1 /* 2 * Copyright (c) 2009, Microsoft Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 * 13 * You should have received a copy of the GNU General Public License along with 14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple 15 * Place - Suite 330, Boston, MA 02111-1307 USA. 16 * 17 * Authors: 18 * Haiyang Zhang <haiyangz@microsoft.com> 19 * Hank Janssen <hjanssen@microsoft.com> 20 */ 21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 22 23 #include <linux/kernel.h> 24 #include <linux/sched.h> 25 #include <linux/wait.h> 26 #include <linux/mm.h> 27 #include <linux/slab.h> 28 #include <linux/list.h> 29 #include <linux/module.h> 30 #include <linux/completion.h> 31 #include <linux/hyperv.h> 32 33 #include "hyperv_vmbus.h" 34 35 static void init_vp_index(struct vmbus_channel *channel, 36 const uuid_le *type_guid); 37 38 /** 39 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message 40 * @icmsghdrp: Pointer to msg header structure 41 * @icmsg_negotiate: Pointer to negotiate message structure 42 * @buf: Raw buffer channel data 43 * 44 * @icmsghdrp is of type &struct icmsg_hdr. 45 * @negop is of type &struct icmsg_negotiate. 46 * Set up and fill in default negotiate response message. 47 * 48 * The fw_version specifies the framework version that 49 * we can support and srv_version specifies the service 50 * version we can support. 51 * 52 * Mainly used by Hyper-V drivers. 53 */ 54 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, 55 struct icmsg_negotiate *negop, u8 *buf, 56 int fw_version, int srv_version) 57 { 58 int icframe_major, icframe_minor; 59 int icmsg_major, icmsg_minor; 60 int fw_major, fw_minor; 61 int srv_major, srv_minor; 62 int i; 63 bool found_match = false; 64 65 icmsghdrp->icmsgsize = 0x10; 66 fw_major = (fw_version >> 16); 67 fw_minor = (fw_version & 0xFFFF); 68 69 srv_major = (srv_version >> 16); 70 srv_minor = (srv_version & 0xFFFF); 71 72 negop = (struct icmsg_negotiate *)&buf[ 73 sizeof(struct vmbuspipe_hdr) + 74 sizeof(struct icmsg_hdr)]; 75 76 icframe_major = negop->icframe_vercnt; 77 icframe_minor = 0; 78 79 icmsg_major = negop->icmsg_vercnt; 80 icmsg_minor = 0; 81 82 /* 83 * Select the framework version number we will 84 * support. 85 */ 86 87 for (i = 0; i < negop->icframe_vercnt; i++) { 88 if ((negop->icversion_data[i].major == fw_major) && 89 (negop->icversion_data[i].minor == fw_minor)) { 90 icframe_major = negop->icversion_data[i].major; 91 icframe_minor = negop->icversion_data[i].minor; 92 found_match = true; 93 } 94 } 95 96 if (!found_match) 97 goto fw_error; 98 99 found_match = false; 100 101 for (i = negop->icframe_vercnt; 102 (i < negop->icframe_vercnt + negop->icmsg_vercnt); i++) { 103 if ((negop->icversion_data[i].major == srv_major) && 104 (negop->icversion_data[i].minor == srv_minor)) { 105 icmsg_major = negop->icversion_data[i].major; 106 icmsg_minor = negop->icversion_data[i].minor; 107 found_match = true; 108 } 109 } 110 111 /* 112 * Respond with the framework and service 113 * version numbers we can support. 114 */ 115 116 fw_error: 117 if (!found_match) { 118 negop->icframe_vercnt = 0; 119 negop->icmsg_vercnt = 0; 120 } else { 121 negop->icframe_vercnt = 1; 122 negop->icmsg_vercnt = 1; 123 } 124 125 negop->icversion_data[0].major = icframe_major; 126 negop->icversion_data[0].minor = icframe_minor; 127 negop->icversion_data[1].major = icmsg_major; 128 negop->icversion_data[1].minor = icmsg_minor; 129 return found_match; 130 } 131 132 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp); 133 134 /* 135 * alloc_channel - Allocate and initialize a vmbus channel object 136 */ 137 static struct vmbus_channel *alloc_channel(void) 138 { 139 static atomic_t chan_num = ATOMIC_INIT(0); 140 struct vmbus_channel *channel; 141 142 channel = kzalloc(sizeof(*channel), GFP_ATOMIC); 143 if (!channel) 144 return NULL; 145 146 channel->id = atomic_inc_return(&chan_num); 147 spin_lock_init(&channel->inbound_lock); 148 spin_lock_init(&channel->lock); 149 150 INIT_LIST_HEAD(&channel->sc_list); 151 INIT_LIST_HEAD(&channel->percpu_list); 152 153 return channel; 154 } 155 156 /* 157 * free_channel - Release the resources used by the vmbus channel object 158 */ 159 static void free_channel(struct vmbus_channel *channel) 160 { 161 kfree(channel); 162 } 163 164 static void percpu_channel_enq(void *arg) 165 { 166 struct vmbus_channel *channel = arg; 167 int cpu = smp_processor_id(); 168 169 list_add_tail(&channel->percpu_list, &hv_context.percpu_list[cpu]); 170 } 171 172 static void percpu_channel_deq(void *arg) 173 { 174 struct vmbus_channel *channel = arg; 175 176 list_del(&channel->percpu_list); 177 } 178 179 180 void hv_process_channel_removal(struct vmbus_channel *channel, u32 relid) 181 { 182 struct vmbus_channel_relid_released msg; 183 unsigned long flags; 184 struct vmbus_channel *primary_channel; 185 186 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released)); 187 msg.child_relid = relid; 188 msg.header.msgtype = CHANNELMSG_RELID_RELEASED; 189 vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released)); 190 191 if (channel == NULL) 192 return; 193 194 if (channel->target_cpu != get_cpu()) { 195 put_cpu(); 196 smp_call_function_single(channel->target_cpu, 197 percpu_channel_deq, channel, true); 198 } else { 199 percpu_channel_deq(channel); 200 put_cpu(); 201 } 202 203 if (channel->primary_channel == NULL) { 204 spin_lock_irqsave(&vmbus_connection.channel_lock, flags); 205 list_del(&channel->listentry); 206 spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags); 207 } else { 208 primary_channel = channel->primary_channel; 209 spin_lock_irqsave(&primary_channel->lock, flags); 210 list_del(&channel->sc_list); 211 primary_channel->num_sc--; 212 spin_unlock_irqrestore(&primary_channel->lock, flags); 213 } 214 free_channel(channel); 215 } 216 217 void vmbus_free_channels(void) 218 { 219 struct vmbus_channel *channel, *tmp; 220 221 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list, 222 listentry) { 223 /* if we don't set rescind to true, vmbus_close_internal() 224 * won't invoke hv_process_channel_removal(). 225 */ 226 channel->rescind = true; 227 228 vmbus_device_unregister(channel->device_obj); 229 } 230 } 231 232 /* 233 * vmbus_process_offer - Process the offer by creating a channel/device 234 * associated with this offer 235 */ 236 static void vmbus_process_offer(struct vmbus_channel *newchannel) 237 { 238 struct vmbus_channel *channel; 239 bool fnew = true; 240 unsigned long flags; 241 242 /* Make sure this is a new offer */ 243 spin_lock_irqsave(&vmbus_connection.channel_lock, flags); 244 245 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 246 if (!uuid_le_cmp(channel->offermsg.offer.if_type, 247 newchannel->offermsg.offer.if_type) && 248 !uuid_le_cmp(channel->offermsg.offer.if_instance, 249 newchannel->offermsg.offer.if_instance)) { 250 fnew = false; 251 break; 252 } 253 } 254 255 if (fnew) 256 list_add_tail(&newchannel->listentry, 257 &vmbus_connection.chn_list); 258 259 spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags); 260 261 if (!fnew) { 262 /* 263 * Check to see if this is a sub-channel. 264 */ 265 if (newchannel->offermsg.offer.sub_channel_index != 0) { 266 /* 267 * Process the sub-channel. 268 */ 269 newchannel->primary_channel = channel; 270 spin_lock_irqsave(&channel->lock, flags); 271 list_add_tail(&newchannel->sc_list, &channel->sc_list); 272 channel->num_sc++; 273 spin_unlock_irqrestore(&channel->lock, flags); 274 } else 275 goto err_free_chan; 276 } 277 278 init_vp_index(newchannel, &newchannel->offermsg.offer.if_type); 279 280 if (newchannel->target_cpu != get_cpu()) { 281 put_cpu(); 282 smp_call_function_single(newchannel->target_cpu, 283 percpu_channel_enq, 284 newchannel, true); 285 } else { 286 percpu_channel_enq(newchannel); 287 put_cpu(); 288 } 289 290 /* 291 * This state is used to indicate a successful open 292 * so that when we do close the channel normally, we 293 * can cleanup properly 294 */ 295 newchannel->state = CHANNEL_OPEN_STATE; 296 297 if (!fnew) { 298 if (channel->sc_creation_callback != NULL) 299 channel->sc_creation_callback(newchannel); 300 return; 301 } 302 303 /* 304 * Start the process of binding this offer to the driver 305 * We need to set the DeviceObject field before calling 306 * vmbus_child_dev_add() 307 */ 308 newchannel->device_obj = vmbus_device_create( 309 &newchannel->offermsg.offer.if_type, 310 &newchannel->offermsg.offer.if_instance, 311 newchannel); 312 if (!newchannel->device_obj) 313 goto err_deq_chan; 314 315 /* 316 * Add the new device to the bus. This will kick off device-driver 317 * binding which eventually invokes the device driver's AddDevice() 318 * method. 319 */ 320 if (vmbus_device_register(newchannel->device_obj) != 0) { 321 pr_err("unable to add child device object (relid %d)\n", 322 newchannel->offermsg.child_relid); 323 kfree(newchannel->device_obj); 324 goto err_deq_chan; 325 } 326 return; 327 328 err_deq_chan: 329 spin_lock_irqsave(&vmbus_connection.channel_lock, flags); 330 list_del(&newchannel->listentry); 331 spin_unlock_irqrestore(&vmbus_connection.channel_lock, flags); 332 333 if (newchannel->target_cpu != get_cpu()) { 334 put_cpu(); 335 smp_call_function_single(newchannel->target_cpu, 336 percpu_channel_deq, newchannel, true); 337 } else { 338 percpu_channel_deq(newchannel); 339 put_cpu(); 340 } 341 342 err_free_chan: 343 free_channel(newchannel); 344 } 345 346 enum { 347 IDE = 0, 348 SCSI, 349 NIC, 350 ND_NIC, 351 MAX_PERF_CHN, 352 }; 353 354 /* 355 * This is an array of device_ids (device types) that are performance critical. 356 * We attempt to distribute the interrupt load for these devices across 357 * all available CPUs. 358 */ 359 static const struct hv_vmbus_device_id hp_devs[] = { 360 /* IDE */ 361 { HV_IDE_GUID, }, 362 /* Storage - SCSI */ 363 { HV_SCSI_GUID, }, 364 /* Network */ 365 { HV_NIC_GUID, }, 366 /* NetworkDirect Guest RDMA */ 367 { HV_ND_GUID, }, 368 }; 369 370 371 /* 372 * We use this state to statically distribute the channel interrupt load. 373 */ 374 static int next_numa_node_id; 375 376 /* 377 * Starting with Win8, we can statically distribute the incoming 378 * channel interrupt load by binding a channel to VCPU. 379 * We do this in a hierarchical fashion: 380 * First distribute the primary channels across available NUMA nodes 381 * and then distribute the subchannels amongst the CPUs in the NUMA 382 * node assigned to the primary channel. 383 * 384 * For pre-win8 hosts or non-performance critical channels we assign the 385 * first CPU in the first NUMA node. 386 */ 387 static void init_vp_index(struct vmbus_channel *channel, const uuid_le *type_guid) 388 { 389 u32 cur_cpu; 390 int i; 391 bool perf_chn = false; 392 struct vmbus_channel *primary = channel->primary_channel; 393 int next_node; 394 struct cpumask available_mask; 395 struct cpumask *alloced_mask; 396 397 for (i = IDE; i < MAX_PERF_CHN; i++) { 398 if (!memcmp(type_guid->b, hp_devs[i].guid, 399 sizeof(uuid_le))) { 400 perf_chn = true; 401 break; 402 } 403 } 404 if ((vmbus_proto_version == VERSION_WS2008) || 405 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) { 406 /* 407 * Prior to win8, all channel interrupts are 408 * delivered on cpu 0. 409 * Also if the channel is not a performance critical 410 * channel, bind it to cpu 0. 411 */ 412 channel->numa_node = 0; 413 channel->target_cpu = 0; 414 channel->target_vp = hv_context.vp_index[0]; 415 return; 416 } 417 418 /* 419 * We distribute primary channels evenly across all the available 420 * NUMA nodes and within the assigned NUMA node we will assign the 421 * first available CPU to the primary channel. 422 * The sub-channels will be assigned to the CPUs available in the 423 * NUMA node evenly. 424 */ 425 if (!primary) { 426 while (true) { 427 next_node = next_numa_node_id++; 428 if (next_node == nr_node_ids) 429 next_node = next_numa_node_id = 0; 430 if (cpumask_empty(cpumask_of_node(next_node))) 431 continue; 432 break; 433 } 434 channel->numa_node = next_node; 435 primary = channel; 436 } 437 alloced_mask = &hv_context.hv_numa_map[primary->numa_node]; 438 439 if (cpumask_weight(alloced_mask) == 440 cpumask_weight(cpumask_of_node(primary->numa_node))) { 441 /* 442 * We have cycled through all the CPUs in the node; 443 * reset the alloced map. 444 */ 445 cpumask_clear(alloced_mask); 446 } 447 448 cpumask_xor(&available_mask, alloced_mask, 449 cpumask_of_node(primary->numa_node)); 450 451 cur_cpu = cpumask_next(-1, &available_mask); 452 cpumask_set_cpu(cur_cpu, alloced_mask); 453 454 channel->target_cpu = cur_cpu; 455 channel->target_vp = hv_context.vp_index[cur_cpu]; 456 } 457 458 /* 459 * vmbus_unload_response - Handler for the unload response. 460 */ 461 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr) 462 { 463 /* 464 * This is a global event; just wakeup the waiting thread. 465 * Once we successfully unload, we can cleanup the monitor state. 466 */ 467 complete(&vmbus_connection.unload_event); 468 } 469 470 void vmbus_initiate_unload(void) 471 { 472 struct vmbus_channel_message_header hdr; 473 474 /* Pre-Win2012R2 hosts don't support reconnect */ 475 if (vmbus_proto_version < VERSION_WIN8_1) 476 return; 477 478 init_completion(&vmbus_connection.unload_event); 479 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header)); 480 hdr.msgtype = CHANNELMSG_UNLOAD; 481 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header)); 482 483 wait_for_completion(&vmbus_connection.unload_event); 484 } 485 486 /* 487 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition. 488 * 489 */ 490 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr) 491 { 492 struct vmbus_channel_offer_channel *offer; 493 struct vmbus_channel *newchannel; 494 495 offer = (struct vmbus_channel_offer_channel *)hdr; 496 497 /* Allocate the channel object and save this offer. */ 498 newchannel = alloc_channel(); 499 if (!newchannel) { 500 pr_err("Unable to allocate channel object\n"); 501 return; 502 } 503 504 /* 505 * By default we setup state to enable batched 506 * reading. A specific service can choose to 507 * disable this prior to opening the channel. 508 */ 509 newchannel->batched_reading = true; 510 511 /* 512 * Setup state for signalling the host. 513 */ 514 newchannel->sig_event = (struct hv_input_signal_event *) 515 (ALIGN((unsigned long) 516 &newchannel->sig_buf, 517 HV_HYPERCALL_PARAM_ALIGN)); 518 519 newchannel->sig_event->connectionid.asu32 = 0; 520 newchannel->sig_event->connectionid.u.id = VMBUS_EVENT_CONNECTION_ID; 521 newchannel->sig_event->flag_number = 0; 522 newchannel->sig_event->rsvdz = 0; 523 524 if (vmbus_proto_version != VERSION_WS2008) { 525 newchannel->is_dedicated_interrupt = 526 (offer->is_dedicated_interrupt != 0); 527 newchannel->sig_event->connectionid.u.id = 528 offer->connection_id; 529 } 530 531 memcpy(&newchannel->offermsg, offer, 532 sizeof(struct vmbus_channel_offer_channel)); 533 newchannel->monitor_grp = (u8)offer->monitorid / 32; 534 newchannel->monitor_bit = (u8)offer->monitorid % 32; 535 536 vmbus_process_offer(newchannel); 537 } 538 539 /* 540 * vmbus_onoffer_rescind - Rescind offer handler. 541 * 542 * We queue a work item to process this offer synchronously 543 */ 544 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr) 545 { 546 struct vmbus_channel_rescind_offer *rescind; 547 struct vmbus_channel *channel; 548 unsigned long flags; 549 struct device *dev; 550 551 rescind = (struct vmbus_channel_rescind_offer *)hdr; 552 channel = relid2channel(rescind->child_relid); 553 554 if (channel == NULL) { 555 hv_process_channel_removal(NULL, rescind->child_relid); 556 return; 557 } 558 559 spin_lock_irqsave(&channel->lock, flags); 560 channel->rescind = true; 561 spin_unlock_irqrestore(&channel->lock, flags); 562 563 if (channel->device_obj) { 564 /* 565 * We will have to unregister this device from the 566 * driver core. 567 */ 568 dev = get_device(&channel->device_obj->device); 569 if (dev) { 570 vmbus_device_unregister(channel->device_obj); 571 put_device(dev); 572 } 573 } else { 574 hv_process_channel_removal(channel, 575 channel->offermsg.child_relid); 576 } 577 } 578 579 /* 580 * vmbus_onoffers_delivered - 581 * This is invoked when all offers have been delivered. 582 * 583 * Nothing to do here. 584 */ 585 static void vmbus_onoffers_delivered( 586 struct vmbus_channel_message_header *hdr) 587 { 588 } 589 590 /* 591 * vmbus_onopen_result - Open result handler. 592 * 593 * This is invoked when we received a response to our channel open request. 594 * Find the matching request, copy the response and signal the requesting 595 * thread. 596 */ 597 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr) 598 { 599 struct vmbus_channel_open_result *result; 600 struct vmbus_channel_msginfo *msginfo; 601 struct vmbus_channel_message_header *requestheader; 602 struct vmbus_channel_open_channel *openmsg; 603 unsigned long flags; 604 605 result = (struct vmbus_channel_open_result *)hdr; 606 607 /* 608 * Find the open msg, copy the result and signal/unblock the wait event 609 */ 610 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 611 612 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 613 msglistentry) { 614 requestheader = 615 (struct vmbus_channel_message_header *)msginfo->msg; 616 617 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) { 618 openmsg = 619 (struct vmbus_channel_open_channel *)msginfo->msg; 620 if (openmsg->child_relid == result->child_relid && 621 openmsg->openid == result->openid) { 622 memcpy(&msginfo->response.open_result, 623 result, 624 sizeof( 625 struct vmbus_channel_open_result)); 626 complete(&msginfo->waitevent); 627 break; 628 } 629 } 630 } 631 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 632 } 633 634 /* 635 * vmbus_ongpadl_created - GPADL created handler. 636 * 637 * This is invoked when we received a response to our gpadl create request. 638 * Find the matching request, copy the response and signal the requesting 639 * thread. 640 */ 641 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr) 642 { 643 struct vmbus_channel_gpadl_created *gpadlcreated; 644 struct vmbus_channel_msginfo *msginfo; 645 struct vmbus_channel_message_header *requestheader; 646 struct vmbus_channel_gpadl_header *gpadlheader; 647 unsigned long flags; 648 649 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr; 650 651 /* 652 * Find the establish msg, copy the result and signal/unblock the wait 653 * event 654 */ 655 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 656 657 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 658 msglistentry) { 659 requestheader = 660 (struct vmbus_channel_message_header *)msginfo->msg; 661 662 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) { 663 gpadlheader = 664 (struct vmbus_channel_gpadl_header *)requestheader; 665 666 if ((gpadlcreated->child_relid == 667 gpadlheader->child_relid) && 668 (gpadlcreated->gpadl == gpadlheader->gpadl)) { 669 memcpy(&msginfo->response.gpadl_created, 670 gpadlcreated, 671 sizeof( 672 struct vmbus_channel_gpadl_created)); 673 complete(&msginfo->waitevent); 674 break; 675 } 676 } 677 } 678 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 679 } 680 681 /* 682 * vmbus_ongpadl_torndown - GPADL torndown handler. 683 * 684 * This is invoked when we received a response to our gpadl teardown request. 685 * Find the matching request, copy the response and signal the requesting 686 * thread. 687 */ 688 static void vmbus_ongpadl_torndown( 689 struct vmbus_channel_message_header *hdr) 690 { 691 struct vmbus_channel_gpadl_torndown *gpadl_torndown; 692 struct vmbus_channel_msginfo *msginfo; 693 struct vmbus_channel_message_header *requestheader; 694 struct vmbus_channel_gpadl_teardown *gpadl_teardown; 695 unsigned long flags; 696 697 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr; 698 699 /* 700 * Find the open msg, copy the result and signal/unblock the wait event 701 */ 702 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 703 704 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 705 msglistentry) { 706 requestheader = 707 (struct vmbus_channel_message_header *)msginfo->msg; 708 709 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) { 710 gpadl_teardown = 711 (struct vmbus_channel_gpadl_teardown *)requestheader; 712 713 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) { 714 memcpy(&msginfo->response.gpadl_torndown, 715 gpadl_torndown, 716 sizeof( 717 struct vmbus_channel_gpadl_torndown)); 718 complete(&msginfo->waitevent); 719 break; 720 } 721 } 722 } 723 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 724 } 725 726 /* 727 * vmbus_onversion_response - Version response handler 728 * 729 * This is invoked when we received a response to our initiate contact request. 730 * Find the matching request, copy the response and signal the requesting 731 * thread. 732 */ 733 static void vmbus_onversion_response( 734 struct vmbus_channel_message_header *hdr) 735 { 736 struct vmbus_channel_msginfo *msginfo; 737 struct vmbus_channel_message_header *requestheader; 738 struct vmbus_channel_version_response *version_response; 739 unsigned long flags; 740 741 version_response = (struct vmbus_channel_version_response *)hdr; 742 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 743 744 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 745 msglistentry) { 746 requestheader = 747 (struct vmbus_channel_message_header *)msginfo->msg; 748 749 if (requestheader->msgtype == 750 CHANNELMSG_INITIATE_CONTACT) { 751 memcpy(&msginfo->response.version_response, 752 version_response, 753 sizeof(struct vmbus_channel_version_response)); 754 complete(&msginfo->waitevent); 755 } 756 } 757 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 758 } 759 760 /* Channel message dispatch table */ 761 struct vmbus_channel_message_table_entry 762 channel_message_table[CHANNELMSG_COUNT] = { 763 {CHANNELMSG_INVALID, 0, NULL}, 764 {CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer}, 765 {CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind}, 766 {CHANNELMSG_REQUESTOFFERS, 0, NULL}, 767 {CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered}, 768 {CHANNELMSG_OPENCHANNEL, 0, NULL}, 769 {CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result}, 770 {CHANNELMSG_CLOSECHANNEL, 0, NULL}, 771 {CHANNELMSG_GPADL_HEADER, 0, NULL}, 772 {CHANNELMSG_GPADL_BODY, 0, NULL}, 773 {CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created}, 774 {CHANNELMSG_GPADL_TEARDOWN, 0, NULL}, 775 {CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown}, 776 {CHANNELMSG_RELID_RELEASED, 0, NULL}, 777 {CHANNELMSG_INITIATE_CONTACT, 0, NULL}, 778 {CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response}, 779 {CHANNELMSG_UNLOAD, 0, NULL}, 780 {CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response}, 781 }; 782 783 /* 784 * vmbus_onmessage - Handler for channel protocol messages. 785 * 786 * This is invoked in the vmbus worker thread context. 787 */ 788 void vmbus_onmessage(void *context) 789 { 790 struct hv_message *msg = context; 791 struct vmbus_channel_message_header *hdr; 792 int size; 793 794 hdr = (struct vmbus_channel_message_header *)msg->u.payload; 795 size = msg->header.payload_size; 796 797 if (hdr->msgtype >= CHANNELMSG_COUNT) { 798 pr_err("Received invalid channel message type %d size %d\n", 799 hdr->msgtype, size); 800 print_hex_dump_bytes("", DUMP_PREFIX_NONE, 801 (unsigned char *)msg->u.payload, size); 802 return; 803 } 804 805 if (channel_message_table[hdr->msgtype].message_handler) 806 channel_message_table[hdr->msgtype].message_handler(hdr); 807 else 808 pr_err("Unhandled channel message type %d\n", hdr->msgtype); 809 } 810 811 /* 812 * vmbus_request_offers - Send a request to get all our pending offers. 813 */ 814 int vmbus_request_offers(void) 815 { 816 struct vmbus_channel_message_header *msg; 817 struct vmbus_channel_msginfo *msginfo; 818 int ret; 819 820 msginfo = kmalloc(sizeof(*msginfo) + 821 sizeof(struct vmbus_channel_message_header), 822 GFP_KERNEL); 823 if (!msginfo) 824 return -ENOMEM; 825 826 msg = (struct vmbus_channel_message_header *)msginfo->msg; 827 828 msg->msgtype = CHANNELMSG_REQUESTOFFERS; 829 830 831 ret = vmbus_post_msg(msg, 832 sizeof(struct vmbus_channel_message_header)); 833 if (ret != 0) { 834 pr_err("Unable to request offers - %d\n", ret); 835 836 goto cleanup; 837 } 838 839 cleanup: 840 kfree(msginfo); 841 842 return ret; 843 } 844 845 /* 846 * Retrieve the (sub) channel on which to send an outgoing request. 847 * When a primary channel has multiple sub-channels, we try to 848 * distribute the load equally amongst all available channels. 849 */ 850 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary) 851 { 852 struct list_head *cur, *tmp; 853 int cur_cpu; 854 struct vmbus_channel *cur_channel; 855 struct vmbus_channel *outgoing_channel = primary; 856 int next_channel; 857 int i = 1; 858 859 if (list_empty(&primary->sc_list)) 860 return outgoing_channel; 861 862 next_channel = primary->next_oc++; 863 864 if (next_channel > (primary->num_sc)) { 865 primary->next_oc = 0; 866 return outgoing_channel; 867 } 868 869 cur_cpu = hv_context.vp_index[get_cpu()]; 870 put_cpu(); 871 list_for_each_safe(cur, tmp, &primary->sc_list) { 872 cur_channel = list_entry(cur, struct vmbus_channel, sc_list); 873 if (cur_channel->state != CHANNEL_OPENED_STATE) 874 continue; 875 876 if (cur_channel->target_vp == cur_cpu) 877 return cur_channel; 878 879 if (i == next_channel) 880 return cur_channel; 881 882 i++; 883 } 884 885 return outgoing_channel; 886 } 887 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel); 888 889 static void invoke_sc_cb(struct vmbus_channel *primary_channel) 890 { 891 struct list_head *cur, *tmp; 892 struct vmbus_channel *cur_channel; 893 894 if (primary_channel->sc_creation_callback == NULL) 895 return; 896 897 list_for_each_safe(cur, tmp, &primary_channel->sc_list) { 898 cur_channel = list_entry(cur, struct vmbus_channel, sc_list); 899 900 primary_channel->sc_creation_callback(cur_channel); 901 } 902 } 903 904 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel, 905 void (*sc_cr_cb)(struct vmbus_channel *new_sc)) 906 { 907 primary_channel->sc_creation_callback = sc_cr_cb; 908 } 909 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback); 910 911 bool vmbus_are_subchannels_present(struct vmbus_channel *primary) 912 { 913 bool ret; 914 915 ret = !list_empty(&primary->sc_list); 916 917 if (ret) { 918 /* 919 * Invoke the callback on sub-channel creation. 920 * This will present a uniform interface to the 921 * clients. 922 */ 923 invoke_sc_cb(primary); 924 } 925 926 return ret; 927 } 928 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present); 929