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 <[email protected]>
  19  *   Hank Janssen  <[email protected]>
  20  */
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/kernel.h>
  24 #include <linux/sched.h>
  25 #include <linux/wait.h>
  26 #include <linux/mm.h>
  27 #include <linux/slab.h>
  28 #include <linux/list.h>
  29 #include <linux/module.h>
  30 #include <linux/completion.h>
  31 #include <linux/delay.h>
  32 #include <linux/hyperv.h>
  33
  34 #include "hyperv_vmbus.h"
  35
  36 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type);
  37
  38 static const struct vmbus_device vmbus_devs[] = {
  39         /* IDE */
  40         { .dev_type = HV_IDE,
  41           HV_IDE_GUID,
  42           .perf_device = true,
  43         },
  44
  45         /* SCSI */
  46         { .dev_type = HV_SCSI,
  47           HV_SCSI_GUID,
  48           .perf_device = true,
  49         },
  50
  51         /* Fibre Channel */
  52         { .dev_type = HV_FC,
  53           HV_SYNTHFC_GUID,
  54           .perf_device = true,
  55         },
  56
  57         /* Synthetic NIC */
  58         { .dev_type = HV_NIC,
  59           HV_NIC_GUID,
  60           .perf_device = true,
  61         },
  62
  63         /* Network Direct */
  64         { .dev_type = HV_ND,
  65           HV_ND_GUID,
  66           .perf_device = true,
  67         },
  68
  69         /* PCIE */
  70         { .dev_type = HV_PCIE,
  71           HV_PCIE_GUID,
  72           .perf_device = true,
  73         },
  74
  75         /* Synthetic Frame Buffer */
  76         { .dev_type = HV_FB,
  77           HV_SYNTHVID_GUID,
  78           .perf_device = false,
  79         },
  80
  81         /* Synthetic Keyboard */
  82         { .dev_type = HV_KBD,
  83           HV_KBD_GUID,
  84           .perf_device = false,
  85         },
  86
  87         /* Synthetic MOUSE */
  88         { .dev_type = HV_MOUSE,
  89           HV_MOUSE_GUID,
  90           .perf_device = false,
  91         },
  92
  93         /* KVP */
  94         { .dev_type = HV_KVP,
  95           HV_KVP_GUID,
  96           .perf_device = false,
  97         },
  98
  99         /* Time Synch */
 100         { .dev_type = HV_TS,
 101           HV_TS_GUID,
 102           .perf_device = false,
 103         },
 104
 105         /* Heartbeat */
 106         { .dev_type = HV_HB,
 107           HV_HEART_BEAT_GUID,
 108           .perf_device = false,
 109         },
 110
 111         /* Shutdown */
 112         { .dev_type = HV_SHUTDOWN,
 113           HV_SHUTDOWN_GUID,
 114           .perf_device = false,
 115         },
 116
 117         /* File copy */
 118         { .dev_type = HV_FCOPY,
 119           HV_FCOPY_GUID,
 120           .perf_device = false,
 121         },
 122
 123         /* Backup */
 124         { .dev_type = HV_BACKUP,
 125           HV_VSS_GUID,
 126           .perf_device = false,
 127         },
 128
 129         /* Dynamic Memory */
 130         { .dev_type = HV_DM,
 131           HV_DM_GUID,
 132           .perf_device = false,
 133         },
 134
 135         /* Unknown GUID */
 136         { .dev_type = HV_UNKOWN,
 137           .perf_device = false,
 138         },
 139 };
 140
 141 static u16 hv_get_dev_type(const uuid_le *guid)
 142 {
 143         u16 i;
 144
 145         for (i = HV_IDE; i < HV_UNKOWN; i++) {
 146                 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid))
 147                         return i;
 148         }
 149         pr_info("Unknown GUID: %pUl\n", guid);
 150         return i;
 151 }
 152
 153 /**
 154  * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message
 155  * @icmsghdrp: Pointer to msg header structure
 156  * @icmsg_negotiate: Pointer to negotiate message structure
 157  * @buf: Raw buffer channel data
 158  *
 159  * @icmsghdrp is of type &struct icmsg_hdr.
 160  * @negop is of type &struct icmsg_negotiate.
 161  * Set up and fill in default negotiate response message.
 162  *
 163  * The fw_version specifies the  framework version that
 164  * we can support and srv_version specifies the service
 165  * version we can support.
 166  *
 167  * Mainly used by Hyper-V drivers.
 168  */
 169 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp,
 170                                 struct icmsg_negotiate *negop, u8 *buf,
 171                                 int fw_version, int srv_version)
 172 {
 173         int icframe_major, icframe_minor;
 174         int icmsg_major, icmsg_minor;
 175         int fw_major, fw_minor;
 176         int srv_major, srv_minor;
 177         int i;
 178         bool found_match = false;
 179
 180         icmsghdrp->icmsgsize = 0x10;
 181         fw_major = (fw_version >> 16);
 182         fw_minor = (fw_version & 0xFFFF);
 183
 184         srv_major = (srv_version >> 16);
 185         srv_minor = (srv_version & 0xFFFF);
 186
 187         negop = (struct icmsg_negotiate *)&buf[
 188                 sizeof(struct vmbuspipe_hdr) +
 189                 sizeof(struct icmsg_hdr)];
 190
 191         icframe_major = negop->icframe_vercnt;
 192         icframe_minor = 0;
 193
 194         icmsg_major = negop->icmsg_vercnt;
 195         icmsg_minor = 0;
 196
 197         /*
 198          * Select the framework version number we will
 199          * support.
 200          */
 201
 202         for (i = 0; i < negop->icframe_vercnt; i++) {
 203                 if ((negop->icversion_data[i].major == fw_major) &&
 204                    (negop->icversion_data[i].minor == fw_minor)) {
 205                         icframe_major = negop->icversion_data[i].major;
 206                         icframe_minor = negop->icversion_data[i].minor;
 207                         found_match = true;
 208                 }
 209         }
 210
 211         if (!found_match)
 212                 goto fw_error;
 213
 214         found_match = false;
 215
 216         for (i = negop->icframe_vercnt;
 217                  (i < negop->icframe_vercnt + negop->icmsg_vercnt); i++) {
 218                 if ((negop->icversion_data[i].major == srv_major) &&
 219                    (negop->icversion_data[i].minor == srv_minor)) {
 220                         icmsg_major = negop->icversion_data[i].major;
 221                         icmsg_minor = negop->icversion_data[i].minor;
 222                         found_match = true;
 223                 }
 224         }
 225
 226         /*
 227          * Respond with the framework and service
 228          * version numbers we can support.
 229          */
 230
 231 fw_error:
 232         if (!found_match) {
 233                 negop->icframe_vercnt = 0;
 234                 negop->icmsg_vercnt = 0;
 235         } else {
 236                 negop->icframe_vercnt = 1;
 237                 negop->icmsg_vercnt = 1;
 238         }
 239
 240         negop->icversion_data[0].major = icframe_major;
 241         negop->icversion_data[0].minor = icframe_minor;
 242         negop->icversion_data[1].major = icmsg_major;
 243         negop->icversion_data[1].minor = icmsg_minor;
 244         return found_match;
 245 }
 246
 247 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp);
 248
 249 /*
 250  * alloc_channel - Allocate and initialize a vmbus channel object
 251  */
 252 static struct vmbus_channel *alloc_channel(void)
 253 {
 254         static atomic_t chan_num = ATOMIC_INIT(0);
 255         struct vmbus_channel *channel;
 256
 257         channel = kzalloc(sizeof(*channel), GFP_ATOMIC);
 258         if (!channel)
 259                 return NULL;
 260
 261         channel->id = atomic_inc_return(&chan_num);
 262         channel->acquire_ring_lock = true;
 263         spin_lock_init(&channel->inbound_lock);
 264         spin_lock_init(&channel->lock);
 265
 266         INIT_LIST_HEAD(&channel->sc_list);
 267         INIT_LIST_HEAD(&channel->percpu_list);
 268
 269         return channel;
 270 }
 271
 272 /*
 273  * free_channel - Release the resources used by the vmbus channel object
 274  */
 275 static void free_channel(struct vmbus_channel *channel)
 276 {
 277         kfree(channel);
 278 }
 279
 280 static void percpu_channel_enq(void *arg)
 281 {
 282         struct vmbus_channel *channel = arg;
 283         int cpu = smp_processor_id();
 284
 285         list_add_tail(&channel->percpu_list, &hv_context.percpu_list[cpu]);
 286 }
 287
 288 static void percpu_channel_deq(void *arg)
 289 {
 290         struct vmbus_channel *channel = arg;
 291
 292         list_del(&channel->percpu_list);
 293 }
 294
 295
 296 static void vmbus_release_relid(u32 relid)
 297 {
 298         struct vmbus_channel_relid_released msg;
 299
 300         memset(&msg, 0, sizeof(struct vmbus_channel_relid_released));
 301         msg.child_relid = relid;
 302         msg.header.msgtype = CHANNELMSG_RELID_RELEASED;
 303         vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released));
 304 }
 305
 306 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid)
 307 {
 308         unsigned long flags;
 309         struct vmbus_channel *primary_channel;
 310
 311         vmbus_release_relid(relid);
 312
 313         BUG_ON(!channel->rescind);
 314         BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
 315
 316         if (channel->target_cpu != get_cpu()) {
 317                 put_cpu();
 318                 smp_call_function_single(channel->target_cpu,
 319                                          percpu_channel_deq, channel, true);
 320         } else {
 321                 percpu_channel_deq(channel);
 322                 put_cpu();
 323         }
 324
 325         if (channel->primary_channel == NULL) {
 326                 list_del(&channel->listentry);
 327
 328                 primary_channel = channel;
 329         } else {
 330                 primary_channel = channel->primary_channel;
 331                 spin_lock_irqsave(&primary_channel->lock, flags);
 332                 list_del(&channel->sc_list);
 333                 primary_channel->num_sc--;
 334                 spin_unlock_irqrestore(&primary_channel->lock, flags);
 335         }
 336
 337         /*
 338          * We need to free the bit for init_vp_index() to work in the case
 339          * of sub-channel, when we reload drivers like hv_netvsc.
 340          */
 341         cpumask_clear_cpu(channel->target_cpu,
 342                           &primary_channel->alloced_cpus_in_node);
 343
 344         free_channel(channel);
 345 }
 346
 347 void vmbus_free_channels(void)
 348 {
 349         struct vmbus_channel *channel, *tmp;
 350
 351         list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
 352                 listentry) {
 353                 /* hv_process_channel_removal() needs this */
 354                 channel->rescind = true;
 355
 356                 vmbus_device_unregister(channel->device_obj);
 357         }
 358 }
 359
 360 /*
 361  * vmbus_process_offer - Process the offer by creating a channel/device
 362  * associated with this offer
 363  */
 364 static void vmbus_process_offer(struct vmbus_channel *newchannel)
 365 {
 366         struct vmbus_channel *channel;
 367         bool fnew = true;
 368         unsigned long flags;
 369         u16 dev_type;
 370         int ret;
 371
 372         /* Make sure this is a new offer */
 373         mutex_lock(&vmbus_connection.channel_mutex);
 374
 375         list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 376                 if (!uuid_le_cmp(channel->offermsg.offer.if_type,
 377                         newchannel->offermsg.offer.if_type) &&
 378                         !uuid_le_cmp(channel->offermsg.offer.if_instance,
 379                                 newchannel->offermsg.offer.if_instance)) {
 380                         fnew = false;
 381                         break;
 382                 }
 383         }
 384
 385         if (fnew)
 386                 list_add_tail(&newchannel->listentry,
 387                               &vmbus_connection.chn_list);
 388
 389         mutex_unlock(&vmbus_connection.channel_mutex);
 390
 391         if (!fnew) {
 392                 /*
 393                  * Check to see if this is a sub-channel.
 394                  */
 395                 if (newchannel->offermsg.offer.sub_channel_index != 0) {
 396                         /*
 397                          * Process the sub-channel.
 398                          */
 399                         newchannel->primary_channel = channel;
 400                         spin_lock_irqsave(&channel->lock, flags);
 401                         list_add_tail(&newchannel->sc_list, &channel->sc_list);
 402                         channel->num_sc++;
 403                         spin_unlock_irqrestore(&channel->lock, flags);
 404                 } else
 405                         goto err_free_chan;
 406         }
 407
 408         dev_type = hv_get_dev_type(&newchannel->offermsg.offer.if_type);
 409
 410         init_vp_index(newchannel, dev_type);
 411
 412         if (newchannel->target_cpu != get_cpu()) {
 413                 put_cpu();
 414                 smp_call_function_single(newchannel->target_cpu,
 415                                          percpu_channel_enq,
 416                                          newchannel, true);
 417         } else {
 418                 percpu_channel_enq(newchannel);
 419                 put_cpu();
 420         }
 421
 422         /*
 423          * This state is used to indicate a successful open
 424          * so that when we do close the channel normally, we
 425          * can cleanup properly
 426          */
 427         newchannel->state = CHANNEL_OPEN_STATE;
 428
 429         if (!fnew) {
 430                 if (channel->sc_creation_callback != NULL)
 431                         channel->sc_creation_callback(newchannel);
 432                 return;
 433         }
 434
 435         /*
 436          * Start the process of binding this offer to the driver
 437          * We need to set the DeviceObject field before calling
 438          * vmbus_child_dev_add()
 439          */
 440         newchannel->device_obj = vmbus_device_create(
 441                 &newchannel->offermsg.offer.if_type,
 442                 &newchannel->offermsg.offer.if_instance,
 443                 newchannel);
 444         if (!newchannel->device_obj)
 445                 goto err_deq_chan;
 446
 447         newchannel->device_obj->device_id = dev_type;
 448         /*
 449          * Add the new device to the bus. This will kick off device-driver
 450          * binding which eventually invokes the device driver's AddDevice()
 451          * method.
 452          */
 453         mutex_lock(&vmbus_connection.channel_mutex);
 454         ret = vmbus_device_register(newchannel->device_obj);
 455         mutex_unlock(&vmbus_connection.channel_mutex);
 456
 457         if (ret != 0) {
 458                 pr_err("unable to add child device object (relid %d)\n",
 459                         newchannel->offermsg.child_relid);
 460                 kfree(newchannel->device_obj);
 461                 goto err_deq_chan;
 462         }
 463         return;
 464
 465 err_deq_chan:
 466         vmbus_release_relid(newchannel->offermsg.child_relid);
 467
 468         mutex_lock(&vmbus_connection.channel_mutex);
 469         list_del(&newchannel->listentry);
 470         mutex_unlock(&vmbus_connection.channel_mutex);
 471
 472         if (newchannel->target_cpu != get_cpu()) {
 473                 put_cpu();
 474                 smp_call_function_single(newchannel->target_cpu,
 475                                          percpu_channel_deq, newchannel, true);
 476         } else {
 477                 percpu_channel_deq(newchannel);
 478                 put_cpu();
 479         }
 480
 481 err_free_chan:
 482         free_channel(newchannel);
 483 }
 484
 485 /*
 486  * We use this state to statically distribute the channel interrupt load.
 487  */
 488 static int next_numa_node_id;
 489
 490 /*
 491  * Starting with Win8, we can statically distribute the incoming
 492  * channel interrupt load by binding a channel to VCPU.
 493  * We do this in a hierarchical fashion:
 494  * First distribute the primary channels across available NUMA nodes
 495  * and then distribute the subchannels amongst the CPUs in the NUMA
 496  * node assigned to the primary channel.
 497  *
 498  * For pre-win8 hosts or non-performance critical channels we assign the
 499  * first CPU in the first NUMA node.
 500  */
 501 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
 502 {
 503         u32 cur_cpu;
 504         bool perf_chn = vmbus_devs[dev_type].perf_device;
 505         struct vmbus_channel *primary = channel->primary_channel;
 506         int next_node;
 507         struct cpumask available_mask;
 508         struct cpumask *alloced_mask;
 509
 510         if ((vmbus_proto_version == VERSION_WS2008) ||
 511             (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
 512                 /*
 513                  * Prior to win8, all channel interrupts are
 514                  * delivered on cpu 0.
 515                  * Also if the channel is not a performance critical
 516                  * channel, bind it to cpu 0.
 517                  */
 518                 channel->numa_node = 0;
 519                 channel->target_cpu = 0;
 520                 channel->target_vp = hv_context.vp_index[0];
 521                 return;
 522         }
 523
 524         /*
 525          * We distribute primary channels evenly across all the available
 526          * NUMA nodes and within the assigned NUMA node we will assign the
 527          * first available CPU to the primary channel.
 528          * The sub-channels will be assigned to the CPUs available in the
 529          * NUMA node evenly.
 530          */
 531         if (!primary) {
 532                 while (true) {
 533                         next_node = next_numa_node_id++;
 534                         if (next_node == nr_node_ids)
 535                                 next_node = next_numa_node_id = 0;
 536                         if (cpumask_empty(cpumask_of_node(next_node)))
 537                                 continue;
 538                         break;
 539                 }
 540                 channel->numa_node = next_node;
 541                 primary = channel;
 542         }
 543         alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
 544
 545         if (cpumask_weight(alloced_mask) ==
 546             cpumask_weight(cpumask_of_node(primary->numa_node))) {
 547                 /*
 548                  * We have cycled through all the CPUs in the node;
 549                  * reset the alloced map.
 550                  */
 551                 cpumask_clear(alloced_mask);
 552         }
 553
 554         cpumask_xor(&available_mask, alloced_mask,
 555                     cpumask_of_node(primary->numa_node));
 556
 557         cur_cpu = -1;
 558
 559         /*
 560          * Normally Hyper-V host doesn't create more subchannels than there
 561          * are VCPUs on the node but it is possible when not all present VCPUs
 562          * on the node are initialized by guest. Clear the alloced_cpus_in_node
 563          * to start over.
 564          */
 565         if (cpumask_equal(&primary->alloced_cpus_in_node,
 566                           cpumask_of_node(primary->numa_node)))
 567                 cpumask_clear(&primary->alloced_cpus_in_node);
 568
 569         while (true) {
 570                 cur_cpu = cpumask_next(cur_cpu, &available_mask);
 571                 if (cur_cpu >= nr_cpu_ids) {
 572                         cur_cpu = -1;
 573                         cpumask_copy(&available_mask,
 574                                      cpumask_of_node(primary->numa_node));
 575                         continue;
 576                 }
 577
 578                 /*
 579                  * NOTE: in the case of sub-channel, we clear the sub-channel
 580                  * related bit(s) in primary->alloced_cpus_in_node in
 581                  * hv_process_channel_removal(), so when we reload drivers
 582                  * like hv_netvsc in SMP guest, here we're able to re-allocate
 583                  * bit from primary->alloced_cpus_in_node.
 584                  */
 585                 if (!cpumask_test_cpu(cur_cpu,
 586                                 &primary->alloced_cpus_in_node)) {
 587                         cpumask_set_cpu(cur_cpu,
 588                                         &primary->alloced_cpus_in_node);
 589                         cpumask_set_cpu(cur_cpu, alloced_mask);
 590                         break;
 591                 }
 592         }
 593
 594         channel->target_cpu = cur_cpu;
 595         channel->target_vp = hv_context.vp_index[cur_cpu];
 596 }
 597
 598 static void vmbus_wait_for_unload(void)
 599 {
 600         int cpu;
 601         void *page_addr;
 602         struct hv_message *msg;
 603         struct vmbus_channel_message_header *hdr;
 604         u32 message_type;
 605
 606         /*
 607          * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was
 608          * used for initial contact or to CPU0 depending on host version. When
 609          * we're crashing on a different CPU let's hope that IRQ handler on
 610          * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still
 611          * functional and vmbus_unload_response() will complete
 612          * vmbus_connection.unload_event. If not, the last thing we can do is
 613          * read message pages for all CPUs directly.
 614          */
 615         while (1) {
 616                 if (completion_done(&vmbus_connection.unload_event))
 617                         break;
 618
 619                 for_each_online_cpu(cpu) {
 620                         page_addr = hv_context.synic_message_page[cpu];
 621                         msg = (struct hv_message *)page_addr +
 622                                 VMBUS_MESSAGE_SINT;
 623
 624                         message_type = READ_ONCE(msg->header.message_type);
 625                         if (message_type == HVMSG_NONE)
 626                                 continue;
 627
 628                         hdr = (struct vmbus_channel_message_header *)
 629                                 msg->u.payload;
 630
 631                         if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
 632                                 complete(&vmbus_connection.unload_event);
 633
 634                         vmbus_signal_eom(msg, message_type);
 635                 }
 636
 637                 mdelay(10);
 638         }
 639
 640         /*
 641          * We're crashing and already got the UNLOAD_RESPONSE, cleanup all
 642          * maybe-pending messages on all CPUs to be able to receive new
 643          * messages after we reconnect.
 644          */
 645         for_each_online_cpu(cpu) {
 646                 page_addr = hv_context.synic_message_page[cpu];
 647                 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT;
 648                 msg->header.message_type = HVMSG_NONE;
 649         }
 650 }
 651
 652 /*
 653  * vmbus_unload_response - Handler for the unload response.
 654  */
 655 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
 656 {
 657         /*
 658          * This is a global event; just wakeup the waiting thread.
 659          * Once we successfully unload, we can cleanup the monitor state.
 660          */
 661         complete(&vmbus_connection.unload_event);
 662 }
 663
 664 void vmbus_initiate_unload(bool crash)
 665 {
 666         struct vmbus_channel_message_header hdr;
 667
 668         /* Pre-Win2012R2 hosts don't support reconnect */
 669         if (vmbus_proto_version < VERSION_WIN8_1)
 670                 return;
 671
 672         init_completion(&vmbus_connection.unload_event);
 673         memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 674         hdr.msgtype = CHANNELMSG_UNLOAD;
 675         vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header));
 676
 677         /*
 678          * vmbus_initiate_unload() is also called on crash and the crash can be
 679          * happening in an interrupt context, where scheduling is impossible.
 680          */
 681         if (!crash)
 682                 wait_for_completion(&vmbus_connection.unload_event);
 683         else
 684                 vmbus_wait_for_unload();
 685 }
 686
 687 /*
 688  * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
 689  *
 690  */
 691 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
 692 {
 693         struct vmbus_channel_offer_channel *offer;
 694         struct vmbus_channel *newchannel;
 695
 696         offer = (struct vmbus_channel_offer_channel *)hdr;
 697
 698         /* Allocate the channel object and save this offer. */
 699         newchannel = alloc_channel();
 700         if (!newchannel) {
 701                 pr_err("Unable to allocate channel object\n");
 702                 return;
 703         }
 704
 705         /*
 706          * By default we setup state to enable batched
 707          * reading. A specific service can choose to
 708          * disable this prior to opening the channel.
 709          */
 710         newchannel->batched_reading = true;
 711
 712         /*
 713          * Setup state for signalling the host.
 714          */
 715         newchannel->sig_event = (struct hv_input_signal_event *)
 716                                 (ALIGN((unsigned long)
 717                                 &newchannel->sig_buf,
 718                                 HV_HYPERCALL_PARAM_ALIGN));
 719
 720         newchannel->sig_event->connectionid.asu32 = 0;
 721         newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
 722         newchannel->sig_event->flag_number = 0;
 723         newchannel->sig_event->rsvdz = 0;
 724
 725         if (vmbus_proto_version != VERSION_WS2008) {
 726                 newchannel->is_dedicated_interrupt =
 727                                 (offer->is_dedicated_interrupt != 0);
 728                 newchannel->sig_event->connectionid.u.id =
 729                                 offer->connection_id;
 730         }
 731
 732         memcpy(&newchannel->offermsg, offer,
 733                sizeof(struct vmbus_channel_offer_channel));
 734         newchannel->monitor_grp = (u8)offer->monitorid / 32;
 735         newchannel->monitor_bit = (u8)offer->monitorid % 32;
 736
 737         vmbus_process_offer(newchannel);
 738 }
 739
 740 /*
 741  * vmbus_onoffer_rescind - Rescind offer handler.
 742  *
 743  * We queue a work item to process this offer synchronously
 744  */
 745 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 746 {
 747         struct vmbus_channel_rescind_offer *rescind;
 748         struct vmbus_channel *channel;
 749         unsigned long flags;
 750         struct device *dev;
 751
 752         rescind = (struct vmbus_channel_rescind_offer *)hdr;
 753
 754         mutex_lock(&vmbus_connection.channel_mutex);
 755         channel = relid2channel(rescind->child_relid);
 756
 757         if (channel == NULL) {
 758                 /*
 759                  * This is very impossible, because in
 760                  * vmbus_process_offer(), we have already invoked
 761                  * vmbus_release_relid() on error.
 762                  */
 763                 goto out;
 764         }
 765
 766         spin_lock_irqsave(&channel->lock, flags);
 767         channel->rescind = true;
 768         spin_unlock_irqrestore(&channel->lock, flags);
 769
 770         if (channel->device_obj) {
 771                 if (channel->chn_rescind_callback) {
 772                         channel->chn_rescind_callback(channel);
 773                         goto out;
 774                 }
 775                 /*
 776                  * We will have to unregister this device from the
 777                  * driver core.
 778                  */
 779                 dev = get_device(&channel->device_obj->device);
 780                 if (dev) {
 781                         vmbus_device_unregister(channel->device_obj);
 782                         put_device(dev);
 783                 }
 784         } else {
 785                 hv_process_channel_removal(channel,
 786                         channel->offermsg.child_relid);
 787         }
 788
 789 out:
 790         mutex_unlock(&vmbus_connection.channel_mutex);
 791 }
 792
 793 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
 794 {
 795         mutex_lock(&vmbus_connection.channel_mutex);
 796
 797         BUG_ON(!is_hvsock_channel(channel));
 798
 799         channel->rescind = true;
 800         vmbus_device_unregister(channel->device_obj);
 801
 802         mutex_unlock(&vmbus_connection.channel_mutex);
 803 }
 804 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
 805
 806
 807 /*
 808  * vmbus_onoffers_delivered -
 809  * This is invoked when all offers have been delivered.
 810  *
 811  * Nothing to do here.
 812  */
 813 static void vmbus_onoffers_delivered(
 814                         struct vmbus_channel_message_header *hdr)
 815 {
 816 }
 817
 818 /*
 819  * vmbus_onopen_result - Open result handler.
 820  *
 821  * This is invoked when we received a response to our channel open request.
 822  * Find the matching request, copy the response and signal the requesting
 823  * thread.
 824  */
 825 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
 826 {
 827         struct vmbus_channel_open_result *result;
 828         struct vmbus_channel_msginfo *msginfo;
 829         struct vmbus_channel_message_header *requestheader;
 830         struct vmbus_channel_open_channel *openmsg;
 831         unsigned long flags;
 832
 833         result = (struct vmbus_channel_open_result *)hdr;
 834
 835         /*
 836          * Find the open msg, copy the result and signal/unblock the wait event
 837          */
 838         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 839
 840         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 841                                 msglistentry) {
 842                 requestheader =
 843                         (struct vmbus_channel_message_header *)msginfo->msg;
 844
 845                 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
 846                         openmsg =
 847                         (struct vmbus_channel_open_channel *)msginfo->msg;
 848                         if (openmsg->child_relid == result->child_relid &&
 849                             openmsg->openid == result->openid) {
 850                                 memcpy(&msginfo->response.open_result,
 851                                        result,
 852                                        sizeof(
 853                                         struct vmbus_channel_open_result));
 854                                 complete(&msginfo->waitevent);
 855                                 break;
 856                         }
 857                 }
 858         }
 859         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 860 }
 861
 862 /*
 863  * vmbus_ongpadl_created - GPADL created handler.
 864  *
 865  * This is invoked when we received a response to our gpadl create request.
 866  * Find the matching request, copy the response and signal the requesting
 867  * thread.
 868  */
 869 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
 870 {
 871         struct vmbus_channel_gpadl_created *gpadlcreated;
 872         struct vmbus_channel_msginfo *msginfo;
 873         struct vmbus_channel_message_header *requestheader;
 874         struct vmbus_channel_gpadl_header *gpadlheader;
 875         unsigned long flags;
 876
 877         gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
 878
 879         /*
 880          * Find the establish msg, copy the result and signal/unblock the wait
 881          * event
 882          */
 883         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 884
 885         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 886                                 msglistentry) {
 887                 requestheader =
 888                         (struct vmbus_channel_message_header *)msginfo->msg;
 889
 890                 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
 891                         gpadlheader =
 892                         (struct vmbus_channel_gpadl_header *)requestheader;
 893
 894                         if ((gpadlcreated->child_relid ==
 895                              gpadlheader->child_relid) &&
 896                             (gpadlcreated->gpadl == gpadlheader->gpadl)) {
 897                                 memcpy(&msginfo->response.gpadl_created,
 898                                        gpadlcreated,
 899                                        sizeof(
 900                                         struct vmbus_channel_gpadl_created));
 901                                 complete(&msginfo->waitevent);
 902                                 break;
 903                         }
 904                 }
 905         }
 906         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 907 }
 908
 909 /*
 910  * vmbus_ongpadl_torndown - GPADL torndown handler.
 911  *
 912  * This is invoked when we received a response to our gpadl teardown request.
 913  * Find the matching request, copy the response and signal the requesting
 914  * thread.
 915  */
 916 static void vmbus_ongpadl_torndown(
 917                         struct vmbus_channel_message_header *hdr)
 918 {
 919         struct vmbus_channel_gpadl_torndown *gpadl_torndown;
 920         struct vmbus_channel_msginfo *msginfo;
 921         struct vmbus_channel_message_header *requestheader;
 922         struct vmbus_channel_gpadl_teardown *gpadl_teardown;
 923         unsigned long flags;
 924
 925         gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
 926
 927         /*
 928          * Find the open msg, copy the result and signal/unblock the wait event
 929          */
 930         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 931
 932         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 933                                 msglistentry) {
 934                 requestheader =
 935                         (struct vmbus_channel_message_header *)msginfo->msg;
 936
 937                 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
 938                         gpadl_teardown =
 939                         (struct vmbus_channel_gpadl_teardown *)requestheader;
 940
 941                         if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
 942                                 memcpy(&msginfo->response.gpadl_torndown,
 943                                        gpadl_torndown,
 944                                        sizeof(
 945                                         struct vmbus_channel_gpadl_torndown));
 946                                 complete(&msginfo->waitevent);
 947                                 break;
 948                         }
 949                 }
 950         }
 951         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 952 }
 953
 954 /*
 955  * vmbus_onversion_response - Version response handler
 956  *
 957  * This is invoked when we received a response to our initiate contact request.
 958  * Find the matching request, copy the response and signal the requesting
 959  * thread.
 960  */
 961 static void vmbus_onversion_response(
 962                 struct vmbus_channel_message_header *hdr)
 963 {
 964         struct vmbus_channel_msginfo *msginfo;
 965         struct vmbus_channel_message_header *requestheader;
 966         struct vmbus_channel_version_response *version_response;
 967         unsigned long flags;
 968
 969         version_response = (struct vmbus_channel_version_response *)hdr;
 970         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 971
 972         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 973                                 msglistentry) {
 974                 requestheader =
 975                         (struct vmbus_channel_message_header *)msginfo->msg;
 976
 977                 if (requestheader->msgtype ==
 978                     CHANNELMSG_INITIATE_CONTACT) {
 979                         memcpy(&msginfo->response.version_response,
 980                               version_response,
 981                               sizeof(struct vmbus_channel_version_response));
 982                         complete(&msginfo->waitevent);
 983                 }
 984         }
 985         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 986 }
 987
 988 /* Channel message dispatch table */
 989 struct vmbus_channel_message_table_entry
 990         channel_message_table[CHANNELMSG_COUNT] = {
 991         {CHANNELMSG_INVALID,                    0, NULL},
 992         {CHANNELMSG_OFFERCHANNEL,               0, vmbus_onoffer},
 993         {CHANNELMSG_RESCIND_CHANNELOFFER,       0, vmbus_onoffer_rescind},
 994         {CHANNELMSG_REQUESTOFFERS,              0, NULL},
 995         {CHANNELMSG_ALLOFFERS_DELIVERED,        1, vmbus_onoffers_delivered},
 996         {CHANNELMSG_OPENCHANNEL,                0, NULL},
 997         {CHANNELMSG_OPENCHANNEL_RESULT,         1, vmbus_onopen_result},
 998         {CHANNELMSG_CLOSECHANNEL,               0, NULL},
 999         {CHANNELMSG_GPADL_HEADER,               0, NULL},
1000         {CHANNELMSG_GPADL_BODY,                 0, NULL},
1001         {CHANNELMSG_GPADL_CREATED,              1, vmbus_ongpadl_created},
1002         {CHANNELMSG_GPADL_TEARDOWN,             0, NULL},
1003         {CHANNELMSG_GPADL_TORNDOWN,             1, vmbus_ongpadl_torndown},
1004         {CHANNELMSG_RELID_RELEASED,             0, NULL},
1005         {CHANNELMSG_INITIATE_CONTACT,           0, NULL},
1006         {CHANNELMSG_VERSION_RESPONSE,           1, vmbus_onversion_response},
1007         {CHANNELMSG_UNLOAD,                     0, NULL},
1008         {CHANNELMSG_UNLOAD_RESPONSE,            1, vmbus_unload_response},
1009         {CHANNELMSG_18,                         0, NULL},
1010         {CHANNELMSG_19,                         0, NULL},
1011         {CHANNELMSG_20,                         0, NULL},
1012         {CHANNELMSG_TL_CONNECT_REQUEST,         0, NULL},
1013 };
1014
1015 /*
1016  * vmbus_onmessage - Handler for channel protocol messages.
1017  *
1018  * This is invoked in the vmbus worker thread context.
1019  */
1020 void vmbus_onmessage(void *context)
1021 {
1022         struct hv_message *msg = context;
1023         struct vmbus_channel_message_header *hdr;
1024         int size;
1025
1026         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1027         size = msg->header.payload_size;
1028
1029         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1030                 pr_err("Received invalid channel message type %d size %d\n",
1031                            hdr->msgtype, size);
1032                 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1033                                      (unsigned char *)msg->u.payload, size);
1034                 return;
1035         }
1036
1037         if (channel_message_table[hdr->msgtype].message_handler)
1038                 channel_message_table[hdr->msgtype].message_handler(hdr);
1039         else
1040                 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1041 }
1042
1043 /*
1044  * vmbus_request_offers - Send a request to get all our pending offers.
1045  */
1046 int vmbus_request_offers(void)
1047 {
1048         struct vmbus_channel_message_header *msg;
1049         struct vmbus_channel_msginfo *msginfo;
1050         int ret;
1051
1052         msginfo = kmalloc(sizeof(*msginfo) +
1053                           sizeof(struct vmbus_channel_message_header),
1054                           GFP_KERNEL);
1055         if (!msginfo)
1056                 return -ENOMEM;
1057
1058         msg = (struct vmbus_channel_message_header *)msginfo->msg;
1059
1060         msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1061
1062
1063         ret = vmbus_post_msg(msg,
1064                                sizeof(struct vmbus_channel_message_header));
1065         if (ret != 0) {
1066                 pr_err("Unable to request offers - %d\n", ret);
1067
1068                 goto cleanup;
1069         }
1070
1071 cleanup:
1072         kfree(msginfo);
1073
1074         return ret;
1075 }
1076
1077 /*
1078  * Retrieve the (sub) channel on which to send an outgoing request.
1079  * When a primary channel has multiple sub-channels, we try to
1080  * distribute the load equally amongst all available channels.
1081  */
1082 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1083 {
1084         struct list_head *cur, *tmp;
1085         int cur_cpu;
1086         struct vmbus_channel *cur_channel;
1087         struct vmbus_channel *outgoing_channel = primary;
1088         int next_channel;
1089         int i = 1;
1090
1091         if (list_empty(&primary->sc_list))
1092                 return outgoing_channel;
1093
1094         next_channel = primary->next_oc++;
1095
1096         if (next_channel > (primary->num_sc)) {
1097                 primary->next_oc = 0;
1098                 return outgoing_channel;
1099         }
1100
1101         cur_cpu = hv_context.vp_index[get_cpu()];
1102         put_cpu();
1103         list_for_each_safe(cur, tmp, &primary->sc_list) {
1104                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1105                 if (cur_channel->state != CHANNEL_OPENED_STATE)
1106                         continue;
1107
1108                 if (cur_channel->target_vp == cur_cpu)
1109                         return cur_channel;
1110
1111                 if (i == next_channel)
1112                         return cur_channel;
1113
1114                 i++;
1115         }
1116
1117         return outgoing_channel;
1118 }
1119 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1120
1121 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1122 {
1123         struct list_head *cur, *tmp;
1124         struct vmbus_channel *cur_channel;
1125
1126         if (primary_channel->sc_creation_callback == NULL)
1127                 return;
1128
1129         list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1130                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1131
1132                 primary_channel->sc_creation_callback(cur_channel);
1133         }
1134 }
1135
1136 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1137                                 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1138 {
1139         primary_channel->sc_creation_callback = sc_cr_cb;
1140 }
1141 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1142
1143 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1144 {
1145         bool ret;
1146
1147         ret = !list_empty(&primary->sc_list);
1148
1149         if (ret) {
1150                 /*
1151                  * Invoke the callback on sub-channel creation.
1152                  * This will present a uniform interface to the
1153                  * clients.
1154                  */
1155                 invoke_sc_cb(primary);
1156         }
1157
1158         return ret;
1159 }
1160 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1161
1162 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1163                 void (*chn_rescind_cb)(struct vmbus_channel *))
1164 {
1165         channel->chn_rescind_callback = chn_rescind_cb;
1166 }
1167 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);