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