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 #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         DECLARE_WAITQUEUE(wait, current);
 762
 763         add_wait_queue(&vcpu->wq, &wait);
 764
 765         /*
 766          * We will block until either an interrupt or a signal wakes us up
 767          */
 768         while (!kvm_cpu_has_interrupt(vcpu)
 769                && !kvm_cpu_has_pending_timer(vcpu)
 770                && !signal_pending(current)
 771                && !kvm_arch_vcpu_runnable(vcpu)) {
 772                 set_current_state(TASK_INTERRUPTIBLE);
 773                 vcpu_put(vcpu);
 774                 schedule();
 775                 vcpu_load(vcpu);
 776         }
 777
 778         __set_current_state(TASK_RUNNING);
 779         remove_wait_queue(&vcpu->wq, &wait);
 780 }
 781
 782 void kvm_resched(struct kvm_vcpu *vcpu)
 783 {
 784         if (!need_resched())
 785                 return;
 786         cond_resched();
 787 }
 788 EXPORT_SYMBOL_GPL(kvm_resched);
 789
 790 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 791 {
 792         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
 793         struct page *page;
 794
 795         if (vmf->pgoff == 0)
 796                 page = virt_to_page(vcpu->run);
 797 #ifdef CONFIG_X86
 798         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
 799                 page = virt_to_page(vcpu->arch.pio_data);
 800 #endif
 801         else
 802                 return VM_FAULT_SIGBUS;
 803         get_page(page);
 804         vmf->page = page;
 805         return 0;
 806 }
 807
 808 static struct vm_operations_struct kvm_vcpu_vm_ops = {
 809         .fault = kvm_vcpu_fault,
 810 };
 811
 812 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
 813 {
 814         vma->vm_ops = &kvm_vcpu_vm_ops;
 815         return 0;
 816 }
 817
 818 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
 819 {
 820         struct kvm_vcpu *vcpu = filp->private_data;
 821
 822         kvm_put_kvm(vcpu->kvm);
 823         return 0;
 824 }
 825
 826 static const struct file_operations kvm_vcpu_fops = {
 827         .release        = kvm_vcpu_release,
 828         .unlocked_ioctl = kvm_vcpu_ioctl,
 829         .compat_ioctl   = kvm_vcpu_ioctl,
 830         .mmap           = kvm_vcpu_mmap,
 831 };
 832
 833 /*
 834  * Allocates an inode for the vcpu.
 835  */
 836 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
 837 {
 838         int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu);
 839         if (fd < 0)
 840                 kvm_put_kvm(vcpu->kvm);
 841         return fd;
 842 }
 843
 844 /*
 845  * Creates some virtual cpus.  Good luck creating more than one.
 846  */
 847 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 848 {
 849         int r;
 850         struct kvm_vcpu *vcpu;
 851
 852         if (!valid_vcpu(n))
 853                 return -EINVAL;
 854
 855         vcpu = kvm_arch_vcpu_create(kvm, n);
 856         if (IS_ERR(vcpu))
 857                 return PTR_ERR(vcpu);
 858
 859         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 860
 861         r = kvm_arch_vcpu_setup(vcpu);
 862         if (r)
 863                 goto vcpu_destroy;
 864
 865         mutex_lock(&kvm->lock);
 866         if (kvm->vcpus[n]) {
 867                 r = -EEXIST;
 868                 mutex_unlock(&kvm->lock);
 869                 goto vcpu_destroy;
 870         }
 871         kvm->vcpus[n] = vcpu;
 872         mutex_unlock(&kvm->lock);
 873
 874         /* Now it's all set up, let userspace reach it */
 875         kvm_get_kvm(kvm);
 876         r = create_vcpu_fd(vcpu);
 877         if (r < 0)
 878                 goto unlink;
 879         return r;
 880
 881 unlink:
 882         mutex_lock(&kvm->lock);
 883         kvm->vcpus[n] = NULL;
 884         mutex_unlock(&kvm->lock);
 885 vcpu_destroy:
 886         kvm_arch_vcpu_destroy(vcpu);
 887         return r;
 888 }
 889
 890 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
 891 {
 892         if (sigset) {
 893                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
 894                 vcpu->sigset_active = 1;
 895                 vcpu->sigset = *sigset;
 896         } else
 897                 vcpu->sigset_active = 0;
 898         return 0;
 899 }
 900
 901 static long kvm_vcpu_ioctl(struct file *filp,
 902                            unsigned int ioctl, unsigned long arg)
 903 {
 904         struct kvm_vcpu *vcpu = filp->private_data;
 905         void __user *argp = (void __user *)arg;
 906         int r;
 907
 908         if (vcpu->kvm->mm != current->mm)
 909                 return -EIO;
 910         switch (ioctl) {
 911         case KVM_RUN:
 912                 r = -EINVAL;
 913                 if (arg)
 914                         goto out;
 915                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
 916                 break;
 917         case KVM_GET_REGS: {
 918                 struct kvm_regs *kvm_regs;
 919
 920                 r = -ENOMEM;
 921                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 922                 if (!kvm_regs)
 923                         goto out;
 924                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
 925                 if (r)
 926                         goto out_free1;
 927                 r = -EFAULT;
 928                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
 929                         goto out_free1;
 930                 r = 0;
 931 out_free1:
 932                 kfree(kvm_regs);
 933                 break;
 934         }
 935         case KVM_SET_REGS: {
 936                 struct kvm_regs *kvm_regs;
 937
 938                 r = -ENOMEM;
 939                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
 940                 if (!kvm_regs)
 941                         goto out;
 942                 r = -EFAULT;
 943                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
 944                         goto out_free2;
 945                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
 946                 if (r)
 947                         goto out_free2;
 948                 r = 0;
 949 out_free2:
 950                 kfree(kvm_regs);
 951                 break;
 952         }
 953         case KVM_GET_SREGS: {
 954                 struct kvm_sregs kvm_sregs;
 955
 956                 memset(&kvm_sregs, 0, sizeof kvm_sregs);
 957                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, &kvm_sregs);
 958                 if (r)
 959                         goto out;
 960                 r = -EFAULT;
 961                 if (copy_to_user(argp, &kvm_sregs, sizeof kvm_sregs))
 962                         goto out;
 963                 r = 0;
 964                 break;
 965         }
 966         case KVM_SET_SREGS: {
 967                 struct kvm_sregs kvm_sregs;
 968
 969                 r = -EFAULT;
 970                 if (copy_from_user(&kvm_sregs, argp, sizeof kvm_sregs))
 971                         goto out;
 972                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, &kvm_sregs);
 973                 if (r)
 974                         goto out;
 975                 r = 0;
 976                 break;
 977         }
 978         case KVM_GET_MP_STATE: {
 979                 struct kvm_mp_state mp_state;
 980
 981                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
 982                 if (r)
 983                         goto out;
 984                 r = -EFAULT;
 985                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
 986                         goto out;
 987                 r = 0;
 988                 break;
 989         }
 990         case KVM_SET_MP_STATE: {
 991                 struct kvm_mp_state mp_state;
 992
 993                 r = -EFAULT;
 994                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
 995                         goto out;
 996                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
 997                 if (r)
 998                         goto out;
 999                 r = 0;
1000                 break;
1001         }
1002         case KVM_TRANSLATE: {
1003                 struct kvm_translation tr;
1004
1005                 r = -EFAULT;
1006                 if (copy_from_user(&tr, argp, sizeof tr))
1007                         goto out;
1008                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1009                 if (r)
1010                         goto out;
1011                 r = -EFAULT;
1012                 if (copy_to_user(argp, &tr, sizeof tr))
1013                         goto out;
1014                 r = 0;
1015                 break;
1016         }
1017         case KVM_DEBUG_GUEST: {
1018                 struct kvm_debug_guest dbg;
1019
1020                 r = -EFAULT;
1021                 if (copy_from_user(&dbg, argp, sizeof dbg))
1022                         goto out;
1023                 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1024                 if (r)
1025                         goto out;
1026                 r = 0;
1027                 break;
1028         }
1029         case KVM_SET_SIGNAL_MASK: {
1030                 struct kvm_signal_mask __user *sigmask_arg = argp;
1031                 struct kvm_signal_mask kvm_sigmask;
1032                 sigset_t sigset, *p;
1033
1034                 p = NULL;
1035                 if (argp) {
1036                         r = -EFAULT;
1037                         if (copy_from_user(&kvm_sigmask, argp,
1038                                            sizeof kvm_sigmask))
1039                                 goto out;
1040                         r = -EINVAL;
1041                         if (kvm_sigmask.len != sizeof sigset)
1042                                 goto out;
1043                         r = -EFAULT;
1044                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1045                                            sizeof sigset))
1046                                 goto out;
1047                         p = &sigset;
1048                 }
1049                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1050                 break;
1051         }
1052         case KVM_GET_FPU: {
1053                 struct kvm_fpu fpu;
1054
1055                 memset(&fpu, 0, sizeof fpu);
1056                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, &fpu);
1057                 if (r)
1058                         goto out;
1059                 r = -EFAULT;
1060                 if (copy_to_user(argp, &fpu, sizeof fpu))
1061                         goto out;
1062                 r = 0;
1063                 break;
1064         }
1065         case KVM_SET_FPU: {
1066                 struct kvm_fpu fpu;
1067
1068                 r = -EFAULT;
1069                 if (copy_from_user(&fpu, argp, sizeof fpu))
1070                         goto out;
1071                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, &fpu);
1072                 if (r)
1073                         goto out;
1074                 r = 0;
1075                 break;
1076         }
1077         default:
1078                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1079         }
1080 out:
1081         return r;
1082 }
1083
1084 static long kvm_vm_ioctl(struct file *filp,
1085                            unsigned int ioctl, unsigned long arg)
1086 {
1087         struct kvm *kvm = filp->private_data;
1088         void __user *argp = (void __user *)arg;
1089         int r;
1090
1091         if (kvm->mm != current->mm)
1092                 return -EIO;
1093         switch (ioctl) {
1094         case KVM_CREATE_VCPU:
1095                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1096                 if (r < 0)
1097                         goto out;
1098                 break;
1099         case KVM_SET_USER_MEMORY_REGION: {
1100                 struct kvm_userspace_memory_region kvm_userspace_mem;
1101
1102                 r = -EFAULT;
1103                 if (copy_from_user(&kvm_userspace_mem, argp,
1104                                                 sizeof kvm_userspace_mem))
1105                         goto out;
1106
1107                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1108                 if (r)
1109                         goto out;
1110                 break;
1111         }
1112         case KVM_GET_DIRTY_LOG: {
1113                 struct kvm_dirty_log log;
1114
1115                 r = -EFAULT;
1116                 if (copy_from_user(&log, argp, sizeof log))
1117                         goto out;
1118                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1119                 if (r)
1120                         goto out;
1121                 break;
1122         }
1123         default:
1124                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1125         }
1126 out:
1127         return r;
1128 }
1129
1130 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1131 {
1132         struct kvm *kvm = vma->vm_file->private_data;
1133         struct page *page;
1134
1135         if (!kvm_is_visible_gfn(kvm, vmf->pgoff))
1136                 return VM_FAULT_SIGBUS;
1137         page = gfn_to_page(kvm, vmf->pgoff);
1138         if (is_error_page(page)) {
1139                 kvm_release_page_clean(page);
1140                 return VM_FAULT_SIGBUS;
1141         }
1142         vmf->page = page;
1143         return 0;
1144 }
1145
1146 static struct vm_operations_struct kvm_vm_vm_ops = {
1147         .fault = kvm_vm_fault,
1148 };
1149
1150 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1151 {
1152         vma->vm_ops = &kvm_vm_vm_ops;
1153         return 0;
1154 }
1155
1156 static const struct file_operations kvm_vm_fops = {
1157         .release        = kvm_vm_release,
1158         .unlocked_ioctl = kvm_vm_ioctl,
1159         .compat_ioctl   = kvm_vm_ioctl,
1160         .mmap           = kvm_vm_mmap,
1161 };
1162
1163 static int kvm_dev_ioctl_create_vm(void)
1164 {
1165         int fd;
1166         struct kvm *kvm;
1167
1168         kvm = kvm_create_vm();
1169         if (IS_ERR(kvm))
1170                 return PTR_ERR(kvm);
1171         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm);
1172         if (fd < 0)
1173                 kvm_put_kvm(kvm);
1174
1175         return fd;
1176 }
1177
1178 static long kvm_dev_ioctl(struct file *filp,
1179                           unsigned int ioctl, unsigned long arg)
1180 {
1181         void __user *argp = (void __user *)arg;
1182         long r = -EINVAL;
1183
1184         switch (ioctl) {
1185         case KVM_GET_API_VERSION:
1186                 r = -EINVAL;
1187                 if (arg)
1188                         goto out;
1189                 r = KVM_API_VERSION;
1190                 break;
1191         case KVM_CREATE_VM:
1192                 r = -EINVAL;
1193                 if (arg)
1194                         goto out;
1195                 r = kvm_dev_ioctl_create_vm();
1196                 break;
1197         case KVM_CHECK_EXTENSION:
1198                 r = kvm_dev_ioctl_check_extension((long)argp);
1199                 break;
1200         case KVM_GET_VCPU_MMAP_SIZE:
1201                 r = -EINVAL;
1202                 if (arg)
1203                         goto out;
1204                 r = PAGE_SIZE;     /* struct kvm_run */
1205 #ifdef CONFIG_X86
1206                 r += PAGE_SIZE;    /* pio data page */
1207 #endif
1208                 break;
1209         case KVM_TRACE_ENABLE:
1210         case KVM_TRACE_PAUSE:
1211         case KVM_TRACE_DISABLE:
1212                 r = kvm_trace_ioctl(ioctl, arg);
1213                 break;
1214         default:
1215                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1216         }
1217 out:
1218         return r;
1219 }
1220
1221 static struct file_operations kvm_chardev_ops = {
1222         .unlocked_ioctl = kvm_dev_ioctl,
1223         .compat_ioctl   = kvm_dev_ioctl,
1224 };
1225
1226 static struct miscdevice kvm_dev = {
1227         KVM_MINOR,
1228         "kvm",
1229         &kvm_chardev_ops,
1230 };
1231
1232 static void hardware_enable(void *junk)
1233 {
1234         int cpu = raw_smp_processor_id();
1235
1236         if (cpu_isset(cpu, cpus_hardware_enabled))
1237                 return;
1238         cpu_set(cpu, cpus_hardware_enabled);
1239         kvm_arch_hardware_enable(NULL);
1240 }
1241
1242 static void hardware_disable(void *junk)
1243 {
1244         int cpu = raw_smp_processor_id();
1245
1246         if (!cpu_isset(cpu, cpus_hardware_enabled))
1247                 return;
1248         cpu_clear(cpu, cpus_hardware_enabled);
1249         decache_vcpus_on_cpu(cpu);
1250         kvm_arch_hardware_disable(NULL);
1251 }
1252
1253 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1254                            void *v)
1255 {
1256         int cpu = (long)v;
1257
1258         val &= ~CPU_TASKS_FROZEN;
1259         switch (val) {
1260         case CPU_DYING:
1261                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1262                        cpu);
1263                 hardware_disable(NULL);
1264                 break;
1265         case CPU_UP_CANCELED:
1266                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1267                        cpu);
1268                 smp_call_function_single(cpu, hardware_disable, NULL, 0, 1);
1269                 break;
1270         case CPU_ONLINE:
1271                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1272                        cpu);
1273                 smp_call_function_single(cpu, hardware_enable, NULL, 0, 1);
1274                 break;
1275         }
1276         return NOTIFY_OK;
1277 }
1278
1279 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1280                       void *v)
1281 {
1282         if (val == SYS_RESTART) {
1283                 /*
1284                  * Some (well, at least mine) BIOSes hang on reboot if
1285                  * in vmx root mode.
1286                  */
1287                 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1288                 on_each_cpu(hardware_disable, NULL, 0, 1);
1289         }
1290         return NOTIFY_OK;
1291 }
1292
1293 static struct notifier_block kvm_reboot_notifier = {
1294         .notifier_call = kvm_reboot,
1295         .priority = 0,
1296 };
1297
1298 void kvm_io_bus_init(struct kvm_io_bus *bus)
1299 {
1300         memset(bus, 0, sizeof(*bus));
1301 }
1302
1303 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1304 {
1305         int i;
1306
1307         for (i = 0; i < bus->dev_count; i++) {
1308                 struct kvm_io_device *pos = bus->devs[i];
1309
1310                 kvm_iodevice_destructor(pos);
1311         }
1312 }
1313
1314 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, gpa_t addr)
1315 {
1316         int i;
1317
1318         for (i = 0; i < bus->dev_count; i++) {
1319                 struct kvm_io_device *pos = bus->devs[i];
1320
1321                 if (pos->in_range(pos, addr))
1322                         return pos;
1323         }
1324
1325         return NULL;
1326 }
1327
1328 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1329 {
1330         BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1331
1332         bus->devs[bus->dev_count++] = dev;
1333 }
1334
1335 static struct notifier_block kvm_cpu_notifier = {
1336         .notifier_call = kvm_cpu_hotplug,
1337         .priority = 20, /* must be > scheduler priority */
1338 };
1339
1340 static int vm_stat_get(void *_offset, u64 *val)
1341 {
1342         unsigned offset = (long)_offset;
1343         struct kvm *kvm;
1344
1345         *val = 0;
1346         spin_lock(&kvm_lock);
1347         list_for_each_entry(kvm, &vm_list, vm_list)
1348                 *val += *(u32 *)((void *)kvm + offset);
1349         spin_unlock(&kvm_lock);
1350         return 0;
1351 }
1352
1353 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1354
1355 static int vcpu_stat_get(void *_offset, u64 *val)
1356 {
1357         unsigned offset = (long)_offset;
1358         struct kvm *kvm;
1359         struct kvm_vcpu *vcpu;
1360         int i;
1361
1362         *val = 0;
1363         spin_lock(&kvm_lock);
1364         list_for_each_entry(kvm, &vm_list, vm_list)
1365                 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1366                         vcpu = kvm->vcpus[i];
1367                         if (vcpu)
1368                                 *val += *(u32 *)((void *)vcpu + offset);
1369                 }
1370         spin_unlock(&kvm_lock);
1371         return 0;
1372 }
1373
1374 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1375
1376 static struct file_operations *stat_fops[] = {
1377         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1378         [KVM_STAT_VM]   = &vm_stat_fops,
1379 };
1380
1381 static void kvm_init_debug(void)
1382 {
1383         struct kvm_stats_debugfs_item *p;
1384
1385         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1386         for (p = debugfs_entries; p->name; ++p)
1387                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1388                                                 (void *)(long)p->offset,
1389                                                 stat_fops[p->kind]);
1390 }
1391
1392 static void kvm_exit_debug(void)
1393 {
1394         struct kvm_stats_debugfs_item *p;
1395
1396         for (p = debugfs_entries; p->name; ++p)
1397                 debugfs_remove(p->dentry);
1398         debugfs_remove(kvm_debugfs_dir);
1399 }
1400
1401 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1402 {
1403         hardware_disable(NULL);
1404         return 0;
1405 }
1406
1407 static int kvm_resume(struct sys_device *dev)
1408 {
1409         hardware_enable(NULL);
1410         return 0;
1411 }
1412
1413 static struct sysdev_class kvm_sysdev_class = {
1414         .name = "kvm",
1415         .suspend = kvm_suspend,
1416         .resume = kvm_resume,
1417 };
1418
1419 static struct sys_device kvm_sysdev = {
1420         .id = 0,
1421         .cls = &kvm_sysdev_class,
1422 };
1423
1424 struct page *bad_page;
1425 pfn_t bad_pfn;
1426
1427 static inline
1428 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1429 {
1430         return container_of(pn, struct kvm_vcpu, preempt_notifier);
1431 }
1432
1433 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1434 {
1435         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1436
1437         kvm_arch_vcpu_load(vcpu, cpu);
1438 }
1439
1440 static void kvm_sched_out(struct preempt_notifier *pn,
1441                           struct task_struct *next)
1442 {
1443         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1444
1445         kvm_arch_vcpu_put(vcpu);
1446 }
1447
1448 int kvm_init(void *opaque, unsigned int vcpu_size,
1449                   struct module *module)
1450 {
1451         int r;
1452         int cpu;
1453
1454         kvm_init_debug();
1455
1456         r = kvm_arch_init(opaque);
1457         if (r)
1458                 goto out_fail;
1459
1460         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
1461
1462         if (bad_page == NULL) {
1463                 r = -ENOMEM;
1464                 goto out;
1465         }
1466
1467         bad_pfn = page_to_pfn(bad_page);
1468
1469         r = kvm_arch_hardware_setup();
1470         if (r < 0)
1471                 goto out_free_0;
1472
1473         for_each_online_cpu(cpu) {
1474                 smp_call_function_single(cpu,
1475                                 kvm_arch_check_processor_compat,
1476                                 &r, 0, 1);
1477                 if (r < 0)
1478                         goto out_free_1;
1479         }
1480
1481         on_each_cpu(hardware_enable, NULL, 0, 1);
1482         r = register_cpu_notifier(&kvm_cpu_notifier);
1483         if (r)
1484                 goto out_free_2;
1485         register_reboot_notifier(&kvm_reboot_notifier);
1486
1487         r = sysdev_class_register(&kvm_sysdev_class);
1488         if (r)
1489                 goto out_free_3;
1490
1491         r = sysdev_register(&kvm_sysdev);
1492         if (r)
1493                 goto out_free_4;
1494
1495         /* A kmem cache lets us meet the alignment requirements of fx_save. */
1496         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
1497                                            __alignof__(struct kvm_vcpu),
1498                                            0, NULL);
1499         if (!kvm_vcpu_cache) {
1500                 r = -ENOMEM;
1501                 goto out_free_5;
1502         }
1503
1504         kvm_chardev_ops.owner = module;
1505
1506         r = misc_register(&kvm_dev);
1507         if (r) {
1508                 printk(KERN_ERR "kvm: misc device register failed\n");
1509                 goto out_free;
1510         }
1511
1512         kvm_preempt_ops.sched_in = kvm_sched_in;
1513         kvm_preempt_ops.sched_out = kvm_sched_out;
1514
1515         return 0;
1516
1517 out_free:
1518         kmem_cache_destroy(kvm_vcpu_cache);
1519 out_free_5:
1520         sysdev_unregister(&kvm_sysdev);
1521 out_free_4:
1522         sysdev_class_unregister(&kvm_sysdev_class);
1523 out_free_3:
1524         unregister_reboot_notifier(&kvm_reboot_notifier);
1525         unregister_cpu_notifier(&kvm_cpu_notifier);
1526 out_free_2:
1527         on_each_cpu(hardware_disable, NULL, 0, 1);
1528 out_free_1:
1529         kvm_arch_hardware_unsetup();
1530 out_free_0:
1531         __free_page(bad_page);
1532 out:
1533         kvm_arch_exit();
1534         kvm_exit_debug();
1535 out_fail:
1536         return r;
1537 }
1538 EXPORT_SYMBOL_GPL(kvm_init);
1539
1540 void kvm_exit(void)
1541 {
1542         kvm_trace_cleanup();
1543         misc_deregister(&kvm_dev);
1544         kmem_cache_destroy(kvm_vcpu_cache);
1545         sysdev_unregister(&kvm_sysdev);
1546         sysdev_class_unregister(&kvm_sysdev_class);
1547         unregister_reboot_notifier(&kvm_reboot_notifier);
1548         unregister_cpu_notifier(&kvm_cpu_notifier);
1549         on_each_cpu(hardware_disable, NULL, 0, 1);
1550         kvm_arch_hardware_unsetup();
1551         kvm_arch_exit();
1552         kvm_exit_debug();
1553         __free_page(bad_page);
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_exit);