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