virt/kvm/kvm_main.c

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