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 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47
  48 #include <asm/processor.h>
  49 #include <asm/io.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/pgtable.h>
  52 #include <asm-generic/bitops/le.h>
  53
  54 #include "coalesced_mmio.h"
  55
  56 #define CREATE_TRACE_POINTS
  57 #include <trace/events/kvm.h>
  58
  59 MODULE_AUTHOR("Qumranet");
  60 MODULE_LICENSE("GPL");
  61
  62 /*
  63  * Ordering of locks:
  64  *
  65  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  66  */
  67
  68 DEFINE_SPINLOCK(kvm_lock);
  69 LIST_HEAD(vm_list);
  70
  71 static cpumask_var_t cpus_hardware_enabled;
  72 static int kvm_usage_count = 0;
  73 static atomic_t hardware_enable_failed;
  74
  75 struct kmem_cache *kvm_vcpu_cache;
  76 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  77
  78 static __read_mostly struct preempt_ops kvm_preempt_ops;
  79
  80 struct dentry *kvm_debugfs_dir;
  81
  82 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  83                            unsigned long arg);
  84 static int hardware_enable_all(void);
  85 static void hardware_disable_all(void);
  86
  87 static bool kvm_rebooting;
  88
  89 static bool largepages_enabled = true;
  90
  91 inline int kvm_is_mmio_pfn(pfn_t pfn)
  92 {
  93         if (pfn_valid(pfn)) {
  94                 struct page *page = compound_head(pfn_to_page(pfn));
  95                 return PageReserved(page);
  96         }
  97
  98         return true;
  99 }
 100
 101 /*
 102  * Switches to specified vcpu, until a matching vcpu_put()
 103  */
 104 void vcpu_load(struct kvm_vcpu *vcpu)
 105 {
 106         int cpu;
 107
 108         mutex_lock(&vcpu->mutex);
 109         cpu = get_cpu();
 110         preempt_notifier_register(&vcpu->preempt_notifier);
 111         kvm_arch_vcpu_load(vcpu, cpu);
 112         put_cpu();
 113 }
 114
 115 void vcpu_put(struct kvm_vcpu *vcpu)
 116 {
 117         preempt_disable();
 118         kvm_arch_vcpu_put(vcpu);
 119         preempt_notifier_unregister(&vcpu->preempt_notifier);
 120         preempt_enable();
 121         mutex_unlock(&vcpu->mutex);
 122 }
 123
 124 static void ack_flush(void *_completed)
 125 {
 126 }
 127
 128 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 129 {
 130         int i, cpu, me;
 131         cpumask_var_t cpus;
 132         bool called = true;
 133         struct kvm_vcpu *vcpu;
 134
 135         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 136
 137         spin_lock(&kvm->requests_lock);
 138         me = smp_processor_id();
 139         kvm_for_each_vcpu(i, vcpu, kvm) {
 140                 if (test_and_set_bit(req, &vcpu->requests))
 141                         continue;
 142                 cpu = vcpu->cpu;
 143                 if (cpus != NULL && cpu != -1 && cpu != me)
 144                         cpumask_set_cpu(cpu, cpus);
 145         }
 146         if (unlikely(cpus == NULL))
 147                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 148         else if (!cpumask_empty(cpus))
 149                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 150         else
 151                 called = false;
 152         spin_unlock(&kvm->requests_lock);
 153         free_cpumask_var(cpus);
 154         return called;
 155 }
 156
 157 void kvm_flush_remote_tlbs(struct kvm *kvm)
 158 {
 159         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 160                 ++kvm->stat.remote_tlb_flush;
 161 }
 162
 163 void kvm_reload_remote_mmus(struct kvm *kvm)
 164 {
 165         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 166 }
 167
 168 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 169 {
 170         struct page *page;
 171         int r;
 172
 173         mutex_init(&vcpu->mutex);
 174         vcpu->cpu = -1;
 175         vcpu->kvm = kvm;
 176         vcpu->vcpu_id = id;
 177         init_waitqueue_head(&vcpu->wq);
 178
 179         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 180         if (!page) {
 181                 r = -ENOMEM;
 182                 goto fail;
 183         }
 184         vcpu->run = page_address(page);
 185
 186         r = kvm_arch_vcpu_init(vcpu);
 187         if (r < 0)
 188                 goto fail_free_run;
 189         return 0;
 190
 191 fail_free_run:
 192         free_page((unsigned long)vcpu->run);
 193 fail:
 194         return r;
 195 }
 196 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 197
 198 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 199 {
 200         kvm_arch_vcpu_uninit(vcpu);
 201         free_page((unsigned long)vcpu->run);
 202 }
 203 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 204
 205 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 206 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 207 {
 208         return container_of(mn, struct kvm, mmu_notifier);
 209 }
 210
 211 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 212                                              struct mm_struct *mm,
 213                                              unsigned long address)
 214 {
 215         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 216         int need_tlb_flush;
 217
 218         /*
 219          * When ->invalidate_page runs, the linux pte has been zapped
 220          * already but the page is still allocated until
 221          * ->invalidate_page returns. So if we increase the sequence
 222          * here the kvm page fault will notice if the spte can't be
 223          * established because the page is going to be freed. If
 224          * instead the kvm page fault establishes the spte before
 225          * ->invalidate_page runs, kvm_unmap_hva will release it
 226          * before returning.
 227          *
 228          * The sequence increase only need to be seen at spin_unlock
 229          * time, and not at spin_lock time.
 230          *
 231          * Increasing the sequence after the spin_unlock would be
 232          * unsafe because the kvm page fault could then establish the
 233          * pte after kvm_unmap_hva returned, without noticing the page
 234          * is going to be freed.
 235          */
 236         spin_lock(&kvm->mmu_lock);
 237         kvm->mmu_notifier_seq++;
 238         need_tlb_flush = kvm_unmap_hva(kvm, address);
 239         spin_unlock(&kvm->mmu_lock);
 240
 241         /* we've to flush the tlb before the pages can be freed */
 242         if (need_tlb_flush)
 243                 kvm_flush_remote_tlbs(kvm);
 244
 245 }
 246
 247 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 248                                         struct mm_struct *mm,
 249                                         unsigned long address,
 250                                         pte_t pte)
 251 {
 252         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 253
 254         spin_lock(&kvm->mmu_lock);
 255         kvm->mmu_notifier_seq++;
 256         kvm_set_spte_hva(kvm, address, pte);
 257         spin_unlock(&kvm->mmu_lock);
 258 }
 259
 260 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 261                                                     struct mm_struct *mm,
 262                                                     unsigned long start,
 263                                                     unsigned long end)
 264 {
 265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 266         int need_tlb_flush = 0;
 267
 268         spin_lock(&kvm->mmu_lock);
 269         /*
 270          * The count increase must become visible at unlock time as no
 271          * spte can be established without taking the mmu_lock and
 272          * count is also read inside the mmu_lock critical section.
 273          */
 274         kvm->mmu_notifier_count++;
 275         for (; start < end; start += PAGE_SIZE)
 276                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 277         spin_unlock(&kvm->mmu_lock);
 278
 279         /* we've to flush the tlb before the pages can be freed */
 280         if (need_tlb_flush)
 281                 kvm_flush_remote_tlbs(kvm);
 282 }
 283
 284 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 285                                                   struct mm_struct *mm,
 286                                                   unsigned long start,
 287                                                   unsigned long end)
 288 {
 289         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 290
 291         spin_lock(&kvm->mmu_lock);
 292         /*
 293          * This sequence increase will notify the kvm page fault that
 294          * the page that is going to be mapped in the spte could have
 295          * been freed.
 296          */
 297         kvm->mmu_notifier_seq++;
 298         /*
 299          * The above sequence increase must be visible before the
 300          * below count decrease but both values are read by the kvm
 301          * page fault under mmu_lock spinlock so we don't need to add
 302          * a smb_wmb() here in between the two.
 303          */
 304         kvm->mmu_notifier_count--;
 305         spin_unlock(&kvm->mmu_lock);
 306
 307         BUG_ON(kvm->mmu_notifier_count < 0);
 308 }
 309
 310 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 311                                               struct mm_struct *mm,
 312                                               unsigned long address)
 313 {
 314         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 315         int young;
 316
 317         spin_lock(&kvm->mmu_lock);
 318         young = kvm_age_hva(kvm, address);
 319         spin_unlock(&kvm->mmu_lock);
 320
 321         if (young)
 322                 kvm_flush_remote_tlbs(kvm);
 323
 324         return young;
 325 }
 326
 327 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 328                                      struct mm_struct *mm)
 329 {
 330         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 331         kvm_arch_flush_shadow(kvm);
 332 }
 333
 334 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 335         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 336         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 337         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 338         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 339         .change_pte             = kvm_mmu_notifier_change_pte,
 340         .release                = kvm_mmu_notifier_release,
 341 };
 342
 343 static int kvm_init_mmu_notifier(struct kvm *kvm)
 344 {
 345         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 346         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 347 }
 348
 349 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 350
 351 static int kvm_init_mmu_notifier(struct kvm *kvm)
 352 {
 353         return 0;
 354 }
 355
 356 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 357
 358 static struct kvm *kvm_create_vm(void)
 359 {
 360         int r = 0;
 361         struct kvm *kvm = kvm_arch_create_vm();
 362 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 363         struct page *page;
 364 #endif
 365
 366         if (IS_ERR(kvm))
 367                 goto out;
 368
 369         r = hardware_enable_all();
 370         if (r)
 371                 goto out_err_nodisable;
 372
 373 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 374         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 375         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 376 #endif
 377
 378         r = -ENOMEM;
 379         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 380         if (!kvm->memslots)
 381                 goto out_err;
 382
 383 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 384         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 385         if (!page)
 386                 goto out_err;
 387
 388         kvm->coalesced_mmio_ring =
 389                         (struct kvm_coalesced_mmio_ring *)page_address(page);
 390 #endif
 391
 392         r = kvm_init_mmu_notifier(kvm);
 393         if (r) {
 394 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 395                 put_page(page);
 396 #endif
 397                 goto out_err;
 398         }
 399
 400         kvm->mm = current->mm;
 401         atomic_inc(&kvm->mm->mm_count);
 402         spin_lock_init(&kvm->mmu_lock);
 403         spin_lock_init(&kvm->requests_lock);
 404         kvm_io_bus_init(&kvm->pio_bus);
 405         kvm_eventfd_init(kvm);
 406         mutex_init(&kvm->lock);
 407         mutex_init(&kvm->irq_lock);
 408         kvm_io_bus_init(&kvm->mmio_bus);
 409         init_rwsem(&kvm->slots_lock);
 410         atomic_set(&kvm->users_count, 1);
 411         spin_lock(&kvm_lock);
 412         list_add(&kvm->vm_list, &vm_list);
 413         spin_unlock(&kvm_lock);
 414 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 415         kvm_coalesced_mmio_init(kvm);
 416 #endif
 417 out:
 418         return kvm;
 419
 420 out_err:
 421         hardware_disable_all();
 422 out_err_nodisable:
 423         kfree(kvm->memslots);
 424         kfree(kvm);
 425         return ERR_PTR(r);
 426 }
 427
 428 /*
 429  * Free any memory in @free but not in @dont.
 430  */
 431 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 432                                   struct kvm_memory_slot *dont)
 433 {
 434         int i;
 435
 436         if (!dont || free->rmap != dont->rmap)
 437                 vfree(free->rmap);
 438
 439         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 440                 vfree(free->dirty_bitmap);
 441
 442
 443         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 444                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 445                         vfree(free->lpage_info[i]);
 446                         free->lpage_info[i] = NULL;
 447                 }
 448         }
 449
 450         free->npages = 0;
 451         free->dirty_bitmap = NULL;
 452         free->rmap = NULL;
 453 }
 454
 455 void kvm_free_physmem(struct kvm *kvm)
 456 {
 457         int i;
 458         struct kvm_memslots *slots = kvm->memslots;
 459
 460         for (i = 0; i < slots->nmemslots; ++i)
 461                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 462
 463         kfree(kvm->memslots);
 464 }
 465
 466 static void kvm_destroy_vm(struct kvm *kvm)
 467 {
 468         struct mm_struct *mm = kvm->mm;
 469
 470         kvm_arch_sync_events(kvm);
 471         spin_lock(&kvm_lock);
 472         list_del(&kvm->vm_list);
 473         spin_unlock(&kvm_lock);
 474         kvm_free_irq_routing(kvm);
 475         kvm_io_bus_destroy(&kvm->pio_bus);
 476         kvm_io_bus_destroy(&kvm->mmio_bus);
 477         kvm_coalesced_mmio_free(kvm);
 478 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 479         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 480 #else
 481         kvm_arch_flush_shadow(kvm);
 482 #endif
 483         kvm_arch_destroy_vm(kvm);
 484         hardware_disable_all();
 485         mmdrop(mm);
 486 }
 487
 488 void kvm_get_kvm(struct kvm *kvm)
 489 {
 490         atomic_inc(&kvm->users_count);
 491 }
 492 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 493
 494 void kvm_put_kvm(struct kvm *kvm)
 495 {
 496         if (atomic_dec_and_test(&kvm->users_count))
 497                 kvm_destroy_vm(kvm);
 498 }
 499 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 500
 501
 502 static int kvm_vm_release(struct inode *inode, struct file *filp)
 503 {
 504         struct kvm *kvm = filp->private_data;
 505
 506         kvm_irqfd_release(kvm);
 507
 508         kvm_put_kvm(kvm);
 509         return 0;
 510 }
 511
 512 /*
 513  * Allocate some memory and give it an address in the guest physical address
 514  * space.
 515  *
 516  * Discontiguous memory is allowed, mostly for framebuffers.
 517  *
 518  * Must be called holding mmap_sem for write.
 519  */
 520 int __kvm_set_memory_region(struct kvm *kvm,
 521                             struct kvm_userspace_memory_region *mem,
 522                             int user_alloc)
 523 {
 524         int r;
 525         gfn_t base_gfn;
 526         unsigned long npages;
 527         unsigned long i;
 528         struct kvm_memory_slot *memslot;
 529         struct kvm_memory_slot old, new;
 530
 531         r = -EINVAL;
 532         /* General sanity checks */
 533         if (mem->memory_size & (PAGE_SIZE - 1))
 534                 goto out;
 535         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 536                 goto out;
 537         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 538                 goto out;
 539         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 540                 goto out;
 541         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 542                 goto out;
 543
 544         memslot = &kvm->memslots->memslots[mem->slot];
 545         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 546         npages = mem->memory_size >> PAGE_SHIFT;
 547
 548         if (!npages)
 549                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 550
 551         new = old = *memslot;
 552
 553         new.base_gfn = base_gfn;
 554         new.npages = npages;
 555         new.flags = mem->flags;
 556
 557         /* Disallow changing a memory slot's size. */
 558         r = -EINVAL;
 559         if (npages && old.npages && npages != old.npages)
 560                 goto out_free;
 561
 562         /* Check for overlaps */
 563         r = -EEXIST;
 564         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 565                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 566
 567                 if (s == memslot || !s->npages)
 568                         continue;
 569                 if (!((base_gfn + npages <= s->base_gfn) ||
 570                       (base_gfn >= s->base_gfn + s->npages)))
 571                         goto out_free;
 572         }
 573
 574         /* Free page dirty bitmap if unneeded */
 575         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 576                 new.dirty_bitmap = NULL;
 577
 578         r = -ENOMEM;
 579
 580         /* Allocate if a slot is being created */
 581 #ifndef CONFIG_S390
 582         if (npages && !new.rmap) {
 583                 new.rmap = vmalloc(npages * sizeof(struct page *));
 584
 585                 if (!new.rmap)
 586                         goto out_free;
 587
 588                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 589
 590                 new.user_alloc = user_alloc;
 591                 /*
 592                  * hva_to_rmmap() serialzies with the mmu_lock and to be
 593                  * safe it has to ignore memslots with !user_alloc &&
 594                  * !userspace_addr.
 595                  */
 596                 if (user_alloc)
 597                         new.userspace_addr = mem->userspace_addr;
 598                 else
 599                         new.userspace_addr = 0;
 600         }
 601         if (!npages)
 602                 goto skip_lpage;
 603
 604         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 605                 unsigned long ugfn;
 606                 unsigned long j;
 607                 int lpages;
 608                 int level = i + 2;
 609
 610                 /* Avoid unused variable warning if no large pages */
 611                 (void)level;
 612
 613                 if (new.lpage_info[i])
 614                         continue;
 615
 616                 lpages = 1 + (base_gfn + npages - 1) /
 617                              KVM_PAGES_PER_HPAGE(level);
 618                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 619
 620                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 621
 622                 if (!new.lpage_info[i])
 623                         goto out_free;
 624
 625                 memset(new.lpage_info[i], 0,
 626                        lpages * sizeof(*new.lpage_info[i]));
 627
 628                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 629                         new.lpage_info[i][0].write_count = 1;
 630                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 631                         new.lpage_info[i][lpages - 1].write_count = 1;
 632                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 633                 /*
 634                  * If the gfn and userspace address are not aligned wrt each
 635                  * other, or if explicitly asked to, disable large page
 636                  * support for this slot
 637                  */
 638                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 639                     !largepages_enabled)
 640                         for (j = 0; j < lpages; ++j)
 641                                 new.lpage_info[i][j].write_count = 1;
 642         }
 643
 644 skip_lpage:
 645
 646         /* Allocate page dirty bitmap if needed */
 647         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 648                 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
 649
 650                 new.dirty_bitmap = vmalloc(dirty_bytes);
 651                 if (!new.dirty_bitmap)
 652                         goto out_free;
 653                 memset(new.dirty_bitmap, 0, dirty_bytes);
 654                 if (old.npages)
 655                         kvm_arch_flush_shadow(kvm);
 656         }
 657 #else  /* not defined CONFIG_S390 */
 658         new.user_alloc = user_alloc;
 659         if (user_alloc)
 660                 new.userspace_addr = mem->userspace_addr;
 661 #endif /* not defined CONFIG_S390 */
 662
 663         if (!npages)
 664                 kvm_arch_flush_shadow(kvm);
 665
 666         spin_lock(&kvm->mmu_lock);
 667         if (mem->slot >= kvm->memslots->nmemslots)
 668                 kvm->memslots->nmemslots = mem->slot + 1;
 669
 670         *memslot = new;
 671         spin_unlock(&kvm->mmu_lock);
 672
 673         r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
 674         if (r) {
 675                 spin_lock(&kvm->mmu_lock);
 676                 *memslot = old;
 677                 spin_unlock(&kvm->mmu_lock);
 678                 goto out_free;
 679         }
 680
 681         kvm_free_physmem_slot(&old, npages ? &new : NULL);
 682         /* Slot deletion case: we have to update the current slot */
 683         spin_lock(&kvm->mmu_lock);
 684         if (!npages)
 685                 *memslot = old;
 686         spin_unlock(&kvm->mmu_lock);
 687 #ifdef CONFIG_DMAR
 688         /* map the pages in iommu page table */
 689         r = kvm_iommu_map_pages(kvm, base_gfn, npages);
 690         if (r)
 691                 goto out;
 692 #endif
 693         return 0;
 694
 695 out_free:
 696         kvm_free_physmem_slot(&new, &old);
 697 out:
 698         return r;
 699
 700 }
 701 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 702
 703 int kvm_set_memory_region(struct kvm *kvm,
 704                           struct kvm_userspace_memory_region *mem,
 705                           int user_alloc)
 706 {
 707         int r;
 708
 709         down_write(&kvm->slots_lock);
 710         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 711         up_write(&kvm->slots_lock);
 712         return r;
 713 }
 714 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 715
 716 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 717                                    struct
 718                                    kvm_userspace_memory_region *mem,
 719                                    int user_alloc)
 720 {
 721         if (mem->slot >= KVM_MEMORY_SLOTS)
 722                 return -EINVAL;
 723         return kvm_set_memory_region(kvm, mem, user_alloc);
 724 }
 725
 726 int kvm_get_dirty_log(struct kvm *kvm,
 727                         struct kvm_dirty_log *log, int *is_dirty)
 728 {
 729         struct kvm_memory_slot *memslot;
 730         int r, i;
 731         int n;
 732         unsigned long any = 0;
 733
 734         r = -EINVAL;
 735         if (log->slot >= KVM_MEMORY_SLOTS)
 736                 goto out;
 737
 738         memslot = &kvm->memslots->memslots[log->slot];
 739         r = -ENOENT;
 740         if (!memslot->dirty_bitmap)
 741                 goto out;
 742
 743         n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
 744
 745         for (i = 0; !any && i < n/sizeof(long); ++i)
 746                 any = memslot->dirty_bitmap[i];
 747
 748         r = -EFAULT;
 749         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 750                 goto out;
 751
 752         if (any)
 753                 *is_dirty = 1;
 754
 755         r = 0;
 756 out:
 757         return r;
 758 }
 759
 760 void kvm_disable_largepages(void)
 761 {
 762         largepages_enabled = false;
 763 }
 764 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 765
 766 int is_error_page(struct page *page)
 767 {
 768         return page == bad_page;
 769 }
 770 EXPORT_SYMBOL_GPL(is_error_page);
 771
 772 int is_error_pfn(pfn_t pfn)
 773 {
 774         return pfn == bad_pfn;
 775 }
 776 EXPORT_SYMBOL_GPL(is_error_pfn);
 777
 778 static inline unsigned long bad_hva(void)
 779 {
 780         return PAGE_OFFSET;
 781 }
 782
 783 int kvm_is_error_hva(unsigned long addr)
 784 {
 785         return addr == bad_hva();
 786 }
 787 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 788
 789 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 790 {
 791         int i;
 792         struct kvm_memslots *slots = kvm->memslots;
 793
 794         for (i = 0; i < slots->nmemslots; ++i) {
 795                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 796
 797                 if (gfn >= memslot->base_gfn
 798                     && gfn < memslot->base_gfn + memslot->npages)
 799                         return memslot;
 800         }
 801         return NULL;
 802 }
 803 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 804
 805 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 806 {
 807         gfn = unalias_gfn(kvm, gfn);
 808         return gfn_to_memslot_unaliased(kvm, gfn);
 809 }
 810
 811 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 812 {
 813         int i;
 814         struct kvm_memslots *slots = kvm->memslots;
 815
 816         gfn = unalias_gfn(kvm, gfn);
 817         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 818                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 819
 820                 if (gfn >= memslot->base_gfn
 821                     && gfn < memslot->base_gfn + memslot->npages)
 822                         return 1;
 823         }
 824         return 0;
 825 }
 826 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 827
 828 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 829 {
 830         struct kvm_memory_slot *slot;
 831
 832         gfn = unalias_gfn(kvm, gfn);
 833         slot = gfn_to_memslot_unaliased(kvm, gfn);
 834         if (!slot)
 835                 return bad_hva();
 836         return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
 837 }
 838 EXPORT_SYMBOL_GPL(gfn_to_hva);
 839
 840 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 841 {
 842         struct page *page[1];
 843         unsigned long addr;
 844         int npages;
 845         pfn_t pfn;
 846
 847         might_sleep();
 848
 849         addr = gfn_to_hva(kvm, gfn);
 850         if (kvm_is_error_hva(addr)) {
 851                 get_page(bad_page);
 852                 return page_to_pfn(bad_page);
 853         }
 854
 855         npages = get_user_pages_fast(addr, 1, 1, page);
 856
 857         if (unlikely(npages != 1)) {
 858                 struct vm_area_struct *vma;
 859
 860                 down_read(&current->mm->mmap_sem);
 861                 vma = find_vma(current->mm, addr);
 862
 863                 if (vma == NULL || addr < vma->vm_start ||
 864                     !(vma->vm_flags & VM_PFNMAP)) {
 865                         up_read(&current->mm->mmap_sem);
 866                         get_page(bad_page);
 867                         return page_to_pfn(bad_page);
 868                 }
 869
 870                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 871                 up_read(&current->mm->mmap_sem);
 872                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 873         } else
 874                 pfn = page_to_pfn(page[0]);
 875
 876         return pfn;
 877 }
 878
 879 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 880
 881 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 882 {
 883         pfn_t pfn;
 884
 885         pfn = gfn_to_pfn(kvm, gfn);
 886         if (!kvm_is_mmio_pfn(pfn))
 887                 return pfn_to_page(pfn);
 888
 889         WARN_ON(kvm_is_mmio_pfn(pfn));
 890
 891         get_page(bad_page);
 892         return bad_page;
 893 }
 894
 895 EXPORT_SYMBOL_GPL(gfn_to_page);
 896
 897 void kvm_release_page_clean(struct page *page)
 898 {
 899         kvm_release_pfn_clean(page_to_pfn(page));
 900 }
 901 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
 902
 903 void kvm_release_pfn_clean(pfn_t pfn)
 904 {
 905         if (!kvm_is_mmio_pfn(pfn))
 906                 put_page(pfn_to_page(pfn));
 907 }
 908 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
 909
 910 void kvm_release_page_dirty(struct page *page)
 911 {
 912         kvm_release_pfn_dirty(page_to_pfn(page));
 913 }
 914 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
 915
 916 void kvm_release_pfn_dirty(pfn_t pfn)
 917 {
 918         kvm_set_pfn_dirty(pfn);
 919         kvm_release_pfn_clean(pfn);
 920 }
 921 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
 922
 923 void kvm_set_page_dirty(struct page *page)
 924 {
 925         kvm_set_pfn_dirty(page_to_pfn(page));
 926 }
 927 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
 928
 929 void kvm_set_pfn_dirty(pfn_t pfn)
 930 {
 931         if (!kvm_is_mmio_pfn(pfn)) {
 932                 struct page *page = pfn_to_page(pfn);
 933                 if (!PageReserved(page))
 934                         SetPageDirty(page);
 935         }
 936 }
 937 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
 938
 939 void kvm_set_pfn_accessed(pfn_t pfn)
 940 {
 941         if (!kvm_is_mmio_pfn(pfn))
 942                 mark_page_accessed(pfn_to_page(pfn));
 943 }
 944 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
 945
 946 void kvm_get_pfn(pfn_t pfn)
 947 {
 948         if (!kvm_is_mmio_pfn(pfn))
 949                 get_page(pfn_to_page(pfn));
 950 }
 951 EXPORT_SYMBOL_GPL(kvm_get_pfn);
 952
 953 static int next_segment(unsigned long len, int offset)
 954 {
 955         if (len > PAGE_SIZE - offset)
 956                 return PAGE_SIZE - offset;
 957         else
 958                 return len;
 959 }
 960
 961 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
 962                         int len)
 963 {
 964         int r;
 965         unsigned long addr;
 966
 967         addr = gfn_to_hva(kvm, gfn);
 968         if (kvm_is_error_hva(addr))
 969                 return -EFAULT;
 970         r = copy_from_user(data, (void __user *)addr + offset, len);
 971         if (r)
 972                 return -EFAULT;
 973         return 0;
 974 }
 975 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
 976
 977 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 978 {
 979         gfn_t gfn = gpa >> PAGE_SHIFT;
 980         int seg;
 981         int offset = offset_in_page(gpa);
 982         int ret;
 983
 984         while ((seg = next_segment(len, offset)) != 0) {
 985                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
 986                 if (ret < 0)
 987                         return ret;
 988                 offset = 0;
 989                 len -= seg;
 990                 data += seg;
 991                 ++gfn;
 992         }
 993         return 0;
 994 }
 995 EXPORT_SYMBOL_GPL(kvm_read_guest);
 996
 997 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
 998                           unsigned long len)
 999 {
1000         int r;
1001         unsigned long addr;
1002         gfn_t gfn = gpa >> PAGE_SHIFT;
1003         int offset = offset_in_page(gpa);
1004
1005         addr = gfn_to_hva(kvm, gfn);
1006         if (kvm_is_error_hva(addr))
1007                 return -EFAULT;
1008         pagefault_disable();
1009         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1010         pagefault_enable();
1011         if (r)
1012                 return -EFAULT;
1013         return 0;
1014 }
1015 EXPORT_SYMBOL(kvm_read_guest_atomic);
1016
1017 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1018                          int offset, int len)
1019 {
1020         int r;
1021         unsigned long addr;
1022
1023         addr = gfn_to_hva(kvm, gfn);
1024         if (kvm_is_error_hva(addr))
1025                 return -EFAULT;
1026         r = copy_to_user((void __user *)addr + offset, data, len);
1027         if (r)
1028                 return -EFAULT;
1029         mark_page_dirty(kvm, gfn);
1030         return 0;
1031 }
1032 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1033
1034 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1035                     unsigned long len)
1036 {
1037         gfn_t gfn = gpa >> PAGE_SHIFT;
1038         int seg;
1039         int offset = offset_in_page(gpa);
1040         int ret;
1041
1042         while ((seg = next_segment(len, offset)) != 0) {
1043                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1044                 if (ret < 0)
1045                         return ret;
1046                 offset = 0;
1047                 len -= seg;
1048                 data += seg;
1049                 ++gfn;
1050         }
1051         return 0;
1052 }
1053
1054 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1055 {
1056         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1059
1060 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1061 {
1062         gfn_t gfn = gpa >> PAGE_SHIFT;
1063         int seg;
1064         int offset = offset_in_page(gpa);
1065         int ret;
1066
1067         while ((seg = next_segment(len, offset)) != 0) {
1068                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1069                 if (ret < 0)
1070                         return ret;
1071                 offset = 0;
1072                 len -= seg;
1073                 ++gfn;
1074         }
1075         return 0;
1076 }
1077 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1078
1079 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1080 {
1081         struct kvm_memory_slot *memslot;
1082
1083         gfn = unalias_gfn(kvm, gfn);
1084         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1085         if (memslot && memslot->dirty_bitmap) {
1086                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1087
1088                 /* avoid RMW */
1089                 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1090                         generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1091         }
1092 }
1093
1094 /*
1095  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1096  */
1097 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1098 {
1099         DEFINE_WAIT(wait);
1100
1101         for (;;) {
1102                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1103
1104                 if (kvm_arch_vcpu_runnable(vcpu)) {
1105                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1106                         break;
1107                 }
1108                 if (kvm_cpu_has_pending_timer(vcpu))
1109                         break;
1110                 if (signal_pending(current))
1111                         break;
1112
1113                 schedule();
1114         }
1115
1116         finish_wait(&vcpu->wq, &wait);
1117 }
1118
1119 void kvm_resched(struct kvm_vcpu *vcpu)
1120 {
1121         if (!need_resched())
1122                 return;
1123         cond_resched();
1124 }
1125 EXPORT_SYMBOL_GPL(kvm_resched);
1126
1127 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1128 {
1129         ktime_t expires;
1130         DEFINE_WAIT(wait);
1131
1132         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1133
1134         /* Sleep for 100 us, and hope lock-holder got scheduled */
1135         expires = ktime_add_ns(ktime_get(), 100000UL);
1136         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1137
1138         finish_wait(&vcpu->wq, &wait);
1139 }
1140 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1141
1142 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1143 {
1144         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1145         struct page *page;
1146
1147         if (vmf->pgoff == 0)
1148                 page = virt_to_page(vcpu->run);
1149 #ifdef CONFIG_X86
1150         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1151                 page = virt_to_page(vcpu->arch.pio_data);
1152 #endif
1153 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1154         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1155                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1156 #endif
1157         else
1158                 return VM_FAULT_SIGBUS;
1159         get_page(page);
1160         vmf->page = page;
1161         return 0;
1162 }
1163
1164 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1165         .fault = kvm_vcpu_fault,
1166 };
1167
1168 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1169 {
1170         vma->vm_ops = &kvm_vcpu_vm_ops;
1171         return 0;
1172 }
1173
1174 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1175 {
1176         struct kvm_vcpu *vcpu = filp->private_data;
1177
1178         kvm_put_kvm(vcpu->kvm);
1179         return 0;
1180 }
1181
1182 static struct file_operations kvm_vcpu_fops = {
1183         .release        = kvm_vcpu_release,
1184         .unlocked_ioctl = kvm_vcpu_ioctl,
1185         .compat_ioctl   = kvm_vcpu_ioctl,
1186         .mmap           = kvm_vcpu_mmap,
1187 };
1188
1189 /*
1190  * Allocates an inode for the vcpu.
1191  */
1192 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1193 {
1194         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1195 }
1196
1197 /*
1198  * Creates some virtual cpus.  Good luck creating more than one.
1199  */
1200 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1201 {
1202         int r;
1203         struct kvm_vcpu *vcpu, *v;
1204
1205         vcpu = kvm_arch_vcpu_create(kvm, id);
1206         if (IS_ERR(vcpu))
1207                 return PTR_ERR(vcpu);
1208
1209         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1210
1211         r = kvm_arch_vcpu_setup(vcpu);
1212         if (r)
1213                 return r;
1214
1215         mutex_lock(&kvm->lock);
1216         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1217                 r = -EINVAL;
1218                 goto vcpu_destroy;
1219         }
1220
1221         kvm_for_each_vcpu(r, v, kvm)
1222                 if (v->vcpu_id == id) {
1223                         r = -EEXIST;
1224                         goto vcpu_destroy;
1225                 }
1226
1227         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1228
1229         /* Now it's all set up, let userspace reach it */
1230         kvm_get_kvm(kvm);
1231         r = create_vcpu_fd(vcpu);
1232         if (r < 0) {
1233                 kvm_put_kvm(kvm);
1234                 goto vcpu_destroy;
1235         }
1236
1237         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1238         smp_wmb();
1239         atomic_inc(&kvm->online_vcpus);
1240
1241 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1242         if (kvm->bsp_vcpu_id == id)
1243                 kvm->bsp_vcpu = vcpu;
1244 #endif
1245         mutex_unlock(&kvm->lock);
1246         return r;
1247
1248 vcpu_destroy:
1249         mutex_unlock(&kvm->lock);
1250         kvm_arch_vcpu_destroy(vcpu);
1251         return r;
1252 }
1253
1254 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1255 {
1256         if (sigset) {
1257                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1258                 vcpu->sigset_active = 1;
1259                 vcpu->sigset = *sigset;
1260         } else
1261                 vcpu->sigset_active = 0;
1262         return 0;
1263 }
1264
1265 static long kvm_vcpu_ioctl(struct file *filp,
1266                            unsigned int ioctl, unsigned long arg)
1267 {
1268         struct kvm_vcpu *vcpu = filp->private_data;
1269         void __user *argp = (void __user *)arg;
1270         int r;
1271         struct kvm_fpu *fpu = NULL;
1272         struct kvm_sregs *kvm_sregs = NULL;
1273
1274         if (vcpu->kvm->mm != current->mm)
1275                 return -EIO;
1276         switch (ioctl) {
1277         case KVM_RUN:
1278                 r = -EINVAL;
1279                 if (arg)
1280                         goto out;
1281                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1282                 break;
1283         case KVM_GET_REGS: {
1284                 struct kvm_regs *kvm_regs;
1285
1286                 r = -ENOMEM;
1287                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1288                 if (!kvm_regs)
1289                         goto out;
1290                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1291                 if (r)
1292                         goto out_free1;
1293                 r = -EFAULT;
1294                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1295                         goto out_free1;
1296                 r = 0;
1297 out_free1:
1298                 kfree(kvm_regs);
1299                 break;
1300         }
1301         case KVM_SET_REGS: {
1302                 struct kvm_regs *kvm_regs;
1303
1304                 r = -ENOMEM;
1305                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1306                 if (!kvm_regs)
1307                         goto out;
1308                 r = -EFAULT;
1309                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1310                         goto out_free2;
1311                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1312                 if (r)
1313                         goto out_free2;
1314                 r = 0;
1315 out_free2:
1316                 kfree(kvm_regs);
1317                 break;
1318         }
1319         case KVM_GET_SREGS: {
1320                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1321                 r = -ENOMEM;
1322                 if (!kvm_sregs)
1323                         goto out;
1324                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1325                 if (r)
1326                         goto out;
1327                 r = -EFAULT;
1328                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1329                         goto out;
1330                 r = 0;
1331                 break;
1332         }
1333         case KVM_SET_SREGS: {
1334                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1335                 r = -ENOMEM;
1336                 if (!kvm_sregs)
1337                         goto out;
1338                 r = -EFAULT;
1339                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1340                         goto out;
1341                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1342                 if (r)
1343                         goto out;
1344                 r = 0;
1345                 break;
1346         }
1347         case KVM_GET_MP_STATE: {
1348                 struct kvm_mp_state mp_state;
1349
1350                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1351                 if (r)
1352                         goto out;
1353                 r = -EFAULT;
1354                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1355                         goto out;
1356                 r = 0;
1357                 break;
1358         }
1359         case KVM_SET_MP_STATE: {
1360                 struct kvm_mp_state mp_state;
1361
1362                 r = -EFAULT;
1363                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1364                         goto out;
1365                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1366                 if (r)
1367                         goto out;
1368                 r = 0;
1369                 break;
1370         }
1371         case KVM_TRANSLATE: {
1372                 struct kvm_translation tr;
1373
1374                 r = -EFAULT;
1375                 if (copy_from_user(&tr, argp, sizeof tr))
1376                         goto out;
1377                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1378                 if (r)
1379                         goto out;
1380                 r = -EFAULT;
1381                 if (copy_to_user(argp, &tr, sizeof tr))
1382                         goto out;
1383                 r = 0;
1384                 break;
1385         }
1386         case KVM_SET_GUEST_DEBUG: {
1387                 struct kvm_guest_debug dbg;
1388
1389                 r = -EFAULT;
1390                 if (copy_from_user(&dbg, argp, sizeof dbg))
1391                         goto out;
1392                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1393                 if (r)
1394                         goto out;
1395                 r = 0;
1396                 break;
1397         }
1398         case KVM_SET_SIGNAL_MASK: {
1399                 struct kvm_signal_mask __user *sigmask_arg = argp;
1400                 struct kvm_signal_mask kvm_sigmask;
1401                 sigset_t sigset, *p;
1402
1403                 p = NULL;
1404                 if (argp) {
1405                         r = -EFAULT;
1406                         if (copy_from_user(&kvm_sigmask, argp,
1407                                            sizeof kvm_sigmask))
1408                                 goto out;
1409                         r = -EINVAL;
1410                         if (kvm_sigmask.len != sizeof sigset)
1411                                 goto out;
1412                         r = -EFAULT;
1413                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1414                                            sizeof sigset))
1415                                 goto out;
1416                         p = &sigset;
1417                 }
1418                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1419                 break;
1420         }
1421         case KVM_GET_FPU: {
1422                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1423                 r = -ENOMEM;
1424                 if (!fpu)
1425                         goto out;
1426                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1427                 if (r)
1428                         goto out;
1429                 r = -EFAULT;
1430                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1431                         goto out;
1432                 r = 0;
1433                 break;
1434         }
1435         case KVM_SET_FPU: {
1436                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1437                 r = -ENOMEM;
1438                 if (!fpu)
1439                         goto out;
1440                 r = -EFAULT;
1441                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1442                         goto out;
1443                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1444                 if (r)
1445                         goto out;
1446                 r = 0;
1447                 break;
1448         }
1449         default:
1450                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1451         }
1452 out:
1453         kfree(fpu);
1454         kfree(kvm_sregs);
1455         return r;
1456 }
1457
1458 static long kvm_vm_ioctl(struct file *filp,
1459                            unsigned int ioctl, unsigned long arg)
1460 {
1461         struct kvm *kvm = filp->private_data;
1462         void __user *argp = (void __user *)arg;
1463         int r;
1464
1465         if (kvm->mm != current->mm)
1466                 return -EIO;
1467         switch (ioctl) {
1468         case KVM_CREATE_VCPU:
1469                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1470                 if (r < 0)
1471                         goto out;
1472                 break;
1473         case KVM_SET_USER_MEMORY_REGION: {
1474                 struct kvm_userspace_memory_region kvm_userspace_mem;
1475
1476                 r = -EFAULT;
1477                 if (copy_from_user(&kvm_userspace_mem, argp,
1478                                                 sizeof kvm_userspace_mem))
1479                         goto out;
1480
1481                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1482                 if (r)
1483                         goto out;
1484                 break;
1485         }
1486         case KVM_GET_DIRTY_LOG: {
1487                 struct kvm_dirty_log log;
1488
1489                 r = -EFAULT;
1490                 if (copy_from_user(&log, argp, sizeof log))
1491                         goto out;
1492                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1493                 if (r)
1494                         goto out;
1495                 break;
1496         }
1497 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1498         case KVM_REGISTER_COALESCED_MMIO: {
1499                 struct kvm_coalesced_mmio_zone zone;
1500                 r = -EFAULT;
1501                 if (copy_from_user(&zone, argp, sizeof zone))
1502                         goto out;
1503                 r = -ENXIO;
1504                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1505                 if (r)
1506                         goto out;
1507                 r = 0;
1508                 break;
1509         }
1510         case KVM_UNREGISTER_COALESCED_MMIO: {
1511                 struct kvm_coalesced_mmio_zone zone;
1512                 r = -EFAULT;
1513                 if (copy_from_user(&zone, argp, sizeof zone))
1514                         goto out;
1515                 r = -ENXIO;
1516                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1517                 if (r)
1518                         goto out;
1519                 r = 0;
1520                 break;
1521         }
1522 #endif
1523         case KVM_IRQFD: {
1524                 struct kvm_irqfd data;
1525
1526                 r = -EFAULT;
1527                 if (copy_from_user(&data, argp, sizeof data))
1528                         goto out;
1529                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1530                 break;
1531         }
1532         case KVM_IOEVENTFD: {
1533                 struct kvm_ioeventfd data;
1534
1535                 r = -EFAULT;
1536                 if (copy_from_user(&data, argp, sizeof data))
1537                         goto out;
1538                 r = kvm_ioeventfd(kvm, &data);
1539                 break;
1540         }
1541 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1542         case KVM_SET_BOOT_CPU_ID:
1543                 r = 0;
1544                 mutex_lock(&kvm->lock);
1545                 if (atomic_read(&kvm->online_vcpus) != 0)
1546                         r = -EBUSY;
1547                 else
1548                         kvm->bsp_vcpu_id = arg;
1549                 mutex_unlock(&kvm->lock);
1550                 break;
1551 #endif
1552         default:
1553                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1554                 if (r == -ENOTTY)
1555                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1556         }
1557 out:
1558         return r;
1559 }
1560
1561 #ifdef CONFIG_COMPAT
1562 struct compat_kvm_dirty_log {
1563         __u32 slot;
1564         __u32 padding1;
1565         union {
1566                 compat_uptr_t dirty_bitmap; /* one bit per page */
1567                 __u64 padding2;
1568         };
1569 };
1570
1571 static long kvm_vm_compat_ioctl(struct file *filp,
1572                            unsigned int ioctl, unsigned long arg)
1573 {
1574         struct kvm *kvm = filp->private_data;
1575         int r;
1576
1577         if (kvm->mm != current->mm)
1578                 return -EIO;
1579         switch (ioctl) {
1580         case KVM_GET_DIRTY_LOG: {
1581                 struct compat_kvm_dirty_log compat_log;
1582                 struct kvm_dirty_log log;
1583
1584                 r = -EFAULT;
1585                 if (copy_from_user(&compat_log, (void __user *)arg,
1586                                    sizeof(compat_log)))
1587                         goto out;
1588                 log.slot         = compat_log.slot;
1589                 log.padding1     = compat_log.padding1;
1590                 log.padding2     = compat_log.padding2;
1591                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1592
1593                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1594                 if (r)
1595                         goto out;
1596                 break;
1597         }
1598         default:
1599                 r = kvm_vm_ioctl(filp, ioctl, arg);
1600         }
1601
1602 out:
1603         return r;
1604 }
1605 #endif
1606
1607 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1608 {
1609         struct page *page[1];
1610         unsigned long addr;
1611         int npages;
1612         gfn_t gfn = vmf->pgoff;
1613         struct kvm *kvm = vma->vm_file->private_data;
1614
1615         addr = gfn_to_hva(kvm, gfn);
1616         if (kvm_is_error_hva(addr))
1617                 return VM_FAULT_SIGBUS;
1618
1619         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1620                                 NULL);
1621         if (unlikely(npages != 1))
1622                 return VM_FAULT_SIGBUS;
1623
1624         vmf->page = page[0];
1625         return 0;
1626 }
1627
1628 static const struct vm_operations_struct kvm_vm_vm_ops = {
1629         .fault = kvm_vm_fault,
1630 };
1631
1632 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1633 {
1634         vma->vm_ops = &kvm_vm_vm_ops;
1635         return 0;
1636 }
1637
1638 static struct file_operations kvm_vm_fops = {
1639         .release        = kvm_vm_release,
1640         .unlocked_ioctl = kvm_vm_ioctl,
1641 #ifdef CONFIG_COMPAT
1642         .compat_ioctl   = kvm_vm_compat_ioctl,
1643 #endif
1644         .mmap           = kvm_vm_mmap,
1645 };
1646
1647 static int kvm_dev_ioctl_create_vm(void)
1648 {
1649         int fd;
1650         struct kvm *kvm;
1651
1652         kvm = kvm_create_vm();
1653         if (IS_ERR(kvm))
1654                 return PTR_ERR(kvm);
1655         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1656         if (fd < 0)
1657                 kvm_put_kvm(kvm);
1658
1659         return fd;
1660 }
1661
1662 static long kvm_dev_ioctl_check_extension_generic(long arg)
1663 {
1664         switch (arg) {
1665         case KVM_CAP_USER_MEMORY:
1666         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1667         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1668 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1669         case KVM_CAP_SET_BOOT_CPU_ID:
1670 #endif
1671         case KVM_CAP_INTERNAL_ERROR_DATA:
1672                 return 1;
1673 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1674         case KVM_CAP_IRQ_ROUTING:
1675                 return KVM_MAX_IRQ_ROUTES;
1676 #endif
1677         default:
1678                 break;
1679         }
1680         return kvm_dev_ioctl_check_extension(arg);
1681 }
1682
1683 static long kvm_dev_ioctl(struct file *filp,
1684                           unsigned int ioctl, unsigned long arg)
1685 {
1686         long r = -EINVAL;
1687
1688         switch (ioctl) {
1689         case KVM_GET_API_VERSION:
1690                 r = -EINVAL;
1691                 if (arg)
1692                         goto out;
1693                 r = KVM_API_VERSION;
1694                 break;
1695         case KVM_CREATE_VM:
1696                 r = -EINVAL;
1697                 if (arg)
1698                         goto out;
1699                 r = kvm_dev_ioctl_create_vm();
1700                 break;
1701         case KVM_CHECK_EXTENSION:
1702                 r = kvm_dev_ioctl_check_extension_generic(arg);
1703                 break;
1704         case KVM_GET_VCPU_MMAP_SIZE:
1705                 r = -EINVAL;
1706                 if (arg)
1707                         goto out;
1708                 r = PAGE_SIZE;     /* struct kvm_run */
1709 #ifdef CONFIG_X86
1710                 r += PAGE_SIZE;    /* pio data page */
1711 #endif
1712 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1713                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1714 #endif
1715                 break;
1716         case KVM_TRACE_ENABLE:
1717         case KVM_TRACE_PAUSE:
1718         case KVM_TRACE_DISABLE:
1719                 r = -EOPNOTSUPP;
1720                 break;
1721         default:
1722                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1723         }
1724 out:
1725         return r;
1726 }
1727
1728 static struct file_operations kvm_chardev_ops = {
1729         .unlocked_ioctl = kvm_dev_ioctl,
1730         .compat_ioctl   = kvm_dev_ioctl,
1731 };
1732
1733 static struct miscdevice kvm_dev = {
1734         KVM_MINOR,
1735         "kvm",
1736         &kvm_chardev_ops,
1737 };
1738
1739 static void hardware_enable(void *junk)
1740 {
1741         int cpu = raw_smp_processor_id();
1742         int r;
1743
1744         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1745                 return;
1746
1747         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1748
1749         r = kvm_arch_hardware_enable(NULL);
1750
1751         if (r) {
1752                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1753                 atomic_inc(&hardware_enable_failed);
1754                 printk(KERN_INFO "kvm: enabling virtualization on "
1755                                  "CPU%d failed\n", cpu);
1756         }
1757 }
1758
1759 static void hardware_disable(void *junk)
1760 {
1761         int cpu = raw_smp_processor_id();
1762
1763         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1764                 return;
1765         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1766         kvm_arch_hardware_disable(NULL);
1767 }
1768
1769 static void hardware_disable_all_nolock(void)
1770 {
1771         BUG_ON(!kvm_usage_count);
1772
1773         kvm_usage_count--;
1774         if (!kvm_usage_count)
1775                 on_each_cpu(hardware_disable, NULL, 1);
1776 }
1777
1778 static void hardware_disable_all(void)
1779 {
1780         spin_lock(&kvm_lock);
1781         hardware_disable_all_nolock();
1782         spin_unlock(&kvm_lock);
1783 }
1784
1785 static int hardware_enable_all(void)
1786 {
1787         int r = 0;
1788
1789         spin_lock(&kvm_lock);
1790
1791         kvm_usage_count++;
1792         if (kvm_usage_count == 1) {
1793                 atomic_set(&hardware_enable_failed, 0);
1794                 on_each_cpu(hardware_enable, NULL, 1);
1795
1796                 if (atomic_read(&hardware_enable_failed)) {
1797                         hardware_disable_all_nolock();
1798                         r = -EBUSY;
1799                 }
1800         }
1801
1802         spin_unlock(&kvm_lock);
1803
1804         return r;
1805 }
1806
1807 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1808                            void *v)
1809 {
1810         int cpu = (long)v;
1811
1812         if (!kvm_usage_count)
1813                 return NOTIFY_OK;
1814
1815         val &= ~CPU_TASKS_FROZEN;
1816         switch (val) {
1817         case CPU_DYING:
1818                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1819                        cpu);
1820                 hardware_disable(NULL);
1821                 break;
1822         case CPU_UP_CANCELED:
1823                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1824                        cpu);
1825                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1826                 break;
1827         case CPU_ONLINE:
1828                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1829                        cpu);
1830                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1831                 break;
1832         }
1833         return NOTIFY_OK;
1834 }
1835
1836
1837 asmlinkage void kvm_handle_fault_on_reboot(void)
1838 {
1839         if (kvm_rebooting)
1840                 /* spin while reset goes on */
1841                 while (true)
1842                         ;
1843         /* Fault while not rebooting.  We want the trace. */
1844         BUG();
1845 }
1846 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1847
1848 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1849                       void *v)
1850 {
1851         /*
1852          * Some (well, at least mine) BIOSes hang on reboot if
1853          * in vmx root mode.
1854          *
1855          * And Intel TXT required VMX off for all cpu when system shutdown.
1856          */
1857         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1858         kvm_rebooting = true;
1859         on_each_cpu(hardware_disable, NULL, 1);
1860         return NOTIFY_OK;
1861 }
1862
1863 static struct notifier_block kvm_reboot_notifier = {
1864         .notifier_call = kvm_reboot,
1865         .priority = 0,
1866 };
1867
1868 void kvm_io_bus_init(struct kvm_io_bus *bus)
1869 {
1870         memset(bus, 0, sizeof(*bus));
1871 }
1872
1873 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1874 {
1875         int i;
1876
1877         for (i = 0; i < bus->dev_count; i++) {
1878                 struct kvm_io_device *pos = bus->devs[i];
1879
1880                 kvm_iodevice_destructor(pos);
1881         }
1882 }
1883
1884 /* kvm_io_bus_write - called under kvm->slots_lock */
1885 int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1886                      int len, const void *val)
1887 {
1888         int i;
1889         for (i = 0; i < bus->dev_count; i++)
1890                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1891                         return 0;
1892         return -EOPNOTSUPP;
1893 }
1894
1895 /* kvm_io_bus_read - called under kvm->slots_lock */
1896 int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1897 {
1898         int i;
1899         for (i = 0; i < bus->dev_count; i++)
1900                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1901                         return 0;
1902         return -EOPNOTSUPP;
1903 }
1904
1905 int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
1906                              struct kvm_io_device *dev)
1907 {
1908         int ret;
1909
1910         down_write(&kvm->slots_lock);
1911         ret = __kvm_io_bus_register_dev(bus, dev);
1912         up_write(&kvm->slots_lock);
1913
1914         return ret;
1915 }
1916
1917 /* An unlocked version. Caller must have write lock on slots_lock. */
1918 int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1919                               struct kvm_io_device *dev)
1920 {
1921         if (bus->dev_count > NR_IOBUS_DEVS-1)
1922                 return -ENOSPC;
1923
1924         bus->devs[bus->dev_count++] = dev;
1925
1926         return 0;
1927 }
1928
1929 void kvm_io_bus_unregister_dev(struct kvm *kvm,
1930                                struct kvm_io_bus *bus,
1931                                struct kvm_io_device *dev)
1932 {
1933         down_write(&kvm->slots_lock);
1934         __kvm_io_bus_unregister_dev(bus, dev);
1935         up_write(&kvm->slots_lock);
1936 }
1937
1938 /* An unlocked version. Caller must have write lock on slots_lock. */
1939 void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1940                                  struct kvm_io_device *dev)
1941 {
1942         int i;
1943
1944         for (i = 0; i < bus->dev_count; i++)
1945                 if (bus->devs[i] == dev) {
1946                         bus->devs[i] = bus->devs[--bus->dev_count];
1947                         break;
1948                 }
1949 }
1950
1951 static struct notifier_block kvm_cpu_notifier = {
1952         .notifier_call = kvm_cpu_hotplug,
1953         .priority = 20, /* must be > scheduler priority */
1954 };
1955
1956 static int vm_stat_get(void *_offset, u64 *val)
1957 {
1958         unsigned offset = (long)_offset;
1959         struct kvm *kvm;
1960
1961         *val = 0;
1962         spin_lock(&kvm_lock);
1963         list_for_each_entry(kvm, &vm_list, vm_list)
1964                 *val += *(u32 *)((void *)kvm + offset);
1965         spin_unlock(&kvm_lock);
1966         return 0;
1967 }
1968
1969 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1970
1971 static int vcpu_stat_get(void *_offset, u64 *val)
1972 {
1973         unsigned offset = (long)_offset;
1974         struct kvm *kvm;
1975         struct kvm_vcpu *vcpu;
1976         int i;
1977
1978         *val = 0;
1979         spin_lock(&kvm_lock);
1980         list_for_each_entry(kvm, &vm_list, vm_list)
1981                 kvm_for_each_vcpu(i, vcpu, kvm)
1982                         *val += *(u32 *)((void *)vcpu + offset);
1983
1984         spin_unlock(&kvm_lock);
1985         return 0;
1986 }
1987
1988 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1989
1990 static const struct file_operations *stat_fops[] = {
1991         [KVM_STAT_VCPU] = &vcpu_stat_fops,
1992         [KVM_STAT_VM]   = &vm_stat_fops,
1993 };
1994
1995 static void kvm_init_debug(void)
1996 {
1997         struct kvm_stats_debugfs_item *p;
1998
1999         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2000         for (p = debugfs_entries; p->name; ++p)
2001                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2002                                                 (void *)(long)p->offset,
2003                                                 stat_fops[p->kind]);
2004 }
2005
2006 static void kvm_exit_debug(void)
2007 {
2008         struct kvm_stats_debugfs_item *p;
2009
2010         for (p = debugfs_entries; p->name; ++p)
2011                 debugfs_remove(p->dentry);
2012         debugfs_remove(kvm_debugfs_dir);
2013 }
2014
2015 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2016 {
2017         if (kvm_usage_count)
2018                 hardware_disable(NULL);
2019         return 0;
2020 }
2021
2022 static int kvm_resume(struct sys_device *dev)
2023 {
2024         if (kvm_usage_count)
2025                 hardware_enable(NULL);
2026         return 0;
2027 }
2028
2029 static struct sysdev_class kvm_sysdev_class = {
2030         .name = "kvm",
2031         .suspend = kvm_suspend,
2032         .resume = kvm_resume,
2033 };
2034
2035 static struct sys_device kvm_sysdev = {
2036         .id = 0,
2037         .cls = &kvm_sysdev_class,
2038 };
2039
2040 struct page *bad_page;
2041 pfn_t bad_pfn;
2042
2043 static inline
2044 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2045 {
2046         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2047 }
2048
2049 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2050 {
2051         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2052
2053         kvm_arch_vcpu_load(vcpu, cpu);
2054 }
2055
2056 static void kvm_sched_out(struct preempt_notifier *pn,
2057                           struct task_struct *next)
2058 {
2059         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2060
2061         kvm_arch_vcpu_put(vcpu);
2062 }
2063
2064 int kvm_init(void *opaque, unsigned int vcpu_size,
2065                   struct module *module)
2066 {
2067         int r;
2068         int cpu;
2069
2070         r = kvm_arch_init(opaque);
2071         if (r)
2072                 goto out_fail;
2073
2074         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2075
2076         if (bad_page == NULL) {
2077                 r = -ENOMEM;
2078                 goto out;
2079         }
2080
2081         bad_pfn = page_to_pfn(bad_page);
2082
2083         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2084                 r = -ENOMEM;
2085                 goto out_free_0;
2086         }
2087
2088         r = kvm_arch_hardware_setup();
2089         if (r < 0)
2090                 goto out_free_0a;
2091
2092         for_each_online_cpu(cpu) {
2093                 smp_call_function_single(cpu,
2094                                 kvm_arch_check_processor_compat,
2095                                 &r, 1);
2096                 if (r < 0)
2097                         goto out_free_1;
2098         }
2099
2100         r = register_cpu_notifier(&kvm_cpu_notifier);
2101         if (r)
2102                 goto out_free_2;
2103         register_reboot_notifier(&kvm_reboot_notifier);
2104
2105         r = sysdev_class_register(&kvm_sysdev_class);
2106         if (r)
2107                 goto out_free_3;
2108
2109         r = sysdev_register(&kvm_sysdev);
2110         if (r)
2111                 goto out_free_4;
2112
2113         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2114         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2115                                            __alignof__(struct kvm_vcpu),
2116                                            0, NULL);
2117         if (!kvm_vcpu_cache) {
2118                 r = -ENOMEM;
2119                 goto out_free_5;
2120         }
2121
2122         kvm_chardev_ops.owner = module;
2123         kvm_vm_fops.owner = module;
2124         kvm_vcpu_fops.owner = module;
2125
2126         r = misc_register(&kvm_dev);
2127         if (r) {
2128                 printk(KERN_ERR "kvm: misc device register failed\n");
2129                 goto out_free;
2130         }
2131
2132         kvm_preempt_ops.sched_in = kvm_sched_in;
2133         kvm_preempt_ops.sched_out = kvm_sched_out;
2134
2135         kvm_init_debug();
2136
2137         return 0;
2138
2139 out_free:
2140         kmem_cache_destroy(kvm_vcpu_cache);
2141 out_free_5:
2142         sysdev_unregister(&kvm_sysdev);
2143 out_free_4:
2144         sysdev_class_unregister(&kvm_sysdev_class);
2145 out_free_3:
2146         unregister_reboot_notifier(&kvm_reboot_notifier);
2147         unregister_cpu_notifier(&kvm_cpu_notifier);
2148 out_free_2:
2149 out_free_1:
2150         kvm_arch_hardware_unsetup();
2151 out_free_0a:
2152         free_cpumask_var(cpus_hardware_enabled);
2153 out_free_0:
2154         __free_page(bad_page);
2155 out:
2156         kvm_arch_exit();
2157 out_fail:
2158         return r;
2159 }
2160 EXPORT_SYMBOL_GPL(kvm_init);
2161
2162 void kvm_exit(void)
2163 {
2164         tracepoint_synchronize_unregister();
2165         kvm_exit_debug();
2166         misc_deregister(&kvm_dev);
2167         kmem_cache_destroy(kvm_vcpu_cache);
2168         sysdev_unregister(&kvm_sysdev);
2169         sysdev_class_unregister(&kvm_sysdev_class);
2170         unregister_reboot_notifier(&kvm_reboot_notifier);
2171         unregister_cpu_notifier(&kvm_cpu_notifier);
2172         on_each_cpu(hardware_disable, NULL, 1);
2173         kvm_arch_hardware_unsetup();
2174         kvm_arch_exit();
2175         free_cpumask_var(cpus_hardware_enabled);
2176         __free_page(bad_page);
2177 }
2178 EXPORT_SYMBOL_GPL(kvm_exit);