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  * Copyright 2010 Red Hat, Inc. and/or its affilates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.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 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50
  51 #include <asm/processor.h>
  52 #include <asm/io.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/pgtable.h>
  55 #include <asm-generic/bitops/le.h>
  56
  57 #include "coalesced_mmio.h"
  58
  59 #define CREATE_TRACE_POINTS
  60 #include <trace/events/kvm.h>
  61
  62 MODULE_AUTHOR("Qumranet");
  63 MODULE_LICENSE("GPL");
  64
  65 /*
  66  * Ordering of locks:
  67  *
  68  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  69  */
  70
  71 DEFINE_SPINLOCK(kvm_lock);
  72 LIST_HEAD(vm_list);
  73
  74 static cpumask_var_t cpus_hardware_enabled;
  75 static int kvm_usage_count = 0;
  76 static atomic_t hardware_enable_failed;
  77
  78 struct kmem_cache *kvm_vcpu_cache;
  79 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  80
  81 static __read_mostly struct preempt_ops kvm_preempt_ops;
  82
  83 struct dentry *kvm_debugfs_dir;
  84
  85 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  86                            unsigned long arg);
  87 static int hardware_enable_all(void);
  88 static void hardware_disable_all(void);
  89
  90 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  91
  92 static bool kvm_rebooting;
  93
  94 static bool largepages_enabled = true;
  95
  96 static struct page *hwpoison_page;
  97 static pfn_t hwpoison_pfn;
  98
  99 static struct page *fault_page;
 100 static pfn_t fault_pfn;
 101
 102 inline int kvm_is_mmio_pfn(pfn_t pfn)
 103 {
 104         if (pfn_valid(pfn)) {
 105                 struct page *page = compound_head(pfn_to_page(pfn));
 106                 return PageReserved(page);
 107         }
 108
 109         return true;
 110 }
 111
 112 /*
 113  * Switches to specified vcpu, until a matching vcpu_put()
 114  */
 115 void vcpu_load(struct kvm_vcpu *vcpu)
 116 {
 117         int cpu;
 118
 119         mutex_lock(&vcpu->mutex);
 120         cpu = get_cpu();
 121         preempt_notifier_register(&vcpu->preempt_notifier);
 122         kvm_arch_vcpu_load(vcpu, cpu);
 123         put_cpu();
 124 }
 125
 126 void vcpu_put(struct kvm_vcpu *vcpu)
 127 {
 128         preempt_disable();
 129         kvm_arch_vcpu_put(vcpu);
 130         preempt_notifier_unregister(&vcpu->preempt_notifier);
 131         preempt_enable();
 132         mutex_unlock(&vcpu->mutex);
 133 }
 134
 135 static void ack_flush(void *_completed)
 136 {
 137 }
 138
 139 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 140 {
 141         int i, cpu, me;
 142         cpumask_var_t cpus;
 143         bool called = true;
 144         struct kvm_vcpu *vcpu;
 145
 146         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 147
 148         raw_spin_lock(&kvm->requests_lock);
 149         me = smp_processor_id();
 150         kvm_for_each_vcpu(i, vcpu, kvm) {
 151                 if (kvm_make_check_request(req, vcpu))
 152                         continue;
 153                 cpu = vcpu->cpu;
 154                 if (cpus != NULL && cpu != -1 && cpu != me)
 155                         cpumask_set_cpu(cpu, cpus);
 156         }
 157         if (unlikely(cpus == NULL))
 158                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 159         else if (!cpumask_empty(cpus))
 160                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 161         else
 162                 called = false;
 163         raw_spin_unlock(&kvm->requests_lock);
 164         free_cpumask_var(cpus);
 165         return called;
 166 }
 167
 168 void kvm_flush_remote_tlbs(struct kvm *kvm)
 169 {
 170         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 171                 ++kvm->stat.remote_tlb_flush;
 172 }
 173
 174 void kvm_reload_remote_mmus(struct kvm *kvm)
 175 {
 176         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 177 }
 178
 179 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 180 {
 181         struct page *page;
 182         int r;
 183
 184         mutex_init(&vcpu->mutex);
 185         vcpu->cpu = -1;
 186         vcpu->kvm = kvm;
 187         vcpu->vcpu_id = id;
 188         init_waitqueue_head(&vcpu->wq);
 189
 190         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 191         if (!page) {
 192                 r = -ENOMEM;
 193                 goto fail;
 194         }
 195         vcpu->run = page_address(page);
 196
 197         r = kvm_arch_vcpu_init(vcpu);
 198         if (r < 0)
 199                 goto fail_free_run;
 200         return 0;
 201
 202 fail_free_run:
 203         free_page((unsigned long)vcpu->run);
 204 fail:
 205         return r;
 206 }
 207 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 208
 209 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 210 {
 211         kvm_arch_vcpu_uninit(vcpu);
 212         free_page((unsigned long)vcpu->run);
 213 }
 214 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 215
 216 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 217 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 218 {
 219         return container_of(mn, struct kvm, mmu_notifier);
 220 }
 221
 222 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 223                                              struct mm_struct *mm,
 224                                              unsigned long address)
 225 {
 226         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 227         int need_tlb_flush, idx;
 228
 229         /*
 230          * When ->invalidate_page runs, the linux pte has been zapped
 231          * already but the page is still allocated until
 232          * ->invalidate_page returns. So if we increase the sequence
 233          * here the kvm page fault will notice if the spte can't be
 234          * established because the page is going to be freed. If
 235          * instead the kvm page fault establishes the spte before
 236          * ->invalidate_page runs, kvm_unmap_hva will release it
 237          * before returning.
 238          *
 239          * The sequence increase only need to be seen at spin_unlock
 240          * time, and not at spin_lock time.
 241          *
 242          * Increasing the sequence after the spin_unlock would be
 243          * unsafe because the kvm page fault could then establish the
 244          * pte after kvm_unmap_hva returned, without noticing the page
 245          * is going to be freed.
 246          */
 247         idx = srcu_read_lock(&kvm->srcu);
 248         spin_lock(&kvm->mmu_lock);
 249         kvm->mmu_notifier_seq++;
 250         need_tlb_flush = kvm_unmap_hva(kvm, address);
 251         spin_unlock(&kvm->mmu_lock);
 252         srcu_read_unlock(&kvm->srcu, idx);
 253
 254         /* we've to flush the tlb before the pages can be freed */
 255         if (need_tlb_flush)
 256                 kvm_flush_remote_tlbs(kvm);
 257
 258 }
 259
 260 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 261                                         struct mm_struct *mm,
 262                                         unsigned long address,
 263                                         pte_t pte)
 264 {
 265         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 266         int idx;
 267
 268         idx = srcu_read_lock(&kvm->srcu);
 269         spin_lock(&kvm->mmu_lock);
 270         kvm->mmu_notifier_seq++;
 271         kvm_set_spte_hva(kvm, address, pte);
 272         spin_unlock(&kvm->mmu_lock);
 273         srcu_read_unlock(&kvm->srcu, idx);
 274 }
 275
 276 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 277                                                     struct mm_struct *mm,
 278                                                     unsigned long start,
 279                                                     unsigned long end)
 280 {
 281         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 282         int need_tlb_flush = 0, idx;
 283
 284         idx = srcu_read_lock(&kvm->srcu);
 285         spin_lock(&kvm->mmu_lock);
 286         /*
 287          * The count increase must become visible at unlock time as no
 288          * spte can be established without taking the mmu_lock and
 289          * count is also read inside the mmu_lock critical section.
 290          */
 291         kvm->mmu_notifier_count++;
 292         for (; start < end; start += PAGE_SIZE)
 293                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 294         spin_unlock(&kvm->mmu_lock);
 295         srcu_read_unlock(&kvm->srcu, idx);
 296
 297         /* we've to flush the tlb before the pages can be freed */
 298         if (need_tlb_flush)
 299                 kvm_flush_remote_tlbs(kvm);
 300 }
 301
 302 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 303                                                   struct mm_struct *mm,
 304                                                   unsigned long start,
 305                                                   unsigned long end)
 306 {
 307         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 308
 309         spin_lock(&kvm->mmu_lock);
 310         /*
 311          * This sequence increase will notify the kvm page fault that
 312          * the page that is going to be mapped in the spte could have
 313          * been freed.
 314          */
 315         kvm->mmu_notifier_seq++;
 316         /*
 317          * The above sequence increase must be visible before the
 318          * below count decrease but both values are read by the kvm
 319          * page fault under mmu_lock spinlock so we don't need to add
 320          * a smb_wmb() here in between the two.
 321          */
 322         kvm->mmu_notifier_count--;
 323         spin_unlock(&kvm->mmu_lock);
 324
 325         BUG_ON(kvm->mmu_notifier_count < 0);
 326 }
 327
 328 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 329                                               struct mm_struct *mm,
 330                                               unsigned long address)
 331 {
 332         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 333         int young, idx;
 334
 335         idx = srcu_read_lock(&kvm->srcu);
 336         spin_lock(&kvm->mmu_lock);
 337         young = kvm_age_hva(kvm, address);
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340
 341         if (young)
 342                 kvm_flush_remote_tlbs(kvm);
 343
 344         return young;
 345 }
 346
 347 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 348                                      struct mm_struct *mm)
 349 {
 350         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 351         int idx;
 352
 353         idx = srcu_read_lock(&kvm->srcu);
 354         kvm_arch_flush_shadow(kvm);
 355         srcu_read_unlock(&kvm->srcu, idx);
 356 }
 357
 358 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 359         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 360         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 361         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 362         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 363         .change_pte             = kvm_mmu_notifier_change_pte,
 364         .release                = kvm_mmu_notifier_release,
 365 };
 366
 367 static int kvm_init_mmu_notifier(struct kvm *kvm)
 368 {
 369         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 370         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 371 }
 372
 373 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 374
 375 static int kvm_init_mmu_notifier(struct kvm *kvm)
 376 {
 377         return 0;
 378 }
 379
 380 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 381
 382 static struct kvm *kvm_create_vm(void)
 383 {
 384         int r = 0, i;
 385         struct kvm *kvm = kvm_arch_create_vm();
 386
 387         if (IS_ERR(kvm))
 388                 goto out;
 389
 390         r = hardware_enable_all();
 391         if (r)
 392                 goto out_err_nodisable;
 393
 394 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 395         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 396         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 397 #endif
 398
 399         r = -ENOMEM;
 400         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 401         if (!kvm->memslots)
 402                 goto out_err;
 403         if (init_srcu_struct(&kvm->srcu))
 404                 goto out_err;
 405         for (i = 0; i < KVM_NR_BUSES; i++) {
 406                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 407                                         GFP_KERNEL);
 408                 if (!kvm->buses[i]) {
 409                         cleanup_srcu_struct(&kvm->srcu);
 410                         goto out_err;
 411                 }
 412         }
 413
 414         r = kvm_init_mmu_notifier(kvm);
 415         if (r) {
 416                 cleanup_srcu_struct(&kvm->srcu);
 417                 goto out_err;
 418         }
 419
 420         kvm->mm = current->mm;
 421         atomic_inc(&kvm->mm->mm_count);
 422         spin_lock_init(&kvm->mmu_lock);
 423         raw_spin_lock_init(&kvm->requests_lock);
 424         kvm_eventfd_init(kvm);
 425         mutex_init(&kvm->lock);
 426         mutex_init(&kvm->irq_lock);
 427         mutex_init(&kvm->slots_lock);
 428         atomic_set(&kvm->users_count, 1);
 429         spin_lock(&kvm_lock);
 430         list_add(&kvm->vm_list, &vm_list);
 431         spin_unlock(&kvm_lock);
 432 out:
 433         return kvm;
 434
 435 out_err:
 436         hardware_disable_all();
 437 out_err_nodisable:
 438         for (i = 0; i < KVM_NR_BUSES; i++)
 439                 kfree(kvm->buses[i]);
 440         kfree(kvm->memslots);
 441         kfree(kvm);
 442         return ERR_PTR(r);
 443 }
 444
 445 /*
 446  * Free any memory in @free but not in @dont.
 447  */
 448 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 449                                   struct kvm_memory_slot *dont)
 450 {
 451         int i;
 452
 453         if (!dont || free->rmap != dont->rmap)
 454                 vfree(free->rmap);
 455
 456         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 457                 vfree(free->dirty_bitmap);
 458
 459
 460         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 461                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 462                         vfree(free->lpage_info[i]);
 463                         free->lpage_info[i] = NULL;
 464                 }
 465         }
 466
 467         free->npages = 0;
 468         free->dirty_bitmap = NULL;
 469         free->rmap = NULL;
 470 }
 471
 472 void kvm_free_physmem(struct kvm *kvm)
 473 {
 474         int i;
 475         struct kvm_memslots *slots = kvm->memslots;
 476
 477         for (i = 0; i < slots->nmemslots; ++i)
 478                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 479
 480         kfree(kvm->memslots);
 481 }
 482
 483 static void kvm_destroy_vm(struct kvm *kvm)
 484 {
 485         int i;
 486         struct mm_struct *mm = kvm->mm;
 487
 488         kvm_arch_sync_events(kvm);
 489         spin_lock(&kvm_lock);
 490         list_del(&kvm->vm_list);
 491         spin_unlock(&kvm_lock);
 492         kvm_free_irq_routing(kvm);
 493         for (i = 0; i < KVM_NR_BUSES; i++)
 494                 kvm_io_bus_destroy(kvm->buses[i]);
 495         kvm_coalesced_mmio_free(kvm);
 496 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 497         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 498 #else
 499         kvm_arch_flush_shadow(kvm);
 500 #endif
 501         kvm_arch_destroy_vm(kvm);
 502         hardware_disable_all();
 503         mmdrop(mm);
 504 }
 505
 506 void kvm_get_kvm(struct kvm *kvm)
 507 {
 508         atomic_inc(&kvm->users_count);
 509 }
 510 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 511
 512 void kvm_put_kvm(struct kvm *kvm)
 513 {
 514         if (atomic_dec_and_test(&kvm->users_count))
 515                 kvm_destroy_vm(kvm);
 516 }
 517 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 518
 519
 520 static int kvm_vm_release(struct inode *inode, struct file *filp)
 521 {
 522         struct kvm *kvm = filp->private_data;
 523
 524         kvm_irqfd_release(kvm);
 525
 526         kvm_put_kvm(kvm);
 527         return 0;
 528 }
 529
 530 /*
 531  * Allocate some memory and give it an address in the guest physical address
 532  * space.
 533  *
 534  * Discontiguous memory is allowed, mostly for framebuffers.
 535  *
 536  * Must be called holding mmap_sem for write.
 537  */
 538 int __kvm_set_memory_region(struct kvm *kvm,
 539                             struct kvm_userspace_memory_region *mem,
 540                             int user_alloc)
 541 {
 542         int r, flush_shadow = 0;
 543         gfn_t base_gfn;
 544         unsigned long npages;
 545         unsigned long i;
 546         struct kvm_memory_slot *memslot;
 547         struct kvm_memory_slot old, new;
 548         struct kvm_memslots *slots, *old_memslots;
 549
 550         r = -EINVAL;
 551         /* General sanity checks */
 552         if (mem->memory_size & (PAGE_SIZE - 1))
 553                 goto out;
 554         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 555                 goto out;
 556         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 557                 goto out;
 558         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 559                 goto out;
 560         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 561                 goto out;
 562
 563         memslot = &kvm->memslots->memslots[mem->slot];
 564         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 565         npages = mem->memory_size >> PAGE_SHIFT;
 566
 567         r = -EINVAL;
 568         if (npages > KVM_MEM_MAX_NR_PAGES)
 569                 goto out;
 570
 571         if (!npages)
 572                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 573
 574         new = old = *memslot;
 575
 576         new.id = mem->slot;
 577         new.base_gfn = base_gfn;
 578         new.npages = npages;
 579         new.flags = mem->flags;
 580
 581         /* Disallow changing a memory slot's size. */
 582         r = -EINVAL;
 583         if (npages && old.npages && npages != old.npages)
 584                 goto out_free;
 585
 586         /* Check for overlaps */
 587         r = -EEXIST;
 588         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 589                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 590
 591                 if (s == memslot || !s->npages)
 592                         continue;
 593                 if (!((base_gfn + npages <= s->base_gfn) ||
 594                       (base_gfn >= s->base_gfn + s->npages)))
 595                         goto out_free;
 596         }
 597
 598         /* Free page dirty bitmap if unneeded */
 599         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 600                 new.dirty_bitmap = NULL;
 601
 602         r = -ENOMEM;
 603
 604         /* Allocate if a slot is being created */
 605 #ifndef CONFIG_S390
 606         if (npages && !new.rmap) {
 607                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
 608
 609                 if (!new.rmap)
 610                         goto out_free;
 611
 612                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 613
 614                 new.user_alloc = user_alloc;
 615                 new.userspace_addr = mem->userspace_addr;
 616         }
 617         if (!npages)
 618                 goto skip_lpage;
 619
 620         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 621                 unsigned long ugfn;
 622                 unsigned long j;
 623                 int lpages;
 624                 int level = i + 2;
 625
 626                 /* Avoid unused variable warning if no large pages */
 627                 (void)level;
 628
 629                 if (new.lpage_info[i])
 630                         continue;
 631
 632                 lpages = 1 + ((base_gfn + npages - 1)
 633                              >> KVM_HPAGE_GFN_SHIFT(level));
 634                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 635
 636                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 637
 638                 if (!new.lpage_info[i])
 639                         goto out_free;
 640
 641                 memset(new.lpage_info[i], 0,
 642                        lpages * sizeof(*new.lpage_info[i]));
 643
 644                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 645                         new.lpage_info[i][0].write_count = 1;
 646                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 647                         new.lpage_info[i][lpages - 1].write_count = 1;
 648                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 649                 /*
 650                  * If the gfn and userspace address are not aligned wrt each
 651                  * other, or if explicitly asked to, disable large page
 652                  * support for this slot
 653                  */
 654                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 655                     !largepages_enabled)
 656                         for (j = 0; j < lpages; ++j)
 657                                 new.lpage_info[i][j].write_count = 1;
 658         }
 659
 660 skip_lpage:
 661
 662         /* Allocate page dirty bitmap if needed */
 663         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 664                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
 665
 666                 new.dirty_bitmap = vmalloc(dirty_bytes);
 667                 if (!new.dirty_bitmap)
 668                         goto out_free;
 669                 memset(new.dirty_bitmap, 0, dirty_bytes);
 670                 /* destroy any largepage mappings for dirty tracking */
 671                 if (old.npages)
 672                         flush_shadow = 1;
 673         }
 674 #else  /* not defined CONFIG_S390 */
 675         new.user_alloc = user_alloc;
 676         if (user_alloc)
 677                 new.userspace_addr = mem->userspace_addr;
 678 #endif /* not defined CONFIG_S390 */
 679
 680         if (!npages) {
 681                 r = -ENOMEM;
 682                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 683                 if (!slots)
 684                         goto out_free;
 685                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 686                 if (mem->slot >= slots->nmemslots)
 687                         slots->nmemslots = mem->slot + 1;
 688                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 689
 690                 old_memslots = kvm->memslots;
 691                 rcu_assign_pointer(kvm->memslots, slots);
 692                 synchronize_srcu_expedited(&kvm->srcu);
 693                 /* From this point no new shadow pages pointing to a deleted
 694                  * memslot will be created.
 695                  *
 696                  * validation of sp->gfn happens in:
 697                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 698                  *      - kvm_is_visible_gfn (mmu_check_roots)
 699                  */
 700                 kvm_arch_flush_shadow(kvm);
 701                 kfree(old_memslots);
 702         }
 703
 704         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 705         if (r)
 706                 goto out_free;
 707
 708 #ifdef CONFIG_DMAR
 709         /* map the pages in iommu page table */
 710         if (npages) {
 711                 r = kvm_iommu_map_pages(kvm, &new);
 712                 if (r)
 713                         goto out_free;
 714         }
 715 #endif
 716
 717         r = -ENOMEM;
 718         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 719         if (!slots)
 720                 goto out_free;
 721         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 722         if (mem->slot >= slots->nmemslots)
 723                 slots->nmemslots = mem->slot + 1;
 724
 725         /* actual memory is freed via old in kvm_free_physmem_slot below */
 726         if (!npages) {
 727                 new.rmap = NULL;
 728                 new.dirty_bitmap = NULL;
 729                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 730                         new.lpage_info[i] = NULL;
 731         }
 732
 733         slots->memslots[mem->slot] = new;
 734         old_memslots = kvm->memslots;
 735         rcu_assign_pointer(kvm->memslots, slots);
 736         synchronize_srcu_expedited(&kvm->srcu);
 737
 738         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 739
 740         kvm_free_physmem_slot(&old, &new);
 741         kfree(old_memslots);
 742
 743         if (flush_shadow)
 744                 kvm_arch_flush_shadow(kvm);
 745
 746         return 0;
 747
 748 out_free:
 749         kvm_free_physmem_slot(&new, &old);
 750 out:
 751         return r;
 752
 753 }
 754 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 755
 756 int kvm_set_memory_region(struct kvm *kvm,
 757                           struct kvm_userspace_memory_region *mem,
 758                           int user_alloc)
 759 {
 760         int r;
 761
 762         mutex_lock(&kvm->slots_lock);
 763         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 764         mutex_unlock(&kvm->slots_lock);
 765         return r;
 766 }
 767 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 768
 769 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 770                                    struct
 771                                    kvm_userspace_memory_region *mem,
 772                                    int user_alloc)
 773 {
 774         if (mem->slot >= KVM_MEMORY_SLOTS)
 775                 return -EINVAL;
 776         return kvm_set_memory_region(kvm, mem, user_alloc);
 777 }
 778
 779 int kvm_get_dirty_log(struct kvm *kvm,
 780                         struct kvm_dirty_log *log, int *is_dirty)
 781 {
 782         struct kvm_memory_slot *memslot;
 783         int r, i;
 784         unsigned long n;
 785         unsigned long any = 0;
 786
 787         r = -EINVAL;
 788         if (log->slot >= KVM_MEMORY_SLOTS)
 789                 goto out;
 790
 791         memslot = &kvm->memslots->memslots[log->slot];
 792         r = -ENOENT;
 793         if (!memslot->dirty_bitmap)
 794                 goto out;
 795
 796         n = kvm_dirty_bitmap_bytes(memslot);
 797
 798         for (i = 0; !any && i < n/sizeof(long); ++i)
 799                 any = memslot->dirty_bitmap[i];
 800
 801         r = -EFAULT;
 802         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 803                 goto out;
 804
 805         if (any)
 806                 *is_dirty = 1;
 807
 808         r = 0;
 809 out:
 810         return r;
 811 }
 812
 813 void kvm_disable_largepages(void)
 814 {
 815         largepages_enabled = false;
 816 }
 817 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 818
 819 int is_error_page(struct page *page)
 820 {
 821         return page == bad_page || page == hwpoison_page || page == fault_page;
 822 }
 823 EXPORT_SYMBOL_GPL(is_error_page);
 824
 825 int is_error_pfn(pfn_t pfn)
 826 {
 827         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 828 }
 829 EXPORT_SYMBOL_GPL(is_error_pfn);
 830
 831 int is_hwpoison_pfn(pfn_t pfn)
 832 {
 833         return pfn == hwpoison_pfn;
 834 }
 835 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 836
 837 int is_fault_pfn(pfn_t pfn)
 838 {
 839         return pfn == fault_pfn;
 840 }
 841 EXPORT_SYMBOL_GPL(is_fault_pfn);
 842
 843 static inline unsigned long bad_hva(void)
 844 {
 845         return PAGE_OFFSET;
 846 }
 847
 848 int kvm_is_error_hva(unsigned long addr)
 849 {
 850         return addr == bad_hva();
 851 }
 852 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 853
 854 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 855 {
 856         int i;
 857         struct kvm_memslots *slots = kvm_memslots(kvm);
 858
 859         for (i = 0; i < slots->nmemslots; ++i) {
 860                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 861
 862                 if (gfn >= memslot->base_gfn
 863                     && gfn < memslot->base_gfn + memslot->npages)
 864                         return memslot;
 865         }
 866         return NULL;
 867 }
 868 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 869
 870 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 871 {
 872         int i;
 873         struct kvm_memslots *slots = kvm_memslots(kvm);
 874
 875         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 876                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 877
 878                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 879                         continue;
 880
 881                 if (gfn >= memslot->base_gfn
 882                     && gfn < memslot->base_gfn + memslot->npages)
 883                         return 1;
 884         }
 885         return 0;
 886 }
 887 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 888
 889 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 890 {
 891         struct vm_area_struct *vma;
 892         unsigned long addr, size;
 893
 894         size = PAGE_SIZE;
 895
 896         addr = gfn_to_hva(kvm, gfn);
 897         if (kvm_is_error_hva(addr))
 898                 return PAGE_SIZE;
 899
 900         down_read(&current->mm->mmap_sem);
 901         vma = find_vma(current->mm, addr);
 902         if (!vma)
 903                 goto out;
 904
 905         size = vma_kernel_pagesize(vma);
 906
 907 out:
 908         up_read(&current->mm->mmap_sem);
 909
 910         return size;
 911 }
 912
 913 int memslot_id(struct kvm *kvm, gfn_t gfn)
 914 {
 915         int i;
 916         struct kvm_memslots *slots = kvm_memslots(kvm);
 917         struct kvm_memory_slot *memslot = NULL;
 918
 919         for (i = 0; i < slots->nmemslots; ++i) {
 920                 memslot = &slots->memslots[i];
 921
 922                 if (gfn >= memslot->base_gfn
 923                     && gfn < memslot->base_gfn + memslot->npages)
 924                         break;
 925         }
 926
 927         return memslot - slots->memslots;
 928 }
 929
 930 static unsigned long gfn_to_hva_many(struct kvm *kvm, gfn_t gfn,
 931                                      gfn_t *nr_pages)
 932 {
 933         struct kvm_memory_slot *slot;
 934
 935         slot = gfn_to_memslot(kvm, gfn);
 936         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 937                 return bad_hva();
 938
 939         if (nr_pages)
 940                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 941
 942         return gfn_to_hva_memslot(slot, gfn);
 943 }
 944
 945 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 946 {
 947         return gfn_to_hva_many(kvm, gfn, NULL);
 948 }
 949 EXPORT_SYMBOL_GPL(gfn_to_hva);
 950
 951 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic)
 952 {
 953         struct page *page[1];
 954         int npages;
 955         pfn_t pfn;
 956
 957         if (atomic)
 958                 npages = __get_user_pages_fast(addr, 1, 1, page);
 959         else {
 960                 might_sleep();
 961                 npages = get_user_pages_fast(addr, 1, 1, page);
 962         }
 963
 964         if (unlikely(npages != 1)) {
 965                 struct vm_area_struct *vma;
 966
 967                 if (atomic)
 968                         goto return_fault_page;
 969
 970                 down_read(&current->mm->mmap_sem);
 971                 if (is_hwpoison_address(addr)) {
 972                         up_read(&current->mm->mmap_sem);
 973                         get_page(hwpoison_page);
 974                         return page_to_pfn(hwpoison_page);
 975                 }
 976
 977                 vma = find_vma(current->mm, addr);
 978
 979                 if (vma == NULL || addr < vma->vm_start ||
 980                     !(vma->vm_flags & VM_PFNMAP)) {
 981                         up_read(&current->mm->mmap_sem);
 982 return_fault_page:
 983                         get_page(fault_page);
 984                         return page_to_pfn(fault_page);
 985                 }
 986
 987                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 988                 up_read(&current->mm->mmap_sem);
 989                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 990         } else
 991                 pfn = page_to_pfn(page[0]);
 992
 993         return pfn;
 994 }
 995
 996 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
 997 {
 998         return hva_to_pfn(kvm, addr, true);
 999 }
1000 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1001
1002 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1003 {
1004         unsigned long addr;
1005
1006         addr = gfn_to_hva(kvm, gfn);
1007         if (kvm_is_error_hva(addr)) {
1008                 get_page(bad_page);
1009                 return page_to_pfn(bad_page);
1010         }
1011
1012         return hva_to_pfn(kvm, addr, false);
1013 }
1014 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1015
1016 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1017                          struct kvm_memory_slot *slot, gfn_t gfn)
1018 {
1019         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1020         return hva_to_pfn(kvm, addr, false);
1021 }
1022
1023 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1024                                                                   int nr_pages)
1025 {
1026         unsigned long addr;
1027         gfn_t entry;
1028
1029         addr = gfn_to_hva_many(kvm, gfn, &entry);
1030         if (kvm_is_error_hva(addr))
1031                 return -1;
1032
1033         if (entry < nr_pages)
1034                 return 0;
1035
1036         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1037 }
1038 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1039
1040 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1041 {
1042         pfn_t pfn;
1043
1044         pfn = gfn_to_pfn(kvm, gfn);
1045         if (!kvm_is_mmio_pfn(pfn))
1046                 return pfn_to_page(pfn);
1047
1048         WARN_ON(kvm_is_mmio_pfn(pfn));
1049
1050         get_page(bad_page);
1051         return bad_page;
1052 }
1053
1054 EXPORT_SYMBOL_GPL(gfn_to_page);
1055
1056 void kvm_release_page_clean(struct page *page)
1057 {
1058         kvm_release_pfn_clean(page_to_pfn(page));
1059 }
1060 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1061
1062 void kvm_release_pfn_clean(pfn_t pfn)
1063 {
1064         if (!kvm_is_mmio_pfn(pfn))
1065                 put_page(pfn_to_page(pfn));
1066 }
1067 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1068
1069 void kvm_release_page_dirty(struct page *page)
1070 {
1071         kvm_release_pfn_dirty(page_to_pfn(page));
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1074
1075 void kvm_release_pfn_dirty(pfn_t pfn)
1076 {
1077         kvm_set_pfn_dirty(pfn);
1078         kvm_release_pfn_clean(pfn);
1079 }
1080 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1081
1082 void kvm_set_page_dirty(struct page *page)
1083 {
1084         kvm_set_pfn_dirty(page_to_pfn(page));
1085 }
1086 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1087
1088 void kvm_set_pfn_dirty(pfn_t pfn)
1089 {
1090         if (!kvm_is_mmio_pfn(pfn)) {
1091                 struct page *page = pfn_to_page(pfn);
1092                 if (!PageReserved(page))
1093                         SetPageDirty(page);
1094         }
1095 }
1096 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1097
1098 void kvm_set_pfn_accessed(pfn_t pfn)
1099 {
1100         if (!kvm_is_mmio_pfn(pfn))
1101                 mark_page_accessed(pfn_to_page(pfn));
1102 }
1103 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1104
1105 void kvm_get_pfn(pfn_t pfn)
1106 {
1107         if (!kvm_is_mmio_pfn(pfn))
1108                 get_page(pfn_to_page(pfn));
1109 }
1110 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1111
1112 static int next_segment(unsigned long len, int offset)
1113 {
1114         if (len > PAGE_SIZE - offset)
1115                 return PAGE_SIZE - offset;
1116         else
1117                 return len;
1118 }
1119
1120 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1121                         int len)
1122 {
1123         int r;
1124         unsigned long addr;
1125
1126         addr = gfn_to_hva(kvm, gfn);
1127         if (kvm_is_error_hva(addr))
1128                 return -EFAULT;
1129         r = copy_from_user(data, (void __user *)addr + offset, len);
1130         if (r)
1131                 return -EFAULT;
1132         return 0;
1133 }
1134 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1135
1136 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1137 {
1138         gfn_t gfn = gpa >> PAGE_SHIFT;
1139         int seg;
1140         int offset = offset_in_page(gpa);
1141         int ret;
1142
1143         while ((seg = next_segment(len, offset)) != 0) {
1144                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1145                 if (ret < 0)
1146                         return ret;
1147                 offset = 0;
1148                 len -= seg;
1149                 data += seg;
1150                 ++gfn;
1151         }
1152         return 0;
1153 }
1154 EXPORT_SYMBOL_GPL(kvm_read_guest);
1155
1156 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1157                           unsigned long len)
1158 {
1159         int r;
1160         unsigned long addr;
1161         gfn_t gfn = gpa >> PAGE_SHIFT;
1162         int offset = offset_in_page(gpa);
1163
1164         addr = gfn_to_hva(kvm, gfn);
1165         if (kvm_is_error_hva(addr))
1166                 return -EFAULT;
1167         pagefault_disable();
1168         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1169         pagefault_enable();
1170         if (r)
1171                 return -EFAULT;
1172         return 0;
1173 }
1174 EXPORT_SYMBOL(kvm_read_guest_atomic);
1175
1176 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1177                          int offset, int len)
1178 {
1179         int r;
1180         unsigned long addr;
1181
1182         addr = gfn_to_hva(kvm, gfn);
1183         if (kvm_is_error_hva(addr))
1184                 return -EFAULT;
1185         r = copy_to_user((void __user *)addr + offset, data, len);
1186         if (r)
1187                 return -EFAULT;
1188         mark_page_dirty(kvm, gfn);
1189         return 0;
1190 }
1191 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1192
1193 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1194                     unsigned long len)
1195 {
1196         gfn_t gfn = gpa >> PAGE_SHIFT;
1197         int seg;
1198         int offset = offset_in_page(gpa);
1199         int ret;
1200
1201         while ((seg = next_segment(len, offset)) != 0) {
1202                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1203                 if (ret < 0)
1204                         return ret;
1205                 offset = 0;
1206                 len -= seg;
1207                 data += seg;
1208                 ++gfn;
1209         }
1210         return 0;
1211 }
1212
1213 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1214 {
1215         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1218
1219 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1220 {
1221         gfn_t gfn = gpa >> PAGE_SHIFT;
1222         int seg;
1223         int offset = offset_in_page(gpa);
1224         int ret;
1225
1226         while ((seg = next_segment(len, offset)) != 0) {
1227                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1228                 if (ret < 0)
1229                         return ret;
1230                 offset = 0;
1231                 len -= seg;
1232                 ++gfn;
1233         }
1234         return 0;
1235 }
1236 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1237
1238 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1239 {
1240         struct kvm_memory_slot *memslot;
1241
1242         memslot = gfn_to_memslot(kvm, gfn);
1243         if (memslot && memslot->dirty_bitmap) {
1244                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1245
1246                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1247         }
1248 }
1249
1250 /*
1251  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1252  */
1253 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1254 {
1255         DEFINE_WAIT(wait);
1256
1257         for (;;) {
1258                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1259
1260                 if (kvm_arch_vcpu_runnable(vcpu)) {
1261                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1262                         break;
1263                 }
1264                 if (kvm_cpu_has_pending_timer(vcpu))
1265                         break;
1266                 if (signal_pending(current))
1267                         break;
1268
1269                 schedule();
1270         }
1271
1272         finish_wait(&vcpu->wq, &wait);
1273 }
1274
1275 void kvm_resched(struct kvm_vcpu *vcpu)
1276 {
1277         if (!need_resched())
1278                 return;
1279         cond_resched();
1280 }
1281 EXPORT_SYMBOL_GPL(kvm_resched);
1282
1283 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1284 {
1285         ktime_t expires;
1286         DEFINE_WAIT(wait);
1287
1288         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1289
1290         /* Sleep for 100 us, and hope lock-holder got scheduled */
1291         expires = ktime_add_ns(ktime_get(), 100000UL);
1292         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1293
1294         finish_wait(&vcpu->wq, &wait);
1295 }
1296 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1297
1298 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1299 {
1300         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1301         struct page *page;
1302
1303         if (vmf->pgoff == 0)
1304                 page = virt_to_page(vcpu->run);
1305 #ifdef CONFIG_X86
1306         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1307                 page = virt_to_page(vcpu->arch.pio_data);
1308 #endif
1309 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1310         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1311                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1312 #endif
1313         else
1314                 return VM_FAULT_SIGBUS;
1315         get_page(page);
1316         vmf->page = page;
1317         return 0;
1318 }
1319
1320 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1321         .fault = kvm_vcpu_fault,
1322 };
1323
1324 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1325 {
1326         vma->vm_ops = &kvm_vcpu_vm_ops;
1327         return 0;
1328 }
1329
1330 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1331 {
1332         struct kvm_vcpu *vcpu = filp->private_data;
1333
1334         kvm_put_kvm(vcpu->kvm);
1335         return 0;
1336 }
1337
1338 static struct file_operations kvm_vcpu_fops = {
1339         .release        = kvm_vcpu_release,
1340         .unlocked_ioctl = kvm_vcpu_ioctl,
1341         .compat_ioctl   = kvm_vcpu_ioctl,
1342         .mmap           = kvm_vcpu_mmap,
1343 };
1344
1345 /*
1346  * Allocates an inode for the vcpu.
1347  */
1348 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1349 {
1350         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1351 }
1352
1353 /*
1354  * Creates some virtual cpus.  Good luck creating more than one.
1355  */
1356 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1357 {
1358         int r;
1359         struct kvm_vcpu *vcpu, *v;
1360
1361         vcpu = kvm_arch_vcpu_create(kvm, id);
1362         if (IS_ERR(vcpu))
1363                 return PTR_ERR(vcpu);
1364
1365         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1366
1367         r = kvm_arch_vcpu_setup(vcpu);
1368         if (r)
1369                 return r;
1370
1371         mutex_lock(&kvm->lock);
1372         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1373                 r = -EINVAL;
1374                 goto vcpu_destroy;
1375         }
1376
1377         kvm_for_each_vcpu(r, v, kvm)
1378                 if (v->vcpu_id == id) {
1379                         r = -EEXIST;
1380                         goto vcpu_destroy;
1381                 }
1382
1383         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1384
1385         /* Now it's all set up, let userspace reach it */
1386         kvm_get_kvm(kvm);
1387         r = create_vcpu_fd(vcpu);
1388         if (r < 0) {
1389                 kvm_put_kvm(kvm);
1390                 goto vcpu_destroy;
1391         }
1392
1393         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1394         smp_wmb();
1395         atomic_inc(&kvm->online_vcpus);
1396
1397 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1398         if (kvm->bsp_vcpu_id == id)
1399                 kvm->bsp_vcpu = vcpu;
1400 #endif
1401         mutex_unlock(&kvm->lock);
1402         return r;
1403
1404 vcpu_destroy:
1405         mutex_unlock(&kvm->lock);
1406         kvm_arch_vcpu_destroy(vcpu);
1407         return r;
1408 }
1409
1410 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1411 {
1412         if (sigset) {
1413                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1414                 vcpu->sigset_active = 1;
1415                 vcpu->sigset = *sigset;
1416         } else
1417                 vcpu->sigset_active = 0;
1418         return 0;
1419 }
1420
1421 static long kvm_vcpu_ioctl(struct file *filp,
1422                            unsigned int ioctl, unsigned long arg)
1423 {
1424         struct kvm_vcpu *vcpu = filp->private_data;
1425         void __user *argp = (void __user *)arg;
1426         int r;
1427         struct kvm_fpu *fpu = NULL;
1428         struct kvm_sregs *kvm_sregs = NULL;
1429
1430         if (vcpu->kvm->mm != current->mm)
1431                 return -EIO;
1432
1433 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1434         /*
1435          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1436          * so vcpu_load() would break it.
1437          */
1438         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1439                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1440 #endif
1441
1442
1443         vcpu_load(vcpu);
1444         switch (ioctl) {
1445         case KVM_RUN:
1446                 r = -EINVAL;
1447                 if (arg)
1448                         goto out;
1449                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1450                 break;
1451         case KVM_GET_REGS: {
1452                 struct kvm_regs *kvm_regs;
1453
1454                 r = -ENOMEM;
1455                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1456                 if (!kvm_regs)
1457                         goto out;
1458                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1459                 if (r)
1460                         goto out_free1;
1461                 r = -EFAULT;
1462                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1463                         goto out_free1;
1464                 r = 0;
1465 out_free1:
1466                 kfree(kvm_regs);
1467                 break;
1468         }
1469         case KVM_SET_REGS: {
1470                 struct kvm_regs *kvm_regs;
1471
1472                 r = -ENOMEM;
1473                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1474                 if (!kvm_regs)
1475                         goto out;
1476                 r = -EFAULT;
1477                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1478                         goto out_free2;
1479                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1480                 if (r)
1481                         goto out_free2;
1482                 r = 0;
1483 out_free2:
1484                 kfree(kvm_regs);
1485                 break;
1486         }
1487         case KVM_GET_SREGS: {
1488                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1489                 r = -ENOMEM;
1490                 if (!kvm_sregs)
1491                         goto out;
1492                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1493                 if (r)
1494                         goto out;
1495                 r = -EFAULT;
1496                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1497                         goto out;
1498                 r = 0;
1499                 break;
1500         }
1501         case KVM_SET_SREGS: {
1502                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1503                 r = -ENOMEM;
1504                 if (!kvm_sregs)
1505                         goto out;
1506                 r = -EFAULT;
1507                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1508                         goto out;
1509                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1510                 if (r)
1511                         goto out;
1512                 r = 0;
1513                 break;
1514         }
1515         case KVM_GET_MP_STATE: {
1516                 struct kvm_mp_state mp_state;
1517
1518                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1519                 if (r)
1520                         goto out;
1521                 r = -EFAULT;
1522                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1523                         goto out;
1524                 r = 0;
1525                 break;
1526         }
1527         case KVM_SET_MP_STATE: {
1528                 struct kvm_mp_state mp_state;
1529
1530                 r = -EFAULT;
1531                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1532                         goto out;
1533                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1534                 if (r)
1535                         goto out;
1536                 r = 0;
1537                 break;
1538         }
1539         case KVM_TRANSLATE: {
1540                 struct kvm_translation tr;
1541
1542                 r = -EFAULT;
1543                 if (copy_from_user(&tr, argp, sizeof tr))
1544                         goto out;
1545                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1546                 if (r)
1547                         goto out;
1548                 r = -EFAULT;
1549                 if (copy_to_user(argp, &tr, sizeof tr))
1550                         goto out;
1551                 r = 0;
1552                 break;
1553         }
1554         case KVM_SET_GUEST_DEBUG: {
1555                 struct kvm_guest_debug dbg;
1556
1557                 r = -EFAULT;
1558                 if (copy_from_user(&dbg, argp, sizeof dbg))
1559                         goto out;
1560                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1561                 if (r)
1562                         goto out;
1563                 r = 0;
1564                 break;
1565         }
1566         case KVM_SET_SIGNAL_MASK: {
1567                 struct kvm_signal_mask __user *sigmask_arg = argp;
1568                 struct kvm_signal_mask kvm_sigmask;
1569                 sigset_t sigset, *p;
1570
1571                 p = NULL;
1572                 if (argp) {
1573                         r = -EFAULT;
1574                         if (copy_from_user(&kvm_sigmask, argp,
1575                                            sizeof kvm_sigmask))
1576                                 goto out;
1577                         r = -EINVAL;
1578                         if (kvm_sigmask.len != sizeof sigset)
1579                                 goto out;
1580                         r = -EFAULT;
1581                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1582                                            sizeof sigset))
1583                                 goto out;
1584                         p = &sigset;
1585                 }
1586                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1587                 break;
1588         }
1589         case KVM_GET_FPU: {
1590                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1591                 r = -ENOMEM;
1592                 if (!fpu)
1593                         goto out;
1594                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1595                 if (r)
1596                         goto out;
1597                 r = -EFAULT;
1598                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1599                         goto out;
1600                 r = 0;
1601                 break;
1602         }
1603         case KVM_SET_FPU: {
1604                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1605                 r = -ENOMEM;
1606                 if (!fpu)
1607                         goto out;
1608                 r = -EFAULT;
1609                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1610                         goto out;
1611                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1612                 if (r)
1613                         goto out;
1614                 r = 0;
1615                 break;
1616         }
1617         default:
1618                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1619         }
1620 out:
1621         vcpu_put(vcpu);
1622         kfree(fpu);
1623         kfree(kvm_sregs);
1624         return r;
1625 }
1626
1627 static long kvm_vm_ioctl(struct file *filp,
1628                            unsigned int ioctl, unsigned long arg)
1629 {
1630         struct kvm *kvm = filp->private_data;
1631         void __user *argp = (void __user *)arg;
1632         int r;
1633
1634         if (kvm->mm != current->mm)
1635                 return -EIO;
1636         switch (ioctl) {
1637         case KVM_CREATE_VCPU:
1638                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1639                 if (r < 0)
1640                         goto out;
1641                 break;
1642         case KVM_SET_USER_MEMORY_REGION: {
1643                 struct kvm_userspace_memory_region kvm_userspace_mem;
1644
1645                 r = -EFAULT;
1646                 if (copy_from_user(&kvm_userspace_mem, argp,
1647                                                 sizeof kvm_userspace_mem))
1648                         goto out;
1649
1650                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1651                 if (r)
1652                         goto out;
1653                 break;
1654         }
1655         case KVM_GET_DIRTY_LOG: {
1656                 struct kvm_dirty_log log;
1657
1658                 r = -EFAULT;
1659                 if (copy_from_user(&log, argp, sizeof log))
1660                         goto out;
1661                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1662                 if (r)
1663                         goto out;
1664                 break;
1665         }
1666 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1667         case KVM_REGISTER_COALESCED_MMIO: {
1668                 struct kvm_coalesced_mmio_zone zone;
1669                 r = -EFAULT;
1670                 if (copy_from_user(&zone, argp, sizeof zone))
1671                         goto out;
1672                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1673                 if (r)
1674                         goto out;
1675                 r = 0;
1676                 break;
1677         }
1678         case KVM_UNREGISTER_COALESCED_MMIO: {
1679                 struct kvm_coalesced_mmio_zone zone;
1680                 r = -EFAULT;
1681                 if (copy_from_user(&zone, argp, sizeof zone))
1682                         goto out;
1683                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1684                 if (r)
1685                         goto out;
1686                 r = 0;
1687                 break;
1688         }
1689 #endif
1690         case KVM_IRQFD: {
1691                 struct kvm_irqfd data;
1692
1693                 r = -EFAULT;
1694                 if (copy_from_user(&data, argp, sizeof data))
1695                         goto out;
1696                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1697                 break;
1698         }
1699         case KVM_IOEVENTFD: {
1700                 struct kvm_ioeventfd data;
1701
1702                 r = -EFAULT;
1703                 if (copy_from_user(&data, argp, sizeof data))
1704                         goto out;
1705                 r = kvm_ioeventfd(kvm, &data);
1706                 break;
1707         }
1708 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1709         case KVM_SET_BOOT_CPU_ID:
1710                 r = 0;
1711                 mutex_lock(&kvm->lock);
1712                 if (atomic_read(&kvm->online_vcpus) != 0)
1713                         r = -EBUSY;
1714                 else
1715                         kvm->bsp_vcpu_id = arg;
1716                 mutex_unlock(&kvm->lock);
1717                 break;
1718 #endif
1719         default:
1720                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1721                 if (r == -ENOTTY)
1722                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1723         }
1724 out:
1725         return r;
1726 }
1727
1728 #ifdef CONFIG_COMPAT
1729 struct compat_kvm_dirty_log {
1730         __u32 slot;
1731         __u32 padding1;
1732         union {
1733                 compat_uptr_t dirty_bitmap; /* one bit per page */
1734                 __u64 padding2;
1735         };
1736 };
1737
1738 static long kvm_vm_compat_ioctl(struct file *filp,
1739                            unsigned int ioctl, unsigned long arg)
1740 {
1741         struct kvm *kvm = filp->private_data;
1742         int r;
1743
1744         if (kvm->mm != current->mm)
1745                 return -EIO;
1746         switch (ioctl) {
1747         case KVM_GET_DIRTY_LOG: {
1748                 struct compat_kvm_dirty_log compat_log;
1749                 struct kvm_dirty_log log;
1750
1751                 r = -EFAULT;
1752                 if (copy_from_user(&compat_log, (void __user *)arg,
1753                                    sizeof(compat_log)))
1754                         goto out;
1755                 log.slot         = compat_log.slot;
1756                 log.padding1     = compat_log.padding1;
1757                 log.padding2     = compat_log.padding2;
1758                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1759
1760                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1761                 if (r)
1762                         goto out;
1763                 break;
1764         }
1765         default:
1766                 r = kvm_vm_ioctl(filp, ioctl, arg);
1767         }
1768
1769 out:
1770         return r;
1771 }
1772 #endif
1773
1774 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1775 {
1776         struct page *page[1];
1777         unsigned long addr;
1778         int npages;
1779         gfn_t gfn = vmf->pgoff;
1780         struct kvm *kvm = vma->vm_file->private_data;
1781
1782         addr = gfn_to_hva(kvm, gfn);
1783         if (kvm_is_error_hva(addr))
1784                 return VM_FAULT_SIGBUS;
1785
1786         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1787                                 NULL);
1788         if (unlikely(npages != 1))
1789                 return VM_FAULT_SIGBUS;
1790
1791         vmf->page = page[0];
1792         return 0;
1793 }
1794
1795 static const struct vm_operations_struct kvm_vm_vm_ops = {
1796         .fault = kvm_vm_fault,
1797 };
1798
1799 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1800 {
1801         vma->vm_ops = &kvm_vm_vm_ops;
1802         return 0;
1803 }
1804
1805 static struct file_operations kvm_vm_fops = {
1806         .release        = kvm_vm_release,
1807         .unlocked_ioctl = kvm_vm_ioctl,
1808 #ifdef CONFIG_COMPAT
1809         .compat_ioctl   = kvm_vm_compat_ioctl,
1810 #endif
1811         .mmap           = kvm_vm_mmap,
1812 };
1813
1814 static int kvm_dev_ioctl_create_vm(void)
1815 {
1816         int fd, r;
1817         struct kvm *kvm;
1818
1819         kvm = kvm_create_vm();
1820         if (IS_ERR(kvm))
1821                 return PTR_ERR(kvm);
1822 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1823         r = kvm_coalesced_mmio_init(kvm);
1824         if (r < 0) {
1825                 kvm_put_kvm(kvm);
1826                 return r;
1827         }
1828 #endif
1829         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1830         if (fd < 0)
1831                 kvm_put_kvm(kvm);
1832
1833         return fd;
1834 }
1835
1836 static long kvm_dev_ioctl_check_extension_generic(long arg)
1837 {
1838         switch (arg) {
1839         case KVM_CAP_USER_MEMORY:
1840         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1841         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1842 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1843         case KVM_CAP_SET_BOOT_CPU_ID:
1844 #endif
1845         case KVM_CAP_INTERNAL_ERROR_DATA:
1846                 return 1;
1847 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1848         case KVM_CAP_IRQ_ROUTING:
1849                 return KVM_MAX_IRQ_ROUTES;
1850 #endif
1851         default:
1852                 break;
1853         }
1854         return kvm_dev_ioctl_check_extension(arg);
1855 }
1856
1857 static long kvm_dev_ioctl(struct file *filp,
1858                           unsigned int ioctl, unsigned long arg)
1859 {
1860         long r = -EINVAL;
1861
1862         switch (ioctl) {
1863         case KVM_GET_API_VERSION:
1864                 r = -EINVAL;
1865                 if (arg)
1866                         goto out;
1867                 r = KVM_API_VERSION;
1868                 break;
1869         case KVM_CREATE_VM:
1870                 r = -EINVAL;
1871                 if (arg)
1872                         goto out;
1873                 r = kvm_dev_ioctl_create_vm();
1874                 break;
1875         case KVM_CHECK_EXTENSION:
1876                 r = kvm_dev_ioctl_check_extension_generic(arg);
1877                 break;
1878         case KVM_GET_VCPU_MMAP_SIZE:
1879                 r = -EINVAL;
1880                 if (arg)
1881                         goto out;
1882                 r = PAGE_SIZE;     /* struct kvm_run */
1883 #ifdef CONFIG_X86
1884                 r += PAGE_SIZE;    /* pio data page */
1885 #endif
1886 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1887                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1888 #endif
1889                 break;
1890         case KVM_TRACE_ENABLE:
1891         case KVM_TRACE_PAUSE:
1892         case KVM_TRACE_DISABLE:
1893                 r = -EOPNOTSUPP;
1894                 break;
1895         default:
1896                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1897         }
1898 out:
1899         return r;
1900 }
1901
1902 static struct file_operations kvm_chardev_ops = {
1903         .unlocked_ioctl = kvm_dev_ioctl,
1904         .compat_ioctl   = kvm_dev_ioctl,
1905 };
1906
1907 static struct miscdevice kvm_dev = {
1908         KVM_MINOR,
1909         "kvm",
1910         &kvm_chardev_ops,
1911 };
1912
1913 static void hardware_enable(void *junk)
1914 {
1915         int cpu = raw_smp_processor_id();
1916         int r;
1917
1918         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1919                 return;
1920
1921         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1922
1923         r = kvm_arch_hardware_enable(NULL);
1924
1925         if (r) {
1926                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1927                 atomic_inc(&hardware_enable_failed);
1928                 printk(KERN_INFO "kvm: enabling virtualization on "
1929                                  "CPU%d failed\n", cpu);
1930         }
1931 }
1932
1933 static void hardware_disable(void *junk)
1934 {
1935         int cpu = raw_smp_processor_id();
1936
1937         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1938                 return;
1939         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1940         kvm_arch_hardware_disable(NULL);
1941 }
1942
1943 static void hardware_disable_all_nolock(void)
1944 {
1945         BUG_ON(!kvm_usage_count);
1946
1947         kvm_usage_count--;
1948         if (!kvm_usage_count)
1949                 on_each_cpu(hardware_disable, NULL, 1);
1950 }
1951
1952 static void hardware_disable_all(void)
1953 {
1954         spin_lock(&kvm_lock);
1955         hardware_disable_all_nolock();
1956         spin_unlock(&kvm_lock);
1957 }
1958
1959 static int hardware_enable_all(void)
1960 {
1961         int r = 0;
1962
1963         spin_lock(&kvm_lock);
1964
1965         kvm_usage_count++;
1966         if (kvm_usage_count == 1) {
1967                 atomic_set(&hardware_enable_failed, 0);
1968                 on_each_cpu(hardware_enable, NULL, 1);
1969
1970                 if (atomic_read(&hardware_enable_failed)) {
1971                         hardware_disable_all_nolock();
1972                         r = -EBUSY;
1973                 }
1974         }
1975
1976         spin_unlock(&kvm_lock);
1977
1978         return r;
1979 }
1980
1981 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1982                            void *v)
1983 {
1984         int cpu = (long)v;
1985
1986         if (!kvm_usage_count)
1987                 return NOTIFY_OK;
1988
1989         val &= ~CPU_TASKS_FROZEN;
1990         switch (val) {
1991         case CPU_DYING:
1992                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1993                        cpu);
1994                 hardware_disable(NULL);
1995                 break;
1996         case CPU_STARTING:
1997                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1998                        cpu);
1999                 spin_lock(&kvm_lock);
2000                 hardware_enable(NULL);
2001                 spin_unlock(&kvm_lock);
2002                 break;
2003         }
2004         return NOTIFY_OK;
2005 }
2006
2007
2008 asmlinkage void kvm_handle_fault_on_reboot(void)
2009 {
2010         if (kvm_rebooting) {
2011                 /* spin while reset goes on */
2012                 local_irq_enable();
2013                 while (true)
2014                         ;
2015         }
2016         /* Fault while not rebooting.  We want the trace. */
2017         BUG();
2018 }
2019 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2020
2021 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2022                       void *v)
2023 {
2024         /*
2025          * Some (well, at least mine) BIOSes hang on reboot if
2026          * in vmx root mode.
2027          *
2028          * And Intel TXT required VMX off for all cpu when system shutdown.
2029          */
2030         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2031         kvm_rebooting = true;
2032         on_each_cpu(hardware_disable, NULL, 1);
2033         return NOTIFY_OK;
2034 }
2035
2036 static struct notifier_block kvm_reboot_notifier = {
2037         .notifier_call = kvm_reboot,
2038         .priority = 0,
2039 };
2040
2041 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2042 {
2043         int i;
2044
2045         for (i = 0; i < bus->dev_count; i++) {
2046                 struct kvm_io_device *pos = bus->devs[i];
2047
2048                 kvm_iodevice_destructor(pos);
2049         }
2050         kfree(bus);
2051 }
2052
2053 /* kvm_io_bus_write - called under kvm->slots_lock */
2054 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2055                      int len, const void *val)
2056 {
2057         int i;
2058         struct kvm_io_bus *bus;
2059
2060         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2061         for (i = 0; i < bus->dev_count; i++)
2062                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2063                         return 0;
2064         return -EOPNOTSUPP;
2065 }
2066
2067 /* kvm_io_bus_read - called under kvm->slots_lock */
2068 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2069                     int len, void *val)
2070 {
2071         int i;
2072         struct kvm_io_bus *bus;
2073
2074         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2075         for (i = 0; i < bus->dev_count; i++)
2076                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2077                         return 0;
2078         return -EOPNOTSUPP;
2079 }
2080
2081 /* Caller must hold slots_lock. */
2082 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2083                             struct kvm_io_device *dev)
2084 {
2085         struct kvm_io_bus *new_bus, *bus;
2086
2087         bus = kvm->buses[bus_idx];
2088         if (bus->dev_count > NR_IOBUS_DEVS-1)
2089                 return -ENOSPC;
2090
2091         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2092         if (!new_bus)
2093                 return -ENOMEM;
2094         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2095         new_bus->devs[new_bus->dev_count++] = dev;
2096         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2097         synchronize_srcu_expedited(&kvm->srcu);
2098         kfree(bus);
2099
2100         return 0;
2101 }
2102
2103 /* Caller must hold slots_lock. */
2104 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2105                               struct kvm_io_device *dev)
2106 {
2107         int i, r;
2108         struct kvm_io_bus *new_bus, *bus;
2109
2110         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2111         if (!new_bus)
2112                 return -ENOMEM;
2113
2114         bus = kvm->buses[bus_idx];
2115         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2116
2117         r = -ENOENT;
2118         for (i = 0; i < new_bus->dev_count; i++)
2119                 if (new_bus->devs[i] == dev) {
2120                         r = 0;
2121                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2122                         break;
2123                 }
2124
2125         if (r) {
2126                 kfree(new_bus);
2127                 return r;
2128         }
2129
2130         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2131         synchronize_srcu_expedited(&kvm->srcu);
2132         kfree(bus);
2133         return r;
2134 }
2135
2136 static struct notifier_block kvm_cpu_notifier = {
2137         .notifier_call = kvm_cpu_hotplug,
2138 };
2139
2140 static int vm_stat_get(void *_offset, u64 *val)
2141 {
2142         unsigned offset = (long)_offset;
2143         struct kvm *kvm;
2144
2145         *val = 0;
2146         spin_lock(&kvm_lock);
2147         list_for_each_entry(kvm, &vm_list, vm_list)
2148                 *val += *(u32 *)((void *)kvm + offset);
2149         spin_unlock(&kvm_lock);
2150         return 0;
2151 }
2152
2153 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2154
2155 static int vcpu_stat_get(void *_offset, u64 *val)
2156 {
2157         unsigned offset = (long)_offset;
2158         struct kvm *kvm;
2159         struct kvm_vcpu *vcpu;
2160         int i;
2161
2162         *val = 0;
2163         spin_lock(&kvm_lock);
2164         list_for_each_entry(kvm, &vm_list, vm_list)
2165                 kvm_for_each_vcpu(i, vcpu, kvm)
2166                         *val += *(u32 *)((void *)vcpu + offset);
2167
2168         spin_unlock(&kvm_lock);
2169         return 0;
2170 }
2171
2172 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2173
2174 static const struct file_operations *stat_fops[] = {
2175         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2176         [KVM_STAT_VM]   = &vm_stat_fops,
2177 };
2178
2179 static void kvm_init_debug(void)
2180 {
2181         struct kvm_stats_debugfs_item *p;
2182
2183         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2184         for (p = debugfs_entries; p->name; ++p)
2185                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2186                                                 (void *)(long)p->offset,
2187                                                 stat_fops[p->kind]);
2188 }
2189
2190 static void kvm_exit_debug(void)
2191 {
2192         struct kvm_stats_debugfs_item *p;
2193
2194         for (p = debugfs_entries; p->name; ++p)
2195                 debugfs_remove(p->dentry);
2196         debugfs_remove(kvm_debugfs_dir);
2197 }
2198
2199 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2200 {
2201         if (kvm_usage_count)
2202                 hardware_disable(NULL);
2203         return 0;
2204 }
2205
2206 static int kvm_resume(struct sys_device *dev)
2207 {
2208         if (kvm_usage_count) {
2209                 WARN_ON(spin_is_locked(&kvm_lock));
2210                 hardware_enable(NULL);
2211         }
2212         return 0;
2213 }
2214
2215 static struct sysdev_class kvm_sysdev_class = {
2216         .name = "kvm",
2217         .suspend = kvm_suspend,
2218         .resume = kvm_resume,
2219 };
2220
2221 static struct sys_device kvm_sysdev = {
2222         .id = 0,
2223         .cls = &kvm_sysdev_class,
2224 };
2225
2226 struct page *bad_page;
2227 pfn_t bad_pfn;
2228
2229 static inline
2230 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2231 {
2232         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2233 }
2234
2235 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2236 {
2237         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2238
2239         kvm_arch_vcpu_load(vcpu, cpu);
2240 }
2241
2242 static void kvm_sched_out(struct preempt_notifier *pn,
2243                           struct task_struct *next)
2244 {
2245         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2246
2247         kvm_arch_vcpu_put(vcpu);
2248 }
2249
2250 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2251                   struct module *module)
2252 {
2253         int r;
2254         int cpu;
2255
2256         r = kvm_arch_init(opaque);
2257         if (r)
2258                 goto out_fail;
2259
2260         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2261
2262         if (bad_page == NULL) {
2263                 r = -ENOMEM;
2264                 goto out;
2265         }
2266
2267         bad_pfn = page_to_pfn(bad_page);
2268
2269         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2270
2271         if (hwpoison_page == NULL) {
2272                 r = -ENOMEM;
2273                 goto out_free_0;
2274         }
2275
2276         hwpoison_pfn = page_to_pfn(hwpoison_page);
2277
2278         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2279
2280         if (fault_page == NULL) {
2281                 r = -ENOMEM;
2282                 goto out_free_0;
2283         }
2284
2285         fault_pfn = page_to_pfn(fault_page);
2286
2287         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2288                 r = -ENOMEM;
2289                 goto out_free_0;
2290         }
2291
2292         r = kvm_arch_hardware_setup();
2293         if (r < 0)
2294                 goto out_free_0a;
2295
2296         for_each_online_cpu(cpu) {
2297                 smp_call_function_single(cpu,
2298                                 kvm_arch_check_processor_compat,
2299                                 &r, 1);
2300                 if (r < 0)
2301                         goto out_free_1;
2302         }
2303
2304         r = register_cpu_notifier(&kvm_cpu_notifier);
2305         if (r)
2306                 goto out_free_2;
2307         register_reboot_notifier(&kvm_reboot_notifier);
2308
2309         r = sysdev_class_register(&kvm_sysdev_class);
2310         if (r)
2311                 goto out_free_3;
2312
2313         r = sysdev_register(&kvm_sysdev);
2314         if (r)
2315                 goto out_free_4;
2316
2317         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2318         if (!vcpu_align)
2319                 vcpu_align = __alignof__(struct kvm_vcpu);
2320         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2321                                            0, NULL);
2322         if (!kvm_vcpu_cache) {
2323                 r = -ENOMEM;
2324                 goto out_free_5;
2325         }
2326
2327         kvm_chardev_ops.owner = module;
2328         kvm_vm_fops.owner = module;
2329         kvm_vcpu_fops.owner = module;
2330
2331         r = misc_register(&kvm_dev);
2332         if (r) {
2333                 printk(KERN_ERR "kvm: misc device register failed\n");
2334                 goto out_free;
2335         }
2336
2337         kvm_preempt_ops.sched_in = kvm_sched_in;
2338         kvm_preempt_ops.sched_out = kvm_sched_out;
2339
2340         kvm_init_debug();
2341
2342         return 0;
2343
2344 out_free:
2345         kmem_cache_destroy(kvm_vcpu_cache);
2346 out_free_5:
2347         sysdev_unregister(&kvm_sysdev);
2348 out_free_4:
2349         sysdev_class_unregister(&kvm_sysdev_class);
2350 out_free_3:
2351         unregister_reboot_notifier(&kvm_reboot_notifier);
2352         unregister_cpu_notifier(&kvm_cpu_notifier);
2353 out_free_2:
2354 out_free_1:
2355         kvm_arch_hardware_unsetup();
2356 out_free_0a:
2357         free_cpumask_var(cpus_hardware_enabled);
2358 out_free_0:
2359         if (fault_page)
2360                 __free_page(fault_page);
2361         if (hwpoison_page)
2362                 __free_page(hwpoison_page);
2363         __free_page(bad_page);
2364 out:
2365         kvm_arch_exit();
2366 out_fail:
2367         return r;
2368 }
2369 EXPORT_SYMBOL_GPL(kvm_init);
2370
2371 void kvm_exit(void)
2372 {
2373         kvm_exit_debug();
2374         misc_deregister(&kvm_dev);
2375         kmem_cache_destroy(kvm_vcpu_cache);
2376         sysdev_unregister(&kvm_sysdev);
2377         sysdev_class_unregister(&kvm_sysdev_class);
2378         unregister_reboot_notifier(&kvm_reboot_notifier);
2379         unregister_cpu_notifier(&kvm_cpu_notifier);
2380         on_each_cpu(hardware_disable, NULL, 1);
2381         kvm_arch_hardware_unsetup();
2382         kvm_arch_exit();
2383         free_cpumask_var(cpus_hardware_enabled);
2384         __free_page(hwpoison_page);
2385         __free_page(bad_page);
2386 }
2387 EXPORT_SYMBOL_GPL(kvm_exit);