 err_free_chan:
 358         free_channel(newchannel);
 359 }
 360
 361 enum {
 362         IDE = 0,
 363         SCSI,
 364         NIC,
 365         ND_NIC,
 366         PCIE,
 367         MAX_PERF_CHN,
 368 };
 369
 370 /*
 371  * This is an array of device_ids (device types) that are performance critical.
 372  * We attempt to distribute the interrupt load for these devices across
 373  * all available CPUs.
 374  */
 375 static const struct hv_vmbus_device_id hp_devs[] = {
 376         /* IDE */
 377         { HV_IDE_GUID, },
 378         /* Storage - SCSI */
 379         { HV_SCSI_GUID, },
 380         /* Network */
 381         { HV_NIC_GUID, },
 382         /* NetworkDirect Guest RDMA */
 383         { HV_ND_GUID, },
 384         /* PCI Express Pass Through */
 385         { HV_PCIE_GUID, },
 386 };
 387
 388
 389 /*
 390  * We use this state to statically distribute the channel interrupt load.
 391  */
 392 static int next_numa_node_id;
 393
 394 /*
 395  * Starting with Win8, we can statically distribute the incoming
 396  * channel interrupt load by binding a channel to VCPU.
 397  * We do this in a hierarchical fashion:
 398  * First distribute the primary channels across available NUMA nodes
 399  * and then distribute the subchannels amongst the CPUs in the NUMA
 400  * node assigned to the primary channel.
 401  *
 402  * For pre-win8 hosts or non-performance critical channels we assign the
 403  * first CPU in the first NUMA node.
 404  */
 405 static void init_vp_index(struct vmbus_channel *channel, const uuid_le *type_guid)
 406 {
 407         u32 cur_cpu;
 408         int i;
 409         bool perf_chn = false;
 410         struct vmbus_channel *primary = channel->primary_channel;
 411         int next_node;
 412         struct cpumask available_mask;
 413         struct cpumask *alloced_mask;
 414
 415         for (i = IDE; i < MAX_PERF_CHN; i++) {
 416                 if (!uuid_le_cmp(*type_guid, hp_devs[i].guid)) {
 417                         perf_chn = true;
 418                         break;
 419                 }
 420         }
 421         if ((vmbus_proto_version == VERSION_WS2008) ||
 422             (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
 423                 /*
 424                  * Prior to win8, all channel interrupts are
 425                  * delivered on cpu 0.
 426                  * Also if the channel is not a performance critical
 427                  * channel, bind it to cpu 0.
 428                  */
 429                 channel->numa_node = 0;
 430                 channel->target_cpu = 0;
 431                 channel->target_vp = hv_context.vp_index[0];
 432                 return;
 433         }
 434
 435         /*
 436          * We distribute primary channels evenly across all the available
 437          * NUMA nodes and within the assigned NUMA node we will assign the
 438          * first available CPU to the primary channel.
 439          * The sub-channels will be assigned to the CPUs available in the
 440          * NUMA node evenly.
 441          */
 442         if (!primary) {
 443                 while (true) {
 444                         next_node = next_numa_node_id++;
 445                         if (next_node == nr_node_ids)
 446                                 next_node = next_numa_node_id = 0;
 447                         if (cpumask_empty(cpumask_of_node(next_node)))
 448                                 continue;
 449                         break;
 450                 }
 451                 channel->numa_node = next_node;
 452                 primary = channel;
 453         }
 454         alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
 455
 456         if (cpumask_weight(alloced_mask) ==
 457             cpumask_weight(cpumask_of_node(primary->numa_node))) {
 458                 /*
 459                  * We have cycled through all the CPUs in the node;
 460                  * reset the alloced map.
 461                  */
 462                 cpumask_clear(alloced_mask);
 463         }
 464
 465         cpumask_xor(&available_mask, alloced_mask,
 466                     cpumask_of_node(primary->numa_node));
 467
 468         cur_cpu = -1;
 469         while (true) {
 470                 cur_cpu = cpumask_next(cur_cpu, &available_mask);
 471                 if (cur_cpu >= nr_cpu_ids) {
 472                         cur_cpu = -1;
 473                         cpumask_copy(&available_mask,
 474                                      cpumask_of_node(primary->numa_node));
 475                         continue;
 476                 }
 477
 478                 /*
 479                  * NOTE: in the case of sub-channel, we clear the sub-channel
 480                  * related bit(s) in primary->alloced_cpus_in_node in
 481                  * hv_process_channel_removal(), so when we reload drivers
 482                  * like hv_netvsc in SMP guest, here we're able to re-allocate
 483                  * bit from primary->alloced_cpus_in_node.
 484                  */
 485                 if (!cpumask_test_cpu(cur_cpu,
 486                                 &primary->alloced_cpus_in_node)) {
 487                         cpumask_set_cpu(cur_cpu,
 488                                         &primary->alloced_cpus_in_node);
 489                         cpumask_set_cpu(cur_cpu, alloced_mask);
 490                         break;
 491                 }
 492         }
 493
 494         channel->target_cpu = cur_cpu;
 495         channel->target_vp = hv_context.vp_index[cur_cpu];
 496 }
 497
 498 /*
 499  * vmbus_unload_response - Handler for the unload response.
 500  */
 501 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
 502 {
 503         /*
 504          * This is a global event; just wakeup the waiting thread.
 505          * Once we successfully unload, we can cleanup the monitor state.
 506          */
 507         complete(&vmbus_connection.unload_event);
 508 }
 509
 510 void vmbus_initiate_unload(void)
 511 {
 512         struct vmbus_channel_message_header hdr;
 513
 514         /* Pre-Win2012R2 hosts don't support reconnect */
 515         if (vmbus_proto_version < VERSION_WIN8_1)
 516                 return;
 517
 518         init_completion(&vmbus_connection.unload_event);
 519         memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 520         hdr.msgtype = CHANNELMSG_UNLOAD;
 521         vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header));
 522
 523         wait_for_completion(&vmbus_connection.unload_event);
 524 }
 525
 526 /*
 527  * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
 528  *
 529  */
 530 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
 531 {
 532         struct vmbus_channel_offer_channel *offer;
 533         struct vmbus_channel *newchannel;
 534
 535         offer = (struct vmbus_channel_offer_channel *)hdr;
 536
 537         /* Allocate the channel object and save this offer. */
 538         newchannel = alloc_channel();
 539         if (!newchannel) {
 540                 pr_err("Unable to allocate channel object\n");
 541                 return;
 542         }
 543
 544         /*
 545          * By default we setup state to enable batched
 546          * reading. A specific service can choose to
 547          * disable this prior to opening the channel.
 548          */
 549         newchannel->batched_reading = true;
 550
 551         /*
 552          * Setup state for signalling the host.
 553          */
 554         newchannel->sig_event = (struct hv_input_signal_event *)
 555                                 (ALIGN((unsigned long)
 556                                 &newchannel->sig_buf,
 557                                 HV_HYPERCALL_PARAM_ALIGN));
 558
 559         newchannel->sig_event->connectionid.asu32 = 0;
 560         newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
 561         newchannel->sig_event->flag_number = 0;
 562         newchannel->sig_event->rsvdz = 0;
 563
 564         if (vmbus_proto_version != VERSION_WS2008) {
 565                 newchannel->is_dedicated_interrupt =
 566                                 (offer->is_dedicated_interrupt != 0);
 567                 newchannel->sig_event->connectionid.u.id =
 568                                 offer->connection_id;
 569         }
 570
 571         memcpy(&newchannel->offermsg, offer,
 572                sizeof(struct vmbus_channel_offer_channel));
 573         newchannel->monitor_grp = (u8)offer->monitorid / 32;
 574         newchannel->monitor_bit = (u8)offer->monitorid % 32;
 575
 576         vmbus_process_offer(newchannel);
 577 }
 578
 579 /*
 580  * vmbus_onoffer_rescind - Rescind offer handler.
 581  *
 582  * We queue a work item to process this offer synchronously
 583  */
 584 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 585 {
 586         struct vmbus_channel_rescind_offer *rescind;
 587         struct vmbus_channel *channel;
 588         unsigned long flags;
 589         struct device *dev;
 590
 591         rescind = (struct vmbus_channel_rescind_offer *)hdr;
 592         channel = relid2channel(rescind->child_relid);
 593
 594         if (channel == NULL) {
 595                 /*
 596                  * This is very impossible, because in
 597                  * vmbus_process_offer(), we have already invoked
 598                  * vmbus_release_relid() on error.
 599                  */
 600                 return;
 601         }
 602
 603         spin_lock_irqsave(&channel->lock, flags);
 604         channel->rescind = true;
 605         spin_unlock_irqrestore(&channel->lock, flags);
 606
 607         if (channel->device_obj) {
 608                 /*
 609                  * We will have to unregister this device from the
 610                  * driver core.
 611                  */
 612                 dev = get_device(&channel->device_obj->device);
 613                 if (dev) {
 614                         vmbus_device_unregister(channel->device_obj);
 615                         put_device(dev);
 616                 }
 617         } else {
 618                 hv_process_channel_removal(channel,
 619                         channel->offermsg.child_relid);
 620         }
 621 }
 622
 623 /*
 624  * vmbus_onoffers_delivered -
 625  * This is invoked when all offers have been delivered.
 626  *
 627  * Nothing to do here.
 628  */
 629 static void vmbus_onoffers_delivered(
 630                         struct vmbus_channel_message_header *hdr)
 631 {
 632 }
 633
 634 /*
 635  * vmbus_onopen_result - Open result handler.
 636  *
 637  * This is invoked when we received a response to our channel open request.
 638  * Find the matching request, copy the response and signal the requesting
 639  * thread.
 640  */
 641 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
 642 {
 643         struct vmbus_channel_open_result *result;
 644         struct vmbus_channel_msginfo *msginfo;
 645         struct vmbus_channel_message_header *requestheader;
 646         struct vmbus_channel_open_channel *openmsg;
 647         unsigned long flags;
 648
 649         result = (struct vmbus_channel_open_result *)hdr;
 650
 651         /*
 652          * Find the open msg, copy the result and signal/unblock the wait event
 653          */
 654         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 655
 656         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 657                                 msglistentry) {
 658                 requestheader =
 659                         (struct vmbus_channel_message_header *)msginfo->msg;
 660
 661                 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
 662                         openmsg =
 663                         (struct vmbus_channel_open_channel *)msginfo->msg;
 664                         if (openmsg->child_relid == result->child_relid &&
 665                             openmsg->openid == result->openid) {
 666                                 memcpy(&msginfo->response.open_result,
 667                                        result,
 668                                        sizeof(
 669                                         struct vmbus_channel_open_result));
 670                                 complete(&msginfo->waitevent);
 671                                 break;
 672                         }
 673                 }
 674         }
 675         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 676 }
 677
 678 /*
 679  * vmbus_ongpadl_created - GPADL created handler.
 680  *
 681  * This is invoked when we received a response to our gpadl create request.
 682  * Find the matching request, copy the response and signal the requesting
 683  * thread.
 684  */
 685 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
 686 {
 687         struct vmbus_channel_gpadl_created *gpadlcreated;
 688         struct vmbus_channel_msginfo *msginfo;
 689         struct vmbus_channel_message_header *requestheader;
 690         struct vmbus_channel_gpadl_header *gpadlheader;
 691         unsigned long flags;
 692
 693         gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
 694
 695         /*
 696          * Find the establish msg, copy the result and signal/unblock the wait
 697          * event
 698          */
 699         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 700
 701         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 702                                 msglistentry) {
 703                 requestheader =
 704                         (struct vmbus_channel_message_header *)msginfo->msg;
 705
 706                 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
 707                         gpadlheader =
 708                         (struct vmbus_channel_gpadl_header *)requestheader;
 709
 710                         if ((gpadlcreated->child_relid ==
 711                              gpadlheader->child_relid) &&
 712                             (gpadlcreated->gpadl == gpadlheader->gpadl)) {
 713                                 memcpy(&msginfo->response.gpadl_created,
 714                                        gpadlcreated,
 715                                        sizeof(
 716                                         struct vmbus_channel_gpadl_created));
 717                                 complete(&msginfo->waitevent);
 718                                 break;
 719                         }
 720                 }
 721         }
 722         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 723 }
 724
 725 /*
 726  * vmbus_ongpadl_torndown - GPADL torndown handler.
 727  *
 728  * This is invoked when we received a response to our gpadl teardown request.
 729  * Find the matching request, copy the response and signal the requesting
 730  * thread.
 731  */
 732 static void vmbus_ongpadl_torndown(
 733                         struct vmbus_channel_message_header *hdr)
 734 {
 735         struct vmbus_channel_gpadl_torndown *gpadl_torndown;
 736         struct vmbus_channel_msginfo *msginfo;
 737         struct vmbus_channel_message_header *requestheader;
 738         struct vmbus_channel_gpadl_teardown *gpadl_teardown;
 739         unsigned long flags;
 740
 741         gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
 742
 743         /*
 744          * Find the open msg, copy the result and signal/unblock the wait event
 745          */
 746         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 747
 748         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 749                                 msglistentry) {
 750                 requestheader =
 751                         (struct vmbus_channel_message_header *)msginfo->msg;
 752
 753                 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
 754                         gpadl_teardown =
 755                         (struct vmbus_channel_gpadl_teardown *)requestheader;
 756
 757                         if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
 758                                 memcpy(&msginfo->response.gpadl_torndown,
 759                                        gpadl_torndown,
 760                                        sizeof(
 761                                         struct vmbus_channel_gpadl_torndown));
 762                                 complete(&msginfo->waitevent);
 763                                 break;
 764                         }
 765                 }
 766         }
 767         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 768 }
 769
 770 /*
 771  * vmbus_onversion_response - Version response handler
 772  *
 773  * This is invoked when we received a response to our initiate contact request.
 774  * Find the matching request, copy the response and signal the requesting
 775  * thread.
 776  */
 777 static void vmbus_onversion_response(
 778                 struct vmbus_channel_message_header *hdr)
 779 {
 780         struct vmbus_channel_msginfo *msginfo;
 781         struct vmbus_channel_message_header *requestheader;
 782         struct vmbus_channel_version_response *version_response;
 783         unsigned long flags;
 784
 785         version_response = (struct vmbus_channel_version_response *)hdr;
 786         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 787
 788         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 789                                 msglistentry) {
 790                 requestheader =
 791                         (struct vmbus_channel_message_header *)msginfo->msg;
 792
 793                 if (requestheader->msgtype ==
 794                     CHANNELMSG_INITIATE_CONTACT) {
 795                         memcpy(&msginfo->response.version_response,
 796                               version_response,
 797                               sizeof(struct vmbus_channel_version_response));
 798                         complete(&msginfo->waitevent);
 799                 }
 800         }
 801         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 802 }
 803
 804 /* Channel message dispatch table */
 805 struct vmbus_channel_message_table_entry
 806         channel_message_table[CHANNELMSG_COUNT] = {
 807         {CHANNELMSG_INVALID,                    0, NULL},
 808         {CHANNELMSG_OFFERCHANNEL,               0, vmbus_onoffer},
 809         {CHANNELMSG_RESCIND_CHANNELOFFER,       0, vmbus_onoffer_rescind},
 810         {CHANNELMSG_REQUESTOFFERS,              0, NULL},
 811         {CHANNELMSG_ALLOFFERS_DELIVERED,        1, vmbus_onoffers_delivered},
 812         {CHANNELMSG_OPENCHANNEL,                0, NULL},
 813         {CHANNELMSG_OPENCHANNEL_RESULT,         1, vmbus_onopen_result},
 814         {CHANNELMSG_CLOSECHANNEL,               0, NULL},
 815         {CHANNELMSG_GPADL_HEADER,               0, NULL},
 816         {CHANNELMSG_GPADL_BODY,                 0, NULL},
 817         {CHANNELMSG_GPADL_CREATED,              1, vmbus_ongpadl_created},
 818         {CHANNELMSG_GPADL_TEARDOWN,             0, NULL},
 819         {CHANNELMSG_GPADL_TORNDOWN,             1, vmbus_ongpadl_torndown},
 820         {CHANNELMSG_RELID_RELEASED,             0, NULL},
 821         {CHANNELMSG_INITIATE_CONTACT,           0, NULL},
 822         {CHANNELMSG_VERSION_RESPONSE,           1, vmbus_onversion_response},
 823         {CHANNELMSG_UNLOAD,                     0, NULL},
 824         {CHANNELMSG_UNLOAD_RESPONSE,            1, vmbus_unload_response},
 825 };
 826
 827 /*
 828  * vmbus_onmessage - Handler for channel protocol messages.
 829  *
 830  * This is invoked in the vmbus worker thread context.
 831  */
 832 void vmbus_onmessage(void *context)
 833 {
 834         struct hv_message *msg = context;
 835         struct vmbus_channel_message_header *hdr;
 836         int size;
 837
 838         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 839         size = msg->header.payload_size;
 840
 841         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 842                 pr_err("Received invalid channel message type %d size %d\n",
 843                            hdr->msgtype, size);
 844                 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
 845                                      (unsigned char *)msg->u.payload, size);
 846                 return;
 847         }
 848
 849         if (channel_message_table[hdr->msgtype].message_handler)
 850                 channel_message_table[hdr->msgtype].message_handler(hdr);
 851         else
 852                 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
 853 }
 854
 855 /*
 856  * vmbus_request_offers - Send a request to get all our pending offers.
 857  */
 858 int vmbus_request_offers(void)
 859 {
 860         struct vmbus_channel_message_header *msg;
 861         struct vmbus_channel_msginfo *msginfo;
 862         int ret;
 863
 864         msginfo = kmalloc(sizeof(*msginfo) +
 865                           sizeof(struct vmbus_channel_message_header),
 866                           GFP_KERNEL);
 867         if (!msginfo)
 868                 return -ENOMEM;
 869
 870         msg = (struct vmbus_channel_message_header *)msginfo->msg;
 871
 872         msg->msgtype = CHANNELMSG_REQUESTOFFERS;
 873
 874
 875         ret = vmbus_post_msg(msg,
 876                                sizeof(struct vmbus_channel_message_header));
 877         if (ret != 0) {
 878                 pr_err("Unable to request offers - %d\n", ret);
 879
 880                 goto cleanup;
 881         }
 882
 883 cleanup:
 884         kfree(msginfo);
 885
 886         return ret;
 887 }
 888
 889 /*
 890  * Retrieve the (sub) channel on which to send an outgoing request.
 891  * When a primary channel has multiple sub-channels, we try to
 892  * distribute the load equally amongst all available channels.
 893  */
 894 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
 895 {
 896         struct list_head *cur, *tmp;
 897         int cur_cpu;
 898         struct vmbus_channel *cur_channel;
 899         struct vmbus_channel *outgoing_channel = primary;
 900         int next_channel;
 901         int i = 1;
 902
 903         if (list_empty(&primary->sc_list))
 904                 return outgoing_channel;
 905
 906         next_channel = primary->next_oc++;
 907
 908         if (next_channel > (primary->num_sc)) {
 909                 primary->next_oc = 0;
 910                 return outgoing_channel;
 911         }
 912
 913         cur_cpu = hv_context.vp_index[get_cpu()];
 914         put_cpu();
 915         list_for_each_safe(cur, tmp, &primary->sc_list) {
 916                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
 917                 if (cur_channel->state != CHANNEL_OPENED_STATE)
 918                         continue;
 919
 920                 if (cur_channel->target_vp == cur_cpu)
 921                         return cur_channel;
 922
 923                 if (i == next_channel)
 924                         return cur_channel;
 925
 926                 i++;
 927         }
 928
 929         return outgoing_channel;
 930 }
 931 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
 932
 933 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
 934 {
 935         struct list_head *cur, *tmp;
 936         struct vmbus_channel *cur_channel;
 937
 938         if (primary_channel->sc_creation_callback == NULL)
 939                 return;
 940
 941         list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
 942                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
 943
 944                 primary_channel->sc_creation_callback(cur_channel);
 945         }
 946 }
 947
 948 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
 949                                 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
 950 {
 951         primary_channel->sc_creation_callback = sc_cr_cb;
 952 }
 953 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
 954
 955 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
 956 {
 957         bool ret;
 958
 959         ret = !list_empty(&primary->sc_list);
 960
 961         if (ret) {
 962                 /*
 963                  * Invoke the callback on sub-channel creation.
 964                  * This will present a uniform interface to the
 965                  * clients.
 966                  */
 967                 invoke_sc_cb(primary);
 968         }
 969
 970         return ret;
 971 }
 972 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);