SAFE public projects git trees. - safe/jmp/linux-2.6/blob - virt/kvm/kvm_main.c

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