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         spin_lock_init(&channel->inbound_lock);
 263         spin_lock_init(&channel->lock);
 264
 265         INIT_LIST_HEAD(&channel->sc_list);
 266         INIT_LIST_HEAD(&channel->percpu_list);
 267
 268         return channel;
 269 }
 270
 271 /*
 272  * free_channel - Release the resources used by the vmbus channel object
 273  */
 274 static void free_channel(struct vmbus_channel *channel)
 275 {
 276         kfree(channel);
 277 }
 278
 279 static void percpu_channel_enq(void *arg)
 280 {
 281         struct vmbus_channel *channel = arg;
 282         int cpu = smp_processor_id();
 283
 284         list_add_tail(&channel->percpu_list, &hv_context.percpu_list[cpu]);
 285 }
 286
 287 static void percpu_channel_deq(void *arg)
 288 {
 289         struct vmbus_channel *channel = arg;
 290
 291         list_del(&channel->percpu_list);
 292 }
 293
 294
 295 static void vmbus_release_relid(u32 relid)
 296 {
 297         struct vmbus_channel_relid_released msg;
 298
 299         memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
 300         msg.child_relid = relid;
 301         msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
 302         vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released));
 303 }
 304
 305 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
 306 {
 307         unsigned long flags;
 308         struct vmbus_channel *primary_channel;
 309
 310         vmbus_release_relid(relid);
 311
 312         BUG_ON(!channel->rescind);
 313
 314         if (channel->target_cpu != get_cpu()) {
 315                 put_cpu();
 316                 smp_call_function_single(channel->target_cpu,
 317                                          percpu_channel_deq, channel, true);
 318         } else {
 319                 percpu_channel_deq(channel);
 320                 put_cpu();
 321         }
 322
 323         if (channel->primary_channel == NULL) {
 324                 mutex_lock(&vmbus_connection.channel_mutex);
 325                 list_del(&channel->listentry);
 326                 mutex_unlock(&vmbus_connection.channel_mutex);
 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
 371         /* Make sure this is a new offer */
 372         mutex_lock(&vmbus_connection.channel_mutex);
 373
 374         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 375                 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
 376                         newchannel->offermsg.offer.if_type) &&
 377                         !uuid_le_cmp(channel->offermsg.offer.if_instance,
 378                                 newchannel->offermsg.offer.if_instance)) {
 379                         fnew = false;
 380                         break;
 381                 }
 382         }
 383
 384         if (fnew)
 385                 list_add_tail(&newchannel->listentry,
 386                               &vmbus_connection.chn_list);
 387
 388         mutex_unlock(&vmbus_connection.channel_mutex);
 389
 390         if (!fnew) {
 391                 /*
 392                  * Check to see if this is a sub-channel.
 393                  */
 394                 if (newchannel->offermsg.offer.sub_channel_index != 0) {
 395                         /*
 396                          * Process the sub-channel.
 397                          */
 398                         newchannel->primary_channel = channel;
 399                         spin_lock_irqsave(&channel->lock, flags);
 400                         list_add_tail(&newchannel->sc_list, &channel->sc_list);
 401                         channel->num_sc++;
 402                         spin_unlock_irqrestore(&channel->lock, flags);
 403                 } else
 404                         goto err_free_chan;
 405         }
 406
 407         dev_type = hv_get_dev_type(&newchannel->offermsg.offer.if_type);
 408
 409         init_vp_index(newchannel, dev_type);
 410
 411         if (newchannel->target_cpu != get_cpu()) {
 412                 put_cpu();
 413                 smp_call_function_single(newchannel->target_cpu,
 414                                          percpu_channel_enq,
 415                                          newchannel, true);
 416         } else {
 417                 percpu_channel_enq(newchannel);
 418                 put_cpu();
 419         }
 420
 421         /*
 422          * This state is used to indicate a successful open
 423          * so that when we do close the channel normally, we
 424          * can cleanup properly
 425          */
 426         newchannel->state = CHANNEL_OPEN_STATE;
 427
 428         if (!fnew) {
 429                 if (channel->sc_creation_callback != NULL)
 430                         channel->sc_creation_callback(newchannel);
 431                 return;
 432         }
 433
 434         /*
 435          * Start the process of binding this offer to the driver
 436          * We need to set the DeviceObject field before calling
 437          * vmbus_child_dev_add()
 438          */
 439         newchannel->device_obj = vmbus_device_create(
 440                 &newchannel->offermsg.offer.if_type,
 441                 &newchannel->offermsg.offer.if_instance,
 442                 newchannel);
 443         if (!newchannel->device_obj)
 444                 goto err_deq_chan;
 445
 446         newchannel->device_obj->device_id = dev_type;
 447         /*
 448          * Add the new device to the bus. This will kick off device-driver
 449          * binding which eventually invokes the device driver's AddDevice()
 450          * method.
 451          */
 452         if (vmbus_device_register(newchannel->device_obj) != 0) {
 453                 pr_err("unable to add child device object (relid %d)\n",
 454                         newchannel->offermsg.child_relid);
 455                 kfree(newchannel->device_obj);
 456                 goto err_deq_chan;
 457         }
 458         return;
 459
 460 err_deq_chan:
 461         vmbus_release_relid(newchannel->offermsg.child_relid);
 462
 463         mutex_lock(&vmbus_connection.channel_mutex);
 464         list_del(&newchannel->listentry);
 465         mutex_unlock(&vmbus_connection.channel_mutex);
 466
 467         if (newchannel->target_cpu != get_cpu()) {
 468                 put_cpu();
 469                 smp_call_function_single(newchannel->target_cpu,
 470                                          percpu_channel_deq, newchannel, true);
 471         } else {
 472                 percpu_channel_deq(newchannel);
 473                 put_cpu();
 474         }
 475
 476 err_free_chan:
 477         free_channel(newchannel);
 478 }
 479
 480 /*
 481  * We use this state to statically distribute the channel interrupt load.
 482  */
 483 static int next_numa_node_id;
 484
 485 /*
 486  * Starting with Win8, we can statically distribute the incoming
 487  * channel interrupt load by binding a channel to VCPU.
 488  * We do this in a hierarchical fashion:
 489  * First distribute the primary channels across available NUMA nodes
 490  * and then distribute the subchannels amongst the CPUs in the NUMA
 491  * node assigned to the primary channel.
 492  *
 493  * For pre-win8 hosts or non-performance critical channels we assign the
 494  * first CPU in the first NUMA node.
 495  */
 496 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
 497 {
 498         u32 cur_cpu;
 499         bool perf_chn = vmbus_devs[dev_type].perf_device;
 500         struct vmbus_channel *primary = channel->primary_channel;
 501         int next_node;
 502         struct cpumask available_mask;
 503         struct cpumask *alloced_mask;
 504
 505         if ((vmbus_proto_version == VERSION_WS2008) ||
 506             (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
 507                 /*
 508                  * Prior to win8, all channel interrupts are
 509                  * delivered on cpu 0.
 510                  * Also if the channel is not a performance critical
 511                  * channel, bind it to cpu 0.
 512                  */
 513                 channel->numa_node = 0;
 514                 channel->target_cpu = 0;
 515                 channel->target_vp = hv_context.vp_index[0];
 516                 return;
 517         }
 518
 519         /*
 520          * We distribute primary channels evenly across all the available
 521          * NUMA nodes and within the assigned NUMA node we will assign the
 522          * first available CPU to the primary channel.
 523          * The sub-channels will be assigned to the CPUs available in the
 524          * NUMA node evenly.
 525          */
 526         if (!primary) {
 527                 while (true) {
 528                         next_node = next_numa_node_id++;
 529                         if (next_node == nr_node_ids)
 530                                 next_node = next_numa_node_id = 0;
 531                         if (cpumask_empty(cpumask_of_node(next_node)))
 532                                 continue;
 533                         break;
 534                 }
 535                 channel->numa_node = next_node;
 536                 primary = channel;
 537         }
 538         alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
 539
 540         if (cpumask_weight(alloced_mask) ==
 541             cpumask_weight(cpumask_of_node(primary->numa_node))) {
 542                 /*
 543                  * We have cycled through all the CPUs in the node;
 544                  * reset the alloced map.
 545                  */
 546                 cpumask_clear(alloced_mask);
 547         }
 548
 549         cpumask_xor(&available_mask, alloced_mask,
 550                     cpumask_of_node(primary->numa_node));
 551
 552         cur_cpu = -1;
 553
 554         /*
 555          * Normally Hyper-V host doesn't create more subchannels than there
 556          * are VCPUs on the node but it is possible when not all present VCPUs
 557          * on the node are initialized by guest. Clear the alloced_cpus_in_node
 558          * to start over.
 559          */
 560         if (cpumask_equal(&primary->alloced_cpus_in_node,
 561                           cpumask_of_node(primary->numa_node)))
 562                 cpumask_clear(&primary->alloced_cpus_in_node);
 563
 564         while (true) {
 565                 cur_cpu = cpumask_next(cur_cpu, &available_mask);
 566                 if (cur_cpu >= nr_cpu_ids) {
 567                         cur_cpu = -1;
 568                         cpumask_copy(&available_mask,
 569                                      cpumask_of_node(primary->numa_node));
 570                         continue;
 571                 }
 572
 573                 /*
 574                  * NOTE: in the case of sub-channel, we clear the sub-channel
 575                  * related bit(s) in primary->alloced_cpus_in_node in
 576                  * hv_process_channel_removal(), so when we reload drivers
 577                  * like hv_netvsc in SMP guest, here we're able to re-allocate
 578                  * bit from primary->alloced_cpus_in_node.
 579                  */
 580                 if (!cpumask_test_cpu(cur_cpu,
 581                                 &primary->alloced_cpus_in_node)) {
 582                         cpumask_set_cpu(cur_cpu,
 583                                         &primary->alloced_cpus_in_node);
 584                         cpumask_set_cpu(cur_cpu, alloced_mask);
 585                         break;
 586                 }
 587         }
 588
 589         channel->target_cpu = cur_cpu;
 590         channel->target_vp = hv_context.vp_index[cur_cpu];
 591 }
 592
 593 static void vmbus_wait_for_unload(void)
 594 {
 595         int cpu = smp_processor_id();
 596         void *page_addr = hv_context.synic_message_page[cpu];
 597         struct hv_message *msg = (struct hv_message *)page_addr +
 598                                   VMBUS_MESSAGE_SINT;
 599         struct vmbus_channel_message_header *hdr;
 600         bool unloaded = false;
 601
 602         while (1) {
 603                 if (msg->header.message_type == HVMSG_NONE) {
 604                         mdelay(10);
 605                         continue;
 606                 }
 607
 608                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 609                 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
 610                         unloaded = true;
 611
 612                 msg->header.message_type = HVMSG_NONE;
 613                 /*
 614                  * header.message_type needs to be written before we do
 615                  * wrmsrl() below.
 616                  */
 617                 mb();
 618
 619                 if (msg->header.message_flags.msg_pending)
 620                         wrmsrl(HV_X64_MSR_EOM, 0);
 621
 622                 if (unloaded)
 623                         break;
 624         }
 625 }
 626
 627 /*
 628  * vmbus_unload_response - Handler for the unload response.
 629  */
 630 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
 631 {
 632         /*
 633          * This is a global event; just wakeup the waiting thread.
 634          * Once we successfully unload, we can cleanup the monitor state.
 635          */
 636         complete(&vmbus_connection.unload_event);
 637 }
 638
 639 void vmbus_initiate_unload(void)
 640 {
 641         struct vmbus_channel_message_header hdr;
 642
 643         /* Pre-Win2012R2 hosts don't support reconnect */
 644         if (vmbus_proto_version < VERSION_WIN8_1)
 645                 return;
 646
 647         init_completion(&vmbus_connection.unload_event);
 648         memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 649         hdr.msgtype = CHANNELMSG_UNLOAD;
 650         vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header));
 651
 652         /*
 653          * vmbus_initiate_unload() is also called on crash and the crash can be
 654          * happening in an interrupt context, where scheduling is impossible.
 655          */
 656         if (!in_interrupt())
 657                 wait_for_completion(&vmbus_connection.unload_event);
 658         else
 659                 vmbus_wait_for_unload();
 660 }
 661
 662 /*
 663  * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
 664  *
 665  */
 666 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
 667 {
 668         struct vmbus_channel_offer_channel *offer;
 669         struct vmbus_channel *newchannel;
 670
 671         offer = (struct vmbus_channel_offer_channel *)hdr;
 672
 673         /* Allocate the channel object and save this offer. */
 674         newchannel = alloc_channel();
 675         if (!newchannel) {
 676                 pr_err("Unable to allocate channel object\n");
 677                 return;
 678         }
 679
 680         /*
 681          * By default we setup state to enable batched
 682          * reading. A specific service can choose to
 683          * disable this prior to opening the channel.
 684          */
 685         newchannel->batched_reading = true;
 686
 687         /*
 688          * Setup state for signalling the host.
 689          */
 690         newchannel->sig_event = (struct hv_input_signal_event *)
 691                                 (ALIGN((unsigned long)
 692                                 &newchannel->sig_buf,
 693                                 HV_HYPERCALL_PARAM_ALIGN));
 694
 695         newchannel->sig_event->connectionid.asu32 = 0;
 696         newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
 697         newchannel->sig_event->flag_number = 0;
 698         newchannel->sig_event->rsvdz = 0;
 699
 700         if (vmbus_proto_version != VERSION_WS2008) {
 701                 newchannel->is_dedicated_interrupt =
 702                                 (offer->is_dedicated_interrupt != 0);
 703                 newchannel->sig_event->connectionid.u.id =
 704                                 offer->connection_id;
 705         }
 706
 707         memcpy(&newchannel->offermsg, offer,
 708                sizeof(struct vmbus_channel_offer_channel));
 709         newchannel->monitor_grp = (u8)offer->monitorid / 32;
 710         newchannel->monitor_bit = (u8)offer->monitorid % 32;
 711
 712         vmbus_process_offer(newchannel);
 713 }
 714
 715 /*
 716  * vmbus_onoffer_rescind - Rescind offer handler.
 717  *
 718  * We queue a work item to process this offer synchronously
 719  */
 720 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 721 {
 722         struct vmbus_channel_rescind_offer *rescind;
 723         struct vmbus_channel *channel;
 724         unsigned long flags;
 725         struct device *dev;
 726
 727         rescind = (struct vmbus_channel_rescind_offer *)hdr;
 728         channel = relid2channel(rescind->child_relid);
 729
 730         if (channel == NULL) {
 731                 /*
 732                  * This is very impossible, because in
 733                  * vmbus_process_offer(), we have already invoked
 734                  * vmbus_release_relid() on error.
 735                  */
 736                 return;
 737         }
 738
 739         spin_lock_irqsave(&channel->lock, flags);
 740         channel->rescind = true;
 741         spin_unlock_irqrestore(&channel->lock, flags);
 742
 743         if (channel->device_obj) {
 744                 /*
 745                  * We will have to unregister this device from the
 746                  * driver core.
 747                  */
 748                 dev = get_device(&channel->device_obj->device);
 749                 if (dev) {
 750                         vmbus_device_unregister(channel->device_obj);
 751                         put_device(dev);
 752                 }
 753         } else {
 754                 hv_process_channel_removal(channel,
 755                         channel->offermsg.child_relid);
 756         }
 757 }
 758
 759 /*
 760  * vmbus_onoffers_delivered -
 761  * This is invoked when all offers have been delivered.
 762  *
 763  * Nothing to do here.
 764  */
 765 static void vmbus_onoffers_delivered(
 766                         struct vmbus_channel_message_header *hdr)
 767 {
 768 }
 769
 770 /*
 771  * vmbus_onopen_result - Open result handler.
 772  *
 773  * This is invoked when we received a response to our channel open request.
 774  * Find the matching request, copy the response and signal the requesting
 775  * thread.
 776  */
 777 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
 778 {
 779         struct vmbus_channel_open_result *result;
 780         struct vmbus_channel_msginfo *msginfo;
 781         struct vmbus_channel_message_header *requestheader;
 782         struct vmbus_channel_open_channel *openmsg;
 783         unsigned long flags;
 784
 785         result = (struct vmbus_channel_open_result *)hdr;
 786
 787         /*
 788          * Find the open msg, copy the result and signal/unblock the wait event
 789          */
 790         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 791
 792         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 793                                 msglistentry) {
 794                 requestheader =
 795                         (struct vmbus_channel_message_header *)msginfo->msg;
 796
 797                 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
 798                         openmsg =
 799                         (struct vmbus_channel_open_channel *)msginfo->msg;
 800                         if (openmsg->child_relid == result->child_relid &&
 801                             openmsg->openid == result->openid) {
 802                                 memcpy(&msginfo->response.open_result,
 803                                        result,
 804                                        sizeof(
 805                                         struct vmbus_channel_open_result));
 806                                 complete(&msginfo->waitevent);
 807                                 break;
 808                         }
 809                 }
 810         }
 811         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 812 }
 813
 814 /*
 815  * vmbus_ongpadl_created - GPADL created handler.
 816  *
 817  * This is invoked when we received a response to our gpadl create request.
 818  * Find the matching request, copy the response and signal the requesting
 819  * thread.
 820  */
 821 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
 822 {
 823         struct vmbus_channel_gpadl_created *gpadlcreated;
 824         struct vmbus_channel_msginfo *msginfo;
 825         struct vmbus_channel_message_header *requestheader;
 826         struct vmbus_channel_gpadl_header *gpadlheader;
 827         unsigned long flags;
 828
 829         gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
 830
 831         /*
 832          * Find the establish msg, copy the result and signal/unblock the wait
 833          * event
 834          */
 835         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 836
 837         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 838                                 msglistentry) {
 839                 requestheader =
 840                         (struct vmbus_channel_message_header *)msginfo->msg;
 841
 842                 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
 843                         gpadlheader =
 844                         (struct vmbus_channel_gpadl_header *)requestheader;
 845
 846                         if ((gpadlcreated->child_relid ==
 847                              gpadlheader->child_relid) &&
 848                             (gpadlcreated->gpadl == gpadlheader->gpadl)) {
 849                                 memcpy(&msginfo->response.gpadl_created,
 850                                        gpadlcreated,
 851                                        sizeof(
 852                                         struct vmbus_channel_gpadl_created));
 853                                 complete(&msginfo->waitevent);
 854                                 break;
 855                         }
 856                 }
 857         }
 858         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 859 }
 860
 861 /*
 862  * vmbus_ongpadl_torndown - GPADL torndown handler.
 863  *
 864  * This is invoked when we received a response to our gpadl teardown request.
 865  * Find the matching request, copy the response and signal the requesting
 866  * thread.
 867  */
 868 static void vmbus_ongpadl_torndown(
 869                         struct vmbus_channel_message_header *hdr)
 870 {
 871         struct vmbus_channel_gpadl_torndown *gpadl_torndown;
 872         struct vmbus_channel_msginfo *msginfo;
 873         struct vmbus_channel_message_header *requestheader;
 874         struct vmbus_channel_gpadl_teardown *gpadl_teardown;
 875         unsigned long flags;
 876
 877         gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
 878
 879         /*
 880          * Find the open msg, copy the result and signal/unblock the wait event
 881          */
 882         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 883
 884         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 885                                 msglistentry) {
 886                 requestheader =
 887                         (struct vmbus_channel_message_header *)msginfo->msg;
 888
 889                 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
 890                         gpadl_teardown =
 891                         (struct vmbus_channel_gpadl_teardown *)requestheader;
 892
 893                         if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
 894                                 memcpy(&msginfo->response.gpadl_torndown,
 895                                        gpadl_torndown,
 896                                        sizeof(
 897                                         struct vmbus_channel_gpadl_torndown));
 898                                 complete(&msginfo->waitevent);
 899                                 break;
 900                         }
 901                 }
 902         }
 903         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 904 }
 905
 906 /*
 907  * vmbus_onversion_response - Version response handler
 908  *
 909  * This is invoked when we received a response to our initiate contact request.
 910  * Find the matching request, copy the response and signal the requesting
 911  * thread.
 912  */
 913 static void vmbus_onversion_response(
 914                 struct vmbus_channel_message_header *hdr)
 915 {
 916         struct vmbus_channel_msginfo *msginfo;
 917         struct vmbus_channel_message_header *requestheader;
 918         struct vmbus_channel_version_response *version_response;
 919         unsigned long flags;
 920
 921         version_response = (struct vmbus_channel_version_response *)hdr;
 922         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 923
 924         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 925                                 msglistentry) {
 926                 requestheader =
 927                         (struct vmbus_channel_message_header *)msginfo->msg;
 928
 929                 if (requestheader->msgtype ==
 930                     CHANNELMSG_INITIATE_CONTACT) {
 931                         memcpy(&msginfo->response.version_response,
 932                               version_response,
 933                               sizeof(struct vmbus_channel_version_response));
 934                         complete(&msginfo->waitevent);
 935                 }
 936         }
 937         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 938 }
 939
 940 /* Channel message dispatch table */
 941 struct vmbus_channel_message_table_entry
 942         channel_message_table[CHANNELMSG_COUNT] = {
 943         {CHANNELMSG_INVALID,                    0, NULL},
 944         {CHANNELMSG_OFFERCHANNEL,               0, vmbus_onoffer},
 945         {CHANNELMSG_RESCIND_CHANNELOFFER,       0, vmbus_onoffer_rescind},
 946         {CHANNELMSG_REQUESTOFFERS,              0, NULL},
 947         {CHANNELMSG_ALLOFFERS_DELIVERED,        1, vmbus_onoffers_delivered},
 948         {CHANNELMSG_OPENCHANNEL,                0, NULL},
 949         {CHANNELMSG_OPENCHANNEL_RESULT,         1, vmbus_onopen_result},
 950         {CHANNELMSG_CLOSECHANNEL,               0, NULL},
 951         {CHANNELMSG_GPADL_HEADER,               0, NULL},
 952         {CHANNELMSG_GPADL_BODY,                 0, NULL},
 953         {CHANNELMSG_GPADL_CREATED,              1, vmbus_ongpadl_created},
 954         {CHANNELMSG_GPADL_TEARDOWN,             0, NULL},
 955         {CHANNELMSG_GPADL_TORNDOWN,             1, vmbus_ongpadl_torndown},
 956         {CHANNELMSG_RELID_RELEASED,             0, NULL},
 957         {CHANNELMSG_INITIATE_CONTACT,           0, NULL},
 958         {CHANNELMSG_VERSION_RESPONSE,           1, vmbus_onversion_response},
 959         {CHANNELMSG_UNLOAD,                     0, NULL},
 960         {CHANNELMSG_UNLOAD_RESPONSE,            1, vmbus_unload_response},
 961         {CHANNELMSG_18,                         0, NULL},
 962         {CHANNELMSG_19,                         0, NULL},
 963         {CHANNELMSG_20,                         0, NULL},
 964         {CHANNELMSG_TL_CONNECT_REQUEST,         0, NULL},
 965 };
 966
 967 /*
 968  * vmbus_onmessage - Handler for channel protocol messages.
 969  *
 970  * This is invoked in the vmbus worker thread context.
 971  */
 972 void vmbus_onmessage(void *context)
 973 {
 974         struct hv_message *msg = context;
 975         struct vmbus_channel_message_header *hdr;
 976         int size;
 977
 978         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 979         size = msg->header.payload_size;
 980
 981         if (hdr->msgtype >= CHANNELMSG_COUNT) {
 982                 pr_err("Received invalid channel message type %d size %d\n",
 983                            hdr->msgtype, size);
 984                 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
 985                                      (unsigned char *)msg->u.payload, size);
 986                 return;
 987         }
 988
 989         if (channel_message_table[hdr->msgtype].message_handler)
 990                 channel_message_table[hdr->msgtype].message_handler(hdr);
 991         else
 992                 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
 993 }
 994
 995 /*
 996  * vmbus_request_offers - Send a request to get all our pending offers.
 997  */
 998 int vmbus_request_offers(void)
 999 {
1000         struct vmbus_channel_message_header *msg;
1001         struct vmbus_channel_msginfo *msginfo;
1002         int ret;
1003
1004         msginfo = kmalloc(sizeof(*msginfo) +
1005                           sizeof(struct vmbus_channel_message_header),
1006                           GFP_KERNEL);
1007         if (!msginfo)
1008                 return -ENOMEM;
1009
1010         msg = (struct vmbus_channel_message_header *)msginfo->msg;
1011
1012         msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1013
1014
1015         ret = vmbus_post_msg(msg,
1016                                sizeof(struct vmbus_channel_message_header));
1017         if (ret != 0) {
1018                 pr_err("Unable to request offers - %d\n", ret);
1019
1020                 goto cleanup;
1021         }
1022
1023 cleanup:
1024         kfree(msginfo);
1025
1026         return ret;
1027 }
1028
1029 /*
1030  * Retrieve the (sub) channel on which to send an outgoing request.
1031  * When a primary channel has multiple sub-channels, we try to
1032  * distribute the load equally amongst all available channels.
1033  */
1034 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1035 {
1036         struct list_head *cur, *tmp;
1037         int cur_cpu;
1038         struct vmbus_channel *cur_channel;
1039         struct vmbus_channel *outgoing_channel = primary;
1040         int next_channel;
1041         int i = 1;
1042
1043         if (list_empty(&primary->sc_list))
1044                 return outgoing_channel;
1045
1046         next_channel = primary->next_oc++;
1047
1048         if (next_channel > (primary->num_sc)) {
1049                 primary->next_oc = 0;
1050                 return outgoing_channel;
1051         }
1052
1053         cur_cpu = hv_context.vp_index[get_cpu()];
1054         put_cpu();
1055         list_for_each_safe(cur, tmp, &primary->sc_list) {
1056                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1057                 if (cur_channel->state != CHANNEL_OPENED_STATE)
1058                         continue;
1059
1060                 if (cur_channel->target_vp == cur_cpu)
1061                         return cur_channel;
1062
1063                 if (i == next_channel)
1064                         return cur_channel;
1065
1066                 i++;
1067         }
1068
1069         return outgoing_channel;
1070 }
1071 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1072
1073 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1074 {
1075         struct list_head *cur, *tmp;
1076         struct vmbus_channel *cur_channel;
1077
1078         if (primary_channel->sc_creation_callback == NULL)
1079                 return;
1080
1081         list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1082                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1083
1084                 primary_channel->sc_creation_callback(cur_channel);
1085         }
1086 }
1087
1088 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1089                                 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1090 {
1091         primary_channel->sc_creation_callback = sc_cr_cb;
1092 }
1093 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1094
1095 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1096 {
1097         bool ret;
1098
1099         ret = !list_empty(&primary->sc_list);
1100
1101         if (ret) {
1102                 /*
1103                  * Invoke the callback on sub-channel creation.
1104                  * This will present a uniform interface to the
1105                  * clients.
1106                  */
1107                 invoke_sc_cb(primary);
1108         }
1109
1110         return ret;
1111 }
1112 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);