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 "irq.h"
  21
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/gfp.h>
  27 #include <linux/mm.h>
  28 #include <linux/miscdevice.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/reboot.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/highmem.h>
  33 #include <linux/file.h>
  34 #include <linux/sysdev.h>
  35 #include <linux/cpu.h>
  36 #include <linux/sched.h>
  37 #include <linux/cpumask.h>
  38 #include <linux/smp.h>
  39 #include <linux/anon_inodes.h>
  40 #include <linux/profile.h>
  41 #include <linux/kvm_para.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mman.h>
  44
  45 #include <asm/processor.h>
  46 #include <asm/io.h>
  47 #include <asm/uaccess.h>
  48 #include <asm/desc.h>
  49
  50 MODULE_AUTHOR("Qumranet");
  51 MODULE_LICENSE("GPL");
  52
  53 DEFINE_SPINLOCK(kvm_lock);
  54 LIST_HEAD(vm_list);
  55
  56 static cpumask_t cpus_hardware_enabled;
  57
  58 struct kmem_cache *kvm_vcpu_cache;
  59 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  60
  61 static __read_mostly struct preempt_ops kvm_preempt_ops;
  62
  63 static struct dentry *debugfs_dir;
  64
  65 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  66                            unsigned long arg);
  67
  68 static inline int valid_vcpu(int n)
  69 {
  70         return likely(n >= 0 && n < KVM_MAX_VCPUS);
  71 }
  72
  73 /*
  74  * Switches to specified vcpu, until a matching vcpu_put()
  75  */
  76 void vcpu_load(struct kvm_vcpu *vcpu)
  77 {
  78         int cpu;
  79
  80         mutex_lock(&vcpu->mutex);
  81         cpu = get_cpu();
  82         preempt_notifier_register(&vcpu->preempt_notifier);
  83         kvm_arch_vcpu_load(vcpu, cpu);
  84         put_cpu();
  85 }
  86
  87 void vcpu_put(struct kvm_vcpu *vcpu)
  88 {
  89         preempt_disable();
  90         kvm_arch_vcpu_put(vcpu);
  91         preempt_notifier_unregister(&vcpu->preempt_notifier);
  92         preempt_enable();
  93         mutex_unlock(&vcpu->mutex);
  94 }
  95
  96 static void ack_flush(void *_completed)
  97 {
  98 }
  99
 100 void kvm_flush_remote_tlbs(struct kvm *kvm)
 101 {
 102         int i, cpu;
 103         cpumask_t cpus;
 104         struct kvm_vcpu *vcpu;
 105
 106         cpus_clear(cpus);
 107         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 108                 vcpu = kvm->vcpus[i];
 109                 if (!vcpu)
 110                         continue;
 111                 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
 112                         continue;
 113                 cpu = vcpu->cpu;
 114                 if (cpu != -1 && cpu != raw_smp_processor_id())
 115                         cpu_set(cpu, cpus);
 116         }
 117         smp_call_function_mask(cpus, ack_flush, NULL, 1);
 118 }
 119
 120 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 121 {
 122         struct page *page;
 123         int r;
 124
 125         mutex_init(&vcpu->mutex);
 126         vcpu->cpu = -1;
 127         vcpu->kvm = kvm;
 128         vcpu->vcpu_id = id;
 129         init_waitqueue_head(&vcpu->wq);
 130
 131         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 132         if (!page) {
 133                 r = -ENOMEM;
 134                 goto fail;
 135         }
 136         vcpu->run = page_address(page);
 137
 138         r = kvm_arch_vcpu_init(vcpu);
 139         if (r < 0)
 140                 goto fail_free_run;
 141         return 0;
 142
 143 fail_free_run:
 144         free_page((unsigned long)vcpu->run);
 145 fail:
 146         return r;
 147 }
 148 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 149
 150 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 151 {
 152         kvm_arch_vcpu_uninit(vcpu);
 153         free_page((unsigned long)vcpu->run);
 154 }
 155 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 156
 157 static struct kvm *kvm_create_vm(void)
 158 {
 159         struct kvm *kvm = kvm_arch_create_vm();
 160
 161         if (IS_ERR(kvm))
 162                 goto out;
 163
 164         kvm_io_bus_init(&kvm->pio_bus);
 165         mutex_init(&kvm->lock);
 166         kvm_io_bus_init(&kvm->mmio_bus);
 167         spin_lock(&kvm_lock);
 168         list_add(&kvm->vm_list, &vm_list);
 169         spin_unlock(&kvm_lock);
 170 out:
 171         return kvm;
 172 }
 173
 174 /*
 175  * Free any memory in @free but not in @dont.
 176  */
 177 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 178                                   struct kvm_memory_slot *dont)
 179 {
 180         if (!dont || free->rmap != dont->rmap)
 181                 vfree(free->rmap);
 182
 183         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 184                 vfree(free->dirty_bitmap);
 185
 186         free->npages = 0;
 187         free->dirty_bitmap = NULL;
 188         free->rmap = NULL;
 189 }
 190
 191 void kvm_free_physmem(struct kvm *kvm)
 192 {
 193         int i;
 194
 195         for (i = 0; i < kvm->nmemslots; ++i)
 196                 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
 197 }
 198
 199 static void kvm_destroy_vm(struct kvm *kvm)
 200 {
 201         spin_lock(&kvm_lock);
 202         list_del(&kvm->vm_list);
 203         spin_unlock(&kvm_lock);
 204         kvm_io_bus_destroy(&kvm->pio_bus);
 205         kvm_io_bus_destroy(&kvm->mmio_bus);
 206         kvm_arch_destroy_vm(kvm);
 207 }
 208
 209 static int kvm_vm_release(struct inode *inode, struct file *filp)
 210 {
 211         struct kvm *kvm = filp->private_data;
 212
 213         kvm_destroy_vm(kvm);
 214         return 0;
 215 }
 216
 217 /*
 218  * Allocate some memory and give it an address in the guest physical address
 219  * space.
 220  *
 221  * Discontiguous memory is allowed, mostly for framebuffers.
 222  *
 223  * Must be called holding kvm->lock.
 224  */
 225 int __kvm_set_memory_region(struct kvm *kvm,
 226                             struct kvm_userspace_memory_region *mem,
 227                             int user_alloc)
 228 {
 229         int r;
 230         gfn_t base_gfn;
 231         unsigned long npages;
 232         unsigned long i;
 233         struct kvm_memory_slot *memslot;
 234         struct kvm_memory_slot old, new;
 235
 236         r = -EINVAL;
 237         /* General sanity checks */
 238         if (mem->memory_size & (PAGE_SIZE - 1))
 239                 goto out;
 240         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 241                 goto out;
 242         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 243                 goto out;
 244         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 245                 goto out;
 246
 247         memslot = &kvm->memslots[mem->slot];
 248         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 249         npages = mem->memory_size >> PAGE_SHIFT;
 250
 251         if (!npages)
 252                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 253
 254         new = old = *memslot;
 255
 256         new.base_gfn = base_gfn;
 257         new.npages = npages;
 258         new.flags = mem->flags;
 259
 260         /* Disallow changing a memory slot's size. */
 261         r = -EINVAL;
 262         if (npages && old.npages && npages != old.npages)
 263                 goto out_free;
 264
 265         /* Check for overlaps */
 266         r = -EEXIST;
 267         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 268                 struct kvm_memory_slot *s = &kvm->memslots[i];
 269
 270                 if (s == memslot)
 271                         continue;
 272                 if (!((base_gfn + npages <= s->base_gfn) ||
 273                       (base_gfn >= s->base_gfn + s->npages)))
 274                         goto out_free;
 275         }
 276
 277         /* Free page dirty bitmap if unneeded */
 278         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 279                 new.dirty_bitmap = NULL;
 280
 281         r = -ENOMEM;
 282
 283         /* Allocate if a slot is being created */
 284         if (npages && !new.rmap) {
 285                 new.rmap = vmalloc(npages * sizeof(struct page *));
 286
 287                 if (!new.rmap)
 288                         goto out_free;
 289
 290                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 291
 292                 new.user_alloc = user_alloc;
 293                 if (user_alloc)
 294                         new.userspace_addr = mem->userspace_addr;
 295                 else {
 296                         down_write(&current->mm->mmap_sem);
 297                         new.userspace_addr = do_mmap(NULL, 0,
 298                                                      npages * PAGE_SIZE,
 299                                                      PROT_READ | PROT_WRITE,
 300                                                      MAP_SHARED | MAP_ANONYMOUS,
 301                                                      0);
 302                         up_write(&current->mm->mmap_sem);
 303
 304                         if (IS_ERR((void *)new.userspace_addr))
 305                                 goto out_free;
 306                 }
 307         } else {
 308                 if (!old.user_alloc && old.rmap) {
 309                         int ret;
 310
 311                         down_write(&current->mm->mmap_sem);
 312                         ret = do_munmap(current->mm, old.userspace_addr,
 313                                         old.npages * PAGE_SIZE);
 314                         up_write(&current->mm->mmap_sem);
 315                         if (ret < 0)
 316                                 printk(KERN_WARNING
 317                                        "kvm_vm_ioctl_set_memory_region: "
 318                                        "failed to munmap memory\n");
 319                 }
 320         }
 321
 322         /* Allocate page dirty bitmap if needed */
 323         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 324                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 325
 326                 new.dirty_bitmap = vmalloc(dirty_bytes);
 327                 if (!new.dirty_bitmap)
 328                         goto out_free;
 329                 memset(new.dirty_bitmap, 0, dirty_bytes);
 330         }
 331
 332         if (mem->slot >= kvm->nmemslots)
 333                 kvm->nmemslots = mem->slot + 1;
 334
 335         if (!kvm->n_requested_mmu_pages) {
 336                 unsigned int n_pages;
 337
 338                 if (npages) {
 339                         n_pages = npages * KVM_PERMILLE_MMU_PAGES / 1000;
 340                         kvm_mmu_change_mmu_pages(kvm, kvm->n_alloc_mmu_pages +
 341                                                  n_pages);
 342                 } else {
 343                         unsigned int nr_mmu_pages;
 344
 345                         n_pages = old.npages * KVM_PERMILLE_MMU_PAGES / 1000;
 346                         nr_mmu_pages = kvm->n_alloc_mmu_pages - n_pages;
 347                         nr_mmu_pages = max(nr_mmu_pages,
 348                                         (unsigned int) KVM_MIN_ALLOC_MMU_PAGES);
 349                         kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
 350                 }
 351         }
 352
 353         *memslot = new;
 354
 355         kvm_mmu_slot_remove_write_access(kvm, mem->slot);
 356         kvm_flush_remote_tlbs(kvm);
 357
 358         kvm_free_physmem_slot(&old, &new);
 359         return 0;
 360
 361 out_free:
 362         kvm_free_physmem_slot(&new, &old);
 363 out:
 364         return r;
 365
 366 }
 367 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 368
 369 int kvm_set_memory_region(struct kvm *kvm,
 370                           struct kvm_userspace_memory_region *mem,
 371                           int user_alloc)
 372 {
 373         int r;
 374
 375         mutex_lock(&kvm->lock);
 376         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 377         mutex_unlock(&kvm->lock);
 378         return r;
 379 }
 380 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 381
 382 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 383                                    struct
 384                                    kvm_userspace_memory_region *mem,
 385                                    int user_alloc)
 386 {
 387         if (mem->slot >= KVM_MEMORY_SLOTS)
 388                 return -EINVAL;
 389         return kvm_set_memory_region(kvm, mem, user_alloc);
 390 }
 391
 392 int kvm_get_dirty_log(struct kvm *kvm,
 393                         struct kvm_dirty_log *log, int *is_dirty)
 394 {
 395         struct kvm_memory_slot *memslot;
 396         int r, i;
 397         int n;
 398         unsigned long any = 0;
 399
 400         r = -EINVAL;
 401         if (log->slot >= KVM_MEMORY_SLOTS)
 402                 goto out;
 403
 404         memslot = &kvm->memslots[log->slot];
 405         r = -ENOENT;
 406         if (!memslot->dirty_bitmap)
 407                 goto out;
 408
 409         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 410
 411         for (i = 0; !any && i < n/sizeof(long); ++i)
 412                 any = memslot->dirty_bitmap[i];
 413
 414         r = -EFAULT;
 415         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 416                 goto out;
 417
 418         if (any)
 419                 *is_dirty = 1;
 420
 421         r = 0;
 422 out:
 423         return r;
 424 }
 425
 426 int is_error_page(struct page *page)
 427 {
 428         return page == bad_page;
 429 }
 430 EXPORT_SYMBOL_GPL(is_error_page);
 431
 432 static inline unsigned long bad_hva(void)
 433 {
 434         return PAGE_OFFSET;
 435 }
 436
 437 int kvm_is_error_hva(unsigned long addr)
 438 {
 439         return addr == bad_hva();
 440 }
 441 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 442
 443 gfn_t unalias_gfn(struct kvm *kvm, gfn_t gfn)
 444 {
 445         int i;
 446         struct kvm_mem_alias *alias;
 447
 448         for (i = 0; i < kvm->naliases; ++i) {
 449                 alias = &kvm->aliases[i];
 450                 if (gfn >= alias->base_gfn
 451                     && gfn < alias->base_gfn + alias->npages)
 452                         return alias->target_gfn + gfn - alias->base_gfn;
 453         }
 454         return gfn;
 455 }
 456
 457 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 458 {
 459         int i;
 460
 461         for (i = 0; i < kvm->nmemslots; ++i) {
 462                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 463
 464                 if (gfn >= memslot->base_gfn
 465                     && gfn < memslot->base_gfn + memslot->npages)
 466                         return memslot;
 467         }
 468         return NULL;
 469 }
 470
 471 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 472 {
 473         gfn = unalias_gfn(kvm, gfn);
 474         return __gfn_to_memslot(kvm, gfn);
 475 }
 476
 477 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 478 {
 479         int i;
 480
 481         gfn = unalias_gfn(kvm, gfn);
 482         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 483                 struct kvm_memory_slot *memslot = &kvm->memslots[i];
 484
 485                 if (gfn >= memslot->base_gfn
 486                     && gfn < memslot->base_gfn + memslot->npages)
 487                         return 1;
 488         }
 489         return 0;
 490 }
 491 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 492
 493 static unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 494 {
 495         struct kvm_memory_slot *slot;
 496
 497         gfn = unalias_gfn(kvm, gfn);
 498         slot = __gfn_to_memslot(kvm, gfn);
 499         if (!slot)
 500                 return bad_hva();
 501         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 502 }
 503
 504 /*
 505  * Requires current->mm->mmap_sem to be held
 506  */
 507 static struct page *__gfn_to_page(struct kvm *kvm, gfn_t gfn)
 508 {
 509         struct page *page[1];
 510         unsigned long addr;
 511         int npages;
 512
 513         might_sleep();
 514
 515         addr = gfn_to_hva(kvm, gfn);
 516         if (kvm_is_error_hva(addr)) {
 517                 get_page(bad_page);
 518                 return bad_page;
 519         }
 520
 521         npages = get_user_pages(current, current->mm, addr, 1, 1, 1, page,
 522                                 NULL);
 523
 524         if (npages != 1) {
 525                 get_page(bad_page);
 526                 return bad_page;
 527         }
 528
 529         return page[0];
 530 }
 531
 532 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 533 {
 534         struct page *page;
 535
 536         down_read(&current->mm->mmap_sem);
 537         page = __gfn_to_page(kvm, gfn);
 538         up_read(&current->mm->mmap_sem);
 539
 540         return page;
 541 }
 542
 543 EXPORT_SYMBOL_GPL(gfn_to_page);
 544
 545 void kvm_release_page(struct page *page)
 546 {
 547         if (!PageReserved(page))
 548                 SetPageDirty(page);
 549         put_page(page);
 550 }
 551 EXPORT_SYMBOL_GPL(kvm_release_page);
 552
 553 static int next_segment(unsigned long len, int offset)
 554 {
 555         if (len > PAGE_SIZE - offset)
 556                 return PAGE_SIZE - offset;
 557         else
 558                 return len;
 559 }
 560
 561 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 562                         int len)
 563 {
 564         int r;
 565         unsigned long addr;
 566
 567         addr = gfn_to_hva(kvm, gfn);
 568         if (kvm_is_error_hva(addr))
 569                 return -EFAULT;
 570         r = copy_from_user(data, (void __user *)addr + offset, len);
 571         if (r)
 572                 return -EFAULT;
 573         return 0;
 574 }
 575 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 576
 577 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 578 {
 579         gfn_t gfn = gpa >> PAGE_SHIFT;
 580         int seg;
 581         int offset = offset_in_page(gpa);
 582         int ret;
 583
 584         while ((seg = next_segment(len, offset)) != 0) {
 585                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 586                 if (ret < 0)
 587                         return ret;
 588                 offset = 0;
 589                 len -= seg;
 590                 data += seg;
 591                 ++gfn;
 592         }
 593         return 0;
 594 }
 595 EXPORT_SYMBOL_GPL(kvm_read_guest);
 596
 597 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
 598                          int offset, int len)
 599 {
 600         int r;
 601         unsigned long addr;
 602
 603         addr = gfn_to_hva(kvm, gfn);
 604         if (kvm_is_error_hva(addr))
 605                 return -EFAULT;
 606         r = copy_to_user((void __user *)addr + offset, data, len);
 607         if (r)
 608                 return -EFAULT;
 609         mark_page_dirty(kvm, gfn);
 610         return 0;
 611 }
 612 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
 613
 614 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
 615                     unsigned long len)
 616 {
 617         gfn_t gfn = gpa >> PAGE_SHIFT;
 618         int seg;
 619         int offset = offset_in_page(gpa);
 620         int ret;
 621
 622         while ((seg = next_segment(len, offset)) != 0) {
 623                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
 624                 if (ret < 0)
 625                         return ret;
 626                 offset = 0;
 627                 len -= seg;
 628                 data += seg;
 629                 ++gfn;
 630         }
 631         return 0;
 632 }
 633
 634 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
 635 {
 636         void *page_virt;
 637         struct page *page;
 638
 639         page = gfn_to_page(kvm, gfn);
 640         if (is_error_page(page)) {
 641                 kvm_release_page(page);
 642                 return -EFAULT;
 643         }
 644         page_virt = kmap_atomic(page, KM_USER0);
 645
 646         memset(page_virt + offset, 0, len);
 647
 648         kunmap_atomic(page_virt, KM_USER0);
 649         kvm_release_page(page);
 650         mark_page_dirty(kvm, gfn);
 651         return 0;
 652 }
 653 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
 654
 655 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
 656 {
 657         gfn_t gfn = gpa >> PAGE_SHIFT;
 658         int seg;
 659         int offset = offset_in_page(gpa);
 660         int ret;
 661
 662         while ((seg = next_segment(len, offset)) != 0) {
 663                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
 664                 if (ret < 0)
 665                         return ret;
 666                 offset = 0;
 667                 len -= seg;
 668                 ++gfn;
 669         }
 670         return 0;
 671 }
 672 EXPORT_SYMBOL_GPL(kvm_clear_guest);
 673
 674 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 675 {
 676         struct kvm_memory_slot *memslot;
 677
 678         gfn = unalias_gfn(kvm, gfn);
 679         memslot = __gfn_to_memslot(kvm, gfn);
 680         if (memslot && memslot->dirty_bitmap) {
 681                 unsigned long rel_gfn = gfn - memslot->base_gfn;
 682
 683                 /* avoid RMW */
 684                 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
 685                         set_bit(rel_gfn, memslot->dirty_bitmap);
 686         }
 687 }
 688
 689 /*
 690  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
 691  */
 692 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
 693 {
 694         DECLARE_WAITQUEUE(wait, current);
 695
 696         add_wait_queue(&vcpu->wq, &wait);
 697
 698         /*
 699          * We will block until either an interrupt or a signal wakes us up
 700          */
 701         while (!kvm_cpu_has_interrupt(vcpu)
 702                && !signal_pending(current)
 703                && vcpu->mp_state != VCPU_MP_STATE_RUNNABLE
 704                && vcpu->mp_state != VCPU_MP_STATE_SIPI_RECEIVED) {
 705                 set_current_state(TASK_INTERRUPTIBLE);
 706                 vcpu_put(vcpu);
 707                 schedule();
 708                 vcpu_load(vcpu);
 709         }
 710
 711         __set_current_state(TASK_RUNNING);
 712         remove_wait_queue(&vcpu->wq, &wait);
 713 }
 714
 715 void kvm_resched(struct kvm_vcpu *vcpu)
 716 {
 717         if (!need_resched())
 718                 return;
 719         cond_resched();
 720 }
 721 EXPORT_SYMBOL_GPL(kvm_resched);
 722
 723 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
 724                                     struct kvm_interrupt *irq)
 725 {
 726         if (irq->irq < 0 || irq->irq >= 256)
 727                 return -EINVAL;
 728         if (irqchip_in_kernel(vcpu->kvm))
 729                 return -ENXIO;
 730         vcpu_load(vcpu);
 731
 732         set_bit(irq->irq, vcpu->irq_pending);
 733         set_bit(irq->irq / BITS_PER_LONG, &vcpu->irq_summary);
 734
 735         vcpu_put(vcpu);
 736
 737         return 0;
 738 }
 739
 740 static struct page *kvm_vcpu_nopage(struct vm_area_struct *vma,
 741                                     unsigned long address,
 742                                     int *type)
 743 {
 744         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 745         unsigned long pgoff;
 746         struct page *page;
 747
 748         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 749         if (pgoff == 0)
 750                 page = virt_to_page(vcpu->run);
 751         else if (pgoff == KVM_PIO_PAGE_OFFSET)
 752                 page = virt_to_page(vcpu->pio_data);
 753         else
 754                 return NOPAGE_SIGBUS;
 755         get_page(page);
 756         if (type != NULL)
 757                 *type = VM_FAULT_MINOR;
 758
 759         return page;
 760 }
 761
 762 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 763         .nopage = kvm_vcpu_nopage,
 764 };
 765
 766 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 767 {
 768         vma->vm_ops = &kvm_vcpu_vm_ops;
 769         return 0;
 770 }
 771
 772 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 773 {
 774         struct kvm_vcpu *vcpu = filp->private_data;
 775
 776         fput(vcpu->kvm->filp);
 777         return 0;
 778 }
 779
 780 static struct file_operations kvm_vcpu_fops = {
 781         .release        = kvm_vcpu_release,
 782         .unlocked_ioctl = kvm_vcpu_ioctl,
 783         .compat_ioctl   = kvm_vcpu_ioctl,
 784         .mmap           = kvm_vcpu_mmap,
 785 };
 786
 787 /*
 788  * Allocates an inode for the vcpu.
 789  */
 790 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 791 {
 792         int fd, r;
 793         struct inode *inode;
 794         struct file *file;
 795
 796         r = anon_inode_getfd(&fd, &inode, &file,
 797                              "kvm-vcpu", &kvm_vcpu_fops, vcpu);
 798         if (r)
 799                 return r;
 800         atomic_inc(&vcpu->kvm->filp->f_count);
 801         return fd;
 802 }
 803
 804 /*
 805  * Creates some virtual cpus.  Good luck creating more than one.
 806  */
 807 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 808 {
 809         int r;
 810         struct kvm_vcpu *vcpu;
 811
 812         if (!valid_vcpu(n))
 813                 return -EINVAL;
 814
 815         vcpu = kvm_arch_vcpu_create(kvm, n);
 816         if (IS_ERR(vcpu))
 817                 return PTR_ERR(vcpu);
 818
 819         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 820
 821         mutex_lock(&kvm->lock);
 822         if (kvm->vcpus[n]) {
 823                 r = -EEXIST;
 824                 mutex_unlock(&kvm->lock);
 825                 goto vcpu_destroy;
 826         }
 827         kvm->vcpus[n] = vcpu;
 828         mutex_unlock(&kvm->lock);
 829
 830         /* Now it's all set up, let userspace reach it */
 831         r = create_vcpu_fd(vcpu);
 832         if (r < 0)
 833                 goto unlink;
 834         return r;
 835
 836 unlink:
 837         mutex_lock(&kvm->lock);
 838         kvm->vcpus[n] = NULL;
 839         mutex_unlock(&kvm->lock);
 840 vcpu_destroy:
 841         kvm_arch_vcpu_destory(vcpu);
 842         return r;
 843 }
 844
 845 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 846 {
 847         if (sigset) {
 848                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 849                 vcpu->sigset_active = 1;
 850                 vcpu->sigset = *sigset;
 851         } else
 852                 vcpu->sigset_active = 0;
 853         return 0;
 854 }
 855
 856 static long kvm_vcpu_ioctl(struct file *filp,
 857                            unsigned int ioctl, unsigned long arg)
 858 {
 859         struct kvm_vcpu *vcpu = filp->private_data;
 860         void __user *argp = (void __user *)arg;
 861         int r;
 862
 863         switch (ioctl) {
 864         case KVM_RUN:
 865                 r = -EINVAL;
 866                 if (arg)
 867                         goto out;
 868                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 869                 break;
 870         case KVM_GET_REGS: {
 871                 struct kvm_regs kvm_regs;
 872
 873                 memset(&kvm_regs, 0, sizeof kvm_regs);
 874                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, &kvm_regs);
 875                 if (r)
 876                         goto out;
 877                 r = -EFAULT;
 878                 if (copy_to_user(argp, &kvm_regs, sizeof kvm_regs))
 879                         goto out;
 880                 r = 0;
 881                 break;
 882         }
 883         case KVM_SET_REGS: {
 884                 struct kvm_regs kvm_regs;
 885
 886                 r = -EFAULT;
 887                 if (copy_from_user(&kvm_regs, argp, sizeof kvm_regs))
 888                         goto out;
 889                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, &kvm_regs);
 890                 if (r)
 891                         goto out;
 892                 r = 0;
 893                 break;
 894         }
 895         case KVM_GET_SREGS: {
 896                 struct kvm_sregs kvm_sregs;
 897
 898                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 899                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 900                 if (r)
 901                         goto out;
 902                 r = -EFAULT;
 903                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 904                         goto out;
 905                 r = 0;
 906                 break;
 907         }
 908         case KVM_SET_SREGS: {
 909                 struct kvm_sregs kvm_sregs;
 910
 911                 r = -EFAULT;
 912                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 913                         goto out;
 914                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 915                 if (r)
 916                         goto out;
 917                 r = 0;
 918                 break;
 919         }
 920         case KVM_TRANSLATE: {
 921                 struct kvm_translation tr;
 922
 923                 r = -EFAULT;
 924                 if (copy_from_user(&tr, argp, sizeof tr))
 925                         goto out;
 926                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
 927                 if (r)
 928                         goto out;
 929                 r = -EFAULT;
 930                 if (copy_to_user(argp, &tr, sizeof tr))
 931                         goto out;
 932                 r = 0;
 933                 break;
 934         }
 935         case KVM_INTERRUPT: {
 936                 struct kvm_interrupt irq;
 937
 938                 r = -EFAULT;
 939                 if (copy_from_user(&irq, argp, sizeof irq))
 940                         goto out;
 941                 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
 942                 if (r)
 943                         goto out;
 944                 r = 0;
 945                 break;
 946         }
 947         case KVM_DEBUG_GUEST: {
 948                 struct kvm_debug_guest dbg;
 949
 950                 r = -EFAULT;
 951                 if (copy_from_user(&dbg, argp, sizeof dbg))
 952                         goto out;
 953                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
 954                 if (r)
 955                         goto out;
 956                 r = 0;
 957                 break;
 958         }
 959         case KVM_SET_SIGNAL_MASK: {
 960                 struct kvm_signal_mask __user *sigmask_arg = argp;
 961                 struct kvm_signal_mask kvm_sigmask;
 962                 sigset_t sigset, *p;
 963
 964                 p = NULL;
 965                 if (argp) {
 966                         r = -EFAULT;
 967                         if (copy_from_user(&kvm_sigmask, argp,
 968                                            sizeof kvm_sigmask))
 969                                 goto out;
 970                         r = -EINVAL;
 971                         if (kvm_sigmask.len != sizeof sigset)
 972                                 goto out;
 973                         r = -EFAULT;
 974                         if (copy_from_user(&sigset, sigmask_arg->sigset,
 975                                            sizeof sigset))
 976                                 goto out;
 977                         p = &sigset;
 978                 }
 979                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
 980                 break;
 981         }
 982         case KVM_GET_FPU: {
 983                 struct kvm_fpu fpu;
 984
 985                 memset(&fpu, 0, sizeof fpu);
 986                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
 987                 if (r)
 988                         goto out;
 989                 r = -EFAULT;
 990                 if (copy_to_user(argp, &fpu, sizeof fpu))
 991                         goto out;
 992                 r = 0;
 993                 break;
 994         }
 995         case KVM_SET_FPU: {
 996                 struct kvm_fpu fpu;
 997
 998                 r = -EFAULT;
 999                 if (copy_from_user(&fpu, argp, sizeof fpu))
1000                         goto out;
1001                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1002                 if (r)
1003                         goto out;
1004                 r = 0;
1005                 break;
1006         }
1007         default:
1008                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1009         }
1010 out:
1011         return r;
1012 }
1013
1014 static long kvm_vm_ioctl(struct file *filp,
1015                            unsigned int ioctl, unsigned long arg)
1016 {
1017         struct kvm *kvm = filp->private_data;
1018         void __user *argp = (void __user *)arg;
1019         int r;
1020
1021         switch (ioctl) {
1022         case KVM_CREATE_VCPU:
1023                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1024                 if (r < 0)
1025                         goto out;
1026                 break;
1027         case KVM_SET_USER_MEMORY_REGION: {
1028                 struct kvm_userspace_memory_region kvm_userspace_mem;
1029
1030                 r = -EFAULT;
1031                 if (copy_from_user(&kvm_userspace_mem, argp,
1032                                                 sizeof kvm_userspace_mem))
1033                         goto out;
1034
1035                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1036                 if (r)
1037                         goto out;
1038                 break;
1039         }
1040         case KVM_GET_DIRTY_LOG: {
1041                 struct kvm_dirty_log log;
1042
1043                 r = -EFAULT;
1044                 if (copy_from_user(&log, argp, sizeof log))
1045                         goto out;
1046                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1047                 if (r)
1048                         goto out;
1049                 break;
1050         }
1051         default:
1052                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1053         }
1054 out:
1055         return r;
1056 }
1057
1058 static struct page *kvm_vm_nopage(struct vm_area_struct *vma,
1059                                   unsigned long address,
1060                                   int *type)
1061 {
1062         struct kvm *kvm = vma->vm_file->private_data;
1063         unsigned long pgoff;
1064         struct page *page;
1065
1066         pgoff = ((address - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1067         if (!kvm_is_visible_gfn(kvm, pgoff))
1068                 return NOPAGE_SIGBUS;
1069         /* current->mm->mmap_sem is already held so call lockless version */
1070         page = __gfn_to_page(kvm, pgoff);
1071         if (is_error_page(page)) {
1072                 kvm_release_page(page);
1073                 return NOPAGE_SIGBUS;
1074         }
1075         if (type != NULL)
1076                 *type = VM_FAULT_MINOR;
1077
1078         return page;
1079 }
1080
1081 static struct vm_operations_struct kvm_vm_vm_ops = {
1082         .nopage = kvm_vm_nopage,
1083 };
1084
1085 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1086 {
1087         vma->vm_ops = &kvm_vm_vm_ops;
1088         return 0;
1089 }
1090
1091 static struct file_operations kvm_vm_fops = {
1092         .release        = kvm_vm_release,
1093         .unlocked_ioctl = kvm_vm_ioctl,
1094         .compat_ioctl   = kvm_vm_ioctl,
1095         .mmap           = kvm_vm_mmap,
1096 };
1097
1098 static int kvm_dev_ioctl_create_vm(void)
1099 {
1100         int fd, r;
1101         struct inode *inode;
1102         struct file *file;
1103         struct kvm *kvm;
1104
1105         kvm = kvm_create_vm();
1106         if (IS_ERR(kvm))
1107                 return PTR_ERR(kvm);
1108         r = anon_inode_getfd(&fd, &inode, &file, "kvm-vm", &kvm_vm_fops, kvm);
1109         if (r) {
1110                 kvm_destroy_vm(kvm);
1111                 return r;
1112         }
1113
1114         kvm->filp = file;
1115
1116         return fd;
1117 }
1118
1119 static long kvm_dev_ioctl(struct file *filp,
1120                           unsigned int ioctl, unsigned long arg)
1121 {
1122         void __user *argp = (void __user *)arg;
1123         long r = -EINVAL;
1124
1125         switch (ioctl) {
1126         case KVM_GET_API_VERSION:
1127                 r = -EINVAL;
1128                 if (arg)
1129                         goto out;
1130                 r = KVM_API_VERSION;
1131                 break;
1132         case KVM_CREATE_VM:
1133                 r = -EINVAL;
1134                 if (arg)
1135                         goto out;
1136                 r = kvm_dev_ioctl_create_vm();
1137                 break;
1138         case KVM_CHECK_EXTENSION:
1139                 r = kvm_dev_ioctl_check_extension((long)argp);
1140                 break;
1141         case KVM_GET_VCPU_MMAP_SIZE:
1142                 r = -EINVAL;
1143                 if (arg)
1144                         goto out;
1145                 r = 2 * PAGE_SIZE;
1146                 break;
1147         default:
1148                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1149         }
1150 out:
1151         return r;
1152 }
1153
1154 static struct file_operations kvm_chardev_ops = {
1155         .unlocked_ioctl = kvm_dev_ioctl,
1156         .compat_ioctl   = kvm_dev_ioctl,
1157 };
1158
1159 static struct miscdevice kvm_dev = {
1160         KVM_MINOR,
1161         "kvm",
1162         &kvm_chardev_ops,
1163 };
1164
1165 static void hardware_enable(void *junk)
1166 {
1167         int cpu = raw_smp_processor_id();
1168
1169         if (cpu_isset(cpu, cpus_hardware_enabled))
1170                 return;
1171         cpu_set(cpu, cpus_hardware_enabled);
1172         kvm_arch_hardware_enable(NULL);
1173 }
1174
1175 static void hardware_disable(void *junk)
1176 {
1177         int cpu = raw_smp_processor_id();
1178
1179         if (!cpu_isset(cpu, cpus_hardware_enabled))
1180                 return;
1181         cpu_clear(cpu, cpus_hardware_enabled);
1182         decache_vcpus_on_cpu(cpu);
1183         kvm_arch_hardware_disable(NULL);
1184 }
1185
1186 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1187                            void *v)
1188 {
1189         int cpu = (long)v;
1190
1191         val &= ~CPU_TASKS_FROZEN;
1192         switch (val) {
1193         case CPU_DYING:
1194                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1195                        cpu);
1196                 hardware_disable(NULL);
1197                 break;
1198         case CPU_UP_CANCELED:
1199                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1200                        cpu);
1201                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1202                 break;
1203         case CPU_ONLINE:
1204                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1205                        cpu);
1206                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1207                 break;
1208         }
1209         return NOTIFY_OK;
1210 }
1211
1212 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1213                       void *v)
1214 {
1215         if (val == SYS_RESTART) {
1216                 /*
1217                  * Some (well, at least mine) BIOSes hang on reboot if
1218                  * in vmx root mode.
1219                  */
1220                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1221                 on_each_cpu(hardware_disable, NULL, 0, 1);
1222         }
1223         return NOTIFY_OK;
1224 }
1225
1226 static struct notifier_block kvm_reboot_notifier = {
1227         .notifier_call = kvm_reboot,
1228         .priority = 0,
1229 };
1230
1231 void kvm_io_bus_init(struct kvm_io_bus *bus)
1232 {
1233         memset(bus, 0, sizeof(*bus));
1234 }
1235
1236 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1237 {
1238         int i;
1239
1240         for (i = 0; i < bus->dev_count; i++) {
1241                 struct kvm_io_device *pos = bus->devs[i];
1242
1243                 kvm_iodevice_destructor(pos);
1244         }
1245 }
1246
1247 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1248 {
1249         int i;
1250
1251         for (i = 0; i < bus->dev_count; i++) {
1252                 struct kvm_io_device *pos = bus->devs[i];
1253
1254                 if (pos->in_range(pos, addr))
1255                         return pos;
1256         }
1257
1258         return NULL;
1259 }
1260
1261 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1262 {
1263         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1264
1265         bus->devs[bus->dev_count++] = dev;
1266 }
1267
1268 static struct notifier_block kvm_cpu_notifier = {
1269         .notifier_call = kvm_cpu_hotplug,
1270         .priority = 20, /* must be > scheduler priority */
1271 };
1272
1273 static u64 vm_stat_get(void *_offset)
1274 {
1275         unsigned offset = (long)_offset;
1276         u64 total = 0;
1277         struct kvm *kvm;
1278
1279         spin_lock(&kvm_lock);
1280         list_for_each_entry(kvm, &vm_list, vm_list)
1281                 total += *(u32 *)((void *)kvm + offset);
1282         spin_unlock(&kvm_lock);
1283         return total;
1284 }
1285
1286 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1287
1288 static u64 vcpu_stat_get(void *_offset)
1289 {
1290         unsigned offset = (long)_offset;
1291         u64 total = 0;
1292         struct kvm *kvm;
1293         struct kvm_vcpu *vcpu;
1294         int i;
1295
1296         spin_lock(&kvm_lock);
1297         list_for_each_entry(kvm, &vm_list, vm_list)
1298                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1299                         vcpu = kvm->vcpus[i];
1300                         if (vcpu)
1301                                 total += *(u32 *)((void *)vcpu + offset);
1302                 }
1303         spin_unlock(&kvm_lock);
1304         return total;
1305 }
1306
1307 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1308
1309 static struct file_operations *stat_fops[] = {
1310         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1311         [KVM_STAT_VM]   = &vm_stat_fops,
1312 };
1313
1314 static void kvm_init_debug(void)
1315 {
1316         struct kvm_stats_debugfs_item *p;
1317
1318         debugfs_dir = debugfs_create_dir("kvm", NULL);
1319         for (p = debugfs_entries; p->name; ++p)
1320                 p->dentry = debugfs_create_file(p->name, 0444, debugfs_dir,
1321                                                 (void *)(long)p->offset,
1322                                                 stat_fops[p->kind]);
1323 }
1324
1325 static void kvm_exit_debug(void)
1326 {
1327         struct kvm_stats_debugfs_item *p;
1328
1329         for (p = debugfs_entries; p->name; ++p)
1330                 debugfs_remove(p->dentry);
1331         debugfs_remove(debugfs_dir);
1332 }
1333
1334 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1335 {
1336         hardware_disable(NULL);
1337         return 0;
1338 }
1339
1340 static int kvm_resume(struct sys_device *dev)
1341 {
1342         hardware_enable(NULL);
1343         return 0;
1344 }
1345
1346 static struct sysdev_class kvm_sysdev_class = {
1347         .name = "kvm",
1348         .suspend = kvm_suspend,
1349         .resume = kvm_resume,
1350 };
1351
1352 static struct sys_device kvm_sysdev = {
1353         .id = 0,
1354         .cls = &kvm_sysdev_class,
1355 };
1356
1357 struct page *bad_page;
1358
1359 static inline
1360 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1361 {
1362         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1363 }
1364
1365 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1366 {
1367         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1368
1369         kvm_arch_vcpu_load(vcpu, cpu);
1370 }
1371
1372 static void kvm_sched_out(struct preempt_notifier *pn,
1373                           struct task_struct *next)
1374 {
1375         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1376
1377         kvm_arch_vcpu_put(vcpu);
1378 }
1379
1380 int kvm_init(void *opaque, unsigned int vcpu_size,
1381                   struct module *module)
1382 {
1383         int r;
1384         int cpu;
1385
1386         r = kvm_mmu_module_init();
1387         if (r)
1388                 goto out4;
1389
1390         kvm_init_debug();
1391
1392         r = kvm_arch_init(opaque);
1393         if (r)
1394                 goto out4;
1395
1396         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1397
1398         if (bad_page == NULL) {
1399                 r = -ENOMEM;
1400                 goto out;
1401         }
1402
1403         r = kvm_arch_hardware_setup();
1404         if (r < 0)
1405                 goto out;
1406
1407         for_each_online_cpu(cpu) {
1408                 smp_call_function_single(cpu,
1409                                 kvm_arch_check_processor_compat,
1410                                 &r, 0, 1);
1411                 if (r < 0)
1412                         goto out_free_0;
1413         }
1414
1415         on_each_cpu(hardware_enable, NULL, 0, 1);
1416         r = register_cpu_notifier(&kvm_cpu_notifier);
1417         if (r)
1418                 goto out_free_1;
1419         register_reboot_notifier(&kvm_reboot_notifier);
1420
1421         r = sysdev_class_register(&kvm_sysdev_class);
1422         if (r)
1423                 goto out_free_2;
1424
1425         r = sysdev_register(&kvm_sysdev);
1426         if (r)
1427                 goto out_free_3;
1428
1429         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1430         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1431                                            __alignof__(struct kvm_vcpu),
1432                                            0, NULL);
1433         if (!kvm_vcpu_cache) {
1434                 r = -ENOMEM;
1435                 goto out_free_4;
1436         }
1437
1438         kvm_chardev_ops.owner = module;
1439
1440         r = misc_register(&kvm_dev);
1441         if (r) {
1442                 printk(KERN_ERR "kvm: misc device register failed\n");
1443                 goto out_free;
1444         }
1445
1446         kvm_preempt_ops.sched_in = kvm_sched_in;
1447         kvm_preempt_ops.sched_out = kvm_sched_out;
1448
1449         kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
1450
1451         return 0;
1452
1453 out_free:
1454         kmem_cache_destroy(kvm_vcpu_cache);
1455 out_free_4:
1456         sysdev_unregister(&kvm_sysdev);
1457 out_free_3:
1458         sysdev_class_unregister(&kvm_sysdev_class);
1459 out_free_2:
1460         unregister_reboot_notifier(&kvm_reboot_notifier);
1461         unregister_cpu_notifier(&kvm_cpu_notifier);
1462 out_free_1:
1463         on_each_cpu(hardware_disable, NULL, 0, 1);
1464 out_free_0:
1465         kvm_arch_hardware_unsetup();
1466 out:
1467         kvm_arch_exit();
1468         kvm_exit_debug();
1469         kvm_mmu_module_exit();
1470 out4:
1471         return r;
1472 }
1473 EXPORT_SYMBOL_GPL(kvm_init);
1474
1475 void kvm_exit(void)
1476 {
1477         misc_deregister(&kvm_dev);
1478         kmem_cache_destroy(kvm_vcpu_cache);
1479         sysdev_unregister(&kvm_sysdev);
1480         sysdev_class_unregister(&kvm_sysdev_class);
1481         unregister_reboot_notifier(&kvm_reboot_notifier);
1482         unregister_cpu_notifier(&kvm_cpu_notifier);
1483         on_each_cpu(hardware_disable, NULL, 0, 1);
1484         kvm_arch_hardware_unsetup();
1485         kvm_arch_exit();
1486         kvm_exit_debug();
1487         __free_page(bad_page);
1488         kvm_mmu_module_exit();
1489 }
1490 EXPORT_SYMBOL_GPL(kvm_exit);