err.no 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                 /*
 379                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 380                  * safe it has to ignore memslots with !user_alloc &&
 381                  * !userspace_addr.
 382                  */
 383                 if (user_alloc)
 384                         new.userspace_addr = mem->userspace_addr;
 385                 else
 386                         new.userspace_addr = 0;
 387         }
 388         if (npages && !new.lpage_info) {
 389                 int largepages = npages / KVM_PAGES_PER_HPAGE;
 390                 if (npages % KVM_PAGES_PER_HPAGE)
 391                         largepages++;
 392                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 393                         largepages++;
 394
 395                 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
 396
 397                 if (!new.lpage_info)
 398                         goto out_free;
 399
 400                 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
 401
 402                 if (base_gfn % KVM_PAGES_PER_HPAGE)
 403                         new.lpage_info[0].write_count = 1;
 404                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
 405                         new.lpage_info[largepages-1].write_count = 1;
 406         }
 407
 408         /* Allocate page dirty bitmap if needed */
 409         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 410                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 411
 412                 new.dirty_bitmap = vmalloc(dirty_bytes);
 413                 if (!new.dirty_bitmap)
 414                         goto out_free;
 415                 memset(new.dirty_bitmap, 0, dirty_bytes);
 416         }
 417 #endif /* not defined CONFIG_S390 */
 418
 419         if (!npages)
 420                 kvm_arch_flush_shadow(kvm);
 421
 422         spin_lock(&kvm->mmu_lock);
 423         if (mem->slot >= kvm->nmemslots)
 424                 kvm->nmemslots = mem->slot + 1;
 425
 426         *memslot = new;
 427         spin_unlock(&kvm->mmu_lock);
 428
 429         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 430         if (r) {
 431                 spin_lock(&kvm->mmu_lock);
 432                 *memslot = old;
 433                 spin_unlock(&kvm->mmu_lock);
 434                 goto out_free;
 435         }
 436
 437         kvm_free_physmem_slot(&old, &new);
 438         return 0;
 439
 440 out_free:
 441         kvm_free_physmem_slot(&new, &old);
 442 out:
 443         return r;
 444
 445 }
 446 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 447
 448 int kvm_set_memory_region(struct kvm *kvm,
 449                           struct kvm_userspace_memory_region *mem,
 450                           int user_alloc)
 451 {
 452         int r;
 453
 454         down_write(&kvm->slots_lock);
 455         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 456         up_write(&kvm->slots_lock);
 457         return r;
 458 }
 459 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 460
 461 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 462                                    struct
 463                                    kvm_userspace_memory_region *mem,
 464                                    int user_alloc)
 465 {
 466         if (mem->slot >= KVM_MEMORY_SLOTS)
 467                 return -EINVAL;
 468         return kvm_set_memory_region(kvm, mem, user_alloc);
 469 }
 470
 471 int kvm_get_dirty_log(struct kvm *kvm,
 472                         struct kvm_dirty_log *log, int *is_dirty)
 473 {
 474         struct kvm_memory_slot *memslot;
 475         int r, i;
 476         int n;
 477         unsigned long any = 0;
 478
 479         r = -EINVAL;
 480         if (log->slot >= KVM_MEMORY_SLOTS)
 481                 goto out;
 482
 483         memslot = &kvm->memslots[log->slot];
 484         r = -ENOENT;
 485         if (!memslot->dirty_bitmap)
 486                 goto out;
 487
 488         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 489
 490         for (i = 0; !any && i < n/sizeof(long); ++i)
 491                 any = memslot->dirty_bitmap[i];
 492
 493         r = -EFAULT;
 494         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 495                 goto out;
 496
 497         if (any)
 498                 *is_dirty = 1;
 499
 500         r = 0;
 501 out:
 502         return r;
 503 }
 504
 505 int is_error_page(struct page *page)
 506 {
 507         return page == bad_page;
 508 }
 509 EXPORT_SYMBOL_GPL(is_error_page);
 510
 511 int is_error_pfn(pfn_t pfn)
 512 {
 513         return pfn == bad_pfn;
 514 }
 515 EXPORT_SYMBOL_GPL(is_error_pfn);
 516
 517 static inline unsigned long bad_hva(void)
 518 {
 519         return PAGE_OFFSET;
 520 }
 521
 522 int kvm_is_error_hva(unsigned long addr)
 523 {
 524         return addr == bad_hva();
 525 }
 526 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 527
 528 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 529 {
 530         int i;
 531
 532         for (i = 0; i < kvm->nmemslots; ++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 memslot;
 538         }
 539         return NULL;
 540 }
 541
 542 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 543 {
 544         gfn = unalias_gfn(kvm, gfn);
 545         return __gfn_to_memslot(kvm, gfn);
 546 }
 547
 548 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 549 {
 550         int i;
 551
 552         gfn = unalias_gfn(kvm, gfn);
 553         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 554                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 555
 556                 if (gfn >= memslot->base_gfn
 557                     && gfn < memslot->base_gfn + memslot->npages)
 558                         return 1;
 559         }
 560         return 0;
 561 }
 562 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 563
 564 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 565 {
 566         struct kvm_memory_slot *slot;
 567
 568         gfn = unalias_gfn(kvm, gfn);
 569         slot = __gfn_to_memslot(kvm, gfn);
 570         if (!slot)
 571                 return bad_hva();
 572         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 573 }
 574 EXPORT_SYMBOL_GPL(gfn_to_hva);
 575
 576 /*
 577  * Requires current->mm->mmap_sem to be held
 578  */
 579 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 580 {
 581         struct page *page[1];
 582         unsigned long addr;
 583         int npages;
 584         pfn_t pfn;
 585
 586         might_sleep();
 587
 588         addr = gfn_to_hva(kvm, gfn);
 589         if (kvm_is_error_hva(addr)) {
 590                 get_page(bad_page);
 591                 return page_to_pfn(bad_page);
 592         }
 593
 594         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 595                                 NULL);
 596
 597         if (unlikely(npages != 1)) {
 598                 struct vm_area_struct *vma;
 599
 600                 vma = find_vma(current->mm, addr);
 601                 if (vma == NULL || addr < vma->vm_start ||
 602                     !(vma->vm_flags & VM_PFNMAP)) {
 603                         get_page(bad_page);
 604                         return page_to_pfn(bad_page);
 605                 }
 606
 607                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 608                 BUG_ON(pfn_valid(pfn));
 609         } else
 610                 pfn = page_to_pfn(page[0]);
 611
 612         return pfn;
 613 }
 614
 615 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 616
 617 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 618 {
 619         pfn_t pfn;
 620
 621         pfn = gfn_to_pfn(kvm, gfn);
 622         if (pfn_valid(pfn))
 623                 return pfn_to_page(pfn);
 624
 625         WARN_ON(!pfn_valid(pfn));
 626
 627         get_page(bad_page);
 628         return bad_page;
 629 }
 630
 631 EXPORT_SYMBOL_GPL(gfn_to_page);
 632
 633 void kvm_release_page_clean(struct page *page)
 634 {
 635         kvm_release_pfn_clean(page_to_pfn(page));
 636 }
 637 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 638
 639 void kvm_release_pfn_clean(pfn_t pfn)
 640 {
 641         if (pfn_valid(pfn))
 642                 put_page(pfn_to_page(pfn));
 643 }
 644 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 645
 646 void kvm_release_page_dirty(struct page *page)
 647 {
 648         kvm_release_pfn_dirty(page_to_pfn(page));
 649 }
 650 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 651
 652 void kvm_release_pfn_dirty(pfn_t pfn)
 653 {
 654         kvm_set_pfn_dirty(pfn);
 655         kvm_release_pfn_clean(pfn);
 656 }
 657 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 658
 659 void kvm_set_page_dirty(struct page *page)
 660 {
 661         kvm_set_pfn_dirty(page_to_pfn(page));
 662 }
 663 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 664
 665 void kvm_set_pfn_dirty(pfn_t pfn)
 666 {
 667         if (pfn_valid(pfn)) {
 668                 struct page *page = pfn_to_page(pfn);
 669                 if (!PageReserved(page))
 670                         SetPageDirty(page);
 671         }
 672 }
 673 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 674
 675 void kvm_set_pfn_accessed(pfn_t pfn)
 676 {
 677         if (pfn_valid(pfn))
 678                 mark_page_accessed(pfn_to_page(pfn));
 679 }
 680 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 681
 682 void kvm_get_pfn(pfn_t pfn)
 683 {
 684         if (pfn_valid(pfn))
 685                 get_page(pfn_to_page(pfn));
 686 }
 687 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 688
 689 static int next_segment(unsigned long len, int offset)
 690 {
 691         if (len > PAGE_SIZE - offset)
 692                 return PAGE_SIZE - offset;
 693         else
 694                 return len;
 695 }
 696
 697 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 698                         int len)
 699 {
 700         int r;
 701         unsigned long addr;
 702
 703         addr = gfn_to_hva(kvm, gfn);
 704         if (kvm_is_error_hva(addr))
 705                 return -EFAULT;
 706         r = copy_from_user(data, (void __user *)addr + offset, len);
 707         if (r)
 708                 return -EFAULT;
 709         return 0;
 710 }
 711 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 712
 713 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 714 {
 715         gfn_t gfn = gpa >> PAGE_SHIFT;
 716         int seg;
 717         int offset = offset_in_page(gpa);
 718         int ret;
 719
 720         while ((seg = next_segment(len, offset)) != 0) {
 721                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 722                 if (ret < 0)
 723                         return ret;
 724                 offset = 0;
 725                 len -= seg;
 726                 data += seg;
 727                 ++gfn;
 728         }
 729         return 0;
 730 }
 731 EXPORT_SYMBOL_GPL(kvm_read_guest);
 732
 733 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 734                           unsigned long len)
 735 {
 736         int r;
 737         unsigned long addr;
 738         gfn_t gfn = gpa >> PAGE_SHIFT;
 739         int offset = offset_in_page(gpa);
 740
 741         addr = gfn_to_hva(kvm, gfn);
 742         if (kvm_is_error_hva(addr))
 743                 return -EFAULT;
 744         pagefault_disable();
 745         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
 746         pagefault_enable();
 747         if (r)
 748                 return -EFAULT;
 749         return 0;
 750 }
 751 EXPORT_SYMBOL(kvm_read_guest_atomic);
 752
 753 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 754                          int offset, int len)
 755 {
 756         int r;
 757         unsigned long addr;
 758
 759         addr = gfn_to_hva(kvm, gfn);
 760         if (kvm_is_error_hva(addr))
 761                 return -EFAULT;
 762         r = copy_to_user((void __user *)addr + offset, data, len);
 763         if (r)
 764                 return -EFAULT;
 765         mark_page_dirty(kvm, gfn);
 766         return 0;
 767 }
 768 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 769
 770 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 771                     unsigned long len)
 772 {
 773         gfn_t gfn = gpa >> PAGE_SHIFT;
 774         int seg;
 775         int offset = offset_in_page(gpa);
 776         int ret;
 777
 778         while ((seg = next_segment(len, offset)) != 0) {
 779                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 780                 if (ret < 0)
 781                         return ret;
 782                 offset = 0;
 783                 len -= seg;
 784                 data += seg;
 785                 ++gfn;
 786         }
 787         return 0;
 788 }
 789
 790 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 791 {
 792         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
 793 }
 794 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 795
 796 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 797 {
 798         gfn_t gfn = gpa >> PAGE_SHIFT;
 799         int seg;
 800         int offset = offset_in_page(gpa);
 801         int ret;
 802
 803         while ((seg = next_segment(len, offset)) != 0) {
 804                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 805                 if (ret < 0)
 806                         return ret;
 807                 offset = 0;
 808                 len -= seg;
 809                 ++gfn;
 810         }
 811         return 0;
 812 }
 813 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 814
 815 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 816 {
 817         struct kvm_memory_slot *memslot;
 818
 819         gfn = unalias_gfn(kvm, gfn);
 820         memslot = __gfn_to_memslot(kvm, gfn);
 821         if (memslot && memslot->dirty_bitmap) {
 822                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 823
 824                 /* avoid RMW */
 825                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 826                         set_bit(rel_gfn, memslot->dirty_bitmap);
 827         }
 828 }
 829
 830 /*
 831  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 832  */
 833 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 834 {
 835         DEFINE_WAIT(wait);
 836
 837         for (;;) {
 838                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
 839
 840                 if (kvm_cpu_has_interrupt(vcpu))
 841                         break;
 842                 if (kvm_cpu_has_pending_timer(vcpu))
 843                         break;
 844                 if (kvm_arch_vcpu_runnable(vcpu))
 845                         break;
 846                 if (signal_pending(current))
 847                         break;
 848
 849                 vcpu_put(vcpu);
 850                 schedule();
 851                 vcpu_load(vcpu);
 852         }
 853
 854         finish_wait(&vcpu->wq, &wait);
 855 }
 856
 857 void kvm_resched(struct kvm_vcpu *vcpu)
 858 {
 859         if (!need_resched())
 860                 return;
 861         cond_resched();
 862 }
 863 EXPORT_SYMBOL_GPL(kvm_resched);
 864
 865 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 866 {
 867         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 868         struct page *page;
 869
 870         if (vmf->pgoff == 0)
 871                 page = virt_to_page(vcpu->run);
 872 #ifdef CONFIG_X86
 873         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 874                 page = virt_to_page(vcpu->arch.pio_data);
 875 #endif
 876 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 877         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
 878                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
 879 #endif
 880         else
 881                 return VM_FAULT_SIGBUS;
 882         get_page(page);
 883         vmf->page = page;
 884         return 0;
 885 }
 886
 887 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 888         .fault = kvm_vcpu_fault,
 889 };
 890
 891 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 892 {
 893         vma->vm_ops = &kvm_vcpu_vm_ops;
 894         return 0;
 895 }
 896
 897 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 898 {
 899         struct kvm_vcpu *vcpu = filp->private_data;
 900
 901         kvm_put_kvm(vcpu->kvm);
 902         return 0;
 903 }
 904
 905 static const struct file_operations kvm_vcpu_fops = {
 906         .release        = kvm_vcpu_release,
 907         .unlocked_ioctl = kvm_vcpu_ioctl,
 908         .compat_ioctl   = kvm_vcpu_ioctl,
 909         .mmap           = kvm_vcpu_mmap,
 910 };
 911
 912 /*
 913  * Allocates an inode for the vcpu.
 914  */
 915 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 916 {
 917         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
 918         if (fd < 0)
 919                 kvm_put_kvm(vcpu->kvm);
 920         return fd;
 921 }
 922
 923 /*
 924  * Creates some virtual cpus.  Good luck creating more than one.
 925  */
 926 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 927 {
 928         int r;
 929         struct kvm_vcpu *vcpu;
 930
 931         if (!valid_vcpu(n))
 932                 return -EINVAL;
 933
 934         vcpu = kvm_arch_vcpu_create(kvm, n);
 935         if (IS_ERR(vcpu))
 936                 return PTR_ERR(vcpu);
 937
 938         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 939
 940         r = kvm_arch_vcpu_setup(vcpu);
 941         if (r)
 942                 goto vcpu_destroy;
 943
 944         mutex_lock(&kvm->lock);
 945         if (kvm->vcpus[n]) {
 946                 r = -EEXIST;
 947                 mutex_unlock(&kvm->lock);
 948                 goto vcpu_destroy;
 949         }
 950         kvm->vcpus[n] = vcpu;
 951         mutex_unlock(&kvm->lock);
 952
 953         /* Now it's all set up, let userspace reach it */
 954         kvm_get_kvm(kvm);
 955         r = create_vcpu_fd(vcpu);
 956         if (r < 0)
 957                 goto unlink;
 958         return r;
 959
 960 unlink:
 961         mutex_lock(&kvm->lock);
 962         kvm->vcpus[n] = NULL;
 963         mutex_unlock(&kvm->lock);
 964 vcpu_destroy:
 965         kvm_arch_vcpu_destroy(vcpu);
 966         return r;
 967 }
 968
 969 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 970 {
 971         if (sigset) {
 972                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 973                 vcpu->sigset_active = 1;
 974                 vcpu->sigset = *sigset;
 975         } else
 976                 vcpu->sigset_active = 0;
 977         return 0;
 978 }
 979
 980 static long kvm_vcpu_ioctl(struct file *filp,
 981                            unsigned int ioctl, unsigned long arg)
 982 {
 983         struct kvm_vcpu *vcpu = filp->private_data;
 984         void __user *argp = (void __user *)arg;
 985         int r;
 986
 987         if (vcpu->kvm->mm != current->mm)
 988                 return -EIO;
 989         switch (ioctl) {
 990         case KVM_RUN:
 991                 r = -EINVAL;
 992                 if (arg)
 993                         goto out;
 994                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 995                 break;
 996         case KVM_GET_REGS: {
 997                 struct kvm_regs *kvm_regs;
 998
 999                 r = -ENOMEM;
1000                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1001                 if (!kvm_regs)
1002                         goto out;
1003                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1004                 if (r)
1005                         goto out_free1;
1006                 r = -EFAULT;
1007                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1008                         goto out_free1;
1009                 r = 0;
1010 out_free1:
1011                 kfree(kvm_regs);
1012                 break;
1013         }
1014         case KVM_SET_REGS: {
1015                 struct kvm_regs *kvm_regs;
1016
1017                 r = -ENOMEM;
1018                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1019                 if (!kvm_regs)
1020                         goto out;
1021                 r = -EFAULT;
1022                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1023                         goto out_free2;
1024                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1025                 if (r)
1026                         goto out_free2;
1027                 r = 0;
1028 out_free2:
1029                 kfree(kvm_regs);
1030                 break;
1031         }
1032         case KVM_GET_SREGS: {
1033                 struct kvm_sregs kvm_sregs;
1034
1035                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1036                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1037                 if (r)
1038                         goto out;
1039                 r = -EFAULT;
1040                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1041                         goto out;
1042                 r = 0;
1043                 break;
1044         }
1045         case KVM_SET_SREGS: {
1046                 struct kvm_sregs kvm_sregs;
1047
1048                 r = -EFAULT;
1049                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1050                         goto out;
1051                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1052                 if (r)
1053                         goto out;
1054                 r = 0;
1055                 break;
1056         }
1057         case KVM_GET_MP_STATE: {
1058                 struct kvm_mp_state mp_state;
1059
1060                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1061                 if (r)
1062                         goto out;
1063                 r = -EFAULT;
1064                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1065                         goto out;
1066                 r = 0;
1067                 break;
1068         }
1069         case KVM_SET_MP_STATE: {
1070                 struct kvm_mp_state mp_state;
1071
1072                 r = -EFAULT;
1073                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1074                         goto out;
1075                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1076                 if (r)
1077                         goto out;
1078                 r = 0;
1079                 break;
1080         }
1081         case KVM_TRANSLATE: {
1082                 struct kvm_translation tr;
1083
1084                 r = -EFAULT;
1085                 if (copy_from_user(&tr, argp, sizeof tr))
1086                         goto out;
1087                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1088                 if (r)
1089                         goto out;
1090                 r = -EFAULT;
1091                 if (copy_to_user(argp, &tr, sizeof tr))
1092                         goto out;
1093                 r = 0;
1094                 break;
1095         }
1096         case KVM_DEBUG_GUEST: {
1097                 struct kvm_debug_guest dbg;
1098
1099                 r = -EFAULT;
1100                 if (copy_from_user(&dbg, argp, sizeof dbg))
1101                         goto out;
1102                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1103                 if (r)
1104                         goto out;
1105                 r = 0;
1106                 break;
1107         }
1108         case KVM_SET_SIGNAL_MASK: {
1109                 struct kvm_signal_mask __user *sigmask_arg = argp;
1110                 struct kvm_signal_mask kvm_sigmask;
1111                 sigset_t sigset, *p;
1112
1113                 p = NULL;
1114                 if (argp) {
1115                         r = -EFAULT;
1116                         if (copy_from_user(&kvm_sigmask, argp,
1117                                            sizeof kvm_sigmask))
1118                                 goto out;
1119                         r = -EINVAL;
1120                         if (kvm_sigmask.len != sizeof sigset)
1121                                 goto out;
1122                         r = -EFAULT;
1123                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1124                                            sizeof sigset))
1125                                 goto out;
1126                         p = &sigset;
1127                 }
1128                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1129                 break;
1130         }
1131         case KVM_GET_FPU: {
1132                 struct kvm_fpu fpu;
1133
1134                 memset(&fpu, 0, sizeof fpu);
1135                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1136                 if (r)
1137                         goto out;
1138                 r = -EFAULT;
1139                 if (copy_to_user(argp, &fpu, sizeof fpu))
1140                         goto out;
1141                 r = 0;
1142                 break;
1143         }
1144         case KVM_SET_FPU: {
1145                 struct kvm_fpu fpu;
1146
1147                 r = -EFAULT;
1148                 if (copy_from_user(&fpu, argp, sizeof fpu))
1149                         goto out;
1150                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1151                 if (r)
1152                         goto out;
1153                 r = 0;
1154                 break;
1155         }
1156         default:
1157                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1158         }
1159 out:
1160         return r;
1161 }
1162
1163 static long kvm_vm_ioctl(struct file *filp,
1164                            unsigned int ioctl, unsigned long arg)
1165 {
1166         struct kvm *kvm = filp->private_data;
1167         void __user *argp = (void __user *)arg;
1168         int r;
1169
1170         if (kvm->mm != current->mm)
1171                 return -EIO;
1172         switch (ioctl) {
1173         case KVM_CREATE_VCPU:
1174                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1175                 if (r < 0)
1176                         goto out;
1177                 break;
1178         case KVM_SET_USER_MEMORY_REGION: {
1179                 struct kvm_userspace_memory_region kvm_userspace_mem;
1180
1181                 r = -EFAULT;
1182                 if (copy_from_user(&kvm_userspace_mem, argp,
1183                                                 sizeof kvm_userspace_mem))
1184                         goto out;
1185
1186                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1187                 if (r)
1188                         goto out;
1189                 break;
1190         }
1191         case KVM_GET_DIRTY_LOG: {
1192                 struct kvm_dirty_log log;
1193
1194                 r = -EFAULT;
1195                 if (copy_from_user(&log, argp, sizeof log))
1196                         goto out;
1197                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1198                 if (r)
1199                         goto out;
1200                 break;
1201         }
1202 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1203         case KVM_REGISTER_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_register_coalesced_mmio(kvm, &zone);
1210                 if (r)
1211                         goto out;
1212                 r = 0;
1213                 break;
1214         }
1215         case KVM_UNREGISTER_COALESCED_MMIO: {
1216                 struct kvm_coalesced_mmio_zone zone;
1217                 r = -EFAULT;
1218                 if (copy_from_user(&zone, argp, sizeof zone))
1219                         goto out;
1220                 r = -ENXIO;
1221                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1222                 if (r)
1223                         goto out;
1224                 r = 0;
1225                 break;
1226         }
1227 #endif
1228         default:
1229                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1230         }
1231 out:
1232         return r;
1233 }
1234
1235 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1236 {
1237         struct kvm *kvm = vma->vm_file->private_data;
1238         struct page *page;
1239
1240         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1241                 return VM_FAULT_SIGBUS;
1242         page = gfn_to_page(kvm, vmf->pgoff);
1243         if (is_error_page(page)) {
1244                 kvm_release_page_clean(page);
1245                 return VM_FAULT_SIGBUS;
1246         }
1247         vmf->page = page;
1248         return 0;
1249 }
1250
1251 static struct vm_operations_struct kvm_vm_vm_ops = {
1252         .fault = kvm_vm_fault,
1253 };
1254
1255 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1256 {
1257         vma->vm_ops = &kvm_vm_vm_ops;
1258         return 0;
1259 }
1260
1261 static const struct file_operations kvm_vm_fops = {
1262         .release        = kvm_vm_release,
1263         .unlocked_ioctl = kvm_vm_ioctl,
1264         .compat_ioctl   = kvm_vm_ioctl,
1265         .mmap           = kvm_vm_mmap,
1266 };
1267
1268 static int kvm_dev_ioctl_create_vm(void)
1269 {
1270         int fd;
1271         struct kvm *kvm;
1272
1273         kvm = kvm_create_vm();
1274         if (IS_ERR(kvm))
1275                 return PTR_ERR(kvm);
1276         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1277         if (fd < 0)
1278                 kvm_put_kvm(kvm);
1279
1280         return fd;
1281 }
1282
1283 static long kvm_dev_ioctl(struct file *filp,
1284                           unsigned int ioctl, unsigned long arg)
1285 {
1286         long r = -EINVAL;
1287
1288         switch (ioctl) {
1289         case KVM_GET_API_VERSION:
1290                 r = -EINVAL;
1291                 if (arg)
1292                         goto out;
1293                 r = KVM_API_VERSION;
1294                 break;
1295         case KVM_CREATE_VM:
1296                 r = -EINVAL;
1297                 if (arg)
1298                         goto out;
1299                 r = kvm_dev_ioctl_create_vm();
1300                 break;
1301         case KVM_CHECK_EXTENSION:
1302                 r = kvm_dev_ioctl_check_extension(arg);
1303                 break;
1304         case KVM_GET_VCPU_MMAP_SIZE:
1305                 r = -EINVAL;
1306                 if (arg)
1307                         goto out;
1308                 r = PAGE_SIZE;     /* struct kvm_run */
1309 #ifdef CONFIG_X86
1310                 r += PAGE_SIZE;    /* pio data page */
1311 #endif
1312 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1313                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1314 #endif
1315                 break;
1316         case KVM_TRACE_ENABLE:
1317         case KVM_TRACE_PAUSE:
1318         case KVM_TRACE_DISABLE:
1319                 r = kvm_trace_ioctl(ioctl, arg);
1320                 break;
1321         default:
1322                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1323         }
1324 out:
1325         return r;
1326 }
1327
1328 static struct file_operations kvm_chardev_ops = {
1329         .unlocked_ioctl = kvm_dev_ioctl,
1330         .compat_ioctl   = kvm_dev_ioctl,
1331 };
1332
1333 static struct miscdevice kvm_dev = {
1334         KVM_MINOR,
1335         "kvm",
1336         &kvm_chardev_ops,
1337 };
1338
1339 static void hardware_enable(void *junk)
1340 {
1341         int cpu = raw_smp_processor_id();
1342
1343         if (cpu_isset(cpu, cpus_hardware_enabled))
1344                 return;
1345         cpu_set(cpu, cpus_hardware_enabled);
1346         kvm_arch_hardware_enable(NULL);
1347 }
1348
1349 static void hardware_disable(void *junk)
1350 {
1351         int cpu = raw_smp_processor_id();
1352
1353         if (!cpu_isset(cpu, cpus_hardware_enabled))
1354                 return;
1355         cpu_clear(cpu, cpus_hardware_enabled);
1356         kvm_arch_hardware_disable(NULL);
1357 }
1358
1359 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1360                            void *v)
1361 {
1362         int cpu = (long)v;
1363
1364         val &= ~CPU_TASKS_FROZEN;
1365         switch (val) {
1366         case CPU_DYING:
1367                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1368                        cpu);
1369                 hardware_disable(NULL);
1370                 break;
1371         case CPU_UP_CANCELED:
1372                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1373                        cpu);
1374                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1375                 break;
1376         case CPU_ONLINE:
1377                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1378                        cpu);
1379                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1380                 break;
1381         }
1382         return NOTIFY_OK;
1383 }
1384
1385
1386 asmlinkage void kvm_handle_fault_on_reboot(void)
1387 {
1388         if (kvm_rebooting)
1389                 /* spin while reset goes on */
1390                 while (true)
1391                         ;
1392         /* Fault while not rebooting.  We want the trace. */
1393         BUG();
1394 }
1395 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1396
1397 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1398                       void *v)
1399 {
1400         if (val == SYS_RESTART) {
1401                 /*
1402                  * Some (well, at least mine) BIOSes hang on reboot if
1403                  * in vmx root mode.
1404                  */
1405                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1406                 kvm_rebooting = true;
1407                 on_each_cpu(hardware_disable, NULL, 1);
1408         }
1409         return NOTIFY_OK;
1410 }
1411
1412 static struct notifier_block kvm_reboot_notifier = {
1413         .notifier_call = kvm_reboot,
1414         .priority = 0,
1415 };
1416
1417 void kvm_io_bus_init(struct kvm_io_bus *bus)
1418 {
1419         memset(bus, 0, sizeof(*bus));
1420 }
1421
1422 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
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                 kvm_iodevice_destructor(pos);
1430         }
1431 }
1432
1433 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1434                                           gpa_t addr, int len, int is_write)
1435 {
1436         int i;
1437
1438         for (i = 0; i < bus->dev_count; i++) {
1439                 struct kvm_io_device *pos = bus->devs[i];
1440
1441                 if (pos->in_range(pos, addr, len, is_write))
1442                         return pos;
1443         }
1444
1445         return NULL;
1446 }
1447
1448 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1449 {
1450         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1451
1452         bus->devs[bus->dev_count++] = dev;
1453 }
1454
1455 static struct notifier_block kvm_cpu_notifier = {
1456         .notifier_call = kvm_cpu_hotplug,
1457         .priority = 20, /* must be > scheduler priority */
1458 };
1459
1460 static int vm_stat_get(void *_offset, u64 *val)
1461 {
1462         unsigned offset = (long)_offset;
1463         struct kvm *kvm;
1464
1465         *val = 0;
1466         spin_lock(&kvm_lock);
1467         list_for_each_entry(kvm, &vm_list, vm_list)
1468                 *val += *(u32 *)((void *)kvm + offset);
1469         spin_unlock(&kvm_lock);
1470         return 0;
1471 }
1472
1473 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1474
1475 static int vcpu_stat_get(void *_offset, u64 *val)
1476 {
1477         unsigned offset = (long)_offset;
1478         struct kvm *kvm;
1479         struct kvm_vcpu *vcpu;
1480         int i;
1481
1482         *val = 0;
1483         spin_lock(&kvm_lock);
1484         list_for_each_entry(kvm, &vm_list, vm_list)
1485                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1486                         vcpu = kvm->vcpus[i];
1487                         if (vcpu)
1488                                 *val += *(u32 *)((void *)vcpu + offset);
1489                 }
1490         spin_unlock(&kvm_lock);
1491         return 0;
1492 }
1493
1494 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1495
1496 static struct file_operations *stat_fops[] = {
1497         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1498         [KVM_STAT_VM]   = &vm_stat_fops,
1499 };
1500
1501 static void kvm_init_debug(void)
1502 {
1503         struct kvm_stats_debugfs_item *p;
1504
1505         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1506         for (p = debugfs_entries; p->name; ++p)
1507                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1508                                                 (void *)(long)p->offset,
1509                                                 stat_fops[p->kind]);
1510 }
1511
1512 static void kvm_exit_debug(void)
1513 {
1514         struct kvm_stats_debugfs_item *p;
1515
1516         for (p = debugfs_entries; p->name; ++p)
1517                 debugfs_remove(p->dentry);
1518         debugfs_remove(kvm_debugfs_dir);
1519 }
1520
1521 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1522 {
1523         hardware_disable(NULL);
1524         return 0;
1525 }
1526
1527 static int kvm_resume(struct sys_device *dev)
1528 {
1529         hardware_enable(NULL);
1530         return 0;
1531 }
1532
1533 static struct sysdev_class kvm_sysdev_class = {
1534         .name = "kvm",
1535         .suspend = kvm_suspend,
1536         .resume = kvm_resume,
1537 };
1538
1539 static struct sys_device kvm_sysdev = {
1540         .id = 0,
1541         .cls = &kvm_sysdev_class,
1542 };
1543
1544 struct page *bad_page;
1545 pfn_t bad_pfn;
1546
1547 static inline
1548 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1549 {
1550         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1551 }
1552
1553 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1554 {
1555         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1556
1557         kvm_arch_vcpu_load(vcpu, cpu);
1558 }
1559
1560 static void kvm_sched_out(struct preempt_notifier *pn,
1561                           struct task_struct *next)
1562 {
1563         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1564
1565         kvm_arch_vcpu_put(vcpu);
1566 }
1567
1568 int kvm_init(void *opaque, unsigned int vcpu_size,
1569                   struct module *module)
1570 {
1571         int r;
1572         int cpu;
1573
1574         kvm_init_debug();
1575
1576         r = kvm_arch_init(opaque);
1577         if (r)
1578                 goto out_fail;
1579
1580         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1581
1582         if (bad_page == NULL) {
1583                 r = -ENOMEM;
1584                 goto out;
1585         }
1586
1587         bad_pfn = page_to_pfn(bad_page);
1588
1589         r = kvm_arch_hardware_setup();
1590         if (r < 0)
1591                 goto out_free_0;
1592
1593         for_each_online_cpu(cpu) {
1594                 smp_call_function_single(cpu,
1595                                 kvm_arch_check_processor_compat,
1596                                 &r, 1);
1597                 if (r < 0)
1598                         goto out_free_1;
1599         }
1600
1601         on_each_cpu(hardware_enable, NULL, 1);
1602         r = register_cpu_notifier(&kvm_cpu_notifier);
1603         if (r)
1604                 goto out_free_2;
1605         register_reboot_notifier(&kvm_reboot_notifier);
1606
1607         r = sysdev_class_register(&kvm_sysdev_class);
1608         if (r)
1609                 goto out_free_3;
1610
1611         r = sysdev_register(&kvm_sysdev);
1612         if (r)
1613                 goto out_free_4;
1614
1615         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1616         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1617                                            __alignof__(struct kvm_vcpu),
1618                                            0, NULL);
1619         if (!kvm_vcpu_cache) {
1620                 r = -ENOMEM;
1621                 goto out_free_5;
1622         }
1623
1624         kvm_chardev_ops.owner = module;
1625
1626         r = misc_register(&kvm_dev);
1627         if (r) {
1628                 printk(KERN_ERR "kvm: misc device register failed\n");
1629                 goto out_free;
1630         }
1631
1632         kvm_preempt_ops.sched_in = kvm_sched_in;
1633         kvm_preempt_ops.sched_out = kvm_sched_out;
1634
1635         return 0;
1636
1637 out_free:
1638         kmem_cache_destroy(kvm_vcpu_cache);
1639 out_free_5:
1640         sysdev_unregister(&kvm_sysdev);
1641 out_free_4:
1642         sysdev_class_unregister(&kvm_sysdev_class);
1643 out_free_3:
1644         unregister_reboot_notifier(&kvm_reboot_notifier);
1645         unregister_cpu_notifier(&kvm_cpu_notifier);
1646 out_free_2:
1647         on_each_cpu(hardware_disable, NULL, 1);
1648 out_free_1:
1649         kvm_arch_hardware_unsetup();
1650 out_free_0:
1651         __free_page(bad_page);
1652 out:
1653         kvm_arch_exit();
1654         kvm_exit_debug();
1655 out_fail:
1656         return r;
1657 }
1658 EXPORT_SYMBOL_GPL(kvm_init);
1659
1660 void kvm_exit(void)
1661 {
1662         kvm_trace_cleanup();
1663         misc_deregister(&kvm_dev);
1664         kmem_cache_destroy(kvm_vcpu_cache);
1665         sysdev_unregister(&kvm_sysdev);
1666         sysdev_class_unregister(&kvm_sysdev_class);
1667         unregister_reboot_notifier(&kvm_reboot_notifier);
1668         unregister_cpu_notifier(&kvm_cpu_notifier);
1669         on_each_cpu(hardware_disable, NULL, 1);
1670         kvm_arch_hardware_unsetup();
1671         kvm_arch_exit();
1672         kvm_exit_debug();
1673         __free_page(bad_page);
1674 }
1675 EXPORT_SYMBOL_GPL(kvm_exit);