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