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 * K. Y. Srinivasan <kys@microsoft.com> 21 * 22 */ 23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 24 25 #include <linux/init.h> 26 #include <linux/module.h> 27 #include <linux/device.h> 28 #include <linux/interrupt.h> 29 #include <linux/sysctl.h> 30 #include <linux/slab.h> 31 #include <linux/acpi.h> 32 #include <linux/completion.h> 33 #include <linux/hyperv.h> 34 #include <linux/kernel_stat.h> 35 #include <linux/clockchips.h> 36 #include <linux/cpu.h> 37 #include <asm/hyperv.h> 38 #include <asm/hypervisor.h> 39 #include <asm/mshyperv.h> 40 #include <linux/notifier.h> 41 #include <linux/ptrace.h> 42 #include <linux/screen_info.h> 43 #include <linux/kdebug.h> 44 #include "hyperv_vmbus.h" 45 46 static struct acpi_device *hv_acpi_dev; 47 48 static struct tasklet_struct msg_dpc; 49 static struct completion probe_event; 50 static int irq; 51 52 53 static void hyperv_report_panic(struct pt_regs *regs) 54 { 55 static bool panic_reported; 56 57 /* 58 * We prefer to report panic on 'die' chain as we have proper 59 * registers to report, but if we miss it (e.g. on BUG()) we need 60 * to report it on 'panic'. 61 */ 62 if (panic_reported) 63 return; 64 panic_reported = true; 65 66 wrmsrl(HV_X64_MSR_CRASH_P0, regs->ip); 67 wrmsrl(HV_X64_MSR_CRASH_P1, regs->ax); 68 wrmsrl(HV_X64_MSR_CRASH_P2, regs->bx); 69 wrmsrl(HV_X64_MSR_CRASH_P3, regs->cx); 70 wrmsrl(HV_X64_MSR_CRASH_P4, regs->dx); 71 72 /* 73 * Let Hyper-V know there is crash data available 74 */ 75 wrmsrl(HV_X64_MSR_CRASH_CTL, HV_CRASH_CTL_CRASH_NOTIFY); 76 } 77 78 static int hyperv_panic_event(struct notifier_block *nb, unsigned long val, 79 void *args) 80 { 81 struct pt_regs *regs; 82 83 regs = current_pt_regs(); 84 85 hyperv_report_panic(regs); 86 return NOTIFY_DONE; 87 } 88 89 static int hyperv_die_event(struct notifier_block *nb, unsigned long val, 90 void *args) 91 { 92 struct die_args *die = (struct die_args *)args; 93 struct pt_regs *regs = die->regs; 94 95 hyperv_report_panic(regs); 96 return NOTIFY_DONE; 97 } 98 99 static struct notifier_block hyperv_die_block = { 100 .notifier_call = hyperv_die_event, 101 }; 102 static struct notifier_block hyperv_panic_block = { 103 .notifier_call = hyperv_panic_event, 104 }; 105 106 struct resource *hyperv_mmio; 107 108 static int vmbus_exists(void) 109 { 110 if (hv_acpi_dev == NULL) 111 return -ENODEV; 112 113 return 0; 114 } 115 116 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2) 117 static void print_alias_name(struct hv_device *hv_dev, char *alias_name) 118 { 119 int i; 120 for (i = 0; i < VMBUS_ALIAS_LEN; i += 2) 121 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]); 122 } 123 124 static u8 channel_monitor_group(struct vmbus_channel *channel) 125 { 126 return (u8)channel->offermsg.monitorid / 32; 127 } 128 129 static u8 channel_monitor_offset(struct vmbus_channel *channel) 130 { 131 return (u8)channel->offermsg.monitorid % 32; 132 } 133 134 static u32 channel_pending(struct vmbus_channel *channel, 135 struct hv_monitor_page *monitor_page) 136 { 137 u8 monitor_group = channel_monitor_group(channel); 138 return monitor_page->trigger_group[monitor_group].pending; 139 } 140 141 static u32 channel_latency(struct vmbus_channel *channel, 142 struct hv_monitor_page *monitor_page) 143 { 144 u8 monitor_group = channel_monitor_group(channel); 145 u8 monitor_offset = channel_monitor_offset(channel); 146 return monitor_page->latency[monitor_group][monitor_offset]; 147 } 148 149 static u32 channel_conn_id(struct vmbus_channel *channel, 150 struct hv_monitor_page *monitor_page) 151 { 152 u8 monitor_group = channel_monitor_group(channel); 153 u8 monitor_offset = channel_monitor_offset(channel); 154 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id; 155 } 156 157 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr, 158 char *buf) 159 { 160 struct hv_device *hv_dev = device_to_hv_device(dev); 161 162 if (!hv_dev->channel) 163 return -ENODEV; 164 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid); 165 } 166 static DEVICE_ATTR_RO(id); 167 168 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr, 169 char *buf) 170 { 171 struct hv_device *hv_dev = device_to_hv_device(dev); 172 173 if (!hv_dev->channel) 174 return -ENODEV; 175 return sprintf(buf, "%d\n", hv_dev->channel->state); 176 } 177 static DEVICE_ATTR_RO(state); 178 179 static ssize_t monitor_id_show(struct device *dev, 180 struct device_attribute *dev_attr, char *buf) 181 { 182 struct hv_device *hv_dev = device_to_hv_device(dev); 183 184 if (!hv_dev->channel) 185 return -ENODEV; 186 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid); 187 } 188 static DEVICE_ATTR_RO(monitor_id); 189 190 static ssize_t class_id_show(struct device *dev, 191 struct device_attribute *dev_attr, char *buf) 192 { 193 struct hv_device *hv_dev = device_to_hv_device(dev); 194 195 if (!hv_dev->channel) 196 return -ENODEV; 197 return sprintf(buf, "{%pUl}\n", 198 hv_dev->channel->offermsg.offer.if_type.b); 199 } 200 static DEVICE_ATTR_RO(class_id); 201 202 static ssize_t device_id_show(struct device *dev, 203 struct device_attribute *dev_attr, char *buf) 204 { 205 struct hv_device *hv_dev = device_to_hv_device(dev); 206 207 if (!hv_dev->channel) 208 return -ENODEV; 209 return sprintf(buf, "{%pUl}\n", 210 hv_dev->channel->offermsg.offer.if_instance.b); 211 } 212 static DEVICE_ATTR_RO(device_id); 213 214 static ssize_t modalias_show(struct device *dev, 215 struct device_attribute *dev_attr, char *buf) 216 { 217 struct hv_device *hv_dev = device_to_hv_device(dev); 218 char alias_name[VMBUS_ALIAS_LEN + 1]; 219 220 print_alias_name(hv_dev, alias_name); 221 return sprintf(buf, "vmbus:%s\n", alias_name); 222 } 223 static DEVICE_ATTR_RO(modalias); 224 225 static ssize_t server_monitor_pending_show(struct device *dev, 226 struct device_attribute *dev_attr, 227 char *buf) 228 { 229 struct hv_device *hv_dev = device_to_hv_device(dev); 230 231 if (!hv_dev->channel) 232 return -ENODEV; 233 return sprintf(buf, "%d\n", 234 channel_pending(hv_dev->channel, 235 vmbus_connection.monitor_pages[1])); 236 } 237 static DEVICE_ATTR_RO(server_monitor_pending); 238 239 static ssize_t client_monitor_pending_show(struct device *dev, 240 struct device_attribute *dev_attr, 241 char *buf) 242 { 243 struct hv_device *hv_dev = device_to_hv_device(dev); 244 245 if (!hv_dev->channel) 246 return -ENODEV; 247 return sprintf(buf, "%d\n", 248 channel_pending(hv_dev->channel, 249 vmbus_connection.monitor_pages[1])); 250 } 251 static DEVICE_ATTR_RO(client_monitor_pending); 252 253 static ssize_t server_monitor_latency_show(struct device *dev, 254 struct device_attribute *dev_attr, 255 char *buf) 256 { 257 struct hv_device *hv_dev = device_to_hv_device(dev); 258 259 if (!hv_dev->channel) 260 return -ENODEV; 261 return sprintf(buf, "%d\n", 262 channel_latency(hv_dev->channel, 263 vmbus_connection.monitor_pages[0])); 264 } 265 static DEVICE_ATTR_RO(server_monitor_latency); 266 267 static ssize_t client_monitor_latency_show(struct device *dev, 268 struct device_attribute *dev_attr, 269 char *buf) 270 { 271 struct hv_device *hv_dev = device_to_hv_device(dev); 272 273 if (!hv_dev->channel) 274 return -ENODEV; 275 return sprintf(buf, "%d\n", 276 channel_latency(hv_dev->channel, 277 vmbus_connection.monitor_pages[1])); 278 } 279 static DEVICE_ATTR_RO(client_monitor_latency); 280 281 static ssize_t server_monitor_conn_id_show(struct device *dev, 282 struct device_attribute *dev_attr, 283 char *buf) 284 { 285 struct hv_device *hv_dev = device_to_hv_device(dev); 286 287 if (!hv_dev->channel) 288 return -ENODEV; 289 return sprintf(buf, "%d\n", 290 channel_conn_id(hv_dev->channel, 291 vmbus_connection.monitor_pages[0])); 292 } 293 static DEVICE_ATTR_RO(server_monitor_conn_id); 294 295 static ssize_t client_monitor_conn_id_show(struct device *dev, 296 struct device_attribute *dev_attr, 297 char *buf) 298 { 299 struct hv_device *hv_dev = device_to_hv_device(dev); 300 301 if (!hv_dev->channel) 302 return -ENODEV; 303 return sprintf(buf, "%d\n", 304 channel_conn_id(hv_dev->channel, 305 vmbus_connection.monitor_pages[1])); 306 } 307 static DEVICE_ATTR_RO(client_monitor_conn_id); 308 309 static ssize_t out_intr_mask_show(struct device *dev, 310 struct device_attribute *dev_attr, char *buf) 311 { 312 struct hv_device *hv_dev = device_to_hv_device(dev); 313 struct hv_ring_buffer_debug_info outbound; 314 315 if (!hv_dev->channel) 316 return -ENODEV; 317 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 318 return sprintf(buf, "%d\n", outbound.current_interrupt_mask); 319 } 320 static DEVICE_ATTR_RO(out_intr_mask); 321 322 static ssize_t out_read_index_show(struct device *dev, 323 struct device_attribute *dev_attr, char *buf) 324 { 325 struct hv_device *hv_dev = device_to_hv_device(dev); 326 struct hv_ring_buffer_debug_info outbound; 327 328 if (!hv_dev->channel) 329 return -ENODEV; 330 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 331 return sprintf(buf, "%d\n", outbound.current_read_index); 332 } 333 static DEVICE_ATTR_RO(out_read_index); 334 335 static ssize_t out_write_index_show(struct device *dev, 336 struct device_attribute *dev_attr, 337 char *buf) 338 { 339 struct hv_device *hv_dev = device_to_hv_device(dev); 340 struct hv_ring_buffer_debug_info outbound; 341 342 if (!hv_dev->channel) 343 return -ENODEV; 344 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 345 return sprintf(buf, "%d\n", outbound.current_write_index); 346 } 347 static DEVICE_ATTR_RO(out_write_index); 348 349 static ssize_t out_read_bytes_avail_show(struct device *dev, 350 struct device_attribute *dev_attr, 351 char *buf) 352 { 353 struct hv_device *hv_dev = device_to_hv_device(dev); 354 struct hv_ring_buffer_debug_info outbound; 355 356 if (!hv_dev->channel) 357 return -ENODEV; 358 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 359 return sprintf(buf, "%d\n", outbound.bytes_avail_toread); 360 } 361 static DEVICE_ATTR_RO(out_read_bytes_avail); 362 363 static ssize_t out_write_bytes_avail_show(struct device *dev, 364 struct device_attribute *dev_attr, 365 char *buf) 366 { 367 struct hv_device *hv_dev = device_to_hv_device(dev); 368 struct hv_ring_buffer_debug_info outbound; 369 370 if (!hv_dev->channel) 371 return -ENODEV; 372 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 373 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite); 374 } 375 static DEVICE_ATTR_RO(out_write_bytes_avail); 376 377 static ssize_t in_intr_mask_show(struct device *dev, 378 struct device_attribute *dev_attr, char *buf) 379 { 380 struct hv_device *hv_dev = device_to_hv_device(dev); 381 struct hv_ring_buffer_debug_info inbound; 382 383 if (!hv_dev->channel) 384 return -ENODEV; 385 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 386 return sprintf(buf, "%d\n", inbound.current_interrupt_mask); 387 } 388 static DEVICE_ATTR_RO(in_intr_mask); 389 390 static ssize_t in_read_index_show(struct device *dev, 391 struct device_attribute *dev_attr, char *buf) 392 { 393 struct hv_device *hv_dev = device_to_hv_device(dev); 394 struct hv_ring_buffer_debug_info inbound; 395 396 if (!hv_dev->channel) 397 return -ENODEV; 398 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 399 return sprintf(buf, "%d\n", inbound.current_read_index); 400 } 401 static DEVICE_ATTR_RO(in_read_index); 402 403 static ssize_t in_write_index_show(struct device *dev, 404 struct device_attribute *dev_attr, char *buf) 405 { 406 struct hv_device *hv_dev = device_to_hv_device(dev); 407 struct hv_ring_buffer_debug_info inbound; 408 409 if (!hv_dev->channel) 410 return -ENODEV; 411 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 412 return sprintf(buf, "%d\n", inbound.current_write_index); 413 } 414 static DEVICE_ATTR_RO(in_write_index); 415 416 static ssize_t in_read_bytes_avail_show(struct device *dev, 417 struct device_attribute *dev_attr, 418 char *buf) 419 { 420 struct hv_device *hv_dev = device_to_hv_device(dev); 421 struct hv_ring_buffer_debug_info inbound; 422 423 if (!hv_dev->channel) 424 return -ENODEV; 425 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 426 return sprintf(buf, "%d\n", inbound.bytes_avail_toread); 427 } 428 static DEVICE_ATTR_RO(in_read_bytes_avail); 429 430 static ssize_t in_write_bytes_avail_show(struct device *dev, 431 struct device_attribute *dev_attr, 432 char *buf) 433 { 434 struct hv_device *hv_dev = device_to_hv_device(dev); 435 struct hv_ring_buffer_debug_info inbound; 436 437 if (!hv_dev->channel) 438 return -ENODEV; 439 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 440 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite); 441 } 442 static DEVICE_ATTR_RO(in_write_bytes_avail); 443 444 static ssize_t channel_vp_mapping_show(struct device *dev, 445 struct device_attribute *dev_attr, 446 char *buf) 447 { 448 struct hv_device *hv_dev = device_to_hv_device(dev); 449 struct vmbus_channel *channel = hv_dev->channel, *cur_sc; 450 unsigned long flags; 451 int buf_size = PAGE_SIZE, n_written, tot_written; 452 struct list_head *cur; 453 454 if (!channel) 455 return -ENODEV; 456 457 tot_written = snprintf(buf, buf_size, "%u:%u\n", 458 channel->offermsg.child_relid, channel->target_cpu); 459 460 spin_lock_irqsave(&channel->lock, flags); 461 462 list_for_each(cur, &channel->sc_list) { 463 if (tot_written >= buf_size - 1) 464 break; 465 466 cur_sc = list_entry(cur, struct vmbus_channel, sc_list); 467 n_written = scnprintf(buf + tot_written, 468 buf_size - tot_written, 469 "%u:%u\n", 470 cur_sc->offermsg.child_relid, 471 cur_sc->target_cpu); 472 tot_written += n_written; 473 } 474 475 spin_unlock_irqrestore(&channel->lock, flags); 476 477 return tot_written; 478 } 479 static DEVICE_ATTR_RO(channel_vp_mapping); 480 481 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */ 482 static struct attribute *vmbus_attrs[] = { 483 &dev_attr_id.attr, 484 &dev_attr_state.attr, 485 &dev_attr_monitor_id.attr, 486 &dev_attr_class_id.attr, 487 &dev_attr_device_id.attr, 488 &dev_attr_modalias.attr, 489 &dev_attr_server_monitor_pending.attr, 490 &dev_attr_client_monitor_pending.attr, 491 &dev_attr_server_monitor_latency.attr, 492 &dev_attr_client_monitor_latency.attr, 493 &dev_attr_server_monitor_conn_id.attr, 494 &dev_attr_client_monitor_conn_id.attr, 495 &dev_attr_out_intr_mask.attr, 496 &dev_attr_out_read_index.attr, 497 &dev_attr_out_write_index.attr, 498 &dev_attr_out_read_bytes_avail.attr, 499 &dev_attr_out_write_bytes_avail.attr, 500 &dev_attr_in_intr_mask.attr, 501 &dev_attr_in_read_index.attr, 502 &dev_attr_in_write_index.attr, 503 &dev_attr_in_read_bytes_avail.attr, 504 &dev_attr_in_write_bytes_avail.attr, 505 &dev_attr_channel_vp_mapping.attr, 506 NULL, 507 }; 508 ATTRIBUTE_GROUPS(vmbus); 509 510 /* 511 * vmbus_uevent - add uevent for our device 512 * 513 * This routine is invoked when a device is added or removed on the vmbus to 514 * generate a uevent to udev in the userspace. The udev will then look at its 515 * rule and the uevent generated here to load the appropriate driver 516 * 517 * The alias string will be of the form vmbus:guid where guid is the string 518 * representation of the device guid (each byte of the guid will be 519 * represented with two hex characters. 520 */ 521 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env) 522 { 523 struct hv_device *dev = device_to_hv_device(device); 524 int ret; 525 char alias_name[VMBUS_ALIAS_LEN + 1]; 526 527 print_alias_name(dev, alias_name); 528 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name); 529 return ret; 530 } 531 532 static const uuid_le null_guid; 533 534 static inline bool is_null_guid(const __u8 *guid) 535 { 536 if (memcmp(guid, &null_guid, sizeof(uuid_le))) 537 return false; 538 return true; 539 } 540 541 /* 542 * Return a matching hv_vmbus_device_id pointer. 543 * If there is no match, return NULL. 544 */ 545 static const struct hv_vmbus_device_id *hv_vmbus_get_id( 546 const struct hv_vmbus_device_id *id, 547 const __u8 *guid) 548 { 549 for (; !is_null_guid(id->guid); id++) 550 if (!memcmp(&id->guid, guid, sizeof(uuid_le))) 551 return id; 552 553 return NULL; 554 } 555 556 557 558 /* 559 * vmbus_match - Attempt to match the specified device to the specified driver 560 */ 561 static int vmbus_match(struct device *device, struct device_driver *driver) 562 { 563 struct hv_driver *drv = drv_to_hv_drv(driver); 564 struct hv_device *hv_dev = device_to_hv_device(device); 565 566 if (hv_vmbus_get_id(drv->id_table, hv_dev->dev_type.b)) 567 return 1; 568 569 return 0; 570 } 571 572 /* 573 * vmbus_probe - Add the new vmbus's child device 574 */ 575 static int vmbus_probe(struct device *child_device) 576 { 577 int ret = 0; 578 struct hv_driver *drv = 579 drv_to_hv_drv(child_device->driver); 580 struct hv_device *dev = device_to_hv_device(child_device); 581 const struct hv_vmbus_device_id *dev_id; 582 583 dev_id = hv_vmbus_get_id(drv->id_table, dev->dev_type.b); 584 if (drv->probe) { 585 ret = drv->probe(dev, dev_id); 586 if (ret != 0) 587 pr_err("probe failed for device %s (%d)\n", 588 dev_name(child_device), ret); 589 590 } else { 591 pr_err("probe not set for driver %s\n", 592 dev_name(child_device)); 593 ret = -ENODEV; 594 } 595 return ret; 596 } 597 598 /* 599 * vmbus_remove - Remove a vmbus device 600 */ 601 static int vmbus_remove(struct device *child_device) 602 { 603 struct hv_driver *drv; 604 struct hv_device *dev = device_to_hv_device(child_device); 605 u32 relid = dev->channel->offermsg.child_relid; 606 607 if (child_device->driver) { 608 drv = drv_to_hv_drv(child_device->driver); 609 if (drv->remove) 610 drv->remove(dev); 611 else { 612 hv_process_channel_removal(dev->channel, relid); 613 pr_err("remove not set for driver %s\n", 614 dev_name(child_device)); 615 } 616 } else { 617 /* 618 * We don't have a driver for this device; deal with the 619 * rescind message by removing the channel. 620 */ 621 hv_process_channel_removal(dev->channel, relid); 622 } 623 624 return 0; 625 } 626 627 628 /* 629 * vmbus_shutdown - Shutdown a vmbus device 630 */ 631 static void vmbus_shutdown(struct device *child_device) 632 { 633 struct hv_driver *drv; 634 struct hv_device *dev = device_to_hv_device(child_device); 635 636 637 /* The device may not be attached yet */ 638 if (!child_device->driver) 639 return; 640 641 drv = drv_to_hv_drv(child_device->driver); 642 643 if (drv->shutdown) 644 drv->shutdown(dev); 645 646 return; 647 } 648 649 650 /* 651 * vmbus_device_release - Final callback release of the vmbus child device 652 */ 653 static void vmbus_device_release(struct device *device) 654 { 655 struct hv_device *hv_dev = device_to_hv_device(device); 656 657 kfree(hv_dev); 658 659 } 660 661 /* The one and only one */ 662 static struct bus_type hv_bus = { 663 .name = "vmbus", 664 .match = vmbus_match, 665 .shutdown = vmbus_shutdown, 666 .remove = vmbus_remove, 667 .probe = vmbus_probe, 668 .uevent = vmbus_uevent, 669 .dev_groups = vmbus_groups, 670 }; 671 672 struct onmessage_work_context { 673 struct work_struct work; 674 struct hv_message msg; 675 }; 676 677 static void vmbus_onmessage_work(struct work_struct *work) 678 { 679 struct onmessage_work_context *ctx; 680 681 /* Do not process messages if we're in DISCONNECTED state */ 682 if (vmbus_connection.conn_state == DISCONNECTED) 683 return; 684 685 ctx = container_of(work, struct onmessage_work_context, 686 work); 687 vmbus_onmessage(&ctx->msg); 688 kfree(ctx); 689 } 690 691 static void hv_process_timer_expiration(struct hv_message *msg, int cpu) 692 { 693 struct clock_event_device *dev = hv_context.clk_evt[cpu]; 694 695 if (dev->event_handler) 696 dev->event_handler(dev); 697 698 msg->header.message_type = HVMSG_NONE; 699 700 /* 701 * Make sure the write to MessageType (ie set to 702 * HVMSG_NONE) happens before we read the 703 * MessagePending and EOMing. Otherwise, the EOMing 704 * will not deliver any more messages since there is 705 * no empty slot 706 */ 707 mb(); 708 709 if (msg->header.message_flags.msg_pending) { 710 /* 711 * This will cause message queue rescan to 712 * possibly deliver another msg from the 713 * hypervisor 714 */ 715 wrmsrl(HV_X64_MSR_EOM, 0); 716 } 717 } 718 719 static void vmbus_on_msg_dpc(unsigned long data) 720 { 721 int cpu = smp_processor_id(); 722 void *page_addr = hv_context.synic_message_page[cpu]; 723 struct hv_message *msg = (struct hv_message *)page_addr + 724 VMBUS_MESSAGE_SINT; 725 struct vmbus_channel_message_header *hdr; 726 struct vmbus_channel_message_table_entry *entry; 727 struct onmessage_work_context *ctx; 728 729 while (1) { 730 if (msg->header.message_type == HVMSG_NONE) 731 /* no msg */ 732 break; 733 734 hdr = (struct vmbus_channel_message_header *)msg->u.payload; 735 736 if (hdr->msgtype >= CHANNELMSG_COUNT) { 737 WARN_ONCE(1, "unknown msgtype=%d\n", hdr->msgtype); 738 goto msg_handled; 739 } 740 741 entry = &channel_message_table[hdr->msgtype]; 742 if (entry->handler_type == VMHT_BLOCKING) { 743 ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC); 744 if (ctx == NULL) 745 continue; 746 747 INIT_WORK(&ctx->work, vmbus_onmessage_work); 748 memcpy(&ctx->msg, msg, sizeof(*msg)); 749 750 queue_work(vmbus_connection.work_queue, &ctx->work); 751 } else 752 entry->message_handler(hdr); 753 754 msg_handled: 755 msg->header.message_type = HVMSG_NONE; 756 757 /* 758 * Make sure the write to MessageType (ie set to 759 * HVMSG_NONE) happens before we read the 760 * MessagePending and EOMing. Otherwise, the EOMing 761 * will not deliver any more messages since there is 762 * no empty slot 763 */ 764 mb(); 765 766 if (msg->header.message_flags.msg_pending) { 767 /* 768 * This will cause message queue rescan to 769 * possibly deliver another msg from the 770 * hypervisor 771 */ 772 wrmsrl(HV_X64_MSR_EOM, 0); 773 } 774 } 775 } 776 777 static void vmbus_isr(void) 778 { 779 int cpu = smp_processor_id(); 780 void *page_addr; 781 struct hv_message *msg; 782 union hv_synic_event_flags *event; 783 bool handled = false; 784 785 page_addr = hv_context.synic_event_page[cpu]; 786 if (page_addr == NULL) 787 return; 788 789 event = (union hv_synic_event_flags *)page_addr + 790 VMBUS_MESSAGE_SINT; 791 /* 792 * Check for events before checking for messages. This is the order 793 * in which events and messages are checked in Windows guests on 794 * Hyper-V, and the Windows team suggested we do the same. 795 */ 796 797 if ((vmbus_proto_version == VERSION_WS2008) || 798 (vmbus_proto_version == VERSION_WIN7)) { 799 800 /* Since we are a child, we only need to check bit 0 */ 801 if (sync_test_and_clear_bit(0, 802 (unsigned long *) &event->flags32[0])) { 803 handled = true; 804 } 805 } else { 806 /* 807 * Our host is win8 or above. The signaling mechanism 808 * has changed and we can directly look at the event page. 809 * If bit n is set then we have an interrup on the channel 810 * whose id is n. 811 */ 812 handled = true; 813 } 814 815 if (handled) 816 tasklet_schedule(hv_context.event_dpc[cpu]); 817 818 819 page_addr = hv_context.synic_message_page[cpu]; 820 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT; 821 822 /* Check if there are actual msgs to be processed */ 823 if (msg->header.message_type != HVMSG_NONE) { 824 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) 825 hv_process_timer_expiration(msg, cpu); 826 else 827 tasklet_schedule(&msg_dpc); 828 } 829 } 830 831 832 /* 833 * vmbus_bus_init -Main vmbus driver initialization routine. 834 * 835 * Here, we 836 * - initialize the vmbus driver context 837 * - invoke the vmbus hv main init routine 838 * - get the irq resource 839 * - retrieve the channel offers 840 */ 841 static int vmbus_bus_init(int irq) 842 { 843 int ret; 844 845 /* Hypervisor initialization...setup hypercall page..etc */ 846 ret = hv_init(); 847 if (ret != 0) { 848 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret); 849 return ret; 850 } 851 852 tasklet_init(&msg_dpc, vmbus_on_msg_dpc, 0); 853 854 ret = bus_register(&hv_bus); 855 if (ret) 856 goto err_cleanup; 857 858 hv_setup_vmbus_irq(vmbus_isr); 859 860 ret = hv_synic_alloc(); 861 if (ret) 862 goto err_alloc; 863 /* 864 * Initialize the per-cpu interrupt state and 865 * connect to the host. 866 */ 867 on_each_cpu(hv_synic_init, NULL, 1); 868 ret = vmbus_connect(); 869 if (ret) 870 goto err_alloc; 871 872 if (vmbus_proto_version > VERSION_WIN7) 873 cpu_hotplug_disable(); 874 875 /* 876 * Only register if the crash MSRs are available 877 */ 878 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) { 879 register_die_notifier(&hyperv_die_block); 880 atomic_notifier_chain_register(&panic_notifier_list, 881 &hyperv_panic_block); 882 } 883 884 vmbus_request_offers(); 885 886 return 0; 887 888 err_alloc: 889 hv_synic_free(); 890 hv_remove_vmbus_irq(); 891 892 bus_unregister(&hv_bus); 893 894 err_cleanup: 895 hv_cleanup(); 896 897 return ret; 898 } 899 900 /** 901 * __vmbus_child_driver_register() - Register a vmbus's driver 902 * @hv_driver: Pointer to driver structure you want to register 903 * @owner: owner module of the drv 904 * @mod_name: module name string 905 * 906 * Registers the given driver with Linux through the 'driver_register()' call 907 * and sets up the hyper-v vmbus handling for this driver. 908 * It will return the state of the 'driver_register()' call. 909 * 910 */ 911 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name) 912 { 913 int ret; 914 915 pr_info("registering driver %s\n", hv_driver->name); 916 917 ret = vmbus_exists(); 918 if (ret < 0) 919 return ret; 920 921 hv_driver->driver.name = hv_driver->name; 922 hv_driver->driver.owner = owner; 923 hv_driver->driver.mod_name = mod_name; 924 hv_driver->driver.bus = &hv_bus; 925 926 ret = driver_register(&hv_driver->driver); 927 928 return ret; 929 } 930 EXPORT_SYMBOL_GPL(__vmbus_driver_register); 931 932 /** 933 * vmbus_driver_unregister() - Unregister a vmbus's driver 934 * @hv_driver: Pointer to driver structure you want to 935 * un-register 936 * 937 * Un-register the given driver that was previous registered with a call to 938 * vmbus_driver_register() 939 */ 940 void vmbus_driver_unregister(struct hv_driver *hv_driver) 941 { 942 pr_info("unregistering driver %s\n", hv_driver->name); 943 944 if (!vmbus_exists()) 945 driver_unregister(&hv_driver->driver); 946 } 947 EXPORT_SYMBOL_GPL(vmbus_driver_unregister); 948 949 /* 950 * vmbus_device_create - Creates and registers a new child device 951 * on the vmbus. 952 */ 953 struct hv_device *vmbus_device_create(const uuid_le *type, 954 const uuid_le *instance, 955 struct vmbus_channel *channel) 956 { 957 struct hv_device *child_device_obj; 958 959 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL); 960 if (!child_device_obj) { 961 pr_err("Unable to allocate device object for child device\n"); 962 return NULL; 963 } 964 965 child_device_obj->channel = channel; 966 memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le)); 967 memcpy(&child_device_obj->dev_instance, instance, 968 sizeof(uuid_le)); 969 970 971 return child_device_obj; 972 } 973 974 /* 975 * vmbus_device_register - Register the child device 976 */ 977 int vmbus_device_register(struct hv_device *child_device_obj) 978 { 979 int ret = 0; 980 981 dev_set_name(&child_device_obj->device, "vmbus_%d", 982 child_device_obj->channel->id); 983 984 child_device_obj->device.bus = &hv_bus; 985 child_device_obj->device.parent = &hv_acpi_dev->dev; 986 child_device_obj->device.release = vmbus_device_release; 987 988 /* 989 * Register with the LDM. This will kick off the driver/device 990 * binding...which will eventually call vmbus_match() and vmbus_probe() 991 */ 992 ret = device_register(&child_device_obj->device); 993 994 if (ret) 995 pr_err("Unable to register child device\n"); 996 else 997 pr_debug("child device %s registered\n", 998 dev_name(&child_device_obj->device)); 999 1000 return ret; 1001 } 1002 1003 /* 1004 * vmbus_device_unregister - Remove the specified child device 1005 * from the vmbus. 1006 */ 1007 void vmbus_device_unregister(struct hv_device *device_obj) 1008 { 1009 pr_debug("child device %s unregistered\n", 1010 dev_name(&device_obj->device)); 1011 1012 /* 1013 * Kick off the process of unregistering the device. 1014 * This will call vmbus_remove() and eventually vmbus_device_release() 1015 */ 1016 device_unregister(&device_obj->device); 1017 } 1018 1019 1020 /* 1021 * VMBUS is an acpi enumerated device. Get the information we 1022 * need from DSDT. 1023 */ 1024 #define VTPM_BASE_ADDRESS 0xfed40000 1025 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx) 1026 { 1027 resource_size_t start = 0; 1028 resource_size_t end = 0; 1029 struct resource *new_res; 1030 struct resource **old_res = &hyperv_mmio; 1031 struct resource **prev_res = NULL; 1032 1033 switch (res->type) { 1034 case ACPI_RESOURCE_TYPE_IRQ: 1035 irq = res->data.irq.interrupts[0]; 1036 return AE_OK; 1037 1038 /* 1039 * "Address" descriptors are for bus windows. Ignore 1040 * "memory" descriptors, which are for registers on 1041 * devices. 1042 */ 1043 case ACPI_RESOURCE_TYPE_ADDRESS32: 1044 start = res->data.address32.address.minimum; 1045 end = res->data.address32.address.maximum; 1046 break; 1047 1048 case ACPI_RESOURCE_TYPE_ADDRESS64: 1049 start = res->data.address64.address.minimum; 1050 end = res->data.address64.address.maximum; 1051 break; 1052 1053 default: 1054 /* Unused resource type */ 1055 return AE_OK; 1056 1057 } 1058 /* 1059 * Ignore ranges that are below 1MB, as they're not 1060 * necessary or useful here. 1061 */ 1062 if (end < 0x100000) 1063 return AE_OK; 1064 1065 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC); 1066 if (!new_res) 1067 return AE_NO_MEMORY; 1068 1069 /* If this range overlaps the virtual TPM, truncate it. */ 1070 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS) 1071 end = VTPM_BASE_ADDRESS; 1072 1073 new_res->name = "hyperv mmio"; 1074 new_res->flags = IORESOURCE_MEM; 1075 new_res->start = start; 1076 new_res->end = end; 1077 1078 do { 1079 if (!*old_res) { 1080 *old_res = new_res; 1081 break; 1082 } 1083 1084 if ((*old_res)->end < new_res->start) { 1085 new_res->sibling = *old_res; 1086 if (prev_res) 1087 (*prev_res)->sibling = new_res; 1088 *old_res = new_res; 1089 break; 1090 } 1091 1092 prev_res = old_res; 1093 old_res = &(*old_res)->sibling; 1094 1095 } while (1); 1096 1097 return AE_OK; 1098 } 1099 1100 static int vmbus_acpi_remove(struct acpi_device *device) 1101 { 1102 struct resource *cur_res; 1103 struct resource *next_res; 1104 1105 if (hyperv_mmio) { 1106 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) { 1107 next_res = cur_res->sibling; 1108 kfree(cur_res); 1109 } 1110 } 1111 1112 return 0; 1113 } 1114 1115 /** 1116 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range. 1117 * @new: If successful, supplied a pointer to the 1118 * allocated MMIO space. 1119 * @device_obj: Identifies the caller 1120 * @min: Minimum guest physical address of the 1121 * allocation 1122 * @max: Maximum guest physical address 1123 * @size: Size of the range to be allocated 1124 * @align: Alignment of the range to be allocated 1125 * @fb_overlap_ok: Whether this allocation can be allowed 1126 * to overlap the video frame buffer. 1127 * 1128 * This function walks the resources granted to VMBus by the 1129 * _CRS object in the ACPI namespace underneath the parent 1130 * "bridge" whether that's a root PCI bus in the Generation 1 1131 * case or a Module Device in the Generation 2 case. It then 1132 * attempts to allocate from the global MMIO pool in a way that 1133 * matches the constraints supplied in these parameters and by 1134 * that _CRS. 1135 * 1136 * Return: 0 on success, -errno on failure 1137 */ 1138 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj, 1139 resource_size_t min, resource_size_t max, 1140 resource_size_t size, resource_size_t align, 1141 bool fb_overlap_ok) 1142 { 1143 struct resource *iter; 1144 resource_size_t range_min, range_max, start, local_min, local_max; 1145 const char *dev_n = dev_name(&device_obj->device); 1146 u32 fb_end = screen_info.lfb_base + (screen_info.lfb_size << 1); 1147 int i; 1148 1149 for (iter = hyperv_mmio; iter; iter = iter->sibling) { 1150 if ((iter->start >= max) || (iter->end <= min)) 1151 continue; 1152 1153 range_min = iter->start; 1154 range_max = iter->end; 1155 1156 /* If this range overlaps the frame buffer, split it into 1157 two tries. */ 1158 for (i = 0; i < 2; i++) { 1159 local_min = range_min; 1160 local_max = range_max; 1161 if (fb_overlap_ok || (range_min >= fb_end) || 1162 (range_max <= screen_info.lfb_base)) { 1163 i++; 1164 } else { 1165 if ((range_min <= screen_info.lfb_base) && 1166 (range_max >= screen_info.lfb_base)) { 1167 /* 1168 * The frame buffer is in this window, 1169 * so trim this into the part that 1170 * preceeds the frame buffer. 1171 */ 1172 local_max = screen_info.lfb_base - 1; 1173 range_min = fb_end; 1174 } else { 1175 range_min = fb_end; 1176 continue; 1177 } 1178 } 1179 1180 start = (local_min + align - 1) & ~(align - 1); 1181 for (; start + size - 1 <= local_max; start += align) { 1182 *new = request_mem_region_exclusive(start, size, 1183 dev_n); 1184 if (*new) 1185 return 0; 1186 } 1187 } 1188 } 1189 1190 return -ENXIO; 1191 } 1192 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio); 1193 1194 static int vmbus_acpi_add(struct acpi_device *device) 1195 { 1196 acpi_status result; 1197 int ret_val = -ENODEV; 1198 struct acpi_device *ancestor; 1199 1200 hv_acpi_dev = device; 1201 1202 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS, 1203 vmbus_walk_resources, NULL); 1204 1205 if (ACPI_FAILURE(result)) 1206 goto acpi_walk_err; 1207 /* 1208 * Some ancestor of the vmbus acpi device (Gen1 or Gen2 1209 * firmware) is the VMOD that has the mmio ranges. Get that. 1210 */ 1211 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) { 1212 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS, 1213 vmbus_walk_resources, NULL); 1214 1215 if (ACPI_FAILURE(result)) 1216 continue; 1217 if (hyperv_mmio) 1218 break; 1219 } 1220 ret_val = 0; 1221 1222 acpi_walk_err: 1223 complete(&probe_event); 1224 if (ret_val) 1225 vmbus_acpi_remove(device); 1226 return ret_val; 1227 } 1228 1229 static const struct acpi_device_id vmbus_acpi_device_ids[] = { 1230 {"VMBUS", 0}, 1231 {"VMBus", 0}, 1232 {"", 0}, 1233 }; 1234 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids); 1235 1236 static struct acpi_driver vmbus_acpi_driver = { 1237 .name = "vmbus", 1238 .ids = vmbus_acpi_device_ids, 1239 .ops = { 1240 .add = vmbus_acpi_add, 1241 .remove = vmbus_acpi_remove, 1242 }, 1243 }; 1244 1245 static void hv_kexec_handler(void) 1246 { 1247 int cpu; 1248 1249 hv_synic_clockevents_cleanup(); 1250 vmbus_initiate_unload(); 1251 for_each_online_cpu(cpu) 1252 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1); 1253 hv_cleanup(); 1254 }; 1255 1256 static void hv_crash_handler(struct pt_regs *regs) 1257 { 1258 vmbus_initiate_unload(); 1259 /* 1260 * In crash handler we can't schedule synic cleanup for all CPUs, 1261 * doing the cleanup for current CPU only. This should be sufficient 1262 * for kdump. 1263 */ 1264 hv_synic_cleanup(NULL); 1265 hv_cleanup(); 1266 }; 1267 1268 static int __init hv_acpi_init(void) 1269 { 1270 int ret, t; 1271 1272 if (x86_hyper != &x86_hyper_ms_hyperv) 1273 return -ENODEV; 1274 1275 init_completion(&probe_event); 1276 1277 /* 1278 * Get irq resources first. 1279 */ 1280 ret = acpi_bus_register_driver(&vmbus_acpi_driver); 1281 1282 if (ret) 1283 return ret; 1284 1285 t = wait_for_completion_timeout(&probe_event, 5*HZ); 1286 if (t == 0) { 1287 ret = -ETIMEDOUT; 1288 goto cleanup; 1289 } 1290 1291 if (irq <= 0) { 1292 ret = -ENODEV; 1293 goto cleanup; 1294 } 1295 1296 ret = vmbus_bus_init(irq); 1297 if (ret) 1298 goto cleanup; 1299 1300 hv_setup_kexec_handler(hv_kexec_handler); 1301 hv_setup_crash_handler(hv_crash_handler); 1302 1303 return 0; 1304 1305 cleanup: 1306 acpi_bus_unregister_driver(&vmbus_acpi_driver); 1307 hv_acpi_dev = NULL; 1308 return ret; 1309 } 1310 1311 static void __exit vmbus_exit(void) 1312 { 1313 int cpu; 1314 1315 hv_remove_kexec_handler(); 1316 hv_remove_crash_handler(); 1317 vmbus_connection.conn_state = DISCONNECTED; 1318 hv_synic_clockevents_cleanup(); 1319 vmbus_disconnect(); 1320 hv_remove_vmbus_irq(); 1321 tasklet_kill(&msg_dpc); 1322 vmbus_free_channels(); 1323 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) { 1324 unregister_die_notifier(&hyperv_die_block); 1325 atomic_notifier_chain_unregister(&panic_notifier_list, 1326 &hyperv_panic_block); 1327 } 1328 bus_unregister(&hv_bus); 1329 hv_cleanup(); 1330 for_each_online_cpu(cpu) { 1331 tasklet_kill(hv_context.event_dpc[cpu]); 1332 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1); 1333 } 1334 hv_synic_free(); 1335 acpi_bus_unregister_driver(&vmbus_acpi_driver); 1336 if (vmbus_proto_version > VERSION_WIN7) 1337 cpu_hotplug_enable(); 1338 } 1339 1340 1341 MODULE_LICENSE("GPL"); 1342 1343 subsys_initcall(hv_acpi_init); 1344 module_exit(vmbus_exit); 1345