git.oblomov.eu Git - 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, me;
 109         cpumask_t cpus;
 110         struct kvm_vcpu *vcpu;
 111
 112         me = get_cpu();
 113         cpus_clear(cpus);
 114         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 115                 vcpu = kvm->vcpus[i];
 116                 if (!vcpu)
 117                         continue;
 118                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 119                         continue;
 120                 cpu = vcpu->cpu;
 121                 if (cpu != -1 && cpu != me)
 122                         cpu_set(cpu, cpus);
 123         }
 124         if (cpus_empty(cpus))
 125                 goto out;
 126         ++kvm->stat.remote_tlb_flush;
 127         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 128 out:
 129         put_cpu();
 130 }
 131
 132 void kvm_reload_remote_mmus(struct kvm *kvm)
 133 {
 134         int i, cpu, me;
 135         cpumask_t cpus;
 136         struct kvm_vcpu *vcpu;
 137
 138         me = get_cpu();
 139         cpus_clear(cpus);
 140         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 141                 vcpu = kvm->vcpus[i];
 142                 if (!vcpu)
 143                         continue;
 144                 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
 145                         continue;
 146                 cpu = vcpu->cpu;
 147                 if (cpu != -1 && cpu != me)
 148                         cpu_set(cpu, cpus);
 149         }
 150         if (cpus_empty(cpus))
 151                 goto out;
 152         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 153 out:
 154         put_cpu();
 155 }
 156
 157
 158 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 159 {
 160         struct page *page;
 161         int r;
 162
 163         mutex_init(&vcpu->mutex);
 164         vcpu->cpu = -1;
 165         vcpu->kvm = kvm;
 166         vcpu->vcpu_id = id;
 167         init_waitqueue_head(&vcpu->wq);
 168
 169         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 170         if (!page) {
 171                 r = -ENOMEM;
 172                 goto fail;
 173         }
 174         vcpu->run = page_address(page);
 175
 176         r = kvm_arch_vcpu_init(vcpu);
 177         if (r < 0)
 178                 goto fail_free_run;
 179         return 0;
 180
 181 fail_free_run:
 182         free_page((unsigned long)vcpu->run);
 183 fail:
 184         return r;
 185 }
 186 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 187
 188 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 189 {
 190         kvm_arch_vcpu_uninit(vcpu);
 191         free_page((unsigned long)vcpu->run);
 192 }
 193 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 194
 195 static struct kvm *kvm_create_vm(void)
 196 {
 197         struct kvm *kvm = kvm_arch_create_vm();
 198 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 199         struct page *page;
 200 #endif
 201
 202         if (IS_ERR(kvm))
 203                 goto out;
 204
 205 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 206         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 207         if (!page) {
 208                 kfree(kvm);
 209                 return ERR_PTR(-ENOMEM);
 210         }
 211         kvm->coalesced_mmio_ring =
 212                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 213 #endif
 214
 215         kvm->mm = current->mm;
 216         atomic_inc(&kvm->mm->mm_count);
 217         spin_lock_init(&kvm->mmu_lock);
 218         kvm_io_bus_init(&kvm->pio_bus);
 219         mutex_init(&kvm->lock);
 220         kvm_io_bus_init(&kvm->mmio_bus);
 221         init_rwsem(&kvm->slots_lock);
 222         atomic_set(&kvm->users_count, 1);
 223         spin_lock(&kvm_lock);
 224         list_add(&kvm->vm_list, &vm_list);
 225         spin_unlock(&kvm_lock);
 226 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 227         kvm_coalesced_mmio_init(kvm);
 228 #endif
 229 out:
 230         return kvm;
 231 }
 232
 233 /*
 234  * Free any memory in @free but not in @dont.
 235  */
 236 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 237                                   struct kvm_memory_slot *dont)
 238 {
 239         if (!dont || free->rmap != dont->rmap)
 240                 vfree(free->rmap);
 241
 242         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 243                 vfree(free->dirty_bitmap);
 244
 245         if (!dont || free->lpage_info != dont->lpage_info)
 246                 vfree(free->lpage_info);
 247
 248         free->npages = 0;
 249         free->dirty_bitmap = NULL;
 250         free->rmap = NULL;
 251         free->lpage_info = NULL;
 252 }
 253
 254 void kvm_free_physmem(struct kvm *kvm)
 255 {
 256         int i;
 257
 258         for (i = 0; i < kvm->nmemslots; ++i)
 259                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 260 }
 261
 262 static void kvm_destroy_vm(struct kvm *kvm)
 263 {
 264         struct mm_struct *mm = kvm->mm;
 265
 266         spin_lock(&kvm_lock);
 267         list_del(&kvm->vm_list);
 268         spin_unlock(&kvm_lock);
 269         kvm_io_bus_destroy(&kvm->pio_bus);
 270         kvm_io_bus_destroy(&kvm->mmio_bus);
 271 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 272         if (kvm->coalesced_mmio_ring != NULL)
 273                 free_page((unsigned long)kvm->coalesced_mmio_ring);
 274 #endif
 275         kvm_arch_destroy_vm(kvm);
 276         mmdrop(mm);
 277 }
 278
 279 void kvm_get_kvm(struct kvm *kvm)
 280 {
 281         atomic_inc(&kvm->users_count);
 282 }
 283 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 284
 285 void kvm_put_kvm(struct kvm *kvm)
 286 {
 287         if (atomic_dec_and_test(&kvm->users_count))
 288                 kvm_destroy_vm(kvm);
 289 }
 290 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 291
 292
 293 static int kvm_vm_release(struct inode *inode, struct file *filp)
 294 {
 295         struct kvm *kvm = filp->private_data;
 296
 297         kvm_put_kvm(kvm);
 298         return 0;
 299 }
 300
 301 /*
 302  * Allocate some memory and give it an address in the guest physical address
 303  * space.
 304  *
 305  * Discontiguous memory is allowed, mostly for framebuffers.
 306  *
 307  * Must be called holding mmap_sem for write.
 308  */
 309 int __kvm_set_memory_region(struct kvm *kvm,
 310                             struct kvm_userspace_memory_region *mem,
 311                             int user_alloc)
 312 {
 313         int r;
 314         gfn_t base_gfn;
 315         unsigned long npages;
 316         unsigned long i;
 317         struct kvm_memory_slot *memslot;
 318         struct kvm_memory_slot old, new;
 319
 320         r = -EINVAL;
 321         /* General sanity checks */
 322         if (mem->memory_size & (PAGE_SIZE - 1))
 323                 goto out;
 324         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 325                 goto out;
 326         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 327                 goto out;
 328         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 329                 goto out;
 330
 331         memslot = &kvm->memslots[mem->slot];
 332         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 333         npages = mem->memory_size >> PAGE_SHIFT;
 334
 335         if (!npages)
 336                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 337
 338         new = old = *memslot;
 339
 340         new.base_gfn = base_gfn;
 341         new.npages = npages;
 342         new.flags = mem->flags;
 343
 344         /* Disallow changing a memory slot's size. */
 345         r = -EINVAL;
 346         if (npages && old.npages && npages != old.npages)
 347                 goto out_free;
 348
 349         /* Check for overlaps */
 350         r = -EEXIST;
 351         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 352                 struct kvm_memory_slot *s = &kvm->memslots[i];
 353
 354                 if (s == memslot)
 355                         continue;
 356                 if (!((base_gfn + npages <= s->base_gfn) ||
 357                       (base_gfn >= s->base_gfn + s->npages)))
 358                         goto out_free;
 359         }
 360
 361         /* Free page dirty bitmap if unneeded */
 362         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 363                 new.dirty_bitmap = NULL;
 364
 365         r = -ENOMEM;
 366
 367         /* Allocate if a slot is being created */
 368 #ifndef CONFIG_S390
 369         if (npages && !new.rmap) {
 370                 new.rmap = vmalloc(npages * sizeof(struct page *));
 371
 372                 if (!new.rmap)
 373                         goto out_free;
 374
 375                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 376
 377                 new.user_alloc = user_alloc;
 378                 new.userspace_addr = mem->userspace_addr;
 379         }
 380         if (npages && !new.lpage_info) {
 381                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 382                 if (npages % KVM_PAGES_PER_HPAGE)
 383                         largepages++;
 384                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 385                         largepages++;
 386
 387                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 388
 389                 if (!new.lpage_info)
 390                         goto out_free;
 391
 392                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 393
 394                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 395                         new.lpage_info[0].write_count = 1;
 396                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 397                         new.lpage_info[largepages-1].write_count = 1;
 398         }
 399
 400         /* Allocate page dirty bitmap if needed */
 401         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 402                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 403
 404                 new.dirty_bitmap = vmalloc(dirty_bytes);
 405                 if (!new.dirty_bitmap)
 406                         goto out_free;
 407                 memset(new.dirty_bitmap, 0, dirty_bytes);
 408         }
 409 #endif /* not defined CONFIG_S390 */
 410
 411         if (mem->slot >= kvm->nmemslots)
 412                 kvm->nmemslots = mem->slot + 1;
 413
 414         if (!npages)
 415                 kvm_arch_flush_shadow(kvm);
 416
 417         *memslot = new;
 418
 419         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 420         if (r) {
 421                 *memslot = old;
 422                 goto out_free;
 423         }
 424
 425         kvm_free_physmem_slot(&old, &new);
 426         return 0;
 427
 428 out_free:
 429         kvm_free_physmem_slot(&new, &old);
 430 out:
 431         return r;
 432
 433 }
 434 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 435
 436 int kvm_set_memory_region(struct kvm *kvm,
 437                           struct kvm_userspace_memory_region *mem,
 438                           int user_alloc)
 439 {
 440         int r;
 441
 442         down_write(&kvm->slots_lock);
 443         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 444         up_write(&kvm->slots_lock);
 445         return r;
 446 }
 447 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 448
 449 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 450                                    struct
 451                                    kvm_userspace_memory_region *mem,
 452                                    int user_alloc)
 453 {
 454         if (mem->slot >= KVM_MEMORY_SLOTS)
 455                 return -EINVAL;
 456         return kvm_set_memory_region(kvm, mem, user_alloc);
 457 }
 458
 459 int kvm_get_dirty_log(struct kvm *kvm,
 460                         struct kvm_dirty_log *log, int *is_dirty)
 461 {
 462         struct kvm_memory_slot *memslot;
 463         int r, i;
 464         int n;
 465         unsigned long any = 0;
 466
 467         r = -EINVAL;
 468         if (log->slot >= KVM_MEMORY_SLOTS)
 469                 goto out;
 470
 471         memslot = &kvm->memslots[log->slot];
 472         r = -ENOENT;
 473         if (!memslot->dirty_bitmap)
 474                 goto out;
 475
 476         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 477
 478         for (i = 0; !any && i < n/sizeof(long); ++i)
 479                 any = memslot->dirty_bitmap[i];
 480
 481         r = -EFAULT;
 482         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 483                 goto out;
 484
 485         if (any)
 486                 *is_dirty = 1;
 487
 488         r = 0;
 489 out:
 490         return r;
 491 }
 492
 493 int is_error_page(struct page *page)
 494 {
 495         return page == bad_page;
 496 }
 497 EXPORT_SYMBOL_GPL(is_error_page);
 498
 499 int is_error_pfn(pfn_t pfn)
 500 {
 501         return pfn == bad_pfn;
 502 }
 503 EXPORT_SYMBOL_GPL(is_error_pfn);
 504
 505 static inline unsigned long bad_hva(void)
 506 {
 507         return PAGE_OFFSET;
 508 }
 509
 510 int kvm_is_error_hva(unsigned long addr)
 511 {
 512         return addr == bad_hva();
 513 }
 514 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 515
 516 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 517 {
 518         int i;
 519
 520         for (i = 0; i < kvm->nmemslots; ++i) {
 521                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 522
 523                 if (gfn >= memslot->base_gfn
 524                     && gfn < memslot->base_gfn + memslot->npages)
 525                         return memslot;
 526         }
 527         return NULL;
 528 }
 529
 530 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 531 {
 532         gfn = unalias_gfn(kvm, gfn);
 533         return __gfn_to_memslot(kvm, gfn);
 534 }
 535
 536 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 537 {
 538         int i;
 539
 540         gfn = unalias_gfn(kvm, gfn);
 541         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 542                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 543
 544                 if (gfn >= memslot->base_gfn
 545                     && gfn < memslot->base_gfn + memslot->npages)
 546                         return 1;
 547         }
 548         return 0;
 549 }
 550 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 551
 552 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 553 {
 554         struct kvm_memory_slot *slot;
 555
 556         gfn = unalias_gfn(kvm, gfn);
 557         slot = __gfn_to_memslot(kvm, gfn);
 558         if (!slot)
 559                 return bad_hva();
 560         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 561 }
 562 EXPORT_SYMBOL_GPL(gfn_to_hva);
 563
 564 /*
 565  * Requires current->mm->mmap_sem to be held
 566  */
 567 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 568 {
 569         struct page *page[1];
 570         unsigned long addr;
 571         int npages;
 572         pfn_t pfn;
 573
 574         might_sleep();
 575
 576         addr = gfn_to_hva(kvm, gfn);
 577         if (kvm_is_error_hva(addr)) {
 578                 get_page(bad_page);
 579                 return page_to_pfn(bad_page);
 580         }
 581
 582         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 583                                 NULL);
 584
 585         if (unlikely(npages != 1)) {
 586                 struct vm_area_struct *vma;
 587
 588                 vma = find_vma(current->mm, addr);
 589                 if (vma == NULL || addr < vma->vm_start ||
 590                     !(vma->vm_flags & VM_PFNMAP)) {
 591                         get_page(bad_page);
 592                         return page_to_pfn(bad_page);
 593                 }
 594
 595                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 596                 BUG_ON(pfn_valid(pfn));
 597         } else
 598                 pfn = page_to_pfn(page[0]);
 599
 600         return pfn;
 601 }
 602
 603 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 604
 605 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 606 {
 607         pfn_t pfn;
 608
 609         pfn = gfn_to_pfn(kvm, gfn);
 610         if (pfn_valid(pfn))
 611                 return pfn_to_page(pfn);
 612
 613         WARN_ON(!pfn_valid(pfn));
 614
 615         get_page(bad_page);
 616         return bad_page;
 617 }
 618
 619 EXPORT_SYMBOL_GPL(gfn_to_page);
 620
 621 void kvm_release_page_clean(struct page *page)
 622 {
 623         kvm_release_pfn_clean(page_to_pfn(page));
 624 }
 625 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 626
 627 void kvm_release_pfn_clean(pfn_t pfn)
 628 {
 629         if (pfn_valid(pfn))
 630                 put_page(pfn_to_page(pfn));
 631 }
 632 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 633
 634 void kvm_release_page_dirty(struct page *page)
 635 {
 636         kvm_release_pfn_dirty(page_to_pfn(page));
 637 }
 638 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 639
 640 void kvm_release_pfn_dirty(pfn_t pfn)
 641 {
 642         kvm_set_pfn_dirty(pfn);
 643         kvm_release_pfn_clean(pfn);
 644 }
 645 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 646
 647 void kvm_set_page_dirty(struct page *page)
 648 {
 649         kvm_set_pfn_dirty(page_to_pfn(page));
 650 }
 651 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 652
 653 void kvm_set_pfn_dirty(pfn_t pfn)
 654 {
 655         if (pfn_valid(pfn)) {
 656                 struct page *page = pfn_to_page(pfn);
 657                 if (!PageReserved(page))
 658                         SetPageDirty(page);
 659         }
 660 }
 661 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 662
 663 void kvm_set_pfn_accessed(pfn_t pfn)
 664 {
 665         if (pfn_valid(pfn))
 666                 mark_page_accessed(pfn_to_page(pfn));
 667 }
 668 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 669
 670 void kvm_get_pfn(pfn_t pfn)
 671 {
 672         if (pfn_valid(pfn))
 673                 get_page(pfn_to_page(pfn));
 674 }
 675 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 676
 677 static int next_segment(unsigned long len, int offset)
 678 {
 679         if (len > PAGE_SIZE - offset)
 680                 return PAGE_SIZE - offset;
 681         else
 682                 return len;
 683 }
 684
 685 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 686                         int len)
 687 {
 688         int r;
 689         unsigned long addr;
 690
 691         addr = gfn_to_hva(kvm, gfn);
 692         if (kvm_is_error_hva(addr))
 693                 return -EFAULT;
 694         r = copy_from_user(data, (void __user *)addr + offset, len);
 695         if (r)
 696                 return -EFAULT;
 697         return 0;
 698 }
 699 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 700
 701 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 702 {
 703         gfn_t gfn = gpa >> PAGE_SHIFT;
 704         int seg;
 705         int offset = offset_in_page(gpa);
 706         int ret;
 707
 708         while ((seg = next_segment(len, offset)) != 0) {
 709                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 710                 if (ret < 0)
 711                         return ret;
 712                 offset = 0;
 713                 len -= seg;
 714                 data += seg;
 715                 ++gfn;
 716         }
 717         return 0;
 718 }
 719 EXPORT_SYMBOL_GPL(kvm_read_guest);
 720
 721 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 722                           unsigned long len)
 723 {
 724         int r;
 725         unsigned long addr;
 726         gfn_t gfn = gpa >> PAGE_SHIFT;
 727         int offset = offset_in_page(gpa);
 728
 729         addr = gfn_to_hva(kvm, gfn);
 730         if (kvm_is_error_hva(addr))
 731                 return -EFAULT;
 732         pagefault_disable();
 733         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
 734         pagefault_enable();
 735         if (r)
 736                 return -EFAULT;
 737         return 0;
 738 }
 739 EXPORT_SYMBOL(kvm_read_guest_atomic);
 740
 741 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 742                          int offset, int len)
 743 {
 744         int r;
 745         unsigned long addr;
 746
 747         addr = gfn_to_hva(kvm, gfn);
 748         if (kvm_is_error_hva(addr))
 749                 return -EFAULT;
 750         r = copy_to_user((void __user *)addr + offset, data, len);
 751         if (r)
 752                 return -EFAULT;
 753         mark_page_dirty(kvm, gfn);
 754         return 0;
 755 }
 756 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 757
 758 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 759                     unsigned long len)
 760 {
 761         gfn_t gfn = gpa >> PAGE_SHIFT;
 762         int seg;
 763         int offset = offset_in_page(gpa);
 764         int ret;
 765
 766         while ((seg = next_segment(len, offset)) != 0) {
 767                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 768                 if (ret < 0)
 769                         return ret;
 770                 offset = 0;
 771                 len -= seg;
 772                 data += seg;
 773                 ++gfn;
 774         }
 775         return 0;
 776 }
 777
 778 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 779 {
 780         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 781 }
 782 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 783
 784 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 785 {
 786         gfn_t gfn = gpa >> PAGE_SHIFT;
 787         int seg;
 788         int offset = offset_in_page(gpa);
 789         int ret;
 790
 791         while ((seg = next_segment(len, offset)) != 0) {
 792                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 793                 if (ret < 0)
 794                         return ret;
 795                 offset = 0;
 796                 len -= seg;
 797                 ++gfn;
 798         }
 799         return 0;
 800 }
 801 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 802
 803 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 804 {
 805         struct kvm_memory_slot *memslot;
 806
 807         gfn = unalias_gfn(kvm, gfn);
 808         memslot = __gfn_to_memslot(kvm, gfn);
 809         if (memslot && memslot->dirty_bitmap) {
 810                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 811
 812                 /* avoid RMW */
 813                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 814                         set_bit(rel_gfn, memslot->dirty_bitmap);
 815         }
 816 }
 817
 818 /*
 819  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 820  */
 821 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 822 {
 823         DEFINE_WAIT(wait);
 824
 825         for (;;) {
 826                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
 827
 828                 if (kvm_cpu_has_interrupt(vcpu))
 829                         break;
 830                 if (kvm_cpu_has_pending_timer(vcpu))
 831                         break;
 832                 if (kvm_arch_vcpu_runnable(vcpu))
 833                         break;
 834                 if (signal_pending(current))
 835                         break;
 836
 837                 vcpu_put(vcpu);
 838                 schedule();
 839                 vcpu_load(vcpu);
 840         }
 841
 842         finish_wait(&vcpu->wq, &wait);
 843 }
 844
 845 void kvm_resched(struct kvm_vcpu *vcpu)
 846 {
 847         if (!need_resched())
 848                 return;
 849         cond_resched();
 850 }
 851 EXPORT_SYMBOL_GPL(kvm_resched);
 852
 853 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 854 {
 855         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 856         struct page *page;
 857
 858         if (vmf->pgoff == 0)
 859                 page = virt_to_page(vcpu->run);
 860 #ifdef CONFIG_X86
 861         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 862                 page = virt_to_page(vcpu->arch.pio_data);
 863 #endif
 864 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 865         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
 866                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
 867 #endif
 868         else
 869                 return VM_FAULT_SIGBUS;
 870         get_page(page);
 871         vmf->page = page;
 872         return 0;
 873 }
 874
 875 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 876         .fault = kvm_vcpu_fault,
 877 };
 878
 879 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 880 {
 881         vma->vm_ops = &kvm_vcpu_vm_ops;
 882         return 0;
 883 }
 884
 885 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 886 {
 887         struct kvm_vcpu *vcpu = filp->private_data;
 888
 889         kvm_put_kvm(vcpu->kvm);
 890         return 0;
 891 }
 892
 893 static const struct file_operations kvm_vcpu_fops = {
 894         .release        = kvm_vcpu_release,
 895         .unlocked_ioctl = kvm_vcpu_ioctl,
 896         .compat_ioctl   = kvm_vcpu_ioctl,
 897         .mmap           = kvm_vcpu_mmap,
 898 };
 899
 900 /*
 901  * Allocates an inode for the vcpu.
 902  */
 903 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 904 {
 905         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
 906         if (fd < 0)
 907                 kvm_put_kvm(vcpu->kvm);
 908         return fd;
 909 }
 910
 911 /*
 912  * Creates some virtual cpus.  Good luck creating more than one.
 913  */
 914 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 915 {
 916         int r;
 917         struct kvm_vcpu *vcpu;
 918
 919         if (!valid_vcpu(n))
 920                 return -EINVAL;
 921
 922         vcpu = kvm_arch_vcpu_create(kvm, n);
 923         if (IS_ERR(vcpu))
 924                 return PTR_ERR(vcpu);
 925
 926         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 927
 928         r = kvm_arch_vcpu_setup(vcpu);
 929         if (r)
 930                 goto vcpu_destroy;
 931
 932         mutex_lock(&kvm->lock);
 933         if (kvm->vcpus[n]) {
 934                 r = -EEXIST;
 935                 mutex_unlock(&kvm->lock);
 936                 goto vcpu_destroy;
 937         }
 938         kvm->vcpus[n] = vcpu;
 939         mutex_unlock(&kvm->lock);
 940
 941         /* Now it's all set up, let userspace reach it */
 942         kvm_get_kvm(kvm);
 943         r = create_vcpu_fd(vcpu);
 944         if (r < 0)
 945                 goto unlink;
 946         return r;
 947
 948 unlink:
 949         mutex_lock(&kvm->lock);
 950         kvm->vcpus[n] = NULL;
 951         mutex_unlock(&kvm->lock);
 952 vcpu_destroy:
 953         kvm_arch_vcpu_destroy(vcpu);
 954         return r;
 955 }
 956
 957 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 958 {
 959         if (sigset) {
 960                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 961                 vcpu->sigset_active = 1;
 962                 vcpu->sigset = *sigset;
 963         } else
 964                 vcpu->sigset_active = 0;
 965         return 0;
 966 }
 967
 968 static long kvm_vcpu_ioctl(struct file *filp,
 969                            unsigned int ioctl, unsigned long arg)
 970 {
 971         struct kvm_vcpu *vcpu = filp->private_data;
 972         void __user *argp = (void __user *)arg;
 973         int r;
 974
 975         if (vcpu->kvm->mm != current->mm)
 976                 return -EIO;
 977         switch (ioctl) {
 978         case KVM_RUN:
 979                 r = -EINVAL;
 980                 if (arg)
 981                         goto out;
 982                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 983                 break;
 984         case KVM_GET_REGS: {
 985                 struct kvm_regs *kvm_regs;
 986
 987                 r = -ENOMEM;
 988                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 989                 if (!kvm_regs)
 990                         goto out;
 991                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
 992                 if (r)
 993                         goto out_free1;
 994                 r = -EFAULT;
 995                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
 996                         goto out_free1;
 997                 r = 0;
 998 out_free1:
 999                 kfree(kvm_regs);
1000                 break;
1001         }
1002         case KVM_SET_REGS: {
1003                 struct kvm_regs *kvm_regs;
1004
1005                 r = -ENOMEM;
1006                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1007                 if (!kvm_regs)
1008                         goto out;
1009                 r = -EFAULT;
1010                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1011                         goto out_free2;
1012                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1013                 if (r)
1014                         goto out_free2;
1015                 r = 0;
1016 out_free2:
1017                 kfree(kvm_regs);
1018                 break;
1019         }
1020         case KVM_GET_SREGS: {
1021                 struct kvm_sregs kvm_sregs;
1022
1023                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1024                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1025                 if (r)
1026                         goto out;
1027                 r = -EFAULT;
1028                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1029                         goto out;
1030                 r = 0;
1031                 break;
1032         }
1033         case KVM_SET_SREGS: {
1034                 struct kvm_sregs kvm_sregs;
1035
1036                 r = -EFAULT;
1037                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1038                         goto out;
1039                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1040                 if (r)
1041                         goto out;
1042                 r = 0;
1043                 break;
1044         }
1045         case KVM_GET_MP_STATE: {
1046                 struct kvm_mp_state mp_state;
1047
1048                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1049                 if (r)
1050                         goto out;
1051                 r = -EFAULT;
1052                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1053                         goto out;
1054                 r = 0;
1055                 break;
1056         }
1057         case KVM_SET_MP_STATE: {
1058                 struct kvm_mp_state mp_state;
1059
1060                 r = -EFAULT;
1061                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1062                         goto out;
1063                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1064                 if (r)
1065                         goto out;
1066                 r = 0;
1067                 break;
1068         }
1069         case KVM_TRANSLATE: {
1070                 struct kvm_translation tr;
1071
1072                 r = -EFAULT;
1073                 if (copy_from_user(&tr, argp, sizeof tr))
1074                         goto out;
1075                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1076                 if (r)
1077                         goto out;
1078                 r = -EFAULT;
1079                 if (copy_to_user(argp, &tr, sizeof tr))
1080                         goto out;
1081                 r = 0;
1082                 break;
1083         }
1084         case KVM_DEBUG_GUEST: {
1085                 struct kvm_debug_guest dbg;
1086
1087                 r = -EFAULT;
1088                 if (copy_from_user(&dbg, argp, sizeof dbg))
1089                         goto out;
1090                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1091                 if (r)
1092                         goto out;
1093                 r = 0;
1094                 break;
1095         }
1096         case KVM_SET_SIGNAL_MASK: {
1097                 struct kvm_signal_mask __user *sigmask_arg = argp;
1098                 struct kvm_signal_mask kvm_sigmask;
1099                 sigset_t sigset, *p;
1100
1101                 p = NULL;
1102                 if (argp) {
1103                         r = -EFAULT;
1104                         if (copy_from_user(&kvm_sigmask, argp,
1105                                            sizeof kvm_sigmask))
1106                                 goto out;
1107                         r = -EINVAL;
1108                         if (kvm_sigmask.len != sizeof sigset)
1109                                 goto out;
1110                         r = -EFAULT;
1111                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1112                                            sizeof sigset))
1113                                 goto out;
1114                         p = &sigset;
1115                 }
1116                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1117                 break;
1118         }
1119         case KVM_GET_FPU: {
1120                 struct kvm_fpu fpu;
1121
1122                 memset(&fpu, 0, sizeof fpu);
1123                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1124                 if (r)
1125                         goto out;
1126                 r = -EFAULT;
1127                 if (copy_to_user(argp, &fpu, sizeof fpu))
1128                         goto out;
1129                 r = 0;
1130                 break;
1131         }
1132         case KVM_SET_FPU: {
1133                 struct kvm_fpu fpu;
1134
1135                 r = -EFAULT;
1136                 if (copy_from_user(&fpu, argp, sizeof fpu))
1137                         goto out;
1138                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1139                 if (r)
1140                         goto out;
1141                 r = 0;
1142                 break;
1143         }
1144         default:
1145                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1146         }
1147 out:
1148         return r;
1149 }
1150
1151 static long kvm_vm_ioctl(struct file *filp,
1152                            unsigned int ioctl, unsigned long arg)
1153 {
1154         struct kvm *kvm = filp->private_data;
1155         void __user *argp = (void __user *)arg;
1156         int r;
1157
1158         if (kvm->mm != current->mm)
1159                 return -EIO;
1160         switch (ioctl) {
1161         case KVM_CREATE_VCPU:
1162                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1163                 if (r < 0)
1164                         goto out;
1165                 break;
1166         case KVM_SET_USER_MEMORY_REGION: {
1167                 struct kvm_userspace_memory_region kvm_userspace_mem;
1168
1169                 r = -EFAULT;
1170                 if (copy_from_user(&kvm_userspace_mem, argp,
1171                                                 sizeof kvm_userspace_mem))
1172                         goto out;
1173
1174                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1175                 if (r)
1176                         goto out;
1177                 break;
1178         }
1179         case KVM_GET_DIRTY_LOG: {
1180                 struct kvm_dirty_log log;
1181
1182                 r = -EFAULT;
1183                 if (copy_from_user(&log, argp, sizeof log))
1184                         goto out;
1185                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1186                 if (r)
1187                         goto out;
1188                 break;
1189         }
1190 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1191         case KVM_REGISTER_COALESCED_MMIO: {
1192                 struct kvm_coalesced_mmio_zone zone;
1193                 r = -EFAULT;
1194                 if (copy_from_user(&zone, argp, sizeof zone))
1195                         goto out;
1196                 r = -ENXIO;
1197                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1198                 if (r)
1199                         goto out;
1200                 r = 0;
1201                 break;
1202         }
1203         case KVM_UNREGISTER_COALESCED_MMIO: {
1204                 struct kvm_coalesced_mmio_zone zone;
1205                 r = -EFAULT;
1206                 if (copy_from_user(&zone, argp, sizeof zone))
1207                         goto out;
1208                 r = -ENXIO;
1209                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1210                 if (r)
1211                         goto out;
1212                 r = 0;
1213                 break;
1214         }
1215 #endif
1216         default:
1217                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1218         }
1219 out:
1220         return r;
1221 }
1222
1223 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1224 {
1225         struct kvm *kvm = vma->vm_file->private_data;
1226         struct page *page;
1227
1228         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1229                 return VM_FAULT_SIGBUS;
1230         page = gfn_to_page(kvm, vmf->pgoff);
1231         if (is_error_page(page)) {
1232                 kvm_release_page_clean(page);
1233                 return VM_FAULT_SIGBUS;
1234         }
1235         vmf->page = page;
1236         return 0;
1237 }
1238
1239 static struct vm_operations_struct kvm_vm_vm_ops = {
1240         .fault = kvm_vm_fault,
1241 };
1242
1243 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1244 {
1245         vma->vm_ops = &kvm_vm_vm_ops;
1246         return 0;
1247 }
1248
1249 static const struct file_operations kvm_vm_fops = {
1250         .release        = kvm_vm_release,
1251         .unlocked_ioctl = kvm_vm_ioctl,
1252         .compat_ioctl   = kvm_vm_ioctl,
1253         .mmap           = kvm_vm_mmap,
1254 };
1255
1256 static int kvm_dev_ioctl_create_vm(void)
1257 {
1258         int fd;
1259         struct kvm *kvm;
1260
1261         kvm = kvm_create_vm();
1262         if (IS_ERR(kvm))
1263                 return PTR_ERR(kvm);
1264         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1265         if (fd < 0)
1266                 kvm_put_kvm(kvm);
1267
1268         return fd;
1269 }
1270
1271 static long kvm_dev_ioctl(struct file *filp,
1272                           unsigned int ioctl, unsigned long arg)
1273 {
1274         long r = -EINVAL;
1275
1276         switch (ioctl) {
1277         case KVM_GET_API_VERSION:
1278                 r = -EINVAL;
1279                 if (arg)
1280                         goto out;
1281                 r = KVM_API_VERSION;
1282                 break;
1283         case KVM_CREATE_VM:
1284                 r = -EINVAL;
1285                 if (arg)
1286                         goto out;
1287                 r = kvm_dev_ioctl_create_vm();
1288                 break;
1289         case KVM_CHECK_EXTENSION:
1290                 r = kvm_dev_ioctl_check_extension(arg);
1291                 break;
1292         case KVM_GET_VCPU_MMAP_SIZE:
1293                 r = -EINVAL;
1294                 if (arg)
1295                         goto out;
1296                 r = PAGE_SIZE;     /* struct kvm_run */
1297 #ifdef CONFIG_X86
1298                 r += PAGE_SIZE;    /* pio data page */
1299 #endif
1300 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1301                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1302 #endif
1303                 break;
1304         case KVM_TRACE_ENABLE:
1305         case KVM_TRACE_PAUSE:
1306         case KVM_TRACE_DISABLE:
1307                 r = kvm_trace_ioctl(ioctl, arg);
1308                 break;
1309         default:
1310                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1311         }
1312 out:
1313         return r;
1314 }
1315
1316 static struct file_operations kvm_chardev_ops = {
1317         .unlocked_ioctl = kvm_dev_ioctl,
1318         .compat_ioctl   = kvm_dev_ioctl,
1319 };
1320
1321 static struct miscdevice kvm_dev = {
1322         KVM_MINOR,
1323         "kvm",
1324         &kvm_chardev_ops,
1325 };
1326
1327 static void hardware_enable(void *junk)
1328 {
1329         int cpu = raw_smp_processor_id();
1330
1331         if (cpu_isset(cpu, cpus_hardware_enabled))
1332                 return;
1333         cpu_set(cpu, cpus_hardware_enabled);
1334         kvm_arch_hardware_enable(NULL);
1335 }
1336
1337 static void hardware_disable(void *junk)
1338 {
1339         int cpu = raw_smp_processor_id();
1340
1341         if (!cpu_isset(cpu, cpus_hardware_enabled))
1342                 return;
1343         cpu_clear(cpu, cpus_hardware_enabled);
1344         kvm_arch_hardware_disable(NULL);
1345 }
1346
1347 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1348                            void *v)
1349 {
1350         int cpu = (long)v;
1351
1352         val &= ~CPU_TASKS_FROZEN;
1353         switch (val) {
1354         case CPU_DYING:
1355                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1356                        cpu);
1357                 hardware_disable(NULL);
1358                 break;
1359         case CPU_UP_CANCELED:
1360                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1361                        cpu);
1362                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1363                 break;
1364         case CPU_ONLINE:
1365                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1366                        cpu);
1367                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1368                 break;
1369         }
1370         return NOTIFY_OK;
1371 }
1372
1373
1374 asmlinkage void kvm_handle_fault_on_reboot(void)
1375 {
1376         if (kvm_rebooting)
1377                 /* spin while reset goes on */
1378                 while (true)
1379                         ;
1380         /* Fault while not rebooting.  We want the trace. */
1381         BUG();
1382 }
1383 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1384
1385 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1386                       void *v)
1387 {
1388         if (val == SYS_RESTART) {
1389                 /*
1390                  * Some (well, at least mine) BIOSes hang on reboot if
1391                  * in vmx root mode.
1392                  */
1393                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1394                 kvm_rebooting = true;
1395                 on_each_cpu(hardware_disable, NULL, 1);
1396         }
1397         return NOTIFY_OK;
1398 }
1399
1400 static struct notifier_block kvm_reboot_notifier = {
1401         .notifier_call = kvm_reboot,
1402         .priority = 0,
1403 };
1404
1405 void kvm_io_bus_init(struct kvm_io_bus *bus)
1406 {
1407         memset(bus, 0, sizeof(*bus));
1408 }
1409
1410 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1411 {
1412         int i;
1413
1414         for (i = 0; i < bus->dev_count; i++) {
1415                 struct kvm_io_device *pos = bus->devs[i];
1416
1417                 kvm_iodevice_destructor(pos);
1418         }
1419 }
1420
1421 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1422                                           gpa_t addr, int len, int is_write)
1423 {
1424         int i;
1425
1426         for (i = 0; i < bus->dev_count; i++) {
1427                 struct kvm_io_device *pos = bus->devs[i];
1428
1429                 if (pos->in_range(pos, addr, len, is_write))
1430                         return pos;
1431         }
1432
1433         return NULL;
1434 }
1435
1436 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1437 {
1438         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1439
1440         bus->devs[bus->dev_count++] = dev;
1441 }
1442
1443 static struct notifier_block kvm_cpu_notifier = {
1444         .notifier_call = kvm_cpu_hotplug,
1445         .priority = 20, /* must be > scheduler priority */
1446 };
1447
1448 static int vm_stat_get(void *_offset, u64 *val)
1449 {
1450         unsigned offset = (long)_offset;
1451         struct kvm *kvm;
1452
1453         *val = 0;
1454         spin_lock(&kvm_lock);
1455         list_for_each_entry(kvm, &vm_list, vm_list)
1456                 *val += *(u32 *)((void *)kvm + offset);
1457         spin_unlock(&kvm_lock);
1458         return 0;
1459 }
1460
1461 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1462
1463 static int vcpu_stat_get(void *_offset, u64 *val)
1464 {
1465         unsigned offset = (long)_offset;
1466         struct kvm *kvm;
1467         struct kvm_vcpu *vcpu;
1468         int i;
1469
1470         *val = 0;
1471         spin_lock(&kvm_lock);
1472         list_for_each_entry(kvm, &vm_list, vm_list)
1473                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1474                         vcpu = kvm->vcpus[i];
1475                         if (vcpu)
1476                                 *val += *(u32 *)((void *)vcpu + offset);
1477                 }
1478         spin_unlock(&kvm_lock);
1479         return 0;
1480 }
1481
1482 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1483
1484 static struct file_operations *stat_fops[] = {
1485         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1486         [KVM_STAT_VM]   = &vm_stat_fops,
1487 };
1488
1489 static void kvm_init_debug(void)
1490 {
1491         struct kvm_stats_debugfs_item *p;
1492
1493         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1494         for (p = debugfs_entries; p->name; ++p)
1495                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1496                                                 (void *)(long)p->offset,
1497                                                 stat_fops[p->kind]);
1498 }
1499
1500 static void kvm_exit_debug(void)
1501 {
1502         struct kvm_stats_debugfs_item *p;
1503
1504         for (p = debugfs_entries; p->name; ++p)
1505                 debugfs_remove(p->dentry);
1506         debugfs_remove(kvm_debugfs_dir);
1507 }
1508
1509 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1510 {
1511         hardware_disable(NULL);
1512         return 0;
1513 }
1514
1515 static int kvm_resume(struct sys_device *dev)
1516 {
1517         hardware_enable(NULL);
1518         return 0;
1519 }
1520
1521 static struct sysdev_class kvm_sysdev_class = {
1522         .name = "kvm",
1523         .suspend = kvm_suspend,
1524         .resume = kvm_resume,
1525 };
1526
1527 static struct sys_device kvm_sysdev = {
1528         .id = 0,
1529         .cls = &kvm_sysdev_class,
1530 };
1531
1532 struct page *bad_page;
1533 pfn_t bad_pfn;
1534
1535 static inline
1536 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1537 {
1538         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1539 }
1540
1541 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1542 {
1543         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1544
1545         kvm_arch_vcpu_load(vcpu, cpu);
1546 }
1547
1548 static void kvm_sched_out(struct preempt_notifier *pn,
1549                           struct task_struct *next)
1550 {
1551         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1552
1553         kvm_arch_vcpu_put(vcpu);
1554 }
1555
1556 int kvm_init(void *opaque, unsigned int vcpu_size,
1557                   struct module *module)
1558 {
1559         int r;
1560         int cpu;
1561
1562         kvm_init_debug();
1563
1564         r = kvm_arch_init(opaque);
1565         if (r)
1566                 goto out_fail;
1567
1568         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1569
1570         if (bad_page == NULL) {
1571                 r = -ENOMEM;
1572                 goto out;
1573         }
1574
1575         bad_pfn = page_to_pfn(bad_page);
1576
1577         r = kvm_arch_hardware_setup();
1578         if (r < 0)
1579                 goto out_free_0;
1580
1581         for_each_online_cpu(cpu) {
1582                 smp_call_function_single(cpu,
1583                                 kvm_arch_check_processor_compat,
1584                                 &r, 1);
1585                 if (r < 0)
1586                         goto out_free_1;
1587         }
1588
1589         on_each_cpu(hardware_enable, NULL, 1);
1590         r = register_cpu_notifier(&kvm_cpu_notifier);
1591         if (r)
1592                 goto out_free_2;
1593         register_reboot_notifier(&kvm_reboot_notifier);
1594
1595         r = sysdev_class_register(&kvm_sysdev_class);
1596         if (r)
1597                 goto out_free_3;
1598
1599         r = sysdev_register(&kvm_sysdev);
1600         if (r)
1601                 goto out_free_4;
1602
1603         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1604         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1605                                            __alignof__(struct kvm_vcpu),
1606                                            0, NULL);
1607         if (!kvm_vcpu_cache) {
1608                 r = -ENOMEM;
1609                 goto out_free_5;
1610         }
1611
1612         kvm_chardev_ops.owner = module;
1613
1614         r = misc_register(&kvm_dev);
1615         if (r) {
1616                 printk(KERN_ERR "kvm: misc device register failed\n");
1617                 goto out_free;
1618         }
1619
1620         kvm_preempt_ops.sched_in = kvm_sched_in;
1621         kvm_preempt_ops.sched_out = kvm_sched_out;
1622
1623         return 0;
1624
1625 out_free:
1626         kmem_cache_destroy(kvm_vcpu_cache);
1627 out_free_5:
1628         sysdev_unregister(&kvm_sysdev);
1629 out_free_4:
1630         sysdev_class_unregister(&kvm_sysdev_class);
1631 out_free_3:
1632         unregister_reboot_notifier(&kvm_reboot_notifier);
1633         unregister_cpu_notifier(&kvm_cpu_notifier);
1634 out_free_2:
1635         on_each_cpu(hardware_disable, NULL, 1);
1636 out_free_1:
1637         kvm_arch_hardware_unsetup();
1638 out_free_0:
1639         __free_page(bad_page);
1640 out:
1641         kvm_arch_exit();
1642         kvm_exit_debug();
1643 out_fail:
1644         return r;
1645 }
1646 EXPORT_SYMBOL_GPL(kvm_init);
1647
1648 void kvm_exit(void)
1649 {
1650         kvm_trace_cleanup();
1651         misc_deregister(&kvm_dev);
1652         kmem_cache_destroy(kvm_vcpu_cache);
1653         sysdev_unregister(&kvm_sysdev);
1654         sysdev_class_unregister(&kvm_sysdev_class);
1655         unregister_reboot_notifier(&kvm_reboot_notifier);
1656         unregister_cpu_notifier(&kvm_cpu_notifier);
1657         on_each_cpu(hardware_disable, NULL, 1);
1658         kvm_arch_hardware_unsetup();
1659         kvm_arch_exit();
1660         kvm_exit_debug();
1661         __free_page(bad_page);
1662 }
1663 EXPORT_SYMBOL_GPL(kvm_exit);