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 <linux/efi.h> 45 #include <linux/random.h> 46 #include "hyperv_vmbus.h" 47 48 static struct acpi_device *hv_acpi_dev; 49 50 static struct completion probe_event; 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 static const char *fb_mmio_name = "fb_range"; 107 static struct resource *fb_mmio; 108 struct resource *hyperv_mmio; 109 DEFINE_SEMAPHORE(hyperv_mmio_lock); 110 111 static int vmbus_exists(void) 112 { 113 if (hv_acpi_dev == NULL) 114 return -ENODEV; 115 116 return 0; 117 } 118 119 #define VMBUS_ALIAS_LEN ((sizeof((struct hv_vmbus_device_id *)0)->guid) * 2) 120 static void print_alias_name(struct hv_device *hv_dev, char *alias_name) 121 { 122 int i; 123 for (i = 0; i < VMBUS_ALIAS_LEN; i += 2) 124 sprintf(&alias_name[i], "%02x", hv_dev->dev_type.b[i/2]); 125 } 126 127 static u8 channel_monitor_group(struct vmbus_channel *channel) 128 { 129 return (u8)channel->offermsg.monitorid / 32; 130 } 131 132 static u8 channel_monitor_offset(struct vmbus_channel *channel) 133 { 134 return (u8)channel->offermsg.monitorid % 32; 135 } 136 137 static u32 channel_pending(struct vmbus_channel *channel, 138 struct hv_monitor_page *monitor_page) 139 { 140 u8 monitor_group = channel_monitor_group(channel); 141 return monitor_page->trigger_group[monitor_group].pending; 142 } 143 144 static u32 channel_latency(struct vmbus_channel *channel, 145 struct hv_monitor_page *monitor_page) 146 { 147 u8 monitor_group = channel_monitor_group(channel); 148 u8 monitor_offset = channel_monitor_offset(channel); 149 return monitor_page->latency[monitor_group][monitor_offset]; 150 } 151 152 static u32 channel_conn_id(struct vmbus_channel *channel, 153 struct hv_monitor_page *monitor_page) 154 { 155 u8 monitor_group = channel_monitor_group(channel); 156 u8 monitor_offset = channel_monitor_offset(channel); 157 return monitor_page->parameter[monitor_group][monitor_offset].connectionid.u.id; 158 } 159 160 static ssize_t id_show(struct device *dev, struct device_attribute *dev_attr, 161 char *buf) 162 { 163 struct hv_device *hv_dev = device_to_hv_device(dev); 164 165 if (!hv_dev->channel) 166 return -ENODEV; 167 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.child_relid); 168 } 169 static DEVICE_ATTR_RO(id); 170 171 static ssize_t state_show(struct device *dev, struct device_attribute *dev_attr, 172 char *buf) 173 { 174 struct hv_device *hv_dev = device_to_hv_device(dev); 175 176 if (!hv_dev->channel) 177 return -ENODEV; 178 return sprintf(buf, "%d\n", hv_dev->channel->state); 179 } 180 static DEVICE_ATTR_RO(state); 181 182 static ssize_t monitor_id_show(struct device *dev, 183 struct device_attribute *dev_attr, char *buf) 184 { 185 struct hv_device *hv_dev = device_to_hv_device(dev); 186 187 if (!hv_dev->channel) 188 return -ENODEV; 189 return sprintf(buf, "%d\n", hv_dev->channel->offermsg.monitorid); 190 } 191 static DEVICE_ATTR_RO(monitor_id); 192 193 static ssize_t class_id_show(struct device *dev, 194 struct device_attribute *dev_attr, char *buf) 195 { 196 struct hv_device *hv_dev = device_to_hv_device(dev); 197 198 if (!hv_dev->channel) 199 return -ENODEV; 200 return sprintf(buf, "{%pUl}\n", 201 hv_dev->channel->offermsg.offer.if_type.b); 202 } 203 static DEVICE_ATTR_RO(class_id); 204 205 static ssize_t device_id_show(struct device *dev, 206 struct device_attribute *dev_attr, char *buf) 207 { 208 struct hv_device *hv_dev = device_to_hv_device(dev); 209 210 if (!hv_dev->channel) 211 return -ENODEV; 212 return sprintf(buf, "{%pUl}\n", 213 hv_dev->channel->offermsg.offer.if_instance.b); 214 } 215 static DEVICE_ATTR_RO(device_id); 216 217 static ssize_t modalias_show(struct device *dev, 218 struct device_attribute *dev_attr, char *buf) 219 { 220 struct hv_device *hv_dev = device_to_hv_device(dev); 221 char alias_name[VMBUS_ALIAS_LEN + 1]; 222 223 print_alias_name(hv_dev, alias_name); 224 return sprintf(buf, "vmbus:%s\n", alias_name); 225 } 226 static DEVICE_ATTR_RO(modalias); 227 228 static ssize_t server_monitor_pending_show(struct device *dev, 229 struct device_attribute *dev_attr, 230 char *buf) 231 { 232 struct hv_device *hv_dev = device_to_hv_device(dev); 233 234 if (!hv_dev->channel) 235 return -ENODEV; 236 return sprintf(buf, "%d\n", 237 channel_pending(hv_dev->channel, 238 vmbus_connection.monitor_pages[1])); 239 } 240 static DEVICE_ATTR_RO(server_monitor_pending); 241 242 static ssize_t client_monitor_pending_show(struct device *dev, 243 struct device_attribute *dev_attr, 244 char *buf) 245 { 246 struct hv_device *hv_dev = device_to_hv_device(dev); 247 248 if (!hv_dev->channel) 249 return -ENODEV; 250 return sprintf(buf, "%d\n", 251 channel_pending(hv_dev->channel, 252 vmbus_connection.monitor_pages[1])); 253 } 254 static DEVICE_ATTR_RO(client_monitor_pending); 255 256 static ssize_t server_monitor_latency_show(struct device *dev, 257 struct device_attribute *dev_attr, 258 char *buf) 259 { 260 struct hv_device *hv_dev = device_to_hv_device(dev); 261 262 if (!hv_dev->channel) 263 return -ENODEV; 264 return sprintf(buf, "%d\n", 265 channel_latency(hv_dev->channel, 266 vmbus_connection.monitor_pages[0])); 267 } 268 static DEVICE_ATTR_RO(server_monitor_latency); 269 270 static ssize_t client_monitor_latency_show(struct device *dev, 271 struct device_attribute *dev_attr, 272 char *buf) 273 { 274 struct hv_device *hv_dev = device_to_hv_device(dev); 275 276 if (!hv_dev->channel) 277 return -ENODEV; 278 return sprintf(buf, "%d\n", 279 channel_latency(hv_dev->channel, 280 vmbus_connection.monitor_pages[1])); 281 } 282 static DEVICE_ATTR_RO(client_monitor_latency); 283 284 static ssize_t server_monitor_conn_id_show(struct device *dev, 285 struct device_attribute *dev_attr, 286 char *buf) 287 { 288 struct hv_device *hv_dev = device_to_hv_device(dev); 289 290 if (!hv_dev->channel) 291 return -ENODEV; 292 return sprintf(buf, "%d\n", 293 channel_conn_id(hv_dev->channel, 294 vmbus_connection.monitor_pages[0])); 295 } 296 static DEVICE_ATTR_RO(server_monitor_conn_id); 297 298 static ssize_t client_monitor_conn_id_show(struct device *dev, 299 struct device_attribute *dev_attr, 300 char *buf) 301 { 302 struct hv_device *hv_dev = device_to_hv_device(dev); 303 304 if (!hv_dev->channel) 305 return -ENODEV; 306 return sprintf(buf, "%d\n", 307 channel_conn_id(hv_dev->channel, 308 vmbus_connection.monitor_pages[1])); 309 } 310 static DEVICE_ATTR_RO(client_monitor_conn_id); 311 312 static ssize_t out_intr_mask_show(struct device *dev, 313 struct device_attribute *dev_attr, char *buf) 314 { 315 struct hv_device *hv_dev = device_to_hv_device(dev); 316 struct hv_ring_buffer_debug_info outbound; 317 318 if (!hv_dev->channel) 319 return -ENODEV; 320 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 321 return sprintf(buf, "%d\n", outbound.current_interrupt_mask); 322 } 323 static DEVICE_ATTR_RO(out_intr_mask); 324 325 static ssize_t out_read_index_show(struct device *dev, 326 struct device_attribute *dev_attr, char *buf) 327 { 328 struct hv_device *hv_dev = device_to_hv_device(dev); 329 struct hv_ring_buffer_debug_info outbound; 330 331 if (!hv_dev->channel) 332 return -ENODEV; 333 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 334 return sprintf(buf, "%d\n", outbound.current_read_index); 335 } 336 static DEVICE_ATTR_RO(out_read_index); 337 338 static ssize_t out_write_index_show(struct device *dev, 339 struct device_attribute *dev_attr, 340 char *buf) 341 { 342 struct hv_device *hv_dev = device_to_hv_device(dev); 343 struct hv_ring_buffer_debug_info outbound; 344 345 if (!hv_dev->channel) 346 return -ENODEV; 347 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 348 return sprintf(buf, "%d\n", outbound.current_write_index); 349 } 350 static DEVICE_ATTR_RO(out_write_index); 351 352 static ssize_t out_read_bytes_avail_show(struct device *dev, 353 struct device_attribute *dev_attr, 354 char *buf) 355 { 356 struct hv_device *hv_dev = device_to_hv_device(dev); 357 struct hv_ring_buffer_debug_info outbound; 358 359 if (!hv_dev->channel) 360 return -ENODEV; 361 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 362 return sprintf(buf, "%d\n", outbound.bytes_avail_toread); 363 } 364 static DEVICE_ATTR_RO(out_read_bytes_avail); 365 366 static ssize_t out_write_bytes_avail_show(struct device *dev, 367 struct device_attribute *dev_attr, 368 char *buf) 369 { 370 struct hv_device *hv_dev = device_to_hv_device(dev); 371 struct hv_ring_buffer_debug_info outbound; 372 373 if (!hv_dev->channel) 374 return -ENODEV; 375 hv_ringbuffer_get_debuginfo(&hv_dev->channel->outbound, &outbound); 376 return sprintf(buf, "%d\n", outbound.bytes_avail_towrite); 377 } 378 static DEVICE_ATTR_RO(out_write_bytes_avail); 379 380 static ssize_t in_intr_mask_show(struct device *dev, 381 struct device_attribute *dev_attr, char *buf) 382 { 383 struct hv_device *hv_dev = device_to_hv_device(dev); 384 struct hv_ring_buffer_debug_info inbound; 385 386 if (!hv_dev->channel) 387 return -ENODEV; 388 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 389 return sprintf(buf, "%d\n", inbound.current_interrupt_mask); 390 } 391 static DEVICE_ATTR_RO(in_intr_mask); 392 393 static ssize_t in_read_index_show(struct device *dev, 394 struct device_attribute *dev_attr, char *buf) 395 { 396 struct hv_device *hv_dev = device_to_hv_device(dev); 397 struct hv_ring_buffer_debug_info inbound; 398 399 if (!hv_dev->channel) 400 return -ENODEV; 401 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 402 return sprintf(buf, "%d\n", inbound.current_read_index); 403 } 404 static DEVICE_ATTR_RO(in_read_index); 405 406 static ssize_t in_write_index_show(struct device *dev, 407 struct device_attribute *dev_attr, char *buf) 408 { 409 struct hv_device *hv_dev = device_to_hv_device(dev); 410 struct hv_ring_buffer_debug_info inbound; 411 412 if (!hv_dev->channel) 413 return -ENODEV; 414 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 415 return sprintf(buf, "%d\n", inbound.current_write_index); 416 } 417 static DEVICE_ATTR_RO(in_write_index); 418 419 static ssize_t in_read_bytes_avail_show(struct device *dev, 420 struct device_attribute *dev_attr, 421 char *buf) 422 { 423 struct hv_device *hv_dev = device_to_hv_device(dev); 424 struct hv_ring_buffer_debug_info inbound; 425 426 if (!hv_dev->channel) 427 return -ENODEV; 428 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 429 return sprintf(buf, "%d\n", inbound.bytes_avail_toread); 430 } 431 static DEVICE_ATTR_RO(in_read_bytes_avail); 432 433 static ssize_t in_write_bytes_avail_show(struct device *dev, 434 struct device_attribute *dev_attr, 435 char *buf) 436 { 437 struct hv_device *hv_dev = device_to_hv_device(dev); 438 struct hv_ring_buffer_debug_info inbound; 439 440 if (!hv_dev->channel) 441 return -ENODEV; 442 hv_ringbuffer_get_debuginfo(&hv_dev->channel->inbound, &inbound); 443 return sprintf(buf, "%d\n", inbound.bytes_avail_towrite); 444 } 445 static DEVICE_ATTR_RO(in_write_bytes_avail); 446 447 static ssize_t channel_vp_mapping_show(struct device *dev, 448 struct device_attribute *dev_attr, 449 char *buf) 450 { 451 struct hv_device *hv_dev = device_to_hv_device(dev); 452 struct vmbus_channel *channel = hv_dev->channel, *cur_sc; 453 unsigned long flags; 454 int buf_size = PAGE_SIZE, n_written, tot_written; 455 struct list_head *cur; 456 457 if (!channel) 458 return -ENODEV; 459 460 tot_written = snprintf(buf, buf_size, "%u:%u\n", 461 channel->offermsg.child_relid, channel->target_cpu); 462 463 spin_lock_irqsave(&channel->lock, flags); 464 465 list_for_each(cur, &channel->sc_list) { 466 if (tot_written >= buf_size - 1) 467 break; 468 469 cur_sc = list_entry(cur, struct vmbus_channel, sc_list); 470 n_written = scnprintf(buf + tot_written, 471 buf_size - tot_written, 472 "%u:%u\n", 473 cur_sc->offermsg.child_relid, 474 cur_sc->target_cpu); 475 tot_written += n_written; 476 } 477 478 spin_unlock_irqrestore(&channel->lock, flags); 479 480 return tot_written; 481 } 482 static DEVICE_ATTR_RO(channel_vp_mapping); 483 484 static ssize_t vendor_show(struct device *dev, 485 struct device_attribute *dev_attr, 486 char *buf) 487 { 488 struct hv_device *hv_dev = device_to_hv_device(dev); 489 return sprintf(buf, "0x%x\n", hv_dev->vendor_id); 490 } 491 static DEVICE_ATTR_RO(vendor); 492 493 static ssize_t device_show(struct device *dev, 494 struct device_attribute *dev_attr, 495 char *buf) 496 { 497 struct hv_device *hv_dev = device_to_hv_device(dev); 498 return sprintf(buf, "0x%x\n", hv_dev->device_id); 499 } 500 static DEVICE_ATTR_RO(device); 501 502 /* Set up per device attributes in /sys/bus/vmbus/devices/<bus device> */ 503 static struct attribute *vmbus_attrs[] = { 504 &dev_attr_id.attr, 505 &dev_attr_state.attr, 506 &dev_attr_monitor_id.attr, 507 &dev_attr_class_id.attr, 508 &dev_attr_device_id.attr, 509 &dev_attr_modalias.attr, 510 &dev_attr_server_monitor_pending.attr, 511 &dev_attr_client_monitor_pending.attr, 512 &dev_attr_server_monitor_latency.attr, 513 &dev_attr_client_monitor_latency.attr, 514 &dev_attr_server_monitor_conn_id.attr, 515 &dev_attr_client_monitor_conn_id.attr, 516 &dev_attr_out_intr_mask.attr, 517 &dev_attr_out_read_index.attr, 518 &dev_attr_out_write_index.attr, 519 &dev_attr_out_read_bytes_avail.attr, 520 &dev_attr_out_write_bytes_avail.attr, 521 &dev_attr_in_intr_mask.attr, 522 &dev_attr_in_read_index.attr, 523 &dev_attr_in_write_index.attr, 524 &dev_attr_in_read_bytes_avail.attr, 525 &dev_attr_in_write_bytes_avail.attr, 526 &dev_attr_channel_vp_mapping.attr, 527 &dev_attr_vendor.attr, 528 &dev_attr_device.attr, 529 NULL, 530 }; 531 ATTRIBUTE_GROUPS(vmbus); 532 533 /* 534 * vmbus_uevent - add uevent for our device 535 * 536 * This routine is invoked when a device is added or removed on the vmbus to 537 * generate a uevent to udev in the userspace. The udev will then look at its 538 * rule and the uevent generated here to load the appropriate driver 539 * 540 * The alias string will be of the form vmbus:guid where guid is the string 541 * representation of the device guid (each byte of the guid will be 542 * represented with two hex characters. 543 */ 544 static int vmbus_uevent(struct device *device, struct kobj_uevent_env *env) 545 { 546 struct hv_device *dev = device_to_hv_device(device); 547 int ret; 548 char alias_name[VMBUS_ALIAS_LEN + 1]; 549 550 print_alias_name(dev, alias_name); 551 ret = add_uevent_var(env, "MODALIAS=vmbus:%s", alias_name); 552 return ret; 553 } 554 555 static const uuid_le null_guid; 556 557 static inline bool is_null_guid(const uuid_le *guid) 558 { 559 if (uuid_le_cmp(*guid, null_guid)) 560 return false; 561 return true; 562 } 563 564 /* 565 * Return a matching hv_vmbus_device_id pointer. 566 * If there is no match, return NULL. 567 */ 568 static const struct hv_vmbus_device_id *hv_vmbus_get_id( 569 const struct hv_vmbus_device_id *id, 570 const uuid_le *guid) 571 { 572 for (; !is_null_guid(&id->guid); id++) 573 if (!uuid_le_cmp(id->guid, *guid)) 574 return id; 575 576 return NULL; 577 } 578 579 580 581 /* 582 * vmbus_match - Attempt to match the specified device to the specified driver 583 */ 584 static int vmbus_match(struct device *device, struct device_driver *driver) 585 { 586 struct hv_driver *drv = drv_to_hv_drv(driver); 587 struct hv_device *hv_dev = device_to_hv_device(device); 588 589 /* The hv_sock driver handles all hv_sock offers. */ 590 if (is_hvsock_channel(hv_dev->channel)) 591 return drv->hvsock; 592 593 if (hv_vmbus_get_id(drv->id_table, &hv_dev->dev_type)) 594 return 1; 595 596 return 0; 597 } 598 599 /* 600 * vmbus_probe - Add the new vmbus's child device 601 */ 602 static int vmbus_probe(struct device *child_device) 603 { 604 int ret = 0; 605 struct hv_driver *drv = 606 drv_to_hv_drv(child_device->driver); 607 struct hv_device *dev = device_to_hv_device(child_device); 608 const struct hv_vmbus_device_id *dev_id; 609 610 dev_id = hv_vmbus_get_id(drv->id_table, &dev->dev_type); 611 if (drv->probe) { 612 ret = drv->probe(dev, dev_id); 613 if (ret != 0) 614 pr_err("probe failed for device %s (%d)\n", 615 dev_name(child_device), ret); 616 617 } else { 618 pr_err("probe not set for driver %s\n", 619 dev_name(child_device)); 620 ret = -ENODEV; 621 } 622 return ret; 623 } 624 625 /* 626 * vmbus_remove - Remove a vmbus device 627 */ 628 static int vmbus_remove(struct device *child_device) 629 { 630 struct hv_driver *drv; 631 struct hv_device *dev = device_to_hv_device(child_device); 632 633 if (child_device->driver) { 634 drv = drv_to_hv_drv(child_device->driver); 635 if (drv->remove) 636 drv->remove(dev); 637 } 638 639 return 0; 640 } 641 642 643 /* 644 * vmbus_shutdown - Shutdown a vmbus device 645 */ 646 static void vmbus_shutdown(struct device *child_device) 647 { 648 struct hv_driver *drv; 649 struct hv_device *dev = device_to_hv_device(child_device); 650 651 652 /* The device may not be attached yet */ 653 if (!child_device->driver) 654 return; 655 656 drv = drv_to_hv_drv(child_device->driver); 657 658 if (drv->shutdown) 659 drv->shutdown(dev); 660 661 return; 662 } 663 664 665 /* 666 * vmbus_device_release - Final callback release of the vmbus child device 667 */ 668 static void vmbus_device_release(struct device *device) 669 { 670 struct hv_device *hv_dev = device_to_hv_device(device); 671 struct vmbus_channel *channel = hv_dev->channel; 672 673 hv_process_channel_removal(channel, 674 channel->offermsg.child_relid); 675 kfree(hv_dev); 676 677 } 678 679 /* The one and only one */ 680 static struct bus_type hv_bus = { 681 .name = "vmbus", 682 .match = vmbus_match, 683 .shutdown = vmbus_shutdown, 684 .remove = vmbus_remove, 685 .probe = vmbus_probe, 686 .uevent = vmbus_uevent, 687 .dev_groups = vmbus_groups, 688 }; 689 690 struct onmessage_work_context { 691 struct work_struct work; 692 struct hv_message msg; 693 }; 694 695 static void vmbus_onmessage_work(struct work_struct *work) 696 { 697 struct onmessage_work_context *ctx; 698 699 /* Do not process messages if we're in DISCONNECTED state */ 700 if (vmbus_connection.conn_state == DISCONNECTED) 701 return; 702 703 ctx = container_of(work, struct onmessage_work_context, 704 work); 705 vmbus_onmessage(&ctx->msg); 706 kfree(ctx); 707 } 708 709 static void hv_process_timer_expiration(struct hv_message *msg, int cpu) 710 { 711 struct clock_event_device *dev = hv_context.clk_evt[cpu]; 712 713 if (dev->event_handler) 714 dev->event_handler(dev); 715 716 vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED); 717 } 718 719 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 u32 message_type = msg->header.message_type; 729 730 if (message_type == HVMSG_NONE) 731 /* no msg */ 732 return; 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 return; 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 vmbus_signal_eom(msg, message_type); 756 } 757 758 static void vmbus_isr(void) 759 { 760 int cpu = smp_processor_id(); 761 void *page_addr; 762 struct hv_message *msg; 763 union hv_synic_event_flags *event; 764 bool handled = false; 765 766 page_addr = hv_context.synic_event_page[cpu]; 767 if (page_addr == NULL) 768 return; 769 770 event = (union hv_synic_event_flags *)page_addr + 771 VMBUS_MESSAGE_SINT; 772 /* 773 * Check for events before checking for messages. This is the order 774 * in which events and messages are checked in Windows guests on 775 * Hyper-V, and the Windows team suggested we do the same. 776 */ 777 778 if ((vmbus_proto_version == VERSION_WS2008) || 779 (vmbus_proto_version == VERSION_WIN7)) { 780 781 /* Since we are a child, we only need to check bit 0 */ 782 if (sync_test_and_clear_bit(0, 783 (unsigned long *) &event->flags32[0])) { 784 handled = true; 785 } 786 } else { 787 /* 788 * Our host is win8 or above. The signaling mechanism 789 * has changed and we can directly look at the event page. 790 * If bit n is set then we have an interrup on the channel 791 * whose id is n. 792 */ 793 handled = true; 794 } 795 796 if (handled) 797 tasklet_schedule(hv_context.event_dpc[cpu]); 798 799 800 page_addr = hv_context.synic_message_page[cpu]; 801 msg = (struct hv_message *)page_addr + VMBUS_MESSAGE_SINT; 802 803 /* Check if there are actual msgs to be processed */ 804 if (msg->header.message_type != HVMSG_NONE) { 805 if (msg->header.message_type == HVMSG_TIMER_EXPIRED) 806 hv_process_timer_expiration(msg, cpu); 807 else 808 tasklet_schedule(hv_context.msg_dpc[cpu]); 809 } 810 811 add_interrupt_randomness(HYPERVISOR_CALLBACK_VECTOR, 0); 812 } 813 814 815 /* 816 * vmbus_bus_init -Main vmbus driver initialization routine. 817 * 818 * Here, we 819 * - initialize the vmbus driver context 820 * - invoke the vmbus hv main init routine 821 * - retrieve the channel offers 822 */ 823 static int vmbus_bus_init(void) 824 { 825 int ret; 826 827 /* Hypervisor initialization...setup hypercall page..etc */ 828 ret = hv_init(); 829 if (ret != 0) { 830 pr_err("Unable to initialize the hypervisor - 0x%x\n", ret); 831 return ret; 832 } 833 834 ret = bus_register(&hv_bus); 835 if (ret) 836 goto err_cleanup; 837 838 hv_setup_vmbus_irq(vmbus_isr); 839 840 ret = hv_synic_alloc(); 841 if (ret) 842 goto err_alloc; 843 /* 844 * Initialize the per-cpu interrupt state and 845 * connect to the host. 846 */ 847 on_each_cpu(hv_synic_init, NULL, 1); 848 ret = vmbus_connect(); 849 if (ret) 850 goto err_connect; 851 852 if (vmbus_proto_version > VERSION_WIN7) 853 cpu_hotplug_disable(); 854 855 /* 856 * Only register if the crash MSRs are available 857 */ 858 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) { 859 register_die_notifier(&hyperv_die_block); 860 atomic_notifier_chain_register(&panic_notifier_list, 861 &hyperv_panic_block); 862 } 863 864 vmbus_request_offers(); 865 866 return 0; 867 868 err_connect: 869 on_each_cpu(hv_synic_cleanup, NULL, 1); 870 err_alloc: 871 hv_synic_free(); 872 hv_remove_vmbus_irq(); 873 874 bus_unregister(&hv_bus); 875 876 err_cleanup: 877 hv_cleanup(); 878 879 return ret; 880 } 881 882 /** 883 * __vmbus_child_driver_register() - Register a vmbus's driver 884 * @hv_driver: Pointer to driver structure you want to register 885 * @owner: owner module of the drv 886 * @mod_name: module name string 887 * 888 * Registers the given driver with Linux through the 'driver_register()' call 889 * and sets up the hyper-v vmbus handling for this driver. 890 * It will return the state of the 'driver_register()' call. 891 * 892 */ 893 int __vmbus_driver_register(struct hv_driver *hv_driver, struct module *owner, const char *mod_name) 894 { 895 int ret; 896 897 pr_info("registering driver %s\n", hv_driver->name); 898 899 ret = vmbus_exists(); 900 if (ret < 0) 901 return ret; 902 903 hv_driver->driver.name = hv_driver->name; 904 hv_driver->driver.owner = owner; 905 hv_driver->driver.mod_name = mod_name; 906 hv_driver->driver.bus = &hv_bus; 907 908 ret = driver_register(&hv_driver->driver); 909 910 return ret; 911 } 912 EXPORT_SYMBOL_GPL(__vmbus_driver_register); 913 914 /** 915 * vmbus_driver_unregister() - Unregister a vmbus's driver 916 * @hv_driver: Pointer to driver structure you want to 917 * un-register 918 * 919 * Un-register the given driver that was previous registered with a call to 920 * vmbus_driver_register() 921 */ 922 void vmbus_driver_unregister(struct hv_driver *hv_driver) 923 { 924 pr_info("unregistering driver %s\n", hv_driver->name); 925 926 if (!vmbus_exists()) 927 driver_unregister(&hv_driver->driver); 928 } 929 EXPORT_SYMBOL_GPL(vmbus_driver_unregister); 930 931 /* 932 * vmbus_device_create - Creates and registers a new child device 933 * on the vmbus. 934 */ 935 struct hv_device *vmbus_device_create(const uuid_le *type, 936 const uuid_le *instance, 937 struct vmbus_channel *channel) 938 { 939 struct hv_device *child_device_obj; 940 941 child_device_obj = kzalloc(sizeof(struct hv_device), GFP_KERNEL); 942 if (!child_device_obj) { 943 pr_err("Unable to allocate device object for child device\n"); 944 return NULL; 945 } 946 947 child_device_obj->channel = channel; 948 memcpy(&child_device_obj->dev_type, type, sizeof(uuid_le)); 949 memcpy(&child_device_obj->dev_instance, instance, 950 sizeof(uuid_le)); 951 child_device_obj->vendor_id = 0x1414; /* MSFT vendor ID */ 952 953 954 return child_device_obj; 955 } 956 957 /* 958 * vmbus_device_register - Register the child device 959 */ 960 int vmbus_device_register(struct hv_device *child_device_obj) 961 { 962 int ret = 0; 963 964 dev_set_name(&child_device_obj->device, "vmbus_%d", 965 child_device_obj->channel->id); 966 967 child_device_obj->device.bus = &hv_bus; 968 child_device_obj->device.parent = &hv_acpi_dev->dev; 969 child_device_obj->device.release = vmbus_device_release; 970 971 /* 972 * Register with the LDM. This will kick off the driver/device 973 * binding...which will eventually call vmbus_match() and vmbus_probe() 974 */ 975 ret = device_register(&child_device_obj->device); 976 977 if (ret) 978 pr_err("Unable to register child device\n"); 979 else 980 pr_debug("child device %s registered\n", 981 dev_name(&child_device_obj->device)); 982 983 return ret; 984 } 985 986 /* 987 * vmbus_device_unregister - Remove the specified child device 988 * from the vmbus. 989 */ 990 void vmbus_device_unregister(struct hv_device *device_obj) 991 { 992 pr_debug("child device %s unregistered\n", 993 dev_name(&device_obj->device)); 994 995 /* 996 * Kick off the process of unregistering the device. 997 * This will call vmbus_remove() and eventually vmbus_device_release() 998 */ 999 device_unregister(&device_obj->device); 1000 } 1001 1002 1003 /* 1004 * VMBUS is an acpi enumerated device. Get the information we 1005 * need from DSDT. 1006 */ 1007 #define VTPM_BASE_ADDRESS 0xfed40000 1008 static acpi_status vmbus_walk_resources(struct acpi_resource *res, void *ctx) 1009 { 1010 resource_size_t start = 0; 1011 resource_size_t end = 0; 1012 struct resource *new_res; 1013 struct resource **old_res = &hyperv_mmio; 1014 struct resource **prev_res = NULL; 1015 1016 switch (res->type) { 1017 1018 /* 1019 * "Address" descriptors are for bus windows. Ignore 1020 * "memory" descriptors, which are for registers on 1021 * devices. 1022 */ 1023 case ACPI_RESOURCE_TYPE_ADDRESS32: 1024 start = res->data.address32.address.minimum; 1025 end = res->data.address32.address.maximum; 1026 break; 1027 1028 case ACPI_RESOURCE_TYPE_ADDRESS64: 1029 start = res->data.address64.address.minimum; 1030 end = res->data.address64.address.maximum; 1031 break; 1032 1033 default: 1034 /* Unused resource type */ 1035 return AE_OK; 1036 1037 } 1038 /* 1039 * Ignore ranges that are below 1MB, as they're not 1040 * necessary or useful here. 1041 */ 1042 if (end < 0x100000) 1043 return AE_OK; 1044 1045 new_res = kzalloc(sizeof(*new_res), GFP_ATOMIC); 1046 if (!new_res) 1047 return AE_NO_MEMORY; 1048 1049 /* If this range overlaps the virtual TPM, truncate it. */ 1050 if (end > VTPM_BASE_ADDRESS && start < VTPM_BASE_ADDRESS) 1051 end = VTPM_BASE_ADDRESS; 1052 1053 new_res->name = "hyperv mmio"; 1054 new_res->flags = IORESOURCE_MEM; 1055 new_res->start = start; 1056 new_res->end = end; 1057 1058 /* 1059 * If two ranges are adjacent, merge them. 1060 */ 1061 do { 1062 if (!*old_res) { 1063 *old_res = new_res; 1064 break; 1065 } 1066 1067 if (((*old_res)->end + 1) == new_res->start) { 1068 (*old_res)->end = new_res->end; 1069 kfree(new_res); 1070 break; 1071 } 1072 1073 if ((*old_res)->start == new_res->end + 1) { 1074 (*old_res)->start = new_res->start; 1075 kfree(new_res); 1076 break; 1077 } 1078 1079 if ((*old_res)->start > new_res->end) { 1080 new_res->sibling = *old_res; 1081 if (prev_res) 1082 (*prev_res)->sibling = new_res; 1083 *old_res = new_res; 1084 break; 1085 } 1086 1087 prev_res = old_res; 1088 old_res = &(*old_res)->sibling; 1089 1090 } while (1); 1091 1092 return AE_OK; 1093 } 1094 1095 static int vmbus_acpi_remove(struct acpi_device *device) 1096 { 1097 struct resource *cur_res; 1098 struct resource *next_res; 1099 1100 if (hyperv_mmio) { 1101 if (fb_mmio) { 1102 __release_region(hyperv_mmio, fb_mmio->start, 1103 resource_size(fb_mmio)); 1104 fb_mmio = NULL; 1105 } 1106 1107 for (cur_res = hyperv_mmio; cur_res; cur_res = next_res) { 1108 next_res = cur_res->sibling; 1109 kfree(cur_res); 1110 } 1111 } 1112 1113 return 0; 1114 } 1115 1116 static void vmbus_reserve_fb(void) 1117 { 1118 int size; 1119 /* 1120 * Make a claim for the frame buffer in the resource tree under the 1121 * first node, which will be the one below 4GB. The length seems to 1122 * be underreported, particularly in a Generation 1 VM. So start out 1123 * reserving a larger area and make it smaller until it succeeds. 1124 */ 1125 1126 if (screen_info.lfb_base) { 1127 if (efi_enabled(EFI_BOOT)) 1128 size = max_t(__u32, screen_info.lfb_size, 0x800000); 1129 else 1130 size = max_t(__u32, screen_info.lfb_size, 0x4000000); 1131 1132 for (; !fb_mmio && (size >= 0x100000); size >>= 1) { 1133 fb_mmio = __request_region(hyperv_mmio, 1134 screen_info.lfb_base, size, 1135 fb_mmio_name, 0); 1136 } 1137 } 1138 } 1139 1140 /** 1141 * vmbus_allocate_mmio() - Pick a memory-mapped I/O range. 1142 * @new: If successful, supplied a pointer to the 1143 * allocated MMIO space. 1144 * @device_obj: Identifies the caller 1145 * @min: Minimum guest physical address of the 1146 * allocation 1147 * @max: Maximum guest physical address 1148 * @size: Size of the range to be allocated 1149 * @align: Alignment of the range to be allocated 1150 * @fb_overlap_ok: Whether this allocation can be allowed 1151 * to overlap the video frame buffer. 1152 * 1153 * This function walks the resources granted to VMBus by the 1154 * _CRS object in the ACPI namespace underneath the parent 1155 * "bridge" whether that's a root PCI bus in the Generation 1 1156 * case or a Module Device in the Generation 2 case. It then 1157 * attempts to allocate from the global MMIO pool in a way that 1158 * matches the constraints supplied in these parameters and by 1159 * that _CRS. 1160 * 1161 * Return: 0 on success, -errno on failure 1162 */ 1163 int vmbus_allocate_mmio(struct resource **new, struct hv_device *device_obj, 1164 resource_size_t min, resource_size_t max, 1165 resource_size_t size, resource_size_t align, 1166 bool fb_overlap_ok) 1167 { 1168 struct resource *iter, *shadow; 1169 resource_size_t range_min, range_max, start; 1170 const char *dev_n = dev_name(&device_obj->device); 1171 int retval; 1172 1173 retval = -ENXIO; 1174 down(&hyperv_mmio_lock); 1175 1176 /* 1177 * If overlaps with frame buffers are allowed, then first attempt to 1178 * make the allocation from within the reserved region. Because it 1179 * is already reserved, no shadow allocation is necessary. 1180 */ 1181 if (fb_overlap_ok && fb_mmio && !(min > fb_mmio->end) && 1182 !(max < fb_mmio->start)) { 1183 1184 range_min = fb_mmio->start; 1185 range_max = fb_mmio->end; 1186 start = (range_min + align - 1) & ~(align - 1); 1187 for (; start + size - 1 <= range_max; start += align) { 1188 *new = request_mem_region_exclusive(start, size, dev_n); 1189 if (*new) { 1190 retval = 0; 1191 goto exit; 1192 } 1193 } 1194 } 1195 1196 for (iter = hyperv_mmio; iter; iter = iter->sibling) { 1197 if ((iter->start >= max) || (iter->end <= min)) 1198 continue; 1199 1200 range_min = iter->start; 1201 range_max = iter->end; 1202 start = (range_min + align - 1) & ~(align - 1); 1203 for (; start + size - 1 <= range_max; start += align) { 1204 shadow = __request_region(iter, start, size, NULL, 1205 IORESOURCE_BUSY); 1206 if (!shadow) 1207 continue; 1208 1209 *new = request_mem_region_exclusive(start, size, dev_n); 1210 if (*new) { 1211 shadow->name = (char *)*new; 1212 retval = 0; 1213 goto exit; 1214 } 1215 1216 __release_region(iter, start, size); 1217 } 1218 } 1219 1220 exit: 1221 up(&hyperv_mmio_lock); 1222 return retval; 1223 } 1224 EXPORT_SYMBOL_GPL(vmbus_allocate_mmio); 1225 1226 /** 1227 * vmbus_free_mmio() - Free a memory-mapped I/O range. 1228 * @start: Base address of region to release. 1229 * @size: Size of the range to be allocated 1230 * 1231 * This function releases anything requested by 1232 * vmbus_mmio_allocate(). 1233 */ 1234 void vmbus_free_mmio(resource_size_t start, resource_size_t size) 1235 { 1236 struct resource *iter; 1237 1238 down(&hyperv_mmio_lock); 1239 for (iter = hyperv_mmio; iter; iter = iter->sibling) { 1240 if ((iter->start >= start + size) || (iter->end <= start)) 1241 continue; 1242 1243 __release_region(iter, start, size); 1244 } 1245 release_mem_region(start, size); 1246 up(&hyperv_mmio_lock); 1247 1248 } 1249 EXPORT_SYMBOL_GPL(vmbus_free_mmio); 1250 1251 /** 1252 * vmbus_cpu_number_to_vp_number() - Map CPU to VP. 1253 * @cpu_number: CPU number in Linux terms 1254 * 1255 * This function returns the mapping between the Linux processor 1256 * number and the hypervisor's virtual processor number, useful 1257 * in making hypercalls and such that talk about specific 1258 * processors. 1259 * 1260 * Return: Virtual processor number in Hyper-V terms 1261 */ 1262 int vmbus_cpu_number_to_vp_number(int cpu_number) 1263 { 1264 return hv_context.vp_index[cpu_number]; 1265 } 1266 EXPORT_SYMBOL_GPL(vmbus_cpu_number_to_vp_number); 1267 1268 static int vmbus_acpi_add(struct acpi_device *device) 1269 { 1270 acpi_status result; 1271 int ret_val = -ENODEV; 1272 struct acpi_device *ancestor; 1273 1274 hv_acpi_dev = device; 1275 1276 result = acpi_walk_resources(device->handle, METHOD_NAME__CRS, 1277 vmbus_walk_resources, NULL); 1278 1279 if (ACPI_FAILURE(result)) 1280 goto acpi_walk_err; 1281 /* 1282 * Some ancestor of the vmbus acpi device (Gen1 or Gen2 1283 * firmware) is the VMOD that has the mmio ranges. Get that. 1284 */ 1285 for (ancestor = device->parent; ancestor; ancestor = ancestor->parent) { 1286 result = acpi_walk_resources(ancestor->handle, METHOD_NAME__CRS, 1287 vmbus_walk_resources, NULL); 1288 1289 if (ACPI_FAILURE(result)) 1290 continue; 1291 if (hyperv_mmio) { 1292 vmbus_reserve_fb(); 1293 break; 1294 } 1295 } 1296 ret_val = 0; 1297 1298 acpi_walk_err: 1299 complete(&probe_event); 1300 if (ret_val) 1301 vmbus_acpi_remove(device); 1302 return ret_val; 1303 } 1304 1305 static const struct acpi_device_id vmbus_acpi_device_ids[] = { 1306 {"VMBUS", 0}, 1307 {"VMBus", 0}, 1308 {"", 0}, 1309 }; 1310 MODULE_DEVICE_TABLE(acpi, vmbus_acpi_device_ids); 1311 1312 static struct acpi_driver vmbus_acpi_driver = { 1313 .name = "vmbus", 1314 .ids = vmbus_acpi_device_ids, 1315 .ops = { 1316 .add = vmbus_acpi_add, 1317 .remove = vmbus_acpi_remove, 1318 }, 1319 }; 1320 1321 static void hv_kexec_handler(void) 1322 { 1323 int cpu; 1324 1325 hv_synic_clockevents_cleanup(); 1326 vmbus_initiate_unload(false); 1327 for_each_online_cpu(cpu) 1328 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1); 1329 hv_cleanup(); 1330 }; 1331 1332 static void hv_crash_handler(struct pt_regs *regs) 1333 { 1334 vmbus_initiate_unload(true); 1335 /* 1336 * In crash handler we can't schedule synic cleanup for all CPUs, 1337 * doing the cleanup for current CPU only. This should be sufficient 1338 * for kdump. 1339 */ 1340 hv_synic_cleanup(NULL); 1341 hv_cleanup(); 1342 }; 1343 1344 static int __init hv_acpi_init(void) 1345 { 1346 int ret, t; 1347 1348 if (x86_hyper != &x86_hyper_ms_hyperv) 1349 return -ENODEV; 1350 1351 init_completion(&probe_event); 1352 1353 /* 1354 * Get ACPI resources first. 1355 */ 1356 ret = acpi_bus_register_driver(&vmbus_acpi_driver); 1357 1358 if (ret) 1359 return ret; 1360 1361 t = wait_for_completion_timeout(&probe_event, 5*HZ); 1362 if (t == 0) { 1363 ret = -ETIMEDOUT; 1364 goto cleanup; 1365 } 1366 1367 ret = vmbus_bus_init(); 1368 if (ret) 1369 goto cleanup; 1370 1371 hv_setup_kexec_handler(hv_kexec_handler); 1372 hv_setup_crash_handler(hv_crash_handler); 1373 1374 return 0; 1375 1376 cleanup: 1377 acpi_bus_unregister_driver(&vmbus_acpi_driver); 1378 hv_acpi_dev = NULL; 1379 return ret; 1380 } 1381 1382 static void __exit vmbus_exit(void) 1383 { 1384 int cpu; 1385 1386 hv_remove_kexec_handler(); 1387 hv_remove_crash_handler(); 1388 vmbus_connection.conn_state = DISCONNECTED; 1389 hv_synic_clockevents_cleanup(); 1390 vmbus_disconnect(); 1391 hv_remove_vmbus_irq(); 1392 for_each_online_cpu(cpu) 1393 tasklet_kill(hv_context.msg_dpc[cpu]); 1394 vmbus_free_channels(); 1395 if (ms_hyperv.misc_features & HV_FEATURE_GUEST_CRASH_MSR_AVAILABLE) { 1396 unregister_die_notifier(&hyperv_die_block); 1397 atomic_notifier_chain_unregister(&panic_notifier_list, 1398 &hyperv_panic_block); 1399 } 1400 bus_unregister(&hv_bus); 1401 hv_cleanup(); 1402 for_each_online_cpu(cpu) { 1403 tasklet_kill(hv_context.event_dpc[cpu]); 1404 smp_call_function_single(cpu, hv_synic_cleanup, NULL, 1); 1405 } 1406 hv_synic_free(); 1407 acpi_bus_unregister_driver(&vmbus_acpi_driver); 1408 if (vmbus_proto_version > VERSION_WIN7) 1409 cpu_hotplug_enable(); 1410 } 1411 1412 1413 MODULE_LICENSE("GPL"); 1414 1415 subsys_initcall(hv_acpi_init); 1416 module_exit(vmbus_exit); 1417