err.no Git - linux-2.6/blob - drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86.h"
  20 #include "irq.h"
  21 #include "iodev.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mman.h>
  45
  46 #include <asm/processor.h>
  47 #include <asm/io.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/desc.h>
  50 #include <asm/pgtable.h>
  51
  52 MODULE_AUTHOR("Qumranet");
  53 MODULE_LICENSE("GPL");
  54
  55 DEFINE_SPINLOCK(kvm_lock);
  56 LIST_HEAD(vm_list);
  57
  58 static cpumask_t cpus_hardware_enabled;
  59
  60 struct kmem_cache *kvm_vcpu_cache;
  61 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  62
  63 static __read_mostly struct preempt_ops kvm_preempt_ops;
  64
  65 static struct dentry *debugfs_dir;
  66
  67 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  68                            unsigned long arg);
  69
  70 static inline int valid_vcpu(int n)
  71 {
  72         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  73 }
  74
  75 /*
  76  * Switches to specified vcpu, until a matching vcpu_put()
  77  */
  78 void vcpu_load(struct kvm_vcpu *vcpu)
  79 {
  80         int cpu;
  81
  82         mutex_lock(&vcpu->mutex);
  83         cpu = get_cpu();
  84         preempt_notifier_register(&vcpu->preempt_notifier);
  85         kvm_arch_vcpu_load(vcpu, cpu);
  86         put_cpu();
  87 }
  88
  89 void vcpu_put(struct kvm_vcpu *vcpu)
  90 {
  91         preempt_disable();
  92         kvm_arch_vcpu_put(vcpu);
  93         preempt_notifier_unregister(&vcpu->preempt_notifier);
  94         preempt_enable();
  95         mutex_unlock(&vcpu->mutex);
  96 }
  97
  98 static void ack_flush(void *_completed)
  99 {
 100 }
 101
 102 void kvm_flush_remote_tlbs(struct kvm *kvm)
 103 {
 104         int i, cpu;
 105         cpumask_t cpus;
 106         struct kvm_vcpu *vcpu;
 107
 108         cpus_clear(cpus);
 109         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 110                 vcpu = kvm->vcpus[i];
 111                 if (!vcpu)
 112                         continue;
 113                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 114                         continue;
 115                 cpu = vcpu->cpu;
 116                 if (cpu != -1 && cpu != raw_smp_processor_id())
 117                         cpu_set(cpu, cpus);
 118         }
 119         if (cpus_empty(cpus))
 120                 return;
 121         ++kvm->stat.remote_tlb_flush;
 122         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 123 }
 124
 125 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 126 {
 127         struct page *page;
 128         int r;
 129
 130         mutex_init(&vcpu->mutex);
 131         vcpu->cpu = -1;
 132         vcpu->kvm = kvm;
 133         vcpu->vcpu_id = id;
 134         init_waitqueue_head(&vcpu->wq);
 135
 136         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 137         if (!page) {
 138                 r = -ENOMEM;
 139                 goto fail;
 140         }
 141         vcpu->run = page_address(page);
 142
 143         r = kvm_arch_vcpu_init(vcpu);
 144         if (r < 0)
 145                 goto fail_free_run;
 146         return 0;
 147
 148 fail_free_run:
 149         free_page((unsigned long)vcpu->run);
 150 fail:
 151         return r;
 152 }
 153 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 154
 155 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 156 {
 157         kvm_arch_vcpu_uninit(vcpu);
 158         free_page((unsigned long)vcpu->run);
 159 }
 160 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 161
 162 static struct kvm *kvm_create_vm(void)
 163 {
 164         struct kvm *kvm = kvm_arch_create_vm();
 165
 166         if (IS_ERR(kvm))
 167                 goto out;
 168
 169         kvm->mm = current->mm;
 170         atomic_inc(&kvm->mm->mm_count);
 171         kvm_io_bus_init(&kvm->pio_bus);
 172         mutex_init(&kvm->lock);
 173         kvm_io_bus_init(&kvm->mmio_bus);
 174         spin_lock(&kvm_lock);
 175         list_add(&kvm->vm_list, &vm_list);
 176         spin_unlock(&kvm_lock);
 177 out:
 178         return kvm;
 179 }
 180
 181 /*
 182  * Free any memory in @free but not in @dont.
 183  */
 184 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 185                                   struct kvm_memory_slot *dont)
 186 {
 187         if (!dont || free->rmap != dont->rmap)
 188                 vfree(free->rmap);
 189
 190         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 191                 vfree(free->dirty_bitmap);
 192
 193         free->npages = 0;
 194         free->dirty_bitmap = NULL;
 195         free->rmap = NULL;
 196 }
 197
 198 void kvm_free_physmem(struct kvm *kvm)
 199 {
 200         int i;
 201
 202         for (i = 0; i < kvm->nmemslots; ++i)
 203                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 204 }
 205
 206 static void kvm_destroy_vm(struct kvm *kvm)
 207 {
 208         struct mm_struct *mm = kvm->mm;
 209
 210         spin_lock(&kvm_lock);
 211         list_del(&kvm->vm_list);
 212         spin_unlock(&kvm_lock);
 213         kvm_io_bus_destroy(&kvm->pio_bus);
 214         kvm_io_bus_destroy(&kvm->mmio_bus);
 215         kvm_arch_destroy_vm(kvm);
 216         mmdrop(mm);
 217 }
 218
 219 static int kvm_vm_release(struct inode *inode, struct file *filp)
 220 {
 221         struct kvm *kvm = filp->private_data;
 222
 223         kvm_destroy_vm(kvm);
 224         return 0;
 225 }
 226
 227 /*
 228  * Allocate some memory and give it an address in the guest physical address
 229  * space.
 230  *
 231  * Discontiguous memory is allowed, mostly for framebuffers.
 232  *
 233  * Must be called holding kvm->lock.
 234  */
 235 int __kvm_set_memory_region(struct kvm *kvm,
 236                             struct kvm_userspace_memory_region *mem,
 237                             int user_alloc)
 238 {
 239         int r;
 240         gfn_t base_gfn;
 241         unsigned long npages;
 242         unsigned long i;
 243         struct kvm_memory_slot *memslot;
 244         struct kvm_memory_slot old, new;
 245
 246         r = -EINVAL;
 247         /* General sanity checks */
 248         if (mem->memory_size & (PAGE_SIZE - 1))
 249                 goto out;
 250         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 251                 goto out;
 252         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 253                 goto out;
 254         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 255                 goto out;
 256
 257         memslot = &kvm->memslots[mem->slot];
 258         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 259         npages = mem->memory_size >> PAGE_SHIFT;
 260
 261         if (!npages)
 262                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 263
 264         new = old = *memslot;
 265
 266         new.base_gfn = base_gfn;
 267         new.npages = npages;
 268         new.flags = mem->flags;
 269
 270         /* Disallow changing a memory slot's size. */
 271         r = -EINVAL;
 272         if (npages && old.npages && npages != old.npages)
 273                 goto out_free;
 274
 275         /* Check for overlaps */
 276         r = -EEXIST;
 277         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 278                 struct kvm_memory_slot *s = &kvm->memslots[i];
 279
 280                 if (s == memslot)
 281                         continue;
 282                 if (!((base_gfn + npages <= s->base_gfn) ||
 283                       (base_gfn >= s->base_gfn + s->npages)))
 284                         goto out_free;
 285         }
 286
 287         /* Free page dirty bitmap if unneeded */
 288         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 289                 new.dirty_bitmap = NULL;
 290
 291         r = -ENOMEM;
 292
 293         /* Allocate if a slot is being created */
 294         if (npages && !new.rmap) {
 295                 new.rmap = vmalloc(npages * sizeof(struct page *));
 296
 297                 if (!new.rmap)
 298                         goto out_free;
 299
 300                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 301
 302                 new.user_alloc = user_alloc;
 303                 new.userspace_addr = mem->userspace_addr;
 304         }
 305
 306         /* Allocate page dirty bitmap if needed */
 307         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 308                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 309
 310                 new.dirty_bitmap = vmalloc(dirty_bytes);
 311                 if (!new.dirty_bitmap)
 312                         goto out_free;
 313                 memset(new.dirty_bitmap, 0, dirty_bytes);
 314         }
 315
 316         if (mem->slot >= kvm->nmemslots)
 317                 kvm->nmemslots = mem->slot + 1;
 318
 319         *memslot = new;
 320
 321         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 322         if (r) {
 323                 *memslot = old;
 324                 goto out_free;
 325         }
 326
 327         kvm_free_physmem_slot(&old, &new);
 328         return 0;
 329
 330 out_free:
 331         kvm_free_physmem_slot(&new, &old);
 332 out:
 333         return r;
 334
 335 }
 336 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 337
 338 int kvm_set_memory_region(struct kvm *kvm,
 339                           struct kvm_userspace_memory_region *mem,
 340                           int user_alloc)
 341 {
 342         int r;
 343
 344         mutex_lock(&kvm->lock);
 345         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 346         mutex_unlock(&kvm->lock);
 347         return r;
 348 }
 349 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 350
 351 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 352                                    struct
 353                                    kvm_userspace_memory_region *mem,
 354                                    int user_alloc)
 355 {
 356         if (mem->slot >= KVM_MEMORY_SLOTS)
 357                 return -EINVAL;
 358         return kvm_set_memory_region(kvm, mem, user_alloc);
 359 }
 360
 361 int kvm_get_dirty_log(struct kvm *kvm,
 362                         struct kvm_dirty_log *log, int *is_dirty)
 363 {
 364         struct kvm_memory_slot *memslot;
 365         int r, i;
 366         int n;
 367         unsigned long any = 0;
 368
 369         r = -EINVAL;
 370         if (log->slot >= KVM_MEMORY_SLOTS)
 371                 goto out;
 372
 373         memslot = &kvm->memslots[log->slot];
 374         r = -ENOENT;
 375         if (!memslot->dirty_bitmap)
 376                 goto out;
 377
 378         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 379
 380         for (i = 0; !any && i < n/sizeof(long); ++i)
 381                 any = memslot->dirty_bitmap[i];
 382
 383         r = -EFAULT;
 384         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 385                 goto out;
 386
 387         if (any)
 388                 *is_dirty = 1;
 389
 390         r = 0;
 391 out:
 392         return r;
 393 }
 394
 395 int is_error_page(struct page *page)
 396 {
 397         return page == bad_page;
 398 }
 399 EXPORT_SYMBOL_GPL(is_error_page);
 400
 401 static inline unsigned long bad_hva(void)
 402 {
 403         return PAGE_OFFSET;
 404 }
 405
 406 int kvm_is_error_hva(unsigned long addr)
 407 {
 408         return addr == bad_hva();
 409 }
 410 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 411
 412 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 413 {
 414         int i;
 415
 416         for (i = 0; i < kvm->nmemslots; ++i) {
 417                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 418
 419                 if (gfn >= memslot->base_gfn
 420                     && gfn < memslot->base_gfn + memslot->npages)
 421                         return memslot;
 422         }
 423         return NULL;
 424 }
 425
 426 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 427 {
 428         gfn = unalias_gfn(kvm, gfn);
 429         return __gfn_to_memslot(kvm, gfn);
 430 }
 431
 432 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 433 {
 434         int i;
 435
 436         gfn = unalias_gfn(kvm, gfn);
 437         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 438                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 439
 440                 if (gfn >= memslot->base_gfn
 441                     && gfn < memslot->base_gfn + memslot->npages)
 442                         return 1;
 443         }
 444         return 0;
 445 }
 446 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 447
 448 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 449 {
 450         struct kvm_memory_slot *slot;
 451
 452         gfn = unalias_gfn(kvm, gfn);
 453         slot = __gfn_to_memslot(kvm, gfn);
 454         if (!slot)
 455                 return bad_hva();
 456         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 457 }
 458
 459 /*
 460  * Requires current->mm->mmap_sem to be held
 461  */
 462 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 463 {
 464         struct page *page[1];
 465         unsigned long addr;
 466         int npages;
 467
 468         might_sleep();
 469
 470         addr = gfn_to_hva(kvm, gfn);
 471         if (kvm_is_error_hva(addr)) {
 472                 get_page(bad_page);
 473                 return bad_page;
 474         }
 475
 476         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 477                                 NULL);
 478
 479         if (npages != 1) {
 480                 get_page(bad_page);
 481                 return bad_page;
 482         }
 483
 484         return page[0];
 485 }
 486
 487 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 488 {
 489         struct page *page;
 490
 491         down_read(&current->mm->mmap_sem);
 492         page = __gfn_to_page(kvm, gfn);
 493         up_read(&current->mm->mmap_sem);
 494
 495         return page;
 496 }
 497
 498 EXPORT_SYMBOL_GPL(gfn_to_page);
 499
 500 void kvm_release_page_clean(struct page *page)
 501 {
 502         put_page(page);
 503 }
 504 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 505
 506 void kvm_release_page_dirty(struct page *page)
 507 {
 508         if (!PageReserved(page))
 509                 SetPageDirty(page);
 510         put_page(page);
 511 }
 512 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 513
 514 static int next_segment(unsigned long len, int offset)
 515 {
 516         if (len > PAGE_SIZE - offset)
 517                 return PAGE_SIZE - offset;
 518         else
 519                 return len;
 520 }
 521
 522 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 523                         int len)
 524 {
 525         int r;
 526         unsigned long addr;
 527
 528         addr = gfn_to_hva(kvm, gfn);
 529         if (kvm_is_error_hva(addr))
 530                 return -EFAULT;
 531         r = copy_from_user(data, (void __user *)addr + offset, len);
 532         if (r)
 533                 return -EFAULT;
 534         return 0;
 535 }
 536 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 537
 538 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 539 {
 540         gfn_t gfn = gpa >> PAGE_SHIFT;
 541         int seg;
 542         int offset = offset_in_page(gpa);
 543         int ret;
 544
 545         while ((seg = next_segment(len, offset)) != 0) {
 546                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 547                 if (ret < 0)
 548                         return ret;
 549                 offset = 0;
 550                 len -= seg;
 551                 data += seg;
 552                 ++gfn;
 553         }
 554         return 0;
 555 }
 556 EXPORT_SYMBOL_GPL(kvm_read_guest);
 557
 558 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 559                          int offset, int len)
 560 {
 561         int r;
 562         unsigned long addr;
 563
 564         addr = gfn_to_hva(kvm, gfn);
 565         if (kvm_is_error_hva(addr))
 566                 return -EFAULT;
 567         r = copy_to_user((void __user *)addr + offset, data, len);
 568         if (r)
 569                 return -EFAULT;
 570         mark_page_dirty(kvm, gfn);
 571         return 0;
 572 }
 573 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 574
 575 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 576                     unsigned long len)
 577 {
 578         gfn_t gfn = gpa >> PAGE_SHIFT;
 579         int seg;
 580         int offset = offset_in_page(gpa);
 581         int ret;
 582
 583         while ((seg = next_segment(len, offset)) != 0) {
 584                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 585                 if (ret < 0)
 586                         return ret;
 587                 offset = 0;
 588                 len -= seg;
 589                 data += seg;
 590                 ++gfn;
 591         }
 592         return 0;
 593 }
 594
 595 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 596 {
 597         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 600
 601 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 602 {
 603         gfn_t gfn = gpa >> PAGE_SHIFT;
 604         int seg;
 605         int offset = offset_in_page(gpa);
 606         int ret;
 607
 608         while ((seg = next_segment(len, offset)) != 0) {
 609                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 610                 if (ret < 0)
 611                         return ret;
 612                 offset = 0;
 613                 len -= seg;
 614                 ++gfn;
 615         }
 616         return 0;
 617 }
 618 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 619
 620 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 621 {
 622         struct kvm_memory_slot *memslot;
 623
 624         gfn = unalias_gfn(kvm, gfn);
 625         memslot = __gfn_to_memslot(kvm, gfn);
 626         if (memslot && memslot->dirty_bitmap) {
 627                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 628
 629                 /* avoid RMW */
 630                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 631                         set_bit(rel_gfn, memslot->dirty_bitmap);
 632         }
 633 }
 634
 635 /*
 636  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 637  */
 638 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 639 {
 640         DECLARE_WAITQUEUE(wait, current);
 641
 642         add_wait_queue(&vcpu->wq, &wait);
 643
 644         /*
 645          * We will block until either an interrupt or a signal wakes us up
 646          */
 647         while (!kvm_cpu_has_interrupt(vcpu)
 648                && !signal_pending(current)
 649                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
 650                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
 651                 set_current_state(TASK_INTERRUPTIBLE);
 652                 vcpu_put(vcpu);
 653                 schedule();
 654                 vcpu_load(vcpu);
 655         }
 656
 657         __set_current_state(TASK_RUNNING);
 658         remove_wait_queue(&vcpu->wq, &wait);
 659 }
 660
 661 void kvm_resched(struct kvm_vcpu *vcpu)
 662 {
 663         if (!need_resched())
 664                 return;
 665         cond_resched();
 666 }
 667 EXPORT_SYMBOL_GPL(kvm_resched);
 668
 669 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
 670                                     unsigned long address,
 671                                     int *type)
 672 {
 673         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 674         unsigned long pgoff;
 675         struct page *page;
 676
 677         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 678         if (pgoff == 0)
 679                 page = virt_to_page(vcpu->run);
 680         else if (pgoff == KVM_PIO_PAGE_OFFSET)
 681                 page = virt_to_page(vcpu->pio_data);
 682         else
 683                 return NOPAGE_SIGBUS;
 684         get_page(page);
 685         if (type != NULL)
 686                 *type = VM_FAULT_MINOR;
 687
 688         return page;
 689 }
 690
 691 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 692         .nopage = kvm_vcpu_nopage,
 693 };
 694
 695 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 696 {
 697         vma->vm_ops = &kvm_vcpu_vm_ops;
 698         return 0;
 699 }
 700
 701 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 702 {
 703         struct kvm_vcpu *vcpu = filp->private_data;
 704
 705         fput(vcpu->kvm->filp);
 706         return 0;
 707 }
 708
 709 static struct file_operations kvm_vcpu_fops = {
 710         .release        = kvm_vcpu_release,
 711         .unlocked_ioctl = kvm_vcpu_ioctl,
 712         .compat_ioctl   = kvm_vcpu_ioctl,
 713         .mmap           = kvm_vcpu_mmap,
 714 };
 715
 716 /*
 717  * Allocates an inode for the vcpu.
 718  */
 719 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 720 {
 721         int fd, r;
 722         struct inode *inode;
 723         struct file *file;
 724
 725         r = anon_inode_getfd(&fd, &inode, &file,
 726                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 727         if (r)
 728                 return r;
 729         atomic_inc(&vcpu->kvm->filp->f_count);
 730         return fd;
 731 }
 732
 733 /*
 734  * Creates some virtual cpus.  Good luck creating more than one.
 735  */
 736 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 737 {
 738         int r;
 739         struct kvm_vcpu *vcpu;
 740
 741         if (!valid_vcpu(n))
 742                 return -EINVAL;
 743
 744         vcpu = kvm_arch_vcpu_create(kvm, n);
 745         if (IS_ERR(vcpu))
 746                 return PTR_ERR(vcpu);
 747
 748         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 749
 750         r = kvm_arch_vcpu_setup(vcpu);
 751         if (r)
 752                 goto vcpu_destroy;
 753
 754         mutex_lock(&kvm->lock);
 755         if (kvm->vcpus[n]) {
 756                 r = -EEXIST;
 757                 mutex_unlock(&kvm->lock);
 758                 goto vcpu_destroy;
 759         }
 760         kvm->vcpus[n] = vcpu;
 761         mutex_unlock(&kvm->lock);
 762
 763         /* Now it's all set up, let userspace reach it */
 764         r = create_vcpu_fd(vcpu);
 765         if (r < 0)
 766                 goto unlink;
 767         return r;
 768
 769 unlink:
 770         mutex_lock(&kvm->lock);
 771         kvm->vcpus[n] = NULL;
 772         mutex_unlock(&kvm->lock);
 773 vcpu_destroy:
 774         kvm_arch_vcpu_destroy(vcpu);
 775         return r;
 776 }
 777
 778 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 779 {
 780         if (sigset) {
 781                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 782                 vcpu->sigset_active = 1;
 783                 vcpu->sigset = *sigset;
 784         } else
 785                 vcpu->sigset_active = 0;
 786         return 0;
 787 }
 788
 789 static long kvm_vcpu_ioctl(struct file *filp,
 790                            unsigned int ioctl, unsigned long arg)
 791 {
 792         struct kvm_vcpu *vcpu = filp->private_data;
 793         void __user *argp = (void __user *)arg;
 794         int r;
 795
 796         if (vcpu->kvm->mm != current->mm)
 797                 return -EIO;
 798         switch (ioctl) {
 799         case KVM_RUN:
 800                 r = -EINVAL;
 801                 if (arg)
 802                         goto out;
 803                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 804                 break;
 805         case KVM_GET_REGS: {
 806                 struct kvm_regs kvm_regs;
 807
 808                 memset(&kvm_regs, 0, sizeof kvm_regs);
 809                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 810                 if (r)
 811                         goto out;
 812                 r = -EFAULT;
 813                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 814                         goto out;
 815                 r = 0;
 816                 break;
 817         }
 818         case KVM_SET_REGS: {
 819                 struct kvm_regs kvm_regs;
 820
 821                 r = -EFAULT;
 822                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 823                         goto out;
 824                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 825                 if (r)
 826                         goto out;
 827                 r = 0;
 828                 break;
 829         }
 830         case KVM_GET_SREGS: {
 831                 struct kvm_sregs kvm_sregs;
 832
 833                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 834                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 835                 if (r)
 836                         goto out;
 837                 r = -EFAULT;
 838                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 839                         goto out;
 840                 r = 0;
 841                 break;
 842         }
 843         case KVM_SET_SREGS: {
 844                 struct kvm_sregs kvm_sregs;
 845
 846                 r = -EFAULT;
 847                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 848                         goto out;
 849                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 850                 if (r)
 851                         goto out;
 852                 r = 0;
 853                 break;
 854         }
 855         case KVM_TRANSLATE: {
 856                 struct kvm_translation tr;
 857
 858                 r = -EFAULT;
 859                 if (copy_from_user(&tr, argp, sizeof tr))
 860                         goto out;
 861                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 862                 if (r)
 863                         goto out;
 864                 r = -EFAULT;
 865                 if (copy_to_user(argp, &tr, sizeof tr))
 866                         goto out;
 867                 r = 0;
 868                 break;
 869         }
 870         case KVM_DEBUG_GUEST: {
 871                 struct kvm_debug_guest dbg;
 872
 873                 r = -EFAULT;
 874                 if (copy_from_user(&dbg, argp, sizeof dbg))
 875                         goto out;
 876                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 877                 if (r)
 878                         goto out;
 879                 r = 0;
 880                 break;
 881         }
 882         case KVM_SET_SIGNAL_MASK: {
 883                 struct kvm_signal_mask __user *sigmask_arg = argp;
 884                 struct kvm_signal_mask kvm_sigmask;
 885                 sigset_t sigset, *p;
 886
 887                 p = NULL;
 888                 if (argp) {
 889                         r = -EFAULT;
 890                         if (copy_from_user(&kvm_sigmask, argp,
 891                                            sizeof kvm_sigmask))
 892                                 goto out;
 893                         r = -EINVAL;
 894                         if (kvm_sigmask.len != sizeof sigset)
 895                                 goto out;
 896                         r = -EFAULT;
 897                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 898                                            sizeof sigset))
 899                                 goto out;
 900                         p = &sigset;
 901                 }
 902                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 903                 break;
 904         }
 905         case KVM_GET_FPU: {
 906                 struct kvm_fpu fpu;
 907
 908                 memset(&fpu, 0, sizeof fpu);
 909                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 910                 if (r)
 911                         goto out;
 912                 r = -EFAULT;
 913                 if (copy_to_user(argp, &fpu, sizeof fpu))
 914                         goto out;
 915                 r = 0;
 916                 break;
 917         }
 918         case KVM_SET_FPU: {
 919                 struct kvm_fpu fpu;
 920
 921                 r = -EFAULT;
 922                 if (copy_from_user(&fpu, argp, sizeof fpu))
 923                         goto out;
 924                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
 925                 if (r)
 926                         goto out;
 927                 r = 0;
 928                 break;
 929         }
 930         default:
 931                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
 932         }
 933 out:
 934         return r;
 935 }
 936
 937 static long kvm_vm_ioctl(struct file *filp,
 938                            unsigned int ioctl, unsigned long arg)
 939 {
 940         struct kvm *kvm = filp->private_data;
 941         void __user *argp = (void __user *)arg;
 942         int r;
 943
 944         if (kvm->mm != current->mm)
 945                 return -EIO;
 946         switch (ioctl) {
 947         case KVM_CREATE_VCPU:
 948                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
 949                 if (r < 0)
 950                         goto out;
 951                 break;
 952         case KVM_SET_USER_MEMORY_REGION: {
 953                 struct kvm_userspace_memory_region kvm_userspace_mem;
 954
 955                 r = -EFAULT;
 956                 if (copy_from_user(&kvm_userspace_mem, argp,
 957                                                 sizeof kvm_userspace_mem))
 958                         goto out;
 959
 960                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
 961                 if (r)
 962                         goto out;
 963                 break;
 964         }
 965         case KVM_GET_DIRTY_LOG: {
 966                 struct kvm_dirty_log log;
 967
 968                 r = -EFAULT;
 969                 if (copy_from_user(&log, argp, sizeof log))
 970                         goto out;
 971                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
 972                 if (r)
 973                         goto out;
 974                 break;
 975         }
 976         default:
 977                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
 978         }
 979 out:
 980         return r;
 981 }
 982
 983 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
 984                                   unsigned long address,
 985                                   int *type)
 986 {
 987         struct kvm *kvm = vma->vm_file->private_data;
 988         unsigned long pgoff;
 989         struct page *page;
 990
 991         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 992         if (!kvm_is_visible_gfn(kvm, pgoff))
 993                 return NOPAGE_SIGBUS;
 994         /* current->mm->mmap_sem is already held so call lockless version */
 995         page = __gfn_to_page(kvm, pgoff);
 996         if (is_error_page(page)) {
 997                 kvm_release_page_clean(page);
 998                 return NOPAGE_SIGBUS;
 999         }
1000         if (type != NULL)
1001                 *type = VM_FAULT_MINOR;
1002
1003         return page;
1004 }
1005
1006 static struct vm_operations_struct kvm_vm_vm_ops = {
1007         .nopage = kvm_vm_nopage,
1008 };
1009
1010 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1011 {
1012         vma->vm_ops = &kvm_vm_vm_ops;
1013         return 0;
1014 }
1015
1016 static struct file_operations kvm_vm_fops = {
1017         .release        = kvm_vm_release,
1018         .unlocked_ioctl = kvm_vm_ioctl,
1019         .compat_ioctl   = kvm_vm_ioctl,
1020         .mmap           = kvm_vm_mmap,
1021 };
1022
1023 static int kvm_dev_ioctl_create_vm(void)
1024 {
1025         int fd, r;
1026         struct inode *inode;
1027         struct file *file;
1028         struct kvm *kvm;
1029
1030         kvm = kvm_create_vm();
1031         if (IS_ERR(kvm))
1032                 return PTR_ERR(kvm);
1033         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1034         if (r) {
1035                 kvm_destroy_vm(kvm);
1036                 return r;
1037         }
1038
1039         kvm->filp = file;
1040
1041         return fd;
1042 }
1043
1044 static long kvm_dev_ioctl(struct file *filp,
1045                           unsigned int ioctl, unsigned long arg)
1046 {
1047         void __user *argp = (void __user *)arg;
1048         long r = -EINVAL;
1049
1050         switch (ioctl) {
1051         case KVM_GET_API_VERSION:
1052                 r = -EINVAL;
1053                 if (arg)
1054                         goto out;
1055                 r = KVM_API_VERSION;
1056                 break;
1057         case KVM_CREATE_VM:
1058                 r = -EINVAL;
1059                 if (arg)
1060                         goto out;
1061                 r = kvm_dev_ioctl_create_vm();
1062                 break;
1063         case KVM_CHECK_EXTENSION:
1064                 r = kvm_dev_ioctl_check_extension((long)argp);
1065                 break;
1066         case KVM_GET_VCPU_MMAP_SIZE:
1067                 r = -EINVAL;
1068                 if (arg)
1069                         goto out;
1070                 r = 2 * PAGE_SIZE;
1071                 break;
1072         default:
1073                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1074         }
1075 out:
1076         return r;
1077 }
1078
1079 static struct file_operations kvm_chardev_ops = {
1080         .unlocked_ioctl = kvm_dev_ioctl,
1081         .compat_ioctl   = kvm_dev_ioctl,
1082 };
1083
1084 static struct miscdevice kvm_dev = {
1085         KVM_MINOR,
1086         "kvm",
1087         &kvm_chardev_ops,
1088 };
1089
1090 static void hardware_enable(void *junk)
1091 {
1092         int cpu = raw_smp_processor_id();
1093
1094         if (cpu_isset(cpu, cpus_hardware_enabled))
1095                 return;
1096         cpu_set(cpu, cpus_hardware_enabled);
1097         kvm_arch_hardware_enable(NULL);
1098 }
1099
1100 static void hardware_disable(void *junk)
1101 {
1102         int cpu = raw_smp_processor_id();
1103
1104         if (!cpu_isset(cpu, cpus_hardware_enabled))
1105                 return;
1106         cpu_clear(cpu, cpus_hardware_enabled);
1107         decache_vcpus_on_cpu(cpu);
1108         kvm_arch_hardware_disable(NULL);
1109 }
1110
1111 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1112                            void *v)
1113 {
1114         int cpu = (long)v;
1115
1116         val &= ~CPU_TASKS_FROZEN;
1117         switch (val) {
1118         case CPU_DYING:
1119                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1120                        cpu);
1121                 hardware_disable(NULL);
1122                 break;
1123         case CPU_UP_CANCELED:
1124                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1125                        cpu);
1126                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1127                 break;
1128         case CPU_ONLINE:
1129                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1130                        cpu);
1131                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1132                 break;
1133         }
1134         return NOTIFY_OK;
1135 }
1136
1137 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1138                       void *v)
1139 {
1140         if (val == SYS_RESTART) {
1141                 /*
1142                  * Some (well, at least mine) BIOSes hang on reboot if
1143                  * in vmx root mode.
1144                  */
1145                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1146                 on_each_cpu(hardware_disable, NULL, 0, 1);
1147         }
1148         return NOTIFY_OK;
1149 }
1150
1151 static struct notifier_block kvm_reboot_notifier = {
1152         .notifier_call = kvm_reboot,
1153         .priority = 0,
1154 };
1155
1156 void kvm_io_bus_init(struct kvm_io_bus *bus)
1157 {
1158         memset(bus, 0, sizeof(*bus));
1159 }
1160
1161 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1162 {
1163         int i;
1164
1165         for (i = 0; i < bus->dev_count; i++) {
1166                 struct kvm_io_device *pos = bus->devs[i];
1167
1168                 kvm_iodevice_destructor(pos);
1169         }
1170 }
1171
1172 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1173 {
1174         int i;
1175
1176         for (i = 0; i < bus->dev_count; i++) {
1177                 struct kvm_io_device *pos = bus->devs[i];
1178
1179                 if (pos->in_range(pos, addr))
1180                         return pos;
1181         }
1182
1183         return NULL;
1184 }
1185
1186 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1187 {
1188         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1189
1190         bus->devs[bus->dev_count++] = dev;
1191 }
1192
1193 static struct notifier_block kvm_cpu_notifier = {
1194         .notifier_call = kvm_cpu_hotplug,
1195         .priority = 20, /* must be > scheduler priority */
1196 };
1197
1198 static u64 vm_stat_get(void *_offset)
1199 {
1200         unsigned offset = (long)_offset;
1201         u64 total = 0;
1202         struct kvm *kvm;
1203
1204         spin_lock(&kvm_lock);
1205         list_for_each_entry(kvm, &vm_list, vm_list)
1206                 total += *(u32 *)((void *)kvm + offset);
1207         spin_unlock(&kvm_lock);
1208         return total;
1209 }
1210
1211 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1212
1213 static u64 vcpu_stat_get(void *_offset)
1214 {
1215         unsigned offset = (long)_offset;
1216         u64 total = 0;
1217         struct kvm *kvm;
1218         struct kvm_vcpu *vcpu;
1219         int i;
1220
1221         spin_lock(&kvm_lock);
1222         list_for_each_entry(kvm, &vm_list, vm_list)
1223                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1224                         vcpu = kvm->vcpus[i];
1225                         if (vcpu)
1226                                 total += *(u32 *)((void *)vcpu + offset);
1227                 }
1228         spin_unlock(&kvm_lock);
1229         return total;
1230 }
1231
1232 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1233
1234 static struct file_operations *stat_fops[] = {
1235         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1236         [KVM_STAT_VM]   = &vm_stat_fops,
1237 };
1238
1239 static void kvm_init_debug(void)
1240 {
1241         struct kvm_stats_debugfs_item *p;
1242
1243         debugfs_dir = debugfs_create_dir("kvm", NULL);
1244         for (p = debugfs_entries; p->name; ++p)
1245                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1246                                                 (void *)(long)p->offset,
1247                                                 stat_fops[p->kind]);
1248 }
1249
1250 static void kvm_exit_debug(void)
1251 {
1252         struct kvm_stats_debugfs_item *p;
1253
1254         for (p = debugfs_entries; p->name; ++p)
1255                 debugfs_remove(p->dentry);
1256         debugfs_remove(debugfs_dir);
1257 }
1258
1259 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1260 {
1261         hardware_disable(NULL);
1262         return 0;
1263 }
1264
1265 static int kvm_resume(struct sys_device *dev)
1266 {
1267         hardware_enable(NULL);
1268         return 0;
1269 }
1270
1271 static struct sysdev_class kvm_sysdev_class = {
1272         .name = "kvm",
1273         .suspend = kvm_suspend,
1274         .resume = kvm_resume,
1275 };
1276
1277 static struct sys_device kvm_sysdev = {
1278         .id = 0,
1279         .cls = &kvm_sysdev_class,
1280 };
1281
1282 struct page *bad_page;
1283
1284 static inline
1285 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1286 {
1287         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1288 }
1289
1290 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1291 {
1292         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1293
1294         kvm_arch_vcpu_load(vcpu, cpu);
1295 }
1296
1297 static void kvm_sched_out(struct preempt_notifier *pn,
1298                           struct task_struct *next)
1299 {
1300         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1301
1302         kvm_arch_vcpu_put(vcpu);
1303 }
1304
1305 int kvm_init(void *opaque, unsigned int vcpu_size,
1306                   struct module *module)
1307 {
1308         int r;
1309         int cpu;
1310
1311         kvm_init_debug();
1312
1313         r = kvm_arch_init(opaque);
1314         if (r)
1315                 goto out_fail;
1316
1317         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1318
1319         if (bad_page == NULL) {
1320                 r = -ENOMEM;
1321                 goto out;
1322         }
1323
1324         r = kvm_arch_hardware_setup();
1325         if (r < 0)
1326                 goto out_free_0;
1327
1328         for_each_online_cpu(cpu) {
1329                 smp_call_function_single(cpu,
1330                                 kvm_arch_check_processor_compat,
1331                                 &r, 0, 1);
1332                 if (r < 0)
1333                         goto out_free_1;
1334         }
1335
1336         on_each_cpu(hardware_enable, NULL, 0, 1);
1337         r = register_cpu_notifier(&kvm_cpu_notifier);
1338         if (r)
1339                 goto out_free_2;
1340         register_reboot_notifier(&kvm_reboot_notifier);
1341
1342         r = sysdev_class_register(&kvm_sysdev_class);
1343         if (r)
1344                 goto out_free_3;
1345
1346         r = sysdev_register(&kvm_sysdev);
1347         if (r)
1348                 goto out_free_4;
1349
1350         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1351         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1352                                            __alignof__(struct kvm_vcpu),
1353                                            0, NULL);
1354         if (!kvm_vcpu_cache) {
1355                 r = -ENOMEM;
1356                 goto out_free_5;
1357         }
1358
1359         kvm_chardev_ops.owner = module;
1360
1361         r = misc_register(&kvm_dev);
1362         if (r) {
1363                 printk(KERN_ERR "kvm: misc device register failed\n");
1364                 goto out_free;
1365         }
1366
1367         kvm_preempt_ops.sched_in = kvm_sched_in;
1368         kvm_preempt_ops.sched_out = kvm_sched_out;
1369
1370         return 0;
1371
1372 out_free:
1373         kmem_cache_destroy(kvm_vcpu_cache);
1374 out_free_5:
1375         sysdev_unregister(&kvm_sysdev);
1376 out_free_4:
1377         sysdev_class_unregister(&kvm_sysdev_class);
1378 out_free_3:
1379         unregister_reboot_notifier(&kvm_reboot_notifier);
1380         unregister_cpu_notifier(&kvm_cpu_notifier);
1381 out_free_2:
1382         on_each_cpu(hardware_disable, NULL, 0, 1);
1383 out_free_1:
1384         kvm_arch_hardware_unsetup();
1385 out_free_0:
1386         __free_page(bad_page);
1387 out:
1388         kvm_arch_exit();
1389         kvm_exit_debug();
1390 out_fail:
1391         return r;
1392 }
1393 EXPORT_SYMBOL_GPL(kvm_init);
1394
1395 void kvm_exit(void)
1396 {
1397         misc_deregister(&kvm_dev);
1398         kmem_cache_destroy(kvm_vcpu_cache);
1399         sysdev_unregister(&kvm_sysdev);
1400         sysdev_class_unregister(&kvm_sysdev_class);
1401         unregister_reboot_notifier(&kvm_reboot_notifier);
1402         unregister_cpu_notifier(&kvm_cpu_notifier);
1403         on_each_cpu(hardware_disable, NULL, 0, 1);
1404         kvm_arch_hardware_unsetup();
1405         kvm_arch_exit();
1406         kvm_exit_debug();
1407         __free_page(bad_page);
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_exit);