SAFE public projects git trees. - safe/jmp/linux-2.6/blob - virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "iodev.h"
  19
  20 #include <linux/kvm_host.h>
  21 #include <linux/kvm.h>
  22 #include <linux/module.h>
  23 #include <linux/errno.h>
  24 #include <linux/percpu.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44
  45 #include <asm/processor.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/pgtable.h>
  49
  50 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
  51 #include "coalesced_mmio.h"
  52 #endif
  53
  54 MODULE_AUTHOR("Qumranet");
  55 MODULE_LICENSE("GPL");
  56
  57 DEFINE_SPINLOCK(kvm_lock);
  58 LIST_HEAD(vm_list);
  59
  60 static cpumask_t cpus_hardware_enabled;
  61
  62 struct kmem_cache *kvm_vcpu_cache;
  63 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  64
  65 static __read_mostly struct preempt_ops kvm_preempt_ops;
  66
  67 struct dentry *kvm_debugfs_dir;
  68
  69 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  70                            unsigned long arg);
  71
  72 bool kvm_rebooting;
  73
  74 static inline int valid_vcpu(int n)
  75 {
  76         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  77 }
  78
  79 /*
  80  * Switches to specified vcpu, until a matching vcpu_put()
  81  */
  82 void vcpu_load(struct kvm_vcpu *vcpu)
  83 {
  84         int cpu;
  85
  86         mutex_lock(&vcpu->mutex);
  87         cpu = get_cpu();
  88         preempt_notifier_register(&vcpu->preempt_notifier);
  89         kvm_arch_vcpu_load(vcpu, cpu);
  90         put_cpu();
  91 }
  92
  93 void vcpu_put(struct kvm_vcpu *vcpu)
  94 {
  95         preempt_disable();
  96         kvm_arch_vcpu_put(vcpu);
  97         preempt_notifier_unregister(&vcpu->preempt_notifier);
  98         preempt_enable();
  99         mutex_unlock(&vcpu->mutex);
 100 }
 101
 102 static void ack_flush(void *_completed)
 103 {
 104 }
 105
 106 void kvm_flush_remote_tlbs(struct kvm *kvm)
 107 {
 108         int i, cpu;
 109         cpumask_t cpus;
 110         struct kvm_vcpu *vcpu;
 111
 112         cpus_clear(cpus);
 113         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 114                 vcpu = kvm->vcpus[i];
 115                 if (!vcpu)
 116                         continue;
 117                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 118                         continue;
 119                 cpu = vcpu->cpu;
 120                 if (cpu != -1 && cpu != raw_smp_processor_id())
 121                         cpu_set(cpu, cpus);
 122         }
 123         if (cpus_empty(cpus))
 124                 return;
 125         ++kvm->stat.remote_tlb_flush;
 126         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 127 }
 128
 129 void kvm_reload_remote_mmus(struct kvm *kvm)
 130 {
 131         int i, cpu;
 132         cpumask_t cpus;
 133         struct kvm_vcpu *vcpu;
 134
 135         cpus_clear(cpus);
 136         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 137                 vcpu = kvm->vcpus[i];
 138                 if (!vcpu)
 139                         continue;
 140                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 141                         continue;
 142                 cpu = vcpu->cpu;
 143                 if (cpu != -1 && cpu != raw_smp_processor_id())
 144                         cpu_set(cpu, cpus);
 145         }
 146         if (cpus_empty(cpus))
 147                 return;
 148         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 149 }
 150
 151
 152 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 153 {
 154         struct page *page;
 155         int r;
 156
 157         mutex_init(&vcpu->mutex);
 158         vcpu->cpu = -1;
 159         vcpu->kvm = kvm;
 160         vcpu->vcpu_id = id;
 161         init_waitqueue_head(&vcpu->wq);
 162
 163         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 164         if (!page) {
 165                 r = -ENOMEM;
 166                 goto fail;
 167         }
 168         vcpu->run = page_address(page);
 169
 170         r = kvm_arch_vcpu_init(vcpu);
 171         if (r < 0)
 172                 goto fail_free_run;
 173         return 0;
 174
 175 fail_free_run:
 176         free_page((unsigned long)vcpu->run);
 177 fail:
 178         return r;
 179 }
 180 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 181
 182 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 183 {
 184         kvm_arch_vcpu_uninit(vcpu);
 185         free_page((unsigned long)vcpu->run);
 186 }
 187 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 188
 189 static struct kvm *kvm_create_vm(void)
 190 {
 191         struct kvm *kvm = kvm_arch_create_vm();
 192 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 193         struct page *page;
 194 #endif
 195
 196         if (IS_ERR(kvm))
 197                 goto out;
 198
 199 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 200         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 201         if (!page) {
 202                 kfree(kvm);
 203                 return ERR_PTR(-ENOMEM);
 204         }
 205         kvm->coalesced_mmio_ring =
 206                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 207 #endif
 208
 209         kvm->mm = current->mm;
 210         atomic_inc(&kvm->mm->mm_count);
 211         spin_lock_init(&kvm->mmu_lock);
 212         kvm_io_bus_init(&kvm->pio_bus);
 213         mutex_init(&kvm->lock);
 214         kvm_io_bus_init(&kvm->mmio_bus);
 215         init_rwsem(&kvm->slots_lock);
 216         atomic_set(&kvm->users_count, 1);
 217         spin_lock(&kvm_lock);
 218         list_add(&kvm->vm_list, &vm_list);
 219         spin_unlock(&kvm_lock);
 220 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 221         kvm_coalesced_mmio_init(kvm);
 222 #endif
 223 out:
 224         return kvm;
 225 }
 226
 227 /*
 228  * Free any memory in @free but not in @dont.
 229  */
 230 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 231                                   struct kvm_memory_slot *dont)
 232 {
 233         if (!dont || free->rmap != dont->rmap)
 234                 vfree(free->rmap);
 235
 236         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 237                 vfree(free->dirty_bitmap);
 238
 239         if (!dont || free->lpage_info != dont->lpage_info)
 240                 vfree(free->lpage_info);
 241
 242         free->npages = 0;
 243         free->dirty_bitmap = NULL;
 244         free->rmap = NULL;
 245         free->lpage_info = NULL;
 246 }
 247
 248 void kvm_free_physmem(struct kvm *kvm)
 249 {
 250         int i;
 251
 252         for (i = 0; i < kvm->nmemslots; ++i)
 253                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 254 }
 255
 256 static void kvm_destroy_vm(struct kvm *kvm)
 257 {
 258         struct mm_struct *mm = kvm->mm;
 259
 260         spin_lock(&kvm_lock);
 261         list_del(&kvm->vm_list);
 262         spin_unlock(&kvm_lock);
 263         kvm_io_bus_destroy(&kvm->pio_bus);
 264         kvm_io_bus_destroy(&kvm->mmio_bus);
 265 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 266         if (kvm->coalesced_mmio_ring != NULL)
 267                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 268 #endif
 269         kvm_arch_destroy_vm(kvm);
 270         mmdrop(mm);
 271 }
 272
 273 void kvm_get_kvm(struct kvm *kvm)
 274 {
 275         atomic_inc(&kvm->users_count);
 276 }
 277 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 278
 279 void kvm_put_kvm(struct kvm *kvm)
 280 {
 281         if (atomic_dec_and_test(&kvm->users_count))
 282                 kvm_destroy_vm(kvm);
 283 }
 284 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 285
 286
 287 static int kvm_vm_release(struct inode *inode, struct file *filp)
 288 {
 289         struct kvm *kvm = filp->private_data;
 290
 291         kvm_put_kvm(kvm);
 292         return 0;
 293 }
 294
 295 /*
 296  * Allocate some memory and give it an address in the guest physical address
 297  * space.
 298  *
 299  * Discontiguous memory is allowed, mostly for framebuffers.
 300  *
 301  * Must be called holding mmap_sem for write.
 302  */
 303 int __kvm_set_memory_region(struct kvm *kvm,
 304                             struct kvm_userspace_memory_region *mem,
 305                             int user_alloc)
 306 {
 307         int r;
 308         gfn_t base_gfn;
 309         unsigned long npages;
 310         unsigned long i;
 311         struct kvm_memory_slot *memslot;
 312         struct kvm_memory_slot old, new;
 313
 314         r = -EINVAL;
 315         /* General sanity checks */
 316         if (mem->memory_size & (PAGE_SIZE - 1))
 317                 goto out;
 318         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 319                 goto out;
 320         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 321                 goto out;
 322         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 323                 goto out;
 324
 325         memslot = &kvm->memslots[mem->slot];
 326         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 327         npages = mem->memory_size >> PAGE_SHIFT;
 328
 329         if (!npages)
 330                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 331
 332         new = old = *memslot;
 333
 334         new.base_gfn = base_gfn;
 335         new.npages = npages;
 336         new.flags = mem->flags;
 337
 338         /* Disallow changing a memory slot's size. */
 339         r = -EINVAL;
 340         if (npages && old.npages && npages != old.npages)
 341                 goto out_free;
 342
 343         /* Check for overlaps */
 344         r = -EEXIST;
 345         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 346                 struct kvm_memory_slot *s = &kvm->memslots[i];
 347
 348                 if (s == memslot)
 349                         continue;
 350                 if (!((base_gfn + npages <= s->base_gfn) ||
 351                       (base_gfn >= s->base_gfn + s->npages)))
 352                         goto out_free;
 353         }
 354
 355         /* Free page dirty bitmap if unneeded */
 356         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 357                 new.dirty_bitmap = NULL;
 358
 359         r = -ENOMEM;
 360
 361         /* Allocate if a slot is being created */
 362 #ifndef CONFIG_S390
 363         if (npages && !new.rmap) {
 364                 new.rmap = vmalloc(npages * sizeof(struct page *));
 365
 366                 if (!new.rmap)
 367                         goto out_free;
 368
 369                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 370
 371                 new.user_alloc = user_alloc;
 372                 new.userspace_addr = mem->userspace_addr;
 373         }
 374         if (npages && !new.lpage_info) {
 375                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 376                 if (npages % KVM_PAGES_PER_HPAGE)
 377                         largepages++;
 378                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 379                         largepages++;
 380
 381                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 382
 383                 if (!new.lpage_info)
 384                         goto out_free;
 385
 386                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 387
 388                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 389                         new.lpage_info[0].write_count = 1;
 390                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 391                         new.lpage_info[largepages-1].write_count = 1;
 392         }
 393
 394         /* Allocate page dirty bitmap if needed */
 395         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 396                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 397
 398                 new.dirty_bitmap = vmalloc(dirty_bytes);
 399                 if (!new.dirty_bitmap)
 400                         goto out_free;
 401                 memset(new.dirty_bitmap, 0, dirty_bytes);
 402         }
 403 #endif /* not defined CONFIG_S390 */
 404
 405         if (mem->slot >= kvm->nmemslots)
 406                 kvm->nmemslots = mem->slot + 1;
 407
 408         *memslot = new;
 409
 410         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 411         if (r) {
 412                 *memslot = old;
 413                 goto out_free;
 414         }
 415
 416         kvm_free_physmem_slot(&old, &new);
 417         return 0;
 418
 419 out_free:
 420         kvm_free_physmem_slot(&new, &old);
 421 out:
 422         return r;
 423
 424 }
 425 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 426
 427 int kvm_set_memory_region(struct kvm *kvm,
 428                           struct kvm_userspace_memory_region *mem,
 429                           int user_alloc)
 430 {
 431         int r;
 432
 433         down_write(&kvm->slots_lock);
 434         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 435         up_write(&kvm->slots_lock);
 436         return r;
 437 }
 438 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 439
 440 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 441                                    struct
 442                                    kvm_userspace_memory_region *mem,
 443                                    int user_alloc)
 444 {
 445         if (mem->slot >= KVM_MEMORY_SLOTS)
 446                 return -EINVAL;
 447         return kvm_set_memory_region(kvm, mem, user_alloc);
 448 }
 449
 450 int kvm_get_dirty_log(struct kvm *kvm,
 451                         struct kvm_dirty_log *log, int *is_dirty)
 452 {
 453         struct kvm_memory_slot *memslot;
 454         int r, i;
 455         int n;
 456         unsigned long any = 0;
 457
 458         r = -EINVAL;
 459         if (log->slot >= KVM_MEMORY_SLOTS)
 460                 goto out;
 461
 462         memslot = &kvm->memslots[log->slot];
 463         r = -ENOENT;
 464         if (!memslot->dirty_bitmap)
 465                 goto out;
 466
 467         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 468
 469         for (i = 0; !any && i < n/sizeof(long); ++i)
 470                 any = memslot->dirty_bitmap[i];
 471
 472         r = -EFAULT;
 473         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 474                 goto out;
 475
 476         if (any)
 477                 *is_dirty = 1;
 478
 479         r = 0;
 480 out:
 481         return r;
 482 }
 483
 484 int is_error_page(struct page *page)
 485 {
 486         return page == bad_page;
 487 }
 488 EXPORT_SYMBOL_GPL(is_error_page);
 489
 490 int is_error_pfn(pfn_t pfn)
 491 {
 492         return pfn == bad_pfn;
 493 }
 494 EXPORT_SYMBOL_GPL(is_error_pfn);
 495
 496 static inline unsigned long bad_hva(void)
 497 {
 498         return PAGE_OFFSET;
 499 }
 500
 501 int kvm_is_error_hva(unsigned long addr)
 502 {
 503         return addr == bad_hva();
 504 }
 505 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 506
 507 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 508 {
 509         int i;
 510
 511         for (i = 0; i < kvm->nmemslots; ++i) {
 512                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 513
 514                 if (gfn >= memslot->base_gfn
 515                     && gfn < memslot->base_gfn + memslot->npages)
 516                         return memslot;
 517         }
 518         return NULL;
 519 }
 520
 521 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 522 {
 523         gfn = unalias_gfn(kvm, gfn);
 524         return __gfn_to_memslot(kvm, gfn);
 525 }
 526
 527 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 528 {
 529         int i;
 530
 531         gfn = unalias_gfn(kvm, gfn);
 532         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 533                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 534
 535                 if (gfn >= memslot->base_gfn
 536                     && gfn < memslot->base_gfn + memslot->npages)
 537                         return 1;
 538         }
 539         return 0;
 540 }
 541 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 542
 543 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 544 {
 545         struct kvm_memory_slot *slot;
 546
 547         gfn = unalias_gfn(kvm, gfn);
 548         slot = __gfn_to_memslot(kvm, gfn);
 549         if (!slot)
 550                 return bad_hva();
 551         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 552 }
 553 EXPORT_SYMBOL_GPL(gfn_to_hva);
 554
 555 /*
 556  * Requires current->mm->mmap_sem to be held
 557  */
 558 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 559 {
 560         struct page *page[1];
 561         unsigned long addr;
 562         int npages;
 563         pfn_t pfn;
 564
 565         might_sleep();
 566
 567         addr = gfn_to_hva(kvm, gfn);
 568         if (kvm_is_error_hva(addr)) {
 569                 get_page(bad_page);
 570                 return page_to_pfn(bad_page);
 571         }
 572
 573         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 574                                 NULL);
 575
 576         if (unlikely(npages != 1)) {
 577                 struct vm_area_struct *vma;
 578
 579                 vma = find_vma(current->mm, addr);
 580                 if (vma == NULL || addr < vma->vm_start ||
 581                     !(vma->vm_flags & VM_PFNMAP)) {
 582                         get_page(bad_page);
 583                         return page_to_pfn(bad_page);
 584                 }
 585
 586                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 587                 BUG_ON(pfn_valid(pfn));
 588         } else
 589                 pfn = page_to_pfn(page[0]);
 590
 591         return pfn;
 592 }
 593
 594 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 595
 596 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 597 {
 598         pfn_t pfn;
 599
 600         pfn = gfn_to_pfn(kvm, gfn);
 601         if (pfn_valid(pfn))
 602                 return pfn_to_page(pfn);
 603
 604         WARN_ON(!pfn_valid(pfn));
 605
 606         get_page(bad_page);
 607         return bad_page;
 608 }
 609
 610 EXPORT_SYMBOL_GPL(gfn_to_page);
 611
 612 void kvm_release_page_clean(struct page *page)
 613 {
 614         kvm_release_pfn_clean(page_to_pfn(page));
 615 }
 616 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 617
 618 void kvm_release_pfn_clean(pfn_t pfn)
 619 {
 620         if (pfn_valid(pfn))
 621                 put_page(pfn_to_page(pfn));
 622 }
 623 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 624
 625 void kvm_release_page_dirty(struct page *page)
 626 {
 627         kvm_release_pfn_dirty(page_to_pfn(page));
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 630
 631 void kvm_release_pfn_dirty(pfn_t pfn)
 632 {
 633         kvm_set_pfn_dirty(pfn);
 634         kvm_release_pfn_clean(pfn);
 635 }
 636 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 637
 638 void kvm_set_page_dirty(struct page *page)
 639 {
 640         kvm_set_pfn_dirty(page_to_pfn(page));
 641 }
 642 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 643
 644 void kvm_set_pfn_dirty(pfn_t pfn)
 645 {
 646         if (pfn_valid(pfn)) {
 647                 struct page *page = pfn_to_page(pfn);
 648                 if (!PageReserved(page))
 649                         SetPageDirty(page);
 650         }
 651 }
 652 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 653
 654 void kvm_set_pfn_accessed(pfn_t pfn)
 655 {
 656         if (pfn_valid(pfn))
 657                 mark_page_accessed(pfn_to_page(pfn));
 658 }
 659 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 660
 661 void kvm_get_pfn(pfn_t pfn)
 662 {
 663         if (pfn_valid(pfn))
 664                 get_page(pfn_to_page(pfn));
 665 }
 666 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 667
 668 static int next_segment(unsigned long len, int offset)
 669 {
 670         if (len > PAGE_SIZE - offset)
 671                 return PAGE_SIZE - offset;
 672         else
 673                 return len;
 674 }
 675
 676 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 677                         int len)
 678 {
 679         int r;
 680         unsigned long addr;
 681
 682         addr = gfn_to_hva(kvm, gfn);
 683         if (kvm_is_error_hva(addr))
 684                 return -EFAULT;
 685         r = copy_from_user(data, (void __user *)addr + offset, len);
 686         if (r)
 687                 return -EFAULT;
 688         return 0;
 689 }
 690 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 691
 692 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 693 {
 694         gfn_t gfn = gpa >> PAGE_SHIFT;
 695         int seg;
 696         int offset = offset_in_page(gpa);
 697         int ret;
 698
 699         while ((seg = next_segment(len, offset)) != 0) {
 700                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 701                 if (ret < 0)
 702                         return ret;
 703                 offset = 0;
 704                 len -= seg;
 705                 data += seg;
 706                 ++gfn;
 707         }
 708         return 0;
 709 }
 710 EXPORT_SYMBOL_GPL(kvm_read_guest);
 711
 712 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 713                           unsigned long len)
 714 {
 715         int r;
 716         unsigned long addr;
 717         gfn_t gfn = gpa >> PAGE_SHIFT;
 718         int offset = offset_in_page(gpa);
 719
 720         addr = gfn_to_hva(kvm, gfn);
 721         if (kvm_is_error_hva(addr))
 722                 return -EFAULT;
 723         pagefault_disable();
 724         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
 725         pagefault_enable();
 726         if (r)
 727                 return -EFAULT;
 728         return 0;
 729 }
 730 EXPORT_SYMBOL(kvm_read_guest_atomic);
 731
 732 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 733                          int offset, int len)
 734 {
 735         int r;
 736         unsigned long addr;
 737
 738         addr = gfn_to_hva(kvm, gfn);
 739         if (kvm_is_error_hva(addr))
 740                 return -EFAULT;
 741         r = copy_to_user((void __user *)addr + offset, data, len);
 742         if (r)
 743                 return -EFAULT;
 744         mark_page_dirty(kvm, gfn);
 745         return 0;
 746 }
 747 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 748
 749 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 750                     unsigned long len)
 751 {
 752         gfn_t gfn = gpa >> PAGE_SHIFT;
 753         int seg;
 754         int offset = offset_in_page(gpa);
 755         int ret;
 756
 757         while ((seg = next_segment(len, offset)) != 0) {
 758                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 759                 if (ret < 0)
 760                         return ret;
 761                 offset = 0;
 762                 len -= seg;
 763                 data += seg;
 764                 ++gfn;
 765         }
 766         return 0;
 767 }
 768
 769 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 770 {
 771         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 772 }
 773 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 774
 775 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 776 {
 777         gfn_t gfn = gpa >> PAGE_SHIFT;
 778         int seg;
 779         int offset = offset_in_page(gpa);
 780         int ret;
 781
 782         while ((seg = next_segment(len, offset)) != 0) {
 783                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 784                 if (ret < 0)
 785                         return ret;
 786                 offset = 0;
 787                 len -= seg;
 788                 ++gfn;
 789         }
 790         return 0;
 791 }
 792 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 793
 794 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 795 {
 796         struct kvm_memory_slot *memslot;
 797
 798         gfn = unalias_gfn(kvm, gfn);
 799         memslot = __gfn_to_memslot(kvm, gfn);
 800         if (memslot && memslot->dirty_bitmap) {
 801                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 802
 803                 /* avoid RMW */
 804                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 805                         set_bit(rel_gfn, memslot->dirty_bitmap);
 806         }
 807 }
 808
 809 /*
 810  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 811  */
 812 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 813 {
 814         DEFINE_WAIT(wait);
 815
 816         for (;;) {
 817                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
 818
 819                 if (kvm_cpu_has_interrupt(vcpu))
 820                         break;
 821                 if (kvm_cpu_has_pending_timer(vcpu))
 822                         break;
 823                 if (kvm_arch_vcpu_runnable(vcpu))
 824                         break;
 825                 if (signal_pending(current))
 826                         break;
 827
 828                 vcpu_put(vcpu);
 829                 schedule();
 830                 vcpu_load(vcpu);
 831         }
 832
 833         finish_wait(&vcpu->wq, &wait);
 834 }
 835
 836 void kvm_resched(struct kvm_vcpu *vcpu)
 837 {
 838         if (!need_resched())
 839                 return;
 840         cond_resched();
 841 }
 842 EXPORT_SYMBOL_GPL(kvm_resched);
 843
 844 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 845 {
 846         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 847         struct page *page;
 848
 849         if (vmf->pgoff == 0)
 850                 page = virt_to_page(vcpu->run);
 851 #ifdef CONFIG_X86
 852         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 853                 page = virt_to_page(vcpu->arch.pio_data);
 854 #endif
 855 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 856         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
 857                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
 858 #endif
 859         else
 860                 return VM_FAULT_SIGBUS;
 861         get_page(page);
 862         vmf->page = page;
 863         return 0;
 864 }
 865
 866 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 867         .fault = kvm_vcpu_fault,
 868 };
 869
 870 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 871 {
 872         vma->vm_ops = &kvm_vcpu_vm_ops;
 873         return 0;
 874 }
 875
 876 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 877 {
 878         struct kvm_vcpu *vcpu = filp->private_data;
 879
 880         kvm_put_kvm(vcpu->kvm);
 881         return 0;
 882 }
 883
 884 static const struct file_operations kvm_vcpu_fops = {
 885         .release        = kvm_vcpu_release,
 886         .unlocked_ioctl = kvm_vcpu_ioctl,
 887         .compat_ioctl   = kvm_vcpu_ioctl,
 888         .mmap           = kvm_vcpu_mmap,
 889 };
 890
 891 /*
 892  * Allocates an inode for the vcpu.
 893  */
 894 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 895 {
 896         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu);
 897         if (fd < 0)
 898                 kvm_put_kvm(vcpu->kvm);
 899         return fd;
 900 }
 901
 902 /*
 903  * Creates some virtual cpus.  Good luck creating more than one.
 904  */
 905 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 906 {
 907         int r;
 908         struct kvm_vcpu *vcpu;
 909
 910         if (!valid_vcpu(n))
 911                 return -EINVAL;
 912
 913         vcpu = kvm_arch_vcpu_create(kvm, n);
 914         if (IS_ERR(vcpu))
 915                 return PTR_ERR(vcpu);
 916
 917         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 918
 919         r = kvm_arch_vcpu_setup(vcpu);
 920         if (r)
 921                 goto vcpu_destroy;
 922
 923         mutex_lock(&kvm->lock);
 924         if (kvm->vcpus[n]) {
 925                 r = -EEXIST;
 926                 mutex_unlock(&kvm->lock);
 927                 goto vcpu_destroy;
 928         }
 929         kvm->vcpus[n] = vcpu;
 930         mutex_unlock(&kvm->lock);
 931
 932         /* Now it's all set up, let userspace reach it */
 933         kvm_get_kvm(kvm);
 934         r = create_vcpu_fd(vcpu);
 935         if (r < 0)
 936                 goto unlink;
 937         return r;
 938
 939 unlink:
 940         mutex_lock(&kvm->lock);
 941         kvm->vcpus[n] = NULL;
 942         mutex_unlock(&kvm->lock);
 943 vcpu_destroy:
 944         kvm_arch_vcpu_destroy(vcpu);
 945         return r;
 946 }
 947
 948 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 949 {
 950         if (sigset) {
 951                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 952                 vcpu->sigset_active = 1;
 953                 vcpu->sigset = *sigset;
 954         } else
 955                 vcpu->sigset_active = 0;
 956         return 0;
 957 }
 958
 959 static long kvm_vcpu_ioctl(struct file *filp,
 960                            unsigned int ioctl, unsigned long arg)
 961 {
 962         struct kvm_vcpu *vcpu = filp->private_data;
 963         void __user *argp = (void __user *)arg;
 964         int r;
 965
 966         if (vcpu->kvm->mm != current->mm)
 967                 return -EIO;
 968         switch (ioctl) {
 969         case KVM_RUN:
 970                 r = -EINVAL;
 971                 if (arg)
 972                         goto out;
 973                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 974                 break;
 975         case KVM_GET_REGS: {
 976                 struct kvm_regs *kvm_regs;
 977
 978                 r = -ENOMEM;
 979                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 980                 if (!kvm_regs)
 981                         goto out;
 982                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
 983                 if (r)
 984                         goto out_free1;
 985                 r = -EFAULT;
 986                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
 987                         goto out_free1;
 988                 r = 0;
 989 out_free1:
 990                 kfree(kvm_regs);
 991                 break;
 992         }
 993         case KVM_SET_REGS: {
 994                 struct kvm_regs *kvm_regs;
 995
 996                 r = -ENOMEM;
 997                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 998                 if (!kvm_regs)
 999                         goto out;
1000                 r = -EFAULT;
1001                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1002                         goto out_free2;
1003                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1004                 if (r)
1005                         goto out_free2;
1006                 r = 0;
1007 out_free2:
1008                 kfree(kvm_regs);
1009                 break;
1010         }
1011         case KVM_GET_SREGS: {
1012                 struct kvm_sregs kvm_sregs;
1013
1014                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1015                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1016                 if (r)
1017                         goto out;
1018                 r = -EFAULT;
1019                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1020                         goto out;
1021                 r = 0;
1022                 break;
1023         }
1024         case KVM_SET_SREGS: {
1025                 struct kvm_sregs kvm_sregs;
1026
1027                 r = -EFAULT;
1028                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1029                         goto out;
1030                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1031                 if (r)
1032                         goto out;
1033                 r = 0;
1034                 break;
1035         }
1036         case KVM_GET_MP_STATE: {
1037                 struct kvm_mp_state mp_state;
1038
1039                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1040                 if (r)
1041                         goto out;
1042                 r = -EFAULT;
1043                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1044                         goto out;
1045                 r = 0;
1046                 break;
1047         }
1048         case KVM_SET_MP_STATE: {
1049                 struct kvm_mp_state mp_state;
1050
1051                 r = -EFAULT;
1052                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1053                         goto out;
1054                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1055                 if (r)
1056                         goto out;
1057                 r = 0;
1058                 break;
1059         }
1060         case KVM_TRANSLATE: {
1061                 struct kvm_translation tr;
1062
1063                 r = -EFAULT;
1064                 if (copy_from_user(&tr, argp, sizeof tr))
1065                         goto out;
1066                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1067                 if (r)
1068                         goto out;
1069                 r = -EFAULT;
1070                 if (copy_to_user(argp, &tr, sizeof tr))
1071                         goto out;
1072                 r = 0;
1073                 break;
1074         }
1075         case KVM_DEBUG_GUEST: {
1076                 struct kvm_debug_guest dbg;
1077
1078                 r = -EFAULT;
1079                 if (copy_from_user(&dbg, argp, sizeof dbg))
1080                         goto out;
1081                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1082                 if (r)
1083                         goto out;
1084                 r = 0;
1085                 break;
1086         }
1087         case KVM_SET_SIGNAL_MASK: {
1088                 struct kvm_signal_mask __user *sigmask_arg = argp;
1089                 struct kvm_signal_mask kvm_sigmask;
1090                 sigset_t sigset, *p;
1091
1092                 p = NULL;
1093                 if (argp) {
1094                         r = -EFAULT;
1095                         if (copy_from_user(&kvm_sigmask, argp,
1096                                            sizeof kvm_sigmask))
1097                                 goto out;
1098                         r = -EINVAL;
1099                         if (kvm_sigmask.len != sizeof sigset)
1100                                 goto out;
1101                         r = -EFAULT;
1102                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1103                                            sizeof sigset))
1104                                 goto out;
1105                         p = &sigset;
1106                 }
1107                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1108                 break;
1109         }
1110         case KVM_GET_FPU: {
1111                 struct kvm_fpu fpu;
1112
1113                 memset(&fpu, 0, sizeof fpu);
1114                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1115                 if (r)
1116                         goto out;
1117                 r = -EFAULT;
1118                 if (copy_to_user(argp, &fpu, sizeof fpu))
1119                         goto out;
1120                 r = 0;
1121                 break;
1122         }
1123         case KVM_SET_FPU: {
1124                 struct kvm_fpu fpu;
1125
1126                 r = -EFAULT;
1127                 if (copy_from_user(&fpu, argp, sizeof fpu))
1128                         goto out;
1129                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1130                 if (r)
1131                         goto out;
1132                 r = 0;
1133                 break;
1134         }
1135         default:
1136                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1137         }
1138 out:
1139         return r;
1140 }
1141
1142 static long kvm_vm_ioctl(struct file *filp,
1143                            unsigned int ioctl, unsigned long arg)
1144 {
1145         struct kvm *kvm = filp->private_data;
1146         void __user *argp = (void __user *)arg;
1147         int r;
1148
1149         if (kvm->mm != current->mm)
1150                 return -EIO;
1151         switch (ioctl) {
1152         case KVM_CREATE_VCPU:
1153                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1154                 if (r < 0)
1155                         goto out;
1156                 break;
1157         case KVM_SET_USER_MEMORY_REGION: {
1158                 struct kvm_userspace_memory_region kvm_userspace_mem;
1159
1160                 r = -EFAULT;
1161                 if (copy_from_user(&kvm_userspace_mem, argp,
1162                                                 sizeof kvm_userspace_mem))
1163                         goto out;
1164
1165                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1166                 if (r)
1167                         goto out;
1168                 break;
1169         }
1170         case KVM_GET_DIRTY_LOG: {
1171                 struct kvm_dirty_log log;
1172
1173                 r = -EFAULT;
1174                 if (copy_from_user(&log, argp, sizeof log))
1175                         goto out;
1176                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1177                 if (r)
1178                         goto out;
1179                 break;
1180         }
1181 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1182         case KVM_REGISTER_COALESCED_MMIO: {
1183                 struct kvm_coalesced_mmio_zone zone;
1184                 r = -EFAULT;
1185                 if (copy_from_user(&zone, argp, sizeof zone))
1186                         goto out;
1187                 r = -ENXIO;
1188                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1189                 if (r)
1190                         goto out;
1191                 r = 0;
1192                 break;
1193         }
1194         case KVM_UNREGISTER_COALESCED_MMIO: {
1195                 struct kvm_coalesced_mmio_zone zone;
1196                 r = -EFAULT;
1197                 if (copy_from_user(&zone, argp, sizeof zone))
1198                         goto out;
1199                 r = -ENXIO;
1200                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1201                 if (r)
1202                         goto out;
1203                 r = 0;
1204                 break;
1205         }
1206 #endif
1207         default:
1208                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1209         }
1210 out:
1211         return r;
1212 }
1213
1214 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1215 {
1216         struct kvm *kvm = vma->vm_file->private_data;
1217         struct page *page;
1218
1219         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1220                 return VM_FAULT_SIGBUS;
1221         page = gfn_to_page(kvm, vmf->pgoff);
1222         if (is_error_page(page)) {
1223                 kvm_release_page_clean(page);
1224                 return VM_FAULT_SIGBUS;
1225         }
1226         vmf->page = page;
1227         return 0;
1228 }
1229
1230 static struct vm_operations_struct kvm_vm_vm_ops = {
1231         .fault = kvm_vm_fault,
1232 };
1233
1234 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1235 {
1236         vma->vm_ops = &kvm_vm_vm_ops;
1237         return 0;
1238 }
1239
1240 static const struct file_operations kvm_vm_fops = {
1241         .release        = kvm_vm_release,
1242         .unlocked_ioctl = kvm_vm_ioctl,
1243         .compat_ioctl   = kvm_vm_ioctl,
1244         .mmap           = kvm_vm_mmap,
1245 };
1246
1247 static int kvm_dev_ioctl_create_vm(void)
1248 {
1249         int fd;
1250         struct kvm *kvm;
1251
1252         kvm = kvm_create_vm();
1253         if (IS_ERR(kvm))
1254                 return PTR_ERR(kvm);
1255         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm);
1256         if (fd < 0)
1257                 kvm_put_kvm(kvm);
1258
1259         return fd;
1260 }
1261
1262 static long kvm_dev_ioctl(struct file *filp,
1263                           unsigned int ioctl, unsigned long arg)
1264 {
1265         long r = -EINVAL;
1266
1267         switch (ioctl) {
1268         case KVM_GET_API_VERSION:
1269                 r = -EINVAL;
1270                 if (arg)
1271                         goto out;
1272                 r = KVM_API_VERSION;
1273                 break;
1274         case KVM_CREATE_VM:
1275                 r = -EINVAL;
1276                 if (arg)
1277                         goto out;
1278                 r = kvm_dev_ioctl_create_vm();
1279                 break;
1280         case KVM_CHECK_EXTENSION:
1281                 r = kvm_dev_ioctl_check_extension(arg);
1282                 break;
1283         case KVM_GET_VCPU_MMAP_SIZE:
1284                 r = -EINVAL;
1285                 if (arg)
1286                         goto out;
1287                 r = PAGE_SIZE;     /* struct kvm_run */
1288 #ifdef CONFIG_X86
1289                 r += PAGE_SIZE;    /* pio data page */
1290 #endif
1291 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1292                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1293 #endif
1294                 break;
1295         case KVM_TRACE_ENABLE:
1296         case KVM_TRACE_PAUSE:
1297         case KVM_TRACE_DISABLE:
1298                 r = kvm_trace_ioctl(ioctl, arg);
1299                 break;
1300         default:
1301                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1302         }
1303 out:
1304         return r;
1305 }
1306
1307 static struct file_operations kvm_chardev_ops = {
1308         .unlocked_ioctl = kvm_dev_ioctl,
1309         .compat_ioctl   = kvm_dev_ioctl,
1310 };
1311
1312 static struct miscdevice kvm_dev = {
1313         KVM_MINOR,
1314         "kvm",
1315         &kvm_chardev_ops,
1316 };
1317
1318 static void hardware_enable(void *junk)
1319 {
1320         int cpu = raw_smp_processor_id();
1321
1322         if (cpu_isset(cpu, cpus_hardware_enabled))
1323                 return;
1324         cpu_set(cpu, cpus_hardware_enabled);
1325         kvm_arch_hardware_enable(NULL);
1326 }
1327
1328 static void hardware_disable(void *junk)
1329 {
1330         int cpu = raw_smp_processor_id();
1331
1332         if (!cpu_isset(cpu, cpus_hardware_enabled))
1333                 return;
1334         cpu_clear(cpu, cpus_hardware_enabled);
1335         kvm_arch_hardware_disable(NULL);
1336 }
1337
1338 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1339                            void *v)
1340 {
1341         int cpu = (long)v;
1342
1343         val &= ~CPU_TASKS_FROZEN;
1344         switch (val) {
1345         case CPU_DYING:
1346                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1347                        cpu);
1348                 hardware_disable(NULL);
1349                 break;
1350         case CPU_UP_CANCELED:
1351                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1352                        cpu);
1353                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1354                 break;
1355         case CPU_ONLINE:
1356                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1357                        cpu);
1358                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1359                 break;
1360         }
1361         return NOTIFY_OK;
1362 }
1363
1364
1365 asmlinkage void kvm_handle_fault_on_reboot(void)
1366 {
1367         if (kvm_rebooting)
1368                 /* spin while reset goes on */
1369                 while (true)
1370                         ;
1371         /* Fault while not rebooting.  We want the trace. */
1372         BUG();
1373 }
1374 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1375
1376 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1377                       void *v)
1378 {
1379         if (val == SYS_RESTART) {
1380                 /*
1381                  * Some (well, at least mine) BIOSes hang on reboot if
1382                  * in vmx root mode.
1383                  */
1384                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1385                 kvm_rebooting = true;
1386                 on_each_cpu(hardware_disable, NULL, 1);
1387         }
1388         return NOTIFY_OK;
1389 }
1390
1391 static struct notifier_block kvm_reboot_notifier = {
1392         .notifier_call = kvm_reboot,
1393         .priority = 0,
1394 };
1395
1396 void kvm_io_bus_init(struct kvm_io_bus *bus)
1397 {
1398         memset(bus, 0, sizeof(*bus));
1399 }
1400
1401 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1402 {
1403         int i;
1404
1405         for (i = 0; i < bus->dev_count; i++) {
1406                 struct kvm_io_device *pos = bus->devs[i];
1407
1408                 kvm_iodevice_destructor(pos);
1409         }
1410 }
1411
1412 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1413                                           gpa_t addr, int len, int is_write)
1414 {
1415         int i;
1416
1417         for (i = 0; i < bus->dev_count; i++) {
1418                 struct kvm_io_device *pos = bus->devs[i];
1419
1420                 if (pos->in_range(pos, addr, len, is_write))
1421                         return pos;
1422         }
1423
1424         return NULL;
1425 }
1426
1427 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1428 {
1429         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1430
1431         bus->devs[bus->dev_count++] = dev;
1432 }
1433
1434 static struct notifier_block kvm_cpu_notifier = {
1435         .notifier_call = kvm_cpu_hotplug,
1436         .priority = 20, /* must be > scheduler priority */
1437 };
1438
1439 static int vm_stat_get(void *_offset, u64 *val)
1440 {
1441         unsigned offset = (long)_offset;
1442         struct kvm *kvm;
1443
1444         *val = 0;
1445         spin_lock(&kvm_lock);
1446         list_for_each_entry(kvm, &vm_list, vm_list)
1447                 *val += *(u32 *)((void *)kvm + offset);
1448         spin_unlock(&kvm_lock);
1449         return 0;
1450 }
1451
1452 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1453
1454 static int vcpu_stat_get(void *_offset, u64 *val)
1455 {
1456         unsigned offset = (long)_offset;
1457         struct kvm *kvm;
1458         struct kvm_vcpu *vcpu;
1459         int i;
1460
1461         *val = 0;
1462         spin_lock(&kvm_lock);
1463         list_for_each_entry(kvm, &vm_list, vm_list)
1464                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1465                         vcpu = kvm->vcpus[i];
1466                         if (vcpu)
1467                                 *val += *(u32 *)((void *)vcpu + offset);
1468                 }
1469         spin_unlock(&kvm_lock);
1470         return 0;
1471 }
1472
1473 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1474
1475 static struct file_operations *stat_fops[] = {
1476         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1477         [KVM_STAT_VM]   = &vm_stat_fops,
1478 };
1479
1480 static void kvm_init_debug(void)
1481 {
1482         struct kvm_stats_debugfs_item *p;
1483
1484         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1485         for (p = debugfs_entries; p->name; ++p)
1486                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1487                                                 (void *)(long)p->offset,
1488                                                 stat_fops[p->kind]);
1489 }
1490
1491 static void kvm_exit_debug(void)
1492 {
1493         struct kvm_stats_debugfs_item *p;
1494
1495         for (p = debugfs_entries; p->name; ++p)
1496                 debugfs_remove(p->dentry);
1497         debugfs_remove(kvm_debugfs_dir);
1498 }
1499
1500 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1501 {
1502         hardware_disable(NULL);
1503         return 0;
1504 }
1505
1506 static int kvm_resume(struct sys_device *dev)
1507 {
1508         hardware_enable(NULL);
1509         return 0;
1510 }
1511
1512 static struct sysdev_class kvm_sysdev_class = {
1513         .name = "kvm",
1514         .suspend = kvm_suspend,
1515         .resume = kvm_resume,
1516 };
1517
1518 static struct sys_device kvm_sysdev = {
1519         .id = 0,
1520         .cls = &kvm_sysdev_class,
1521 };
1522
1523 struct page *bad_page;
1524 pfn_t bad_pfn;
1525
1526 static inline
1527 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1528 {
1529         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1530 }
1531
1532 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1533 {
1534         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1535
1536         kvm_arch_vcpu_load(vcpu, cpu);
1537 }
1538
1539 static void kvm_sched_out(struct preempt_notifier *pn,
1540                           struct task_struct *next)
1541 {
1542         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1543
1544         kvm_arch_vcpu_put(vcpu);
1545 }
1546
1547 int kvm_init(void *opaque, unsigned int vcpu_size,
1548                   struct module *module)
1549 {
1550         int r;
1551         int cpu;
1552
1553         kvm_init_debug();
1554
1555         r = kvm_arch_init(opaque);
1556         if (r)
1557                 goto out_fail;
1558
1559         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1560
1561         if (bad_page == NULL) {
1562                 r = -ENOMEM;
1563                 goto out;
1564         }
1565
1566         bad_pfn = page_to_pfn(bad_page);
1567
1568         r = kvm_arch_hardware_setup();
1569         if (r < 0)
1570                 goto out_free_0;
1571
1572         for_each_online_cpu(cpu) {
1573                 smp_call_function_single(cpu,
1574                                 kvm_arch_check_processor_compat,
1575                                 &r, 1);
1576                 if (r < 0)
1577                         goto out_free_1;
1578         }
1579
1580         on_each_cpu(hardware_enable, NULL, 1);
1581         r = register_cpu_notifier(&kvm_cpu_notifier);
1582         if (r)
1583                 goto out_free_2;
1584         register_reboot_notifier(&kvm_reboot_notifier);
1585
1586         r = sysdev_class_register(&kvm_sysdev_class);
1587         if (r)
1588                 goto out_free_3;
1589
1590         r = sysdev_register(&kvm_sysdev);
1591         if (r)
1592                 goto out_free_4;
1593
1594         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1595         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1596                                            __alignof__(struct kvm_vcpu),
1597                                            0, NULL);
1598         if (!kvm_vcpu_cache) {
1599                 r = -ENOMEM;
1600                 goto out_free_5;
1601         }
1602
1603         kvm_chardev_ops.owner = module;
1604
1605         r = misc_register(&kvm_dev);
1606         if (r) {
1607                 printk(KERN_ERR "kvm: misc device register failed\n");
1608                 goto out_free;
1609         }
1610
1611         kvm_preempt_ops.sched_in = kvm_sched_in;
1612         kvm_preempt_ops.sched_out = kvm_sched_out;
1613
1614         return 0;
1615
1616 out_free:
1617         kmem_cache_destroy(kvm_vcpu_cache);
1618 out_free_5:
1619         sysdev_unregister(&kvm_sysdev);
1620 out_free_4:
1621         sysdev_class_unregister(&kvm_sysdev_class);
1622 out_free_3:
1623         unregister_reboot_notifier(&kvm_reboot_notifier);
1624         unregister_cpu_notifier(&kvm_cpu_notifier);
1625 out_free_2:
1626         on_each_cpu(hardware_disable, NULL, 1);
1627 out_free_1:
1628         kvm_arch_hardware_unsetup();
1629 out_free_0:
1630         __free_page(bad_page);
1631 out:
1632         kvm_arch_exit();
1633         kvm_exit_debug();
1634 out_fail:
1635         return r;
1636 }
1637 EXPORT_SYMBOL_GPL(kvm_init);
1638
1639 void kvm_exit(void)
1640 {
1641         kvm_trace_cleanup();
1642         misc_deregister(&kvm_dev);
1643         kmem_cache_destroy(kvm_vcpu_cache);
1644         sysdev_unregister(&kvm_sysdev);
1645         sysdev_class_unregister(&kvm_sysdev_class);
1646         unregister_reboot_notifier(&kvm_reboot_notifier);
1647         unregister_cpu_notifier(&kvm_cpu_notifier);
1648         on_each_cpu(hardware_disable, NULL, 1);
1649         kvm_arch_hardware_unsetup();
1650         kvm_arch_exit();
1651         kvm_exit_debug();
1652         __free_page(bad_page);
1653 }
1654 EXPORT_SYMBOL_GPL(kvm_exit);