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         BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex));
 314
 315         if (channel->target_cpu != get_cpu()) {
 316                 put_cpu();
 317                 smp_call_function_single(channel->target_cpu,
 318                                          percpu_channel_deq, channel, true);
 319         } else {
 320                 percpu_channel_deq(channel);
 321                 put_cpu();
 322         }
 323
 324         if (channel->primary_channel == NULL) {
 325                 list_del(&channel->listentry);
 326
 327                 primary_channel = channel;
 328         } else {
 329                 primary_channel = channel->primary_channel;
 330                 spin_lock_irqsave(&primary_channel->lock, flags);
 331                 list_del(&channel->sc_list);
 332                 primary_channel->num_sc--;
 333                 spin_unlock_irqrestore(&primary_channel->lock, flags);
 334         }
 335
 336         /*
 337          * We need to free the bit for init_vp_index() to work in the case
 338          * of sub-channel, when we reload drivers like hv_netvsc.
 339          */
 340         cpumask_clear_cpu(channel->target_cpu,
 341                           &primary_channel->alloced_cpus_in_node);
 342
 343         free_channel(channel);
 344 }
 345
 346 void vmbus_free_channels(void)
 347 {
 348         struct vmbus_channel *channel, *tmp;
 349
 350         list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list,
 351                 listentry) {
 352                 /* hv_process_channel_removal() needs this */
 353                 channel->rescind = true;
 354
 355                 vmbus_device_unregister(channel->device_obj);
 356         }
 357 }
 358
 359 /*
 360  * vmbus_process_offer - Process the offer by creating a channel/device
 361  * associated with this offer
 362  */
 363 static void vmbus_process_offer(struct vmbus_channel *newchannel)
 364 {
 365         struct vmbus_channel *channel;
 366         bool fnew = true;
 367         unsigned long flags;
 368         u16 dev_type;
 369         int ret;
 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         mutex_lock(&vmbus_connection.channel_mutex);
 453         ret = vmbus_device_register(newchannel->device_obj);
 454         mutex_unlock(&vmbus_connection.channel_mutex);
 455
 456         if (ret != 0) {
 457                 pr_err("unable to add child device object (relid %d)\n",
 458                         newchannel->offermsg.child_relid);
 459                 kfree(newchannel->device_obj);
 460                 goto err_deq_chan;
 461         }
 462         return;
 463
 464 err_deq_chan:
 465         vmbus_release_relid(newchannel->offermsg.child_relid);
 466
 467         mutex_lock(&vmbus_connection.channel_mutex);
 468         list_del(&newchannel->listentry);
 469         mutex_unlock(&vmbus_connection.channel_mutex);
 470
 471         if (newchannel->target_cpu != get_cpu()) {
 472                 put_cpu();
 473                 smp_call_function_single(newchannel->target_cpu,
 474                                          percpu_channel_deq, newchannel, true);
 475         } else {
 476                 percpu_channel_deq(newchannel);
 477                 put_cpu();
 478         }
 479
 480 err_free_chan:
 481         free_channel(newchannel);
 482 }
 483
 484 /*
 485  * We use this state to statically distribute the channel interrupt load.
 486  */
 487 static int next_numa_node_id;
 488
 489 /*
 490  * Starting with Win8, we can statically distribute the incoming
 491  * channel interrupt load by binding a channel to VCPU.
 492  * We do this in a hierarchical fashion:
 493  * First distribute the primary channels across available NUMA nodes
 494  * and then distribute the subchannels amongst the CPUs in the NUMA
 495  * node assigned to the primary channel.
 496  *
 497  * For pre-win8 hosts or non-performance critical channels we assign the
 498  * first CPU in the first NUMA node.
 499  */
 500 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type)
 501 {
 502         u32 cur_cpu;
 503         bool perf_chn = vmbus_devs[dev_type].perf_device;
 504         struct vmbus_channel *primary = channel->primary_channel;
 505         int next_node;
 506         struct cpumask available_mask;
 507         struct cpumask *alloced_mask;
 508
 509         if ((vmbus_proto_version == VERSION_WS2008) ||
 510             (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) {
 511                 /*
 512                  * Prior to win8, all channel interrupts are
 513                  * delivered on cpu 0.
 514                  * Also if the channel is not a performance critical
 515                  * channel, bind it to cpu 0.
 516                  */
 517                 channel->numa_node = 0;
 518                 channel->target_cpu = 0;
 519                 channel->target_vp = hv_context.vp_index[0];
 520                 return;
 521         }
 522
 523         /*
 524          * We distribute primary channels evenly across all the available
 525          * NUMA nodes and within the assigned NUMA node we will assign the
 526          * first available CPU to the primary channel.
 527          * The sub-channels will be assigned to the CPUs available in the
 528          * NUMA node evenly.
 529          */
 530         if (!primary) {
 531                 while (true) {
 532                         next_node = next_numa_node_id++;
 533                         if (next_node == nr_node_ids)
 534                                 next_node = next_numa_node_id = 0;
 535                         if (cpumask_empty(cpumask_of_node(next_node)))
 536                                 continue;
 537                         break;
 538                 }
 539                 channel->numa_node = next_node;
 540                 primary = channel;
 541         }
 542         alloced_mask = &hv_context.hv_numa_map[primary->numa_node];
 543
 544         if (cpumask_weight(alloced_mask) ==
 545             cpumask_weight(cpumask_of_node(primary->numa_node))) {
 546                 /*
 547                  * We have cycled through all the CPUs in the node;
 548                  * reset the alloced map.
 549                  */
 550                 cpumask_clear(alloced_mask);
 551         }
 552
 553         cpumask_xor(&available_mask, alloced_mask,
 554                     cpumask_of_node(primary->numa_node));
 555
 556         cur_cpu = -1;
 557
 558         /*
 559          * Normally Hyper-V host doesn't create more subchannels than there
 560          * are VCPUs on the node but it is possible when not all present VCPUs
 561          * on the node are initialized by guest. Clear the alloced_cpus_in_node
 562          * to start over.
 563          */
 564         if (cpumask_equal(&primary->alloced_cpus_in_node,
 565                           cpumask_of_node(primary->numa_node)))
 566                 cpumask_clear(&primary->alloced_cpus_in_node);
 567
 568         while (true) {
 569                 cur_cpu = cpumask_next(cur_cpu, &available_mask);
 570                 if (cur_cpu >= nr_cpu_ids) {
 571                         cur_cpu = -1;
 572                         cpumask_copy(&available_mask,
 573                                      cpumask_of_node(primary->numa_node));
 574                         continue;
 575                 }
 576
 577                 /*
 578                  * NOTE: in the case of sub-channel, we clear the sub-channel
 579                  * related bit(s) in primary->alloced_cpus_in_node in
 580                  * hv_process_channel_removal(), so when we reload drivers
 581                  * like hv_netvsc in SMP guest, here we're able to re-allocate
 582                  * bit from primary->alloced_cpus_in_node.
 583                  */
 584                 if (!cpumask_test_cpu(cur_cpu,
 585                                 &primary->alloced_cpus_in_node)) {
 586                         cpumask_set_cpu(cur_cpu,
 587                                         &primary->alloced_cpus_in_node);
 588                         cpumask_set_cpu(cur_cpu, alloced_mask);
 589                         break;
 590                 }
 591         }
 592
 593         channel->target_cpu = cur_cpu;
 594         channel->target_vp = hv_context.vp_index[cur_cpu];
 595 }
 596
 597 static void vmbus_wait_for_unload(void)
 598 {
 599         int cpu = smp_processor_id();
 600         void *page_addr = hv_context.synic_message_page[cpu];
 601         struct hv_message *msg = (struct hv_message *)page_addr +
 602                                   VMBUS_MESSAGE_SINT;
 603         struct vmbus_channel_message_header *hdr;
 604         bool unloaded = false;
 605
 606         while (1) {
 607                 if (msg->header.message_type == HVMSG_NONE) {
 608                         mdelay(10);
 609                         continue;
 610                 }
 611
 612                 hdr = (struct vmbus_channel_message_header *)msg->u.payload;
 613                 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE)
 614                         unloaded = true;
 615
 616                 msg->header.message_type = HVMSG_NONE;
 617                 /*
 618                  * header.message_type needs to be written before we do
 619                  * wrmsrl() below.
 620                  */
 621                 mb();
 622
 623                 if (msg->header.message_flags.msg_pending)
 624                         wrmsrl(HV_X64_MSR_EOM, 0);
 625
 626                 if (unloaded)
 627                         break;
 628         }
 629 }
 630
 631 /*
 632  * vmbus_unload_response - Handler for the unload response.
 633  */
 634 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr)
 635 {
 636         /*
 637          * This is a global event; just wakeup the waiting thread.
 638          * Once we successfully unload, we can cleanup the monitor state.
 639          */
 640         complete(&vmbus_connection.unload_event);
 641 }
 642
 643 void vmbus_initiate_unload(void)
 644 {
 645         struct vmbus_channel_message_header hdr;
 646
 647         /* Pre-Win2012R2 hosts don't support reconnect */
 648         if (vmbus_proto_version < VERSION_WIN8_1)
 649                 return;
 650
 651         init_completion(&vmbus_connection.unload_event);
 652         memset(&hdr, 0, sizeof(struct vmbus_channel_message_header));
 653         hdr.msgtype = CHANNELMSG_UNLOAD;
 654         vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header));
 655
 656         /*
 657          * vmbus_initiate_unload() is also called on crash and the crash can be
 658          * happening in an interrupt context, where scheduling is impossible.
 659          */
 660         if (!in_interrupt())
 661                 wait_for_completion(&vmbus_connection.unload_event);
 662         else
 663                 vmbus_wait_for_unload();
 664 }
 665
 666 /*
 667  * vmbus_onoffer - Handler for channel offers from vmbus in parent partition.
 668  *
 669  */
 670 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr)
 671 {
 672         struct vmbus_channel_offer_channel *offer;
 673         struct vmbus_channel *newchannel;
 674
 675         offer = (struct vmbus_channel_offer_channel *)hdr;
 676
 677         /* Allocate the channel object and save this offer. */
 678         newchannel = alloc_channel();
 679         if (!newchannel) {
 680                 pr_err("Unable to allocate channel object\n");
 681                 return;
 682         }
 683
 684         /*
 685          * By default we setup state to enable batched
 686          * reading. A specific service can choose to
 687          * disable this prior to opening the channel.
 688          */
 689         newchannel->batched_reading = true;
 690
 691         /*
 692          * Setup state for signalling the host.
 693          */
 694         newchannel->sig_event = (struct hv_input_signal_event *)
 695                                 (ALIGN((unsigned long)
 696                                 &newchannel->sig_buf,
 697                                 HV_HYPERCALL_PARAM_ALIGN));
 698
 699         newchannel->sig_event->connectionid.asu32 = 0;
 700         newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID;
 701         newchannel->sig_event->flag_number = 0;
 702         newchannel->sig_event->rsvdz = 0;
 703
 704         if (vmbus_proto_version != VERSION_WS2008) {
 705                 newchannel->is_dedicated_interrupt =
 706                                 (offer->is_dedicated_interrupt != 0);
 707                 newchannel->sig_event->connectionid.u.id =
 708                                 offer->connection_id;
 709         }
 710
 711         memcpy(&newchannel->offermsg, offer,
 712                sizeof(struct vmbus_channel_offer_channel));
 713         newchannel->monitor_grp = (u8)offer->monitorid / 32;
 714         newchannel->monitor_bit = (u8)offer->monitorid % 32;
 715
 716         vmbus_process_offer(newchannel);
 717 }
 718
 719 /*
 720  * vmbus_onoffer_rescind - Rescind offer handler.
 721  *
 722  * We queue a work item to process this offer synchronously
 723  */
 724 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr)
 725 {
 726         struct vmbus_channel_rescind_offer *rescind;
 727         struct vmbus_channel *channel;
 728         unsigned long flags;
 729         struct device *dev;
 730
 731         rescind = (struct vmbus_channel_rescind_offer *)hdr;
 732
 733         mutex_lock(&vmbus_connection.channel_mutex);
 734         channel = relid2channel(rescind->child_relid);
 735
 736         if (channel == NULL) {
 737                 /*
 738                  * This is very impossible, because in
 739                  * vmbus_process_offer(), we have already invoked
 740                  * vmbus_release_relid() on error.
 741                  */
 742                 goto out;
 743         }
 744
 745         spin_lock_irqsave(&channel->lock, flags);
 746         channel->rescind = true;
 747         spin_unlock_irqrestore(&channel->lock, flags);
 748
 749         if (channel->device_obj) {
 750                 if (channel->chn_rescind_callback) {
 751                         channel->chn_rescind_callback(channel);
 752                         goto out;
 753                 }
 754                 /*
 755                  * We will have to unregister this device from the
 756                  * driver core.
 757                  */
 758                 dev = get_device(&channel->device_obj->device);
 759                 if (dev) {
 760                         vmbus_device_unregister(channel->device_obj);
 761                         put_device(dev);
 762                 }
 763         } else {
 764                 hv_process_channel_removal(channel,
 765                         channel->offermsg.child_relid);
 766         }
 767
 768 out:
 769         mutex_unlock(&vmbus_connection.channel_mutex);
 770 }
 771
 772 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel)
 773 {
 774         mutex_lock(&vmbus_connection.channel_mutex);
 775
 776         BUG_ON(!is_hvsock_channel(channel));
 777
 778         channel->rescind = true;
 779         vmbus_device_unregister(channel->device_obj);
 780
 781         mutex_unlock(&vmbus_connection.channel_mutex);
 782 }
 783 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister);
 784
 785
 786 /*
 787  * vmbus_onoffers_delivered -
 788  * This is invoked when all offers have been delivered.
 789  *
 790  * Nothing to do here.
 791  */
 792 static void vmbus_onoffers_delivered(
 793                         struct vmbus_channel_message_header *hdr)
 794 {
 795 }
 796
 797 /*
 798  * vmbus_onopen_result - Open result handler.
 799  *
 800  * This is invoked when we received a response to our channel open request.
 801  * Find the matching request, copy the response and signal the requesting
 802  * thread.
 803  */
 804 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr)
 805 {
 806         struct vmbus_channel_open_result *result;
 807         struct vmbus_channel_msginfo *msginfo;
 808         struct vmbus_channel_message_header *requestheader;
 809         struct vmbus_channel_open_channel *openmsg;
 810         unsigned long flags;
 811
 812         result = (struct vmbus_channel_open_result *)hdr;
 813
 814         /*
 815          * Find the open msg, copy the result and signal/unblock the wait event
 816          */
 817         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 818
 819         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 820                                 msglistentry) {
 821                 requestheader =
 822                         (struct vmbus_channel_message_header *)msginfo->msg;
 823
 824                 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) {
 825                         openmsg =
 826                         (struct vmbus_channel_open_channel *)msginfo->msg;
 827                         if (openmsg->child_relid == result->child_relid &&
 828                             openmsg->openid == result->openid) {
 829                                 memcpy(&msginfo->response.open_result,
 830                                        result,
 831                                        sizeof(
 832                                         struct vmbus_channel_open_result));
 833                                 complete(&msginfo->waitevent);
 834                                 break;
 835                         }
 836                 }
 837         }
 838         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 839 }
 840
 841 /*
 842  * vmbus_ongpadl_created - GPADL created handler.
 843  *
 844  * This is invoked when we received a response to our gpadl create request.
 845  * Find the matching request, copy the response and signal the requesting
 846  * thread.
 847  */
 848 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr)
 849 {
 850         struct vmbus_channel_gpadl_created *gpadlcreated;
 851         struct vmbus_channel_msginfo *msginfo;
 852         struct vmbus_channel_message_header *requestheader;
 853         struct vmbus_channel_gpadl_header *gpadlheader;
 854         unsigned long flags;
 855
 856         gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr;
 857
 858         /*
 859          * Find the establish msg, copy the result and signal/unblock the wait
 860          * event
 861          */
 862         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 863
 864         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 865                                 msglistentry) {
 866                 requestheader =
 867                         (struct vmbus_channel_message_header *)msginfo->msg;
 868
 869                 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) {
 870                         gpadlheader =
 871                         (struct vmbus_channel_gpadl_header *)requestheader;
 872
 873                         if ((gpadlcreated->child_relid ==
 874                              gpadlheader->child_relid) &&
 875                             (gpadlcreated->gpadl == gpadlheader->gpadl)) {
 876                                 memcpy(&msginfo->response.gpadl_created,
 877                                        gpadlcreated,
 878                                        sizeof(
 879                                         struct vmbus_channel_gpadl_created));
 880                                 complete(&msginfo->waitevent);
 881                                 break;
 882                         }
 883                 }
 884         }
 885         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 886 }
 887
 888 /*
 889  * vmbus_ongpadl_torndown - GPADL torndown handler.
 890  *
 891  * This is invoked when we received a response to our gpadl teardown request.
 892  * Find the matching request, copy the response and signal the requesting
 893  * thread.
 894  */
 895 static void vmbus_ongpadl_torndown(
 896                         struct vmbus_channel_message_header *hdr)
 897 {
 898         struct vmbus_channel_gpadl_torndown *gpadl_torndown;
 899         struct vmbus_channel_msginfo *msginfo;
 900         struct vmbus_channel_message_header *requestheader;
 901         struct vmbus_channel_gpadl_teardown *gpadl_teardown;
 902         unsigned long flags;
 903
 904         gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr;
 905
 906         /*
 907          * Find the open msg, copy the result and signal/unblock the wait event
 908          */
 909         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 910
 911         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 912                                 msglistentry) {
 913                 requestheader =
 914                         (struct vmbus_channel_message_header *)msginfo->msg;
 915
 916                 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) {
 917                         gpadl_teardown =
 918                         (struct vmbus_channel_gpadl_teardown *)requestheader;
 919
 920                         if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) {
 921                                 memcpy(&msginfo->response.gpadl_torndown,
 922                                        gpadl_torndown,
 923                                        sizeof(
 924                                         struct vmbus_channel_gpadl_torndown));
 925                                 complete(&msginfo->waitevent);
 926                                 break;
 927                         }
 928                 }
 929         }
 930         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 931 }
 932
 933 /*
 934  * vmbus_onversion_response - Version response handler
 935  *
 936  * This is invoked when we received a response to our initiate contact request.
 937  * Find the matching request, copy the response and signal the requesting
 938  * thread.
 939  */
 940 static void vmbus_onversion_response(
 941                 struct vmbus_channel_message_header *hdr)
 942 {
 943         struct vmbus_channel_msginfo *msginfo;
 944         struct vmbus_channel_message_header *requestheader;
 945         struct vmbus_channel_version_response *version_response;
 946         unsigned long flags;
 947
 948         version_response = (struct vmbus_channel_version_response *)hdr;
 949         spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags);
 950
 951         list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list,
 952                                 msglistentry) {
 953                 requestheader =
 954                         (struct vmbus_channel_message_header *)msginfo->msg;
 955
 956                 if (requestheader->msgtype ==
 957                     CHANNELMSG_INITIATE_CONTACT) {
 958                         memcpy(&msginfo->response.version_response,
 959                               version_response,
 960                               sizeof(struct vmbus_channel_version_response));
 961                         complete(&msginfo->waitevent);
 962                 }
 963         }
 964         spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags);
 965 }
 966
 967 /* Channel message dispatch table */
 968 struct vmbus_channel_message_table_entry
 969         channel_message_table[CHANNELMSG_COUNT] = {
 970         {CHANNELMSG_INVALID,                    0, NULL},
 971         {CHANNELMSG_OFFERCHANNEL,               0, vmbus_onoffer},
 972         {CHANNELMSG_RESCIND_CHANNELOFFER,       0, vmbus_onoffer_rescind},
 973         {CHANNELMSG_REQUESTOFFERS,              0, NULL},
 974         {CHANNELMSG_ALLOFFERS_DELIVERED,        1, vmbus_onoffers_delivered},
 975         {CHANNELMSG_OPENCHANNEL,                0, NULL},
 976         {CHANNELMSG_OPENCHANNEL_RESULT,         1, vmbus_onopen_result},
 977         {CHANNELMSG_CLOSECHANNEL,               0, NULL},
 978         {CHANNELMSG_GPADL_HEADER,               0, NULL},
 979         {CHANNELMSG_GPADL_BODY,                 0, NULL},
 980         {CHANNELMSG_GPADL_CREATED,              1, vmbus_ongpadl_created},
 981         {CHANNELMSG_GPADL_TEARDOWN,             0, NULL},
 982         {CHANNELMSG_GPADL_TORNDOWN,             1, vmbus_ongpadl_torndown},
 983         {CHANNELMSG_RELID_RELEASED,             0, NULL},
 984         {CHANNELMSG_INITIATE_CONTACT,           0, NULL},
 985         {CHANNELMSG_VERSION_RESPONSE,           1, vmbus_onversion_response},
 986         {CHANNELMSG_UNLOAD,                     0, NULL},
 987         {CHANNELMSG_UNLOAD_RESPONSE,            1, vmbus_unload_response},
 988         {CHANNELMSG_18,                         0, NULL},
 989         {CHANNELMSG_19,                         0, NULL},
 990         {CHANNELMSG_20,                         0, NULL},
 991         {CHANNELMSG_TL_CONNECT_REQUEST,         0, NULL},
 992 };
 993
 994 /*
 995  * vmbus_onmessage - Handler for channel protocol messages.
 996  *
 997  * This is invoked in the vmbus worker thread context.
 998  */
 999 void vmbus_onmessage(void *context)
1000 {
1001         struct hv_message *msg = context;
1002         struct vmbus_channel_message_header *hdr;
1003         int size;
1004
1005         hdr = (struct vmbus_channel_message_header *)msg->u.payload;
1006         size = msg->header.payload_size;
1007
1008         if (hdr->msgtype >= CHANNELMSG_COUNT) {
1009                 pr_err("Received invalid channel message type %d size %d\n",
1010                            hdr->msgtype, size);
1011                 print_hex_dump_bytes("", DUMP_PREFIX_NONE,
1012                                      (unsigned char *)msg->u.payload, size);
1013                 return;
1014         }
1015
1016         if (channel_message_table[hdr->msgtype].message_handler)
1017                 channel_message_table[hdr->msgtype].message_handler(hdr);
1018         else
1019                 pr_err("Unhandled channel message type %d\n", hdr->msgtype);
1020 }
1021
1022 /*
1023  * vmbus_request_offers - Send a request to get all our pending offers.
1024  */
1025 int vmbus_request_offers(void)
1026 {
1027         struct vmbus_channel_message_header *msg;
1028         struct vmbus_channel_msginfo *msginfo;
1029         int ret;
1030
1031         msginfo = kmalloc(sizeof(*msginfo) +
1032                           sizeof(struct vmbus_channel_message_header),
1033                           GFP_KERNEL);
1034         if (!msginfo)
1035                 return -ENOMEM;
1036
1037         msg = (struct vmbus_channel_message_header *)msginfo->msg;
1038
1039         msg->msgtype = CHANNELMSG_REQUESTOFFERS;
1040
1041
1042         ret = vmbus_post_msg(msg,
1043                                sizeof(struct vmbus_channel_message_header));
1044         if (ret != 0) {
1045                 pr_err("Unable to request offers - %d\n", ret);
1046
1047                 goto cleanup;
1048         }
1049
1050 cleanup:
1051         kfree(msginfo);
1052
1053         return ret;
1054 }
1055
1056 /*
1057  * Retrieve the (sub) channel on which to send an outgoing request.
1058  * When a primary channel has multiple sub-channels, we try to
1059  * distribute the load equally amongst all available channels.
1060  */
1061 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary)
1062 {
1063         struct list_head *cur, *tmp;
1064         int cur_cpu;
1065         struct vmbus_channel *cur_channel;
1066         struct vmbus_channel *outgoing_channel = primary;
1067         int next_channel;
1068         int i = 1;
1069
1070         if (list_empty(&primary->sc_list))
1071                 return outgoing_channel;
1072
1073         next_channel = primary->next_oc++;
1074
1075         if (next_channel > (primary->num_sc)) {
1076                 primary->next_oc = 0;
1077                 return outgoing_channel;
1078         }
1079
1080         cur_cpu = hv_context.vp_index[get_cpu()];
1081         put_cpu();
1082         list_for_each_safe(cur, tmp, &primary->sc_list) {
1083                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1084                 if (cur_channel->state != CHANNEL_OPENED_STATE)
1085                         continue;
1086
1087                 if (cur_channel->target_vp == cur_cpu)
1088                         return cur_channel;
1089
1090                 if (i == next_channel)
1091                         return cur_channel;
1092
1093                 i++;
1094         }
1095
1096         return outgoing_channel;
1097 }
1098 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel);
1099
1100 static void invoke_sc_cb(struct vmbus_channel *primary_channel)
1101 {
1102         struct list_head *cur, *tmp;
1103         struct vmbus_channel *cur_channel;
1104
1105         if (primary_channel->sc_creation_callback == NULL)
1106                 return;
1107
1108         list_for_each_safe(cur, tmp, &primary_channel->sc_list) {
1109                 cur_channel = list_entry(cur, struct vmbus_channel, sc_list);
1110
1111                 primary_channel->sc_creation_callback(cur_channel);
1112         }
1113 }
1114
1115 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel,
1116                                 void (*sc_cr_cb)(struct vmbus_channel *new_sc))
1117 {
1118         primary_channel->sc_creation_callback = sc_cr_cb;
1119 }
1120 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback);
1121
1122 bool vmbus_are_subchannels_present(struct vmbus_channel *primary)
1123 {
1124         bool ret;
1125
1126         ret = !list_empty(&primary->sc_list);
1127
1128         if (ret) {
1129                 /*
1130                  * Invoke the callback on sub-channel creation.
1131                  * This will present a uniform interface to the
1132                  * clients.
1133                  */
1134                 invoke_sc_cb(primary);
1135         }
1136
1137         return ret;
1138 }
1139 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present);
1140
1141 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel,
1142                 void (*chn_rescind_cb)(struct vmbus_channel *))
1143 {
1144         channel->chn_rescind_callback = chn_rescind_cb;
1145 }
1146 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback);