virt/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 "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43
  44 #include <asm/processor.h>
  45 #include <asm/io.h>
  46 #include <asm/uaccess.h>
  47 #include <asm/pgtable.h>
  48
  49 MODULE_AUTHOR("Qumranet");
  50 MODULE_LICENSE("GPL");
  51
  52 DEFINE_SPINLOCK(kvm_lock);
  53 LIST_HEAD(vm_list);
  54
  55 static cpumask_t cpus_hardware_enabled;
  56
  57 struct kmem_cache *kvm_vcpu_cache;
  58 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  59
  60 static __read_mostly struct preempt_ops kvm_preempt_ops;
  61
  62 static struct dentry *debugfs_dir;
  63
  64 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  65                            unsigned long arg);
  66
  67 static inline int valid_vcpu(int n)
  68 {
  69         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  70 }
  71
  72 /*
  73  * Switches to specified vcpu, until a matching vcpu_put()
  74  */
  75 void vcpu_load(struct kvm_vcpu *vcpu)
  76 {
  77         int cpu;
  78
  79         mutex_lock(&vcpu->mutex);
  80         cpu = get_cpu();
  81         preempt_notifier_register(&vcpu->preempt_notifier);
  82         kvm_arch_vcpu_load(vcpu, cpu);
  83         put_cpu();
  84 }
  85
  86 void vcpu_put(struct kvm_vcpu *vcpu)
  87 {
  88         preempt_disable();
  89         kvm_arch_vcpu_put(vcpu);
  90         preempt_notifier_unregister(&vcpu->preempt_notifier);
  91         preempt_enable();
  92         mutex_unlock(&vcpu->mutex);
  93 }
  94
  95 static void ack_flush(void *_completed)
  96 {
  97 }
  98
  99 void kvm_flush_remote_tlbs(struct kvm *kvm)
 100 {
 101         int i, cpu;
 102         cpumask_t cpus;
 103         struct kvm_vcpu *vcpu;
 104
 105         cpus_clear(cpus);
 106         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 107                 vcpu = kvm->vcpus[i];
 108                 if (!vcpu)
 109                         continue;
 110                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 111                         continue;
 112                 cpu = vcpu->cpu;
 113                 if (cpu != -1 && cpu != raw_smp_processor_id())
 114                         cpu_set(cpu, cpus);
 115         }
 116         if (cpus_empty(cpus))
 117                 return;
 118         ++kvm->stat.remote_tlb_flush;
 119         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 120 }
 121
 122 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 123 {
 124         struct page *page;
 125         int r;
 126
 127         mutex_init(&vcpu->mutex);
 128         vcpu->cpu = -1;
 129         vcpu->kvm = kvm;
 130         vcpu->vcpu_id = id;
 131         init_waitqueue_head(&vcpu->wq);
 132
 133         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 134         if (!page) {
 135                 r = -ENOMEM;
 136                 goto fail;
 137         }
 138         vcpu->run = page_address(page);
 139
 140         r = kvm_arch_vcpu_init(vcpu);
 141         if (r < 0)
 142                 goto fail_free_run;
 143         return 0;
 144
 145 fail_free_run:
 146         free_page((unsigned long)vcpu->run);
 147 fail:
 148         return r;
 149 }
 150 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 151
 152 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 153 {
 154         kvm_arch_vcpu_uninit(vcpu);
 155         free_page((unsigned long)vcpu->run);
 156 }
 157 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 158
 159 static struct kvm *kvm_create_vm(void)
 160 {
 161         struct kvm *kvm = kvm_arch_create_vm();
 162
 163         if (IS_ERR(kvm))
 164                 goto out;
 165
 166         kvm->mm = current->mm;
 167         atomic_inc(&kvm->mm->mm_count);
 168         spin_lock_init(&kvm->mmu_lock);
 169         kvm_io_bus_init(&kvm->pio_bus);
 170         mutex_init(&kvm->lock);
 171         kvm_io_bus_init(&kvm->mmio_bus);
 172         init_rwsem(&kvm->slots_lock);
 173         spin_lock(&kvm_lock);
 174         list_add(&kvm->vm_list, &vm_list);
 175         spin_unlock(&kvm_lock);
 176 out:
 177         return kvm;
 178 }
 179
 180 /*
 181  * Free any memory in @free but not in @dont.
 182  */
 183 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 184                                   struct kvm_memory_slot *dont)
 185 {
 186         if (!dont || free->rmap != dont->rmap)
 187                 vfree(free->rmap);
 188
 189         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 190                 vfree(free->dirty_bitmap);
 191
 192         free->npages = 0;
 193         free->dirty_bitmap = NULL;
 194         free->rmap = NULL;
 195 }
 196
 197 void kvm_free_physmem(struct kvm *kvm)
 198 {
 199         int i;
 200
 201         for (i = 0; i < kvm->nmemslots; ++i)
 202                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 203 }
 204
 205 static void kvm_destroy_vm(struct kvm *kvm)
 206 {
 207         struct mm_struct *mm = kvm->mm;
 208
 209         spin_lock(&kvm_lock);
 210         list_del(&kvm->vm_list);
 211         spin_unlock(&kvm_lock);
 212         kvm_io_bus_destroy(&kvm->pio_bus);
 213         kvm_io_bus_destroy(&kvm->mmio_bus);
 214         kvm_arch_destroy_vm(kvm);
 215         mmdrop(mm);
 216 }
 217
 218 static int kvm_vm_release(struct inode *inode, struct file *filp)
 219 {
 220         struct kvm *kvm = filp->private_data;
 221
 222         kvm_destroy_vm(kvm);
 223         return 0;
 224 }
 225
 226 /*
 227  * Allocate some memory and give it an address in the guest physical address
 228  * space.
 229  *
 230  * Discontiguous memory is allowed, mostly for framebuffers.
 231  *
 232  * Must be called holding mmap_sem for write.
 233  */
 234 int __kvm_set_memory_region(struct kvm *kvm,
 235                             struct kvm_userspace_memory_region *mem,
 236                             int user_alloc)
 237 {
 238         int r;
 239         gfn_t base_gfn;
 240         unsigned long npages;
 241         unsigned long i;
 242         struct kvm_memory_slot *memslot;
 243         struct kvm_memory_slot old, new;
 244
 245         r = -EINVAL;
 246         /* General sanity checks */
 247         if (mem->memory_size & (PAGE_SIZE - 1))
 248                 goto out;
 249         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 250                 goto out;
 251         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 252                 goto out;
 253         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 254                 goto out;
 255
 256         memslot = &kvm->memslots[mem->slot];
 257         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 258         npages = mem->memory_size >> PAGE_SHIFT;
 259
 260         if (!npages)
 261                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 262
 263         new = old = *memslot;
 264
 265         new.base_gfn = base_gfn;
 266         new.npages = npages;
 267         new.flags = mem->flags;
 268
 269         /* Disallow changing a memory slot's size. */
 270         r = -EINVAL;
 271         if (npages && old.npages && npages != old.npages)
 272                 goto out_free;
 273
 274         /* Check for overlaps */
 275         r = -EEXIST;
 276         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 277                 struct kvm_memory_slot *s = &kvm->memslots[i];
 278
 279                 if (s == memslot)
 280                         continue;
 281                 if (!((base_gfn + npages <= s->base_gfn) ||
 282                       (base_gfn >= s->base_gfn + s->npages)))
 283                         goto out_free;
 284         }
 285
 286         /* Free page dirty bitmap if unneeded */
 287         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 288                 new.dirty_bitmap = NULL;
 289
 290         r = -ENOMEM;
 291
 292         /* Allocate if a slot is being created */
 293         if (npages && !new.rmap) {
 294                 new.rmap = vmalloc(npages * sizeof(struct page *));
 295
 296                 if (!new.rmap)
 297                         goto out_free;
 298
 299                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 300
 301                 new.user_alloc = user_alloc;
 302                 new.userspace_addr = mem->userspace_addr;
 303         }
 304
 305         /* Allocate page dirty bitmap if needed */
 306         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 307                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 308
 309                 new.dirty_bitmap = vmalloc(dirty_bytes);
 310                 if (!new.dirty_bitmap)
 311                         goto out_free;
 312                 memset(new.dirty_bitmap, 0, dirty_bytes);
 313         }
 314
 315         if (mem->slot >= kvm->nmemslots)
 316                 kvm->nmemslots = mem->slot + 1;
 317
 318         *memslot = new;
 319
 320         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 321         if (r) {
 322                 *memslot = old;
 323                 goto out_free;
 324         }
 325
 326         kvm_free_physmem_slot(&old, &new);
 327         return 0;
 328
 329 out_free:
 330         kvm_free_physmem_slot(&new, &old);
 331 out:
 332         return r;
 333
 334 }
 335 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 336
 337 int kvm_set_memory_region(struct kvm *kvm,
 338                           struct kvm_userspace_memory_region *mem,
 339                           int user_alloc)
 340 {
 341         int r;
 342
 343         down_write(&kvm->slots_lock);
 344         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 345         up_write(&kvm->slots_lock);
 346         return r;
 347 }
 348 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 349
 350 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 351                                    struct
 352                                    kvm_userspace_memory_region *mem,
 353                                    int user_alloc)
 354 {
 355         if (mem->slot >= KVM_MEMORY_SLOTS)
 356                 return -EINVAL;
 357         return kvm_set_memory_region(kvm, mem, user_alloc);
 358 }
 359
 360 int kvm_get_dirty_log(struct kvm *kvm,
 361                         struct kvm_dirty_log *log, int *is_dirty)
 362 {
 363         struct kvm_memory_slot *memslot;
 364         int r, i;
 365         int n;
 366         unsigned long any = 0;
 367
 368         r = -EINVAL;
 369         if (log->slot >= KVM_MEMORY_SLOTS)
 370                 goto out;
 371
 372         memslot = &kvm->memslots[log->slot];
 373         r = -ENOENT;
 374         if (!memslot->dirty_bitmap)
 375                 goto out;
 376
 377         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 378
 379         for (i = 0; !any && i < n/sizeof(long); ++i)
 380                 any = memslot->dirty_bitmap[i];
 381
 382         r = -EFAULT;
 383         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 384                 goto out;
 385
 386         if (any)
 387                 *is_dirty = 1;
 388
 389         r = 0;
 390 out:
 391         return r;
 392 }
 393
 394 int is_error_page(struct page *page)
 395 {
 396         return page == bad_page;
 397 }
 398 EXPORT_SYMBOL_GPL(is_error_page);
 399
 400 static inline unsigned long bad_hva(void)
 401 {
 402         return PAGE_OFFSET;
 403 }
 404
 405 int kvm_is_error_hva(unsigned long addr)
 406 {
 407         return addr == bad_hva();
 408 }
 409 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 410
 411 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 412 {
 413         int i;
 414
 415         for (i = 0; i < kvm->nmemslots; ++i) {
 416                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 417
 418                 if (gfn >= memslot->base_gfn
 419                     && gfn < memslot->base_gfn + memslot->npages)
 420                         return memslot;
 421         }
 422         return NULL;
 423 }
 424
 425 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 426 {
 427         gfn = unalias_gfn(kvm, gfn);
 428         return __gfn_to_memslot(kvm, gfn);
 429 }
 430
 431 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 432 {
 433         int i;
 434
 435         gfn = unalias_gfn(kvm, gfn);
 436         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 437                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 438
 439                 if (gfn >= memslot->base_gfn
 440                     && gfn < memslot->base_gfn + memslot->npages)
 441                         return 1;
 442         }
 443         return 0;
 444 }
 445 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 446
 447 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 448 {
 449         struct kvm_memory_slot *slot;
 450
 451         gfn = unalias_gfn(kvm, gfn);
 452         slot = __gfn_to_memslot(kvm, gfn);
 453         if (!slot)
 454                 return bad_hva();
 455         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 456 }
 457
 458 /*
 459  * Requires current->mm->mmap_sem to be held
 460  */
 461 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 462 {
 463         struct page *page[1];
 464         unsigned long addr;
 465         int npages;
 466
 467         might_sleep();
 468
 469         addr = gfn_to_hva(kvm, gfn);
 470         if (kvm_is_error_hva(addr)) {
 471                 get_page(bad_page);
 472                 return bad_page;
 473         }
 474
 475         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 476                                 NULL);
 477
 478         if (npages != 1) {
 479                 get_page(bad_page);
 480                 return bad_page;
 481         }
 482
 483         return page[0];
 484 }
 485
 486 EXPORT_SYMBOL_GPL(gfn_to_page);
 487
 488 void kvm_release_page_clean(struct page *page)
 489 {
 490         put_page(page);
 491 }
 492 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 493
 494 void kvm_release_page_dirty(struct page *page)
 495 {
 496         if (!PageReserved(page))
 497                 SetPageDirty(page);
 498         put_page(page);
 499 }
 500 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 501
 502 static int next_segment(unsigned long len, int offset)
 503 {
 504         if (len > PAGE_SIZE - offset)
 505                 return PAGE_SIZE - offset;
 506         else
 507                 return len;
 508 }
 509
 510 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 511                         int len)
 512 {
 513         int r;
 514         unsigned long addr;
 515
 516         addr = gfn_to_hva(kvm, gfn);
 517         if (kvm_is_error_hva(addr))
 518                 return -EFAULT;
 519         r = copy_from_user(data, (void __user *)addr + offset, len);
 520         if (r)
 521                 return -EFAULT;
 522         return 0;
 523 }
 524 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 525
 526 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 527 {
 528         gfn_t gfn = gpa >> PAGE_SHIFT;
 529         int seg;
 530         int offset = offset_in_page(gpa);
 531         int ret;
 532
 533         while ((seg = next_segment(len, offset)) != 0) {
 534                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 535                 if (ret < 0)
 536                         return ret;
 537                 offset = 0;
 538                 len -= seg;
 539                 data += seg;
 540                 ++gfn;
 541         }
 542         return 0;
 543 }
 544 EXPORT_SYMBOL_GPL(kvm_read_guest);
 545
 546 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 547                           unsigned long len)
 548 {
 549         int r;
 550         unsigned long addr;
 551         gfn_t gfn = gpa >> PAGE_SHIFT;
 552         int offset = offset_in_page(gpa);
 553
 554         addr = gfn_to_hva(kvm, gfn);
 555         if (kvm_is_error_hva(addr))
 556                 return -EFAULT;
 557         pagefault_disable();
 558         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
 559         pagefault_enable();
 560         if (r)
 561                 return -EFAULT;
 562         return 0;
 563 }
 564 EXPORT_SYMBOL(kvm_read_guest_atomic);
 565
 566 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 567                          int offset, int len)
 568 {
 569         int r;
 570         unsigned long addr;
 571
 572         addr = gfn_to_hva(kvm, gfn);
 573         if (kvm_is_error_hva(addr))
 574                 return -EFAULT;
 575         r = copy_to_user((void __user *)addr + offset, data, len);
 576         if (r)
 577                 return -EFAULT;
 578         mark_page_dirty(kvm, gfn);
 579         return 0;
 580 }
 581 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 582
 583 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 584                     unsigned long len)
 585 {
 586         gfn_t gfn = gpa >> PAGE_SHIFT;
 587         int seg;
 588         int offset = offset_in_page(gpa);
 589         int ret;
 590
 591         while ((seg = next_segment(len, offset)) != 0) {
 592                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 593                 if (ret < 0)
 594                         return ret;
 595                 offset = 0;
 596                 len -= seg;
 597                 data += seg;
 598                 ++gfn;
 599         }
 600         return 0;
 601 }
 602
 603 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 604 {
 605         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 606 }
 607 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 608
 609 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 610 {
 611         gfn_t gfn = gpa >> PAGE_SHIFT;
 612         int seg;
 613         int offset = offset_in_page(gpa);
 614         int ret;
 615
 616         while ((seg = next_segment(len, offset)) != 0) {
 617                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 618                 if (ret < 0)
 619                         return ret;
 620                 offset = 0;
 621                 len -= seg;
 622                 ++gfn;
 623         }
 624         return 0;
 625 }
 626 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 627
 628 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 629 {
 630         struct kvm_memory_slot *memslot;
 631
 632         gfn = unalias_gfn(kvm, gfn);
 633         memslot = __gfn_to_memslot(kvm, gfn);
 634         if (memslot && memslot->dirty_bitmap) {
 635                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 636
 637                 /* avoid RMW */
 638                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 639                         set_bit(rel_gfn, memslot->dirty_bitmap);
 640         }
 641 }
 642
 643 /*
 644  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 645  */
 646 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 647 {
 648         DECLARE_WAITQUEUE(wait, current);
 649
 650         add_wait_queue(&vcpu->wq, &wait);
 651
 652         /*
 653          * We will block until either an interrupt or a signal wakes us up
 654          */
 655         while (!kvm_cpu_has_interrupt(vcpu)
 656                && !signal_pending(current)
 657                && !kvm_arch_vcpu_runnable(vcpu)) {
 658                 set_current_state(TASK_INTERRUPTIBLE);
 659                 vcpu_put(vcpu);
 660                 schedule();
 661                 vcpu_load(vcpu);
 662         }
 663
 664         __set_current_state(TASK_RUNNING);
 665         remove_wait_queue(&vcpu->wq, &wait);
 666 }
 667
 668 void kvm_resched(struct kvm_vcpu *vcpu)
 669 {
 670         if (!need_resched())
 671                 return;
 672         cond_resched();
 673 }
 674 EXPORT_SYMBOL_GPL(kvm_resched);
 675
 676 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 677 {
 678         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 679         struct page *page;
 680
 681         if (vmf->pgoff == 0)
 682                 page = virt_to_page(vcpu->run);
 683 #ifdef CONFIG_X86
 684         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 685                 page = virt_to_page(vcpu->arch.pio_data);
 686 #endif
 687         else
 688                 return VM_FAULT_SIGBUS;
 689         get_page(page);
 690         vmf->page = page;
 691         return 0;
 692 }
 693
 694 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 695         .fault = kvm_vcpu_fault,
 696 };
 697
 698 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 699 {
 700         vma->vm_ops = &kvm_vcpu_vm_ops;
 701         return 0;
 702 }
 703
 704 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 705 {
 706         struct kvm_vcpu *vcpu = filp->private_data;
 707
 708         fput(vcpu->kvm->filp);
 709         return 0;
 710 }
 711
 712 static const struct file_operations kvm_vcpu_fops = {
 713         .release        = kvm_vcpu_release,
 714         .unlocked_ioctl = kvm_vcpu_ioctl,
 715         .compat_ioctl   = kvm_vcpu_ioctl,
 716         .mmap           = kvm_vcpu_mmap,
 717 };
 718
 719 /*
 720  * Allocates an inode for the vcpu.
 721  */
 722 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 723 {
 724         int fd, r;
 725         struct inode *inode;
 726         struct file *file;
 727
 728         r = anon_inode_getfd(&fd, &inode, &file,
 729                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 730         if (r)
 731                 return r;
 732         atomic_inc(&vcpu->kvm->filp->f_count);
 733         return fd;
 734 }
 735
 736 /*
 737  * Creates some virtual cpus.  Good luck creating more than one.
 738  */
 739 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 740 {
 741         int r;
 742         struct kvm_vcpu *vcpu;
 743
 744         if (!valid_vcpu(n))
 745                 return -EINVAL;
 746
 747         vcpu = kvm_arch_vcpu_create(kvm, n);
 748         if (IS_ERR(vcpu))
 749                 return PTR_ERR(vcpu);
 750
 751         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 752
 753         r = kvm_arch_vcpu_setup(vcpu);
 754         if (r)
 755                 goto vcpu_destroy;
 756
 757         mutex_lock(&kvm->lock);
 758         if (kvm->vcpus[n]) {
 759                 r = -EEXIST;
 760                 mutex_unlock(&kvm->lock);
 761                 goto vcpu_destroy;
 762         }
 763         kvm->vcpus[n] = vcpu;
 764         mutex_unlock(&kvm->lock);
 765
 766         /* Now it's all set up, let userspace reach it */
 767         r = create_vcpu_fd(vcpu);
 768         if (r < 0)
 769                 goto unlink;
 770         return r;
 771
 772 unlink:
 773         mutex_lock(&kvm->lock);
 774         kvm->vcpus[n] = NULL;
 775         mutex_unlock(&kvm->lock);
 776 vcpu_destroy:
 777         kvm_arch_vcpu_destroy(vcpu);
 778         return r;
 779 }
 780
 781 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 782 {
 783         if (sigset) {
 784                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 785                 vcpu->sigset_active = 1;
 786                 vcpu->sigset = *sigset;
 787         } else
 788                 vcpu->sigset_active = 0;
 789         return 0;
 790 }
 791
 792 static long kvm_vcpu_ioctl(struct file *filp,
 793                            unsigned int ioctl, unsigned long arg)
 794 {
 795         struct kvm_vcpu *vcpu = filp->private_data;
 796         void __user *argp = (void __user *)arg;
 797         int r;
 798
 799         if (vcpu->kvm->mm != current->mm)
 800                 return -EIO;
 801         switch (ioctl) {
 802         case KVM_RUN:
 803                 r = -EINVAL;
 804                 if (arg)
 805                         goto out;
 806                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 807                 break;
 808         case KVM_GET_REGS: {
 809                 struct kvm_regs kvm_regs;
 810
 811                 memset(&kvm_regs, 0, sizeof kvm_regs);
 812                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 813                 if (r)
 814                         goto out;
 815                 r = -EFAULT;
 816                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 817                         goto out;
 818                 r = 0;
 819                 break;
 820         }
 821         case KVM_SET_REGS: {
 822                 struct kvm_regs kvm_regs;
 823
 824                 r = -EFAULT;
 825                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 826                         goto out;
 827                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 828                 if (r)
 829                         goto out;
 830                 r = 0;
 831                 break;
 832         }
 833         case KVM_GET_SREGS: {
 834                 struct kvm_sregs kvm_sregs;
 835
 836                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 837                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 838                 if (r)
 839                         goto out;
 840                 r = -EFAULT;
 841                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 842                         goto out;
 843                 r = 0;
 844                 break;
 845         }
 846         case KVM_SET_SREGS: {
 847                 struct kvm_sregs kvm_sregs;
 848
 849                 r = -EFAULT;
 850                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 851                         goto out;
 852                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 853                 if (r)
 854                         goto out;
 855                 r = 0;
 856                 break;
 857         }
 858         case KVM_TRANSLATE: {
 859                 struct kvm_translation tr;
 860
 861                 r = -EFAULT;
 862                 if (copy_from_user(&tr, argp, sizeof tr))
 863                         goto out;
 864                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 865                 if (r)
 866                         goto out;
 867                 r = -EFAULT;
 868                 if (copy_to_user(argp, &tr, sizeof tr))
 869                         goto out;
 870                 r = 0;
 871                 break;
 872         }
 873         case KVM_DEBUG_GUEST: {
 874                 struct kvm_debug_guest dbg;
 875
 876                 r = -EFAULT;
 877                 if (copy_from_user(&dbg, argp, sizeof dbg))
 878                         goto out;
 879                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 880                 if (r)
 881                         goto out;
 882                 r = 0;
 883                 break;
 884         }
 885         case KVM_SET_SIGNAL_MASK: {
 886                 struct kvm_signal_mask __user *sigmask_arg = argp;
 887                 struct kvm_signal_mask kvm_sigmask;
 888                 sigset_t sigset, *p;
 889
 890                 p = NULL;
 891                 if (argp) {
 892                         r = -EFAULT;
 893                         if (copy_from_user(&kvm_sigmask, argp,
 894                                            sizeof kvm_sigmask))
 895                                 goto out;
 896                         r = -EINVAL;
 897                         if (kvm_sigmask.len != sizeof sigset)
 898                                 goto out;
 899                         r = -EFAULT;
 900                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 901                                            sizeof sigset))
 902                                 goto out;
 903                         p = &sigset;
 904                 }
 905                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 906                 break;
 907         }
 908         case KVM_GET_FPU: {
 909                 struct kvm_fpu fpu;
 910
 911                 memset(&fpu, 0, sizeof fpu);
 912                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 913                 if (r)
 914                         goto out;
 915                 r = -EFAULT;
 916                 if (copy_to_user(argp, &fpu, sizeof fpu))
 917                         goto out;
 918                 r = 0;
 919                 break;
 920         }
 921         case KVM_SET_FPU: {
 922                 struct kvm_fpu fpu;
 923
 924                 r = -EFAULT;
 925                 if (copy_from_user(&fpu, argp, sizeof fpu))
 926                         goto out;
 927                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
 928                 if (r)
 929                         goto out;
 930                 r = 0;
 931                 break;
 932         }
 933         default:
 934                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
 935         }
 936 out:
 937         return r;
 938 }
 939
 940 static long kvm_vm_ioctl(struct file *filp,
 941                            unsigned int ioctl, unsigned long arg)
 942 {
 943         struct kvm *kvm = filp->private_data;
 944         void __user *argp = (void __user *)arg;
 945         int r;
 946
 947         if (kvm->mm != current->mm)
 948                 return -EIO;
 949         switch (ioctl) {
 950         case KVM_CREATE_VCPU:
 951                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
 952                 if (r < 0)
 953                         goto out;
 954                 break;
 955         case KVM_SET_USER_MEMORY_REGION: {
 956                 struct kvm_userspace_memory_region kvm_userspace_mem;
 957
 958                 r = -EFAULT;
 959                 if (copy_from_user(&kvm_userspace_mem, argp,
 960                                                 sizeof kvm_userspace_mem))
 961                         goto out;
 962
 963                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
 964                 if (r)
 965                         goto out;
 966                 break;
 967         }
 968         case KVM_GET_DIRTY_LOG: {
 969                 struct kvm_dirty_log log;
 970
 971                 r = -EFAULT;
 972                 if (copy_from_user(&log, argp, sizeof log))
 973                         goto out;
 974                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
 975                 if (r)
 976                         goto out;
 977                 break;
 978         }
 979         default:
 980                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
 981         }
 982 out:
 983         return r;
 984 }
 985
 986 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 987 {
 988         struct kvm *kvm = vma->vm_file->private_data;
 989         struct page *page;
 990
 991         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
 992                 return VM_FAULT_SIGBUS;
 993         page = gfn_to_page(kvm, vmf->pgoff);
 994         if (is_error_page(page)) {
 995                 kvm_release_page_clean(page);
 996                 return VM_FAULT_SIGBUS;
 997         }
 998         vmf->page = page;
 999         return 0;
1000 }
1001
1002 static struct vm_operations_struct kvm_vm_vm_ops = {
1003         .fault = kvm_vm_fault,
1004 };
1005
1006 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1007 {
1008         vma->vm_ops = &kvm_vm_vm_ops;
1009         return 0;
1010 }
1011
1012 static const struct file_operations kvm_vm_fops = {
1013         .release        = kvm_vm_release,
1014         .unlocked_ioctl = kvm_vm_ioctl,
1015         .compat_ioctl   = kvm_vm_ioctl,
1016         .mmap           = kvm_vm_mmap,
1017 };
1018
1019 static int kvm_dev_ioctl_create_vm(void)
1020 {
1021         int fd, r;
1022         struct inode *inode;
1023         struct file *file;
1024         struct kvm *kvm;
1025
1026         kvm = kvm_create_vm();
1027         if (IS_ERR(kvm))
1028                 return PTR_ERR(kvm);
1029         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1030         if (r) {
1031                 kvm_destroy_vm(kvm);
1032                 return r;
1033         }
1034
1035         kvm->filp = file;
1036
1037         return fd;
1038 }
1039
1040 static long kvm_dev_ioctl(struct file *filp,
1041                           unsigned int ioctl, unsigned long arg)
1042 {
1043         void __user *argp = (void __user *)arg;
1044         long r = -EINVAL;
1045
1046         switch (ioctl) {
1047         case KVM_GET_API_VERSION:
1048                 r = -EINVAL;
1049                 if (arg)
1050                         goto out;
1051                 r = KVM_API_VERSION;
1052                 break;
1053         case KVM_CREATE_VM:
1054                 r = -EINVAL;
1055                 if (arg)
1056                         goto out;
1057                 r = kvm_dev_ioctl_create_vm();
1058                 break;
1059         case KVM_CHECK_EXTENSION:
1060                 r = kvm_dev_ioctl_check_extension((long)argp);
1061                 break;
1062         case KVM_GET_VCPU_MMAP_SIZE:
1063                 r = -EINVAL;
1064                 if (arg)
1065                         goto out;
1066                 r = PAGE_SIZE;     /* struct kvm_run */
1067 #ifdef CONFIG_X86
1068                 r += PAGE_SIZE;    /* pio data page */
1069 #endif
1070                 break;
1071         default:
1072                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1073         }
1074 out:
1075         return r;
1076 }
1077
1078 static struct file_operations kvm_chardev_ops = {
1079         .unlocked_ioctl = kvm_dev_ioctl,
1080         .compat_ioctl   = kvm_dev_ioctl,
1081 };
1082
1083 static struct miscdevice kvm_dev = {
1084         KVM_MINOR,
1085         "kvm",
1086         &kvm_chardev_ops,
1087 };
1088
1089 static void hardware_enable(void *junk)
1090 {
1091         int cpu = raw_smp_processor_id();
1092
1093         if (cpu_isset(cpu, cpus_hardware_enabled))
1094                 return;
1095         cpu_set(cpu, cpus_hardware_enabled);
1096         kvm_arch_hardware_enable(NULL);
1097 }
1098
1099 static void hardware_disable(void *junk)
1100 {
1101         int cpu = raw_smp_processor_id();
1102
1103         if (!cpu_isset(cpu, cpus_hardware_enabled))
1104                 return;
1105         cpu_clear(cpu, cpus_hardware_enabled);
1106         decache_vcpus_on_cpu(cpu);
1107         kvm_arch_hardware_disable(NULL);
1108 }
1109
1110 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1111                            void *v)
1112 {
1113         int cpu = (long)v;
1114
1115         val &= ~CPU_TASKS_FROZEN;
1116         switch (val) {
1117         case CPU_DYING:
1118                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1119                        cpu);
1120                 hardware_disable(NULL);
1121                 break;
1122         case CPU_UP_CANCELED:
1123                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1124                        cpu);
1125                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1126                 break;
1127         case CPU_ONLINE:
1128                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1129                        cpu);
1130                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1131                 break;
1132         }
1133         return NOTIFY_OK;
1134 }
1135
1136 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1137                       void *v)
1138 {
1139         if (val == SYS_RESTART) {
1140                 /*
1141                  * Some (well, at least mine) BIOSes hang on reboot if
1142                  * in vmx root mode.
1143                  */
1144                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1145                 on_each_cpu(hardware_disable, NULL, 0, 1);
1146         }
1147         return NOTIFY_OK;
1148 }
1149
1150 static struct notifier_block kvm_reboot_notifier = {
1151         .notifier_call = kvm_reboot,
1152         .priority = 0,
1153 };
1154
1155 void kvm_io_bus_init(struct kvm_io_bus *bus)
1156 {
1157         memset(bus, 0, sizeof(*bus));
1158 }
1159
1160 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1161 {
1162         int i;
1163
1164         for (i = 0; i < bus->dev_count; i++) {
1165                 struct kvm_io_device *pos = bus->devs[i];
1166
1167                 kvm_iodevice_destructor(pos);
1168         }
1169 }
1170
1171 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1172 {
1173         int i;
1174
1175         for (i = 0; i < bus->dev_count; i++) {
1176                 struct kvm_io_device *pos = bus->devs[i];
1177
1178                 if (pos->in_range(pos, addr))
1179                         return pos;
1180         }
1181
1182         return NULL;
1183 }
1184
1185 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1186 {
1187         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1188
1189         bus->devs[bus->dev_count++] = dev;
1190 }
1191
1192 static struct notifier_block kvm_cpu_notifier = {
1193         .notifier_call = kvm_cpu_hotplug,
1194         .priority = 20, /* must be > scheduler priority */
1195 };
1196
1197 static int vm_stat_get(void *_offset, u64 *val)
1198 {
1199         unsigned offset = (long)_offset;
1200         struct kvm *kvm;
1201
1202         *val = 0;
1203         spin_lock(&kvm_lock);
1204         list_for_each_entry(kvm, &vm_list, vm_list)
1205                 *val += *(u32 *)((void *)kvm + offset);
1206         spin_unlock(&kvm_lock);
1207         return 0;
1208 }
1209
1210 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1211
1212 static int vcpu_stat_get(void *_offset, u64 *val)
1213 {
1214         unsigned offset = (long)_offset;
1215         struct kvm *kvm;
1216         struct kvm_vcpu *vcpu;
1217         int i;
1218
1219         *val = 0;
1220         spin_lock(&kvm_lock);
1221         list_for_each_entry(kvm, &vm_list, vm_list)
1222                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1223                         vcpu = kvm->vcpus[i];
1224                         if (vcpu)
1225                                 *val += *(u32 *)((void *)vcpu + offset);
1226                 }
1227         spin_unlock(&kvm_lock);
1228         return 0;
1229 }
1230
1231 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1232
1233 static struct file_operations *stat_fops[] = {
1234         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1235         [KVM_STAT_VM]   = &vm_stat_fops,
1236 };
1237
1238 static void kvm_init_debug(void)
1239 {
1240         struct kvm_stats_debugfs_item *p;
1241
1242         debugfs_dir = debugfs_create_dir("kvm", NULL);
1243         for (p = debugfs_entries; p->name; ++p)
1244                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1245                                                 (void *)(long)p->offset,
1246                                                 stat_fops[p->kind]);
1247 }
1248
1249 static void kvm_exit_debug(void)
1250 {
1251         struct kvm_stats_debugfs_item *p;
1252
1253         for (p = debugfs_entries; p->name; ++p)
1254                 debugfs_remove(p->dentry);
1255         debugfs_remove(debugfs_dir);
1256 }
1257
1258 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1259 {
1260         hardware_disable(NULL);
1261         return 0;
1262 }
1263
1264 static int kvm_resume(struct sys_device *dev)
1265 {
1266         hardware_enable(NULL);
1267         return 0;
1268 }
1269
1270 static struct sysdev_class kvm_sysdev_class = {
1271         .name = "kvm",
1272         .suspend = kvm_suspend,
1273         .resume = kvm_resume,
1274 };
1275
1276 static struct sys_device kvm_sysdev = {
1277         .id = 0,
1278         .cls = &kvm_sysdev_class,
1279 };
1280
1281 struct page *bad_page;
1282
1283 static inline
1284 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1285 {
1286         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1287 }
1288
1289 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1290 {
1291         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1292
1293         kvm_arch_vcpu_load(vcpu, cpu);
1294 }
1295
1296 static void kvm_sched_out(struct preempt_notifier *pn,
1297                           struct task_struct *next)
1298 {
1299         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1300
1301         kvm_arch_vcpu_put(vcpu);
1302 }
1303
1304 int kvm_init(void *opaque, unsigned int vcpu_size,
1305                   struct module *module)
1306 {
1307         int r;
1308         int cpu;
1309
1310         kvm_init_debug();
1311
1312         r = kvm_arch_init(opaque);
1313         if (r)
1314                 goto out_fail;
1315
1316         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1317
1318         if (bad_page == NULL) {
1319                 r = -ENOMEM;
1320                 goto out;
1321         }
1322
1323         r = kvm_arch_hardware_setup();
1324         if (r < 0)
1325                 goto out_free_0;
1326
1327         for_each_online_cpu(cpu) {
1328                 smp_call_function_single(cpu,
1329                                 kvm_arch_check_processor_compat,
1330                                 &r, 0, 1);
1331                 if (r < 0)
1332                         goto out_free_1;
1333         }
1334
1335         on_each_cpu(hardware_enable, NULL, 0, 1);
1336         r = register_cpu_notifier(&kvm_cpu_notifier);
1337         if (r)
1338                 goto out_free_2;
1339         register_reboot_notifier(&kvm_reboot_notifier);
1340
1341         r = sysdev_class_register(&kvm_sysdev_class);
1342         if (r)
1343                 goto out_free_3;
1344
1345         r = sysdev_register(&kvm_sysdev);
1346         if (r)
1347                 goto out_free_4;
1348
1349         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1350         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1351                                            __alignof__(struct kvm_vcpu),
1352                                            0, NULL);
1353         if (!kvm_vcpu_cache) {
1354                 r = -ENOMEM;
1355                 goto out_free_5;
1356         }
1357
1358         kvm_chardev_ops.owner = module;
1359
1360         r = misc_register(&kvm_dev);
1361         if (r) {
1362                 printk(KERN_ERR "kvm: misc device register failed\n");
1363                 goto out_free;
1364         }
1365
1366         kvm_preempt_ops.sched_in = kvm_sched_in;
1367         kvm_preempt_ops.sched_out = kvm_sched_out;
1368
1369         return 0;
1370
1371 out_free:
1372         kmem_cache_destroy(kvm_vcpu_cache);
1373 out_free_5:
1374         sysdev_unregister(&kvm_sysdev);
1375 out_free_4:
1376         sysdev_class_unregister(&kvm_sysdev_class);
1377 out_free_3:
1378         unregister_reboot_notifier(&kvm_reboot_notifier);
1379         unregister_cpu_notifier(&kvm_cpu_notifier);
1380 out_free_2:
1381         on_each_cpu(hardware_disable, NULL, 0, 1);
1382 out_free_1:
1383         kvm_arch_hardware_unsetup();
1384 out_free_0:
1385         __free_page(bad_page);
1386 out:
1387         kvm_arch_exit();
1388         kvm_exit_debug();
1389 out_fail:
1390         return r;
1391 }
1392 EXPORT_SYMBOL_GPL(kvm_init);
1393
1394 void kvm_exit(void)
1395 {
1396         misc_deregister(&kvm_dev);
1397         kmem_cache_destroy(kvm_vcpu_cache);
1398         sysdev_unregister(&kvm_sysdev);
1399         sysdev_class_unregister(&kvm_sysdev_class);
1400         unregister_reboot_notifier(&kvm_reboot_notifier);
1401         unregister_cpu_notifier(&kvm_cpu_notifier);
1402         on_each_cpu(hardware_disable, NULL, 0, 1);
1403         kvm_arch_hardware_unsetup();
1404         kvm_arch_exit();
1405         kvm_exit_debug();
1406         __free_page(bad_page);
1407 }
1408 EXPORT_SYMBOL_GPL(kvm_exit);