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
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40 #include <linux/profile.h>
  41 #include <linux/kvm_para.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mman.h>
  44
  45 #include <asm/processor.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/desc.h>
  49 #include <asm/pgtable.h>
  50
  51 MODULE_AUTHOR("Qumranet");
  52 MODULE_LICENSE("GPL");
  53
  54 DEFINE_SPINLOCK(kvm_lock);
  55 LIST_HEAD(vm_list);
  56
  57 static cpumask_t cpus_hardware_enabled;
  58
  59 struct kmem_cache *kvm_vcpu_cache;
  60 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  61
  62 static __read_mostly struct preempt_ops kvm_preempt_ops;
  63
  64 static struct dentry *debugfs_dir;
  65
  66 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  67                            unsigned long arg);
  68
  69 static inline int valid_vcpu(int n)
  70 {
  71         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  72 }
  73
  74 /*
  75  * Switches to specified vcpu, until a matching vcpu_put()
  76  */
  77 void vcpu_load(struct kvm_vcpu *vcpu)
  78 {
  79         int cpu;
  80
  81         mutex_lock(&vcpu->mutex);
  82         cpu = get_cpu();
  83         preempt_notifier_register(&vcpu->preempt_notifier);
  84         kvm_arch_vcpu_load(vcpu, cpu);
  85         put_cpu();
  86 }
  87
  88 void vcpu_put(struct kvm_vcpu *vcpu)
  89 {
  90         preempt_disable();
  91         kvm_arch_vcpu_put(vcpu);
  92         preempt_notifier_unregister(&vcpu->preempt_notifier);
  93         preempt_enable();
  94         mutex_unlock(&vcpu->mutex);
  95 }
  96
  97 static void ack_flush(void *_completed)
  98 {
  99 }
 100
 101 void kvm_flush_remote_tlbs(struct kvm *kvm)
 102 {
 103         int i, cpu;
 104         cpumask_t cpus;
 105         struct kvm_vcpu *vcpu;
 106
 107         cpus_clear(cpus);
 108         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 109                 vcpu = kvm->vcpus[i];
 110                 if (!vcpu)
 111                         continue;
 112                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 113                         continue;
 114                 cpu = vcpu->cpu;
 115                 if (cpu != -1 && cpu != raw_smp_processor_id())
 116                         cpu_set(cpu, cpus);
 117         }
 118         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 119 }
 120
 121 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 122 {
 123         struct page *page;
 124         int r;
 125
 126         mutex_init(&vcpu->mutex);
 127         vcpu->cpu = -1;
 128         vcpu->kvm = kvm;
 129         vcpu->vcpu_id = id;
 130         init_waitqueue_head(&vcpu->wq);
 131
 132         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 133         if (!page) {
 134                 r = -ENOMEM;
 135                 goto fail;
 136         }
 137         vcpu->run = page_address(page);
 138
 139         r = kvm_arch_vcpu_init(vcpu);
 140         if (r < 0)
 141                 goto fail_free_run;
 142         return 0;
 143
 144 fail_free_run:
 145         free_page((unsigned long)vcpu->run);
 146 fail:
 147         return r;
 148 }
 149 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 150
 151 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 152 {
 153         kvm_arch_vcpu_uninit(vcpu);
 154         free_page((unsigned long)vcpu->run);
 155 }
 156 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 157
 158 static struct kvm *kvm_create_vm(void)
 159 {
 160         struct kvm *kvm = kvm_arch_create_vm();
 161
 162         if (IS_ERR(kvm))
 163                 goto out;
 164
 165         kvm_io_bus_init(&kvm->pio_bus);
 166         mutex_init(&kvm->lock);
 167         kvm_io_bus_init(&kvm->mmio_bus);
 168         spin_lock(&kvm_lock);
 169         list_add(&kvm->vm_list, &vm_list);
 170         spin_unlock(&kvm_lock);
 171 out:
 172         return kvm;
 173 }
 174
 175 /*
 176  * Free any memory in @free but not in @dont.
 177  */
 178 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 179                                   struct kvm_memory_slot *dont)
 180 {
 181         if (!dont || free->rmap != dont->rmap)
 182                 vfree(free->rmap);
 183
 184         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 185                 vfree(free->dirty_bitmap);
 186
 187         free->npages = 0;
 188         free->dirty_bitmap = NULL;
 189         free->rmap = NULL;
 190 }
 191
 192 void kvm_free_physmem(struct kvm *kvm)
 193 {
 194         int i;
 195
 196         for (i = 0; i < kvm->nmemslots; ++i)
 197                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 198 }
 199
 200 static void kvm_destroy_vm(struct kvm *kvm)
 201 {
 202         spin_lock(&kvm_lock);
 203         list_del(&kvm->vm_list);
 204         spin_unlock(&kvm_lock);
 205         kvm_io_bus_destroy(&kvm->pio_bus);
 206         kvm_io_bus_destroy(&kvm->mmio_bus);
 207         kvm_arch_destroy_vm(kvm);
 208 }
 209
 210 static int kvm_vm_release(struct inode *inode, struct file *filp)
 211 {
 212         struct kvm *kvm = filp->private_data;
 213
 214         kvm_destroy_vm(kvm);
 215         return 0;
 216 }
 217
 218 /*
 219  * Allocate some memory and give it an address in the guest physical address
 220  * space.
 221  *
 222  * Discontiguous memory is allowed, mostly for framebuffers.
 223  *
 224  * Must be called holding kvm->lock.
 225  */
 226 int __kvm_set_memory_region(struct kvm *kvm,
 227                             struct kvm_userspace_memory_region *mem,
 228                             int user_alloc)
 229 {
 230         int r;
 231         gfn_t base_gfn;
 232         unsigned long npages;
 233         unsigned long i;
 234         struct kvm_memory_slot *memslot;
 235         struct kvm_memory_slot old, new;
 236
 237         r = -EINVAL;
 238         /* General sanity checks */
 239         if (mem->memory_size & (PAGE_SIZE - 1))
 240                 goto out;
 241         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 242                 goto out;
 243         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 244                 goto out;
 245         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 246                 goto out;
 247
 248         memslot = &kvm->memslots[mem->slot];
 249         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 250         npages = mem->memory_size >> PAGE_SHIFT;
 251
 252         if (!npages)
 253                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 254
 255         new = old = *memslot;
 256
 257         new.base_gfn = base_gfn;
 258         new.npages = npages;
 259         new.flags = mem->flags;
 260
 261         /* Disallow changing a memory slot's size. */
 262         r = -EINVAL;
 263         if (npages && old.npages && npages != old.npages)
 264                 goto out_free;
 265
 266         /* Check for overlaps */
 267         r = -EEXIST;
 268         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 269                 struct kvm_memory_slot *s = &kvm->memslots[i];
 270
 271                 if (s == memslot)
 272                         continue;
 273                 if (!((base_gfn + npages <= s->base_gfn) ||
 274                       (base_gfn >= s->base_gfn + s->npages)))
 275                         goto out_free;
 276         }
 277
 278         /* Free page dirty bitmap if unneeded */
 279         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 280                 new.dirty_bitmap = NULL;
 281
 282         r = -ENOMEM;
 283
 284         /* Allocate if a slot is being created */
 285         if (npages && !new.rmap) {
 286                 new.rmap = vmalloc(npages * sizeof(struct page *));
 287
 288                 if (!new.rmap)
 289                         goto out_free;
 290
 291                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 292
 293                 new.user_alloc = user_alloc;
 294                 new.userspace_addr = mem->userspace_addr;
 295         }
 296
 297         /* Allocate page dirty bitmap if needed */
 298         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 299                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 300
 301                 new.dirty_bitmap = vmalloc(dirty_bytes);
 302                 if (!new.dirty_bitmap)
 303                         goto out_free;
 304                 memset(new.dirty_bitmap, 0, dirty_bytes);
 305         }
 306
 307         if (mem->slot >= kvm->nmemslots)
 308                 kvm->nmemslots = mem->slot + 1;
 309
 310         *memslot = new;
 311
 312         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 313         if (r) {
 314                 *memslot = old;
 315                 goto out_free;
 316         }
 317
 318         kvm_free_physmem_slot(&old, &new);
 319         return 0;
 320
 321 out_free:
 322         kvm_free_physmem_slot(&new, &old);
 323 out:
 324         return r;
 325
 326 }
 327 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 328
 329 int kvm_set_memory_region(struct kvm *kvm,
 330                           struct kvm_userspace_memory_region *mem,
 331                           int user_alloc)
 332 {
 333         int r;
 334
 335         mutex_lock(&kvm->lock);
 336         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 337         mutex_unlock(&kvm->lock);
 338         return r;
 339 }
 340 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 341
 342 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 343                                    struct
 344                                    kvm_userspace_memory_region *mem,
 345                                    int user_alloc)
 346 {
 347         if (mem->slot >= KVM_MEMORY_SLOTS)
 348                 return -EINVAL;
 349         return kvm_set_memory_region(kvm, mem, user_alloc);
 350 }
 351
 352 int kvm_get_dirty_log(struct kvm *kvm,
 353                         struct kvm_dirty_log *log, int *is_dirty)
 354 {
 355         struct kvm_memory_slot *memslot;
 356         int r, i;
 357         int n;
 358         unsigned long any = 0;
 359
 360         r = -EINVAL;
 361         if (log->slot >= KVM_MEMORY_SLOTS)
 362                 goto out;
 363
 364         memslot = &kvm->memslots[log->slot];
 365         r = -ENOENT;
 366         if (!memslot->dirty_bitmap)
 367                 goto out;
 368
 369         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 370
 371         for (i = 0; !any && i < n/sizeof(long); ++i)
 372                 any = memslot->dirty_bitmap[i];
 373
 374         r = -EFAULT;
 375         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 376                 goto out;
 377
 378         if (any)
 379                 *is_dirty = 1;
 380
 381         r = 0;
 382 out:
 383         return r;
 384 }
 385
 386 int is_error_page(struct page *page)
 387 {
 388         return page == bad_page;
 389 }
 390 EXPORT_SYMBOL_GPL(is_error_page);
 391
 392 static inline unsigned long bad_hva(void)
 393 {
 394         return PAGE_OFFSET;
 395 }
 396
 397 int kvm_is_error_hva(unsigned long addr)
 398 {
 399         return addr == bad_hva();
 400 }
 401 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 402
 403 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 404 {
 405         int i;
 406         struct kvm_mem_alias *alias;
 407
 408         for (i = 0; i < kvm->naliases; ++i) {
 409                 alias = &kvm->aliases[i];
 410                 if (gfn >= alias->base_gfn
 411                     && gfn < alias->base_gfn + alias->npages)
 412                         return alias->target_gfn + gfn - alias->base_gfn;
 413         }
 414         return gfn;
 415 }
 416
 417 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 418 {
 419         int i;
 420
 421         for (i = 0; i < kvm->nmemslots; ++i) {
 422                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 423
 424                 if (gfn >= memslot->base_gfn
 425                     && gfn < memslot->base_gfn + memslot->npages)
 426                         return memslot;
 427         }
 428         return NULL;
 429 }
 430
 431 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 432 {
 433         gfn = unalias_gfn(kvm, gfn);
 434         return __gfn_to_memslot(kvm, gfn);
 435 }
 436
 437 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 438 {
 439         int i;
 440
 441         gfn = unalias_gfn(kvm, gfn);
 442         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 443                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 444
 445                 if (gfn >= memslot->base_gfn
 446                     && gfn < memslot->base_gfn + memslot->npages)
 447                         return 1;
 448         }
 449         return 0;
 450 }
 451 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 452
 453 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 454 {
 455         struct kvm_memory_slot *slot;
 456
 457         gfn = unalias_gfn(kvm, gfn);
 458         slot = __gfn_to_memslot(kvm, gfn);
 459         if (!slot)
 460                 return bad_hva();
 461         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 462 }
 463
 464 /*
 465  * Requires current->mm->mmap_sem to be held
 466  */
 467 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 468 {
 469         struct page *page[1];
 470         unsigned long addr;
 471         int npages;
 472
 473         might_sleep();
 474
 475         addr = gfn_to_hva(kvm, gfn);
 476         if (kvm_is_error_hva(addr)) {
 477                 get_page(bad_page);
 478                 return bad_page;
 479         }
 480
 481         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 482                                 NULL);
 483
 484         if (npages != 1) {
 485                 get_page(bad_page);
 486                 return bad_page;
 487         }
 488
 489         return page[0];
 490 }
 491
 492 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 493 {
 494         struct page *page;
 495
 496         down_read(&current->mm->mmap_sem);
 497         page = __gfn_to_page(kvm, gfn);
 498         up_read(&current->mm->mmap_sem);
 499
 500         return page;
 501 }
 502
 503 EXPORT_SYMBOL_GPL(gfn_to_page);
 504
 505 void kvm_release_page_clean(struct page *page)
 506 {
 507         put_page(page);
 508 }
 509 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 510
 511 void kvm_release_page_dirty(struct page *page)
 512 {
 513         if (!PageReserved(page))
 514                 SetPageDirty(page);
 515         put_page(page);
 516 }
 517 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 518
 519 static int next_segment(unsigned long len, int offset)
 520 {
 521         if (len > PAGE_SIZE - offset)
 522                 return PAGE_SIZE - offset;
 523         else
 524                 return len;
 525 }
 526
 527 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 528                         int len)
 529 {
 530         int r;
 531         unsigned long addr;
 532
 533         addr = gfn_to_hva(kvm, gfn);
 534         if (kvm_is_error_hva(addr))
 535                 return -EFAULT;
 536         r = copy_from_user(data, (void __user *)addr + offset, len);
 537         if (r)
 538                 return -EFAULT;
 539         return 0;
 540 }
 541 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 542
 543 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 544 {
 545         gfn_t gfn = gpa >> PAGE_SHIFT;
 546         int seg;
 547         int offset = offset_in_page(gpa);
 548         int ret;
 549
 550         while ((seg = next_segment(len, offset)) != 0) {
 551                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 552                 if (ret < 0)
 553                         return ret;
 554                 offset = 0;
 555                 len -= seg;
 556                 data += seg;
 557                 ++gfn;
 558         }
 559         return 0;
 560 }
 561 EXPORT_SYMBOL_GPL(kvm_read_guest);
 562
 563 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 564                          int offset, int len)
 565 {
 566         int r;
 567         unsigned long addr;
 568
 569         addr = gfn_to_hva(kvm, gfn);
 570         if (kvm_is_error_hva(addr))
 571                 return -EFAULT;
 572         r = copy_to_user((void __user *)addr + offset, data, len);
 573         if (r)
 574                 return -EFAULT;
 575         mark_page_dirty(kvm, gfn);
 576         return 0;
 577 }
 578 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 579
 580 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 581                     unsigned long len)
 582 {
 583         gfn_t gfn = gpa >> PAGE_SHIFT;
 584         int seg;
 585         int offset = offset_in_page(gpa);
 586         int ret;
 587
 588         while ((seg = next_segment(len, offset)) != 0) {
 589                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 590                 if (ret < 0)
 591                         return ret;
 592                 offset = 0;
 593                 len -= seg;
 594                 data += seg;
 595                 ++gfn;
 596         }
 597         return 0;
 598 }
 599
 600 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 601 {
 602         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 603 }
 604 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 605
 606 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 607 {
 608         gfn_t gfn = gpa >> PAGE_SHIFT;
 609         int seg;
 610         int offset = offset_in_page(gpa);
 611         int ret;
 612
 613         while ((seg = next_segment(len, offset)) != 0) {
 614                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 615                 if (ret < 0)
 616                         return ret;
 617                 offset = 0;
 618                 len -= seg;
 619                 ++gfn;
 620         }
 621         return 0;
 622 }
 623 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 624
 625 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 626 {
 627         struct kvm_memory_slot *memslot;
 628
 629         gfn = unalias_gfn(kvm, gfn);
 630         memslot = __gfn_to_memslot(kvm, gfn);
 631         if (memslot && memslot->dirty_bitmap) {
 632                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 633
 634                 /* avoid RMW */
 635                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 636                         set_bit(rel_gfn, memslot->dirty_bitmap);
 637         }
 638 }
 639
 640 /*
 641  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 642  */
 643 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 644 {
 645         DECLARE_WAITQUEUE(wait, current);
 646
 647         add_wait_queue(&vcpu->wq, &wait);
 648
 649         /*
 650          * We will block until either an interrupt or a signal wakes us up
 651          */
 652         while (!kvm_cpu_has_interrupt(vcpu)
 653                && !signal_pending(current)
 654                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
 655                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
 656                 set_current_state(TASK_INTERRUPTIBLE);
 657                 vcpu_put(vcpu);
 658                 schedule();
 659                 vcpu_load(vcpu);
 660         }
 661
 662         __set_current_state(TASK_RUNNING);
 663         remove_wait_queue(&vcpu->wq, &wait);
 664 }
 665
 666 void kvm_resched(struct kvm_vcpu *vcpu)
 667 {
 668         if (!need_resched())
 669                 return;
 670         cond_resched();
 671 }
 672 EXPORT_SYMBOL_GPL(kvm_resched);
 673
 674 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
 675                                     struct kvm_interrupt *irq)
 676 {
 677         if (irq->irq < 0 || irq->irq >= 256)
 678                 return -EINVAL;
 679         if (irqchip_in_kernel(vcpu->kvm))
 680                 return -ENXIO;
 681         vcpu_load(vcpu);
 682
 683         set_bit(irq->irq, vcpu->irq_pending);
 684         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
 685
 686         vcpu_put(vcpu);
 687
 688         return 0;
 689 }
 690
 691 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
 692                                     unsigned long address,
 693                                     int *type)
 694 {
 695         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 696         unsigned long pgoff;
 697         struct page *page;
 698
 699         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 700         if (pgoff == 0)
 701                 page = virt_to_page(vcpu->run);
 702         else if (pgoff == KVM_PIO_PAGE_OFFSET)
 703                 page = virt_to_page(vcpu->pio_data);
 704         else
 705                 return NOPAGE_SIGBUS;
 706         get_page(page);
 707         if (type != NULL)
 708                 *type = VM_FAULT_MINOR;
 709
 710         return page;
 711 }
 712
 713 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 714         .nopage = kvm_vcpu_nopage,
 715 };
 716
 717 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 718 {
 719         vma->vm_ops = &kvm_vcpu_vm_ops;
 720         return 0;
 721 }
 722
 723 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 724 {
 725         struct kvm_vcpu *vcpu = filp->private_data;
 726
 727         fput(vcpu->kvm->filp);
 728         return 0;
 729 }
 730
 731 static struct file_operations kvm_vcpu_fops = {
 732         .release        = kvm_vcpu_release,
 733         .unlocked_ioctl = kvm_vcpu_ioctl,
 734         .compat_ioctl   = kvm_vcpu_ioctl,
 735         .mmap           = kvm_vcpu_mmap,
 736 };
 737
 738 /*
 739  * Allocates an inode for the vcpu.
 740  */
 741 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 742 {
 743         int fd, r;
 744         struct inode *inode;
 745         struct file *file;
 746
 747         r = anon_inode_getfd(&fd, &inode, &file,
 748                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 749         if (r)
 750                 return r;
 751         atomic_inc(&vcpu->kvm->filp->f_count);
 752         return fd;
 753 }
 754
 755 /*
 756  * Creates some virtual cpus.  Good luck creating more than one.
 757  */
 758 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 759 {
 760         int r;
 761         struct kvm_vcpu *vcpu;
 762
 763         if (!valid_vcpu(n))
 764                 return -EINVAL;
 765
 766         vcpu = kvm_arch_vcpu_create(kvm, n);
 767         if (IS_ERR(vcpu))
 768                 return PTR_ERR(vcpu);
 769
 770         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 771
 772         r = kvm_arch_vcpu_setup(vcpu);
 773         if (r)
 774                 goto vcpu_destroy;
 775
 776         mutex_lock(&kvm->lock);
 777         if (kvm->vcpus[n]) {
 778                 r = -EEXIST;
 779                 mutex_unlock(&kvm->lock);
 780                 goto vcpu_destroy;
 781         }
 782         kvm->vcpus[n] = vcpu;
 783         mutex_unlock(&kvm->lock);
 784
 785         /* Now it's all set up, let userspace reach it */
 786         r = create_vcpu_fd(vcpu);
 787         if (r < 0)
 788                 goto unlink;
 789         return r;
 790
 791 unlink:
 792         mutex_lock(&kvm->lock);
 793         kvm->vcpus[n] = NULL;
 794         mutex_unlock(&kvm->lock);
 795 vcpu_destroy:
 796         kvm_arch_vcpu_destroy(vcpu);
 797         return r;
 798 }
 799
 800 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 801 {
 802         if (sigset) {
 803                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 804                 vcpu->sigset_active = 1;
 805                 vcpu->sigset = *sigset;
 806         } else
 807                 vcpu->sigset_active = 0;
 808         return 0;
 809 }
 810
 811 static long kvm_vcpu_ioctl(struct file *filp,
 812                            unsigned int ioctl, unsigned long arg)
 813 {
 814         struct kvm_vcpu *vcpu = filp->private_data;
 815         void __user *argp = (void __user *)arg;
 816         int r;
 817
 818         switch (ioctl) {
 819         case KVM_RUN:
 820                 r = -EINVAL;
 821                 if (arg)
 822                         goto out;
 823                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 824                 break;
 825         case KVM_GET_REGS: {
 826                 struct kvm_regs kvm_regs;
 827
 828                 memset(&kvm_regs, 0, sizeof kvm_regs);
 829                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 830                 if (r)
 831                         goto out;
 832                 r = -EFAULT;
 833                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 834                         goto out;
 835                 r = 0;
 836                 break;
 837         }
 838         case KVM_SET_REGS: {
 839                 struct kvm_regs kvm_regs;
 840
 841                 r = -EFAULT;
 842                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 843                         goto out;
 844                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 845                 if (r)
 846                         goto out;
 847                 r = 0;
 848                 break;
 849         }
 850         case KVM_GET_SREGS: {
 851                 struct kvm_sregs kvm_sregs;
 852
 853                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 854                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 855                 if (r)
 856                         goto out;
 857                 r = -EFAULT;
 858                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 859                         goto out;
 860                 r = 0;
 861                 break;
 862         }
 863         case KVM_SET_SREGS: {
 864                 struct kvm_sregs kvm_sregs;
 865
 866                 r = -EFAULT;
 867                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 868                         goto out;
 869                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 870                 if (r)
 871                         goto out;
 872                 r = 0;
 873                 break;
 874         }
 875         case KVM_TRANSLATE: {
 876                 struct kvm_translation tr;
 877
 878                 r = -EFAULT;
 879                 if (copy_from_user(&tr, argp, sizeof tr))
 880                         goto out;
 881                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 882                 if (r)
 883                         goto out;
 884                 r = -EFAULT;
 885                 if (copy_to_user(argp, &tr, sizeof tr))
 886                         goto out;
 887                 r = 0;
 888                 break;
 889         }
 890         case KVM_INTERRUPT: {
 891                 struct kvm_interrupt irq;
 892
 893                 r = -EFAULT;
 894                 if (copy_from_user(&irq, argp, sizeof irq))
 895                         goto out;
 896                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 897                 if (r)
 898                         goto out;
 899                 r = 0;
 900                 break;
 901         }
 902         case KVM_DEBUG_GUEST: {
 903                 struct kvm_debug_guest dbg;
 904
 905                 r = -EFAULT;
 906                 if (copy_from_user(&dbg, argp, sizeof dbg))
 907                         goto out;
 908                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 909                 if (r)
 910                         goto out;
 911                 r = 0;
 912                 break;
 913         }
 914         case KVM_SET_SIGNAL_MASK: {
 915                 struct kvm_signal_mask __user *sigmask_arg = argp;
 916                 struct kvm_signal_mask kvm_sigmask;
 917                 sigset_t sigset, *p;
 918
 919                 p = NULL;
 920                 if (argp) {
 921                         r = -EFAULT;
 922                         if (copy_from_user(&kvm_sigmask, argp,
 923                                            sizeof kvm_sigmask))
 924                                 goto out;
 925                         r = -EINVAL;
 926                         if (kvm_sigmask.len != sizeof sigset)
 927                                 goto out;
 928                         r = -EFAULT;
 929                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 930                                            sizeof sigset))
 931                                 goto out;
 932                         p = &sigset;
 933                 }
 934                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 935                 break;
 936         }
 937         case KVM_GET_FPU: {
 938                 struct kvm_fpu fpu;
 939
 940                 memset(&fpu, 0, sizeof fpu);
 941                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 942                 if (r)
 943                         goto out;
 944                 r = -EFAULT;
 945                 if (copy_to_user(argp, &fpu, sizeof fpu))
 946                         goto out;
 947                 r = 0;
 948                 break;
 949         }
 950         case KVM_SET_FPU: {
 951                 struct kvm_fpu fpu;
 952
 953                 r = -EFAULT;
 954                 if (copy_from_user(&fpu, argp, sizeof fpu))
 955                         goto out;
 956                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
 957                 if (r)
 958                         goto out;
 959                 r = 0;
 960                 break;
 961         }
 962         default:
 963                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
 964         }
 965 out:
 966         return r;
 967 }
 968
 969 static long kvm_vm_ioctl(struct file *filp,
 970                            unsigned int ioctl, unsigned long arg)
 971 {
 972         struct kvm *kvm = filp->private_data;
 973         void __user *argp = (void __user *)arg;
 974         int r;
 975
 976         switch (ioctl) {
 977         case KVM_CREATE_VCPU:
 978                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
 979                 if (r < 0)
 980                         goto out;
 981                 break;
 982         case KVM_SET_USER_MEMORY_REGION: {
 983                 struct kvm_userspace_memory_region kvm_userspace_mem;
 984
 985                 r = -EFAULT;
 986                 if (copy_from_user(&kvm_userspace_mem, argp,
 987                                                 sizeof kvm_userspace_mem))
 988                         goto out;
 989
 990                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
 991                 if (r)
 992                         goto out;
 993                 break;
 994         }
 995         case KVM_GET_DIRTY_LOG: {
 996                 struct kvm_dirty_log log;
 997
 998                 r = -EFAULT;
 999                 if (copy_from_user(&log, argp, sizeof log))
1000                         goto out;
1001                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1002                 if (r)
1003                         goto out;
1004                 break;
1005         }
1006         default:
1007                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1008         }
1009 out:
1010         return r;
1011 }
1012
1013 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1014                                   unsigned long address,
1015                                   int *type)
1016 {
1017         struct kvm *kvm = vma->vm_file->private_data;
1018         unsigned long pgoff;
1019         struct page *page;
1020
1021         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1022         if (!kvm_is_visible_gfn(kvm, pgoff))
1023                 return NOPAGE_SIGBUS;
1024         /* current->mm->mmap_sem is already held so call lockless version */
1025         page = __gfn_to_page(kvm, pgoff);
1026         if (is_error_page(page)) {
1027                 kvm_release_page_clean(page);
1028                 return NOPAGE_SIGBUS;
1029         }
1030         if (type != NULL)
1031                 *type = VM_FAULT_MINOR;
1032
1033         return page;
1034 }
1035
1036 static struct vm_operations_struct kvm_vm_vm_ops = {
1037         .nopage = kvm_vm_nopage,
1038 };
1039
1040 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1041 {
1042         vma->vm_ops = &kvm_vm_vm_ops;
1043         return 0;
1044 }
1045
1046 static struct file_operations kvm_vm_fops = {
1047         .release        = kvm_vm_release,
1048         .unlocked_ioctl = kvm_vm_ioctl,
1049         .compat_ioctl   = kvm_vm_ioctl,
1050         .mmap           = kvm_vm_mmap,
1051 };
1052
1053 static int kvm_dev_ioctl_create_vm(void)
1054 {
1055         int fd, r;
1056         struct inode *inode;
1057         struct file *file;
1058         struct kvm *kvm;
1059
1060         kvm = kvm_create_vm();
1061         if (IS_ERR(kvm))
1062                 return PTR_ERR(kvm);
1063         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1064         if (r) {
1065                 kvm_destroy_vm(kvm);
1066                 return r;
1067         }
1068
1069         kvm->filp = file;
1070
1071         return fd;
1072 }
1073
1074 static long kvm_dev_ioctl(struct file *filp,
1075                           unsigned int ioctl, unsigned long arg)
1076 {
1077         void __user *argp = (void __user *)arg;
1078         long r = -EINVAL;
1079
1080         switch (ioctl) {
1081         case KVM_GET_API_VERSION:
1082                 r = -EINVAL;
1083                 if (arg)
1084                         goto out;
1085                 r = KVM_API_VERSION;
1086                 break;
1087         case KVM_CREATE_VM:
1088                 r = -EINVAL;
1089                 if (arg)
1090                         goto out;
1091                 r = kvm_dev_ioctl_create_vm();
1092                 break;
1093         case KVM_CHECK_EXTENSION:
1094                 r = kvm_dev_ioctl_check_extension((long)argp);
1095                 break;
1096         case KVM_GET_VCPU_MMAP_SIZE:
1097                 r = -EINVAL;
1098                 if (arg)
1099                         goto out;
1100                 r = 2 * PAGE_SIZE;
1101                 break;
1102         default:
1103                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1104         }
1105 out:
1106         return r;
1107 }
1108
1109 static struct file_operations kvm_chardev_ops = {
1110         .unlocked_ioctl = kvm_dev_ioctl,
1111         .compat_ioctl   = kvm_dev_ioctl,
1112 };
1113
1114 static struct miscdevice kvm_dev = {
1115         KVM_MINOR,
1116         "kvm",
1117         &kvm_chardev_ops,
1118 };
1119
1120 static void hardware_enable(void *junk)
1121 {
1122         int cpu = raw_smp_processor_id();
1123
1124         if (cpu_isset(cpu, cpus_hardware_enabled))
1125                 return;
1126         cpu_set(cpu, cpus_hardware_enabled);
1127         kvm_arch_hardware_enable(NULL);
1128 }
1129
1130 static void hardware_disable(void *junk)
1131 {
1132         int cpu = raw_smp_processor_id();
1133
1134         if (!cpu_isset(cpu, cpus_hardware_enabled))
1135                 return;
1136         cpu_clear(cpu, cpus_hardware_enabled);
1137         decache_vcpus_on_cpu(cpu);
1138         kvm_arch_hardware_disable(NULL);
1139 }
1140
1141 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1142                            void *v)
1143 {
1144         int cpu = (long)v;
1145
1146         val &= ~CPU_TASKS_FROZEN;
1147         switch (val) {
1148         case CPU_DYING:
1149                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1150                        cpu);
1151                 hardware_disable(NULL);
1152                 break;
1153         case CPU_UP_CANCELED:
1154                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1155                        cpu);
1156                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1157                 break;
1158         case CPU_ONLINE:
1159                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1160                        cpu);
1161                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1162                 break;
1163         }
1164         return NOTIFY_OK;
1165 }
1166
1167 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1168                       void *v)
1169 {
1170         if (val == SYS_RESTART) {
1171                 /*
1172                  * Some (well, at least mine) BIOSes hang on reboot if
1173                  * in vmx root mode.
1174                  */
1175                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1176                 on_each_cpu(hardware_disable, NULL, 0, 1);
1177         }
1178         return NOTIFY_OK;
1179 }
1180
1181 static struct notifier_block kvm_reboot_notifier = {
1182         .notifier_call = kvm_reboot,
1183         .priority = 0,
1184 };
1185
1186 void kvm_io_bus_init(struct kvm_io_bus *bus)
1187 {
1188         memset(bus, 0, sizeof(*bus));
1189 }
1190
1191 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1192 {
1193         int i;
1194
1195         for (i = 0; i < bus->dev_count; i++) {
1196                 struct kvm_io_device *pos = bus->devs[i];
1197
1198                 kvm_iodevice_destructor(pos);
1199         }
1200 }
1201
1202 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1203 {
1204         int i;
1205
1206         for (i = 0; i < bus->dev_count; i++) {
1207                 struct kvm_io_device *pos = bus->devs[i];
1208
1209                 if (pos->in_range(pos, addr))
1210                         return pos;
1211         }
1212
1213         return NULL;
1214 }
1215
1216 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1217 {
1218         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1219
1220         bus->devs[bus->dev_count++] = dev;
1221 }
1222
1223 static struct notifier_block kvm_cpu_notifier = {
1224         .notifier_call = kvm_cpu_hotplug,
1225         .priority = 20, /* must be > scheduler priority */
1226 };
1227
1228 static u64 vm_stat_get(void *_offset)
1229 {
1230         unsigned offset = (long)_offset;
1231         u64 total = 0;
1232         struct kvm *kvm;
1233
1234         spin_lock(&kvm_lock);
1235         list_for_each_entry(kvm, &vm_list, vm_list)
1236                 total += *(u32 *)((void *)kvm + offset);
1237         spin_unlock(&kvm_lock);
1238         return total;
1239 }
1240
1241 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1242
1243 static u64 vcpu_stat_get(void *_offset)
1244 {
1245         unsigned offset = (long)_offset;
1246         u64 total = 0;
1247         struct kvm *kvm;
1248         struct kvm_vcpu *vcpu;
1249         int i;
1250
1251         spin_lock(&kvm_lock);
1252         list_for_each_entry(kvm, &vm_list, vm_list)
1253                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1254                         vcpu = kvm->vcpus[i];
1255                         if (vcpu)
1256                                 total += *(u32 *)((void *)vcpu + offset);
1257                 }
1258         spin_unlock(&kvm_lock);
1259         return total;
1260 }
1261
1262 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1263
1264 static struct file_operations *stat_fops[] = {
1265         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1266         [KVM_STAT_VM]   = &vm_stat_fops,
1267 };
1268
1269 static void kvm_init_debug(void)
1270 {
1271         struct kvm_stats_debugfs_item *p;
1272
1273         debugfs_dir = debugfs_create_dir("kvm", NULL);
1274         for (p = debugfs_entries; p->name; ++p)
1275                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1276                                                 (void *)(long)p->offset,
1277                                                 stat_fops[p->kind]);
1278 }
1279
1280 static void kvm_exit_debug(void)
1281 {
1282         struct kvm_stats_debugfs_item *p;
1283
1284         for (p = debugfs_entries; p->name; ++p)
1285                 debugfs_remove(p->dentry);
1286         debugfs_remove(debugfs_dir);
1287 }
1288
1289 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1290 {
1291         hardware_disable(NULL);
1292         return 0;
1293 }
1294
1295 static int kvm_resume(struct sys_device *dev)
1296 {
1297         hardware_enable(NULL);
1298         return 0;
1299 }
1300
1301 static struct sysdev_class kvm_sysdev_class = {
1302         .name = "kvm",
1303         .suspend = kvm_suspend,
1304         .resume = kvm_resume,
1305 };
1306
1307 static struct sys_device kvm_sysdev = {
1308         .id = 0,
1309         .cls = &kvm_sysdev_class,
1310 };
1311
1312 struct page *bad_page;
1313
1314 static inline
1315 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1316 {
1317         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1318 }
1319
1320 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1321 {
1322         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1323
1324         kvm_arch_vcpu_load(vcpu, cpu);
1325 }
1326
1327 static void kvm_sched_out(struct preempt_notifier *pn,
1328                           struct task_struct *next)
1329 {
1330         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1331
1332         kvm_arch_vcpu_put(vcpu);
1333 }
1334
1335 int kvm_init(void *opaque, unsigned int vcpu_size,
1336                   struct module *module)
1337 {
1338         int r;
1339         int cpu;
1340
1341         kvm_init_debug();
1342
1343         r = kvm_arch_init(opaque);
1344         if (r)
1345                 goto out4;
1346
1347         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1348
1349         if (bad_page == NULL) {
1350                 r = -ENOMEM;
1351                 goto out;
1352         }
1353
1354         r = kvm_arch_hardware_setup();
1355         if (r < 0)
1356                 goto out;
1357
1358         for_each_online_cpu(cpu) {
1359                 smp_call_function_single(cpu,
1360                                 kvm_arch_check_processor_compat,
1361                                 &r, 0, 1);
1362                 if (r < 0)
1363                         goto out_free_0;
1364         }
1365
1366         on_each_cpu(hardware_enable, NULL, 0, 1);
1367         r = register_cpu_notifier(&kvm_cpu_notifier);
1368         if (r)
1369                 goto out_free_1;
1370         register_reboot_notifier(&kvm_reboot_notifier);
1371
1372         r = sysdev_class_register(&kvm_sysdev_class);
1373         if (r)
1374                 goto out_free_2;
1375
1376         r = sysdev_register(&kvm_sysdev);
1377         if (r)
1378                 goto out_free_3;
1379
1380         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1381         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1382                                            __alignof__(struct kvm_vcpu),
1383                                            0, NULL);
1384         if (!kvm_vcpu_cache) {
1385                 r = -ENOMEM;
1386                 goto out_free_4;
1387         }
1388
1389         kvm_chardev_ops.owner = module;
1390
1391         r = misc_register(&kvm_dev);
1392         if (r) {
1393                 printk(KERN_ERR "kvm: misc device register failed\n");
1394                 goto out_free;
1395         }
1396
1397         kvm_preempt_ops.sched_in = kvm_sched_in;
1398         kvm_preempt_ops.sched_out = kvm_sched_out;
1399
1400         return 0;
1401
1402 out_free:
1403         kmem_cache_destroy(kvm_vcpu_cache);
1404 out_free_4:
1405         sysdev_unregister(&kvm_sysdev);
1406 out_free_3:
1407         sysdev_class_unregister(&kvm_sysdev_class);
1408 out_free_2:
1409         unregister_reboot_notifier(&kvm_reboot_notifier);
1410         unregister_cpu_notifier(&kvm_cpu_notifier);
1411 out_free_1:
1412         on_each_cpu(hardware_disable, NULL, 0, 1);
1413 out_free_0:
1414         kvm_arch_hardware_unsetup();
1415 out:
1416         kvm_arch_exit();
1417         kvm_exit_debug();
1418 out4:
1419         return r;
1420 }
1421 EXPORT_SYMBOL_GPL(kvm_init);
1422
1423 void kvm_exit(void)
1424 {
1425         misc_deregister(&kvm_dev);
1426         kmem_cache_destroy(kvm_vcpu_cache);
1427         sysdev_unregister(&kvm_sysdev);
1428         sysdev_class_unregister(&kvm_sysdev_class);
1429         unregister_reboot_notifier(&kvm_reboot_notifier);
1430         unregister_cpu_notifier(&kvm_cpu_notifier);
1431         on_each_cpu(hardware_disable, NULL, 0, 1);
1432         kvm_arch_hardware_unsetup();
1433         kvm_arch_exit();
1434         kvm_exit_debug();
1435         __free_page(bad_page);
1436 }
1437 EXPORT_SYMBOL_GPL(kvm_exit);