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_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 931 {
 932         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
 933 }
 934
 935 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 936 {
 937         struct kvm_memory_slot *slot;
 938
 939         slot = gfn_to_memslot(kvm, gfn);
 940         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 941                 return bad_hva();
 942         return gfn_to_hva_memslot(slot, gfn);
 943 }
 944 EXPORT_SYMBOL_GPL(gfn_to_hva);
 945
 946 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 947 {
 948         struct page *page[1];
 949         int npages;
 950         pfn_t pfn;
 951
 952         might_sleep();
 953
 954         npages = get_user_pages_fast(addr, 1, 1, page);
 955
 956         if (unlikely(npages != 1)) {
 957                 struct vm_area_struct *vma;
 958
 959                 down_read(&current->mm->mmap_sem);
 960                 if (is_hwpoison_address(addr)) {
 961                         up_read(&current->mm->mmap_sem);
 962                         get_page(hwpoison_page);
 963                         return page_to_pfn(hwpoison_page);
 964                 }
 965
 966                 vma = find_vma(current->mm, addr);
 967
 968                 if (vma == NULL || addr < vma->vm_start ||
 969                     !(vma->vm_flags & VM_PFNMAP)) {
 970                         up_read(&current->mm->mmap_sem);
 971                         get_page(fault_page);
 972                         return page_to_pfn(fault_page);
 973                 }
 974
 975                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 976                 up_read(&current->mm->mmap_sem);
 977                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 978         } else
 979                 pfn = page_to_pfn(page[0]);
 980
 981         return pfn;
 982 }
 983
 984 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 985 {
 986         unsigned long addr;
 987
 988         addr = gfn_to_hva(kvm, gfn);
 989         if (kvm_is_error_hva(addr)) {
 990                 get_page(bad_page);
 991                 return page_to_pfn(bad_page);
 992         }
 993
 994         return hva_to_pfn(kvm, addr);
 995 }
 996 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 997
 998 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 999                          struct kvm_memory_slot *slot, gfn_t gfn)
1000 {
1001         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1002         return hva_to_pfn(kvm, addr);
1003 }
1004
1005 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1006 {
1007         pfn_t pfn;
1008
1009         pfn = gfn_to_pfn(kvm, gfn);
1010         if (!kvm_is_mmio_pfn(pfn))
1011                 return pfn_to_page(pfn);
1012
1013         WARN_ON(kvm_is_mmio_pfn(pfn));
1014
1015         get_page(bad_page);
1016         return bad_page;
1017 }
1018
1019 EXPORT_SYMBOL_GPL(gfn_to_page);
1020
1021 void kvm_release_page_clean(struct page *page)
1022 {
1023         kvm_release_pfn_clean(page_to_pfn(page));
1024 }
1025 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1026
1027 void kvm_release_pfn_clean(pfn_t pfn)
1028 {
1029         if (!kvm_is_mmio_pfn(pfn))
1030                 put_page(pfn_to_page(pfn));
1031 }
1032 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1033
1034 void kvm_release_page_dirty(struct page *page)
1035 {
1036         kvm_release_pfn_dirty(page_to_pfn(page));
1037 }
1038 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1039
1040 void kvm_release_pfn_dirty(pfn_t pfn)
1041 {
1042         kvm_set_pfn_dirty(pfn);
1043         kvm_release_pfn_clean(pfn);
1044 }
1045 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1046
1047 void kvm_set_page_dirty(struct page *page)
1048 {
1049         kvm_set_pfn_dirty(page_to_pfn(page));
1050 }
1051 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1052
1053 void kvm_set_pfn_dirty(pfn_t pfn)
1054 {
1055         if (!kvm_is_mmio_pfn(pfn)) {
1056                 struct page *page = pfn_to_page(pfn);
1057                 if (!PageReserved(page))
1058                         SetPageDirty(page);
1059         }
1060 }
1061 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1062
1063 void kvm_set_pfn_accessed(pfn_t pfn)
1064 {
1065         if (!kvm_is_mmio_pfn(pfn))
1066                 mark_page_accessed(pfn_to_page(pfn));
1067 }
1068 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1069
1070 void kvm_get_pfn(pfn_t pfn)
1071 {
1072         if (!kvm_is_mmio_pfn(pfn))
1073                 get_page(pfn_to_page(pfn));
1074 }
1075 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1076
1077 static int next_segment(unsigned long len, int offset)
1078 {
1079         if (len > PAGE_SIZE - offset)
1080                 return PAGE_SIZE - offset;
1081         else
1082                 return len;
1083 }
1084
1085 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1086                         int len)
1087 {
1088         int r;
1089         unsigned long addr;
1090
1091         addr = gfn_to_hva(kvm, gfn);
1092         if (kvm_is_error_hva(addr))
1093                 return -EFAULT;
1094         r = copy_from_user(data, (void __user *)addr + offset, len);
1095         if (r)
1096                 return -EFAULT;
1097         return 0;
1098 }
1099 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1100
1101 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1102 {
1103         gfn_t gfn = gpa >> PAGE_SHIFT;
1104         int seg;
1105         int offset = offset_in_page(gpa);
1106         int ret;
1107
1108         while ((seg = next_segment(len, offset)) != 0) {
1109                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1110                 if (ret < 0)
1111                         return ret;
1112                 offset = 0;
1113                 len -= seg;
1114                 data += seg;
1115                 ++gfn;
1116         }
1117         return 0;
1118 }
1119 EXPORT_SYMBOL_GPL(kvm_read_guest);
1120
1121 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1122                           unsigned long len)
1123 {
1124         int r;
1125         unsigned long addr;
1126         gfn_t gfn = gpa >> PAGE_SHIFT;
1127         int offset = offset_in_page(gpa);
1128
1129         addr = gfn_to_hva(kvm, gfn);
1130         if (kvm_is_error_hva(addr))
1131                 return -EFAULT;
1132         pagefault_disable();
1133         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1134         pagefault_enable();
1135         if (r)
1136                 return -EFAULT;
1137         return 0;
1138 }
1139 EXPORT_SYMBOL(kvm_read_guest_atomic);
1140
1141 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1142                          int offset, int len)
1143 {
1144         int r;
1145         unsigned long addr;
1146
1147         addr = gfn_to_hva(kvm, gfn);
1148         if (kvm_is_error_hva(addr))
1149                 return -EFAULT;
1150         r = copy_to_user((void __user *)addr + offset, data, len);
1151         if (r)
1152                 return -EFAULT;
1153         mark_page_dirty(kvm, gfn);
1154         return 0;
1155 }
1156 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1157
1158 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1159                     unsigned long len)
1160 {
1161         gfn_t gfn = gpa >> PAGE_SHIFT;
1162         int seg;
1163         int offset = offset_in_page(gpa);
1164         int ret;
1165
1166         while ((seg = next_segment(len, offset)) != 0) {
1167                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1168                 if (ret < 0)
1169                         return ret;
1170                 offset = 0;
1171                 len -= seg;
1172                 data += seg;
1173                 ++gfn;
1174         }
1175         return 0;
1176 }
1177
1178 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1179 {
1180         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1181 }
1182 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1183
1184 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1185 {
1186         gfn_t gfn = gpa >> PAGE_SHIFT;
1187         int seg;
1188         int offset = offset_in_page(gpa);
1189         int ret;
1190
1191         while ((seg = next_segment(len, offset)) != 0) {
1192                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1193                 if (ret < 0)
1194                         return ret;
1195                 offset = 0;
1196                 len -= seg;
1197                 ++gfn;
1198         }
1199         return 0;
1200 }
1201 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1202
1203 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1204 {
1205         struct kvm_memory_slot *memslot;
1206
1207         memslot = gfn_to_memslot(kvm, gfn);
1208         if (memslot && memslot->dirty_bitmap) {
1209                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1210
1211                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1212         }
1213 }
1214
1215 /*
1216  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1217  */
1218 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1219 {
1220         DEFINE_WAIT(wait);
1221
1222         for (;;) {
1223                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1224
1225                 if (kvm_arch_vcpu_runnable(vcpu)) {
1226                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1227                         break;
1228                 }
1229                 if (kvm_cpu_has_pending_timer(vcpu))
1230                         break;
1231                 if (signal_pending(current))
1232                         break;
1233
1234                 schedule();
1235         }
1236
1237         finish_wait(&vcpu->wq, &wait);
1238 }
1239
1240 void kvm_resched(struct kvm_vcpu *vcpu)
1241 {
1242         if (!need_resched())
1243                 return;
1244         cond_resched();
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_resched);
1247
1248 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1249 {
1250         ktime_t expires;
1251         DEFINE_WAIT(wait);
1252
1253         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1254
1255         /* Sleep for 100 us, and hope lock-holder got scheduled */
1256         expires = ktime_add_ns(ktime_get(), 100000UL);
1257         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1258
1259         finish_wait(&vcpu->wq, &wait);
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1262
1263 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1264 {
1265         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1266         struct page *page;
1267
1268         if (vmf->pgoff == 0)
1269                 page = virt_to_page(vcpu->run);
1270 #ifdef CONFIG_X86
1271         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1272                 page = virt_to_page(vcpu->arch.pio_data);
1273 #endif
1274 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1275         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1276                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1277 #endif
1278         else
1279                 return VM_FAULT_SIGBUS;
1280         get_page(page);
1281         vmf->page = page;
1282         return 0;
1283 }
1284
1285 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1286         .fault = kvm_vcpu_fault,
1287 };
1288
1289 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1290 {
1291         vma->vm_ops = &kvm_vcpu_vm_ops;
1292         return 0;
1293 }
1294
1295 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1296 {
1297         struct kvm_vcpu *vcpu = filp->private_data;
1298
1299         kvm_put_kvm(vcpu->kvm);
1300         return 0;
1301 }
1302
1303 static struct file_operations kvm_vcpu_fops = {
1304         .release        = kvm_vcpu_release,
1305         .unlocked_ioctl = kvm_vcpu_ioctl,
1306         .compat_ioctl   = kvm_vcpu_ioctl,
1307         .mmap           = kvm_vcpu_mmap,
1308 };
1309
1310 /*
1311  * Allocates an inode for the vcpu.
1312  */
1313 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1314 {
1315         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1316 }
1317
1318 /*
1319  * Creates some virtual cpus.  Good luck creating more than one.
1320  */
1321 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1322 {
1323         int r;
1324         struct kvm_vcpu *vcpu, *v;
1325
1326         vcpu = kvm_arch_vcpu_create(kvm, id);
1327         if (IS_ERR(vcpu))
1328                 return PTR_ERR(vcpu);
1329
1330         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1331
1332         r = kvm_arch_vcpu_setup(vcpu);
1333         if (r)
1334                 return r;
1335
1336         mutex_lock(&kvm->lock);
1337         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1338                 r = -EINVAL;
1339                 goto vcpu_destroy;
1340         }
1341
1342         kvm_for_each_vcpu(r, v, kvm)
1343                 if (v->vcpu_id == id) {
1344                         r = -EEXIST;
1345                         goto vcpu_destroy;
1346                 }
1347
1348         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1349
1350         /* Now it's all set up, let userspace reach it */
1351         kvm_get_kvm(kvm);
1352         r = create_vcpu_fd(vcpu);
1353         if (r < 0) {
1354                 kvm_put_kvm(kvm);
1355                 goto vcpu_destroy;
1356         }
1357
1358         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1359         smp_wmb();
1360         atomic_inc(&kvm->online_vcpus);
1361
1362 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1363         if (kvm->bsp_vcpu_id == id)
1364                 kvm->bsp_vcpu = vcpu;
1365 #endif
1366         mutex_unlock(&kvm->lock);
1367         return r;
1368
1369 vcpu_destroy:
1370         mutex_unlock(&kvm->lock);
1371         kvm_arch_vcpu_destroy(vcpu);
1372         return r;
1373 }
1374
1375 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1376 {
1377         if (sigset) {
1378                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1379                 vcpu->sigset_active = 1;
1380                 vcpu->sigset = *sigset;
1381         } else
1382                 vcpu->sigset_active = 0;
1383         return 0;
1384 }
1385
1386 static long kvm_vcpu_ioctl(struct file *filp,
1387                            unsigned int ioctl, unsigned long arg)
1388 {
1389         struct kvm_vcpu *vcpu = filp->private_data;
1390         void __user *argp = (void __user *)arg;
1391         int r;
1392         struct kvm_fpu *fpu = NULL;
1393         struct kvm_sregs *kvm_sregs = NULL;
1394
1395         if (vcpu->kvm->mm != current->mm)
1396                 return -EIO;
1397
1398 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1399         /*
1400          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1401          * so vcpu_load() would break it.
1402          */
1403         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1404                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1405 #endif
1406
1407
1408         vcpu_load(vcpu);
1409         switch (ioctl) {
1410         case KVM_RUN:
1411                 r = -EINVAL;
1412                 if (arg)
1413                         goto out;
1414                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1415                 break;
1416         case KVM_GET_REGS: {
1417                 struct kvm_regs *kvm_regs;
1418
1419                 r = -ENOMEM;
1420                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1421                 if (!kvm_regs)
1422                         goto out;
1423                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1424                 if (r)
1425                         goto out_free1;
1426                 r = -EFAULT;
1427                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1428                         goto out_free1;
1429                 r = 0;
1430 out_free1:
1431                 kfree(kvm_regs);
1432                 break;
1433         }
1434         case KVM_SET_REGS: {
1435                 struct kvm_regs *kvm_regs;
1436
1437                 r = -ENOMEM;
1438                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1439                 if (!kvm_regs)
1440                         goto out;
1441                 r = -EFAULT;
1442                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1443                         goto out_free2;
1444                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1445                 if (r)
1446                         goto out_free2;
1447                 r = 0;
1448 out_free2:
1449                 kfree(kvm_regs);
1450                 break;
1451         }
1452         case KVM_GET_SREGS: {
1453                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1454                 r = -ENOMEM;
1455                 if (!kvm_sregs)
1456                         goto out;
1457                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1458                 if (r)
1459                         goto out;
1460                 r = -EFAULT;
1461                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1462                         goto out;
1463                 r = 0;
1464                 break;
1465         }
1466         case KVM_SET_SREGS: {
1467                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1468                 r = -ENOMEM;
1469                 if (!kvm_sregs)
1470                         goto out;
1471                 r = -EFAULT;
1472                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1473                         goto out;
1474                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1475                 if (r)
1476                         goto out;
1477                 r = 0;
1478                 break;
1479         }
1480         case KVM_GET_MP_STATE: {
1481                 struct kvm_mp_state mp_state;
1482
1483                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1484                 if (r)
1485                         goto out;
1486                 r = -EFAULT;
1487                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1488                         goto out;
1489                 r = 0;
1490                 break;
1491         }
1492         case KVM_SET_MP_STATE: {
1493                 struct kvm_mp_state mp_state;
1494
1495                 r = -EFAULT;
1496                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1497                         goto out;
1498                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1499                 if (r)
1500                         goto out;
1501                 r = 0;
1502                 break;
1503         }
1504         case KVM_TRANSLATE: {
1505                 struct kvm_translation tr;
1506
1507                 r = -EFAULT;
1508                 if (copy_from_user(&tr, argp, sizeof tr))
1509                         goto out;
1510                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1511                 if (r)
1512                         goto out;
1513                 r = -EFAULT;
1514                 if (copy_to_user(argp, &tr, sizeof tr))
1515                         goto out;
1516                 r = 0;
1517                 break;
1518         }
1519         case KVM_SET_GUEST_DEBUG: {
1520                 struct kvm_guest_debug dbg;
1521
1522                 r = -EFAULT;
1523                 if (copy_from_user(&dbg, argp, sizeof dbg))
1524                         goto out;
1525                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1526                 if (r)
1527                         goto out;
1528                 r = 0;
1529                 break;
1530         }
1531         case KVM_SET_SIGNAL_MASK: {
1532                 struct kvm_signal_mask __user *sigmask_arg = argp;
1533                 struct kvm_signal_mask kvm_sigmask;
1534                 sigset_t sigset, *p;
1535
1536                 p = NULL;
1537                 if (argp) {
1538                         r = -EFAULT;
1539                         if (copy_from_user(&kvm_sigmask, argp,
1540                                            sizeof kvm_sigmask))
1541                                 goto out;
1542                         r = -EINVAL;
1543                         if (kvm_sigmask.len != sizeof sigset)
1544                                 goto out;
1545                         r = -EFAULT;
1546                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1547                                            sizeof sigset))
1548                                 goto out;
1549                         p = &sigset;
1550                 }
1551                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1552                 break;
1553         }
1554         case KVM_GET_FPU: {
1555                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1556                 r = -ENOMEM;
1557                 if (!fpu)
1558                         goto out;
1559                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1560                 if (r)
1561                         goto out;
1562                 r = -EFAULT;
1563                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1564                         goto out;
1565                 r = 0;
1566                 break;
1567         }
1568         case KVM_SET_FPU: {
1569                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1570                 r = -ENOMEM;
1571                 if (!fpu)
1572                         goto out;
1573                 r = -EFAULT;
1574                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1575                         goto out;
1576                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1577                 if (r)
1578                         goto out;
1579                 r = 0;
1580                 break;
1581         }
1582         default:
1583                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1584         }
1585 out:
1586         vcpu_put(vcpu);
1587         kfree(fpu);
1588         kfree(kvm_sregs);
1589         return r;
1590 }
1591
1592 static long kvm_vm_ioctl(struct file *filp,
1593                            unsigned int ioctl, unsigned long arg)
1594 {
1595         struct kvm *kvm = filp->private_data;
1596         void __user *argp = (void __user *)arg;
1597         int r;
1598
1599         if (kvm->mm != current->mm)
1600                 return -EIO;
1601         switch (ioctl) {
1602         case KVM_CREATE_VCPU:
1603                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1604                 if (r < 0)
1605                         goto out;
1606                 break;
1607         case KVM_SET_USER_MEMORY_REGION: {
1608                 struct kvm_userspace_memory_region kvm_userspace_mem;
1609
1610                 r = -EFAULT;
1611                 if (copy_from_user(&kvm_userspace_mem, argp,
1612                                                 sizeof kvm_userspace_mem))
1613                         goto out;
1614
1615                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1616                 if (r)
1617                         goto out;
1618                 break;
1619         }
1620         case KVM_GET_DIRTY_LOG: {
1621                 struct kvm_dirty_log log;
1622
1623                 r = -EFAULT;
1624                 if (copy_from_user(&log, argp, sizeof log))
1625                         goto out;
1626                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1627                 if (r)
1628                         goto out;
1629                 break;
1630         }
1631 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1632         case KVM_REGISTER_COALESCED_MMIO: {
1633                 struct kvm_coalesced_mmio_zone zone;
1634                 r = -EFAULT;
1635                 if (copy_from_user(&zone, argp, sizeof zone))
1636                         goto out;
1637                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1638                 if (r)
1639                         goto out;
1640                 r = 0;
1641                 break;
1642         }
1643         case KVM_UNREGISTER_COALESCED_MMIO: {
1644                 struct kvm_coalesced_mmio_zone zone;
1645                 r = -EFAULT;
1646                 if (copy_from_user(&zone, argp, sizeof zone))
1647                         goto out;
1648                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1649                 if (r)
1650                         goto out;
1651                 r = 0;
1652                 break;
1653         }
1654 #endif
1655         case KVM_IRQFD: {
1656                 struct kvm_irqfd data;
1657
1658                 r = -EFAULT;
1659                 if (copy_from_user(&data, argp, sizeof data))
1660                         goto out;
1661                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1662                 break;
1663         }
1664         case KVM_IOEVENTFD: {
1665                 struct kvm_ioeventfd data;
1666
1667                 r = -EFAULT;
1668                 if (copy_from_user(&data, argp, sizeof data))
1669                         goto out;
1670                 r = kvm_ioeventfd(kvm, &data);
1671                 break;
1672         }
1673 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1674         case KVM_SET_BOOT_CPU_ID:
1675                 r = 0;
1676                 mutex_lock(&kvm->lock);
1677                 if (atomic_read(&kvm->online_vcpus) != 0)
1678                         r = -EBUSY;
1679                 else
1680                         kvm->bsp_vcpu_id = arg;
1681                 mutex_unlock(&kvm->lock);
1682                 break;
1683 #endif
1684         default:
1685                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1686                 if (r == -ENOTTY)
1687                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1688         }
1689 out:
1690         return r;
1691 }
1692
1693 #ifdef CONFIG_COMPAT
1694 struct compat_kvm_dirty_log {
1695         __u32 slot;
1696         __u32 padding1;
1697         union {
1698                 compat_uptr_t dirty_bitmap; /* one bit per page */
1699                 __u64 padding2;
1700         };
1701 };
1702
1703 static long kvm_vm_compat_ioctl(struct file *filp,
1704                            unsigned int ioctl, unsigned long arg)
1705 {
1706         struct kvm *kvm = filp->private_data;
1707         int r;
1708
1709         if (kvm->mm != current->mm)
1710                 return -EIO;
1711         switch (ioctl) {
1712         case KVM_GET_DIRTY_LOG: {
1713                 struct compat_kvm_dirty_log compat_log;
1714                 struct kvm_dirty_log log;
1715
1716                 r = -EFAULT;
1717                 if (copy_from_user(&compat_log, (void __user *)arg,
1718                                    sizeof(compat_log)))
1719                         goto out;
1720                 log.slot         = compat_log.slot;
1721                 log.padding1     = compat_log.padding1;
1722                 log.padding2     = compat_log.padding2;
1723                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1724
1725                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1726                 if (r)
1727                         goto out;
1728                 break;
1729         }
1730         default:
1731                 r = kvm_vm_ioctl(filp, ioctl, arg);
1732         }
1733
1734 out:
1735         return r;
1736 }
1737 #endif
1738
1739 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1740 {
1741         struct page *page[1];
1742         unsigned long addr;
1743         int npages;
1744         gfn_t gfn = vmf->pgoff;
1745         struct kvm *kvm = vma->vm_file->private_data;
1746
1747         addr = gfn_to_hva(kvm, gfn);
1748         if (kvm_is_error_hva(addr))
1749                 return VM_FAULT_SIGBUS;
1750
1751         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1752                                 NULL);
1753         if (unlikely(npages != 1))
1754                 return VM_FAULT_SIGBUS;
1755
1756         vmf->page = page[0];
1757         return 0;
1758 }
1759
1760 static const struct vm_operations_struct kvm_vm_vm_ops = {
1761         .fault = kvm_vm_fault,
1762 };
1763
1764 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1765 {
1766         vma->vm_ops = &kvm_vm_vm_ops;
1767         return 0;
1768 }
1769
1770 static struct file_operations kvm_vm_fops = {
1771         .release        = kvm_vm_release,
1772         .unlocked_ioctl = kvm_vm_ioctl,
1773 #ifdef CONFIG_COMPAT
1774         .compat_ioctl   = kvm_vm_compat_ioctl,
1775 #endif
1776         .mmap           = kvm_vm_mmap,
1777 };
1778
1779 static int kvm_dev_ioctl_create_vm(void)
1780 {
1781         int fd, r;
1782         struct kvm *kvm;
1783
1784         kvm = kvm_create_vm();
1785         if (IS_ERR(kvm))
1786                 return PTR_ERR(kvm);
1787 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1788         r = kvm_coalesced_mmio_init(kvm);
1789         if (r < 0) {
1790                 kvm_put_kvm(kvm);
1791                 return r;
1792         }
1793 #endif
1794         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1795         if (fd < 0)
1796                 kvm_put_kvm(kvm);
1797
1798         return fd;
1799 }
1800
1801 static long kvm_dev_ioctl_check_extension_generic(long arg)
1802 {
1803         switch (arg) {
1804         case KVM_CAP_USER_MEMORY:
1805         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1806         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1807 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1808         case KVM_CAP_SET_BOOT_CPU_ID:
1809 #endif
1810         case KVM_CAP_INTERNAL_ERROR_DATA:
1811                 return 1;
1812 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1813         case KVM_CAP_IRQ_ROUTING:
1814                 return KVM_MAX_IRQ_ROUTES;
1815 #endif
1816         default:
1817                 break;
1818         }
1819         return kvm_dev_ioctl_check_extension(arg);
1820 }
1821
1822 static long kvm_dev_ioctl(struct file *filp,
1823                           unsigned int ioctl, unsigned long arg)
1824 {
1825         long r = -EINVAL;
1826
1827         switch (ioctl) {
1828         case KVM_GET_API_VERSION:
1829                 r = -EINVAL;
1830                 if (arg)
1831                         goto out;
1832                 r = KVM_API_VERSION;
1833                 break;
1834         case KVM_CREATE_VM:
1835                 r = -EINVAL;
1836                 if (arg)
1837                         goto out;
1838                 r = kvm_dev_ioctl_create_vm();
1839                 break;
1840         case KVM_CHECK_EXTENSION:
1841                 r = kvm_dev_ioctl_check_extension_generic(arg);
1842                 break;
1843         case KVM_GET_VCPU_MMAP_SIZE:
1844                 r = -EINVAL;
1845                 if (arg)
1846                         goto out;
1847                 r = PAGE_SIZE;     /* struct kvm_run */
1848 #ifdef CONFIG_X86
1849                 r += PAGE_SIZE;    /* pio data page */
1850 #endif
1851 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1852                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1853 #endif
1854                 break;
1855         case KVM_TRACE_ENABLE:
1856         case KVM_TRACE_PAUSE:
1857         case KVM_TRACE_DISABLE:
1858                 r = -EOPNOTSUPP;
1859                 break;
1860         default:
1861                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1862         }
1863 out:
1864         return r;
1865 }
1866
1867 static struct file_operations kvm_chardev_ops = {
1868         .unlocked_ioctl = kvm_dev_ioctl,
1869         .compat_ioctl   = kvm_dev_ioctl,
1870 };
1871
1872 static struct miscdevice kvm_dev = {
1873         KVM_MINOR,
1874         "kvm",
1875         &kvm_chardev_ops,
1876 };
1877
1878 static void hardware_enable(void *junk)
1879 {
1880         int cpu = raw_smp_processor_id();
1881         int r;
1882
1883         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1884                 return;
1885
1886         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1887
1888         r = kvm_arch_hardware_enable(NULL);
1889
1890         if (r) {
1891                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1892                 atomic_inc(&hardware_enable_failed);
1893                 printk(KERN_INFO "kvm: enabling virtualization on "
1894                                  "CPU%d failed\n", cpu);
1895         }
1896 }
1897
1898 static void hardware_disable(void *junk)
1899 {
1900         int cpu = raw_smp_processor_id();
1901
1902         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1903                 return;
1904         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1905         kvm_arch_hardware_disable(NULL);
1906 }
1907
1908 static void hardware_disable_all_nolock(void)
1909 {
1910         BUG_ON(!kvm_usage_count);
1911
1912         kvm_usage_count--;
1913         if (!kvm_usage_count)
1914                 on_each_cpu(hardware_disable, NULL, 1);
1915 }
1916
1917 static void hardware_disable_all(void)
1918 {
1919         spin_lock(&kvm_lock);
1920         hardware_disable_all_nolock();
1921         spin_unlock(&kvm_lock);
1922 }
1923
1924 static int hardware_enable_all(void)
1925 {
1926         int r = 0;
1927
1928         spin_lock(&kvm_lock);
1929
1930         kvm_usage_count++;
1931         if (kvm_usage_count == 1) {
1932                 atomic_set(&hardware_enable_failed, 0);
1933                 on_each_cpu(hardware_enable, NULL, 1);
1934
1935                 if (atomic_read(&hardware_enable_failed)) {
1936                         hardware_disable_all_nolock();
1937                         r = -EBUSY;
1938                 }
1939         }
1940
1941         spin_unlock(&kvm_lock);
1942
1943         return r;
1944 }
1945
1946 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1947                            void *v)
1948 {
1949         int cpu = (long)v;
1950
1951         if (!kvm_usage_count)
1952                 return NOTIFY_OK;
1953
1954         val &= ~CPU_TASKS_FROZEN;
1955         switch (val) {
1956         case CPU_DYING:
1957                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1958                        cpu);
1959                 hardware_disable(NULL);
1960                 break;
1961         case CPU_ONLINE:
1962                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1963                        cpu);
1964                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1965                 break;
1966         }
1967         return NOTIFY_OK;
1968 }
1969
1970
1971 asmlinkage void kvm_handle_fault_on_reboot(void)
1972 {
1973         if (kvm_rebooting)
1974                 /* spin while reset goes on */
1975                 while (true)
1976                         ;
1977         /* Fault while not rebooting.  We want the trace. */
1978         BUG();
1979 }
1980 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1981
1982 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1983                       void *v)
1984 {
1985         /*
1986          * Some (well, at least mine) BIOSes hang on reboot if
1987          * in vmx root mode.
1988          *
1989          * And Intel TXT required VMX off for all cpu when system shutdown.
1990          */
1991         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1992         kvm_rebooting = true;
1993         on_each_cpu(hardware_disable, NULL, 1);
1994         return NOTIFY_OK;
1995 }
1996
1997 static struct notifier_block kvm_reboot_notifier = {
1998         .notifier_call = kvm_reboot,
1999         .priority = 0,
2000 };
2001
2002 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2003 {
2004         int i;
2005
2006         for (i = 0; i < bus->dev_count; i++) {
2007                 struct kvm_io_device *pos = bus->devs[i];
2008
2009                 kvm_iodevice_destructor(pos);
2010         }
2011         kfree(bus);
2012 }
2013
2014 /* kvm_io_bus_write - called under kvm->slots_lock */
2015 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2016                      int len, const void *val)
2017 {
2018         int i;
2019         struct kvm_io_bus *bus;
2020
2021         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2022         for (i = 0; i < bus->dev_count; i++)
2023                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2024                         return 0;
2025         return -EOPNOTSUPP;
2026 }
2027
2028 /* kvm_io_bus_read - called under kvm->slots_lock */
2029 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2030                     int len, void *val)
2031 {
2032         int i;
2033         struct kvm_io_bus *bus;
2034
2035         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2036         for (i = 0; i < bus->dev_count; i++)
2037                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2038                         return 0;
2039         return -EOPNOTSUPP;
2040 }
2041
2042 /* Caller must hold slots_lock. */
2043 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2044                             struct kvm_io_device *dev)
2045 {
2046         struct kvm_io_bus *new_bus, *bus;
2047
2048         bus = kvm->buses[bus_idx];
2049         if (bus->dev_count > NR_IOBUS_DEVS-1)
2050                 return -ENOSPC;
2051
2052         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2053         if (!new_bus)
2054                 return -ENOMEM;
2055         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2056         new_bus->devs[new_bus->dev_count++] = dev;
2057         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2058         synchronize_srcu_expedited(&kvm->srcu);
2059         kfree(bus);
2060
2061         return 0;
2062 }
2063
2064 /* Caller must hold slots_lock. */
2065 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2066                               struct kvm_io_device *dev)
2067 {
2068         int i, r;
2069         struct kvm_io_bus *new_bus, *bus;
2070
2071         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2072         if (!new_bus)
2073                 return -ENOMEM;
2074
2075         bus = kvm->buses[bus_idx];
2076         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2077
2078         r = -ENOENT;
2079         for (i = 0; i < new_bus->dev_count; i++)
2080                 if (new_bus->devs[i] == dev) {
2081                         r = 0;
2082                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2083                         break;
2084                 }
2085
2086         if (r) {
2087                 kfree(new_bus);
2088                 return r;
2089         }
2090
2091         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2092         synchronize_srcu_expedited(&kvm->srcu);
2093         kfree(bus);
2094         return r;
2095 }
2096
2097 static struct notifier_block kvm_cpu_notifier = {
2098         .notifier_call = kvm_cpu_hotplug,
2099         .priority = 20, /* must be > scheduler priority */
2100 };
2101
2102 static int vm_stat_get(void *_offset, u64 *val)
2103 {
2104         unsigned offset = (long)_offset;
2105         struct kvm *kvm;
2106
2107         *val = 0;
2108         spin_lock(&kvm_lock);
2109         list_for_each_entry(kvm, &vm_list, vm_list)
2110                 *val += *(u32 *)((void *)kvm + offset);
2111         spin_unlock(&kvm_lock);
2112         return 0;
2113 }
2114
2115 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2116
2117 static int vcpu_stat_get(void *_offset, u64 *val)
2118 {
2119         unsigned offset = (long)_offset;
2120         struct kvm *kvm;
2121         struct kvm_vcpu *vcpu;
2122         int i;
2123
2124         *val = 0;
2125         spin_lock(&kvm_lock);
2126         list_for_each_entry(kvm, &vm_list, vm_list)
2127                 kvm_for_each_vcpu(i, vcpu, kvm)
2128                         *val += *(u32 *)((void *)vcpu + offset);
2129
2130         spin_unlock(&kvm_lock);
2131         return 0;
2132 }
2133
2134 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2135
2136 static const struct file_operations *stat_fops[] = {
2137         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2138         [KVM_STAT_VM]   = &vm_stat_fops,
2139 };
2140
2141 static void kvm_init_debug(void)
2142 {
2143         struct kvm_stats_debugfs_item *p;
2144
2145         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2146         for (p = debugfs_entries; p->name; ++p)
2147                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2148                                                 (void *)(long)p->offset,
2149                                                 stat_fops[p->kind]);
2150 }
2151
2152 static void kvm_exit_debug(void)
2153 {
2154         struct kvm_stats_debugfs_item *p;
2155
2156         for (p = debugfs_entries; p->name; ++p)
2157                 debugfs_remove(p->dentry);
2158         debugfs_remove(kvm_debugfs_dir);
2159 }
2160
2161 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2162 {
2163         if (kvm_usage_count)
2164                 hardware_disable(NULL);
2165         return 0;
2166 }
2167
2168 static int kvm_resume(struct sys_device *dev)
2169 {
2170         if (kvm_usage_count)
2171                 hardware_enable(NULL);
2172         return 0;
2173 }
2174
2175 static struct sysdev_class kvm_sysdev_class = {
2176         .name = "kvm",
2177         .suspend = kvm_suspend,
2178         .resume = kvm_resume,
2179 };
2180
2181 static struct sys_device kvm_sysdev = {
2182         .id = 0,
2183         .cls = &kvm_sysdev_class,
2184 };
2185
2186 struct page *bad_page;
2187 pfn_t bad_pfn;
2188
2189 static inline
2190 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2191 {
2192         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2193 }
2194
2195 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2196 {
2197         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2198
2199         kvm_arch_vcpu_load(vcpu, cpu);
2200 }
2201
2202 static void kvm_sched_out(struct preempt_notifier *pn,
2203                           struct task_struct *next)
2204 {
2205         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2206
2207         kvm_arch_vcpu_put(vcpu);
2208 }
2209
2210 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2211                   struct module *module)
2212 {
2213         int r;
2214         int cpu;
2215
2216         r = kvm_arch_init(opaque);
2217         if (r)
2218                 goto out_fail;
2219
2220         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2221
2222         if (bad_page == NULL) {
2223                 r = -ENOMEM;
2224                 goto out;
2225         }
2226
2227         bad_pfn = page_to_pfn(bad_page);
2228
2229         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2230
2231         if (hwpoison_page == NULL) {
2232                 r = -ENOMEM;
2233                 goto out_free_0;
2234         }
2235
2236         hwpoison_pfn = page_to_pfn(hwpoison_page);
2237
2238         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2239
2240         if (fault_page == NULL) {
2241                 r = -ENOMEM;
2242                 goto out_free_0;
2243         }
2244
2245         fault_pfn = page_to_pfn(fault_page);
2246
2247         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2248                 r = -ENOMEM;
2249                 goto out_free_0;
2250         }
2251
2252         r = kvm_arch_hardware_setup();
2253         if (r < 0)
2254                 goto out_free_0a;
2255
2256         for_each_online_cpu(cpu) {
2257                 smp_call_function_single(cpu,
2258                                 kvm_arch_check_processor_compat,
2259                                 &r, 1);
2260                 if (r < 0)
2261                         goto out_free_1;
2262         }
2263
2264         r = register_cpu_notifier(&kvm_cpu_notifier);
2265         if (r)
2266                 goto out_free_2;
2267         register_reboot_notifier(&kvm_reboot_notifier);
2268
2269         r = sysdev_class_register(&kvm_sysdev_class);
2270         if (r)
2271                 goto out_free_3;
2272
2273         r = sysdev_register(&kvm_sysdev);
2274         if (r)
2275                 goto out_free_4;
2276
2277         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2278         if (!vcpu_align)
2279                 vcpu_align = __alignof__(struct kvm_vcpu);
2280         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2281                                            0, NULL);
2282         if (!kvm_vcpu_cache) {
2283                 r = -ENOMEM;
2284                 goto out_free_5;
2285         }
2286
2287         kvm_chardev_ops.owner = module;
2288         kvm_vm_fops.owner = module;
2289         kvm_vcpu_fops.owner = module;
2290
2291         r = misc_register(&kvm_dev);
2292         if (r) {
2293                 printk(KERN_ERR "kvm: misc device register failed\n");
2294                 goto out_free;
2295         }
2296
2297         kvm_preempt_ops.sched_in = kvm_sched_in;
2298         kvm_preempt_ops.sched_out = kvm_sched_out;
2299
2300         kvm_init_debug();
2301
2302         return 0;
2303
2304 out_free:
2305         kmem_cache_destroy(kvm_vcpu_cache);
2306 out_free_5:
2307         sysdev_unregister(&kvm_sysdev);
2308 out_free_4:
2309         sysdev_class_unregister(&kvm_sysdev_class);
2310 out_free_3:
2311         unregister_reboot_notifier(&kvm_reboot_notifier);
2312         unregister_cpu_notifier(&kvm_cpu_notifier);
2313 out_free_2:
2314 out_free_1:
2315         kvm_arch_hardware_unsetup();
2316 out_free_0a:
2317         free_cpumask_var(cpus_hardware_enabled);
2318 out_free_0:
2319         if (fault_page)
2320                 __free_page(fault_page);
2321         if (hwpoison_page)
2322                 __free_page(hwpoison_page);
2323         __free_page(bad_page);
2324 out:
2325         kvm_arch_exit();
2326 out_fail:
2327         return r;
2328 }
2329 EXPORT_SYMBOL_GPL(kvm_init);
2330
2331 void kvm_exit(void)
2332 {
2333         kvm_exit_debug();
2334         misc_deregister(&kvm_dev);
2335         kmem_cache_destroy(kvm_vcpu_cache);
2336         sysdev_unregister(&kvm_sysdev);
2337         sysdev_class_unregister(&kvm_sysdev_class);
2338         unregister_reboot_notifier(&kvm_reboot_notifier);
2339         unregister_cpu_notifier(&kvm_cpu_notifier);
2340         on_each_cpu(hardware_disable, NULL, 1);
2341         kvm_arch_hardware_unsetup();
2342         kvm_arch_exit();
2343         free_cpumask_var(cpus_hardware_enabled);
2344         __free_page(hwpoison_page);
2345         __free_page(bad_page);
2346 }
2347 EXPORT_SYMBOL_GPL(kvm_exit);