virt/kvm/kvm_main.c

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