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