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/interrupt.h> 25 #include <linux/sched.h> 26 #include <linux/wait.h> 27 #include <linux/mm.h> 28 #include <linux/slab.h> 29 #include <linux/list.h> 30 #include <linux/module.h> 31 #include <linux/completion.h> 32 #include <linux/delay.h> 33 #include <linux/hyperv.h> 34 #include <asm/mshyperv.h> 35 36 #include "hyperv_vmbus.h" 37 38 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type); 39 40 static const struct vmbus_device vmbus_devs[] = { 41 /* IDE */ 42 { .dev_type = HV_IDE, 43 HV_IDE_GUID, 44 .perf_device = true, 45 }, 46 47 /* SCSI */ 48 { .dev_type = HV_SCSI, 49 HV_SCSI_GUID, 50 .perf_device = true, 51 }, 52 53 /* Fibre Channel */ 54 { .dev_type = HV_FC, 55 HV_SYNTHFC_GUID, 56 .perf_device = true, 57 }, 58 59 /* Synthetic NIC */ 60 { .dev_type = HV_NIC, 61 HV_NIC_GUID, 62 .perf_device = true, 63 }, 64 65 /* Network Direct */ 66 { .dev_type = HV_ND, 67 HV_ND_GUID, 68 .perf_device = true, 69 }, 70 71 /* PCIE */ 72 { .dev_type = HV_PCIE, 73 HV_PCIE_GUID, 74 .perf_device = true, 75 }, 76 77 /* Synthetic Frame Buffer */ 78 { .dev_type = HV_FB, 79 HV_SYNTHVID_GUID, 80 .perf_device = false, 81 }, 82 83 /* Synthetic Keyboard */ 84 { .dev_type = HV_KBD, 85 HV_KBD_GUID, 86 .perf_device = false, 87 }, 88 89 /* Synthetic MOUSE */ 90 { .dev_type = HV_MOUSE, 91 HV_MOUSE_GUID, 92 .perf_device = false, 93 }, 94 95 /* KVP */ 96 { .dev_type = HV_KVP, 97 HV_KVP_GUID, 98 .perf_device = false, 99 }, 100 101 /* Time Synch */ 102 { .dev_type = HV_TS, 103 HV_TS_GUID, 104 .perf_device = false, 105 }, 106 107 /* Heartbeat */ 108 { .dev_type = HV_HB, 109 HV_HEART_BEAT_GUID, 110 .perf_device = false, 111 }, 112 113 /* Shutdown */ 114 { .dev_type = HV_SHUTDOWN, 115 HV_SHUTDOWN_GUID, 116 .perf_device = false, 117 }, 118 119 /* File copy */ 120 { .dev_type = HV_FCOPY, 121 HV_FCOPY_GUID, 122 .perf_device = false, 123 }, 124 125 /* Backup */ 126 { .dev_type = HV_BACKUP, 127 HV_VSS_GUID, 128 .perf_device = false, 129 }, 130 131 /* Dynamic Memory */ 132 { .dev_type = HV_DM, 133 HV_DM_GUID, 134 .perf_device = false, 135 }, 136 137 /* Unknown GUID */ 138 { .dev_type = HV_UNKNOWN, 139 .perf_device = false, 140 }, 141 }; 142 143 static const struct { 144 uuid_le guid; 145 } vmbus_unsupported_devs[] = { 146 { HV_AVMA1_GUID }, 147 { HV_AVMA2_GUID }, 148 { HV_RDV_GUID }, 149 }; 150 151 /* 152 * The rescinded channel may be blocked waiting for a response from the host; 153 * take care of that. 154 */ 155 static void vmbus_rescind_cleanup(struct vmbus_channel *channel) 156 { 157 struct vmbus_channel_msginfo *msginfo; 158 unsigned long flags; 159 160 161 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 162 channel->rescind = true; 163 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 164 msglistentry) { 165 166 if (msginfo->waiting_channel == channel) { 167 complete(&msginfo->waitevent); 168 break; 169 } 170 } 171 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 172 } 173 174 static bool is_unsupported_vmbus_devs(const uuid_le *guid) 175 { 176 int i; 177 178 for (i = 0; i < ARRAY_SIZE(vmbus_unsupported_devs); i++) 179 if (!uuid_le_cmp(*guid, vmbus_unsupported_devs[i].guid)) 180 return true; 181 return false; 182 } 183 184 static u16 hv_get_dev_type(const struct vmbus_channel *channel) 185 { 186 const uuid_le *guid = &channel->offermsg.offer.if_type; 187 u16 i; 188 189 if (is_hvsock_channel(channel) || is_unsupported_vmbus_devs(guid)) 190 return HV_UNKNOWN; 191 192 for (i = HV_IDE; i < HV_UNKNOWN; i++) { 193 if (!uuid_le_cmp(*guid, vmbus_devs[i].guid)) 194 return i; 195 } 196 pr_info("Unknown GUID: %pUl\n", guid); 197 return i; 198 } 199 200 /** 201 * vmbus_prep_negotiate_resp() - Create default response for Hyper-V Negotiate message 202 * @icmsghdrp: Pointer to msg header structure 203 * @icmsg_negotiate: Pointer to negotiate message structure 204 * @buf: Raw buffer channel data 205 * 206 * @icmsghdrp is of type &struct icmsg_hdr. 207 * Set up and fill in default negotiate response message. 208 * 209 * The fw_version and fw_vercnt specifies the framework version that 210 * we can support. 211 * 212 * The srv_version and srv_vercnt specifies the service 213 * versions we can support. 214 * 215 * Versions are given in decreasing order. 216 * 217 * nego_fw_version and nego_srv_version store the selected protocol versions. 218 * 219 * Mainly used by Hyper-V drivers. 220 */ 221 bool vmbus_prep_negotiate_resp(struct icmsg_hdr *icmsghdrp, 222 u8 *buf, const int *fw_version, int fw_vercnt, 223 const int *srv_version, int srv_vercnt, 224 int *nego_fw_version, int *nego_srv_version) 225 { 226 int icframe_major, icframe_minor; 227 int icmsg_major, icmsg_minor; 228 int fw_major, fw_minor; 229 int srv_major, srv_minor; 230 int i, j; 231 bool found_match = false; 232 struct icmsg_negotiate *negop; 233 234 icmsghdrp->icmsgsize = 0x10; 235 negop = (struct icmsg_negotiate *)&buf[ 236 sizeof(struct vmbuspipe_hdr) + 237 sizeof(struct icmsg_hdr)]; 238 239 icframe_major = negop->icframe_vercnt; 240 icframe_minor = 0; 241 242 icmsg_major = negop->icmsg_vercnt; 243 icmsg_minor = 0; 244 245 /* 246 * Select the framework version number we will 247 * support. 248 */ 249 250 for (i = 0; i < fw_vercnt; i++) { 251 fw_major = (fw_version[i] >> 16); 252 fw_minor = (fw_version[i] & 0xFFFF); 253 254 for (j = 0; j < negop->icframe_vercnt; j++) { 255 if ((negop->icversion_data[j].major == fw_major) && 256 (negop->icversion_data[j].minor == fw_minor)) { 257 icframe_major = negop->icversion_data[j].major; 258 icframe_minor = negop->icversion_data[j].minor; 259 found_match = true; 260 break; 261 } 262 } 263 264 if (found_match) 265 break; 266 } 267 268 if (!found_match) 269 goto fw_error; 270 271 found_match = false; 272 273 for (i = 0; i < srv_vercnt; i++) { 274 srv_major = (srv_version[i] >> 16); 275 srv_minor = (srv_version[i] & 0xFFFF); 276 277 for (j = negop->icframe_vercnt; 278 (j < negop->icframe_vercnt + negop->icmsg_vercnt); 279 j++) { 280 281 if ((negop->icversion_data[j].major == srv_major) && 282 (negop->icversion_data[j].minor == srv_minor)) { 283 284 icmsg_major = negop->icversion_data[j].major; 285 icmsg_minor = negop->icversion_data[j].minor; 286 found_match = true; 287 break; 288 } 289 } 290 291 if (found_match) 292 break; 293 } 294 295 /* 296 * Respond with the framework and service 297 * version numbers we can support. 298 */ 299 300 fw_error: 301 if (!found_match) { 302 negop->icframe_vercnt = 0; 303 negop->icmsg_vercnt = 0; 304 } else { 305 negop->icframe_vercnt = 1; 306 negop->icmsg_vercnt = 1; 307 } 308 309 if (nego_fw_version) 310 *nego_fw_version = (icframe_major << 16) | icframe_minor; 311 312 if (nego_srv_version) 313 *nego_srv_version = (icmsg_major << 16) | icmsg_minor; 314 315 negop->icversion_data[0].major = icframe_major; 316 negop->icversion_data[0].minor = icframe_minor; 317 negop->icversion_data[1].major = icmsg_major; 318 negop->icversion_data[1].minor = icmsg_minor; 319 return found_match; 320 } 321 322 EXPORT_SYMBOL_GPL(vmbus_prep_negotiate_resp); 323 324 /* 325 * alloc_channel - Allocate and initialize a vmbus channel object 326 */ 327 static struct vmbus_channel *alloc_channel(void) 328 { 329 struct vmbus_channel *channel; 330 331 channel = kzalloc(sizeof(*channel), GFP_ATOMIC); 332 if (!channel) 333 return NULL; 334 335 spin_lock_init(&channel->lock); 336 337 INIT_LIST_HEAD(&channel->sc_list); 338 INIT_LIST_HEAD(&channel->percpu_list); 339 340 tasklet_init(&channel->callback_event, 341 vmbus_on_event, (unsigned long)channel); 342 343 return channel; 344 } 345 346 /* 347 * free_channel - Release the resources used by the vmbus channel object 348 */ 349 static void free_channel(struct vmbus_channel *channel) 350 { 351 tasklet_kill(&channel->callback_event); 352 353 kobject_put(&channel->kobj); 354 } 355 356 static void percpu_channel_enq(void *arg) 357 { 358 struct vmbus_channel *channel = arg; 359 struct hv_per_cpu_context *hv_cpu 360 = this_cpu_ptr(hv_context.cpu_context); 361 362 list_add_tail_rcu(&channel->percpu_list, &hv_cpu->chan_list); 363 } 364 365 static void percpu_channel_deq(void *arg) 366 { 367 struct vmbus_channel *channel = arg; 368 369 list_del_rcu(&channel->percpu_list); 370 } 371 372 373 static void vmbus_release_relid(u32 relid) 374 { 375 struct vmbus_channel_relid_released msg; 376 377 memset(&msg, 0, sizeof(struct vmbus_channel_relid_released)); 378 msg.child_relid = relid; 379 msg.header.msgtype = CHANNELMSG_RELID_RELEASED; 380 vmbus_post_msg(&msg, sizeof(struct vmbus_channel_relid_released), 381 true); 382 } 383 384 void hv_process_channel_removal(u32 relid) 385 { 386 unsigned long flags; 387 struct vmbus_channel *primary_channel, *channel; 388 389 BUG_ON(!mutex_is_locked(&vmbus_connection.channel_mutex)); 390 391 /* 392 * Make sure channel is valid as we may have raced. 393 */ 394 channel = relid2channel(relid); 395 if (!channel) 396 return; 397 398 BUG_ON(!channel->rescind); 399 if (channel->target_cpu != get_cpu()) { 400 put_cpu(); 401 smp_call_function_single(channel->target_cpu, 402 percpu_channel_deq, channel, true); 403 } else { 404 percpu_channel_deq(channel); 405 put_cpu(); 406 } 407 408 if (channel->primary_channel == NULL) { 409 list_del(&channel->listentry); 410 411 primary_channel = channel; 412 } else { 413 primary_channel = channel->primary_channel; 414 spin_lock_irqsave(&primary_channel->lock, flags); 415 list_del(&channel->sc_list); 416 primary_channel->num_sc--; 417 spin_unlock_irqrestore(&primary_channel->lock, flags); 418 } 419 420 /* 421 * We need to free the bit for init_vp_index() to work in the case 422 * of sub-channel, when we reload drivers like hv_netvsc. 423 */ 424 if (channel->affinity_policy == HV_LOCALIZED) 425 cpumask_clear_cpu(channel->target_cpu, 426 &primary_channel->alloced_cpus_in_node); 427 428 vmbus_release_relid(relid); 429 430 free_channel(channel); 431 } 432 433 void vmbus_free_channels(void) 434 { 435 struct vmbus_channel *channel, *tmp; 436 437 list_for_each_entry_safe(channel, tmp, &vmbus_connection.chn_list, 438 listentry) { 439 /* hv_process_channel_removal() needs this */ 440 channel->rescind = true; 441 442 vmbus_device_unregister(channel->device_obj); 443 } 444 } 445 446 /* 447 * vmbus_process_offer - Process the offer by creating a channel/device 448 * associated with this offer 449 */ 450 static void vmbus_process_offer(struct vmbus_channel *newchannel) 451 { 452 struct vmbus_channel *channel; 453 bool fnew = true; 454 unsigned long flags; 455 u16 dev_type; 456 int ret; 457 458 /* Make sure this is a new offer */ 459 mutex_lock(&vmbus_connection.channel_mutex); 460 461 /* 462 * Now that we have acquired the channel_mutex, 463 * we can release the potentially racing rescind thread. 464 */ 465 atomic_dec(&vmbus_connection.offer_in_progress); 466 467 list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) { 468 if (!uuid_le_cmp(channel->offermsg.offer.if_type, 469 newchannel->offermsg.offer.if_type) && 470 !uuid_le_cmp(channel->offermsg.offer.if_instance, 471 newchannel->offermsg.offer.if_instance)) { 472 fnew = false; 473 break; 474 } 475 } 476 477 if (fnew) 478 list_add_tail(&newchannel->listentry, 479 &vmbus_connection.chn_list); 480 481 mutex_unlock(&vmbus_connection.channel_mutex); 482 483 if (!fnew) { 484 /* 485 * Check to see if this is a sub-channel. 486 */ 487 if (newchannel->offermsg.offer.sub_channel_index != 0) { 488 /* 489 * Process the sub-channel. 490 */ 491 newchannel->primary_channel = channel; 492 spin_lock_irqsave(&channel->lock, flags); 493 list_add_tail(&newchannel->sc_list, &channel->sc_list); 494 channel->num_sc++; 495 spin_unlock_irqrestore(&channel->lock, flags); 496 } else { 497 goto err_free_chan; 498 } 499 } 500 501 dev_type = hv_get_dev_type(newchannel); 502 503 init_vp_index(newchannel, dev_type); 504 505 if (newchannel->target_cpu != get_cpu()) { 506 put_cpu(); 507 smp_call_function_single(newchannel->target_cpu, 508 percpu_channel_enq, 509 newchannel, true); 510 } else { 511 percpu_channel_enq(newchannel); 512 put_cpu(); 513 } 514 515 /* 516 * This state is used to indicate a successful open 517 * so that when we do close the channel normally, we 518 * can cleanup properly 519 */ 520 newchannel->state = CHANNEL_OPEN_STATE; 521 522 if (!fnew) { 523 struct hv_device *dev 524 = newchannel->primary_channel->device_obj; 525 526 if (vmbus_add_channel_kobj(dev, newchannel)) { 527 atomic_dec(&vmbus_connection.offer_in_progress); 528 goto err_free_chan; 529 } 530 531 if (channel->sc_creation_callback != NULL) 532 channel->sc_creation_callback(newchannel); 533 newchannel->probe_done = true; 534 return; 535 } 536 537 /* 538 * Start the process of binding this offer to the driver 539 * We need to set the DeviceObject field before calling 540 * vmbus_child_dev_add() 541 */ 542 newchannel->device_obj = vmbus_device_create( 543 &newchannel->offermsg.offer.if_type, 544 &newchannel->offermsg.offer.if_instance, 545 newchannel); 546 if (!newchannel->device_obj) 547 goto err_deq_chan; 548 549 newchannel->device_obj->device_id = dev_type; 550 /* 551 * Add the new device to the bus. This will kick off device-driver 552 * binding which eventually invokes the device driver's AddDevice() 553 * method. 554 */ 555 ret = vmbus_device_register(newchannel->device_obj); 556 557 if (ret != 0) { 558 pr_err("unable to add child device object (relid %d)\n", 559 newchannel->offermsg.child_relid); 560 kfree(newchannel->device_obj); 561 goto err_deq_chan; 562 } 563 564 newchannel->probe_done = true; 565 return; 566 567 err_deq_chan: 568 mutex_lock(&vmbus_connection.channel_mutex); 569 list_del(&newchannel->listentry); 570 mutex_unlock(&vmbus_connection.channel_mutex); 571 572 if (newchannel->target_cpu != get_cpu()) { 573 put_cpu(); 574 smp_call_function_single(newchannel->target_cpu, 575 percpu_channel_deq, newchannel, true); 576 } else { 577 percpu_channel_deq(newchannel); 578 put_cpu(); 579 } 580 581 vmbus_release_relid(newchannel->offermsg.child_relid); 582 583 err_free_chan: 584 free_channel(newchannel); 585 } 586 587 /* 588 * We use this state to statically distribute the channel interrupt load. 589 */ 590 static int next_numa_node_id; 591 592 /* 593 * Starting with Win8, we can statically distribute the incoming 594 * channel interrupt load by binding a channel to VCPU. 595 * We do this in a hierarchical fashion: 596 * First distribute the primary channels across available NUMA nodes 597 * and then distribute the subchannels amongst the CPUs in the NUMA 598 * node assigned to the primary channel. 599 * 600 * For pre-win8 hosts or non-performance critical channels we assign the 601 * first CPU in the first NUMA node. 602 */ 603 static void init_vp_index(struct vmbus_channel *channel, u16 dev_type) 604 { 605 u32 cur_cpu; 606 bool perf_chn = vmbus_devs[dev_type].perf_device; 607 struct vmbus_channel *primary = channel->primary_channel; 608 int next_node; 609 struct cpumask available_mask; 610 struct cpumask *alloced_mask; 611 612 if ((vmbus_proto_version == VERSION_WS2008) || 613 (vmbus_proto_version == VERSION_WIN7) || (!perf_chn)) { 614 /* 615 * Prior to win8, all channel interrupts are 616 * delivered on cpu 0. 617 * Also if the channel is not a performance critical 618 * channel, bind it to cpu 0. 619 */ 620 channel->numa_node = 0; 621 channel->target_cpu = 0; 622 channel->target_vp = hv_cpu_number_to_vp_number(0); 623 return; 624 } 625 626 /* 627 * Based on the channel affinity policy, we will assign the NUMA 628 * nodes. 629 */ 630 631 if ((channel->affinity_policy == HV_BALANCED) || (!primary)) { 632 while (true) { 633 next_node = next_numa_node_id++; 634 if (next_node == nr_node_ids) { 635 next_node = next_numa_node_id = 0; 636 continue; 637 } 638 if (cpumask_empty(cpumask_of_node(next_node))) 639 continue; 640 break; 641 } 642 channel->numa_node = next_node; 643 primary = channel; 644 } 645 alloced_mask = &hv_context.hv_numa_map[primary->numa_node]; 646 647 if (cpumask_weight(alloced_mask) == 648 cpumask_weight(cpumask_of_node(primary->numa_node))) { 649 /* 650 * We have cycled through all the CPUs in the node; 651 * reset the alloced map. 652 */ 653 cpumask_clear(alloced_mask); 654 } 655 656 cpumask_xor(&available_mask, alloced_mask, 657 cpumask_of_node(primary->numa_node)); 658 659 cur_cpu = -1; 660 661 if (primary->affinity_policy == HV_LOCALIZED) { 662 /* 663 * Normally Hyper-V host doesn't create more subchannels 664 * than there are VCPUs on the node but it is possible when not 665 * all present VCPUs on the node are initialized by guest. 666 * Clear the alloced_cpus_in_node to start over. 667 */ 668 if (cpumask_equal(&primary->alloced_cpus_in_node, 669 cpumask_of_node(primary->numa_node))) 670 cpumask_clear(&primary->alloced_cpus_in_node); 671 } 672 673 while (true) { 674 cur_cpu = cpumask_next(cur_cpu, &available_mask); 675 if (cur_cpu >= nr_cpu_ids) { 676 cur_cpu = -1; 677 cpumask_copy(&available_mask, 678 cpumask_of_node(primary->numa_node)); 679 continue; 680 } 681 682 if (primary->affinity_policy == HV_LOCALIZED) { 683 /* 684 * NOTE: in the case of sub-channel, we clear the 685 * sub-channel related bit(s) in 686 * primary->alloced_cpus_in_node in 687 * hv_process_channel_removal(), so when we 688 * reload drivers like hv_netvsc in SMP guest, here 689 * we're able to re-allocate 690 * bit from primary->alloced_cpus_in_node. 691 */ 692 if (!cpumask_test_cpu(cur_cpu, 693 &primary->alloced_cpus_in_node)) { 694 cpumask_set_cpu(cur_cpu, 695 &primary->alloced_cpus_in_node); 696 cpumask_set_cpu(cur_cpu, alloced_mask); 697 break; 698 } 699 } else { 700 cpumask_set_cpu(cur_cpu, alloced_mask); 701 break; 702 } 703 } 704 705 channel->target_cpu = cur_cpu; 706 channel->target_vp = hv_cpu_number_to_vp_number(cur_cpu); 707 } 708 709 static void vmbus_wait_for_unload(void) 710 { 711 int cpu; 712 void *page_addr; 713 struct hv_message *msg; 714 struct vmbus_channel_message_header *hdr; 715 u32 message_type; 716 717 /* 718 * CHANNELMSG_UNLOAD_RESPONSE is always delivered to the CPU which was 719 * used for initial contact or to CPU0 depending on host version. When 720 * we're crashing on a different CPU let's hope that IRQ handler on 721 * the cpu which receives CHANNELMSG_UNLOAD_RESPONSE is still 722 * functional and vmbus_unload_response() will complete 723 * vmbus_connection.unload_event. If not, the last thing we can do is 724 * read message pages for all CPUs directly. 725 */ 726 while (1) { 727 if (completion_done(&vmbus_connection.unload_event)) 728 break; 729 730 for_each_online_cpu(cpu) { 731 struct hv_per_cpu_context *hv_cpu 732 = per_cpu_ptr(hv_context.cpu_context, cpu); 733 734 page_addr = hv_cpu->synic_message_page; 735 msg = (struct hv_message *)page_addr 736 + VMBUS_MESSAGE_SINT; 737 738 message_type = READ_ONCE(msg->header.message_type); 739 if (message_type == HVMSG_NONE) 740 continue; 741 742 hdr = (struct vmbus_channel_message_header *) 743 msg->u.payload; 744 745 if (hdr->msgtype == CHANNELMSG_UNLOAD_RESPONSE) 746 complete(&vmbus_connection.unload_event); 747 748 vmbus_signal_eom(msg, message_type); 749 } 750 751 mdelay(10); 752 } 753 754 /* 755 * We're crashing and already got the UNLOAD_RESPONSE, cleanup all 756 * maybe-pending messages on all CPUs to be able to receive new 757 * messages after we reconnect. 758 */ 759 for_each_online_cpu(cpu) { 760 struct hv_per_cpu_context *hv_cpu 761 = per_cpu_ptr(hv_context.cpu_context, cpu); 762 763 page_addr = hv_cpu->synic_message_page; 764 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT; 765 msg->header.message_type = HVMSG_NONE; 766 } 767 } 768 769 /* 770 * vmbus_unload_response - Handler for the unload response. 771 */ 772 static void vmbus_unload_response(struct vmbus_channel_message_header *hdr) 773 { 774 /* 775 * This is a global event; just wakeup the waiting thread. 776 * Once we successfully unload, we can cleanup the monitor state. 777 */ 778 complete(&vmbus_connection.unload_event); 779 } 780 781 void vmbus_initiate_unload(bool crash) 782 { 783 struct vmbus_channel_message_header hdr; 784 785 /* Pre-Win2012R2 hosts don't support reconnect */ 786 if (vmbus_proto_version < VERSION_WIN8_1) 787 return; 788 789 init_completion(&vmbus_connection.unload_event); 790 memset(&hdr, 0, sizeof(struct vmbus_channel_message_header)); 791 hdr.msgtype = CHANNELMSG_UNLOAD; 792 vmbus_post_msg(&hdr, sizeof(struct vmbus_channel_message_header), 793 !crash); 794 795 /* 796 * vmbus_initiate_unload() is also called on crash and the crash can be 797 * happening in an interrupt context, where scheduling is impossible. 798 */ 799 if (!crash) 800 wait_for_completion(&vmbus_connection.unload_event); 801 else 802 vmbus_wait_for_unload(); 803 } 804 805 /* 806 * vmbus_onoffer - Handler for channel offers from vmbus in parent partition. 807 * 808 */ 809 static void vmbus_onoffer(struct vmbus_channel_message_header *hdr) 810 { 811 struct vmbus_channel_offer_channel *offer; 812 struct vmbus_channel *newchannel; 813 814 offer = (struct vmbus_channel_offer_channel *)hdr; 815 816 trace_vmbus_onoffer(offer); 817 818 /* Allocate the channel object and save this offer. */ 819 newchannel = alloc_channel(); 820 if (!newchannel) { 821 vmbus_release_relid(offer->child_relid); 822 atomic_dec(&vmbus_connection.offer_in_progress); 823 pr_err("Unable to allocate channel object\n"); 824 return; 825 } 826 827 /* 828 * Setup state for signalling the host. 829 */ 830 newchannel->sig_event = VMBUS_EVENT_CONNECTION_ID; 831 832 if (vmbus_proto_version != VERSION_WS2008) { 833 newchannel->is_dedicated_interrupt = 834 (offer->is_dedicated_interrupt != 0); 835 newchannel->sig_event = offer->connection_id; 836 } 837 838 memcpy(&newchannel->offermsg, offer, 839 sizeof(struct vmbus_channel_offer_channel)); 840 newchannel->monitor_grp = (u8)offer->monitorid / 32; 841 newchannel->monitor_bit = (u8)offer->monitorid % 32; 842 843 vmbus_process_offer(newchannel); 844 } 845 846 /* 847 * vmbus_onoffer_rescind - Rescind offer handler. 848 * 849 * We queue a work item to process this offer synchronously 850 */ 851 static void vmbus_onoffer_rescind(struct vmbus_channel_message_header *hdr) 852 { 853 struct vmbus_channel_rescind_offer *rescind; 854 struct vmbus_channel *channel; 855 struct device *dev; 856 857 rescind = (struct vmbus_channel_rescind_offer *)hdr; 858 859 trace_vmbus_onoffer_rescind(rescind); 860 861 /* 862 * The offer msg and the corresponding rescind msg 863 * from the host are guranteed to be ordered - 864 * offer comes in first and then the rescind. 865 * Since we process these events in work elements, 866 * and with preemption, we may end up processing 867 * the events out of order. Given that we handle these 868 * work elements on the same CPU, this is possible only 869 * in the case of preemption. In any case wait here 870 * until the offer processing has moved beyond the 871 * point where the channel is discoverable. 872 */ 873 874 while (atomic_read(&vmbus_connection.offer_in_progress) != 0) { 875 /* 876 * We wait here until any channel offer is currently 877 * being processed. 878 */ 879 msleep(1); 880 } 881 882 mutex_lock(&vmbus_connection.channel_mutex); 883 channel = relid2channel(rescind->child_relid); 884 mutex_unlock(&vmbus_connection.channel_mutex); 885 886 if (channel == NULL) { 887 /* 888 * We failed in processing the offer message; 889 * we would have cleaned up the relid in that 890 * failure path. 891 */ 892 return; 893 } 894 895 /* 896 * Now wait for offer handling to complete. 897 */ 898 while (READ_ONCE(channel->probe_done) == false) { 899 /* 900 * We wait here until any channel offer is currently 901 * being processed. 902 */ 903 msleep(1); 904 } 905 906 /* 907 * At this point, the rescind handling can proceed safely. 908 */ 909 910 if (channel->device_obj) { 911 if (channel->chn_rescind_callback) { 912 channel->chn_rescind_callback(channel); 913 vmbus_rescind_cleanup(channel); 914 return; 915 } 916 /* 917 * We will have to unregister this device from the 918 * driver core. 919 */ 920 dev = get_device(&channel->device_obj->device); 921 if (dev) { 922 vmbus_rescind_cleanup(channel); 923 vmbus_device_unregister(channel->device_obj); 924 put_device(dev); 925 } 926 } 927 if (channel->primary_channel != NULL) { 928 /* 929 * Sub-channel is being rescinded. Following is the channel 930 * close sequence when initiated from the driveri (refer to 931 * vmbus_close() for details): 932 * 1. Close all sub-channels first 933 * 2. Then close the primary channel. 934 */ 935 mutex_lock(&vmbus_connection.channel_mutex); 936 vmbus_rescind_cleanup(channel); 937 if (channel->state == CHANNEL_OPEN_STATE) { 938 /* 939 * The channel is currently not open; 940 * it is safe for us to cleanup the channel. 941 */ 942 hv_process_channel_removal(rescind->child_relid); 943 } 944 mutex_unlock(&vmbus_connection.channel_mutex); 945 } 946 } 947 948 void vmbus_hvsock_device_unregister(struct vmbus_channel *channel) 949 { 950 BUG_ON(!is_hvsock_channel(channel)); 951 952 /* We always get a rescind msg when a connection is closed. */ 953 while (!READ_ONCE(channel->probe_done) || !READ_ONCE(channel->rescind)) 954 msleep(1); 955 956 vmbus_device_unregister(channel->device_obj); 957 } 958 EXPORT_SYMBOL_GPL(vmbus_hvsock_device_unregister); 959 960 961 /* 962 * vmbus_onoffers_delivered - 963 * This is invoked when all offers have been delivered. 964 * 965 * Nothing to do here. 966 */ 967 static void vmbus_onoffers_delivered( 968 struct vmbus_channel_message_header *hdr) 969 { 970 } 971 972 /* 973 * vmbus_onopen_result - Open result handler. 974 * 975 * This is invoked when we received a response to our channel open request. 976 * Find the matching request, copy the response and signal the requesting 977 * thread. 978 */ 979 static void vmbus_onopen_result(struct vmbus_channel_message_header *hdr) 980 { 981 struct vmbus_channel_open_result *result; 982 struct vmbus_channel_msginfo *msginfo; 983 struct vmbus_channel_message_header *requestheader; 984 struct vmbus_channel_open_channel *openmsg; 985 unsigned long flags; 986 987 result = (struct vmbus_channel_open_result *)hdr; 988 989 trace_vmbus_onopen_result(result); 990 991 /* 992 * Find the open msg, copy the result and signal/unblock the wait event 993 */ 994 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 995 996 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 997 msglistentry) { 998 requestheader = 999 (struct vmbus_channel_message_header *)msginfo->msg; 1000 1001 if (requestheader->msgtype == CHANNELMSG_OPENCHANNEL) { 1002 openmsg = 1003 (struct vmbus_channel_open_channel *)msginfo->msg; 1004 if (openmsg->child_relid == result->child_relid && 1005 openmsg->openid == result->openid) { 1006 memcpy(&msginfo->response.open_result, 1007 result, 1008 sizeof( 1009 struct vmbus_channel_open_result)); 1010 complete(&msginfo->waitevent); 1011 break; 1012 } 1013 } 1014 } 1015 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 1016 } 1017 1018 /* 1019 * vmbus_ongpadl_created - GPADL created handler. 1020 * 1021 * This is invoked when we received a response to our gpadl create request. 1022 * Find the matching request, copy the response and signal the requesting 1023 * thread. 1024 */ 1025 static void vmbus_ongpadl_created(struct vmbus_channel_message_header *hdr) 1026 { 1027 struct vmbus_channel_gpadl_created *gpadlcreated; 1028 struct vmbus_channel_msginfo *msginfo; 1029 struct vmbus_channel_message_header *requestheader; 1030 struct vmbus_channel_gpadl_header *gpadlheader; 1031 unsigned long flags; 1032 1033 gpadlcreated = (struct vmbus_channel_gpadl_created *)hdr; 1034 1035 trace_vmbus_ongpadl_created(gpadlcreated); 1036 1037 /* 1038 * Find the establish msg, copy the result and signal/unblock the wait 1039 * event 1040 */ 1041 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 1042 1043 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 1044 msglistentry) { 1045 requestheader = 1046 (struct vmbus_channel_message_header *)msginfo->msg; 1047 1048 if (requestheader->msgtype == CHANNELMSG_GPADL_HEADER) { 1049 gpadlheader = 1050 (struct vmbus_channel_gpadl_header *)requestheader; 1051 1052 if ((gpadlcreated->child_relid == 1053 gpadlheader->child_relid) && 1054 (gpadlcreated->gpadl == gpadlheader->gpadl)) { 1055 memcpy(&msginfo->response.gpadl_created, 1056 gpadlcreated, 1057 sizeof( 1058 struct vmbus_channel_gpadl_created)); 1059 complete(&msginfo->waitevent); 1060 break; 1061 } 1062 } 1063 } 1064 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 1065 } 1066 1067 /* 1068 * vmbus_ongpadl_torndown - GPADL torndown handler. 1069 * 1070 * This is invoked when we received a response to our gpadl teardown request. 1071 * Find the matching request, copy the response and signal the requesting 1072 * thread. 1073 */ 1074 static void vmbus_ongpadl_torndown( 1075 struct vmbus_channel_message_header *hdr) 1076 { 1077 struct vmbus_channel_gpadl_torndown *gpadl_torndown; 1078 struct vmbus_channel_msginfo *msginfo; 1079 struct vmbus_channel_message_header *requestheader; 1080 struct vmbus_channel_gpadl_teardown *gpadl_teardown; 1081 unsigned long flags; 1082 1083 gpadl_torndown = (struct vmbus_channel_gpadl_torndown *)hdr; 1084 1085 trace_vmbus_ongpadl_torndown(gpadl_torndown); 1086 1087 /* 1088 * Find the open msg, copy the result and signal/unblock the wait event 1089 */ 1090 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 1091 1092 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 1093 msglistentry) { 1094 requestheader = 1095 (struct vmbus_channel_message_header *)msginfo->msg; 1096 1097 if (requestheader->msgtype == CHANNELMSG_GPADL_TEARDOWN) { 1098 gpadl_teardown = 1099 (struct vmbus_channel_gpadl_teardown *)requestheader; 1100 1101 if (gpadl_torndown->gpadl == gpadl_teardown->gpadl) { 1102 memcpy(&msginfo->response.gpadl_torndown, 1103 gpadl_torndown, 1104 sizeof( 1105 struct vmbus_channel_gpadl_torndown)); 1106 complete(&msginfo->waitevent); 1107 break; 1108 } 1109 } 1110 } 1111 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 1112 } 1113 1114 /* 1115 * vmbus_onversion_response - Version response handler 1116 * 1117 * This is invoked when we received a response to our initiate contact request. 1118 * Find the matching request, copy the response and signal the requesting 1119 * thread. 1120 */ 1121 static void vmbus_onversion_response( 1122 struct vmbus_channel_message_header *hdr) 1123 { 1124 struct vmbus_channel_msginfo *msginfo; 1125 struct vmbus_channel_message_header *requestheader; 1126 struct vmbus_channel_version_response *version_response; 1127 unsigned long flags; 1128 1129 version_response = (struct vmbus_channel_version_response *)hdr; 1130 1131 trace_vmbus_onversion_response(version_response); 1132 1133 spin_lock_irqsave(&vmbus_connection.channelmsg_lock, flags); 1134 1135 list_for_each_entry(msginfo, &vmbus_connection.chn_msg_list, 1136 msglistentry) { 1137 requestheader = 1138 (struct vmbus_channel_message_header *)msginfo->msg; 1139 1140 if (requestheader->msgtype == 1141 CHANNELMSG_INITIATE_CONTACT) { 1142 memcpy(&msginfo->response.version_response, 1143 version_response, 1144 sizeof(struct vmbus_channel_version_response)); 1145 complete(&msginfo->waitevent); 1146 } 1147 } 1148 spin_unlock_irqrestore(&vmbus_connection.channelmsg_lock, flags); 1149 } 1150 1151 /* Channel message dispatch table */ 1152 const struct vmbus_channel_message_table_entry 1153 channel_message_table[CHANNELMSG_COUNT] = { 1154 { CHANNELMSG_INVALID, 0, NULL }, 1155 { CHANNELMSG_OFFERCHANNEL, 0, vmbus_onoffer }, 1156 { CHANNELMSG_RESCIND_CHANNELOFFER, 0, vmbus_onoffer_rescind }, 1157 { CHANNELMSG_REQUESTOFFERS, 0, NULL }, 1158 { CHANNELMSG_ALLOFFERS_DELIVERED, 1, vmbus_onoffers_delivered }, 1159 { CHANNELMSG_OPENCHANNEL, 0, NULL }, 1160 { CHANNELMSG_OPENCHANNEL_RESULT, 1, vmbus_onopen_result }, 1161 { CHANNELMSG_CLOSECHANNEL, 0, NULL }, 1162 { CHANNELMSG_GPADL_HEADER, 0, NULL }, 1163 { CHANNELMSG_GPADL_BODY, 0, NULL }, 1164 { CHANNELMSG_GPADL_CREATED, 1, vmbus_ongpadl_created }, 1165 { CHANNELMSG_GPADL_TEARDOWN, 0, NULL }, 1166 { CHANNELMSG_GPADL_TORNDOWN, 1, vmbus_ongpadl_torndown }, 1167 { CHANNELMSG_RELID_RELEASED, 0, NULL }, 1168 { CHANNELMSG_INITIATE_CONTACT, 0, NULL }, 1169 { CHANNELMSG_VERSION_RESPONSE, 1, vmbus_onversion_response }, 1170 { CHANNELMSG_UNLOAD, 0, NULL }, 1171 { CHANNELMSG_UNLOAD_RESPONSE, 1, vmbus_unload_response }, 1172 { CHANNELMSG_18, 0, NULL }, 1173 { CHANNELMSG_19, 0, NULL }, 1174 { CHANNELMSG_20, 0, NULL }, 1175 { CHANNELMSG_TL_CONNECT_REQUEST, 0, NULL }, 1176 }; 1177 1178 /* 1179 * vmbus_onmessage - Handler for channel protocol messages. 1180 * 1181 * This is invoked in the vmbus worker thread context. 1182 */ 1183 void vmbus_onmessage(void *context) 1184 { 1185 struct hv_message *msg = context; 1186 struct vmbus_channel_message_header *hdr; 1187 int size; 1188 1189 hdr = (struct vmbus_channel_message_header *)msg->u.payload; 1190 size = msg->header.payload_size; 1191 1192 trace_vmbus_on_message(hdr); 1193 1194 if (hdr->msgtype >= CHANNELMSG_COUNT) { 1195 pr_err("Received invalid channel message type %d size %d\n", 1196 hdr->msgtype, size); 1197 print_hex_dump_bytes("", DUMP_PREFIX_NONE, 1198 (unsigned char *)msg->u.payload, size); 1199 return; 1200 } 1201 1202 if (channel_message_table[hdr->msgtype].message_handler) 1203 channel_message_table[hdr->msgtype].message_handler(hdr); 1204 else 1205 pr_err("Unhandled channel message type %d\n", hdr->msgtype); 1206 } 1207 1208 /* 1209 * vmbus_request_offers - Send a request to get all our pending offers. 1210 */ 1211 int vmbus_request_offers(void) 1212 { 1213 struct vmbus_channel_message_header *msg; 1214 struct vmbus_channel_msginfo *msginfo; 1215 int ret; 1216 1217 msginfo = kmalloc(sizeof(*msginfo) + 1218 sizeof(struct vmbus_channel_message_header), 1219 GFP_KERNEL); 1220 if (!msginfo) 1221 return -ENOMEM; 1222 1223 msg = (struct vmbus_channel_message_header *)msginfo->msg; 1224 1225 msg->msgtype = CHANNELMSG_REQUESTOFFERS; 1226 1227 1228 ret = vmbus_post_msg(msg, sizeof(struct vmbus_channel_message_header), 1229 true); 1230 if (ret != 0) { 1231 pr_err("Unable to request offers - %d\n", ret); 1232 1233 goto cleanup; 1234 } 1235 1236 cleanup: 1237 kfree(msginfo); 1238 1239 return ret; 1240 } 1241 1242 /* 1243 * Retrieve the (sub) channel on which to send an outgoing request. 1244 * When a primary channel has multiple sub-channels, we try to 1245 * distribute the load equally amongst all available channels. 1246 */ 1247 struct vmbus_channel *vmbus_get_outgoing_channel(struct vmbus_channel *primary) 1248 { 1249 struct list_head *cur, *tmp; 1250 int cur_cpu; 1251 struct vmbus_channel *cur_channel; 1252 struct vmbus_channel *outgoing_channel = primary; 1253 int next_channel; 1254 int i = 1; 1255 1256 if (list_empty(&primary->sc_list)) 1257 return outgoing_channel; 1258 1259 next_channel = primary->next_oc++; 1260 1261 if (next_channel > (primary->num_sc)) { 1262 primary->next_oc = 0; 1263 return outgoing_channel; 1264 } 1265 1266 cur_cpu = hv_cpu_number_to_vp_number(smp_processor_id()); 1267 list_for_each_safe(cur, tmp, &primary->sc_list) { 1268 cur_channel = list_entry(cur, struct vmbus_channel, sc_list); 1269 if (cur_channel->state != CHANNEL_OPENED_STATE) 1270 continue; 1271 1272 if (cur_channel->target_vp == cur_cpu) 1273 return cur_channel; 1274 1275 if (i == next_channel) 1276 return cur_channel; 1277 1278 i++; 1279 } 1280 1281 return outgoing_channel; 1282 } 1283 EXPORT_SYMBOL_GPL(vmbus_get_outgoing_channel); 1284 1285 static void invoke_sc_cb(struct vmbus_channel *primary_channel) 1286 { 1287 struct list_head *cur, *tmp; 1288 struct vmbus_channel *cur_channel; 1289 1290 if (primary_channel->sc_creation_callback == NULL) 1291 return; 1292 1293 list_for_each_safe(cur, tmp, &primary_channel->sc_list) { 1294 cur_channel = list_entry(cur, struct vmbus_channel, sc_list); 1295 1296 primary_channel->sc_creation_callback(cur_channel); 1297 } 1298 } 1299 1300 void vmbus_set_sc_create_callback(struct vmbus_channel *primary_channel, 1301 void (*sc_cr_cb)(struct vmbus_channel *new_sc)) 1302 { 1303 primary_channel->sc_creation_callback = sc_cr_cb; 1304 } 1305 EXPORT_SYMBOL_GPL(vmbus_set_sc_create_callback); 1306 1307 bool vmbus_are_subchannels_present(struct vmbus_channel *primary) 1308 { 1309 bool ret; 1310 1311 ret = !list_empty(&primary->sc_list); 1312 1313 if (ret) { 1314 /* 1315 * Invoke the callback on sub-channel creation. 1316 * This will present a uniform interface to the 1317 * clients. 1318 */ 1319 invoke_sc_cb(primary); 1320 } 1321 1322 return ret; 1323 } 1324 EXPORT_SYMBOL_GPL(vmbus_are_subchannels_present); 1325 1326 void vmbus_set_chn_rescind_callback(struct vmbus_channel *channel, 1327 void (*chn_rescind_cb)(struct vmbus_channel *)) 1328 { 1329 channel->chn_rescind_callback = chn_rescind_cb; 1330 } 1331 EXPORT_SYMBOL_GPL(vmbus_set_chn_rescind_callback); 1332