git.oblomov.eu 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
  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         if (cpus_empty(cpus))
 119                 return;
 120         ++kvm->stat.remote_tlb_flush;
 121         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 122 }
 123
 124 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 125 {
 126         struct page *page;
 127         int r;
 128
 129         mutex_init(&vcpu->mutex);
 130         vcpu->cpu = -1;
 131         vcpu->kvm = kvm;
 132         vcpu->vcpu_id = id;
 133         init_waitqueue_head(&vcpu->wq);
 134
 135         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 136         if (!page) {
 137                 r = -ENOMEM;
 138                 goto fail;
 139         }
 140         vcpu->run = page_address(page);
 141
 142         r = kvm_arch_vcpu_init(vcpu);
 143         if (r < 0)
 144                 goto fail_free_run;
 145         return 0;
 146
 147 fail_free_run:
 148         free_page((unsigned long)vcpu->run);
 149 fail:
 150         return r;
 151 }
 152 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 153
 154 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 155 {
 156         kvm_arch_vcpu_uninit(vcpu);
 157         free_page((unsigned long)vcpu->run);
 158 }
 159 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 160
 161 static struct kvm *kvm_create_vm(void)
 162 {
 163         struct kvm *kvm = kvm_arch_create_vm();
 164
 165         if (IS_ERR(kvm))
 166                 goto out;
 167
 168         kvm->mm = current->mm;
 169         atomic_inc(&kvm->mm->mm_count);
 170         kvm_io_bus_init(&kvm->pio_bus);
 171         mutex_init(&kvm->lock);
 172         kvm_io_bus_init(&kvm->mmio_bus);
 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 kvm->lock.
 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         mutex_lock(&kvm->lock);
 344         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 345         mutex_unlock(&kvm->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 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 412 {
 413         int i;
 414         struct kvm_mem_alias *alias;
 415
 416         for (i = 0; i < kvm->naliases; ++i) {
 417                 alias = &kvm->aliases[i];
 418                 if (gfn >= alias->base_gfn
 419                     && gfn < alias->base_gfn + alias->npages)
 420                         return alias->target_gfn + gfn - alias->base_gfn;
 421         }
 422         return gfn;
 423 }
 424
 425 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 426 {
 427         int i;
 428
 429         for (i = 0; i < kvm->nmemslots; ++i) {
 430                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 431
 432                 if (gfn >= memslot->base_gfn
 433                     && gfn < memslot->base_gfn + memslot->npages)
 434                         return memslot;
 435         }
 436         return NULL;
 437 }
 438
 439 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 440 {
 441         gfn = unalias_gfn(kvm, gfn);
 442         return __gfn_to_memslot(kvm, gfn);
 443 }
 444
 445 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 446 {
 447         int i;
 448
 449         gfn = unalias_gfn(kvm, gfn);
 450         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 451                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 452
 453                 if (gfn >= memslot->base_gfn
 454                     && gfn < memslot->base_gfn + memslot->npages)
 455                         return 1;
 456         }
 457         return 0;
 458 }
 459 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 460
 461 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 462 {
 463         struct kvm_memory_slot *slot;
 464
 465         gfn = unalias_gfn(kvm, gfn);
 466         slot = __gfn_to_memslot(kvm, gfn);
 467         if (!slot)
 468                 return bad_hva();
 469         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 470 }
 471
 472 /*
 473  * Requires current->mm->mmap_sem to be held
 474  */
 475 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 476 {
 477         struct page *page[1];
 478         unsigned long addr;
 479         int npages;
 480
 481         might_sleep();
 482
 483         addr = gfn_to_hva(kvm, gfn);
 484         if (kvm_is_error_hva(addr)) {
 485                 get_page(bad_page);
 486                 return bad_page;
 487         }
 488
 489         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 490                                 NULL);
 491
 492         if (npages != 1) {
 493                 get_page(bad_page);
 494                 return bad_page;
 495         }
 496
 497         return page[0];
 498 }
 499
 500 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 501 {
 502         struct page *page;
 503
 504         down_read(&current->mm->mmap_sem);
 505         page = __gfn_to_page(kvm, gfn);
 506         up_read(&current->mm->mmap_sem);
 507
 508         return page;
 509 }
 510
 511 EXPORT_SYMBOL_GPL(gfn_to_page);
 512
 513 void kvm_release_page_clean(struct page *page)
 514 {
 515         put_page(page);
 516 }
 517 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 518
 519 void kvm_release_page_dirty(struct page *page)
 520 {
 521         if (!PageReserved(page))
 522                 SetPageDirty(page);
 523         put_page(page);
 524 }
 525 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 526
 527 static int next_segment(unsigned long len, int offset)
 528 {
 529         if (len > PAGE_SIZE - offset)
 530                 return PAGE_SIZE - offset;
 531         else
 532                 return len;
 533 }
 534
 535 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 536                         int len)
 537 {
 538         int r;
 539         unsigned long addr;
 540
 541         addr = gfn_to_hva(kvm, gfn);
 542         if (kvm_is_error_hva(addr))
 543                 return -EFAULT;
 544         r = copy_from_user(data, (void __user *)addr + offset, len);
 545         if (r)
 546                 return -EFAULT;
 547         return 0;
 548 }
 549 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 550
 551 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 552 {
 553         gfn_t gfn = gpa >> PAGE_SHIFT;
 554         int seg;
 555         int offset = offset_in_page(gpa);
 556         int ret;
 557
 558         while ((seg = next_segment(len, offset)) != 0) {
 559                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 560                 if (ret < 0)
 561                         return ret;
 562                 offset = 0;
 563                 len -= seg;
 564                 data += seg;
 565                 ++gfn;
 566         }
 567         return 0;
 568 }
 569 EXPORT_SYMBOL_GPL(kvm_read_guest);
 570
 571 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 572                          int offset, int len)
 573 {
 574         int r;
 575         unsigned long addr;
 576
 577         addr = gfn_to_hva(kvm, gfn);
 578         if (kvm_is_error_hva(addr))
 579                 return -EFAULT;
 580         r = copy_to_user((void __user *)addr + offset, data, len);
 581         if (r)
 582                 return -EFAULT;
 583         mark_page_dirty(kvm, gfn);
 584         return 0;
 585 }
 586 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 587
 588 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 589                     unsigned long len)
 590 {
 591         gfn_t gfn = gpa >> PAGE_SHIFT;
 592         int seg;
 593         int offset = offset_in_page(gpa);
 594         int ret;
 595
 596         while ((seg = next_segment(len, offset)) != 0) {
 597                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 598                 if (ret < 0)
 599                         return ret;
 600                 offset = 0;
 601                 len -= seg;
 602                 data += seg;
 603                 ++gfn;
 604         }
 605         return 0;
 606 }
 607
 608 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 609 {
 610         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 611 }
 612 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 613
 614 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 615 {
 616         gfn_t gfn = gpa >> PAGE_SHIFT;
 617         int seg;
 618         int offset = offset_in_page(gpa);
 619         int ret;
 620
 621         while ((seg = next_segment(len, offset)) != 0) {
 622                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 623                 if (ret < 0)
 624                         return ret;
 625                 offset = 0;
 626                 len -= seg;
 627                 ++gfn;
 628         }
 629         return 0;
 630 }
 631 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 632
 633 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 634 {
 635         struct kvm_memory_slot *memslot;
 636
 637         gfn = unalias_gfn(kvm, gfn);
 638         memslot = __gfn_to_memslot(kvm, gfn);
 639         if (memslot && memslot->dirty_bitmap) {
 640                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 641
 642                 /* avoid RMW */
 643                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 644                         set_bit(rel_gfn, memslot->dirty_bitmap);
 645         }
 646 }
 647
 648 /*
 649  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 650  */
 651 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 652 {
 653         DECLARE_WAITQUEUE(wait, current);
 654
 655         add_wait_queue(&vcpu->wq, &wait);
 656
 657         /*
 658          * We will block until either an interrupt or a signal wakes us up
 659          */
 660         while (!kvm_cpu_has_interrupt(vcpu)
 661                && !signal_pending(current)
 662                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
 663                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
 664                 set_current_state(TASK_INTERRUPTIBLE);
 665                 vcpu_put(vcpu);
 666                 schedule();
 667                 vcpu_load(vcpu);
 668         }
 669
 670         __set_current_state(TASK_RUNNING);
 671         remove_wait_queue(&vcpu->wq, &wait);
 672 }
 673
 674 void kvm_resched(struct kvm_vcpu *vcpu)
 675 {
 676         if (!need_resched())
 677                 return;
 678         cond_resched();
 679 }
 680 EXPORT_SYMBOL_GPL(kvm_resched);
 681
 682 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
 683                                     struct kvm_interrupt *irq)
 684 {
 685         if (irq->irq < 0 || irq->irq >= 256)
 686                 return -EINVAL;
 687         if (irqchip_in_kernel(vcpu->kvm))
 688                 return -ENXIO;
 689         vcpu_load(vcpu);
 690
 691         set_bit(irq->irq, vcpu->irq_pending);
 692         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
 693
 694         vcpu_put(vcpu);
 695
 696         return 0;
 697 }
 698
 699 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
 700                                     unsigned long address,
 701                                     int *type)
 702 {
 703         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 704         unsigned long pgoff;
 705         struct page *page;
 706
 707         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 708         if (pgoff == 0)
 709                 page = virt_to_page(vcpu->run);
 710         else if (pgoff == KVM_PIO_PAGE_OFFSET)
 711                 page = virt_to_page(vcpu->pio_data);
 712         else
 713                 return NOPAGE_SIGBUS;
 714         get_page(page);
 715         if (type != NULL)
 716                 *type = VM_FAULT_MINOR;
 717
 718         return page;
 719 }
 720
 721 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 722         .nopage = kvm_vcpu_nopage,
 723 };
 724
 725 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 726 {
 727         vma->vm_ops = &kvm_vcpu_vm_ops;
 728         return 0;
 729 }
 730
 731 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 732 {
 733         struct kvm_vcpu *vcpu = filp->private_data;
 734
 735         fput(vcpu->kvm->filp);
 736         return 0;
 737 }
 738
 739 static struct file_operations kvm_vcpu_fops = {
 740         .release        = kvm_vcpu_release,
 741         .unlocked_ioctl = kvm_vcpu_ioctl,
 742         .compat_ioctl   = kvm_vcpu_ioctl,
 743         .mmap           = kvm_vcpu_mmap,
 744 };
 745
 746 /*
 747  * Allocates an inode for the vcpu.
 748  */
 749 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 750 {
 751         int fd, r;
 752         struct inode *inode;
 753         struct file *file;
 754
 755         r = anon_inode_getfd(&fd, &inode, &file,
 756                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 757         if (r)
 758                 return r;
 759         atomic_inc(&vcpu->kvm->filp->f_count);
 760         return fd;
 761 }
 762
 763 /*
 764  * Creates some virtual cpus.  Good luck creating more than one.
 765  */
 766 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 767 {
 768         int r;
 769         struct kvm_vcpu *vcpu;
 770
 771         if (!valid_vcpu(n))
 772                 return -EINVAL;
 773
 774         vcpu = kvm_arch_vcpu_create(kvm, n);
 775         if (IS_ERR(vcpu))
 776                 return PTR_ERR(vcpu);
 777
 778         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 779
 780         r = kvm_arch_vcpu_setup(vcpu);
 781         if (r)
 782                 goto vcpu_destroy;
 783
 784         mutex_lock(&kvm->lock);
 785         if (kvm->vcpus[n]) {
 786                 r = -EEXIST;
 787                 mutex_unlock(&kvm->lock);
 788                 goto vcpu_destroy;
 789         }
 790         kvm->vcpus[n] = vcpu;
 791         mutex_unlock(&kvm->lock);
 792
 793         /* Now it's all set up, let userspace reach it */
 794         r = create_vcpu_fd(vcpu);
 795         if (r < 0)
 796                 goto unlink;
 797         return r;
 798
 799 unlink:
 800         mutex_lock(&kvm->lock);
 801         kvm->vcpus[n] = NULL;
 802         mutex_unlock(&kvm->lock);
 803 vcpu_destroy:
 804         kvm_arch_vcpu_destroy(vcpu);
 805         return r;
 806 }
 807
 808 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 809 {
 810         if (sigset) {
 811                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 812                 vcpu->sigset_active = 1;
 813                 vcpu->sigset = *sigset;
 814         } else
 815                 vcpu->sigset_active = 0;
 816         return 0;
 817 }
 818
 819 static long kvm_vcpu_ioctl(struct file *filp,
 820                            unsigned int ioctl, unsigned long arg)
 821 {
 822         struct kvm_vcpu *vcpu = filp->private_data;
 823         void __user *argp = (void __user *)arg;
 824         int r;
 825
 826         if (vcpu->kvm->mm != current->mm)
 827                 return -EIO;
 828         switch (ioctl) {
 829         case KVM_RUN:
 830                 r = -EINVAL;
 831                 if (arg)
 832                         goto out;
 833                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 834                 break;
 835         case KVM_GET_REGS: {
 836                 struct kvm_regs kvm_regs;
 837
 838                 memset(&kvm_regs, 0, sizeof kvm_regs);
 839                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 840                 if (r)
 841                         goto out;
 842                 r = -EFAULT;
 843                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 844                         goto out;
 845                 r = 0;
 846                 break;
 847         }
 848         case KVM_SET_REGS: {
 849                 struct kvm_regs kvm_regs;
 850
 851                 r = -EFAULT;
 852                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 853                         goto out;
 854                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 855                 if (r)
 856                         goto out;
 857                 r = 0;
 858                 break;
 859         }
 860         case KVM_GET_SREGS: {
 861                 struct kvm_sregs kvm_sregs;
 862
 863                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 864                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 865                 if (r)
 866                         goto out;
 867                 r = -EFAULT;
 868                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 869                         goto out;
 870                 r = 0;
 871                 break;
 872         }
 873         case KVM_SET_SREGS: {
 874                 struct kvm_sregs kvm_sregs;
 875
 876                 r = -EFAULT;
 877                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 878                         goto out;
 879                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 880                 if (r)
 881                         goto out;
 882                 r = 0;
 883                 break;
 884         }
 885         case KVM_TRANSLATE: {
 886                 struct kvm_translation tr;
 887
 888                 r = -EFAULT;
 889                 if (copy_from_user(&tr, argp, sizeof tr))
 890                         goto out;
 891                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 892                 if (r)
 893                         goto out;
 894                 r = -EFAULT;
 895                 if (copy_to_user(argp, &tr, sizeof tr))
 896                         goto out;
 897                 r = 0;
 898                 break;
 899         }
 900         case KVM_INTERRUPT: {
 901                 struct kvm_interrupt irq;
 902
 903                 r = -EFAULT;
 904                 if (copy_from_user(&irq, argp, sizeof irq))
 905                         goto out;
 906                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 907                 if (r)
 908                         goto out;
 909                 r = 0;
 910                 break;
 911         }
 912         case KVM_DEBUG_GUEST: {
 913                 struct kvm_debug_guest dbg;
 914
 915                 r = -EFAULT;
 916                 if (copy_from_user(&dbg, argp, sizeof dbg))
 917                         goto out;
 918                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 919                 if (r)
 920                         goto out;
 921                 r = 0;
 922                 break;
 923         }
 924         case KVM_SET_SIGNAL_MASK: {
 925                 struct kvm_signal_mask __user *sigmask_arg = argp;
 926                 struct kvm_signal_mask kvm_sigmask;
 927                 sigset_t sigset, *p;
 928
 929                 p = NULL;
 930                 if (argp) {
 931                         r = -EFAULT;
 932                         if (copy_from_user(&kvm_sigmask, argp,
 933                                            sizeof kvm_sigmask))
 934                                 goto out;
 935                         r = -EINVAL;
 936                         if (kvm_sigmask.len != sizeof sigset)
 937                                 goto out;
 938                         r = -EFAULT;
 939                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 940                                            sizeof sigset))
 941                                 goto out;
 942                         p = &sigset;
 943                 }
 944                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 945                 break;
 946         }
 947         case KVM_GET_FPU: {
 948                 struct kvm_fpu fpu;
 949
 950                 memset(&fpu, 0, sizeof fpu);
 951                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 952                 if (r)
 953                         goto out;
 954                 r = -EFAULT;
 955                 if (copy_to_user(argp, &fpu, sizeof fpu))
 956                         goto out;
 957                 r = 0;
 958                 break;
 959         }
 960         case KVM_SET_FPU: {
 961                 struct kvm_fpu fpu;
 962
 963                 r = -EFAULT;
 964                 if (copy_from_user(&fpu, argp, sizeof fpu))
 965                         goto out;
 966                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
 967                 if (r)
 968                         goto out;
 969                 r = 0;
 970                 break;
 971         }
 972         default:
 973                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
 974         }
 975 out:
 976         return r;
 977 }
 978
 979 static long kvm_vm_ioctl(struct file *filp,
 980                            unsigned int ioctl, unsigned long arg)
 981 {
 982         struct kvm *kvm = filp->private_data;
 983         void __user *argp = (void __user *)arg;
 984         int r;
 985
 986         if (kvm->mm != current->mm)
 987                 return -EIO;
 988         switch (ioctl) {
 989         case KVM_CREATE_VCPU:
 990                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
 991                 if (r < 0)
 992                         goto out;
 993                 break;
 994         case KVM_SET_USER_MEMORY_REGION: {
 995                 struct kvm_userspace_memory_region kvm_userspace_mem;
 996
 997                 r = -EFAULT;
 998                 if (copy_from_user(&kvm_userspace_mem, argp,
 999                                                 sizeof kvm_userspace_mem))
1000                         goto out;
1001
1002                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1003                 if (r)
1004                         goto out;
1005                 break;
1006         }
1007         case KVM_GET_DIRTY_LOG: {
1008                 struct kvm_dirty_log log;
1009
1010                 r = -EFAULT;
1011                 if (copy_from_user(&log, argp, sizeof log))
1012                         goto out;
1013                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1014                 if (r)
1015                         goto out;
1016                 break;
1017         }
1018         default:
1019                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1020         }
1021 out:
1022         return r;
1023 }
1024
1025 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1026                                   unsigned long address,
1027                                   int *type)
1028 {
1029         struct kvm *kvm = vma->vm_file->private_data;
1030         unsigned long pgoff;
1031         struct page *page;
1032
1033         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1034         if (!kvm_is_visible_gfn(kvm, pgoff))
1035                 return NOPAGE_SIGBUS;
1036         /* current->mm->mmap_sem is already held so call lockless version */
1037         page = __gfn_to_page(kvm, pgoff);
1038         if (is_error_page(page)) {
1039                 kvm_release_page_clean(page);
1040                 return NOPAGE_SIGBUS;
1041         }
1042         if (type != NULL)
1043                 *type = VM_FAULT_MINOR;
1044
1045         return page;
1046 }
1047
1048 static struct vm_operations_struct kvm_vm_vm_ops = {
1049         .nopage = kvm_vm_nopage,
1050 };
1051
1052 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1053 {
1054         vma->vm_ops = &kvm_vm_vm_ops;
1055         return 0;
1056 }
1057
1058 static struct file_operations kvm_vm_fops = {
1059         .release        = kvm_vm_release,
1060         .unlocked_ioctl = kvm_vm_ioctl,
1061         .compat_ioctl   = kvm_vm_ioctl,
1062         .mmap           = kvm_vm_mmap,
1063 };
1064
1065 static int kvm_dev_ioctl_create_vm(void)
1066 {
1067         int fd, r;
1068         struct inode *inode;
1069         struct file *file;
1070         struct kvm *kvm;
1071
1072         kvm = kvm_create_vm();
1073         if (IS_ERR(kvm))
1074                 return PTR_ERR(kvm);
1075         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1076         if (r) {
1077                 kvm_destroy_vm(kvm);
1078                 return r;
1079         }
1080
1081         kvm->filp = file;
1082
1083         return fd;
1084 }
1085
1086 static long kvm_dev_ioctl(struct file *filp,
1087                           unsigned int ioctl, unsigned long arg)
1088 {
1089         void __user *argp = (void __user *)arg;
1090         long r = -EINVAL;
1091
1092         switch (ioctl) {
1093         case KVM_GET_API_VERSION:
1094                 r = -EINVAL;
1095                 if (arg)
1096                         goto out;
1097                 r = KVM_API_VERSION;
1098                 break;
1099         case KVM_CREATE_VM:
1100                 r = -EINVAL;
1101                 if (arg)
1102                         goto out;
1103                 r = kvm_dev_ioctl_create_vm();
1104                 break;
1105         case KVM_CHECK_EXTENSION:
1106                 r = kvm_dev_ioctl_check_extension((long)argp);
1107                 break;
1108         case KVM_GET_VCPU_MMAP_SIZE:
1109                 r = -EINVAL;
1110                 if (arg)
1111                         goto out;
1112                 r = 2 * PAGE_SIZE;
1113                 break;
1114         default:
1115                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1116         }
1117 out:
1118         return r;
1119 }
1120
1121 static struct file_operations kvm_chardev_ops = {
1122         .unlocked_ioctl = kvm_dev_ioctl,
1123         .compat_ioctl   = kvm_dev_ioctl,
1124 };
1125
1126 static struct miscdevice kvm_dev = {
1127         KVM_MINOR,
1128         "kvm",
1129         &kvm_chardev_ops,
1130 };
1131
1132 static void hardware_enable(void *junk)
1133 {
1134         int cpu = raw_smp_processor_id();
1135
1136         if (cpu_isset(cpu, cpus_hardware_enabled))
1137                 return;
1138         cpu_set(cpu, cpus_hardware_enabled);
1139         kvm_arch_hardware_enable(NULL);
1140 }
1141
1142 static void hardware_disable(void *junk)
1143 {
1144         int cpu = raw_smp_processor_id();
1145
1146         if (!cpu_isset(cpu, cpus_hardware_enabled))
1147                 return;
1148         cpu_clear(cpu, cpus_hardware_enabled);
1149         decache_vcpus_on_cpu(cpu);
1150         kvm_arch_hardware_disable(NULL);
1151 }
1152
1153 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1154                            void *v)
1155 {
1156         int cpu = (long)v;
1157
1158         val &= ~CPU_TASKS_FROZEN;
1159         switch (val) {
1160         case CPU_DYING:
1161                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1162                        cpu);
1163                 hardware_disable(NULL);
1164                 break;
1165         case CPU_UP_CANCELED:
1166                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1167                        cpu);
1168                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1169                 break;
1170         case CPU_ONLINE:
1171                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1172                        cpu);
1173                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1174                 break;
1175         }
1176         return NOTIFY_OK;
1177 }
1178
1179 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1180                       void *v)
1181 {
1182         if (val == SYS_RESTART) {
1183                 /*
1184                  * Some (well, at least mine) BIOSes hang on reboot if
1185                  * in vmx root mode.
1186                  */
1187                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1188                 on_each_cpu(hardware_disable, NULL, 0, 1);
1189         }
1190         return NOTIFY_OK;
1191 }
1192
1193 static struct notifier_block kvm_reboot_notifier = {
1194         .notifier_call = kvm_reboot,
1195         .priority = 0,
1196 };
1197
1198 void kvm_io_bus_init(struct kvm_io_bus *bus)
1199 {
1200         memset(bus, 0, sizeof(*bus));
1201 }
1202
1203 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1204 {
1205         int i;
1206
1207         for (i = 0; i < bus->dev_count; i++) {
1208                 struct kvm_io_device *pos = bus->devs[i];
1209
1210                 kvm_iodevice_destructor(pos);
1211         }
1212 }
1213
1214 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1215 {
1216         int i;
1217
1218         for (i = 0; i < bus->dev_count; i++) {
1219                 struct kvm_io_device *pos = bus->devs[i];
1220
1221                 if (pos->in_range(pos, addr))
1222                         return pos;
1223         }
1224
1225         return NULL;
1226 }
1227
1228 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1229 {
1230         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1231
1232         bus->devs[bus->dev_count++] = dev;
1233 }
1234
1235 static struct notifier_block kvm_cpu_notifier = {
1236         .notifier_call = kvm_cpu_hotplug,
1237         .priority = 20, /* must be > scheduler priority */
1238 };
1239
1240 static u64 vm_stat_get(void *_offset)
1241 {
1242         unsigned offset = (long)_offset;
1243         u64 total = 0;
1244         struct kvm *kvm;
1245
1246         spin_lock(&kvm_lock);
1247         list_for_each_entry(kvm, &vm_list, vm_list)
1248                 total += *(u32 *)((void *)kvm + offset);
1249         spin_unlock(&kvm_lock);
1250         return total;
1251 }
1252
1253 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1254
1255 static u64 vcpu_stat_get(void *_offset)
1256 {
1257         unsigned offset = (long)_offset;
1258         u64 total = 0;
1259         struct kvm *kvm;
1260         struct kvm_vcpu *vcpu;
1261         int i;
1262
1263         spin_lock(&kvm_lock);
1264         list_for_each_entry(kvm, &vm_list, vm_list)
1265                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1266                         vcpu = kvm->vcpus[i];
1267                         if (vcpu)
1268                                 total += *(u32 *)((void *)vcpu + offset);
1269                 }
1270         spin_unlock(&kvm_lock);
1271         return total;
1272 }
1273
1274 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1275
1276 static struct file_operations *stat_fops[] = {
1277         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1278         [KVM_STAT_VM]   = &vm_stat_fops,
1279 };
1280
1281 static void kvm_init_debug(void)
1282 {
1283         struct kvm_stats_debugfs_item *p;
1284
1285         debugfs_dir = debugfs_create_dir("kvm", NULL);
1286         for (p = debugfs_entries; p->name; ++p)
1287                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1288                                                 (void *)(long)p->offset,
1289                                                 stat_fops[p->kind]);
1290 }
1291
1292 static void kvm_exit_debug(void)
1293 {
1294         struct kvm_stats_debugfs_item *p;
1295
1296         for (p = debugfs_entries; p->name; ++p)
1297                 debugfs_remove(p->dentry);
1298         debugfs_remove(debugfs_dir);
1299 }
1300
1301 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1302 {
1303         hardware_disable(NULL);
1304         return 0;
1305 }
1306
1307 static int kvm_resume(struct sys_device *dev)
1308 {
1309         hardware_enable(NULL);
1310         return 0;
1311 }
1312
1313 static struct sysdev_class kvm_sysdev_class = {
1314         .name = "kvm",
1315         .suspend = kvm_suspend,
1316         .resume = kvm_resume,
1317 };
1318
1319 static struct sys_device kvm_sysdev = {
1320         .id = 0,
1321         .cls = &kvm_sysdev_class,
1322 };
1323
1324 struct page *bad_page;
1325
1326 static inline
1327 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1328 {
1329         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1330 }
1331
1332 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1333 {
1334         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1335
1336         kvm_arch_vcpu_load(vcpu, cpu);
1337 }
1338
1339 static void kvm_sched_out(struct preempt_notifier *pn,
1340                           struct task_struct *next)
1341 {
1342         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1343
1344         kvm_arch_vcpu_put(vcpu);
1345 }
1346
1347 int kvm_init(void *opaque, unsigned int vcpu_size,
1348                   struct module *module)
1349 {
1350         int r;
1351         int cpu;
1352
1353         kvm_init_debug();
1354
1355         r = kvm_arch_init(opaque);
1356         if (r)
1357                 goto out4;
1358
1359         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1360
1361         if (bad_page == NULL) {
1362                 r = -ENOMEM;
1363                 goto out;
1364         }
1365
1366         r = kvm_arch_hardware_setup();
1367         if (r < 0)
1368                 goto out;
1369
1370         for_each_online_cpu(cpu) {
1371                 smp_call_function_single(cpu,
1372                                 kvm_arch_check_processor_compat,
1373                                 &r, 0, 1);
1374                 if (r < 0)
1375                         goto out_free_0;
1376         }
1377
1378         on_each_cpu(hardware_enable, NULL, 0, 1);
1379         r = register_cpu_notifier(&kvm_cpu_notifier);
1380         if (r)
1381                 goto out_free_1;
1382         register_reboot_notifier(&kvm_reboot_notifier);
1383
1384         r = sysdev_class_register(&kvm_sysdev_class);
1385         if (r)
1386                 goto out_free_2;
1387
1388         r = sysdev_register(&kvm_sysdev);
1389         if (r)
1390                 goto out_free_3;
1391
1392         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1393         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1394                                            __alignof__(struct kvm_vcpu),
1395                                            0, NULL);
1396         if (!kvm_vcpu_cache) {
1397                 r = -ENOMEM;
1398                 goto out_free_4;
1399         }
1400
1401         kvm_chardev_ops.owner = module;
1402
1403         r = misc_register(&kvm_dev);
1404         if (r) {
1405                 printk(KERN_ERR "kvm: misc device register failed\n");
1406                 goto out_free;
1407         }
1408
1409         kvm_preempt_ops.sched_in = kvm_sched_in;
1410         kvm_preempt_ops.sched_out = kvm_sched_out;
1411
1412         return 0;
1413
1414 out_free:
1415         kmem_cache_destroy(kvm_vcpu_cache);
1416 out_free_4:
1417         sysdev_unregister(&kvm_sysdev);
1418 out_free_3:
1419         sysdev_class_unregister(&kvm_sysdev_class);
1420 out_free_2:
1421         unregister_reboot_notifier(&kvm_reboot_notifier);
1422         unregister_cpu_notifier(&kvm_cpu_notifier);
1423 out_free_1:
1424         on_each_cpu(hardware_disable, NULL, 0, 1);
1425 out_free_0:
1426         kvm_arch_hardware_unsetup();
1427 out:
1428         kvm_arch_exit();
1429         kvm_exit_debug();
1430 out4:
1431         return r;
1432 }
1433 EXPORT_SYMBOL_GPL(kvm_init);
1434
1435 void kvm_exit(void)
1436 {
1437         misc_deregister(&kvm_dev);
1438         kmem_cache_destroy(kvm_vcpu_cache);
1439         sysdev_unregister(&kvm_sysdev);
1440         sysdev_class_unregister(&kvm_sysdev_class);
1441         unregister_reboot_notifier(&kvm_reboot_notifier);
1442         unregister_cpu_notifier(&kvm_cpu_notifier);
1443         on_each_cpu(hardware_disable, NULL, 0, 1);
1444         kvm_arch_hardware_unsetup();
1445         kvm_arch_exit();
1446         kvm_exit_debug();
1447         __free_page(bad_page);
1448 }
1449 EXPORT_SYMBOL_GPL(kvm_exit);