drivers/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  *
   9  * Authors:
  10  *   Avi Kivity   <avi@qumranet.com>
  11  *   Yaniv Kamay  <yaniv@qumranet.com>
  12  *
  13  * This work is licensed under the terms of the GNU GPL, version 2.  See
  14  * the COPYING file in the top-level directory.
  15  *
  16  */
  17
  18 #include "kvm.h"
  19 #include "x86.h"
  20 #include "x86_emulate.h"
  21 #include "irq.h"
  22
  23 #include <linux/kvm.h>
  24 #include <linux/module.h>
  25 #include <linux/errno.h>
  26 #include <linux/percpu.h>
  27 #include <linux/gfp.h>
  28 #include <linux/mm.h>
  29 #include <linux/miscdevice.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/reboot.h>
  32 #include <linux/debugfs.h>
  33 #include <linux/highmem.h>
  34 #include <linux/file.h>
  35 #include <linux/sysdev.h>
  36 #include <linux/cpu.h>
  37 #include <linux/sched.h>
  38 #include <linux/cpumask.h>
  39 #include <linux/smp.h>
  40 #include <linux/anon_inodes.h>
  41 #include <linux/profile.h>
  42 #include <linux/kvm_para.h>
  43 #include <linux/pagemap.h>
  44 #include <linux/mman.h>
  45
  46 #include <asm/processor.h>
  47 #include <asm/msr.h>
  48 #include <asm/io.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/desc.h>
  51
  52 MODULE_AUTHOR("Qumranet");
  53 MODULE_LICENSE("GPL");
  54
  55 static DEFINE_SPINLOCK(kvm_lock);
  56 static LIST_HEAD(vm_list);
  57
  58 static cpumask_t cpus_hardware_enabled;
  59
  60 struct kvm_x86_ops *kvm_x86_ops;
  61 struct kmem_cache *kvm_vcpu_cache;
  62 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  63
  64 static __read_mostly struct preempt_ops kvm_preempt_ops;
  65
  66 static struct dentry *debugfs_dir;
  67
  68 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  69                            unsigned long arg);
  70
  71 static inline int valid_vcpu(int n)
  72 {
  73         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  74 }
  75
  76 /*
  77  * Switches to specified vcpu, until a matching vcpu_put()
  78  */
  79 void vcpu_load(struct kvm_vcpu *vcpu)
  80 {
  81         int cpu;
  82
  83         mutex_lock(&vcpu->mutex);
  84         cpu = get_cpu();
  85         preempt_notifier_register(&vcpu->preempt_notifier);
  86         kvm_arch_vcpu_load(vcpu, cpu);
  87         put_cpu();
  88 }
  89
  90 void vcpu_put(struct kvm_vcpu *vcpu)
  91 {
  92         preempt_disable();
  93         kvm_arch_vcpu_put(vcpu);
  94         preempt_notifier_unregister(&vcpu->preempt_notifier);
  95         preempt_enable();
  96         mutex_unlock(&vcpu->mutex);
  97 }
  98
  99 static void ack_flush(void *_completed)
 100 {
 101 }
 102
 103 void kvm_flush_remote_tlbs(struct kvm *kvm)
 104 {
 105         int i, cpu;
 106         cpumask_t cpus;
 107         struct kvm_vcpu *vcpu;
 108
 109         cpus_clear(cpus);
 110         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 111                 vcpu = kvm->vcpus[i];
 112                 if (!vcpu)
 113                         continue;
 114                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 115                         continue;
 116                 cpu = vcpu->cpu;
 117                 if (cpu != -1 && cpu != raw_smp_processor_id())
 118                         cpu_set(cpu, cpus);
 119         }
 120         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 121 }
 122
 123 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 124 {
 125         struct page *page;
 126         int r;
 127
 128         mutex_init(&vcpu->mutex);
 129         vcpu->cpu = -1;
 130         vcpu->mmu.root_hpa = INVALID_PAGE;
 131         vcpu->kvm = kvm;
 132         vcpu->vcpu_id = id;
 133         if (!irqchip_in_kernel(kvm) || id == 0)
 134                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 135         else
 136                 vcpu->mp_state = VCPU_MP_STATE_UNINITIALIZED;
 137         init_waitqueue_head(&vcpu->wq);
 138
 139         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 140         if (!page) {
 141                 r = -ENOMEM;
 142                 goto fail;
 143         }
 144         vcpu->run = page_address(page);
 145
 146         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 147         if (!page) {
 148                 r = -ENOMEM;
 149                 goto fail_free_run;
 150         }
 151         vcpu->pio_data = page_address(page);
 152
 153         r = kvm_mmu_create(vcpu);
 154         if (r < 0)
 155                 goto fail_free_pio_data;
 156
 157         if (irqchip_in_kernel(kvm)) {
 158                 r = kvm_create_lapic(vcpu);
 159                 if (r < 0)
 160                         goto fail_mmu_destroy;
 161         }
 162
 163         return 0;
 164
 165 fail_mmu_destroy:
 166         kvm_mmu_destroy(vcpu);
 167 fail_free_pio_data:
 168         free_page((unsigned long)vcpu->pio_data);
 169 fail_free_run:
 170         free_page((unsigned long)vcpu->run);
 171 fail:
 172         return r;
 173 }
 174 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 175
 176 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 177 {
 178         kvm_free_lapic(vcpu);
 179         kvm_mmu_destroy(vcpu);
 180         free_page((unsigned long)vcpu->pio_data);
 181         free_page((unsigned long)vcpu->run);
 182 }
 183 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 184
 185 static struct kvm *kvm_create_vm(void)
 186 {
 187         struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
 188
 189         if (!kvm)
 190                 return ERR_PTR(-ENOMEM);
 191
 192         kvm_io_bus_init(&kvm->pio_bus);
 193         mutex_init(&kvm->lock);
 194         INIT_LIST_HEAD(&kvm->active_mmu_pages);
 195         kvm_io_bus_init(&kvm->mmio_bus);
 196         spin_lock(&kvm_lock);
 197         list_add(&kvm->vm_list, &vm_list);
 198         spin_unlock(&kvm_lock);
 199         return kvm;
 200 }
 201
 202 /*
 203  * Free any memory in @free but not in @dont.
 204  */
 205 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 206                                   struct kvm_memory_slot *dont)
 207 {
 208         if (!dont || free->rmap != dont->rmap)
 209                 vfree(free->rmap);
 210
 211         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 212                 vfree(free->dirty_bitmap);
 213
 214         free->npages = 0;
 215         free->dirty_bitmap = NULL;
 216         free->rmap = NULL;
 217 }
 218
 219 static void kvm_free_physmem(struct kvm *kvm)
 220 {
 221         int i;
 222
 223         for (i = 0; i < kvm->nmemslots; ++i)
 224                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 225 }
 226
 227 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
 228 {
 229         vcpu_load(vcpu);
 230         kvm_mmu_unload(vcpu);
 231         vcpu_put(vcpu);
 232 }
 233
 234 static void kvm_free_vcpus(struct kvm *kvm)
 235 {
 236         unsigned int i;
 237
 238         /*
 239          * Unpin any mmu pages first.
 240          */
 241         for (i = 0; i < KVM_MAX_VCPUS; ++i)
 242                 if (kvm->vcpus[i])
 243                         kvm_unload_vcpu_mmu(kvm->vcpus[i]);
 244         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 245                 if (kvm->vcpus[i]) {
 246                         kvm_x86_ops->vcpu_free(kvm->vcpus[i]);
 247                         kvm->vcpus[i] = NULL;
 248                 }
 249         }
 250
 251 }
 252
 253 static void kvm_destroy_vm(struct kvm *kvm)
 254 {
 255         spin_lock(&kvm_lock);
 256         list_del(&kvm->vm_list);
 257         spin_unlock(&kvm_lock);
 258         kvm_io_bus_destroy(&kvm->pio_bus);
 259         kvm_io_bus_destroy(&kvm->mmio_bus);
 260         kfree(kvm->vpic);
 261         kfree(kvm->vioapic);
 262         kvm_free_vcpus(kvm);
 263         kvm_free_physmem(kvm);
 264         kfree(kvm);
 265 }
 266
 267 static int kvm_vm_release(struct inode *inode, struct file *filp)
 268 {
 269         struct kvm *kvm = filp->private_data;
 270
 271         kvm_destroy_vm(kvm);
 272         return 0;
 273 }
 274
 275 /*
 276  * Allocate some memory and give it an address in the guest physical address
 277  * space.
 278  *
 279  * Discontiguous memory is allowed, mostly for framebuffers.
 280  *
 281  * Must be called holding kvm->lock.
 282  */
 283 int __kvm_set_memory_region(struct kvm *kvm,
 284                             struct kvm_userspace_memory_region *mem,
 285                             int user_alloc)
 286 {
 287         int r;
 288         gfn_t base_gfn;
 289         unsigned long npages;
 290         unsigned long i;
 291         struct kvm_memory_slot *memslot;
 292         struct kvm_memory_slot old, new;
 293
 294         r = -EINVAL;
 295         /* General sanity checks */
 296         if (mem->memory_size & (PAGE_SIZE - 1))
 297                 goto out;
 298         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 299                 goto out;
 300         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 301                 goto out;
 302         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 303                 goto out;
 304
 305         memslot = &kvm->memslots[mem->slot];
 306         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 307         npages = mem->memory_size >> PAGE_SHIFT;
 308
 309         if (!npages)
 310                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 311
 312         new = old = *memslot;
 313
 314         new.base_gfn = base_gfn;
 315         new.npages = npages;
 316         new.flags = mem->flags;
 317
 318         /* Disallow changing a memory slot's size. */
 319         r = -EINVAL;
 320         if (npages && old.npages && npages != old.npages)
 321                 goto out_free;
 322
 323         /* Check for overlaps */
 324         r = -EEXIST;
 325         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 326                 struct kvm_memory_slot *s = &kvm->memslots[i];
 327
 328                 if (s == memslot)
 329                         continue;
 330                 if (!((base_gfn + npages <= s->base_gfn) ||
 331                       (base_gfn >= s->base_gfn + s->npages)))
 332                         goto out_free;
 333         }
 334
 335         /* Free page dirty bitmap if unneeded */
 336         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 337                 new.dirty_bitmap = NULL;
 338
 339         r = -ENOMEM;
 340
 341         /* Allocate if a slot is being created */
 342         if (npages && !new.rmap) {
 343                 new.rmap = vmalloc(npages * sizeof(struct page *));
 344
 345                 if (!new.rmap)
 346                         goto out_free;
 347
 348                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 349
 350                 new.user_alloc = user_alloc;
 351                 if (user_alloc)
 352                         new.userspace_addr = mem->userspace_addr;
 353                 else {
 354                         down_write(&current->mm->mmap_sem);
 355                         new.userspace_addr = do_mmap(NULL, 0,
 356                                                      npages * PAGE_SIZE,
 357                                                      PROT_READ | PROT_WRITE,
 358                                                      MAP_SHARED | MAP_ANONYMOUS,
 359                                                      0);
 360                         up_write(&current->mm->mmap_sem);
 361
 362                         if (IS_ERR((void *)new.userspace_addr))
 363                                 goto out_free;
 364                 }
 365         } else {
 366                 if (!old.user_alloc && old.rmap) {
 367                         int ret;
 368
 369                         down_write(&current->mm->mmap_sem);
 370                         ret = do_munmap(current->mm, old.userspace_addr,
 371                                         old.npages * PAGE_SIZE);
 372                         up_write(&current->mm->mmap_sem);
 373                         if (ret < 0)
 374                                 printk(KERN_WARNING
 375                                        "kvm_vm_ioctl_set_memory_region: "
 376                                        "failed to munmap memory\n");
 377                 }
 378         }
 379
 380         /* Allocate page dirty bitmap if needed */
 381         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 382                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 383
 384                 new.dirty_bitmap = vmalloc(dirty_bytes);
 385                 if (!new.dirty_bitmap)
 386                         goto out_free;
 387                 memset(new.dirty_bitmap, 0, dirty_bytes);
 388         }
 389
 390         if (mem->slot >= kvm->nmemslots)
 391                 kvm->nmemslots = mem->slot + 1;
 392
 393         if (!kvm->n_requested_mmu_pages) {
 394                 unsigned int n_pages;
 395
 396                 if (npages) {
 397                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 398                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 399                                                  n_pages);
 400                 } else {
 401                         unsigned int nr_mmu_pages;
 402
 403                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 404                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 405                         nr_mmu_pages = max(nr_mmu_pages,
 406                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 407                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 408                 }
 409         }
 410
 411         *memslot = new;
 412
 413         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 414         kvm_flush_remote_tlbs(kvm);
 415
 416         kvm_free_physmem_slot(&old, &new);
 417         return 0;
 418
 419 out_free:
 420         kvm_free_physmem_slot(&new, &old);
 421 out:
 422         return r;
 423
 424 }
 425 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 426
 427 int kvm_set_memory_region(struct kvm *kvm,
 428                           struct kvm_userspace_memory_region *mem,
 429                           int user_alloc)
 430 {
 431         int r;
 432
 433         mutex_lock(&kvm->lock);
 434         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 435         mutex_unlock(&kvm->lock);
 436         return r;
 437 }
 438 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 439
 440 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 441                                    struct
 442                                    kvm_userspace_memory_region *mem,
 443                                    int user_alloc)
 444 {
 445         if (mem->slot >= KVM_MEMORY_SLOTS)
 446                 return -EINVAL;
 447         return kvm_set_memory_region(kvm, mem, user_alloc);
 448 }
 449
 450 /*
 451  * Get (and clear) the dirty memory log for a memory slot.
 452  */
 453 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
 454                                       struct kvm_dirty_log *log)
 455 {
 456         struct kvm_memory_slot *memslot;
 457         int r, i;
 458         int n;
 459         unsigned long any = 0;
 460
 461         mutex_lock(&kvm->lock);
 462
 463         r = -EINVAL;
 464         if (log->slot >= KVM_MEMORY_SLOTS)
 465                 goto out;
 466
 467         memslot = &kvm->memslots[log->slot];
 468         r = -ENOENT;
 469         if (!memslot->dirty_bitmap)
 470                 goto out;
 471
 472         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 473
 474         for (i = 0; !any && i < n/sizeof(long); ++i)
 475                 any = memslot->dirty_bitmap[i];
 476
 477         r = -EFAULT;
 478         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 479                 goto out;
 480
 481         /* If nothing is dirty, don't bother messing with page tables. */
 482         if (any) {
 483                 kvm_mmu_slot_remove_write_access(kvm, log->slot);
 484                 kvm_flush_remote_tlbs(kvm);
 485                 memset(memslot->dirty_bitmap, 0, n);
 486         }
 487
 488         r = 0;
 489
 490 out:
 491         mutex_unlock(&kvm->lock);
 492         return r;
 493 }
 494
 495 int is_error_page(struct page *page)
 496 {
 497         return page == bad_page;
 498 }
 499 EXPORT_SYMBOL_GPL(is_error_page);
 500
 501 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 502 {
 503         int i;
 504         struct kvm_mem_alias *alias;
 505
 506         for (i = 0; i < kvm->naliases; ++i) {
 507                 alias = &kvm->aliases[i];
 508                 if (gfn >= alias->base_gfn
 509                     && gfn < alias->base_gfn + alias->npages)
 510                         return alias->target_gfn + gfn - alias->base_gfn;
 511         }
 512         return gfn;
 513 }
 514
 515 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 516 {
 517         int i;
 518
 519         for (i = 0; i < kvm->nmemslots; ++i) {
 520                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 521
 522                 if (gfn >= memslot->base_gfn
 523                     && gfn < memslot->base_gfn + memslot->npages)
 524                         return memslot;
 525         }
 526         return NULL;
 527 }
 528
 529 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 530 {
 531         gfn = unalias_gfn(kvm, gfn);
 532         return __gfn_to_memslot(kvm, gfn);
 533 }
 534
 535 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 536 {
 537         int i;
 538
 539         gfn = unalias_gfn(kvm, gfn);
 540         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 541                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 542
 543                 if (gfn >= memslot->base_gfn
 544                     && gfn < memslot->base_gfn + memslot->npages)
 545                         return 1;
 546         }
 547         return 0;
 548 }
 549 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 550
 551 /*
 552  * Requires current->mm->mmap_sem to be held
 553  */
 554 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 555 {
 556         struct kvm_memory_slot *slot;
 557         struct page *page[1];
 558         int npages;
 559
 560         might_sleep();
 561
 562         gfn = unalias_gfn(kvm, gfn);
 563         slot = __gfn_to_memslot(kvm, gfn);
 564         if (!slot) {
 565                 get_page(bad_page);
 566                 return bad_page;
 567         }
 568
 569         npages = get_user_pages(current, current->mm,
 570                                 slot->userspace_addr
 571                                 + (gfn - slot->base_gfn) * PAGE_SIZE, 1,
 572                                 1, 1, page, NULL);
 573         if (npages != 1) {
 574                 get_page(bad_page);
 575                 return bad_page;
 576         }
 577
 578         return page[0];
 579 }
 580
 581 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 582 {
 583         struct page *page;
 584
 585         down_read(&current->mm->mmap_sem);
 586         page = __gfn_to_page(kvm, gfn);
 587         up_read(&current->mm->mmap_sem);
 588
 589         return page;
 590 }
 591
 592 EXPORT_SYMBOL_GPL(gfn_to_page);
 593
 594 void kvm_release_page(struct page *page)
 595 {
 596         if (!PageReserved(page))
 597                 SetPageDirty(page);
 598         put_page(page);
 599 }
 600 EXPORT_SYMBOL_GPL(kvm_release_page);
 601
 602 static int next_segment(unsigned long len, int offset)
 603 {
 604         if (len > PAGE_SIZE - offset)
 605                 return PAGE_SIZE - offset;
 606         else
 607                 return len;
 608 }
 609
 610 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 611                         int len)
 612 {
 613         void *page_virt;
 614         struct page *page;
 615
 616         page = gfn_to_page(kvm, gfn);
 617         if (is_error_page(page)) {
 618                 kvm_release_page(page);
 619                 return -EFAULT;
 620         }
 621         page_virt = kmap_atomic(page, KM_USER0);
 622
 623         memcpy(data, page_virt + offset, len);
 624
 625         kunmap_atomic(page_virt, KM_USER0);
 626         kvm_release_page(page);
 627         return 0;
 628 }
 629 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 630
 631 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 632 {
 633         gfn_t gfn = gpa >> PAGE_SHIFT;
 634         int seg;
 635         int offset = offset_in_page(gpa);
 636         int ret;
 637
 638         while ((seg = next_segment(len, offset)) != 0) {
 639                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 640                 if (ret < 0)
 641                         return ret;
 642                 offset = 0;
 643                 len -= seg;
 644                 data += seg;
 645                 ++gfn;
 646         }
 647         return 0;
 648 }
 649 EXPORT_SYMBOL_GPL(kvm_read_guest);
 650
 651 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 652                          int offset, int len)
 653 {
 654         void *page_virt;
 655         struct page *page;
 656
 657         page = gfn_to_page(kvm, gfn);
 658         if (is_error_page(page)) {
 659                 kvm_release_page(page);
 660                 return -EFAULT;
 661         }
 662         page_virt = kmap_atomic(page, KM_USER0);
 663
 664         memcpy(page_virt + offset, data, len);
 665
 666         kunmap_atomic(page_virt, KM_USER0);
 667         mark_page_dirty(kvm, gfn);
 668         kvm_release_page(page);
 669         return 0;
 670 }
 671 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 672
 673 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 674                     unsigned long len)
 675 {
 676         gfn_t gfn = gpa >> PAGE_SHIFT;
 677         int seg;
 678         int offset = offset_in_page(gpa);
 679         int ret;
 680
 681         while ((seg = next_segment(len, offset)) != 0) {
 682                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 683                 if (ret < 0)
 684                         return ret;
 685                 offset = 0;
 686                 len -= seg;
 687                 data += seg;
 688                 ++gfn;
 689         }
 690         return 0;
 691 }
 692
 693 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 694 {
 695         void *page_virt;
 696         struct page *page;
 697
 698         page = gfn_to_page(kvm, gfn);
 699         if (is_error_page(page)) {
 700                 kvm_release_page(page);
 701                 return -EFAULT;
 702         }
 703         page_virt = kmap_atomic(page, KM_USER0);
 704
 705         memset(page_virt + offset, 0, len);
 706
 707         kunmap_atomic(page_virt, KM_USER0);
 708         kvm_release_page(page);
 709         return 0;
 710 }
 711 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 712
 713 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, 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_clear_guest_page(kvm, gfn, offset, seg);
 722                 if (ret < 0)
 723                         return ret;
 724                 offset = 0;
 725                 len -= seg;
 726                 ++gfn;
 727         }
 728         return 0;
 729 }
 730 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 731
 732 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 733 {
 734         struct kvm_memory_slot *memslot;
 735
 736         gfn = unalias_gfn(kvm, gfn);
 737         memslot = __gfn_to_memslot(kvm, gfn);
 738         if (memslot && memslot->dirty_bitmap) {
 739                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 740
 741                 /* avoid RMW */
 742                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 743                         set_bit(rel_gfn, memslot->dirty_bitmap);
 744         }
 745 }
 746
 747 /*
 748  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 749  */
 750 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 751 {
 752         DECLARE_WAITQUEUE(wait, current);
 753
 754         add_wait_queue(&vcpu->wq, &wait);
 755
 756         /*
 757          * We will block until either an interrupt or a signal wakes us up
 758          */
 759         while (!kvm_cpu_has_interrupt(vcpu)
 760                && !signal_pending(current)
 761                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
 762                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
 763                 set_current_state(TASK_INTERRUPTIBLE);
 764                 vcpu_put(vcpu);
 765                 schedule();
 766                 vcpu_load(vcpu);
 767         }
 768
 769         __set_current_state(TASK_RUNNING);
 770         remove_wait_queue(&vcpu->wq, &wait);
 771 }
 772
 773 void kvm_resched(struct kvm_vcpu *vcpu)
 774 {
 775         if (!need_resched())
 776                 return;
 777         cond_resched();
 778 }
 779 EXPORT_SYMBOL_GPL(kvm_resched);
 780
 781 /*
 782  * Check if userspace requested an interrupt window, and that the
 783  * interrupt window is open.
 784  *
 785  * No need to exit to userspace if we already have an interrupt queued.
 786  */
 787 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
 788                                           struct kvm_run *kvm_run)
 789 {
 790         return (!vcpu->irq_summary &&
 791                 kvm_run->request_interrupt_window &&
 792                 vcpu->interrupt_window_open &&
 793                 (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
 794 }
 795
 796 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
 797                               struct kvm_run *kvm_run)
 798 {
 799         kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
 800         kvm_run->cr8 = get_cr8(vcpu);
 801         kvm_run->apic_base = kvm_get_apic_base(vcpu);
 802         if (irqchip_in_kernel(vcpu->kvm))
 803                 kvm_run->ready_for_interrupt_injection = 1;
 804         else
 805                 kvm_run->ready_for_interrupt_injection =
 806                                         (vcpu->interrupt_window_open &&
 807                                          vcpu->irq_summary == 0);
 808 }
 809
 810 static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 811 {
 812         int r;
 813
 814         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
 815                 pr_debug("vcpu %d received sipi with vector # %x\n",
 816                        vcpu->vcpu_id, vcpu->sipi_vector);
 817                 kvm_lapic_reset(vcpu);
 818                 r = kvm_x86_ops->vcpu_reset(vcpu);
 819                 if (r)
 820                         return r;
 821                 vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
 822         }
 823
 824 preempted:
 825         if (vcpu->guest_debug.enabled)
 826                 kvm_x86_ops->guest_debug_pre(vcpu);
 827
 828 again:
 829         r = kvm_mmu_reload(vcpu);
 830         if (unlikely(r))
 831                 goto out;
 832
 833         kvm_inject_pending_timer_irqs(vcpu);
 834
 835         preempt_disable();
 836
 837         kvm_x86_ops->prepare_guest_switch(vcpu);
 838         kvm_load_guest_fpu(vcpu);
 839
 840         local_irq_disable();
 841
 842         if (signal_pending(current)) {
 843                 local_irq_enable();
 844                 preempt_enable();
 845                 r = -EINTR;
 846                 kvm_run->exit_reason = KVM_EXIT_INTR;
 847                 ++vcpu->stat.signal_exits;
 848                 goto out;
 849         }
 850
 851         if (irqchip_in_kernel(vcpu->kvm))
 852                 kvm_x86_ops->inject_pending_irq(vcpu);
 853         else if (!vcpu->mmio_read_completed)
 854                 kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
 855
 856         vcpu->guest_mode = 1;
 857         kvm_guest_enter();
 858
 859         if (vcpu->requests)
 860                 if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 861                         kvm_x86_ops->tlb_flush(vcpu);
 862
 863         kvm_x86_ops->run(vcpu, kvm_run);
 864
 865         vcpu->guest_mode = 0;
 866         local_irq_enable();
 867
 868         ++vcpu->stat.exits;
 869
 870         /*
 871          * We must have an instruction between local_irq_enable() and
 872          * kvm_guest_exit(), so the timer interrupt isn't delayed by
 873          * the interrupt shadow.  The stat.exits increment will do nicely.
 874          * But we need to prevent reordering, hence this barrier():
 875          */
 876         barrier();
 877
 878         kvm_guest_exit();
 879
 880         preempt_enable();
 881
 882         /*
 883          * Profile KVM exit RIPs:
 884          */
 885         if (unlikely(prof_on == KVM_PROFILING)) {
 886                 kvm_x86_ops->cache_regs(vcpu);
 887                 profile_hit(KVM_PROFILING, (void *)vcpu->rip);
 888         }
 889
 890         r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
 891
 892         if (r > 0) {
 893                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
 894                         r = -EINTR;
 895                         kvm_run->exit_reason = KVM_EXIT_INTR;
 896                         ++vcpu->stat.request_irq_exits;
 897                         goto out;
 898                 }
 899                 if (!need_resched()) {
 900                         ++vcpu->stat.light_exits;
 901                         goto again;
 902                 }
 903         }
 904
 905 out:
 906         if (r > 0) {
 907                 kvm_resched(vcpu);
 908                 goto preempted;
 909         }
 910
 911         post_kvm_run_save(vcpu, kvm_run);
 912
 913         return r;
 914 }
 915
 916
 917 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 918 {
 919         int r;
 920         sigset_t sigsaved;
 921
 922         vcpu_load(vcpu);
 923
 924         if (unlikely(vcpu->mp_state == VCPU_MP_STATE_UNINITIALIZED)) {
 925                 kvm_vcpu_block(vcpu);
 926                 vcpu_put(vcpu);
 927                 return -EAGAIN;
 928         }
 929
 930         if (vcpu->sigset_active)
 931                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 932
 933         /* re-sync apic's tpr */
 934         if (!irqchip_in_kernel(vcpu->kvm))
 935                 set_cr8(vcpu, kvm_run->cr8);
 936
 937         if (vcpu->pio.cur_count) {
 938                 r = complete_pio(vcpu);
 939                 if (r)
 940                         goto out;
 941         }
 942 #if CONFIG_HAS_IOMEM
 943         if (vcpu->mmio_needed) {
 944                 memcpy(vcpu->mmio_data, kvm_run->mmio.data, 8);
 945                 vcpu->mmio_read_completed = 1;
 946                 vcpu->mmio_needed = 0;
 947                 r = emulate_instruction(vcpu, kvm_run,
 948                                         vcpu->mmio_fault_cr2, 0, 1);
 949                 if (r == EMULATE_DO_MMIO) {
 950                         /*
 951                          * Read-modify-write.  Back to userspace.
 952                          */
 953                         r = 0;
 954                         goto out;
 955                 }
 956         }
 957 #endif
 958         if (kvm_run->exit_reason == KVM_EXIT_HYPERCALL) {
 959                 kvm_x86_ops->cache_regs(vcpu);
 960                 vcpu->regs[VCPU_REGS_RAX] = kvm_run->hypercall.ret;
 961                 kvm_x86_ops->decache_regs(vcpu);
 962         }
 963
 964         r = __vcpu_run(vcpu, kvm_run);
 965
 966 out:
 967         if (vcpu->sigset_active)
 968                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 969
 970         vcpu_put(vcpu);
 971         return r;
 972 }
 973
 974 static int kvm_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu,
 975                                    struct kvm_regs *regs)
 976 {
 977         vcpu_load(vcpu);
 978
 979         kvm_x86_ops->cache_regs(vcpu);
 980
 981         regs->rax = vcpu->regs[VCPU_REGS_RAX];
 982         regs->rbx = vcpu->regs[VCPU_REGS_RBX];
 983         regs->rcx = vcpu->regs[VCPU_REGS_RCX];
 984         regs->rdx = vcpu->regs[VCPU_REGS_RDX];
 985         regs->rsi = vcpu->regs[VCPU_REGS_RSI];
 986         regs->rdi = vcpu->regs[VCPU_REGS_RDI];
 987         regs->rsp = vcpu->regs[VCPU_REGS_RSP];
 988         regs->rbp = vcpu->regs[VCPU_REGS_RBP];
 989 #ifdef CONFIG_X86_64
 990         regs->r8 = vcpu->regs[VCPU_REGS_R8];
 991         regs->r9 = vcpu->regs[VCPU_REGS_R9];
 992         regs->r10 = vcpu->regs[VCPU_REGS_R10];
 993         regs->r11 = vcpu->regs[VCPU_REGS_R11];
 994         regs->r12 = vcpu->regs[VCPU_REGS_R12];
 995         regs->r13 = vcpu->regs[VCPU_REGS_R13];
 996         regs->r14 = vcpu->regs[VCPU_REGS_R14];
 997         regs->r15 = vcpu->regs[VCPU_REGS_R15];
 998 #endif
 999
1000         regs->rip = vcpu->rip;
1001         regs->rflags = kvm_x86_ops->get_rflags(vcpu);
1002
1003         /*
1004          * Don't leak debug flags in case they were set for guest debugging
1005          */
1006         if (vcpu->guest_debug.enabled && vcpu->guest_debug.singlestep)
1007                 regs->rflags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1008
1009         vcpu_put(vcpu);
1010
1011         return 0;
1012 }
1013
1014 static int kvm_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu,
1015                                    struct kvm_regs *regs)
1016 {
1017         vcpu_load(vcpu);
1018
1019         vcpu->regs[VCPU_REGS_RAX] = regs->rax;
1020         vcpu->regs[VCPU_REGS_RBX] = regs->rbx;
1021         vcpu->regs[VCPU_REGS_RCX] = regs->rcx;
1022         vcpu->regs[VCPU_REGS_RDX] = regs->rdx;
1023         vcpu->regs[VCPU_REGS_RSI] = regs->rsi;
1024         vcpu->regs[VCPU_REGS_RDI] = regs->rdi;
1025         vcpu->regs[VCPU_REGS_RSP] = regs->rsp;
1026         vcpu->regs[VCPU_REGS_RBP] = regs->rbp;
1027 #ifdef CONFIG_X86_64
1028         vcpu->regs[VCPU_REGS_R8] = regs->r8;
1029         vcpu->regs[VCPU_REGS_R9] = regs->r9;
1030         vcpu->regs[VCPU_REGS_R10] = regs->r10;
1031         vcpu->regs[VCPU_REGS_R11] = regs->r11;
1032         vcpu->regs[VCPU_REGS_R12] = regs->r12;
1033         vcpu->regs[VCPU_REGS_R13] = regs->r13;
1034         vcpu->regs[VCPU_REGS_R14] = regs->r14;
1035         vcpu->regs[VCPU_REGS_R15] = regs->r15;
1036 #endif
1037
1038         vcpu->rip = regs->rip;
1039         kvm_x86_ops->set_rflags(vcpu, regs->rflags);
1040
1041         kvm_x86_ops->decache_regs(vcpu);
1042
1043         vcpu_put(vcpu);
1044
1045         return 0;
1046 }
1047
1048 static void get_segment(struct kvm_vcpu *vcpu,
1049                         struct kvm_segment *var, int seg)
1050 {
1051         return kvm_x86_ops->get_segment(vcpu, var, seg);
1052 }
1053
1054 static int kvm_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
1055                                     struct kvm_sregs *sregs)
1056 {
1057         struct descriptor_table dt;
1058         int pending_vec;
1059
1060         vcpu_load(vcpu);
1061
1062         get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1063         get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1064         get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1065         get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1066         get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1067         get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1068
1069         get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1070         get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1071
1072         kvm_x86_ops->get_idt(vcpu, &dt);
1073         sregs->idt.limit = dt.limit;
1074         sregs->idt.base = dt.base;
1075         kvm_x86_ops->get_gdt(vcpu, &dt);
1076         sregs->gdt.limit = dt.limit;
1077         sregs->gdt.base = dt.base;
1078
1079         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1080         sregs->cr0 = vcpu->cr0;
1081         sregs->cr2 = vcpu->cr2;
1082         sregs->cr3 = vcpu->cr3;
1083         sregs->cr4 = vcpu->cr4;
1084         sregs->cr8 = get_cr8(vcpu);
1085         sregs->efer = vcpu->shadow_efer;
1086         sregs->apic_base = kvm_get_apic_base(vcpu);
1087
1088         if (irqchip_in_kernel(vcpu->kvm)) {
1089                 memset(sregs->interrupt_bitmap, 0,
1090                        sizeof sregs->interrupt_bitmap);
1091                 pending_vec = kvm_x86_ops->get_irq(vcpu);
1092                 if (pending_vec >= 0)
1093                         set_bit(pending_vec,
1094                                 (unsigned long *)sregs->interrupt_bitmap);
1095         } else
1096                 memcpy(sregs->interrupt_bitmap, vcpu->irq_pending,
1097                        sizeof sregs->interrupt_bitmap);
1098
1099         vcpu_put(vcpu);
1100
1101         return 0;
1102 }
1103
1104 static void set_segment(struct kvm_vcpu *vcpu,
1105                         struct kvm_segment *var, int seg)
1106 {
1107         return kvm_x86_ops->set_segment(vcpu, var, seg);
1108 }
1109
1110 static int kvm_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
1111                                     struct kvm_sregs *sregs)
1112 {
1113         int mmu_reset_needed = 0;
1114         int i, pending_vec, max_bits;
1115         struct descriptor_table dt;
1116
1117         vcpu_load(vcpu);
1118
1119         dt.limit = sregs->idt.limit;
1120         dt.base = sregs->idt.base;
1121         kvm_x86_ops->set_idt(vcpu, &dt);
1122         dt.limit = sregs->gdt.limit;
1123         dt.base = sregs->gdt.base;
1124         kvm_x86_ops->set_gdt(vcpu, &dt);
1125
1126         vcpu->cr2 = sregs->cr2;
1127         mmu_reset_needed |= vcpu->cr3 != sregs->cr3;
1128         vcpu->cr3 = sregs->cr3;
1129
1130         set_cr8(vcpu, sregs->cr8);
1131
1132         mmu_reset_needed |= vcpu->shadow_efer != sregs->efer;
1133 #ifdef CONFIG_X86_64
1134         kvm_x86_ops->set_efer(vcpu, sregs->efer);
1135 #endif
1136         kvm_set_apic_base(vcpu, sregs->apic_base);
1137
1138         kvm_x86_ops->decache_cr4_guest_bits(vcpu);
1139
1140         mmu_reset_needed |= vcpu->cr0 != sregs->cr0;
1141         vcpu->cr0 = sregs->cr0;
1142         kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
1143
1144         mmu_reset_needed |= vcpu->cr4 != sregs->cr4;
1145         kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
1146         if (!is_long_mode(vcpu) && is_pae(vcpu))
1147                 load_pdptrs(vcpu, vcpu->cr3);
1148
1149         if (mmu_reset_needed)
1150                 kvm_mmu_reset_context(vcpu);
1151
1152         if (!irqchip_in_kernel(vcpu->kvm)) {
1153                 memcpy(vcpu->irq_pending, sregs->interrupt_bitmap,
1154                        sizeof vcpu->irq_pending);
1155                 vcpu->irq_summary = 0;
1156                 for (i = 0; i < ARRAY_SIZE(vcpu->irq_pending); ++i)
1157                         if (vcpu->irq_pending[i])
1158                                 __set_bit(i, &vcpu->irq_summary);
1159         } else {
1160                 max_bits = (sizeof sregs->interrupt_bitmap) << 3;
1161                 pending_vec = find_first_bit(
1162                         (const unsigned long *)sregs->interrupt_bitmap,
1163                         max_bits);
1164                 /* Only pending external irq is handled here */
1165                 if (pending_vec < max_bits) {
1166                         kvm_x86_ops->set_irq(vcpu, pending_vec);
1167                         pr_debug("Set back pending irq %d\n",
1168                                  pending_vec);
1169                 }
1170         }
1171
1172         set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
1173         set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
1174         set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
1175         set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
1176         set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
1177         set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
1178
1179         set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
1180         set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
1181
1182         vcpu_put(vcpu);
1183
1184         return 0;
1185 }
1186
1187 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1188 {
1189         struct kvm_segment cs;
1190
1191         get_segment(vcpu, &cs, VCPU_SREG_CS);
1192         *db = cs.db;
1193         *l = cs.l;
1194 }
1195 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
1196
1197 /*
1198  * Translate a guest virtual address to a guest physical address.
1199  */
1200 static int kvm_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
1201                                     struct kvm_translation *tr)
1202 {
1203         unsigned long vaddr = tr->linear_address;
1204         gpa_t gpa;
1205
1206         vcpu_load(vcpu);
1207         mutex_lock(&vcpu->kvm->lock);
1208         gpa = vcpu->mmu.gva_to_gpa(vcpu, vaddr);
1209         tr->physical_address = gpa;
1210         tr->valid = gpa != UNMAPPED_GVA;
1211         tr->writeable = 1;
1212         tr->usermode = 0;
1213         mutex_unlock(&vcpu->kvm->lock);
1214         vcpu_put(vcpu);
1215
1216         return 0;
1217 }
1218
1219 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
1220                                     struct kvm_interrupt *irq)
1221 {
1222         if (irq->irq < 0 || irq->irq >= 256)
1223                 return -EINVAL;
1224         if (irqchip_in_kernel(vcpu->kvm))
1225                 return -ENXIO;
1226         vcpu_load(vcpu);
1227
1228         set_bit(irq->irq, vcpu->irq_pending);
1229         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
1230
1231         vcpu_put(vcpu);
1232
1233         return 0;
1234 }
1235
1236 static int kvm_vcpu_ioctl_debug_guest(struct kvm_vcpu *vcpu,
1237                                       struct kvm_debug_guest *dbg)
1238 {
1239         int r;
1240
1241         vcpu_load(vcpu);
1242
1243         r = kvm_x86_ops->set_guest_debug(vcpu, dbg);
1244
1245         vcpu_put(vcpu);
1246
1247         return r;
1248 }
1249
1250 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
1251                                     unsigned long address,
1252                                     int *type)
1253 {
1254         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1255         unsigned long pgoff;
1256         struct page *page;
1257
1258         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1259         if (pgoff == 0)
1260                 page = virt_to_page(vcpu->run);
1261         else if (pgoff == KVM_PIO_PAGE_OFFSET)
1262                 page = virt_to_page(vcpu->pio_data);
1263         else
1264                 return NOPAGE_SIGBUS;
1265         get_page(page);
1266         if (type != NULL)
1267                 *type = VM_FAULT_MINOR;
1268
1269         return page;
1270 }
1271
1272 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1273         .nopage = kvm_vcpu_nopage,
1274 };
1275
1276 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1277 {
1278         vma->vm_ops = &kvm_vcpu_vm_ops;
1279         return 0;
1280 }
1281
1282 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1283 {
1284         struct kvm_vcpu *vcpu = filp->private_data;
1285
1286         fput(vcpu->kvm->filp);
1287         return 0;
1288 }
1289
1290 static struct file_operations kvm_vcpu_fops = {
1291         .release        = kvm_vcpu_release,
1292         .unlocked_ioctl = kvm_vcpu_ioctl,
1293         .compat_ioctl   = kvm_vcpu_ioctl,
1294         .mmap           = kvm_vcpu_mmap,
1295 };
1296
1297 /*
1298  * Allocates an inode for the vcpu.
1299  */
1300 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1301 {
1302         int fd, r;
1303         struct inode *inode;
1304         struct file *file;
1305
1306         r = anon_inode_getfd(&fd, &inode, &file,
1307                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
1308         if (r)
1309                 return r;
1310         atomic_inc(&vcpu->kvm->filp->f_count);
1311         return fd;
1312 }
1313
1314 /*
1315  * Creates some virtual cpus.  Good luck creating more than one.
1316  */
1317 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1318 {
1319         int r;
1320         struct kvm_vcpu *vcpu;
1321
1322         if (!valid_vcpu(n))
1323                 return -EINVAL;
1324
1325         vcpu = kvm_x86_ops->vcpu_create(kvm, n);
1326         if (IS_ERR(vcpu))
1327                 return PTR_ERR(vcpu);
1328
1329         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1330
1331         /* We do fxsave: this must be aligned. */
1332         BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
1333
1334         vcpu_load(vcpu);
1335         r = kvm_x86_ops->vcpu_reset(vcpu);
1336         if (r == 0)
1337                 r = kvm_mmu_setup(vcpu);
1338         vcpu_put(vcpu);
1339         if (r < 0)
1340                 goto free_vcpu;
1341
1342         mutex_lock(&kvm->lock);
1343         if (kvm->vcpus[n]) {
1344                 r = -EEXIST;
1345                 mutex_unlock(&kvm->lock);
1346                 goto mmu_unload;
1347         }
1348         kvm->vcpus[n] = vcpu;
1349         mutex_unlock(&kvm->lock);
1350
1351         /* Now it's all set up, let userspace reach it */
1352         r = create_vcpu_fd(vcpu);
1353         if (r < 0)
1354                 goto unlink;
1355         return r;
1356
1357 unlink:
1358         mutex_lock(&kvm->lock);
1359         kvm->vcpus[n] = NULL;
1360         mutex_unlock(&kvm->lock);
1361
1362 mmu_unload:
1363         vcpu_load(vcpu);
1364         kvm_mmu_unload(vcpu);
1365         vcpu_put(vcpu);
1366
1367 free_vcpu:
1368         kvm_x86_ops->vcpu_free(vcpu);
1369         return r;
1370 }
1371
1372 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1373 {
1374         if (sigset) {
1375                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1376                 vcpu->sigset_active = 1;
1377                 vcpu->sigset = *sigset;
1378         } else
1379                 vcpu->sigset_active = 0;
1380         return 0;
1381 }
1382
1383 static long kvm_vcpu_ioctl(struct file *filp,
1384                            unsigned int ioctl, unsigned long arg)
1385 {
1386         struct kvm_vcpu *vcpu = filp->private_data;
1387         void __user *argp = (void __user *)arg;
1388         int r;
1389
1390         switch (ioctl) {
1391         case KVM_RUN:
1392                 r = -EINVAL;
1393                 if (arg)
1394                         goto out;
1395                 r = kvm_vcpu_ioctl_run(vcpu, vcpu->run);
1396                 break;
1397         case KVM_GET_REGS: {
1398                 struct kvm_regs kvm_regs;
1399
1400                 memset(&kvm_regs, 0, sizeof kvm_regs);
1401                 r = kvm_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
1402                 if (r)
1403                         goto out;
1404                 r = -EFAULT;
1405                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
1406                         goto out;
1407                 r = 0;
1408                 break;
1409         }
1410         case KVM_SET_REGS: {
1411                 struct kvm_regs kvm_regs;
1412
1413                 r = -EFAULT;
1414                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
1415                         goto out;
1416                 r = kvm_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
1417                 if (r)
1418                         goto out;
1419                 r = 0;
1420                 break;
1421         }
1422         case KVM_GET_SREGS: {
1423                 struct kvm_sregs kvm_sregs;
1424
1425                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
1426                 r = kvm_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
1427                 if (r)
1428                         goto out;
1429                 r = -EFAULT;
1430                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
1431                         goto out;
1432                 r = 0;
1433                 break;
1434         }
1435         case KVM_SET_SREGS: {
1436                 struct kvm_sregs kvm_sregs;
1437
1438                 r = -EFAULT;
1439                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
1440                         goto out;
1441                 r = kvm_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
1442                 if (r)
1443                         goto out;
1444                 r = 0;
1445                 break;
1446         }
1447         case KVM_TRANSLATE: {
1448                 struct kvm_translation tr;
1449
1450                 r = -EFAULT;
1451                 if (copy_from_user(&tr, argp, sizeof tr))
1452                         goto out;
1453                 r = kvm_vcpu_ioctl_translate(vcpu, &tr);
1454                 if (r)
1455                         goto out;
1456                 r = -EFAULT;
1457                 if (copy_to_user(argp, &tr, sizeof tr))
1458                         goto out;
1459                 r = 0;
1460                 break;
1461         }
1462         case KVM_INTERRUPT: {
1463                 struct kvm_interrupt irq;
1464
1465                 r = -EFAULT;
1466                 if (copy_from_user(&irq, argp, sizeof irq))
1467                         goto out;
1468                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
1469                 if (r)
1470                         goto out;
1471                 r = 0;
1472                 break;
1473         }
1474         case KVM_DEBUG_GUEST: {
1475                 struct kvm_debug_guest dbg;
1476
1477                 r = -EFAULT;
1478                 if (copy_from_user(&dbg, argp, sizeof dbg))
1479                         goto out;
1480                 r = kvm_vcpu_ioctl_debug_guest(vcpu, &dbg);
1481                 if (r)
1482                         goto out;
1483                 r = 0;
1484                 break;
1485         }
1486         case KVM_SET_SIGNAL_MASK: {
1487                 struct kvm_signal_mask __user *sigmask_arg = argp;
1488                 struct kvm_signal_mask kvm_sigmask;
1489                 sigset_t sigset, *p;
1490
1491                 p = NULL;
1492                 if (argp) {
1493                         r = -EFAULT;
1494                         if (copy_from_user(&kvm_sigmask, argp,
1495                                            sizeof kvm_sigmask))
1496                                 goto out;
1497                         r = -EINVAL;
1498                         if (kvm_sigmask.len != sizeof sigset)
1499                                 goto out;
1500                         r = -EFAULT;
1501                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1502                                            sizeof sigset))
1503                                 goto out;
1504                         p = &sigset;
1505                 }
1506                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1507                 break;
1508         }
1509         case KVM_GET_FPU: {
1510                 struct kvm_fpu fpu;
1511
1512                 memset(&fpu, 0, sizeof fpu);
1513                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1514                 if (r)
1515                         goto out;
1516                 r = -EFAULT;
1517                 if (copy_to_user(argp, &fpu, sizeof fpu))
1518                         goto out;
1519                 r = 0;
1520                 break;
1521         }
1522         case KVM_SET_FPU: {
1523                 struct kvm_fpu fpu;
1524
1525                 r = -EFAULT;
1526                 if (copy_from_user(&fpu, argp, sizeof fpu))
1527                         goto out;
1528                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1529                 if (r)
1530                         goto out;
1531                 r = 0;
1532                 break;
1533         }
1534         default:
1535                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1536         }
1537 out:
1538         return r;
1539 }
1540
1541 static long kvm_vm_ioctl(struct file *filp,
1542                            unsigned int ioctl, unsigned long arg)
1543 {
1544         struct kvm *kvm = filp->private_data;
1545         void __user *argp = (void __user *)arg;
1546         int r;
1547
1548         switch (ioctl) {
1549         case KVM_CREATE_VCPU:
1550                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1551                 if (r < 0)
1552                         goto out;
1553                 break;
1554         case KVM_SET_USER_MEMORY_REGION: {
1555                 struct kvm_userspace_memory_region kvm_userspace_mem;
1556
1557                 r = -EFAULT;
1558                 if (copy_from_user(&kvm_userspace_mem, argp,
1559                                                 sizeof kvm_userspace_mem))
1560                         goto out;
1561
1562                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1563                 if (r)
1564                         goto out;
1565                 break;
1566         }
1567         case KVM_GET_DIRTY_LOG: {
1568                 struct kvm_dirty_log log;
1569
1570                 r = -EFAULT;
1571                 if (copy_from_user(&log, argp, sizeof log))
1572                         goto out;
1573                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1574                 if (r)
1575                         goto out;
1576                 break;
1577         }
1578         default:
1579                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1580         }
1581 out:
1582         return r;
1583 }
1584
1585 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1586                                   unsigned long address,
1587                                   int *type)
1588 {
1589         struct kvm *kvm = vma->vm_file->private_data;
1590         unsigned long pgoff;
1591         struct page *page;
1592
1593         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1594         if (!kvm_is_visible_gfn(kvm, pgoff))
1595                 return NOPAGE_SIGBUS;
1596         /* current->mm->mmap_sem is already held so call lockless version */
1597         page = __gfn_to_page(kvm, pgoff);
1598         if (is_error_page(page)) {
1599                 kvm_release_page(page);
1600                 return NOPAGE_SIGBUS;
1601         }
1602         if (type != NULL)
1603                 *type = VM_FAULT_MINOR;
1604
1605         return page;
1606 }
1607
1608 static struct vm_operations_struct kvm_vm_vm_ops = {
1609         .nopage = kvm_vm_nopage,
1610 };
1611
1612 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1613 {
1614         vma->vm_ops = &kvm_vm_vm_ops;
1615         return 0;
1616 }
1617
1618 static struct file_operations kvm_vm_fops = {
1619         .release        = kvm_vm_release,
1620         .unlocked_ioctl = kvm_vm_ioctl,
1621         .compat_ioctl   = kvm_vm_ioctl,
1622         .mmap           = kvm_vm_mmap,
1623 };
1624
1625 static int kvm_dev_ioctl_create_vm(void)
1626 {
1627         int fd, r;
1628         struct inode *inode;
1629         struct file *file;
1630         struct kvm *kvm;
1631
1632         kvm = kvm_create_vm();
1633         if (IS_ERR(kvm))
1634                 return PTR_ERR(kvm);
1635         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1636         if (r) {
1637                 kvm_destroy_vm(kvm);
1638                 return r;
1639         }
1640
1641         kvm->filp = file;
1642
1643         return fd;
1644 }
1645
1646 static long kvm_dev_ioctl(struct file *filp,
1647                           unsigned int ioctl, unsigned long arg)
1648 {
1649         void __user *argp = (void __user *)arg;
1650         long r = -EINVAL;
1651
1652         switch (ioctl) {
1653         case KVM_GET_API_VERSION:
1654                 r = -EINVAL;
1655                 if (arg)
1656                         goto out;
1657                 r = KVM_API_VERSION;
1658                 break;
1659         case KVM_CREATE_VM:
1660                 r = -EINVAL;
1661                 if (arg)
1662                         goto out;
1663                 r = kvm_dev_ioctl_create_vm();
1664                 break;
1665         case KVM_CHECK_EXTENSION: {
1666                 int ext = (long)argp;
1667
1668                 switch (ext) {
1669                 case KVM_CAP_IRQCHIP:
1670                 case KVM_CAP_HLT:
1671                 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
1672                 case KVM_CAP_USER_MEMORY:
1673                 case KVM_CAP_SET_TSS_ADDR:
1674                         r = 1;
1675                         break;
1676                 default:
1677                         r = 0;
1678                         break;
1679                 }
1680                 break;
1681         }
1682         case KVM_GET_VCPU_MMAP_SIZE:
1683                 r = -EINVAL;
1684                 if (arg)
1685                         goto out;
1686                 r = 2 * PAGE_SIZE;
1687                 break;
1688         default:
1689                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1690         }
1691 out:
1692         return r;
1693 }
1694
1695 static struct file_operations kvm_chardev_ops = {
1696         .unlocked_ioctl = kvm_dev_ioctl,
1697         .compat_ioctl   = kvm_dev_ioctl,
1698 };
1699
1700 static struct miscdevice kvm_dev = {
1701         KVM_MINOR,
1702         "kvm",
1703         &kvm_chardev_ops,
1704 };
1705
1706 /*
1707  * Make sure that a cpu that is being hot-unplugged does not have any vcpus
1708  * cached on it.
1709  */
1710 static void decache_vcpus_on_cpu(int cpu)
1711 {
1712         struct kvm *vm;
1713         struct kvm_vcpu *vcpu;
1714         int i;
1715
1716         spin_lock(&kvm_lock);
1717         list_for_each_entry(vm, &vm_list, vm_list)
1718                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1719                         vcpu = vm->vcpus[i];
1720                         if (!vcpu)
1721                                 continue;
1722                         /*
1723                          * If the vcpu is locked, then it is running on some
1724                          * other cpu and therefore it is not cached on the
1725                          * cpu in question.
1726                          *
1727                          * If it's not locked, check the last cpu it executed
1728                          * on.
1729                          */
1730                         if (mutex_trylock(&vcpu->mutex)) {
1731                                 if (vcpu->cpu == cpu) {
1732                                         kvm_x86_ops->vcpu_decache(vcpu);
1733                                         vcpu->cpu = -1;
1734                                 }
1735                                 mutex_unlock(&vcpu->mutex);
1736                         }
1737                 }
1738         spin_unlock(&kvm_lock);
1739 }
1740
1741 static void hardware_enable(void *junk)
1742 {
1743         int cpu = raw_smp_processor_id();
1744
1745         if (cpu_isset(cpu, cpus_hardware_enabled))
1746                 return;
1747         cpu_set(cpu, cpus_hardware_enabled);
1748         kvm_x86_ops->hardware_enable(NULL);
1749 }
1750
1751 static void hardware_disable(void *junk)
1752 {
1753         int cpu = raw_smp_processor_id();
1754
1755         if (!cpu_isset(cpu, cpus_hardware_enabled))
1756                 return;
1757         cpu_clear(cpu, cpus_hardware_enabled);
1758         decache_vcpus_on_cpu(cpu);
1759         kvm_x86_ops->hardware_disable(NULL);
1760 }
1761
1762 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1763                            void *v)
1764 {
1765         int cpu = (long)v;
1766
1767         switch (val) {
1768         case CPU_DYING:
1769         case CPU_DYING_FROZEN:
1770                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1771                        cpu);
1772                 hardware_disable(NULL);
1773                 break;
1774         case CPU_UP_CANCELED:
1775         case CPU_UP_CANCELED_FROZEN:
1776                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1777                        cpu);
1778                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1779                 break;
1780         case CPU_ONLINE:
1781         case CPU_ONLINE_FROZEN:
1782                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1783                        cpu);
1784                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1785                 break;
1786         }
1787         return NOTIFY_OK;
1788 }
1789
1790 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1791                       void *v)
1792 {
1793         if (val == SYS_RESTART) {
1794                 /*
1795                  * Some (well, at least mine) BIOSes hang on reboot if
1796                  * in vmx root mode.
1797                  */
1798                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1799                 on_each_cpu(hardware_disable, NULL, 0, 1);
1800         }
1801         return NOTIFY_OK;
1802 }
1803
1804 static struct notifier_block kvm_reboot_notifier = {
1805         .notifier_call = kvm_reboot,
1806         .priority = 0,
1807 };
1808
1809 void kvm_io_bus_init(struct kvm_io_bus *bus)
1810 {
1811         memset(bus, 0, sizeof(*bus));
1812 }
1813
1814 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1815 {
1816         int i;
1817
1818         for (i = 0; i < bus->dev_count; i++) {
1819                 struct kvm_io_device *pos = bus->devs[i];
1820
1821                 kvm_iodevice_destructor(pos);
1822         }
1823 }
1824
1825 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1826 {
1827         int i;
1828
1829         for (i = 0; i < bus->dev_count; i++) {
1830                 struct kvm_io_device *pos = bus->devs[i];
1831
1832                 if (pos->in_range(pos, addr))
1833                         return pos;
1834         }
1835
1836         return NULL;
1837 }
1838
1839 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1840 {
1841         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1842
1843         bus->devs[bus->dev_count++] = dev;
1844 }
1845
1846 static struct notifier_block kvm_cpu_notifier = {
1847         .notifier_call = kvm_cpu_hotplug,
1848         .priority = 20, /* must be > scheduler priority */
1849 };
1850
1851 static u64 stat_get(void *_offset)
1852 {
1853         unsigned offset = (long)_offset;
1854         u64 total = 0;
1855         struct kvm *kvm;
1856         struct kvm_vcpu *vcpu;
1857         int i;
1858
1859         spin_lock(&kvm_lock);
1860         list_for_each_entry(kvm, &vm_list, vm_list)
1861                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1862                         vcpu = kvm->vcpus[i];
1863                         if (vcpu)
1864                                 total += *(u32 *)((void *)vcpu + offset);
1865                 }
1866         spin_unlock(&kvm_lock);
1867         return total;
1868 }
1869
1870 DEFINE_SIMPLE_ATTRIBUTE(stat_fops, stat_get, NULL, "%llu\n");
1871
1872 static __init void kvm_init_debug(void)
1873 {
1874         struct kvm_stats_debugfs_item *p;
1875
1876         debugfs_dir = debugfs_create_dir("kvm", NULL);
1877         for (p = debugfs_entries; p->name; ++p)
1878                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1879                                                 (void *)(long)p->offset,
1880                                                 &stat_fops);
1881 }
1882
1883 static void kvm_exit_debug(void)
1884 {
1885         struct kvm_stats_debugfs_item *p;
1886
1887         for (p = debugfs_entries; p->name; ++p)
1888                 debugfs_remove(p->dentry);
1889         debugfs_remove(debugfs_dir);
1890 }
1891
1892 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1893 {
1894         hardware_disable(NULL);
1895         return 0;
1896 }
1897
1898 static int kvm_resume(struct sys_device *dev)
1899 {
1900         hardware_enable(NULL);
1901         return 0;
1902 }
1903
1904 static struct sysdev_class kvm_sysdev_class = {
1905         .name = "kvm",
1906         .suspend = kvm_suspend,
1907         .resume = kvm_resume,
1908 };
1909
1910 static struct sys_device kvm_sysdev = {
1911         .id = 0,
1912         .cls = &kvm_sysdev_class,
1913 };
1914
1915 struct page *bad_page;
1916
1917 static inline
1918 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1919 {
1920         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1921 }
1922
1923 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1924 {
1925         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1926
1927         kvm_x86_ops->vcpu_load(vcpu, cpu);
1928 }
1929
1930 static void kvm_sched_out(struct preempt_notifier *pn,
1931                           struct task_struct *next)
1932 {
1933         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1934
1935         kvm_x86_ops->vcpu_put(vcpu);
1936 }
1937
1938 int kvm_init_x86(struct kvm_x86_ops *ops, unsigned int vcpu_size,
1939                   struct module *module)
1940 {
1941         int r;
1942         int cpu;
1943
1944         if (kvm_x86_ops) {
1945                 printk(KERN_ERR "kvm: already loaded the other module\n");
1946                 return -EEXIST;
1947         }
1948
1949         if (!ops->cpu_has_kvm_support()) {
1950                 printk(KERN_ERR "kvm: no hardware support\n");
1951                 return -EOPNOTSUPP;
1952         }
1953         if (ops->disabled_by_bios()) {
1954                 printk(KERN_ERR "kvm: disabled by bios\n");
1955                 return -EOPNOTSUPP;
1956         }
1957
1958         kvm_x86_ops = ops;
1959
1960         r = kvm_x86_ops->hardware_setup();
1961         if (r < 0)
1962                 goto out;
1963
1964         for_each_online_cpu(cpu) {
1965                 smp_call_function_single(cpu,
1966                                 kvm_x86_ops->check_processor_compatibility,
1967                                 &r, 0, 1);
1968                 if (r < 0)
1969                         goto out_free_0;
1970         }
1971
1972         on_each_cpu(hardware_enable, NULL, 0, 1);
1973         r = register_cpu_notifier(&kvm_cpu_notifier);
1974         if (r)
1975                 goto out_free_1;
1976         register_reboot_notifier(&kvm_reboot_notifier);
1977
1978         r = sysdev_class_register(&kvm_sysdev_class);
1979         if (r)
1980                 goto out_free_2;
1981
1982         r = sysdev_register(&kvm_sysdev);
1983         if (r)
1984                 goto out_free_3;
1985
1986         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1987         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1988                                            __alignof__(struct kvm_vcpu), 0, 0);
1989         if (!kvm_vcpu_cache) {
1990                 r = -ENOMEM;
1991                 goto out_free_4;
1992         }
1993
1994         kvm_chardev_ops.owner = module;
1995
1996         r = misc_register(&kvm_dev);
1997         if (r) {
1998                 printk(KERN_ERR "kvm: misc device register failed\n");
1999                 goto out_free;
2000         }
2001
2002         kvm_preempt_ops.sched_in = kvm_sched_in;
2003         kvm_preempt_ops.sched_out = kvm_sched_out;
2004
2005         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
2006
2007         return 0;
2008
2009 out_free:
2010         kmem_cache_destroy(kvm_vcpu_cache);
2011 out_free_4:
2012         sysdev_unregister(&kvm_sysdev);
2013 out_free_3:
2014         sysdev_class_unregister(&kvm_sysdev_class);
2015 out_free_2:
2016         unregister_reboot_notifier(&kvm_reboot_notifier);
2017         unregister_cpu_notifier(&kvm_cpu_notifier);
2018 out_free_1:
2019         on_each_cpu(hardware_disable, NULL, 0, 1);
2020 out_free_0:
2021         kvm_x86_ops->hardware_unsetup();
2022 out:
2023         kvm_x86_ops = NULL;
2024         return r;
2025 }
2026 EXPORT_SYMBOL_GPL(kvm_init_x86);
2027
2028 void kvm_exit_x86(void)
2029 {
2030         misc_deregister(&kvm_dev);
2031         kmem_cache_destroy(kvm_vcpu_cache);
2032         sysdev_unregister(&kvm_sysdev);
2033         sysdev_class_unregister(&kvm_sysdev_class);
2034         unregister_reboot_notifier(&kvm_reboot_notifier);
2035         unregister_cpu_notifier(&kvm_cpu_notifier);
2036         on_each_cpu(hardware_disable, NULL, 0, 1);
2037         kvm_x86_ops->hardware_unsetup();
2038         kvm_x86_ops = NULL;
2039 }
2040 EXPORT_SYMBOL_GPL(kvm_exit_x86);
2041
2042 static __init int kvm_init(void)
2043 {
2044         int r;
2045
2046         r = kvm_mmu_module_init();
2047         if (r)
2048                 goto out4;
2049
2050         kvm_init_debug();
2051
2052         kvm_arch_init();
2053
2054         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2055
2056         if (bad_page == NULL) {
2057                 r = -ENOMEM;
2058                 goto out;
2059         }
2060
2061         return 0;
2062
2063 out:
2064         kvm_exit_debug();
2065         kvm_mmu_module_exit();
2066 out4:
2067         return r;
2068 }
2069
2070 static __exit void kvm_exit(void)
2071 {
2072         kvm_exit_debug();
2073         __free_page(bad_page);
2074         kvm_mmu_module_exit();
2075 }
2076
2077 module_init(kvm_init)
2078 module_exit(kvm_exit)